Adding upstream version 3.40.1.upstream/3.40.1upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

45 files changed, 33308 insertions, 0 deletions

diff --git a/ext/lsm1/Makefile b/ext/lsm1/Makefile
new file mode 100644
index 0000000..7056432
--- /dev/null
+++ b/ext/lsm1/Makefile
@@ -0,0 +1,56 @@
+#
+# This Makefile is designed for use with main.mk in the root directory of
+# this project. After including main.mk, the users makefile should contain:
+#
+#    LSMDIR=$(TOP)/ext/lsm1/
+#    LSMOPTS=-fPIC
+#    include $(LSMDIR)/Makefile
+#
+# The most useful targets are [lsmtest] and [lsm.so].
+#
+
+LSMOBJ    = \
+  lsm_ckpt.o \
+  lsm_file.o \
+  lsm_log.o \
+  lsm_main.o \
+  lsm_mem.o \
+  lsm_mutex.o \
+  lsm_shared.o \
+  lsm_sorted.o \
+  lsm_str.o \
+  lsm_tree.o \
+  lsm_unix.o \
+  lsm_win32.o \
+  lsm_varint.o \
+  lsm_vtab.o
+
+LSMHDR   = \
+  $(LSMDIR)/lsm.h \
+  $(LSMDIR)/lsmInt.h
+
+LSMTESTSRC = $(LSMDIR)/lsm-test/lsmtest1.c $(LSMDIR)/lsm-test/lsmtest2.c     \
+             $(LSMDIR)/lsm-test/lsmtest3.c $(LSMDIR)/lsm-test/lsmtest4.c     \
+             $(LSMDIR)/lsm-test/lsmtest5.c $(LSMDIR)/lsm-test/lsmtest6.c     \
+             $(LSMDIR)/lsm-test/lsmtest7.c $(LSMDIR)/lsm-test/lsmtest8.c     \
+             $(LSMDIR)/lsm-test/lsmtest9.c                                   \
+             $(LSMDIR)/lsm-test/lsmtest_datasource.c \
+             $(LSMDIR)/lsm-test/lsmtest_func.c $(LSMDIR)/lsm-test/lsmtest_io.c  \
+             $(LSMDIR)/lsm-test/lsmtest_main.c $(LSMDIR)/lsm-test/lsmtest_mem.c \
+             $(LSMDIR)/lsm-test/lsmtest_tdb.c $(LSMDIR)/lsm-test/lsmtest_tdb3.c \
+             $(LSMDIR)/lsm-test/lsmtest_util.c $(LSMDIR)/lsm-test/lsmtest_win32.c
+
+
+# all: lsm.so
+
+LSMOPTS += -fPIC -DLSM_MUTEX_PTHREADS=1 -I$(LSMDIR) -DHAVE_ZLIB
+
+lsm.so:	$(LSMOBJ)
+	$(TCCX) -shared -fPIC -o lsm.so $(LSMOBJ)
+
+%.o:	$(LSMDIR)/%.c $(LSMHDR) sqlite3.h
+	$(TCCX) $(LSMOPTS) -c $<
+	
+lsmtest$(EXE): $(LSMOBJ) $(LSMTESTSRC) $(LSMTESTHDR) sqlite3.o
+	# $(TCPPX) -c $(TOP)/lsm-test/lsmtest_tdb2.cc
+	$(TCCX) $(LSMOPTS) $(LSMTESTSRC) $(LSMOBJ) sqlite3.o -o lsmtest$(EXE) $(THREADLIB) -lz
diff --git a/ext/lsm1/Makefile.msc b/ext/lsm1/Makefile.msc
new file mode 100644
index 0000000..3e5a3b3
--- /dev/null
+++ b/ext/lsm1/Makefile.msc
@@ -0,0 +1,102 @@
+#
+# This Makefile is designed for use with Makefile.msc in the root directory
+# of this project.  The Makefile.msc should contain:
+#
+#    LSMDIR=$(TOP)\ext\lsm1
+#    !INCLUDE $(LSMDIR)\Makefile.msc
+#
+# The most useful targets are [lsmtest.exe] and [lsm.dll].
+#
+
+LSMOBJ    = \
+  lsm_ckpt.lo \
+  lsm_file.lo \
+  lsm_log.lo \
+  lsm_main.lo \
+  lsm_mem.lo \
+  lsm_mutex.lo \
+  lsm_shared.lo \
+  lsm_sorted.lo \
+  lsm_str.lo \
+  lsm_tree.lo \
+  lsm_unix.lo \
+  lsm_win32.lo \
+  lsm_varint.lo \
+  lsm_vtab.lo
+
+LSMHDR   = \
+  $(LSMDIR)\lsm.h \
+  $(LSMDIR)\lsmInt.h
+
+LSMTESTSRC = $(LSMDIR)\lsm-test\lsmtest1.c $(LSMDIR)\lsm-test\lsmtest2.c     \
+             $(LSMDIR)\lsm-test\lsmtest3.c $(LSMDIR)\lsm-test\lsmtest4.c     \
+             $(LSMDIR)\lsm-test\lsmtest5.c $(LSMDIR)\lsm-test\lsmtest6.c     \
+             $(LSMDIR)\lsm-test\lsmtest7.c $(LSMDIR)\lsm-test\lsmtest8.c     \
+             $(LSMDIR)\lsm-test\lsmtest9.c                                   \
+             $(LSMDIR)\lsm-test\lsmtest_datasource.c \
+             $(LSMDIR)\lsm-test\lsmtest_func.c $(LSMDIR)\lsm-test\lsmtest_io.c  \
+             $(LSMDIR)\lsm-test\lsmtest_main.c $(LSMDIR)\lsm-test\lsmtest_mem.c \
+             $(LSMDIR)\lsm-test\lsmtest_tdb.c $(LSMDIR)\lsm-test\lsmtest_tdb3.c \
+             $(LSMDIR)\lsm-test\lsmtest_util.c $(LSMDIR)\lsm-test\lsmtest_win32.c
+
+# all: lsm.dll lsmtest.exe
+
+LSMOPTS = $(NO_WARN) -DLSM_MUTEX_WIN32=1 -I$(LSMDIR)
+
+!IF $(DEBUG)>2
+LSMOPTS = $(LSMOPTS) -DLSM_DEBUG=1
+!ENDIF
+
+!IF $(MEMDEBUG)!=0
+LSMOPTS = $(LSMOPTS) -DLSM_DEBUG_MEM=1
+!ENDIF
+
+lsm_ckpt.lo:	$(LSMDIR)\lsm_ckpt.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_ckpt.c
+
+lsm_file.lo:	$(LSMDIR)\lsm_file.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_file.c
+
+lsm_log.lo:	$(LSMDIR)\lsm_log.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_log.c
+
+lsm_main.lo:	$(LSMDIR)\lsm_main.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_main.c
+
+lsm_mem.lo:	$(LSMDIR)\lsm_mem.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_mem.c
+
+lsm_mutex.lo:	$(LSMDIR)\lsm_mutex.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_mutex.c
+
+lsm_shared.lo:	$(LSMDIR)\lsm_shared.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_shared.c
+
+lsm_sorted.lo:	$(LSMDIR)\lsm_sorted.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_sorted.c
+
+lsm_str.lo:	$(LSMDIR)\lsm_str.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_str.c
+
+lsm_tree.lo:	$(LSMDIR)\lsm_tree.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_tree.c
+
+lsm_unix.lo:	$(LSMDIR)\lsm_unix.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_unix.c
+
+lsm_win32.lo:	$(LSMDIR)\lsm_win32.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_win32.c
+
+lsm_varint.lo:	$(LSMDIR)\lsm_varint.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_varint.c
+
+lsm_vtab.lo:	$(LSMDIR)\lsm_vtab.c $(LSMHDR) $(SQLITE3H)
+	$(LTCOMPILE) $(LSMOPTS) -c $(LSMDIR)\lsm_vtab.c
+
+lsm.dll:	$(LSMOBJ)
+	$(LD) $(LDFLAGS) $(LTLINKOPTS) $(LTLIBPATHS) /DLL /OUT:$@ $(LSMOBJ)
+	copy /Y $@ $(LSMDIR)\$@
+
+lsmtest.exe: $(LSMOBJ) $(LSMTESTSRC) $(LSMTESTHDR) $(LIBOBJ)
+	$(LTLINK) $(LSMOPTS) $(LSMTESTSRC) /link $(LSMOBJ) $(LIBOBJ)
+	copy /Y $@ $(LSMDIR)\$@
diff --git a/ext/lsm1/lsm-test/README b/ext/lsm1/lsm-test/README
new file mode 100644
index 0000000..80654ee
--- /dev/null
+++ b/ext/lsm1/lsm-test/README
@@ -0,0 +1,40 @@
+
+
+Organization of test case files:
+
+  lsmtest1.c: Data tests. Tests that perform many inserts and deletes on a 
+              database file, then verify that the contents of the database can
+              be queried.
+
+  lsmtest2.c: Crash tests. Tests that attempt to verify that the database 
+              recovers correctly following an application or system crash.
+
+  lsmtest3.c: Rollback tests. Tests that focus on the explicit rollback of
+              transactions and sub-transactions.
+
+  lsmtest4.c: Multi-client tests.
+
+  lsmtest5.c: Multi-client tests with a different thread for each client.
+
+  lsmtest6.c: OOM injection tests.
+
+  lsmtest7.c: API tests.
+
+  lsmtest8.c: Writer crash tests. Tests in this file attempt to verify that
+              the system recovers and other clients proceed unaffected if
+              a process fails in the middle of a write transaction.
+
+              The difference from lsmtest2.c is that this file tests
+              live-recovery (recovery from a failure that occurs while other
+              clients are still running) whereas lsmtest2.c tests recovery
+              from a system or power failure.
+
+  lsmtest9.c: More data tests. These focus on testing that calling
+              lsm_work(nMerge=1) to compact the database does not corrupt it.
+              In other words, that databases containing block-redirects
+              can be read and written.
+
+
+
+
+
diff --git a/ext/lsm1/lsm-test/lsmtest.h b/ext/lsm1/lsm-test/lsmtest.h
new file mode 100644
index 0000000..ca60424
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest.h
@@ -0,0 +1,303 @@
+
+#ifndef __WRAPPER_INT_H_
+#define __WRAPPER_INT_H_
+
+#include "lsmtest_tdb.h"
+#include "sqlite3.h"
+#include "lsm.h"
+
+#include <assert.h>
+#include <stdarg.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#ifndef _WIN32
+# include <unistd.h>
+#endif
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <ctype.h>
+#include <stdlib.h>
+#include <errno.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef _WIN32
+# include "windows.h"
+# define gettimeofday win32GetTimeOfDay
+# define F_OK  (0)
+# define sleep(sec) Sleep(1000 * (sec))
+# define usleep(usec) Sleep(((usec) + 999) / 1000)
+# ifdef _MSC_VER
+#  include <io.h>
+#  define snprintf _snprintf
+#  define fsync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
+#  define fdatasync(fd) FlushFileBuffers((HANDLE)_get_osfhandle((fd)))
+#  define __va_copy(dst,src) ((dst) = (src))
+#  define ftruncate(fd,sz) ((_chsize_s((fd), (sz))==0) ? 0 : -1)
+# else
+#  error Unsupported C compiler for Windows.
+# endif
+int win32GetTimeOfDay(struct timeval *, void *);
+#endif
+
+#ifndef _LSM_INT_H
+typedef unsigned int  u32;
+typedef unsigned char u8;
+typedef long long int i64;
+typedef unsigned long long int u64;
+#endif
+
+
+#define ArraySize(x) ((int)(sizeof(x) / sizeof((x)[0])))
+
+#define MIN(x,y) ((x)<(y) ? (x) : (y))
+#define MAX(x,y) ((x)>(y) ? (x) : (y))
+
+#define unused_parameter(x) (void)(x)
+
+#define TESTDB_DEFAULT_PAGE_SIZE   4096
+#define TESTDB_DEFAULT_CACHE_SIZE  2048
+
+#ifndef _O_BINARY
+# define _O_BINARY (0)
+#endif
+
+/*
+** Ideally, these should be in wrapper.c. But they are here instead so that 
+** they can be used by the C++ database wrappers in wrapper2.cc.
+*/
+typedef struct DatabaseMethods DatabaseMethods;
+struct TestDb {
+  DatabaseMethods const *pMethods;          /* Database methods */
+  const char *zLibrary;                     /* Library name for tdb_open() */
+};
+struct DatabaseMethods {
+  int (*xClose)(TestDb *);
+  int (*xWrite)(TestDb *, void *, int , void *, int);
+  int (*xDelete)(TestDb *, void *, int);
+  int (*xDeleteRange)(TestDb *, void *, int, void *, int);
+  int (*xFetch)(TestDb *, void *, int, void **, int *);
+  int (*xScan)(TestDb *, void *, int, void *, int, void *, int,
+    void (*)(void *, void *, int , void *, int)
+  );
+  int (*xBegin)(TestDb *, int);
+  int (*xCommit)(TestDb *, int);
+  int (*xRollback)(TestDb *, int);
+};
+
+/* 
+** Functions in wrapper2.cc (a C++ source file). wrapper2.cc contains the
+** wrapper for Kyoto Cabinet. Kyoto cabinet has a C API, but
+** the primary interface is the C++ API.
+*/
+int test_kc_open(const char*, const char *zFilename, int bClear, TestDb **ppDb);
+int test_kc_close(TestDb *);
+int test_kc_write(TestDb *, void *, int , void *, int);
+int test_kc_delete(TestDb *, void *, int);
+int test_kc_delete_range(TestDb *, void *, int, void *, int);
+int test_kc_fetch(TestDb *, void *, int, void **, int *);
+int test_kc_scan(TestDb *, void *, int, void *, int, void *, int,
+  void (*)(void *, void *, int , void *, int)
+);
+
+int test_mdb_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_mdb_close(TestDb *);
+int test_mdb_write(TestDb *, void *, int , void *, int);
+int test_mdb_delete(TestDb *, void *, int);
+int test_mdb_fetch(TestDb *, void *, int, void **, int *);
+int test_mdb_scan(TestDb *, void *, int, void *, int, void *, int,
+  void (*)(void *, void *, int , void *, int)
+);
+
+/* 
+** Functions in wrapper3.c. This file contains the tdb wrapper for lsm.
+** The wrapper for lsm is a bit more involved than the others, as it 
+** includes code for a couple of different lsm configurations, and for
+** various types of fault injection and robustness testing.
+*/
+int test_lsm_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_lsm_lomem_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_lomem2_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_zip_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_small_open(const char*, const char*, int bClear, TestDb **ppDb);
+int test_lsm_mt2(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_lsm_mt3(const char*, const char *zFile, int bClear, TestDb **ppDb);
+
+int tdb_lsm_configure(lsm_db *, const char *);
+
+/* Functions in lsmtest_tdb4.c */
+int test_bt_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_fbt_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+int test_fbts_open(const char*, const char *zFile, int bClear, TestDb **ppDb);
+
+
+/* Functions in testutil.c. */
+int  testPrngInit(void);
+u32  testPrngValue(u32 iVal);
+void testPrngArray(u32 iVal, u32 *aOut, int nOut);
+void testPrngString(u32 iVal, char *aOut, int nOut);
+
+void testErrorInit(int argc, char **);
+void testPrintError(const char *zFormat, ...);
+void testPrintUsage(const char *zArgs);
+void testPrintFUsage(const char *zFormat, ...);
+void testTimeInit(void);
+int  testTimeGet(void);
+
+/* Functions in testmem.c. */
+void testMallocInstall(lsm_env *pEnv);
+void testMallocUninstall(lsm_env *pEnv);
+void testMallocCheck(lsm_env *pEnv, int *, int *, FILE *);
+void testMallocOom(lsm_env *pEnv, int, int, void(*)(void*), void *);
+void testMallocOomEnable(lsm_env *pEnv, int);
+
+/* lsmtest.c */
+TestDb *testOpen(const char *zSystem, int, int *pRc);
+void testReopen(TestDb **ppDb, int *pRc);
+void testClose(TestDb **ppDb);
+
+void testFetch(TestDb *, void *, int, void *, int, int *);
+void testWrite(TestDb *, void *, int, void *, int, int *);
+void testDelete(TestDb *, void *, int, int *);
+void testDeleteRange(TestDb *, void *, int, void *, int, int *);
+void testWriteStr(TestDb *, const char *, const char *zVal, int *pRc);
+void testFetchStr(TestDb *, const char *, const char *, int *pRc);
+
+void testBegin(TestDb *pDb, int iTrans, int *pRc);
+void testCommit(TestDb *pDb, int iTrans, int *pRc);
+
+void test_failed(void);
+
+char *testMallocPrintf(const char *zFormat, ...);
+char *testMallocVPrintf(const char *zFormat, va_list ap);
+int testGlobMatch(const char *zPattern, const char *zStr);
+
+void testScanCompare(TestDb *, TestDb *, int, void *, int, void *, int, int *);
+void testFetchCompare(TestDb *, TestDb *, void *, int, int *);
+
+void *testMalloc(int);
+void *testMallocCopy(void *pCopy, int nByte);
+void *testRealloc(void *, int);
+void testFree(void *);
+
+/* lsmtest_bt.c */
+int do_bt(int nArg, char **azArg);
+
+/* testio.c */
+int testVfsConfigureDb(TestDb *pDb);
+
+/* testfunc.c */
+int do_show(int nArg, char **azArg);
+int do_work(int nArg, char **azArg);
+
+/* testio.c */
+int do_io(int nArg, char **azArg);
+
+/* lsmtest2.c */
+void do_crash_test(const char *zPattern, int *pRc);
+int do_rollback_test(int nArg, char **azArg);
+
+/* test3.c */
+void test_rollback(const char *zSystem, const char *zPattern, int *pRc);
+
+/* test4.c */
+void test_mc(const char *zSystem, const char *zPattern, int *pRc);
+
+/* test5.c */
+void test_mt(const char *zSystem, const char *zPattern, int *pRc);
+
+/* lsmtest6.c */
+void test_oom(const char *zPattern, int *pRc);
+void testDeleteLsmdb(const char *zFile);
+
+void testSaveDb(const char *zFile, const char *zAuxExt);
+void testRestoreDb(const char *zFile, const char *zAuxExt);
+void testCopyLsmdb(const char *zFrom, const char *zTo);
+
+/* lsmtest7.c */
+void test_api(const char *zPattern, int *pRc);
+
+/* lsmtest8.c */
+void do_writer_crash_test(const char *zPattern, int *pRc);
+
+/*************************************************************************
+** Interface to functionality in test_datasource.c.
+*/
+typedef struct Datasource Datasource;
+typedef struct DatasourceDefn DatasourceDefn;
+
+struct DatasourceDefn {
+  int eType;                      /* A TEST_DATASOURCE_* value */
+  int nMinKey;                    /* Minimum key size */
+  int nMaxKey;                    /* Maximum key size */
+  int nMinVal;                    /* Minimum value size */
+  int nMaxVal;                    /* Maximum value size */
+};
+
+#define TEST_DATASOURCE_RANDOM    1
+#define TEST_DATASOURCE_SEQUENCE  2
+
+char *testDatasourceName(const DatasourceDefn *);
+Datasource *testDatasourceNew(const DatasourceDefn *);
+void testDatasourceFree(Datasource *);
+void testDatasourceEntry(Datasource *, int, void **, int *, void **, int *);
+/* End of test_datasource.c interface.
+*************************************************************************/
+void testDatasourceFetch(
+  TestDb *pDb,                    /* Database handle */
+  Datasource *pData,
+  int iKey,
+  int *pRc                        /* IN/OUT: Error code */
+);
+
+void testWriteDatasource(TestDb *, Datasource *, int, int *);
+void testWriteDatasourceRange(TestDb *, Datasource *, int, int, int *);
+void testDeleteDatasource(TestDb *, Datasource *, int, int *);
+void testDeleteDatasourceRange(TestDb *, Datasource *, int, int, int *);
+
+
+/* test1.c */
+void test_data_1(const char *, const char *, int *pRc);
+void test_data_2(const char *, const char *, int *pRc);
+void test_data_3(const char *, const char *, int *pRc);
+void testDbContents(TestDb *, Datasource *, int, int, int, int, int, int *);
+void testCaseProgress(int, int, int, int *);
+int testCaseNDot(void);
+
+void testCompareDb(Datasource *, int, int, TestDb *, TestDb *, int *);
+int testControlDb(TestDb **ppDb);
+
+typedef struct CksumDb CksumDb;
+CksumDb *testCksumArrayNew(Datasource *, int, int, int);
+char *testCksumArrayGet(CksumDb *, int);
+void testCksumArrayFree(CksumDb *);
+void testCaseStart(int *pRc, char *zFmt, ...);
+void testCaseFinish(int rc);
+void testCaseSkip(void);
+int testCaseBegin(int *, const char *, const char *, ...);
+
+#define TEST_CKSUM_BYTES 29
+int testCksumDatabase(TestDb *pDb, char *zOut);
+int testCountDatabase(TestDb *pDb);
+void testCompareInt(int, int, int *);
+void testCompareStr(const char *z1, const char *z2, int *pRc);
+
+/* lsmtest9.c */
+void test_data_4(const char *, const char *, int *pRc);
+
+
+/*
+** Similar to the Tcl_GetIndexFromObjStruct() Tcl library function.
+*/
+#define testArgSelect(w,x,y,z) testArgSelectX(w,x,sizeof(w[0]),y,z)
+int testArgSelectX(void *, const char *, int, const char *, int *);
+
+#ifdef __cplusplus
+}  /* End of the 'extern "C"' block */
+#endif
+
+#endif
diff --git a/ext/lsm1/lsm-test/lsmtest1.c b/ext/lsm1/lsm-test/lsmtest1.c
new file mode 100644
index 0000000..1ce2cc0
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest1.c
@@ -0,0 +1,656 @@
+
+#include "lsmtest.h"
+
+#define DATA_SEQUENTIAL TEST_DATASOURCE_SEQUENCE
+#define DATA_RANDOM     TEST_DATASOURCE_RANDOM
+
+typedef struct Datatest1 Datatest1;
+typedef struct Datatest2 Datatest2;
+
+/*
+** An instance of the following structure contains parameters used to
+** customize the test function in this file. Test procedure:
+**
+**   1. Create a data-source based on the "datasource definition" vars.
+**
+**   2. Insert nRow key value pairs into the database.
+**
+**   3. Delete all keys from the database. Deletes are done in the same 
+**      order as the inserts.
+**
+** During steps 2 and 3 above, after each Datatest1.nVerify inserts or
+** deletes, the following:
+**
+**   a. Run Datasource.nTest key lookups and check the results are as expected.
+**
+**   b. If Datasource.bTestScan is true, run a handful (8) of range
+**      queries (scanning forwards and backwards). Check that the results
+**      are as expected.
+**
+**   c. Close and reopen the database. Then run (a) and (b) again.
+*/
+struct Datatest1 {
+  /* Datasource definition */
+  DatasourceDefn defn;
+
+  /* Test procedure parameters */
+  int nRow;                       /* Number of rows to insert then delete */
+  int nVerify;                    /* How often to verify the db contents */
+  int nTest;                      /* Number of keys to test (0==all) */
+  int bTestScan;                  /* True to do scan tests */
+};
+
+/*
+** An instance of the following data structure is used to describe the
+** second type of test case in this file. The chief difference between 
+** these tests and those described by Datatest1 is that these tests also
+** experiment with range-delete operations. Tests proceed as follows:
+**
+**     1. Open the datasource described by Datatest2.defn. 
+**
+**     2. Open a connection on an empty database.
+**
+**     3. Do this Datatest2.nIter times:
+**
+**        a) Insert Datatest2.nWrite key-value pairs from the datasource.
+**
+**        b) Select two pseudo-random keys and use them as the start
+**           and end points of a range-delete operation.
+**
+**        c) Verify that the contents of the database are as expected (see
+**           below for details).
+**
+**        d) Close and then reopen the database handle.
+**
+**        e) Verify that the contents of the database are still as expected.
+**
+** The inserts and range deletes are run twice - once on the database being
+** tested and once using a control system (sqlite3, kc etc. - something that 
+** works). In order to verify that the contents of the db being tested are
+** correct, the test runs a bunch of scans and lookups on both the test and
+** control databases. If the results are the same, the test passes.
+*/
+struct Datatest2 {
+  DatasourceDefn defn;
+  int nRange;
+  int nWrite;                     /* Number of writes per iteration */
+  int nIter;                      /* Total number of iterations to run */
+};
+
+/*
+** Generate a unique name for the test case pTest with database system
+** zSystem.
+*/
+static char *getName(const char *zSystem, int bRecover, Datatest1 *pTest){
+  char *zRet;
+  char *zData;
+  zData = testDatasourceName(&pTest->defn);
+  zRet = testMallocPrintf("data.%s.%s.rec=%d.%d.%d", 
+      zSystem, zData, bRecover, pTest->nRow, pTest->nVerify
+  );
+  testFree(zData);
+  return zRet;
+}
+
+int testControlDb(TestDb **ppDb){
+#ifdef HAVE_KYOTOCABINET
+  return tdb_open("kyotocabinet", "tmp.db", 1, ppDb);
+#else
+  return tdb_open("sqlite3", "", 1, ppDb);
+#endif
+}
+
+void testDatasourceFetch(
+  TestDb *pDb,                    /* Database handle */
+  Datasource *pData,
+  int iKey,
+  int *pRc                        /* IN/OUT: Error code */
+){
+  void *pKey; int nKey;           /* Database key to query for */
+  void *pVal; int nVal;           /* Expected result of query */
+
+  testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+  testFetch(pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+/*
+** This function is called to test that the contents of database pDb
+** are as expected. In this case, expected is defined as containing
+** key-value pairs iFirst through iLast, inclusive, from data source 
+** pData. In other words, a loop like the following could be used to
+** construct a database with identical contents from scratch.
+**
+**   for(i=iFirst; i<=iLast; i++){
+**     testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
+**     // insert (pKey, nKey) -> (pVal, nVal) into database
+**   }
+**
+** The key domain consists of keys 0 to (nRow-1), inclusive, from
+** data source pData. For both scan and lookup tests, keys are selected
+** pseudo-randomly from within this set.
+**
+** This function runs nLookupTest lookup tests and nScanTest scan tests.
+**
+** A lookup test consists of selecting a key from the domain and querying
+** pDb for it. The test fails if the presence of the key and, if present,
+** the associated value do not match the expectations defined above.
+**
+** A scan test involves selecting a key from the domain and running
+** the following queries:
+**
+**   1. Scan all keys equal to or greater than the key, in ascending order.
+**   2. Scan all keys equal to or smaller than the key, in descending order.
+**
+** Additionally, if nLookupTest is greater than zero, the following are
+** run once:
+**
+**   1. Scan all keys in the db, in ascending order.
+**   2. Scan all keys in the db, in descending order.
+**
+** As you would assume, the test fails if the returned values do not match
+** expectations.
+*/
+void testDbContents(
+  TestDb *pDb,                    /* Database handle being tested */
+  Datasource *pData,              /* pDb contains data from here */
+  int nRow,                       /* Size of key domain */
+  int iFirst,                     /* Index of first key from pData in pDb */
+  int iLast,                      /* Index of last key from pData in pDb */
+  int nLookupTest,                /* Number of lookup tests to run */
+  int nScanTest,                  /* Number of scan tests to run */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int j;
+  int rc = *pRc;
+
+  if( rc==0 && nScanTest ){
+    TestDb *pDb2 = 0;
+
+    /* Open a control db (i.e. one that we assume works) */
+    rc = testControlDb(&pDb2);
+
+    for(j=iFirst; rc==0 && j<=iLast; j++){
+      void *pKey; int nKey;         /* Database key to insert */
+      void *pVal; int nVal;         /* Database value to insert */
+      testDatasourceEntry(pData, j, &pKey, &nKey, &pVal, &nVal);
+      rc = tdb_write(pDb2, pKey, nKey, pVal, nVal);
+    }
+
+    if( rc==0 ){
+      int iKey1;
+      int iKey2;
+      void *pKey1; int nKey1;       /* Start key */
+      void *pKey2; int nKey2;       /* Final key */
+
+      iKey1 = testPrngValue((iFirst<<8) + (iLast<<16)) % nRow;
+      iKey2 = testPrngValue((iLast<<8) + (iFirst<<16)) % nRow;
+      testDatasourceEntry(pData, iKey1, &pKey2, &nKey1, 0, 0);
+      pKey1 = testMalloc(nKey1+1);
+      memcpy(pKey1, pKey2, nKey1+1);
+      testDatasourceEntry(pData, iKey2, &pKey2, &nKey2, 0, 0);
+
+      testScanCompare(pDb2, pDb, 0, 0, 0,         0, 0,         &rc);
+      testScanCompare(pDb2, pDb, 0, 0, 0,         pKey2, nKey2, &rc);
+      testScanCompare(pDb2, pDb, 0, pKey1, nKey1, 0, 0,         &rc);
+      testScanCompare(pDb2, pDb, 0, pKey1, nKey1, pKey2, nKey2, &rc);
+      testScanCompare(pDb2, pDb, 1, 0, 0,         0, 0,         &rc);
+      testScanCompare(pDb2, pDb, 1, 0, 0,         pKey2, nKey2, &rc);
+      testScanCompare(pDb2, pDb, 1, pKey1, nKey1, 0, 0,         &rc);
+      testScanCompare(pDb2, pDb, 1, pKey1, nKey1, pKey2, nKey2, &rc);
+      testFree(pKey1);
+    }
+    tdb_close(pDb2);
+  }
+
+  /* Test some lookups. */
+  for(j=0; rc==0 && j<nLookupTest; j++){
+    int iKey;                     /* Datasource key to test */
+    void *pKey; int nKey;         /* Database key to query for */
+    void *pVal; int nVal;         /* Expected result of query */
+
+    if( nLookupTest>=nRow ){
+      iKey = j;
+    }else{
+      iKey = testPrngValue(j + (iFirst<<8) + (iLast<<16)) % nRow;
+    }
+
+    testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+    if( iFirst>iKey || iKey>iLast ){
+      pVal = 0;
+      nVal = -1;
+    }
+
+    testFetch(pDb, pKey, nKey, pVal, nVal, &rc);
+  }
+
+  *pRc = rc;
+}
+
+/*
+** This function should be called during long running test cases to output
+** the progress dots (...) to stdout.
+*/
+void testCaseProgress(int i, int n, int nDot, int *piDot){
+  int iDot = *piDot;
+  while( iDot < ( ((nDot*2+1) * i) / (n*2) ) ){
+    printf(".");
+    fflush(stdout);
+    iDot++;
+  }
+  *piDot = iDot;
+}
+
+int testCaseNDot(void){ return 20; }
+
+#if 0
+static void printScanCb(
+    void *pCtx, void *pKey, int nKey, void *pVal, int nVal
+){
+  printf("%s\n", (char *)pKey);
+  fflush(stdout);
+}
+#endif
+
+void testReopenRecover(TestDb **ppDb, int *pRc){
+  if( *pRc==0 ){
+    const char *zLib = tdb_library_name(*ppDb);
+    const char *zDflt = tdb_default_db(zLib);
+    testCopyLsmdb(zDflt, "bak.db");
+    testClose(ppDb);
+    testCopyLsmdb("bak.db", zDflt);
+    *pRc = tdb_open(zLib, 0, 0, ppDb);
+  }
+}
+
+
+static void doDataTest1(
+  const char *zSystem,            /* Database system to test */
+  int bRecover,
+  Datatest1 *p,                   /* Structure containing test parameters */
+  int *pRc                        /* OUT: Error code */
+){
+  int i;
+  int iDot;
+  int rc = LSM_OK;
+  Datasource *pData;
+  TestDb *pDb;
+  int iToggle = 0;
+
+  /* Start the test case, open a database and allocate the datasource. */
+  pDb = testOpen(zSystem, 1, &rc);
+  pData = testDatasourceNew(&p->defn);
+
+  i = 0;
+  iDot = 0;
+  while( rc==LSM_OK && i<p->nRow ){
+
+    /* Insert some data */
+    testWriteDatasourceRange(pDb, pData, i, p->nVerify, &rc);
+    i += p->nVerify;
+
+    if( iToggle ) testBegin(pDb, 1, &rc);
+    /* Check that the db content is correct. */
+    testDbContents(pDb, pData, p->nRow, 0, i-1, p->nTest, p->bTestScan, &rc);
+    if( iToggle ) testCommit(pDb, 0, &rc);
+    iToggle = (iToggle+1)%2;
+
+    if( bRecover ){
+      testReopenRecover(&pDb, &rc);
+    }else{
+      testReopen(&pDb, &rc);
+    }
+
+    /* Check that the db content is still correct. */
+    testDbContents(pDb, pData, p->nRow, 0, i-1, p->nTest, p->bTestScan, &rc);
+
+    /* Update the progress dots... */
+    testCaseProgress(i, p->nRow, testCaseNDot()/2, &iDot);
+  }
+
+  i = 0;
+  iDot = 0;
+  while( rc==LSM_OK && i<p->nRow ){
+
+    /* Delete some entries */
+    testDeleteDatasourceRange(pDb, pData, i, p->nVerify, &rc);
+    i += p->nVerify;
+
+    /* Check that the db content is correct. */
+    testDbContents(pDb, pData, p->nRow, i, p->nRow-1,p->nTest,p->bTestScan,&rc);
+
+    /* Close and reopen the database. */
+    if( bRecover ){
+      testReopenRecover(&pDb, &rc);
+    }else{
+      testReopen(&pDb, &rc);
+    }
+
+    /* Check that the db content is still correct. */
+    testDbContents(pDb, pData, p->nRow, i, p->nRow-1,p->nTest,p->bTestScan,&rc);
+
+    /* Update the progress dots... */
+    testCaseProgress(i, p->nRow, testCaseNDot()/2, &iDot);
+  }
+
+  /* Free the datasource, close the database and finish the test case. */
+  testDatasourceFree(pData);
+  tdb_close(pDb);
+  testCaseFinish(rc);
+  *pRc = rc;
+}
+
+
+void test_data_1(
+  const char *zSystem,            /* Database system name */
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  Datatest1 aTest[] = {
+    { {DATA_RANDOM,     500,600,   1000,2000},     1000,  100,  10,  0},
+    { {DATA_RANDOM,     20,25,     100,200},       1000,  250, 1000, 1},
+    { {DATA_RANDOM,     8,10,      100,200},       1000,  250, 1000, 1},
+    { {DATA_RANDOM,     8,10,      10,20},         1000,  250, 1000, 1},
+    { {DATA_RANDOM,     8,10,      1000,2000},     1000,  250, 1000, 1},
+    { {DATA_RANDOM,     8,100,     10000,20000},    100,   25,  100, 1},
+    { {DATA_RANDOM,     80,100,    10,20},         1000,  250, 1000, 1},
+    { {DATA_RANDOM,     5000,6000, 10,20},          100,   25,  100, 1},
+    { {DATA_SEQUENTIAL, 5,10,      10,20},         1000,  250, 1000, 1},
+    { {DATA_SEQUENTIAL, 5,10,      100,200},       1000,  250, 1000, 1},
+    { {DATA_SEQUENTIAL, 5,10,      1000,2000},     1000,  250, 1000, 1},
+    { {DATA_SEQUENTIAL, 5,100,     10000,20000},    100,   25,  100, 1},
+    { {DATA_RANDOM,     10,10,     100,100},     100000, 1000,  100, 0},
+    { {DATA_SEQUENTIAL, 10,10,     100,100},     100000, 1000,  100, 0},
+  };
+
+  int i;
+  int bRecover;
+
+  for(bRecover=0; bRecover<2; bRecover++){
+    if( bRecover==1 && memcmp(zSystem, "lsm", 3) ) break;
+    for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+      char *zName = getName(zSystem, bRecover, &aTest[i]);
+      if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+        doDataTest1(zSystem, bRecover, &aTest[i], pRc);
+      }
+      testFree(zName);
+    }
+  }
+}
+
+void testCompareDb(
+  Datasource *pData,
+  int nData,
+  int iSeed,
+  TestDb *pControl,
+  TestDb *pDb,
+  int *pRc
+){
+  int i;
+
+  static int nCall = 0;
+  nCall++;
+
+  testScanCompare(pControl, pDb, 0, 0, 0,         0, 0,         pRc);
+  testScanCompare(pControl, pDb, 1, 0, 0,         0, 0,         pRc);
+
+  if( *pRc==0 ){
+    int iKey1;
+    int iKey2;
+    void *pKey1; int nKey1;       /* Start key */
+    void *pKey2; int nKey2;       /* Final key */
+
+    iKey1 = testPrngValue(iSeed) % nData;
+    iKey2 = testPrngValue(iSeed+1) % nData;
+    testDatasourceEntry(pData, iKey1, &pKey2, &nKey1, 0, 0);
+    pKey1 = testMalloc(nKey1+1);
+    memcpy(pKey1, pKey2, nKey1+1);
+    testDatasourceEntry(pData, iKey2, &pKey2, &nKey2, 0, 0);
+
+    testScanCompare(pControl, pDb, 0, 0, 0,         pKey2, nKey2, pRc);
+    testScanCompare(pControl, pDb, 0, pKey1, nKey1, 0, 0,         pRc);
+    testScanCompare(pControl, pDb, 0, pKey1, nKey1, pKey2, nKey2, pRc);
+    testScanCompare(pControl, pDb, 1, 0, 0,         pKey2, nKey2, pRc);
+    testScanCompare(pControl, pDb, 1, pKey1, nKey1, 0, 0,         pRc);
+    testScanCompare(pControl, pDb, 1, pKey1, nKey1, pKey2, nKey2, pRc);
+    testFree(pKey1);
+  }
+
+  for(i=0; i<nData && *pRc==0; i++){
+    void *pKey; int nKey;
+    testDatasourceEntry(pData, i, &pKey, &nKey, 0, 0);
+    testFetchCompare(pControl, pDb, pKey, nKey, pRc);
+  }
+}
+
+static void doDataTest2(
+  const char *zSystem,            /* Database system to test */
+  int bRecover,
+  Datatest2 *p,                   /* Structure containing test parameters */
+  int *pRc                        /* OUT: Error code */
+){
+  TestDb *pDb;
+  TestDb *pControl;
+  Datasource *pData;
+  int i;
+  int rc = LSM_OK;
+  int iDot = 0;
+
+  /* Start the test case, open a database and allocate the datasource. */
+  pDb = testOpen(zSystem, 1, &rc);
+  pData = testDatasourceNew(&p->defn);
+  rc = testControlDb(&pControl);
+
+  if( tdb_lsm(pDb) ){
+    int nBuf = 32 * 1024 * 1024;
+    lsm_config(tdb_lsm(pDb), LSM_CONFIG_AUTOFLUSH, &nBuf);
+  }
+
+  for(i=0; rc==0 && i<p->nIter; i++){
+    void *pKey1; int nKey1;
+    void *pKey2; int nKey2;
+    int ii;
+    int nRange = MIN(p->nIter*p->nWrite, p->nRange);
+
+    for(ii=0; rc==0 && ii<p->nWrite; ii++){
+      int iKey = (i*p->nWrite + ii) % p->nRange;
+      testWriteDatasource(pControl, pData, iKey, &rc);
+      testWriteDatasource(pDb, pData, iKey, &rc);
+    }
+
+    testDatasourceEntry(pData, i+1000000, &pKey1, &nKey1, 0, 0);
+    pKey1 = testMallocCopy(pKey1, nKey1);
+    testDatasourceEntry(pData, i+2000000, &pKey2, &nKey2, 0, 0);
+
+    testDeleteRange(pDb, pKey1, nKey1, pKey2, nKey2, &rc);
+    testDeleteRange(pControl, pKey1, nKey1, pKey2, nKey2, &rc);
+    testFree(pKey1);
+
+    testCompareDb(pData, nRange, i, pControl, pDb, &rc);
+    if( bRecover ){
+      testReopenRecover(&pDb, &rc);
+    }else{
+      testReopen(&pDb, &rc);
+    }
+    testCompareDb(pData, nRange, i, pControl, pDb, &rc);
+
+    /* Update the progress dots... */
+    testCaseProgress(i, p->nIter, testCaseNDot(), &iDot);
+  }
+
+  testClose(&pDb);
+  testClose(&pControl);
+  testDatasourceFree(pData);
+  testCaseFinish(rc);
+  *pRc = rc;
+}
+
+static char *getName2(const char *zSystem, int bRecover, Datatest2 *pTest){
+  char *zRet;
+  char *zData;
+  zData = testDatasourceName(&pTest->defn);
+  zRet = testMallocPrintf("data2.%s.%s.rec=%d.%d.%d.%d", 
+      zSystem, zData, bRecover, pTest->nRange, pTest->nWrite, pTest->nIter
+  );
+  testFree(zData);
+  return zRet;
+}
+
+void test_data_2(
+  const char *zSystem,            /* Database system name */
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  Datatest2 aTest[] = {
+      /* defn,                                 nRange, nWrite, nIter */
+    { {DATA_RANDOM,     20,25,     100,200},   10000,  10,     50   },
+    { {DATA_RANDOM,     20,25,     100,200},   10000,  200,    50   },
+    { {DATA_RANDOM,     20,25,     100,200},   100,    10,     1000 },
+    { {DATA_RANDOM,     20,25,     100,200},   100,    200,    50   },
+  };
+
+  int i;
+  int bRecover;
+
+  for(bRecover=0; bRecover<2; bRecover++){
+    if( bRecover==1 && memcmp(zSystem, "lsm", 3) ) break;
+    for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+      char *zName = getName2(zSystem, bRecover, &aTest[i]);
+      if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+        doDataTest2(zSystem, bRecover, &aTest[i], pRc);
+      }
+      testFree(zName);
+    }
+  }
+}
+
+/*************************************************************************
+** Test case data3.*
+*/
+
+typedef struct Datatest3 Datatest3;
+struct Datatest3 {
+  int nRange;                     /* Keys are between 1 and this value, incl. */
+  int nIter;                      /* Number of iterations */
+  int nWrite;                     /* Number of writes per iteration */
+  int nDelete;                    /* Number of deletes per iteration */
+
+  int nValMin;                    /* Minimum value size for writes */
+  int nValMax;                    /* Maximum value size for writes */
+};
+
+void testPutU32(u8 *aBuf, u32 iVal){
+  aBuf[0] = (iVal >> 24) & 0xFF;
+  aBuf[1] = (iVal >> 16) & 0xFF;
+  aBuf[2] = (iVal >>  8) & 0xFF;
+  aBuf[3] = (iVal >>  0) & 0xFF;
+}
+
+void dt3PutKey(u8 *aBuf, int iKey){
+  assert( iKey<100000 && iKey>=0 );
+  sprintf((char *)aBuf, "%.5d", iKey);
+}
+
+static void doDataTest3(
+  const char *zSystem,            /* Database system to test */
+  Datatest3 *p,                   /* Structure containing test parameters */
+  int *pRc                        /* OUT: Error code */
+){
+  int iDot = 0;
+  int rc = *pRc;
+  TestDb *pDb;
+  u8 *abPresent;                  /* Array of boolean */
+  char *aVal;                     /* Buffer to hold values */
+  int i;
+  u32 iSeq = 10;                  /* prng counter */
+
+  abPresent = (u8 *)testMalloc(p->nRange+1);
+  aVal = (char *)testMalloc(p->nValMax+1);
+  pDb = testOpen(zSystem, 1, &rc);
+
+  for(i=0; i<p->nIter && rc==0; i++){
+    int ii;
+
+    testCaseProgress(i, p->nIter, testCaseNDot(), &iDot);
+
+    /* Perform nWrite inserts */
+    for(ii=0; ii<p->nWrite; ii++){
+      u8 aKey[6];
+      u32 iKey;
+      int nVal;
+
+      iKey = (testPrngValue(iSeq++) % p->nRange) + 1;
+      nVal = (testPrngValue(iSeq++) % (p->nValMax - p->nValMin)) + p->nValMin;
+      testPrngString(testPrngValue(iSeq++), aVal, nVal);
+      dt3PutKey(aKey, iKey);
+
+      testWrite(pDb, aKey, sizeof(aKey)-1, aVal, nVal, &rc);
+      abPresent[iKey] = 1;
+    }
+
+    /* Perform nDelete deletes */
+    for(ii=0; ii<p->nDelete; ii++){
+      u8 aKey1[6];
+      u8 aKey2[6];
+      u32 iKey;
+
+      iKey = (testPrngValue(iSeq++) % p->nRange) + 1;
+      dt3PutKey(aKey1, iKey-1);
+      dt3PutKey(aKey2, iKey+1);
+
+      testDeleteRange(pDb, aKey1, sizeof(aKey1)-1, aKey2, sizeof(aKey2)-1, &rc);
+      abPresent[iKey] = 0;
+    }
+
+    testReopen(&pDb, &rc);
+
+    for(ii=1; rc==0 && ii<=p->nRange; ii++){
+      int nDbVal;
+      void *pDbVal;
+      u8 aKey[6];
+      int dbrc;
+
+      dt3PutKey(aKey, ii);
+      dbrc = tdb_fetch(pDb, aKey, sizeof(aKey)-1, &pDbVal, &nDbVal);
+      testCompareInt(0, dbrc, &rc);
+
+      if( abPresent[ii] ){
+        testCompareInt(1, (nDbVal>0), &rc);
+      }else{
+        testCompareInt(1, (nDbVal<0), &rc);
+      }
+    }
+  }
+
+  testClose(&pDb);
+  testCaseFinish(rc);
+  *pRc = rc;
+}
+
+static char *getName3(const char *zSystem, Datatest3 *p){
+  return testMallocPrintf("data3.%s.%d.%d.%d.%d.(%d..%d)",
+      zSystem, p->nRange, p->nIter, p->nWrite, p->nDelete, 
+      p->nValMin, p->nValMax
+  );
+}
+
+void test_data_3(
+  const char *zSystem,            /* Database system name */
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  Datatest3 aTest[] = {
+    /* nRange, nIter, nWrite, nDelete, nValMin, nValMax */
+    {  100,    1000,  5,      5,       50,      100 },
+    {  100,    1000,  2,      2,        5,       10 },
+  };
+
+  int i;
+
+  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+    char *zName = getName3(zSystem, &aTest[i]);
+    if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+      doDataTest3(zSystem, &aTest[i], pRc);
+    }
+    testFree(zName);
+  }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest2.c b/ext/lsm1/lsm-test/lsmtest2.c
new file mode 100644
index 0000000..d2ef0eb
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest2.c
@@ -0,0 +1,488 @@
+
+/*
+** This file contains tests related to recovery following application 
+** and system crashes (power failures) while writing to the database.
+*/
+
+#include "lsmtest.h"
+
+/*
+** Structure used by testCksumDatabase() to accumulate checksum values in.
+*/
+typedef struct Cksum Cksum;
+struct Cksum {
+  int nRow;
+  int cksum1;
+  int cksum2;
+};
+
+/*
+** tdb_scan() callback used by testCksumDatabase()
+*/
+static void scanCksumDb(
+  void *pCtx, 
+  void *pKey, int nKey,
+  void *pVal, int nVal
+){
+  Cksum *p = (Cksum *)pCtx;
+  int i;
+
+  p->nRow++;
+  for(i=0; i<nKey; i++){
+    p->cksum1 += ((u8 *)pKey)[i];
+    p->cksum2 += p->cksum1;
+  }
+  for(i=0; i<nVal; i++){
+    p->cksum1 += ((u8 *)pVal)[i];
+    p->cksum2 += p->cksum1;
+  }
+}
+
+/*
+** tdb_scan() callback used by testCountDatabase()
+*/
+static void scanCountDb(
+  void *pCtx, 
+  void *pKey, int nKey,
+  void *pVal, int nVal
+){
+  Cksum *p = (Cksum *)pCtx;
+  p->nRow++;
+
+  unused_parameter(pKey);
+  unused_parameter(nKey);
+  unused_parameter(pVal);
+  unused_parameter(nVal);
+}
+
+
+/*
+** Iterate through the entire contents of database pDb. Write a checksum
+** string based on the db contents into buffer zOut before returning. A
+** checksum string is at most 29 (TEST_CKSUM_BYTES) bytes in size:
+**
+**    * 32-bit integer (10 bytes)
+**    * 1 space        (1 byte)
+**    * 32-bit hex     (8 bytes)
+**    * 1 space        (1 byte)
+**    * 32-bit hex     (8 bytes)
+**    * nul-terminator (1 byte)
+**
+** The number of entries in the database is returned.
+*/
+int testCksumDatabase(
+  TestDb *pDb,                    /* Database handle */
+  char *zOut                      /* Buffer to write checksum to */
+){
+  Cksum cksum;
+  memset(&cksum, 0, sizeof(Cksum));
+  tdb_scan(pDb, (void *)&cksum, 0, 0, 0, 0, 0, scanCksumDb);
+  sprintf(zOut, "%d %x %x", 
+      cksum.nRow, (u32)cksum.cksum1, (u32)cksum.cksum2
+  );
+  assert( strlen(zOut)<TEST_CKSUM_BYTES );
+  return cksum.nRow;
+}
+
+int testCountDatabase(TestDb *pDb){
+  Cksum cksum;
+  memset(&cksum, 0, sizeof(Cksum));
+  tdb_scan(pDb, (void *)&cksum, 0, 0, 0, 0, 0, scanCountDb);
+  return cksum.nRow;
+}
+
+/*
+** This function is a no-op if *pRc is not 0 when it is called.
+**
+** Otherwise, the two nul-terminated strings z1 and z1 are compared. If
+** they are the same, the function returns without doing anything. Otherwise,
+** an error message is printed, *pRc is set to 1 and the test_failed()
+** function called.
+*/
+void testCompareStr(const char *z1, const char *z2, int *pRc){
+  if( *pRc==0 ){
+    if( strcmp(z1, z2) ){
+      testPrintError("testCompareStr: \"%s\" != \"%s\"\n", z1, z2);
+      *pRc = 1;
+      test_failed();
+    }
+  }
+}
+
+/*
+** This function is a no-op if *pRc is not 0 when it is called.
+**
+** Otherwise, the two integers i1 and i2 are compared. If they are equal,
+** the function returns without doing anything. Otherwise, an error message 
+** is printed, *pRc is set to 1 and the test_failed() function called.
+*/
+void testCompareInt(int i1, int i2, int *pRc){
+  if( *pRc==0 && i1!=i2 ){
+    testPrintError("testCompareInt: %d != %d\n", i1, i2);
+    *pRc = 1;
+    test_failed();
+  }
+}
+
+void testCaseStart(int *pRc, char *zFmt, ...){
+  va_list ap;
+  va_start(ap, zFmt);
+  vprintf(zFmt, ap);
+  printf(" ...");
+  va_end(ap);
+  *pRc = 0;
+  fflush(stdout);
+}
+
+/*
+** This function is a no-op if *pRc is non-zero when it is called. Zero
+** is returned in this case.
+**
+** Otherwise, the zFmt (a printf style format string) and following arguments 
+** are used to create a test case name. If zPattern is NULL or a glob pattern
+** that matches the test case name, 1 is returned and the test case started.
+** Otherwise, zero is returned and the test case does not start.
+*/
+int testCaseBegin(int *pRc, const char *zPattern, const char *zFmt, ...){
+  int res = 0;
+  if( *pRc==0 ){
+    char *zTest;
+    va_list ap;
+
+    va_start(ap, zFmt);
+    zTest = testMallocVPrintf(zFmt, ap);
+    va_end(ap);
+    if( zPattern==0 || testGlobMatch(zPattern, zTest) ){
+      printf("%-50s ...", zTest);
+      res = 1;
+    }
+    testFree(zTest);
+    fflush(stdout);
+  }
+
+  return res;
+}
+
+void testCaseFinish(int rc){
+  if( rc==0 ){
+    printf("Ok\n");
+  }else{
+    printf("FAILED\n");
+  }
+  fflush(stdout);
+}
+
+void testCaseSkip(){
+  printf("Skipped\n");
+}
+
+void testSetupSavedLsmdb(
+  const char *zCfg,
+  const char *zFile,
+  Datasource *pData,
+  int nRow,
+  int *pRc
+){
+  if( *pRc==0 ){
+    int rc;
+    TestDb *pDb;
+    rc = tdb_lsm_open(zCfg, zFile, 1, &pDb);
+    if( rc==0 ){
+      testWriteDatasourceRange(pDb, pData, 0, nRow, &rc);
+      testClose(&pDb);
+      if( rc==0 ) testSaveDb(zFile, "log");
+    }
+    *pRc = rc;
+  }
+}
+
+/*
+** This function is a no-op if *pRc is non-zero when it is called.
+**
+** Open the LSM database identified by zFile and compute its checksum
+** (a string, as returned by testCksumDatabase()). If the checksum is
+** identical to zExpect1 or, if it is not NULL, zExpect2, the test passes.
+** Otherwise, print an error message and set *pRc to 1.
+*/
+static void testCompareCksumLsmdb(
+  const char *zFile,              /* Path to LSM database */
+  int bCompress,                  /* True if db is compressed */
+  const char *zExpect1,           /* Expected checksum 1 */
+  const char *zExpect2,           /* Expected checksum 2 (or NULL) */
+  int *pRc                        /* IN/OUT: Test case error code */
+){
+  if( *pRc==0 ){
+    char zCksum[TEST_CKSUM_BYTES];
+    TestDb *pDb;
+
+    *pRc = tdb_lsm_open((bCompress?"compression=1 mmap=0":""), zFile, 0, &pDb);
+    testCksumDatabase(pDb, zCksum);
+    testClose(&pDb);
+
+    if( *pRc==0 ){
+      int r1 = 0;
+      int r2 = -1;
+
+      r1 = strcmp(zCksum, zExpect1);
+      if( zExpect2 ) r2 = strcmp(zCksum, zExpect2);
+      if( r1 && r2 ){
+        if( zExpect2 ){
+          testPrintError("testCompareCksumLsmdb: \"%s\" != (\"%s\" OR \"%s\")",
+              zCksum, zExpect1, zExpect2
+          );
+        }else{
+          testPrintError("testCompareCksumLsmdb: \"%s\" != \"%s\"",
+              zCksum, zExpect1
+          );
+        }
+        *pRc = 1;
+        test_failed();
+      }
+    }
+  }
+}
+
+#if 0 /* not used */
+static void testCompareCksumBtdb(
+  const char *zFile,              /* Path to LSM database */
+  const char *zExpect1,           /* Expected checksum 1 */
+  const char *zExpect2,           /* Expected checksum 2 (or NULL) */
+  int *pRc                        /* IN/OUT: Test case error code */
+){
+  if( *pRc==0 ){
+    char zCksum[TEST_CKSUM_BYTES];
+    TestDb *pDb;
+
+    *pRc = tdb_open("bt", zFile, 0, &pDb);
+    testCksumDatabase(pDb, zCksum);
+    testClose(&pDb);
+
+    if( *pRc==0 ){
+      int r1 = 0;
+      int r2 = -1;
+
+      r1 = strcmp(zCksum, zExpect1);
+      if( zExpect2 ) r2 = strcmp(zCksum, zExpect2);
+      if( r1 && r2 ){
+        if( zExpect2 ){
+          testPrintError("testCompareCksumLsmdb: \"%s\" != (\"%s\" OR \"%s\")",
+              zCksum, zExpect1, zExpect2
+          );
+        }else{
+          testPrintError("testCompareCksumLsmdb: \"%s\" != \"%s\"",
+              zCksum, zExpect1
+          );
+        }
+        *pRc = 1;
+        test_failed();
+      }
+    }
+  }
+}
+#endif /* not used */
+
+/* Above this point are reusable test routines. Not clear that they
+** should really be in this file.
+*************************************************************************/
+
+/*
+** This test verifies that if a system crash occurs while doing merge work
+** on the db, no data is lost.
+*/
+static void crash_test1(int bCompress, int *pRc){
+  const char *DBNAME = "testdb.lsm";
+  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 200, 200};
+
+  const int nRow = 5000;          /* Database size */
+  const int nIter = 200;          /* Number of test iterations */
+  const int nWork = 20;           /* Maximum lsm_work() calls per iteration */
+  const int nPage = 15;           /* Pages per lsm_work call */
+
+  int i;
+  int iDot = 0;
+  Datasource *pData;
+  CksumDb *pCksumDb;
+  TestDb *pDb;
+  char *zCfg;
+
+  const char *azConfig[2] = {
+    "page_size=1024 block_size=65536 autoflush=16384 safety=2 mmap=0", 
+    "page_size=1024 block_size=65536 autoflush=16384 safety=2 "
+    " compression=1 mmap=0"
+  };
+  assert( bCompress==0 || bCompress==1 );
+
+  /* Allocate datasource. And calculate the expected checksums. */
+  pData = testDatasourceNew(&defn);
+  pCksumDb = testCksumArrayNew(pData, nRow, nRow, 1);
+
+  /* Setup and save the initial database. */
+
+  zCfg = testMallocPrintf("%s automerge=7", azConfig[bCompress]);
+  testSetupSavedLsmdb(zCfg, DBNAME, pData, 5000, pRc);
+  testFree(zCfg);
+
+  for(i=0; i<nIter && *pRc==0; i++){
+    int iWork;
+    int testrc = 0;
+
+    testCaseProgress(i, nIter, testCaseNDot(), &iDot);
+
+    /* Restore and open the database. */
+    testRestoreDb(DBNAME, "log");
+    testrc = tdb_lsm_open(azConfig[bCompress], DBNAME, 0, &pDb);
+    assert( testrc==0 );
+
+    /* Call lsm_work() on the db */
+    tdb_lsm_prepare_sync_crash(pDb, 1 + (i%(nWork*2)));
+    for(iWork=0; testrc==0 && iWork<nWork; iWork++){
+      int nWrite = 0;
+      lsm_db *db = tdb_lsm(pDb);
+      testrc = lsm_work(db, 0, nPage, &nWrite);
+      /* assert( testrc!=0 || nWrite>0 ); */
+      if( testrc==0 ) testrc = lsm_checkpoint(db, 0);
+    }
+    tdb_close(pDb);
+
+    /* Check that the database content is still correct */
+    testCompareCksumLsmdb(DBNAME, 
+        bCompress, testCksumArrayGet(pCksumDb, nRow), 0, pRc);
+  }
+
+  testCksumArrayFree(pCksumDb);
+  testDatasourceFree(pData);
+}
+
+/*
+** This test verifies that if a system crash occurs while committing a
+** transaction to the log file, no earlier transactions are lost or damaged.
+*/
+static void crash_test2(int bCompress, int *pRc){
+  const char *DBNAME = "testdb.lsm";
+  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 1000, 1000};
+
+  const int nIter = 200;
+  const int nInsert = 20;
+
+  int i;
+  int iDot = 0;
+  Datasource *pData;
+  CksumDb *pCksumDb;
+  TestDb *pDb;
+
+  /* Allocate datasource. And calculate the expected checksums. */
+  pData = testDatasourceNew(&defn);
+  pCksumDb = testCksumArrayNew(pData, 100, 100+nInsert, 1);
+
+  /* Setup and save the initial database. */
+  testSetupSavedLsmdb("", DBNAME, pData, 100, pRc);
+
+  for(i=0; i<nIter && *pRc==0; i++){
+    int iIns;
+    int testrc = 0;
+
+    testCaseProgress(i, nIter, testCaseNDot(), &iDot);
+
+    /* Restore and open the database. */
+    testRestoreDb(DBNAME, "log");
+    testrc = tdb_lsm_open("safety=2", DBNAME, 0, &pDb);
+    assert( testrc==0 );
+
+    /* Insert nInsert records into the database. Crash midway through. */
+    tdb_lsm_prepare_sync_crash(pDb, 1 + (i%(nInsert+2)));
+    for(iIns=0; iIns<nInsert; iIns++){
+      void *pKey; int nKey;
+      void *pVal; int nVal;
+
+      testDatasourceEntry(pData, 100+iIns, &pKey, &nKey, &pVal, &nVal);
+      testrc = tdb_write(pDb, pKey, nKey, pVal, nVal);
+      if( testrc ) break;
+    }
+    tdb_close(pDb);
+
+    /* Check that no data was lost when the system crashed. */
+    testCompareCksumLsmdb(DBNAME, bCompress,
+      testCksumArrayGet(pCksumDb, 100 + iIns),
+      testCksumArrayGet(pCksumDb, 100 + iIns + 1),
+      pRc
+    );
+  }
+
+  testDatasourceFree(pData);
+  testCksumArrayFree(pCksumDb);
+}
+
+
+/*
+** This test verifies that if a system crash occurs when checkpointing
+** the database, data is not lost (assuming that any writes not synced
+** to the db have been synced into the log file).
+*/
+static void crash_test3(int bCompress, int *pRc){
+  const char *DBNAME = "testdb.lsm";
+  const int nIter = 100;
+  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 1000, 1000};
+
+  int i;
+  int iDot = 0;
+  Datasource *pData;
+  CksumDb *pCksumDb;
+  TestDb *pDb;
+
+  /* Allocate datasource. And calculate the expected checksums. */
+  pData = testDatasourceNew(&defn);
+  pCksumDb = testCksumArrayNew(pData, 110, 150, 10);
+
+  /* Setup and save the initial database. */
+  testSetupSavedLsmdb("", DBNAME, pData, 100, pRc);
+
+  for(i=0; i<nIter && *pRc==0; i++){
+    int iOpen;
+    testCaseProgress(i, nIter, testCaseNDot(), &iDot);
+    testRestoreDb(DBNAME, "log");
+
+    for(iOpen=0; iOpen<5; iOpen++){
+      /* Open the database. Insert 10 more records. */
+      pDb = testOpen("lsm", 0, pRc);
+      testWriteDatasourceRange(pDb, pData, 100+iOpen*10, 10, pRc);
+
+      /* Schedule a crash simulation then close the db. */
+      tdb_lsm_prepare_sync_crash(pDb, 1 + (i%2));
+      tdb_close(pDb);
+
+      /* Open the database and check that the crash did not cause any
+      ** data loss.  */
+      testCompareCksumLsmdb(DBNAME, bCompress,
+        testCksumArrayGet(pCksumDb, 110 + iOpen*10), 0,
+        pRc
+      );
+    }
+  }
+
+  testDatasourceFree(pData);
+  testCksumArrayFree(pCksumDb);
+}
+
+void do_crash_test(const char *zPattern, int *pRc){
+  struct Test {
+    const char *zTest;
+    void (*x)(int, int *);
+    int bCompress;
+  } aTest [] = {
+    { "crash.lsm.1",     crash_test1, 0 },
+#ifdef HAVE_ZLIB
+    { "crash.lsm_zip.1", crash_test1, 1 },
+#endif
+    { "crash.lsm.2",     crash_test2, 0 },
+    { "crash.lsm.3",     crash_test3, 0 },
+  };
+  int i;
+
+  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+    struct Test *p = &aTest[i];
+    if( testCaseBegin(pRc, zPattern, "%s", p->zTest) ){
+      p->x(p->bCompress, pRc);
+      testCaseFinish(*pRc);
+    }
+  }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest3.c b/ext/lsm1/lsm-test/lsmtest3.c
new file mode 100644
index 0000000..760dec3
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest3.c
@@ -0,0 +1,238 @@
+
+
+/*
+** This file contains tests related to the explicit rollback of database
+** transactions and sub-transactions.
+*/
+
+
+/*
+** Repeat 2000 times (until the db contains 100,000 entries):
+**
+**   1. Open a transaction and insert 500 rows, opening a nested 
+**      sub-transaction each 100 rows.
+**
+**   2. Roll back to each sub-transaction savepoint. Check the database
+**      checksum looks Ok.
+**
+**   3. Every second iteration, roll back the main transaction. Check the
+**      db checksum is correct. Every other iteration, commit the main
+**      transaction (increasing the size of the db by 100 rows).
+*/
+
+
+#include "lsmtest.h"
+
+struct CksumDb {
+  int nFirst;
+  int nLast;
+  int nStep;
+  char **azCksum;
+};
+
+CksumDb *testCksumArrayNew(
+  Datasource *pData, 
+  int nFirst, 
+  int nLast, 
+  int nStep
+){
+  TestDb *pDb;
+  CksumDb *pRet;
+  int i;
+  int nEntry;
+  int rc = 0;
+
+  assert( nLast>=nFirst && ((nLast-nFirst)%nStep)==0 );
+ 
+  pRet = malloc(sizeof(CksumDb));
+  memset(pRet, 0, sizeof(CksumDb));
+  pRet->nFirst = nFirst;
+  pRet->nLast = nLast;
+  pRet->nStep = nStep;
+  nEntry = 1 + ((nLast - nFirst) / nStep);
+
+  /* Allocate space so that azCksum is an array of nEntry pointers to
+  ** buffers each TEST_CKSUM_BYTES in size.  */
+  pRet->azCksum = (char **)malloc(nEntry * (sizeof(char *) + TEST_CKSUM_BYTES));
+  for(i=0; i<nEntry; i++){
+    char *pStart = (char *)(&pRet->azCksum[nEntry]);
+    pRet->azCksum[i] = &pStart[i * TEST_CKSUM_BYTES];
+  }
+
+  tdb_open("lsm", "tempdb.lsm", 1, &pDb);
+  testWriteDatasourceRange(pDb, pData, 0, nFirst, &rc);
+  for(i=0; i<nEntry; i++){
+    testCksumDatabase(pDb, pRet->azCksum[i]);
+    if( i==nEntry ) break;
+    testWriteDatasourceRange(pDb, pData, nFirst+i*nStep, nStep, &rc);
+  }
+
+  tdb_close(pDb);
+
+  return pRet;
+}
+
+char *testCksumArrayGet(CksumDb *p, int nRow){
+  int i;
+  assert( nRow>=p->nFirst );
+  assert( nRow<=p->nLast );
+  assert( ((nRow-p->nFirst) % p->nStep)==0 );
+
+  i = (nRow - p->nFirst) / p->nStep;
+  return p->azCksum[i];
+}
+
+void testCksumArrayFree(CksumDb *p){
+  free(p->azCksum);
+  memset(p, 0x55, sizeof(*p));
+  free(p);
+}
+
+/* End of CksumDb code.
+**************************************************************************/
+
+/*
+** Test utility function. Write key-value pair $i from datasource pData 
+** into database pDb.
+*/
+void testWriteDatasource(TestDb *pDb, Datasource *pData, int i, int *pRc){
+  void *pKey; int nKey;
+  void *pVal; int nVal;
+  testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
+  testWrite(pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+/*
+** Test utility function. Delete datasource pData key $i from database pDb.
+*/
+void testDeleteDatasource(TestDb *pDb, Datasource *pData, int i, int *pRc){
+  void *pKey; int nKey;
+  testDatasourceEntry(pData, i, &pKey, &nKey, 0, 0);
+  testDelete(pDb, pKey, nKey, pRc);
+}
+
+/*
+** This function inserts nWrite key/value pairs into database pDb - the
+** nWrite key value pairs starting at iFirst from data source pData.
+*/
+void testWriteDatasourceRange(
+  TestDb *pDb,                    /* Database to write to */
+  Datasource *pData,              /* Data source to read values from */
+  int iFirst,                     /* Index of first key/value pair */
+  int nWrite,                     /* Number of key/value pairs to write */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int i;
+  for(i=0; i<nWrite; i++){
+    testWriteDatasource(pDb, pData, iFirst+i, pRc);
+  }
+}
+
+void testDeleteDatasourceRange(
+  TestDb *pDb,                    /* Database to write to */
+  Datasource *pData,              /* Data source to read keys from */
+  int iFirst,                     /* Index of first key */
+  int nWrite,                     /* Number of keys to delete */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int i;
+  for(i=0; i<nWrite; i++){
+    testDeleteDatasource(pDb, pData, iFirst+i, pRc);
+  }
+}
+
+static char *getName(const char *zSystem){ 
+  char *zRet; 
+  zRet = testMallocPrintf("rollback.%s", zSystem);
+  return zRet;
+}
+
+static int rollback_test_1(
+  const char *zSystem,
+  Datasource *pData
+){
+  const int nRepeat = 100;
+
+  TestDb *pDb;
+  int rc;
+  int i;
+  CksumDb *pCksum;
+  char *zName;
+
+  zName = getName(zSystem);
+  testCaseStart(&rc, zName);
+  testFree(zName);
+
+  pCksum = testCksumArrayNew(pData, 0, nRepeat*100, 100);
+  pDb = 0;
+  rc = tdb_open(zSystem, 0, 1, &pDb);
+  if( pDb && tdb_transaction_support(pDb)==0 ){
+    testCaseSkip();
+    goto skip_rollback_test;
+  }
+
+  for(i=0; i<nRepeat && rc==0; i++){
+    char zCksum[TEST_CKSUM_BYTES];
+    int nCurrent = (((i+1)/2) * 100);
+    int nDbRow;
+    int iTrans;
+
+    /* Check that the database is the expected size. */
+    nDbRow = testCountDatabase(pDb);
+    testCompareInt(nCurrent, nDbRow, &rc);
+
+    for(iTrans=2; iTrans<=6 && rc==0; iTrans++){
+      tdb_begin(pDb, iTrans);
+      testWriteDatasourceRange(pDb, pData, nCurrent, 100, &rc);
+      nCurrent += 100;
+    }
+
+    testCksumDatabase(pDb, zCksum);
+    testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
+
+    for(iTrans=6; iTrans>2 && rc==0; iTrans--){
+      tdb_rollback(pDb, iTrans);
+      nCurrent -= 100;
+      testCksumDatabase(pDb, zCksum);
+      testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
+    }
+
+    if( i%2 ){
+      tdb_rollback(pDb, 0);
+      nCurrent -= 100;
+      testCksumDatabase(pDb, zCksum);
+      testCompareStr(zCksum, testCksumArrayGet(pCksum, nCurrent), &rc);
+    }else{
+      tdb_commit(pDb, 0);
+    }
+  }
+  testCaseFinish(rc);
+
+ skip_rollback_test:
+  tdb_close(pDb);
+  testCksumArrayFree(pCksum);
+  return rc;
+}
+
+void test_rollback(
+  const char *zSystem, 
+  const char *zPattern, 
+  int *pRc
+){
+  if( *pRc==0 ){
+    int bRun = 1;
+
+    if( zPattern ){
+      char *zName = getName(zSystem);
+      bRun = testGlobMatch(zPattern, zName);
+      testFree(zName);
+    }
+
+    if( bRun ){
+      DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 50, 100 };
+      Datasource *pData = testDatasourceNew(&defn);
+      *pRc = rollback_test_1(zSystem, pData);
+      testDatasourceFree(pData);
+    }
+  }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest4.c b/ext/lsm1/lsm-test/lsmtest4.c
new file mode 100644
index 0000000..a47241d
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest4.c
@@ -0,0 +1,127 @@
+
+/*
+** This file contains test cases involving multiple database clients.
+*/
+
+#include "lsmtest.h"
+
+/*
+** The following code implements test cases "mc1.*".
+**
+** This test case uses one writer and $nReader readers. All connections
+** are driven by a single thread. All connections are opened at the start
+** of the test and remain open until the test is finished.
+**
+** The test consists of $nStep steps. Each step the following is performed:
+**
+**   1. The writer inserts $nWriteStep records into the db.
+**
+**   2. The writer checks that the contents of the db are as expected.
+**
+**   3. Each reader that currently has an open read transaction also checks
+**      that the contents of the db are as expected (according to the snapshot
+**      the read transaction is reading - see below).
+**
+** After step 1, reader 1 opens a read transaction. After step 2, reader
+** 2 opens a read transaction, and so on. At step ($nReader+1), reader 1
+** closes the current read transaction and opens a new one. And so on.
+** The result is that at step N (for N > $nReader), there exists a reader
+** with an open read transaction reading the snapshot committed following
+** steps (N-$nReader-1) to N. 
+*/
+typedef struct Mctest Mctest;
+struct Mctest {
+  DatasourceDefn defn;            /* Datasource to use */
+  int nStep;                      /* Total number of steps in test */
+  int nWriteStep;                 /* Number of rows to insert each step */
+  int nReader;                    /* Number of read connections */
+};
+static void do_mc_test(
+  const char *zSystem,            /* Database system to test */
+  Mctest *pTest,
+  int *pRc                        /* IN/OUT: return code */
+){
+  const int nDomain = pTest->nStep * pTest->nWriteStep;
+  Datasource *pData;              /* Source of data */
+  TestDb *pDb;                    /* First database connection (writer) */
+  int iReader;                    /* Used to iterate through aReader */
+  int iStep;                      /* Current step in test */
+  int iDot = 0;                   /* Current step in test */
+
+  /* Array of reader connections */
+  struct Reader {
+    TestDb *pDb;                  /* Connection handle */
+    int iLast;                    /* Current snapshot contains keys 0..iLast */
+  } *aReader;
+
+  /* Create a data source */
+  pData = testDatasourceNew(&pTest->defn);
+
+  /* Open the writer connection */
+  pDb = testOpen(zSystem, 1, pRc);
+
+  /* Allocate aReader */
+  aReader = (struct Reader *)testMalloc(sizeof(aReader[0]) * pTest->nReader);
+  for(iReader=0; iReader<pTest->nReader; iReader++){
+    aReader[iReader].pDb = testOpen(zSystem, 0, pRc);
+  }
+
+  for(iStep=0; iStep<pTest->nStep; iStep++){
+    int iLast;
+    int iBegin;                   /* Start read trans using aReader[iBegin] */
+
+    /* Insert nWriteStep more records into the database */
+    int iFirst = iStep*pTest->nWriteStep;
+    testWriteDatasourceRange(pDb, pData, iFirst, pTest->nWriteStep, pRc);
+
+    /* Check that the db is Ok according to the writer */
+    iLast = (iStep+1) * pTest->nWriteStep - 1;
+    testDbContents(pDb, pData, nDomain, 0, iLast, iLast, 1, pRc);
+
+    /* Have reader (iStep % nReader) open a read transaction here. */
+    iBegin = (iStep % pTest->nReader);
+    if( iBegin<iStep ) tdb_commit(aReader[iBegin].pDb, 0);
+    tdb_begin(aReader[iBegin].pDb, 1);
+    aReader[iBegin].iLast = iLast;
+
+    /* Check that the db is Ok for each open reader */
+    for(iReader=0; iReader<pTest->nReader && aReader[iReader].iLast; iReader++){
+      iLast = aReader[iReader].iLast;
+      testDbContents(
+          aReader[iReader].pDb, pData, nDomain, 0, iLast, iLast, 1, pRc
+      );
+    }
+
+    /* Report progress */
+    testCaseProgress(iStep, pTest->nStep, testCaseNDot(), &iDot);
+  }
+
+  /* Close all readers */
+  for(iReader=0; iReader<pTest->nReader; iReader++){
+    testClose(&aReader[iReader].pDb);
+  }
+  testFree(aReader);
+
+  /* Close the writer-connection and free the datasource */
+  testClose(&pDb);
+  testDatasourceFree(pData);
+}
+
+
+void test_mc(
+  const char *zSystem,            /* Database system name */
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int i;
+  Mctest aTest[] = {
+    { { TEST_DATASOURCE_RANDOM, 10,10, 100,100 }, 100, 10, 5 },
+  };
+
+  for(i=0; i<ArraySize(aTest); i++){
+    if( testCaseBegin(pRc, zPattern, "mc1.%s.%d", zSystem, i) ){
+      do_mc_test(zSystem, &aTest[i], pRc);
+      testCaseFinish(*pRc);
+    }
+  }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest5.c b/ext/lsm1/lsm-test/lsmtest5.c
new file mode 100644
index 0000000..f36184e
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest5.c
@@ -0,0 +1,633 @@
+
+/*
+** This file is broken into three semi-autonomous parts:
+**
+**   1. The database functions.
+**   2. The thread wrappers.
+**   3. The implementation of the mt1.* tests.
+*/
+
+/*************************************************************************
+** DATABASE CONTENTS:
+**
+**   The database contains up to N key/value pairs, where N is some large 
+**   number (say 10,000,000). Keys are integer values between 0 and (N-1).
+**   The value associated with each key is a pseudo-random blob of data.
+**
+**   Key/value pair keys are encoded as the two bytes "k." followed by a 
+**   10-digit decimal number. i.e. key 45 -> "k.0000000045".
+**
+**   As well as the key/value pairs, the database also contains checksum 
+**   entries. The checksums form a hierarchy - for every F key/value
+**   entries there is one level 1 checksum. And for each F level 1 checksums
+**   there is one level 2 checksum. And so on.
+**
+**   Checksum keys are encoded as the two byte "c." followed by the 
+**   checksum level, followed by a 10 digit decimal number containing
+**   the value of the first key that contributes to the checksum value.
+**   For example, assuming F==10, the level 1 checksum that spans keys
+**   10 to 19 is "c.1.0000000010".
+**
+**   Clients may perform one of two operations on the database: a read
+**   or a write.
+** 
+** READ OPERATIONS:
+**
+**   A read operation scans a range of F key/value pairs. It computes
+**   the expected checksum and then compares the computed value to the
+**   actual value stored in the level 1 checksum entry. It then scans 
+**   the group of F level 1 checksums, and compares the computed checksum 
+**   to the associated level 2 checksum value, and so on until the 
+**   highest level checksum value has been verified.
+**
+**   If a checksum ever fails to match the expected value, the test 
+**   has failed.
+**
+** WRITE OPERATIONS:
+**
+**   A write operation involves writing (possibly clobbering) a single
+**   key/value pair. The associated level 1 checksum is then recalculated
+**   updated. Then the level 2 checksum, and so on until the highest
+**   level checksum has been modified.
+**
+**   All updates occur inside a single transaction.
+**
+** INTERFACE:
+**
+**   The interface used by test cases to read and write the db consists
+**   of type DbParameters and the following functions:
+**
+**       dbReadOperation()
+**       dbWriteOperation()
+*/
+
+#include "lsmtest.h"
+
+typedef struct DbParameters DbParameters;
+struct DbParameters {
+  int nFanout;                    /* Checksum fanout (F) */
+  int nKey;                       /* Size of key space (N) */
+};
+
+#define DB_KEY_BYTES          (2+5+10+1)
+
+/*
+** Argument aBuf[] must point to a buffer at least DB_KEY_BYTES in size.
+** This function populates the buffer with a nul-terminated key string 
+** corresponding to key iKey.
+*/
+static void dbFormatKey(
+  DbParameters *pParam,
+  int iLevel,
+  int iKey,                       /* Key value */
+  char *aBuf                      /* Write key string here */
+){
+  if( iLevel==0 ){
+    snprintf(aBuf, DB_KEY_BYTES, "k.%.10d", iKey);
+  }else{
+    int f = 1;
+    int i;
+    for(i=0; i<iLevel; i++) f = f * pParam->nFanout;
+    snprintf(aBuf, DB_KEY_BYTES, "c.%d.%.10d", iLevel, f*(iKey/f));
+  }
+}
+
+/*
+** Argument aBuf[] must point to a buffer at least DB_KEY_BYTES in size.
+** This function populates the buffer with the string representation of
+** checksum value iVal.
+*/
+static void dbFormatCksumValue(u32 iVal, char *aBuf){
+  snprintf(aBuf, DB_KEY_BYTES, "%.10u", iVal);
+}
+
+/*
+** Return the highest level of checksum in the database described
+** by *pParam.
+*/
+static int dbMaxLevel(DbParameters *pParam){
+  int iMax;
+  int n = 1;
+  for(iMax=0; n<pParam->nKey; iMax++){
+    n = n * pParam->nFanout;
+  }
+  return iMax;
+}
+
+static void dbCksum(
+  void *pCtx,                     /* IN/OUT: Pointer to u32 containing cksum */
+  void *pKey, int nKey,           /* Database key. Unused. */
+  void *pVal, int nVal            /* Database value. Checksum this. */
+){
+  u8 *aVal = (u8 *)pVal;
+  u32 *pCksum = (u32 *)pCtx;
+  u32 cksum = *pCksum;
+  int i;
+
+  unused_parameter(pKey);
+  unused_parameter(nKey);
+
+  for(i=0; i<nVal; i++){
+    cksum += (cksum<<3) + (int)aVal[i];
+  }
+
+  *pCksum = cksum;
+}
+
+/*
+** Compute the value of the checksum stored on level iLevel that contains
+** data from key iKey by scanning the pParam->nFanout entries at level 
+** iLevel-1.
+*/
+static u32 dbComputeCksum(
+  DbParameters *pParam,           /* Database parameters */
+  TestDb *pDb,                    /* Database connection handle */
+  int iLevel,                     /* Level of checksum to compute */
+  int iKey,                       /* Compute checksum for this key */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  u32 cksum = 0;
+  if( *pRc==0 ){
+    int nFirst;
+    int nLast;
+    int iFirst = 0;
+    int iLast = 0;
+    int i;
+    int f = 1;
+    char zFirst[DB_KEY_BYTES];
+    char zLast[DB_KEY_BYTES];
+
+    assert( iLevel>=1 );
+    for(i=0; i<iLevel; i++) f = f * pParam->nFanout;
+
+    iFirst = f*(iKey/f);
+    iLast = iFirst + f - 1;
+    dbFormatKey(pParam, iLevel-1, iFirst, zFirst);
+    dbFormatKey(pParam, iLevel-1, iLast, zLast);
+    nFirst = strlen(zFirst);
+    nLast = strlen(zLast);
+
+    *pRc = tdb_scan(pDb, (u32*)&cksum, 0, zFirst, nFirst, zLast, nLast,dbCksum);
+  }
+
+  return cksum;
+}
+
+static void dbReadOperation(
+  DbParameters *pParam,           /* Database parameters */
+  TestDb *pDb,                    /* Database connection handle */
+  void (*xDelay)(void *),
+  void *pDelayCtx,
+  int iKey,                       /* Key to read */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  const int iMax = dbMaxLevel(pParam);
+  int i;
+
+  if( tdb_transaction_support(pDb) ) testBegin(pDb, 1, pRc);
+  for(i=1; *pRc==0 && i<=iMax; i++){
+    char zCksum[DB_KEY_BYTES];
+    char zKey[DB_KEY_BYTES];
+    u32 iCksum = 0;
+
+    iCksum = dbComputeCksum(pParam, pDb, i, iKey, pRc);
+    if( iCksum ){
+      if( xDelay && i==1 ) xDelay(pDelayCtx);
+      dbFormatCksumValue(iCksum, zCksum);
+      dbFormatKey(pParam, i, iKey, zKey);
+      testFetchStr(pDb, zKey, zCksum, pRc);
+    }
+  }
+  if( tdb_transaction_support(pDb) ) testCommit(pDb, 0, pRc);
+}
+
+static int dbWriteOperation(
+  DbParameters *pParam,           /* Database parameters */
+  TestDb *pDb,                    /* Database connection handle */
+  int iKey,                       /* Key to write to */
+  const char *zValue,             /* Nul-terminated value to write */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  const int iMax = dbMaxLevel(pParam);
+  char zKey[DB_KEY_BYTES];
+  int i;
+  int rc;
+
+  assert( iKey>=0 && iKey<pParam->nKey );
+  dbFormatKey(pParam, 0, iKey, zKey);
+
+  /* Open a write transaction. This may fail - SQLITE4_BUSY */
+  if( *pRc==0 && tdb_transaction_support(pDb) ){
+    rc = tdb_begin(pDb, 2);
+    if( rc==5 ) return 0;
+    *pRc = rc;
+  }
+
+  testWriteStr(pDb, zKey, zValue, pRc);
+  for(i=1; i<=iMax; i++){
+    char zCksum[DB_KEY_BYTES];
+    u32 iCksum = 0;
+
+    iCksum = dbComputeCksum(pParam, pDb, i, iKey, pRc);
+    dbFormatCksumValue(iCksum, zCksum);
+    dbFormatKey(pParam, i, iKey, zKey);
+    testWriteStr(pDb, zKey, zCksum, pRc);
+  }
+  if( tdb_transaction_support(pDb) ) testCommit(pDb, 0, pRc);
+  return 1;
+}
+
+/*************************************************************************
+** The following block contains testXXX() functions that implement a
+** wrapper around the systems native multi-thread support. There are no
+** synchronization primitives - just functions to launch and join 
+** threads. Wrapper functions are:
+**
+**    testThreadSupport()
+**
+**    testThreadInit()
+**    testThreadShutdown()
+**    testThreadLaunch()
+**    testThreadWait()
+**
+**    testThreadSetHalt()
+**    testThreadGetHalt()
+**    testThreadSetResult()
+**    testThreadGetResult()
+**
+**    testThreadEnterMutex()
+**    testThreadLeaveMutex()
+*/
+typedef struct ThreadSet ThreadSet;
+#ifdef LSM_MUTEX_PTHREADS
+
+#include <pthread.h>
+#include <unistd.h>
+
+typedef struct Thread Thread;
+struct Thread {
+  int rc;
+  char *zMsg;
+  pthread_t id;
+  void (*xMain)(ThreadSet *, int, void *);
+  void *pCtx;
+  ThreadSet *pThreadSet;
+};
+
+struct ThreadSet {
+  int bHalt;                      /* Halt flag */
+  int nThread;                    /* Number of threads */
+  Thread *aThread;                /* Array of Thread structures */
+  pthread_mutex_t mutex;          /* Mutex used for cheating */
+};
+
+/*
+** Return true if this build supports threads, or false otherwise. If
+** this function returns false, no other testThreadXXX() functions should
+** be called.
+*/
+static int testThreadSupport(){ return 1; }
+
+/*
+** Allocate and return a thread-set handle with enough space allocated
+** to handle up to nMax threads. Each call to this function should be
+** matched by a call to testThreadShutdown() to delete the object.
+*/
+static ThreadSet *testThreadInit(int nMax){
+  int nByte;                      /* Total space to allocate */
+  ThreadSet *p;                   /* Return value */
+
+  nByte = sizeof(ThreadSet) + sizeof(struct Thread) * nMax;
+  p = (ThreadSet *)testMalloc(nByte);
+  p->nThread = nMax;
+  p->aThread = (Thread *)&p[1];
+  pthread_mutex_init(&p->mutex, 0);
+
+  return p;
+}
+
+/*
+** Delete a thread-set object and release all resources held by it.
+*/
+static void testThreadShutdown(ThreadSet *p){
+  int i;
+  for(i=0; i<p->nThread; i++){
+    testFree(p->aThread[i].zMsg);
+  }
+  pthread_mutex_destroy(&p->mutex);
+  testFree(p);
+}
+
+static void *ttMain(void *pArg){
+  Thread *pThread = (Thread *)pArg;
+  int iThread;
+  iThread = (pThread - pThread->pThreadSet->aThread);
+  pThread->xMain(pThread->pThreadSet, iThread, pThread->pCtx);
+  return 0;
+}
+
+/*
+** Launch a new thread.
+*/
+static int testThreadLaunch(
+  ThreadSet *p,
+  int iThread,
+  void (*xMain)(ThreadSet *, int, void *),
+  void *pCtx
+){
+  int rc;
+  Thread *pThread;
+
+  assert( iThread>=0 && iThread<p->nThread );
+
+  pThread = &p->aThread[iThread];
+  assert( pThread->pThreadSet==0 );
+  pThread->xMain = xMain;
+  pThread->pCtx = pCtx;
+  pThread->pThreadSet = p;
+  rc = pthread_create(&pThread->id, 0, ttMain, (void *)pThread);
+
+  return rc;
+}
+
+/*
+** Set the thread-set "halt" flag.
+*/
+static void testThreadSetHalt(ThreadSet *pThreadSet){
+  pThreadSet->bHalt = 1;
+}
+
+/*
+** Return the current value of the thread-set "halt" flag.
+*/
+static int testThreadGetHalt(ThreadSet *pThreadSet){
+  return pThreadSet->bHalt;
+}
+
+static void testThreadSleep(ThreadSet *pThreadSet, int nMs){
+  int nRem = nMs;
+  while( nRem>0 && testThreadGetHalt(pThreadSet)==0 ){
+    usleep(50000);
+    nRem -= 50;
+  }
+}
+
+/*
+** Wait for all threads launched to finish before returning. If nMs
+** is greater than zero, set the "halt" flag to tell all threads
+** to halt after waiting nMs milliseconds.
+*/
+static void testThreadWait(ThreadSet *pThreadSet, int nMs){
+  int i;
+
+  testThreadSleep(pThreadSet, nMs);
+  testThreadSetHalt(pThreadSet);
+  for(i=0; i<pThreadSet->nThread; i++){
+    Thread *pThread = &pThreadSet->aThread[i];
+    if( pThread->xMain ){
+      pthread_join(pThread->id, 0);
+    }
+  }
+}
+
+/*
+** Set the result for thread iThread. 
+*/
+static void testThreadSetResult(
+  ThreadSet *pThreadSet,          /* Thread-set handle */
+  int iThread,                    /* Set result for this thread */
+  int rc,                         /* Result error code */
+  char *zFmt,                     /* Result string format */
+  ...                             /* Result string formatting args... */
+){
+  va_list ap;
+
+  testFree(pThreadSet->aThread[iThread].zMsg);
+  pThreadSet->aThread[iThread].rc = rc;
+  pThreadSet->aThread[iThread].zMsg = 0;
+  if( zFmt ){
+    va_start(ap, zFmt);
+    pThreadSet->aThread[iThread].zMsg = testMallocVPrintf(zFmt, ap);
+    va_end(ap);
+  }
+}
+
+/*
+** Retrieve the result for thread iThread. 
+*/
+static int testThreadGetResult(
+  ThreadSet *pThreadSet,          /* Thread-set handle */
+  int iThread,                    /* Get result for this thread */
+  const char **pzRes              /* OUT: Pointer to result string */
+){
+  if( pzRes ) *pzRes = pThreadSet->aThread[iThread].zMsg;
+  return pThreadSet->aThread[iThread].rc;
+}
+
+/*
+** Enter and leave the test case mutex.
+*/
+#if 0
+static void testThreadEnterMutex(ThreadSet *p){
+  pthread_mutex_lock(&p->mutex);
+}
+static void testThreadLeaveMutex(ThreadSet *p){
+  pthread_mutex_unlock(&p->mutex);
+}
+#endif
+#endif
+
+#if !defined(LSM_MUTEX_PTHREADS)
+static int testThreadSupport(){ return 0; }
+
+#define testThreadInit(a) 0
+#define testThreadShutdown(a)
+#define testThreadLaunch(a,b,c,d) 0
+#define testThreadWait(a,b)
+#define testThreadSetHalt(a)
+#define testThreadGetHalt(a) 0
+#define testThreadGetResult(a,b,c) 0
+#define testThreadSleep(a,b) 0
+
+static void testThreadSetResult(ThreadSet *a, int b, int c, char *d, ...){
+  unused_parameter(a);
+  unused_parameter(b);
+  unused_parameter(c);
+  unused_parameter(d);
+}
+#endif
+/* End of threads wrapper.
+*************************************************************************/
+
+/*************************************************************************
+** Below this point is the third part of this file - the implementation
+** of the mt1.* tests.
+*/
+typedef struct Mt1Test Mt1Test;
+struct Mt1Test {
+  DbParameters param;             /* Description of database to read/write */
+  int nReadwrite;                 /* Number of read/write threads */
+  int nFastReader;                /* Number of fast reader threads */
+  int nSlowReader;                /* Number of slow reader threads */
+  int nMs;                        /* How long to run for */
+  const char *zSystem;            /* Database system to test */
+};
+
+typedef struct Mt1DelayCtx Mt1DelayCtx;
+struct Mt1DelayCtx {
+  ThreadSet *pSet;                /* Threadset to sleep within */
+  int nMs;                        /* Sleep in ms */
+};
+
+static void xMt1Delay(void *pCtx){
+  Mt1DelayCtx *p = (Mt1DelayCtx *)pCtx;
+  testThreadSleep(p->pSet, p->nMs);
+}
+
+#define MT1_THREAD_RDWR 0
+#define MT1_THREAD_SLOW 1
+#define MT1_THREAD_FAST 2
+
+static void xMt1Work(lsm_db *pDb, void *pCtx){
+#if 0
+  char *z = 0;
+  lsm_info(pDb, LSM_INFO_DB_STRUCTURE, &z);
+  printf("%s\n", z);
+  fflush(stdout);
+#endif
+}
+
+/*
+** This is the main() proc for all threads in test case "mt1".
+*/
+static void mt1Main(ThreadSet *pThreadSet, int iThread, void *pCtx){
+  Mt1Test *p = (Mt1Test *)pCtx;   /* Test parameters */
+  Mt1DelayCtx delay;
+  int nRead = 0;                  /* Number of calls to dbReadOperation() */
+  int nWrite = 0;                 /* Number of completed database writes */
+  int rc = 0;                     /* Error code */
+  int iPrng;                      /* Prng argument variable */
+  TestDb *pDb;                    /* Database handle */
+  int eType;
+
+  delay.pSet = pThreadSet;
+  delay.nMs = 0;
+  if( iThread<p->nReadwrite ){
+    eType = MT1_THREAD_RDWR;
+  }else if( iThread<(p->nReadwrite+p->nFastReader) ){
+    eType = MT1_THREAD_FAST;
+  }else{
+    eType = MT1_THREAD_SLOW;
+    delay.nMs = (p->nMs / 20);
+  }
+
+  /* Open a new database connection. Initialize the pseudo-random number
+  ** argument based on the thread number.  */
+  iPrng = testPrngValue(iThread);
+  pDb = testOpen(p->zSystem, 0, &rc);
+
+  if( rc==0 ){
+    tdb_lsm_config_work_hook(pDb, xMt1Work, 0);
+  }
+
+  /* Loop until either an error occurs or some other thread sets the
+  ** halt flag.  */
+  while( rc==0 && testThreadGetHalt(pThreadSet)==0 ){
+    int iKey;
+
+    /* Perform a read operation on an arbitrarily selected key. */
+    iKey = (testPrngValue(iPrng++) % p->param.nKey);
+    dbReadOperation(&p->param, pDb, xMt1Delay, (void *)&delay, iKey, &rc);
+    if( rc ) continue;
+    nRead++;
+
+    /* Attempt to write an arbitrary key value pair (and update the associated
+    ** checksum entries). dbWriteOperation() returns 1 if the write is
+    ** successful, or 0 if it failed with an LSM_BUSY error.  */
+    if( eType==MT1_THREAD_RDWR ){
+      char aValue[50];
+      char aRnd[25];
+
+      iKey = (testPrngValue(iPrng++) % p->param.nKey);
+      testPrngString(iPrng, aRnd, sizeof(aRnd));
+      iPrng += sizeof(aRnd);
+      snprintf(aValue, sizeof(aValue), "%d.%s", iThread, aRnd);
+      nWrite += dbWriteOperation(&p->param, pDb, iKey, aValue, &rc);
+    }
+  }
+  testClose(&pDb);
+
+  /* If an error has occured, set the thread error code and the threadset 
+  ** halt flag to tell the other test threads to halt. Otherwise, set the
+  ** thread error code to 0 and post a message with the number of read
+  ** and write operations completed.  */
+  if( rc ){
+    testThreadSetResult(pThreadSet, iThread, rc, 0);
+    testThreadSetHalt(pThreadSet);
+  }else{
+    testThreadSetResult(pThreadSet, iThread, 0, "r/w: %d/%d", nRead, nWrite);
+  }
+}
+
+static void do_test_mt1(
+  const char *zSystem,            /* Database system name */
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  Mt1Test aTest[] = {
+    /* param, nReadwrite, nFastReader, nSlowReader, nMs, zSystem */
+    { {10, 1000},     4, 0, 0,   10000,   0 },
+    { {10, 1000},     4, 4, 2,   100000,  0 },
+    { {10, 100000},   4, 0, 0,   10000,   0 },
+    { {10, 100000},   4, 4, 2,   100000,  0 },
+  };
+  int i;
+
+  for(i=0; *pRc==0 && i<ArraySize(aTest); i++){
+    Mt1Test *p = &aTest[i];
+    int bRun = testCaseBegin(pRc, zPattern, 
+        "mt1.%s.db=%d,%d.ms=%d.rdwr=%d.fast=%d.slow=%d", 
+        zSystem, p->param.nFanout, p->param.nKey, 
+        p->nMs, p->nReadwrite, p->nFastReader, p->nSlowReader
+    );
+    if( bRun ){
+      TestDb *pDb;
+      ThreadSet *pSet;
+      int iThread;
+      int nThread;
+
+      p->zSystem = zSystem;
+      pDb = testOpen(zSystem, 1, pRc);
+
+      nThread = p->nReadwrite + p->nFastReader + p->nSlowReader;
+      pSet = testThreadInit(nThread);
+      for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
+        testThreadLaunch(pSet, iThread, mt1Main, (void *)p);
+      }
+
+      testThreadWait(pSet, p->nMs);
+      for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
+        *pRc = testThreadGetResult(pSet, iThread, 0);
+      }
+      testCaseFinish(*pRc);
+
+      for(iThread=0; *pRc==0 && iThread<nThread; iThread++){
+        const char *zMsg = 0;
+        *pRc = testThreadGetResult(pSet, iThread, &zMsg);
+        printf("  Info: thread %d (%d): %s\n", iThread, *pRc, zMsg);
+      }
+
+      testThreadShutdown(pSet);
+      testClose(&pDb);
+    }
+  }
+}
+
+void test_mt(
+  const char *zSystem,            /* Database system name */
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  if( testThreadSupport()==0 ) return;
+  do_test_mt1(zSystem, zPattern, pRc);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest6.c b/ext/lsm1/lsm-test/lsmtest6.c
new file mode 100644
index 0000000..a61b738
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest6.c
@@ -0,0 +1,661 @@
+
+#include "lsmtest.h"
+
+typedef struct OomTest OomTest;
+struct OomTest {
+  lsm_env *pEnv;
+  int iNext;                      /* Next value to pass to testMallocOom() */
+  int nFail;                      /* Number of OOM events injected */
+  int bEnable;
+  int rc;                         /* Test case error code */
+};
+
+static void testOomStart(OomTest *p){
+  memset(p, 0, sizeof(OomTest));
+  p->iNext = 1;
+  p->bEnable = 1;
+  p->nFail = 1;
+  p->pEnv = tdb_lsm_env();
+}
+
+static void xOomHook(OomTest *p){
+  p->nFail++;
+}
+
+static int testOomContinue(OomTest *p){
+  if( p->rc!=0 || (p->iNext>1 && p->nFail==0) ){
+    return 0;
+  }
+  p->nFail = 0;
+  testMallocOom(p->pEnv, p->iNext, 0, (void (*)(void*))xOomHook, (void *)p);
+  return 1;
+}
+
+static void testOomEnable(OomTest *p, int bEnable){
+  p->bEnable = bEnable;
+  testMallocOomEnable(p->pEnv, bEnable);
+}
+
+static void testOomNext(OomTest *p){
+  p->iNext++;
+}
+
+static int testOomHit(OomTest *p){
+  return (p->nFail>0);
+}
+
+static int testOomFinish(OomTest *p){
+  return p->rc;
+}
+
+static void testOomAssert(OomTest *p, int bVal){
+  if( bVal==0 ){
+    test_failed();
+    p->rc = 1;
+  }
+}
+
+/*
+** Test that the error code matches the state of the OomTest object passed
+** as the first argument. Specifically, check that rc is LSM_NOMEM if an 
+** OOM error has already been injected, or LSM_OK if not.
+*/
+static void testOomAssertRc(OomTest *p, int rc){
+  testOomAssert(p, rc==LSM_OK || rc==LSM_NOMEM);
+  testOomAssert(p, testOomHit(p)==(rc==LSM_NOMEM) || p->bEnable==0 );
+}
+
+static void testOomOpen(
+  OomTest *pOom,
+  const char *zName,
+  lsm_db **ppDb,
+  int *pRc
+){
+  if( *pRc==LSM_OK ){
+    int rc;
+    rc = lsm_new(tdb_lsm_env(), ppDb);
+    if( rc==LSM_OK ) rc = lsm_open(*ppDb, zName);
+    testOomAssertRc(pOom, rc);
+    *pRc = rc;
+  }
+}
+
+static void testOomFetch(
+  OomTest *pOom,
+  lsm_db *pDb,
+  void *pKey, int nKey,
+  void *pVal, int nVal,
+  int *pRc
+){
+  testOomAssertRc(pOom, *pRc);
+  if( *pRc==LSM_OK ){
+    lsm_cursor *pCsr;
+    int rc;
+
+    rc = lsm_csr_open(pDb, &pCsr);
+    if( rc==LSM_OK ) rc = lsm_csr_seek(pCsr, pKey, nKey, 0);
+    testOomAssertRc(pOom, rc);
+
+    if( rc==LSM_OK ){
+      const void *p; int n;
+      testOomAssert(pOom, lsm_csr_valid(pCsr));
+
+      rc = lsm_csr_key(pCsr, &p, &n);
+      testOomAssertRc(pOom, rc);
+      testOomAssert(pOom, rc!=LSM_OK || (n==nKey && memcmp(pKey, p, nKey)==0) );
+    }
+
+    if( rc==LSM_OK ){
+      const void *p; int n;
+      testOomAssert(pOom, lsm_csr_valid(pCsr));
+
+      rc = lsm_csr_value(pCsr, &p, &n);
+      testOomAssertRc(pOom, rc);
+      testOomAssert(pOom, rc!=LSM_OK || (n==nVal && memcmp(pVal, p, nVal)==0) );
+    }
+
+    lsm_csr_close(pCsr);
+    *pRc = rc;
+  }
+}
+
+static void testOomWrite(
+  OomTest *pOom,
+  lsm_db *pDb,
+  void *pKey, int nKey,
+  void *pVal, int nVal,
+  int *pRc
+){
+  testOomAssertRc(pOom, *pRc);
+  if( *pRc==LSM_OK ){
+    int rc;
+
+    rc = lsm_insert(pDb, pKey, nKey, pVal, nVal);
+    testOomAssertRc(pOom, rc);
+
+    *pRc = rc;
+  }
+}
+
+
+static void testOomFetchStr(
+  OomTest *pOom,
+  lsm_db *pDb,
+  const char *zKey,
+  const char *zVal,
+  int *pRc
+){
+  int nKey = strlen(zKey);
+  int nVal = strlen(zVal);
+  testOomFetch(pOom, pDb, (void *)zKey, nKey, (void *)zVal, nVal, pRc);
+}
+
+static void testOomFetchData(
+  OomTest *pOom,
+  lsm_db *pDb,
+  Datasource *pData,
+  int iKey,
+  int *pRc
+){
+  void *pKey; int nKey;
+  void *pVal; int nVal;
+  testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+  testOomFetch(pOom, pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+static void testOomWriteStr(
+  OomTest *pOom,
+  lsm_db *pDb,
+  const char *zKey,
+  const char *zVal,
+  int *pRc
+){
+  int nKey = strlen(zKey);
+  int nVal = strlen(zVal);
+  testOomWrite(pOom, pDb, (void *)zKey, nKey, (void *)zVal, nVal, pRc);
+}
+
+static void testOomWriteData(
+  OomTest *pOom,
+  lsm_db *pDb,
+  Datasource *pData,
+  int iKey,
+  int *pRc
+){
+  void *pKey; int nKey;
+  void *pVal; int nVal;
+  testDatasourceEntry(pData, iKey, &pKey, &nKey, &pVal, &nVal);
+  testOomWrite(pOom, pDb, pKey, nKey, pVal, nVal, pRc);
+}
+
+static void testOomScan(
+  OomTest *pOom, 
+  lsm_db *pDb, 
+  int bReverse,
+  const void *pKey, int nKey,
+  int nScan,
+  int *pRc
+){
+  if( *pRc==0 ){
+    int rc;
+    int iScan = 0;
+    lsm_cursor *pCsr;
+    int (*xAdvance)(lsm_cursor *) = 0;
+    
+
+    rc = lsm_csr_open(pDb, &pCsr);
+    testOomAssertRc(pOom, rc);
+
+    if( rc==LSM_OK ){
+      if( bReverse ){
+        rc = lsm_csr_seek(pCsr, pKey, nKey, LSM_SEEK_LE);
+        xAdvance = lsm_csr_prev;
+      }else{
+        rc = lsm_csr_seek(pCsr, pKey, nKey, LSM_SEEK_GE);
+        xAdvance = lsm_csr_next;
+      }
+    }
+    testOomAssertRc(pOom, rc);
+
+    while( rc==LSM_OK && lsm_csr_valid(pCsr) && iScan<nScan ){
+      const void *p; int n;
+
+      rc = lsm_csr_key(pCsr, &p, &n);
+      testOomAssertRc(pOom, rc);
+      if( rc==LSM_OK ){
+        rc = lsm_csr_value(pCsr, &p, &n);
+        testOomAssertRc(pOom, rc);
+      }
+      if( rc==LSM_OK ){
+        rc = xAdvance(pCsr);
+        testOomAssertRc(pOom, rc);
+      }
+      iScan++;
+    }
+
+    lsm_csr_close(pCsr);
+    *pRc = rc;
+  }
+}
+
+#define LSMTEST6_TESTDB "testdb.lsm" 
+
+void testDeleteLsmdb(const char *zFile){
+  char *zLog = testMallocPrintf("%s-log", zFile);
+  char *zShm = testMallocPrintf("%s-shm", zFile);
+  unlink(zFile);
+  unlink(zLog);
+  unlink(zShm);
+  testFree(zLog);
+  testFree(zShm);
+}
+
+static void copy_file(const char *zFrom, const char *zTo, int isDatabase){
+
+  if( access(zFrom, F_OK) ){
+    unlink(zTo);
+  }else{
+    int fd1;
+    int fd2;
+    off_t sz;
+    off_t i;
+    struct stat buf;
+    u8 *aBuf;
+
+    fd1 = open(zFrom, O_RDONLY | _O_BINARY, 0644);
+    fd2 = open(zTo, O_RDWR | O_CREAT | _O_BINARY, 0644);
+
+    fstat(fd1, &buf);
+    sz = buf.st_size;
+    ftruncate(fd2, sz);
+
+    aBuf = testMalloc(4096);
+    for(i=0; i<sz; i+=4096){
+      int bLockPage = isDatabase && i == 0;
+      int nByte = MIN((bLockPage ? 4066 : 4096), sz - i);
+      memset(aBuf, 0, 4096);
+      read(fd1, aBuf, nByte);
+      write(fd2, aBuf, nByte);
+      if( bLockPage ){
+        lseek(fd1, 4096, SEEK_SET);
+        lseek(fd2, 4096, SEEK_SET);
+      }
+    }
+    testFree(aBuf);
+
+    close(fd1);
+    close(fd2);
+  }
+}
+
+void testCopyLsmdb(const char *zFrom, const char *zTo){
+  char *zLog1 = testMallocPrintf("%s-log", zFrom);
+  char *zLog2 = testMallocPrintf("%s-log", zTo);
+  char *zShm1 = testMallocPrintf("%s-shm", zFrom);
+  char *zShm2 = testMallocPrintf("%s-shm", zTo);
+
+  unlink(zShm2);
+  unlink(zLog2);
+  unlink(zTo);
+  copy_file(zFrom, zTo, 1);
+  copy_file(zLog1, zLog2, 0);
+  copy_file(zShm1, zShm2, 0);
+
+  testFree(zLog1); testFree(zLog2); testFree(zShm1); testFree(zShm2);
+}
+
+/*
+** File zFile is the path to a database. This function makes backups
+** of the database file and its log as follows:
+**
+**     cp $(zFile)         $(zFile)-save
+**     cp $(zFile)-$(zAux) $(zFile)-save-$(zAux)
+**
+** Function testRestoreDb() can be used to copy the files back in the
+** other direction.
+*/
+void testSaveDb(const char *zFile, const char *zAux){
+  char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
+  char *zFileSave = testMallocPrintf("%s-save", zFile);
+  char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);
+
+  unlink(zFileSave);
+  unlink(zLogSave);
+  copy_file(zFile, zFileSave, 1);
+  copy_file(zLog, zLogSave, 0);
+
+  testFree(zLog); testFree(zFileSave); testFree(zLogSave);
+}
+
+/*
+** File zFile is the path to a database. This function restores
+** a backup of the database made by a previous call to testSaveDb().
+** Specifically, it does the equivalent of:
+**
+**     cp $(zFile)-save         $(zFile)
+**     cp $(zFile)-save-$(zAux) $(zFile)-$(zAux)
+*/
+void testRestoreDb(const char *zFile, const char *zAux){
+  char *zLog = testMallocPrintf("%s-%s", zFile, zAux);
+  char *zFileSave = testMallocPrintf("%s-save", zFile);
+  char *zLogSave = testMallocPrintf("%s-%s-save", zFile, zAux);
+
+  copy_file(zFileSave, zFile, 1);
+  copy_file(zLogSave, zLog, 0);
+
+  testFree(zLog); testFree(zFileSave); testFree(zLogSave);
+}
+
+
+static int lsmWriteStr(lsm_db *pDb, const char *zKey, const char *zVal){
+  int nKey = strlen(zKey);
+  int nVal = strlen(zVal);
+  return lsm_insert(pDb, (void *)zKey, nKey, (void *)zVal, nVal);
+}
+
+static void setup_delete_db(void){
+  testDeleteLsmdb(LSMTEST6_TESTDB);
+}
+
+/*
+** Create a small database. With the following content:
+**
+**    "one"   -> "one"
+**    "two"   -> "four"
+**    "three" -> "nine"
+**    "four"  -> "sixteen"
+**    "five"  -> "twentyfive"
+**    "six"   -> "thirtysix"
+**    "seven" -> "fourtynine"
+**    "eight" -> "sixtyfour"
+*/
+static void setup_populate_db(void){
+  const char *azStr[] = {
+    "one",   "one",
+    "two",   "four",
+    "three", "nine",
+    "four",  "sixteen",
+    "five",  "twentyfive",
+    "six",   "thirtysix",
+    "seven", "fourtynine",
+    "eight", "sixtyfour",
+  };
+  int rc;
+  int ii;
+  lsm_db *pDb;
+
+  testDeleteLsmdb(LSMTEST6_TESTDB);
+
+  rc = lsm_new(tdb_lsm_env(), &pDb);
+  if( rc==LSM_OK ) rc = lsm_open(pDb, LSMTEST6_TESTDB);
+
+  for(ii=0; rc==LSM_OK && ii<ArraySize(azStr); ii+=2){
+    rc = lsmWriteStr(pDb, azStr[ii], azStr[ii+1]);
+  }
+  lsm_close(pDb);
+
+  testSaveDb(LSMTEST6_TESTDB, "log");
+  assert( rc==LSM_OK );
+}
+
+static Datasource *getDatasource(void){
+  const DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 200, 250 };
+  return testDatasourceNew(&defn);
+}
+
+/*
+** Set up a database file with the following properties:
+**
+**   * Page size is 1024 bytes.
+**   * Block size is 64 KB.
+**   * Contains 5000 key-value pairs starting at 0 from the
+**     datasource returned getDatasource().
+*/
+static void setup_populate_db2(void){
+  Datasource *pData;
+  int ii;
+  int rc;
+  int nBlocksize = 64*1024;
+  int nPagesize = 1024;
+  int nWritebuffer = 4*1024;
+  lsm_db *pDb;
+
+  testDeleteLsmdb(LSMTEST6_TESTDB);
+  rc = lsm_new(tdb_lsm_env(), &pDb);
+  if( rc==LSM_OK ) rc = lsm_open(pDb, LSMTEST6_TESTDB);
+
+  lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &nBlocksize); 
+  lsm_config(pDb, LSM_CONFIG_PAGE_SIZE, &nPagesize); 
+  lsm_config(pDb, LSM_CONFIG_AUTOFLUSH, &nWritebuffer); 
+
+  pData = getDatasource();
+  for(ii=0; rc==LSM_OK && ii<5000; ii++){
+    void *pKey; int nKey;
+    void *pVal; int nVal;
+    testDatasourceEntry(pData, ii, &pKey, &nKey, &pVal, &nVal);
+    lsm_insert(pDb, pKey, nKey, pVal, nVal);
+  }
+  testDatasourceFree(pData);
+  lsm_close(pDb);
+
+  testSaveDb(LSMTEST6_TESTDB, "log");
+  assert( rc==LSM_OK );
+}
+
+/*
+** Test the results of OOM conditions in lsm_new().
+*/
+static void simple_oom_1(OomTest *pOom){
+  int rc;
+  lsm_db *pDb;
+
+  rc = lsm_new(tdb_lsm_env(), &pDb);
+  testOomAssertRc(pOom, rc);
+
+  lsm_close(pDb);
+}
+
+/*
+** Test the results of OOM conditions in lsm_open().
+*/
+static void simple_oom_2(OomTest *pOom){
+  int rc;
+  lsm_db *pDb;
+
+  rc = lsm_new(tdb_lsm_env(), &pDb);
+  if( rc==LSM_OK ){
+    rc = lsm_open(pDb, "testdb.lsm");
+  }
+  testOomAssertRc(pOom, rc);
+
+  lsm_close(pDb);
+}
+
+/*
+** Test the results of OOM conditions in simple fetch operations.
+*/
+static void simple_oom_3(OomTest *pOom){
+  int rc = LSM_OK;
+  lsm_db *pDb;
+
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+  testOomFetchStr(pOom, pDb, "four",  "sixteen",    &rc);
+  testOomFetchStr(pOom, pDb, "seven", "fourtynine", &rc);
+  testOomFetchStr(pOom, pDb, "one",   "one",        &rc);
+  testOomFetchStr(pOom, pDb, "eight", "sixtyfour",  &rc);
+
+  lsm_close(pDb);
+}
+
+/*
+** Test the results of OOM conditions in simple write operations.
+*/
+static void simple_oom_4(OomTest *pOom){
+  int rc = LSM_OK;
+  lsm_db *pDb;
+
+  testDeleteLsmdb(LSMTEST6_TESTDB);
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+  testOomWriteStr(pOom, pDb, "123", "onetwothree", &rc);
+  testOomWriteStr(pOom, pDb, "456", "fourfivesix", &rc);
+  testOomWriteStr(pOom, pDb, "789", "seveneightnine", &rc);
+  testOomWriteStr(pOom, pDb, "123", "teneleventwelve", &rc);
+  testOomWriteStr(pOom, pDb, "456", "fourteenfifteensixteen", &rc);
+
+  lsm_close(pDb);
+}
+
+static void simple_oom_5(OomTest *pOom){
+  Datasource *pData = getDatasource();
+  int rc = LSM_OK;
+  lsm_db *pDb;
+
+  testRestoreDb(LSMTEST6_TESTDB, "log");
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+  testOomFetchData(pOom, pDb, pData, 3333, &rc);
+  testOomFetchData(pOom, pDb, pData, 0, &rc);
+  testOomFetchData(pOom, pDb, pData, 4999, &rc);
+
+  lsm_close(pDb);
+  testDatasourceFree(pData);
+}
+
+static void simple_oom_6(OomTest *pOom){
+  Datasource *pData = getDatasource();
+  int rc = LSM_OK;
+  lsm_db *pDb;
+
+  testRestoreDb(LSMTEST6_TESTDB, "log");
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+
+  testOomWriteData(pOom, pDb, pData, 5000, &rc);
+  testOomWriteData(pOom, pDb, pData, 5001, &rc);
+  testOomWriteData(pOom, pDb, pData, 5002, &rc);
+  testOomFetchData(pOom, pDb, pData, 5001, &rc);
+  testOomFetchData(pOom, pDb, pData, 1234, &rc);
+
+  lsm_close(pDb);
+  testDatasourceFree(pData);
+}
+
+static void simple_oom_7(OomTest *pOom){
+  Datasource *pData = getDatasource();
+  int rc = LSM_OK;
+  lsm_db *pDb;
+
+  testRestoreDb(LSMTEST6_TESTDB, "log");
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+  testOomScan(pOom, pDb, 0, "abc", 3, 20, &rc);
+  lsm_close(pDb);
+  testDatasourceFree(pData);
+}
+
+static void simple_oom_8(OomTest *pOom){
+  Datasource *pData = getDatasource();
+  int rc = LSM_OK;
+  lsm_db *pDb;
+  testRestoreDb(LSMTEST6_TESTDB, "log");
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb, &rc);
+  testOomScan(pOom, pDb, 1, "xyz", 3, 20, &rc);
+  lsm_close(pDb);
+  testDatasourceFree(pData);
+}
+
+/*
+** This test case has two clients connected to a database. The first client
+** hits an OOM while writing to the database. Check that the second 
+** connection is still able to query the db following the OOM.
+*/
+static void simple_oom2_1(OomTest *pOom){
+  const int nRecord = 100;        /* Number of records initially in db */
+  const int nIns = 10;            /* Number of records inserted with OOM */
+
+  Datasource *pData = getDatasource();
+  int rc = LSM_OK;
+  lsm_db *pDb1;
+  lsm_db *pDb2;
+  int i;
+
+  testDeleteLsmdb(LSMTEST6_TESTDB);
+
+  /* Open the two connections. Initialize the in-memory tree so that it
+  ** contains 100 records. Do all this with OOM injection disabled. */
+  testOomEnable(pOom, 0);
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb1, &rc);
+  testOomOpen(pOom, LSMTEST6_TESTDB, &pDb2, &rc);
+  for(i=0; i<nRecord; i++){
+    testOomWriteData(pOom, pDb1, pData, i, &rc);
+  }
+  testOomEnable(pOom, 1);
+  assert( rc==0 );
+
+  /* Insert 10 more records using pDb1. Stop when an OOM is encountered. */
+  for(i=nRecord; i<nRecord+nIns; i++){
+    testOomWriteData(pOom, pDb1, pData, i, &rc);
+    if( rc ) break;
+  }
+  testOomAssertRc(pOom, rc);
+
+  /* Switch off OOM injection. Write a few rows using pDb2. Then check
+  ** that the database may be successfully queried.  */
+  testOomEnable(pOom, 0);
+  rc = 0;
+  for(; i<nRecord+nIns && rc==0; i++){
+    testOomWriteData(pOom, pDb2, pData, i, &rc);
+  }
+  for(i=0; i<nRecord+nIns; i++) testOomFetchData(pOom, pDb2, pData, i, &rc);
+  testOomEnable(pOom, 1);
+
+  lsm_close(pDb1);
+  lsm_close(pDb2);
+  testDatasourceFree(pData);
+}
+
+
+static void do_test_oom1(const char *zPattern, int *pRc){
+  struct SimpleOom {
+    const char *zName;
+    void (*xSetup)(void);
+    void (*xFunc)(OomTest *);
+  } aSimple[] = {
+    { "oom1.lsm.1", setup_delete_db,    simple_oom_1 },
+    { "oom1.lsm.2", setup_delete_db,    simple_oom_2 },
+    { "oom1.lsm.3", setup_populate_db,  simple_oom_3 },
+    { "oom1.lsm.4", setup_delete_db,    simple_oom_4 },
+    { "oom1.lsm.5", setup_populate_db2, simple_oom_5 },
+    { "oom1.lsm.6", setup_populate_db2, simple_oom_6 },
+    { "oom1.lsm.7", setup_populate_db2, simple_oom_7 },
+    { "oom1.lsm.8", setup_populate_db2, simple_oom_8 },
+
+    { "oom2.lsm.1", setup_delete_db,    simple_oom2_1 },
+  };
+  int i;
+
+  for(i=0; i<ArraySize(aSimple); i++){
+    if( *pRc==0 && testCaseBegin(pRc, zPattern, "%s", aSimple[i].zName) ){
+      OomTest t;
+
+      if( aSimple[i].xSetup ){
+        aSimple[i].xSetup();
+      }
+
+      for(testOomStart(&t); testOomContinue(&t); testOomNext(&t)){
+        aSimple[i].xFunc(&t);
+      }
+
+      printf("(%d injections).", t.iNext-2);
+      testCaseFinish( (*pRc = testOomFinish(&t)) );
+      testMallocOom(tdb_lsm_env(), 0, 0, 0, 0);
+    }
+  }
+}
+
+void test_oom(
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  do_test_oom1(zPattern, pRc);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest7.c b/ext/lsm1/lsm-test/lsmtest7.c
new file mode 100644
index 0000000..2d26b53
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest7.c
@@ -0,0 +1,206 @@
+
+
+#include "lsmtest.h"
+
+
+/*
+** Test that the rules for when lsm_csr_next() and lsm_csr_prev() are
+** enforced. Specifically:
+**
+**   * Both functions always return LSM_MISUSE if the cursor is at EOF
+**     when they are called.
+**
+**   * lsm_csr_next() may only be used after lsm_csr_seek(LSM_SEEK_GE) or 
+**     lsm_csr_first(). 
+**
+**   * lsm_csr_prev() may only be used after lsm_csr_seek(LSM_SEEK_LE) or 
+**     lsm_csr_last().
+*/
+static void do_test_api1_lsm(lsm_db *pDb, int *pRc){
+  int ret;
+  lsm_cursor *pCsr;
+  lsm_cursor *pCsr2;
+  int nKey;
+  const void *pKey;
+
+  ret = lsm_csr_open(pDb, &pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_GE);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_LE);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+
+  ret = lsm_csr_seek(pCsr, "jjj", 3, LSM_SEEK_LEFAST);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  ret = lsm_csr_key(pCsr, &pKey, &nKey);
+  testCompareInt(LSM_OK, ret, pRc);
+
+  ret = lsm_csr_open(pDb, &pCsr2);
+  testCompareInt(LSM_OK, ret, pRc);
+
+  ret = lsm_csr_seek(pCsr2, pKey, nKey, LSM_SEEK_EQ);
+  testCompareInt(LSM_OK, ret, pRc);
+  testCompareInt(1, lsm_csr_valid(pCsr2), pRc);
+  ret = lsm_csr_next(pCsr2);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+  ret = lsm_csr_prev(pCsr2);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  lsm_csr_close(pCsr2);
+
+  ret = lsm_csr_first(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  ret = lsm_csr_last(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  ret = lsm_csr_first(pCsr);
+  while( lsm_csr_valid(pCsr) ){
+    ret = lsm_csr_next(pCsr);
+    testCompareInt(LSM_OK, ret, pRc);
+  }
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  ret = lsm_csr_last(pCsr);
+  while( lsm_csr_valid(pCsr) ){
+    ret = lsm_csr_prev(pCsr);
+    testCompareInt(LSM_OK, ret, pRc);
+  }
+  ret = lsm_csr_prev(pCsr);
+  testCompareInt(LSM_OK, ret, pRc);
+  ret = lsm_csr_next(pCsr);
+  testCompareInt(LSM_MISUSE, ret, pRc);
+
+  lsm_csr_close(pCsr);
+}
+
+static void do_test_api1(const char *zPattern, int *pRc){
+  if( testCaseBegin(pRc, zPattern, "api1.lsm") ){
+    const DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 10, 15, 200, 250 };
+    Datasource *pData;
+    TestDb *pDb;
+    int rc = 0;
+
+    pDb = testOpen("lsm_lomem", 1, &rc);
+    pData = testDatasourceNew(&defn);
+    testWriteDatasourceRange(pDb, pData, 0, 1000, pRc);
+
+    do_test_api1_lsm(tdb_lsm(pDb), pRc);
+
+    testDatasourceFree(pData);
+    testClose(&pDb);
+
+    testCaseFinish(*pRc);
+  }
+}
+
+static lsm_db *newLsmConnection(
+  const char *zDb, 
+  int nPgsz, 
+  int nBlksz,
+  int *pRc
+){
+  lsm_db *db = 0;
+  if( *pRc==0 ){
+    int n1 = nPgsz;
+    int n2 = nBlksz;
+    *pRc = lsm_new(tdb_lsm_env(), &db);
+    if( *pRc==0 ){
+      if( n1 ) lsm_config(db, LSM_CONFIG_PAGE_SIZE, &n1);
+      if( n2 ) lsm_config(db, LSM_CONFIG_BLOCK_SIZE, &n2);
+      *pRc = lsm_open(db, "testdb.lsm");
+    }
+  }
+  return db;
+}
+
+static void testPagesize(lsm_db *db, int nPgsz, int nBlksz, int *pRc){
+  if( *pRc==0 ){
+    int n1 = 0;
+    int n2 = 0;
+
+    lsm_config(db, LSM_CONFIG_PAGE_SIZE, &n1);
+    lsm_config(db, LSM_CONFIG_BLOCK_SIZE, &n2);
+
+    testCompareInt(n1, nPgsz, pRc);
+    testCompareInt(n2, nBlksz, pRc);
+  }
+}
+
+/*
+** Test case "api2" tests that the default page and block sizes of a 
+** database may only be modified before lsm_open() is called. And that
+** after lsm_open() is called lsm_config() may be used to read the 
+** actual page and block size of the db.
+*/
+static void do_test_api2(const char *zPattern, int *pRc){
+  if( *pRc==0 && testCaseBegin(pRc, zPattern, "api2.lsm") ){
+    lsm_db *db1 = 0;
+    lsm_db *db2 = 0;
+
+    testDeleteLsmdb("testdb.lsm");
+    db1 = newLsmConnection("testdb.lsm", 0, 0, pRc);
+    testPagesize(db1, 4096, 1024, pRc);
+    db2 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
+    testPagesize(db2, 4096, 1024, pRc);
+    lsm_close(db1);
+    lsm_close(db2);
+
+    testDeleteLsmdb("testdb.lsm");
+    db1 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
+    testPagesize(db1, 1024, 64*1024, pRc);
+    db2 = newLsmConnection("testdb.lsm", 0, 0, pRc);
+    testPagesize(db2, 1024, 64*1024, pRc);
+    lsm_close(db1);
+    lsm_close(db2);
+
+    testDeleteLsmdb("testdb.lsm");
+    db1 = newLsmConnection("testdb.lsm", 8192, 2*1024, pRc);
+    testPagesize(db1, 8192, 2*1024, pRc);
+    db2 = newLsmConnection("testdb.lsm", 1024, 64*1024, pRc);
+    testPagesize(db2, 8192, 2*1024, pRc);
+    lsm_close(db1);
+    lsm_close(db2);
+
+    testCaseFinish(*pRc);
+  }
+}
+
+void test_api(
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  do_test_api1(zPattern, pRc);
+  do_test_api2(zPattern, pRc);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest8.c b/ext/lsm1/lsm-test/lsmtest8.c
new file mode 100644
index 0000000..7efa0df
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest8.c
@@ -0,0 +1,324 @@
+
+/*
+** This file contains test cases to verify that "live-recovery" following
+** a mid-transaction failure of a writer process.
+*/
+
+
+/* 
+** This test file includes lsmInt.h to get access to the definition of the
+** ShmHeader structure. This is required to cause strategic damage to the
+** shared memory header as part of recovery testing.
+*/
+#include "lsmInt.h"
+
+#include "lsmtest.h"
+
+typedef struct SetupStep SetupStep;
+struct SetupStep {
+  int bFlush;                     /* Flush to disk and checkpoint */
+  int iInsStart;                  /* First key-value from ds to insert */
+  int nIns;                       /* Number of rows to insert */
+  int iDelStart;                  /* First key from ds to delete */
+  int nDel;                       /* Number of rows to delete */
+};
+
+static void doSetupStep(
+  TestDb *pDb, 
+  Datasource *pData, 
+  const SetupStep *pStep, 
+  int *pRc
+){
+  testWriteDatasourceRange(pDb, pData, pStep->iInsStart, pStep->nIns, pRc);
+  testDeleteDatasourceRange(pDb, pData, pStep->iDelStart, pStep->nDel, pRc);
+  if( *pRc==0 ){
+    int nSave = -1;
+    int nBuf = 64;
+    lsm_db *db = tdb_lsm(pDb);
+
+    lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nSave);
+    lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
+    lsm_begin(db, 1);
+    lsm_commit(db, 0);
+    lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nSave);
+
+    *pRc = lsm_work(db, 0, 0, 0);
+    if( *pRc==0 ){
+      *pRc = lsm_checkpoint(db, 0);
+    }
+  }
+}
+
+static void doSetupStepArray(
+  TestDb *pDb, 
+  Datasource *pData, 
+  const SetupStep *aStep, 
+  int nStep
+){
+  int i;
+  for(i=0; i<nStep; i++){
+    int rc = 0;
+    doSetupStep(pDb, pData, &aStep[i], &rc);
+    assert( rc==0 );
+  }
+}
+
+static void setupDatabase1(TestDb *pDb, Datasource **ppData){
+  const SetupStep aStep[] = {
+    { 0,                                  1,     2000, 0, 0 },
+    { 1,                                  0,     0, 0, 0 },
+    { 0,                                  10001, 1000, 0, 0 },
+  };
+  const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 12, 16, 100, 500};
+  Datasource *pData;
+
+  pData = testDatasourceNew(&defn);
+  doSetupStepArray(pDb, pData, aStep, ArraySize(aStep));
+  if( ppData ){
+    *ppData = pData;
+  }else{
+    testDatasourceFree(pData);
+  }
+}
+
+#include <stdio.h>
+void testReadFile(const char *zFile, int iOff, void *pOut, int nByte, int *pRc){
+  if( *pRc==0 ){
+    FILE *fd;
+    fd = fopen(zFile, "rb");
+    if( fd==0 ){
+      *pRc = 1;
+    }else{
+      if( 0!=fseek(fd, iOff, SEEK_SET) ){
+        *pRc = 1;
+      }else{
+        assert( nByte>=0 );
+        if( (size_t)nByte!=fread(pOut, 1, nByte, fd) ){
+          *pRc = 1;
+        }
+      }
+      fclose(fd);
+    }
+  }
+}
+
+void testWriteFile(
+  const char *zFile, 
+  int iOff, 
+  void *pOut, 
+  int nByte, 
+  int *pRc
+){
+  if( *pRc==0 ){
+    FILE *fd;
+    fd = fopen(zFile, "r+b");
+    if( fd==0 ){
+      *pRc = 1;
+    }else{
+      if( 0!=fseek(fd, iOff, SEEK_SET) ){
+        *pRc = 1;
+      }else{
+        assert( nByte>=0 );
+        if( (size_t)nByte!=fwrite(pOut, 1, nByte, fd) ){
+          *pRc = 1;
+        }
+      }
+      fclose(fd);
+    }
+  }
+}
+
+static ShmHeader *getShmHeader(const char *zDb){
+  int rc = 0;
+  char *zShm = testMallocPrintf("%s-shm", zDb);
+  ShmHeader *pHdr;
+
+  pHdr = testMalloc(sizeof(ShmHeader));
+  testReadFile(zShm, 0, (void *)pHdr, sizeof(ShmHeader), &rc);
+  assert( rc==0 );
+
+  return pHdr;
+}
+
+/*
+** This function makes a copy of the three files associated with LSM 
+** database zDb (i.e. if zDb is "test.db", it makes copies of "test.db",
+** "test.db-log" and "test.db-shm").
+**
+** It then opens a new database connection to the copy with the xLock() call
+** instrumented so that it appears that some other process already connected
+** to the db (holding a shared lock on DMS2). This prevents recovery from
+** running. Then:
+**
+**    1) Check that the checksum of the database is zCksum. 
+**    2) Write a few keys to the database. Then delete the same keys. 
+**    3) Check that the checksum is zCksum.
+**    4) Flush the db to disk and run a checkpoint. 
+**    5) Check once more that the checksum is still zCksum.
+*/
+static void doLiveRecovery(const char *zDb, const char *zCksum, int *pRc){
+  if( *pRc==LSM_OK ){
+    const DatasourceDefn defn = {TEST_DATASOURCE_RANDOM, 20, 25, 100, 500};
+    Datasource *pData;
+    const char *zCopy = "testcopy.lsm";
+    char zCksum2[TEST_CKSUM_BYTES];
+    TestDb *pDb = 0;
+    int rc;
+
+    pData = testDatasourceNew(&defn);
+
+    testCopyLsmdb(zDb, zCopy);
+    rc = tdb_lsm_open("test_no_recovery=1", zCopy, 0, &pDb);
+    if( rc==0 ){
+      ShmHeader *pHdr;
+      lsm_db *db;
+      testCksumDatabase(pDb, zCksum2);
+      testCompareStr(zCksum, zCksum2, &rc);
+
+      testWriteDatasourceRange(pDb, pData, 1, 10, &rc);
+      testDeleteDatasourceRange(pDb, pData, 1, 10, &rc);
+
+      /* Test that the two tree-headers are now consistent. */
+      pHdr = getShmHeader(zCopy);
+      if( rc==0 && memcmp(&pHdr->hdr1, &pHdr->hdr2, sizeof(pHdr->hdr1)) ){
+        rc = 1;
+      }
+      testFree(pHdr);
+
+      if( rc==0 ){
+        int nBuf = 64;
+        db = tdb_lsm(pDb);
+        lsm_config(db, LSM_CONFIG_AUTOFLUSH, &nBuf);
+        lsm_begin(db, 1);
+        lsm_commit(db, 0);
+        rc = lsm_work(db, 0, 0, 0);
+      }
+
+      testCksumDatabase(pDb, zCksum2);
+      testCompareStr(zCksum, zCksum2, &rc);
+    }
+
+    testDatasourceFree(pData);
+    testClose(&pDb);
+    testDeleteLsmdb(zCopy);
+    *pRc = rc;
+  }
+}
+
+static void doWriterCrash1(int *pRc){
+  const int nWrite = 2000;
+  const int nStep = 10;
+  const int iWriteStart = 20000;
+  int rc = 0;
+  TestDb *pDb = 0;
+  Datasource *pData = 0;
+
+  rc = tdb_lsm_open("autowork=0", "testdb.lsm", 1, &pDb);
+  if( rc==0 ){
+    int iDot = 0;
+    char zCksum[TEST_CKSUM_BYTES];
+    int i;
+    setupDatabase1(pDb, &pData);
+    testCksumDatabase(pDb, zCksum);
+    testBegin(pDb, 2, &rc);
+    for(i=0; rc==0 && i<nWrite; i+=nStep){
+      testCaseProgress(i, nWrite, testCaseNDot(), &iDot);
+      testWriteDatasourceRange(pDb, pData, iWriteStart+i, nStep, &rc);
+      doLiveRecovery("testdb.lsm", zCksum, &rc);
+    }
+  }
+  testCommit(pDb, 0, &rc);
+  testClose(&pDb);
+  testDatasourceFree(pData);
+  *pRc = rc;
+}
+
+/*
+** This test case verifies that inconsistent tree-headers in shared-memory
+** are resolved correctly. 
+*/
+static void doWriterCrash2(int *pRc){
+  int rc = 0;
+  TestDb *pDb = 0;
+  Datasource *pData = 0;
+
+  rc = tdb_lsm_open("autowork=0", "testdb.lsm", 1, &pDb);
+  if( rc==0 ){
+    ShmHeader *pHdr1;
+    ShmHeader *pHdr2;
+    char zCksum1[TEST_CKSUM_BYTES];
+    char zCksum2[TEST_CKSUM_BYTES];
+
+    pHdr1 = testMalloc(sizeof(ShmHeader));
+    pHdr2 = testMalloc(sizeof(ShmHeader));
+    setupDatabase1(pDb, &pData);
+
+    /* Grab a copy of the shared-memory header. And the db checksum */
+    testReadFile("testdb.lsm-shm", 0, (void *)pHdr1, sizeof(ShmHeader), &rc);
+    testCksumDatabase(pDb, zCksum1);
+
+    /* Modify the database */
+    testBegin(pDb, 2, &rc);
+    testWriteDatasourceRange(pDb, pData, 30000, 200, &rc);
+    testCommit(pDb, 0, &rc);
+
+    /* Grab a second copy of the shared-memory header. And the db checksum */
+    testReadFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+    testCksumDatabase(pDb, zCksum2);
+    doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+    /* If both tree-headers are valid, tree-header-1 is used. */
+    memcpy(&pHdr2->hdr1, &pHdr1->hdr1, sizeof(pHdr1->hdr1));
+    pHdr2->bWriter = 1;
+    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+    doLiveRecovery("testdb.lsm", zCksum1, &rc);
+
+    /* If both tree-headers are valid, tree-header-1 is used. */
+    memcpy(&pHdr2->hdr1, &pHdr2->hdr2, sizeof(pHdr1->hdr1));
+    memcpy(&pHdr2->hdr2, &pHdr1->hdr1, sizeof(pHdr1->hdr1));
+    pHdr2->bWriter = 1;
+    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+    doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+    /* If tree-header 1 is invalid, tree-header-2 is used */
+    memcpy(&pHdr2->hdr2, &pHdr2->hdr1, sizeof(pHdr1->hdr1));
+    pHdr2->hdr1.aCksum[0] = 5;
+    pHdr2->hdr1.aCksum[0] = 6;
+    pHdr2->bWriter = 1;
+    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+    doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+    /* If tree-header 2 is invalid, tree-header-1 is used */
+    memcpy(&pHdr2->hdr1, &pHdr2->hdr2, sizeof(pHdr1->hdr1));
+    pHdr2->hdr2.aCksum[0] = 5;
+    pHdr2->hdr2.aCksum[0] = 6;
+    pHdr2->bWriter = 1;
+    testWriteFile("testdb.lsm-shm", 0, (void *)pHdr2, sizeof(ShmHeader), &rc);
+    doLiveRecovery("testdb.lsm", zCksum2, &rc);
+
+    testFree(pHdr1);
+    testFree(pHdr2);
+    testClose(&pDb);
+  }
+
+  *pRc = rc;
+}
+
+void do_writer_crash_test(const char *zPattern, int *pRc){
+  struct Test {
+    const char *zName;
+    void (*xFunc)(int *);
+  } aTest[] = {
+    { "writercrash1.lsm", doWriterCrash1 },
+    { "writercrash2.lsm", doWriterCrash2 },
+  };
+  int i;
+  for(i=0; i<ArraySize(aTest); i++){
+    struct Test *p = &aTest[i];
+    if( testCaseBegin(pRc, zPattern, p->zName) ){
+      p->xFunc(pRc);
+      testCaseFinish(*pRc);
+    }
+  }
+
+}
diff --git a/ext/lsm1/lsm-test/lsmtest9.c b/ext/lsm1/lsm-test/lsmtest9.c
new file mode 100644
index 0000000..b01de0d
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest9.c
@@ -0,0 +1,140 @@
+
+#include "lsmtest.h"
+
+#define DATA_SEQUENTIAL TEST_DATASOURCE_SEQUENCE
+#define DATA_RANDOM     TEST_DATASOURCE_RANDOM
+
+typedef struct Datatest4 Datatest4;
+
+/*
+** Test overview:
+**
+**   1. Insert (Datatest4.nRec) records into a database.
+**
+**   2. Repeat (Datatest4.nRepeat) times:
+**
+**      2a. Delete 2/3 of the records in the database.
+**
+**      2b. Run lsm_work(nMerge=1).
+**
+**      2c. Insert as many records as were deleted in 2a.
+**
+**      2d. Check database content is as expected.
+**
+**      2e. If (Datatest4.bReopen) is true, close and reopen the database.
+*/
+struct Datatest4 {
+  /* Datasource definition */
+  DatasourceDefn defn;
+
+  int nRec;
+  int nRepeat;
+  int bReopen;
+};
+
+static void doDataTest4(
+  const char *zSystem,            /* Database system to test */
+  Datatest4 *p,                   /* Structure containing test parameters */
+  int *pRc                        /* OUT: Error code */
+){
+  lsm_db *db = 0;
+  TestDb *pDb;
+  TestDb *pControl;
+  Datasource *pData;
+  int i;
+  int rc = 0;
+  int iDot = 0;
+  int bMultiThreaded = 0;         /* True for MT LSM database */
+
+  int nRecOn3 = (p->nRec / 3);
+  int iData = 0;
+
+  /* Start the test case, open a database and allocate the datasource. */
+  rc = testControlDb(&pControl);
+  pDb = testOpen(zSystem, 1, &rc);
+  pData = testDatasourceNew(&p->defn);
+  if( rc==0 ){
+    db = tdb_lsm(pDb);
+    bMultiThreaded = tdb_lsm_multithread(pDb);
+  }
+
+  testWriteDatasourceRange(pControl, pData, iData, nRecOn3*3, &rc);
+  testWriteDatasourceRange(pDb,      pData, iData, nRecOn3*3, &rc);
+
+  for(i=0; rc==0 && i<p->nRepeat; i++){
+
+    testDeleteDatasourceRange(pControl, pData, iData, nRecOn3*2, &rc);
+    testDeleteDatasourceRange(pDb,      pData, iData, nRecOn3*2, &rc);
+
+    if( db ){
+      int nDone;
+#if 0
+      fprintf(stderr, "lsm_work() start...\n"); fflush(stderr);
+#endif
+      do {
+        nDone = 0;
+        rc = lsm_work(db, 1, (1<<30), &nDone);
+      }while( rc==0 && nDone>0 );
+      if( bMultiThreaded && rc==LSM_BUSY ) rc = LSM_OK;
+#if 0 
+      fprintf(stderr, "lsm_work() done...\n"); fflush(stderr);
+#endif
+    }
+
+if( i+1<p->nRepeat ){
+    iData += (nRecOn3*2);
+    testWriteDatasourceRange(pControl, pData, iData+nRecOn3, nRecOn3*2, &rc);
+    testWriteDatasourceRange(pDb,      pData, iData+nRecOn3, nRecOn3*2, &rc);
+
+    testCompareDb(pData, nRecOn3*3, iData, pControl, pDb, &rc);
+
+    /* If Datatest4.bReopen is true, close and reopen the database */
+    if( p->bReopen ){
+      testReopen(&pDb, &rc);
+      if( rc==0 ) db = tdb_lsm(pDb);
+    }
+}
+
+    /* Update the progress dots... */
+    testCaseProgress(i, p->nRepeat, testCaseNDot(), &iDot);
+  }
+
+  testClose(&pDb);
+  testClose(&pControl);
+  testDatasourceFree(pData);
+  testCaseFinish(rc);
+  *pRc = rc;
+}
+
+static char *getName4(const char *zSystem, Datatest4 *pTest){
+  char *zRet;
+  char *zData;
+  zData = testDatasourceName(&pTest->defn);
+  zRet = testMallocPrintf("data4.%s.%s.%d.%d.%d", 
+      zSystem, zData, pTest->nRec, pTest->nRepeat, pTest->bReopen
+  );
+  testFree(zData);
+  return zRet;
+}
+
+void test_data_4(
+  const char *zSystem,            /* Database system name */
+  const char *zPattern,           /* Run test cases that match this pattern */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  Datatest4 aTest[] = {
+      /* defn,                                 nRec, nRepeat, bReopen */
+    { {DATA_RANDOM,     20,25,     500,600}, 10000,      10,       0   },
+    { {DATA_RANDOM,     20,25,     500,600}, 10000,      10,       1   },
+  };
+
+  int i;
+
+  for(i=0; *pRc==LSM_OK && i<ArraySize(aTest); i++){
+    char *zName = getName4(zSystem, &aTest[i]);
+    if( testCaseBegin(pRc, zPattern, "%s", zName) ){
+      doDataTest4(zSystem, &aTest[i], pRc);
+    }
+    testFree(zName);
+  }
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_bt.c b/ext/lsm1/lsm-test/lsmtest_bt.c
new file mode 100644
index 0000000..8a4f54a
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_bt.c
@@ -0,0 +1,71 @@
+
+#include "lsmtest.h"
+#include "bt.h"
+
+int do_bt(int nArg, char **azArg){
+  struct Option {
+    const char *zName;
+    int bPgno;
+    int eOpt;
+  } aOpt [] = { 
+    { "dbhdr",          0, BT_INFO_HDRDUMP },
+    { "filename",       0, BT_INFO_FILENAME },
+    { "block_freelist", 0, BT_INFO_BLOCK_FREELIST },
+    { "page_freelist",  0, BT_INFO_PAGE_FREELIST },
+    { "filename",       0, BT_INFO_FILENAME },
+    { "page",           1, BT_INFO_PAGEDUMP },
+    { "page_ascii",     1, BT_INFO_PAGEDUMP_ASCII },
+    { "leaks",          0, BT_INFO_PAGE_LEAKS },
+    { 0, 0 } 
+  };
+  int iOpt;
+  int rc;
+  bt_info buf;
+  char *zOpt;
+  char *zFile;
+
+  bt_db *db = 0;
+
+  if( nArg<2 ){
+    testPrintUsage("FILENAME OPTION ...");
+    return -1;
+  }
+  zFile = azArg[0];
+  zOpt = azArg[1];
+
+  rc = testArgSelect(aOpt, "option", zOpt, &iOpt);
+  if( rc!=0 ) return rc;
+  if( nArg!=2+aOpt[iOpt].bPgno ){
+    testPrintFUsage("FILENAME %s %s", zOpt, aOpt[iOpt].bPgno ? "PGNO" : "");
+    return -4;
+  }
+
+  rc = sqlite4BtNew(sqlite4_env_default(), 0, &db);
+  if( rc!=SQLITE4_OK ){
+    testPrintError("sqlite4BtNew() failed: %d", rc);
+    return -2;
+  }
+  rc = sqlite4BtOpen(db, zFile);
+  if( rc!=SQLITE4_OK ){
+    testPrintError("sqlite4BtOpen() failed: %d", rc);
+    return -3;
+  }
+
+  buf.eType = aOpt[iOpt].eOpt;
+  buf.pgno = 0;
+  sqlite4_buffer_init(&buf.output, 0);
+
+  if( aOpt[iOpt].bPgno ){
+    buf.pgno = (u32)atoi(azArg[2]);
+  }
+
+  rc = sqlite4BtControl(db, BT_CONTROL_INFO, &buf);
+  if( rc!=SQLITE4_OK ){
+    testPrintError("sqlite4BtControl() failed: %d\n", rc);
+    return -4;
+  }
+
+  printf("%s\n", (char*)buf.output.p);
+  sqlite4_buffer_clear(&buf.output);
+  return 0;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_datasource.c b/ext/lsm1/lsm-test/lsmtest_datasource.c
new file mode 100644
index 0000000..0b0fd94
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_datasource.c
@@ -0,0 +1,96 @@
+
+
+#include "lsmtest.h"
+
+struct Datasource {
+  int eType;
+
+  int nMinKey;
+  int nMaxKey;
+  int nMinVal;
+  int nMaxVal;
+
+  char *aKey;
+  char *aVal;
+};
+
+void testDatasourceEntry(
+  Datasource *p, 
+  int iData, 
+  void **ppKey, int *pnKey,
+  void **ppVal, int *pnVal
+){
+  assert( (ppKey==0)==(pnKey==0) );
+  assert( (ppVal==0)==(pnVal==0) );
+
+  if( ppKey ){
+    int nKey = 0;
+    switch( p->eType ){
+      case TEST_DATASOURCE_RANDOM: {
+        int nRange = (1 + p->nMaxKey - p->nMinKey);
+        nKey = (int)( testPrngValue((u32)iData) % nRange ) + p->nMinKey; 
+        testPrngString((u32)iData, p->aKey, nKey);
+        break;
+      }
+      case TEST_DATASOURCE_SEQUENCE:
+        nKey = sprintf(p->aKey, "%012d", iData);
+        break;
+    }
+    *ppKey = p->aKey;
+    *pnKey = nKey;
+  }
+  if( ppVal ){
+    u32 nVal = testPrngValue((u32)iData)%(1+p->nMaxVal-p->nMinVal)+p->nMinVal;
+    testPrngString((u32)~iData, p->aVal, (int)nVal);
+    *ppVal = p->aVal;
+    *pnVal = (int)nVal;
+  }
+}
+
+void testDatasourceFree(Datasource *p){
+  testFree(p);
+}
+
+/*
+** Return a pointer to a nul-terminated string that corresponds to the
+** contents of the datasource-definition passed as the first argument.
+** The caller should eventually free the returned pointer using testFree().
+*/
+char *testDatasourceName(const DatasourceDefn *p){
+  char *zRet;
+  zRet = testMallocPrintf("%s.(%d-%d).(%d-%d)",
+      (p->eType==TEST_DATASOURCE_SEQUENCE ? "seq" : "rnd"),
+      p->nMinKey, p->nMaxKey,
+      p->nMinVal, p->nMaxVal
+  );
+  return zRet;
+}
+
+Datasource *testDatasourceNew(const DatasourceDefn *pDefn){
+  Datasource *p;
+  int nMinKey; 
+  int nMaxKey;
+  int nMinVal;
+  int nMaxVal; 
+
+  if( pDefn->eType==TEST_DATASOURCE_SEQUENCE ){
+    nMinKey = 128;
+    nMaxKey = 128;
+  }else{
+    nMinKey = MAX(0, pDefn->nMinKey);
+    nMaxKey = MAX(nMinKey, pDefn->nMaxKey);
+  }
+  nMinVal = MAX(0, pDefn->nMinVal);
+  nMaxVal = MAX(nMinVal, pDefn->nMaxVal);
+
+  p = (Datasource *)testMalloc(sizeof(Datasource) + nMaxKey + nMaxVal + 1);
+  p->eType = pDefn->eType;
+  p->nMinKey = nMinKey;
+  p->nMinVal = nMinVal;
+  p->nMaxKey = nMaxKey;
+  p->nMaxVal = nMaxVal;
+  
+  p->aKey = (char *)&p[1];
+  p->aVal = &p->aKey[nMaxKey];
+  return p;
+};
diff --git a/ext/lsm1/lsm-test/lsmtest_func.c b/ext/lsm1/lsm-test/lsmtest_func.c
new file mode 100644
index 0000000..eb8346a
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_func.c
@@ -0,0 +1,177 @@
+
+#include "lsmtest.h"
+
+
+int do_work(int nArg, char **azArg){
+  struct Option {
+    const char *zName;
+  } aOpt [] = {
+    { "-nmerge" },
+    { "-nkb" },
+    { 0 }
+  };
+
+  lsm_db *pDb;
+  int rc;
+  int i;
+  const char *zDb;
+  int nMerge = 1;
+  int nKB = (1<<30);
+
+  if( nArg==0 ) goto usage;
+  zDb = azArg[nArg-1];
+  for(i=0; i<(nArg-1); i++){
+    int iSel;
+    rc = testArgSelect(aOpt, "option", azArg[i], &iSel);
+    if( rc ) return rc;
+    switch( iSel ){
+      case 0:
+        i++;
+        if( i==(nArg-1) ) goto usage;
+        nMerge = atoi(azArg[i]);
+        break;
+      case 1:
+        i++;
+        if( i==(nArg-1) ) goto usage;
+        nKB = atoi(azArg[i]);
+        break;
+    }
+  }
+
+  rc = lsm_new(0, &pDb);
+  if( rc!=LSM_OK ){
+    testPrintError("lsm_open(): rc=%d\n", rc);
+  }else{
+    rc = lsm_open(pDb, zDb);
+    if( rc!=LSM_OK ){
+      testPrintError("lsm_open(): rc=%d\n", rc);
+    }else{
+      int n = -1;
+      lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &n);
+      n = n*2;
+      lsm_config(pDb, LSM_CONFIG_AUTOCHECKPOINT, &n);
+
+      rc = lsm_work(pDb, nMerge, nKB, 0);
+      if( rc!=LSM_OK ){
+        testPrintError("lsm_work(): rc=%d\n", rc);
+      }
+    }
+  }
+  if( rc==LSM_OK ){
+    rc = lsm_checkpoint(pDb, 0);
+  }
+
+  lsm_close(pDb);
+  return rc;
+
+ usage:
+  testPrintUsage("?-optimize? ?-n N? DATABASE");
+  return -1;
+}
+
+
+/*
+**   lsmtest show ?-config LSM-CONFIG? DATABASE ?COMMAND ?PGNO??
+*/
+int do_show(int nArg, char **azArg){
+  lsm_db *pDb;
+  int rc;
+  const char *zDb;
+
+  int eOpt = LSM_INFO_DB_STRUCTURE;
+  unsigned int iPg = 0;
+  int bConfig = 0;
+  const char *zConfig = "";
+
+  struct Option {
+    const char *zName;
+    int bConfig;
+    int eOpt;
+  } aOpt [] = { 
+    { "array",       0, LSM_INFO_ARRAY_STRUCTURE },
+    { "array-pages", 0, LSM_INFO_ARRAY_PAGES },
+    { "blocksize",   1, LSM_CONFIG_BLOCK_SIZE },
+    { "pagesize",    1, LSM_CONFIG_PAGE_SIZE },
+    { "freelist",    0, LSM_INFO_FREELIST },
+    { "page-ascii",  0, LSM_INFO_PAGE_ASCII_DUMP },
+    { "page-hex",    0, LSM_INFO_PAGE_HEX_DUMP },
+    { 0, 0 } 
+  };
+
+  char *z = 0; 
+  int iDb = 0;                    /* Index of DATABASE in azArg[] */
+
+  /* Check if there is a "-config" option: */
+  if( nArg>2 && strlen(azArg[0])>1 
+   && memcmp(azArg[0], "-config", strlen(azArg[0]))==0
+  ){
+    zConfig = azArg[1];
+    iDb = 2;
+  }
+  if( nArg<(iDb+1) ) goto usage;
+
+  if( nArg>(iDb+1) ){
+    rc = testArgSelect(aOpt, "option", azArg[iDb+1], &eOpt);
+    if( rc!=0 ) return rc;
+    bConfig = aOpt[eOpt].bConfig;
+    eOpt = aOpt[eOpt].eOpt;
+    if( (bConfig==0 && eOpt==LSM_INFO_FREELIST)
+     || (bConfig==1 && eOpt==LSM_CONFIG_BLOCK_SIZE)
+     || (bConfig==1 && eOpt==LSM_CONFIG_PAGE_SIZE)
+    ){
+      if( nArg!=(iDb+2) ) goto usage;
+    }else{
+      if( nArg!=(iDb+3) ) goto usage;
+      iPg = atoi(azArg[iDb+2]);
+    }
+  }
+  zDb = azArg[iDb];
+
+  rc = lsm_new(0, &pDb);
+  tdb_lsm_configure(pDb, zConfig);
+  if( rc!=LSM_OK ){
+    testPrintError("lsm_new(): rc=%d\n", rc);
+  }else{
+    rc = lsm_open(pDb, zDb);
+    if( rc!=LSM_OK ){
+      testPrintError("lsm_open(): rc=%d\n", rc);
+    }
+  }
+
+  if( rc==LSM_OK ){
+    if( bConfig==0 ){
+      switch( eOpt ){
+        case LSM_INFO_DB_STRUCTURE:
+        case LSM_INFO_FREELIST:
+          rc = lsm_info(pDb, eOpt, &z);
+          break;
+        case LSM_INFO_ARRAY_STRUCTURE:
+        case LSM_INFO_ARRAY_PAGES:
+        case LSM_INFO_PAGE_ASCII_DUMP:
+        case LSM_INFO_PAGE_HEX_DUMP:
+          rc = lsm_info(pDb, eOpt, iPg, &z);
+          break;
+        default:
+          assert( !"no chance" );
+      }
+
+      if( rc==LSM_OK ){
+        printf("%s\n", z ? z : "");
+        fflush(stdout);
+      }
+      lsm_free(lsm_get_env(pDb), z);
+    }else{
+      int iRes = -1;
+      lsm_config(pDb, eOpt, &iRes);
+      printf("%d\n", iRes);
+      fflush(stdout);
+    }
+  }
+
+  lsm_close(pDb);
+  return rc;
+
+ usage:
+  testPrintUsage("DATABASE ?array|page-ascii|page-hex PGNO?");
+  return -1;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_io.c b/ext/lsm1/lsm-test/lsmtest_io.c
new file mode 100644
index 0000000..7aa5d10
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_io.c
@@ -0,0 +1,248 @@
+
+/*
+** SUMMARY
+**
+**   This file implements the 'io' subcommand of the test program. It is used
+**   for testing the performance of various combinations of write() and fsync()
+**   system calls. All operations occur on a single file, which may or may not
+**   exist when a test is started.
+**
+**   A test consists of a series of commands. Each command is either a write
+**   or an fsync. A write is specified as "<amount>@<offset>", where <amount>
+**   is the amount of data written, and <offset> is the offset of the file
+**   to write to. An <amount> or an <offset> is specified as an integer number
+**   of bytes. Or, if postfixed with a "K", "M" or "G", an integer number of
+**   KB, MB or GB, respectively. An fsync is simply "S". All commands are
+**   case-insensitive.
+**
+**   Example test program:
+**
+**        2M@6M 1492K@4M S 4096@4K S
+**
+**   This program writes 2 MB of data starting at the offset 6MB offset of
+**   the file, followed by 1492 KB of data written at the 4MB offset of the
+**   file, followed by a call to fsync(), a write of 4KB of data at byte
+**   offset 4096, and finally another call to fsync().
+**
+**   Commands may either be specified on the command line (one command per
+**   command line argument) or read from stdin. Commands read from stdin
+**   must be separated by white-space.
+**
+** COMMAND LINE INVOCATION
+**
+**   The sub-command implemented in this file must be invoked with at least
+**   two arguments - the path to the file to write to and the page-size to
+**   use for writing. If there are more than two arguments, then each
+**   subsequent argument is assumed to be a test command. If there are exactly
+**   two arguments, the test commands are read from stdin.
+**
+**   A write command does not result in a single call to system call write().
+**   Instead, the specified region is written sequentially using one or
+**   more calls to write(), each of which writes not more than one page of
+**   data. For example, if the page-size is 4KB, the command "2M@6M" results
+**   in 512 calls to write(), each of which writes 4KB of data.
+**
+** EXAMPLES
+**
+**   Two equivalent examples:
+**
+**     $ lsmtest io testfile.db 4KB 2M@6M 1492K@4M S 4096@4K S
+**     3544K written in 129 ms
+**     $ echo "2M@6M 1492K@4M S 4096@4K S" | lsmtest io testfile.db 4096 
+**     3544K written in 127 ms
+**
+*/
+
+#include "lsmtest.h"
+
+typedef struct IoContext IoContext;
+
+struct IoContext {
+  int fd;
+  int nWrite;
+};
+
+/*
+** As isspace(3)
+*/
+static int safe_isspace(char c){
+  if( c&0x80) return 0;
+  return isspace(c);
+}
+
+/*
+** As isdigit(3)
+*/
+static int safe_isdigit(char c){
+  if( c&0x80) return 0;
+  return isdigit(c);
+}
+
+static i64 getNextSize(char *zIn, char **pzOut, int *pRc){
+  i64 iRet = 0;
+  if( *pRc==0 ){
+    char *z = zIn;
+
+    if( !safe_isdigit(*z) ){
+      *pRc = 1;
+      return 0;
+    }
+
+    /* Process digits */
+    while( safe_isdigit(*z) ){
+      iRet = iRet*10 + (*z - '0');
+      z++;
+    }
+
+    /* Process suffix */
+    switch( *z ){
+      case 'k': case 'K':
+        iRet = iRet * 1024;
+        z++;
+        break;
+
+      case 'm': case 'M':
+        iRet = iRet * 1024 * 1024;
+        z++;
+        break;
+
+      case 'g': case 'G':
+        iRet = iRet * 1024 * 1024 * 1024;
+        z++;
+        break;
+    }
+
+    if( pzOut ) *pzOut = z;
+  }
+  return iRet;
+}
+
+static int doOneCmd(
+  IoContext *pCtx,
+  u8 *aData,
+  int pgsz,
+  char *zCmd,
+  char **pzOut
+){
+  char c;
+  char *z = zCmd;
+
+  while( safe_isspace(*z) ) z++;
+  c = *z;
+
+  if( c==0 ){
+    if( pzOut ) *pzOut = z;
+    return 0;
+  }
+
+  if( c=='s' || c=='S' ){
+    if( pzOut ) *pzOut = &z[1];
+    return fdatasync(pCtx->fd);
+  }
+
+  if( safe_isdigit(c) ){
+    i64 iOff = 0;
+    int nByte = 0;
+    int rc = 0;
+    int nPg;
+    int iPg;
+
+    nByte = (int)getNextSize(z, &z, &rc);
+    if( rc || *z!='@' ) goto bad_command;
+    z++;
+    iOff = getNextSize(z, &z, &rc);
+    if( rc || (safe_isspace(*z)==0 && *z!='\0') ) goto bad_command;
+    if( pzOut ) *pzOut = z;
+
+    nPg = (nByte+pgsz-1) / pgsz;
+    lseek(pCtx->fd, (off_t)iOff, SEEK_SET);
+    for(iPg=0; iPg<nPg; iPg++){
+      write(pCtx->fd, aData, pgsz);
+    }
+    pCtx->nWrite += nByte/1024;
+
+    return 0;
+  }
+
+ bad_command:
+  testPrintError("unrecognized command: %s", zCmd);
+  return 1;
+}
+
+static int readStdin(char **pzOut){
+  int nAlloc = 128;
+  char *zOut = 0;
+  int nOut = 0;
+
+  while( !feof(stdin) ){
+    int nRead;
+
+    nAlloc = nAlloc*2;
+    zOut = realloc(zOut, nAlloc);
+    nRead = fread(&zOut[nOut], 1, nAlloc-nOut-1, stdin);
+
+    if( nRead==0 ) break;
+    nOut += nRead;
+    zOut[nOut] = '\0';
+  }
+
+  *pzOut = zOut;
+  return 0;
+}
+
+int do_io(int nArg, char **azArg){
+  IoContext ctx;
+  int pgsz;
+  char *zFile;
+  char *zPgsz;
+  int i;
+  int rc = 0;
+
+  char *zStdin = 0;
+  char *z;
+
+  u8 *aData;
+
+  memset(&ctx, 0, sizeof(IoContext));
+  if( nArg<2 ){
+    testPrintUsage("FILE PGSZ ?CMD-1 ...?");
+    return -1;
+  }
+  zFile = azArg[0];
+  zPgsz = azArg[1];
+
+  pgsz = (int)getNextSize(zPgsz, 0, &rc);
+  if( pgsz<=0 ){
+    testPrintError("Ridiculous page size: %d", pgsz);
+    return -1;
+  }
+  aData = malloc(pgsz);
+  memset(aData, 0x77, pgsz);
+
+  ctx.fd = open(zFile, O_RDWR|O_CREAT|_O_BINARY, 0644);
+  if( ctx.fd<0 ){
+    perror("open: ");
+    return -1;
+  }
+
+  if( nArg==2 ){
+    readStdin(&zStdin);
+    testTimeInit();
+    z = zStdin;
+    while( *z && rc==0 ){
+      rc = doOneCmd(&ctx, aData, pgsz, z, &z);
+    }
+  }else{
+    testTimeInit();
+    for(i=2; i<nArg; i++){
+      rc = doOneCmd(&ctx, aData, pgsz, azArg[i], 0);
+    }
+  }
+
+  printf("%dK written in %d ms\n", ctx.nWrite, testTimeGet());
+
+  free(zStdin);
+  close(ctx.fd);
+
+  return 0;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_main.c b/ext/lsm1/lsm-test/lsmtest_main.c
new file mode 100644
index 0000000..f4a3ac0
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_main.c
@@ -0,0 +1,1548 @@
+
+#include "lsmtest.h"
+#include <sqlite3.h>
+
+void test_failed(){ 
+  assert( 0 );
+  return; 
+}
+
+#define testSetError(rc) testSetErrorFunc(rc, pRc, __FILE__, __LINE__)
+static void testSetErrorFunc(int rc, int *pRc, const char *zFile, int iLine){
+  if( rc ){
+    *pRc = rc;
+    fprintf(stderr, "FAILED (%s:%d) rc=%d ", zFile, iLine, rc);
+    test_failed();
+  }
+}
+
+static int lsm_memcmp(u8 *a, u8 *b, int c){
+  int i;
+  for(i=0; i<c; i++){
+    if( a[i]!=b[i] ) return a[i] - b[i];
+  }
+  return 0;
+}
+
+/*
+** A test utility function.
+*/
+void testFetch(
+  TestDb *pDb,                    /* Database handle */
+  void *pKey, int nKey,           /* Key to query database for */
+  void *pVal, int nVal,           /* Expected value */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  if( *pRc==0 ){
+    void *pDbVal;
+    int nDbVal;
+    int rc;
+
+    static int nCall = 0; nCall++;
+
+    rc = tdb_fetch(pDb, pKey, nKey, &pDbVal, &nDbVal);
+    testSetError(rc);
+    if( rc==0 && (nVal!=nDbVal || (nVal>0 && lsm_memcmp(pVal, pDbVal, nVal))) ){
+      testSetError(1);
+    }
+  }
+}
+
+void testWrite(
+  TestDb *pDb,                    /* Database handle */
+  void *pKey, int nKey,           /* Key to query database for */
+  void *pVal, int nVal,           /* Value to write */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  if( *pRc==0 ){
+    int rc;
+static int nCall = 0;
+nCall++;
+    rc = tdb_write(pDb, pKey, nKey, pVal, nVal);
+    testSetError(rc);
+  }
+}
+void testDelete(
+  TestDb *pDb,                    /* Database handle */
+  void *pKey, int nKey,           /* Key to query database for */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  if( *pRc==0 ){
+    int rc;
+    *pRc = rc = tdb_delete(pDb, pKey, nKey);
+    testSetError(rc);
+  }
+}
+void testDeleteRange(
+  TestDb *pDb,                    /* Database handle */
+  void *pKey1, int nKey1,
+  void *pKey2, int nKey2,
+  int *pRc                        /* IN/OUT: Error code */
+){
+  if( *pRc==0 ){
+    int rc;
+    *pRc = rc = tdb_delete_range(pDb, pKey1, nKey1, pKey2, nKey2);
+    testSetError(rc);
+  }
+}
+
+void testBegin(TestDb *pDb, int iTrans, int *pRc){
+  if( *pRc==0 ){
+    int rc;
+    rc = tdb_begin(pDb, iTrans);
+    testSetError(rc);
+  }
+}
+void testCommit(TestDb *pDb, int iTrans, int *pRc){
+  if( *pRc==0 ){
+    int rc;
+    rc = tdb_commit(pDb, iTrans);
+    testSetError(rc);
+  }
+}
+#if 0 /* unused */
+static void testRollback(TestDb *pDb, int iTrans, int *pRc){
+  if( *pRc==0 ){
+    int rc;
+    rc = tdb_rollback(pDb, iTrans);
+    testSetError(rc);
+  }
+}
+#endif
+
+void testWriteStr(
+  TestDb *pDb,                    /* Database handle */
+  const char *zKey,               /* Key to query database for */
+  const char *zVal,               /* Value to write */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int nVal = (zVal ? strlen(zVal) : 0);
+  testWrite(pDb, (void *)zKey, strlen(zKey), (void *)zVal, nVal, pRc);
+}
+
+#if 0 /* unused */
+static void testDeleteStr(TestDb *pDb, const char *zKey, int *pRc){
+  testDelete(pDb, (void *)zKey, strlen(zKey), pRc);
+}
+#endif
+void testFetchStr(
+  TestDb *pDb,                    /* Database handle */
+  const char *zKey,               /* Key to query database for */
+  const char *zVal,               /* Value to write */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int nVal = (zVal ? strlen(zVal) : 0);
+  testFetch(pDb, (void *)zKey, strlen(zKey), (void *)zVal, nVal, pRc);
+}
+
+void testFetchCompare(
+  TestDb *pControl, 
+  TestDb *pDb, 
+  void *pKey, int nKey, 
+  int *pRc
+){
+  int rc;
+  void *pDbVal1;
+  void *pDbVal2;
+  int nDbVal1;
+  int nDbVal2;
+
+  static int nCall = 0;
+  nCall++;
+
+  rc = tdb_fetch(pControl, pKey, nKey, &pDbVal1, &nDbVal1);
+  testSetError(rc);
+
+  rc = tdb_fetch(pDb, pKey, nKey, &pDbVal2, &nDbVal2);
+  testSetError(rc);
+
+  if( *pRc==0 
+   && (nDbVal1!=nDbVal2 || (nDbVal1>0 && memcmp(pDbVal1, pDbVal2, nDbVal1)))
+  ){
+    testSetError(1);
+  }
+}
+
+typedef struct ScanResult ScanResult;
+struct ScanResult {
+  TestDb *pDb;
+
+  int nRow;
+  u32 cksum1;
+  u32 cksum2;
+  void *pKey1; int nKey1;
+  void *pKey2; int nKey2;
+
+  int bReverse;
+  int nPrevKey;
+  u8 aPrevKey[256];
+};
+
+static int keyCompare(void *pKey1, int nKey1, void *pKey2, int nKey2){
+  int res;
+  res = memcmp(pKey1, pKey2, MIN(nKey1, nKey2));
+  if( res==0 ){
+    res = nKey1 - nKey2;
+  }
+  return res;
+}
+
+int test_scan_debug = 0;
+
+static void scanCompareCb(
+  void *pCtx, 
+  void *pKey, int nKey,
+  void *pVal, int nVal
+){
+  ScanResult *p = (ScanResult *)pCtx;
+  u8 *aKey = (u8 *)pKey;
+  u8 *aVal = (u8 *)pVal;
+  int i;
+
+  if( test_scan_debug ){
+    printf("%d: %.*s\n", p->nRow, nKey, (char *)pKey);
+    fflush(stdout);
+  }
+#if 0
+  if( test_scan_debug ) printf("%.20s\n", (char *)pVal);
+#endif
+
+#if 0
+  /* Check tdb_fetch() matches */
+  int rc = 0;
+  testFetch(p->pDb, pKey, nKey, pVal, nVal, &rc);
+  assert( rc==0 );
+#endif
+
+  /* Update the checksum data */
+  p->nRow++;
+  for(i=0; i<nKey; i++){
+    p->cksum1 += ((int)aKey[i] << (i&0x0F));
+    p->cksum2 += p->cksum1;
+  }
+  for(i=0; i<nVal; i++){
+    p->cksum1 += ((int)aVal[i] << (i&0x0F));
+    p->cksum2 += p->cksum1;
+  }
+
+  /* Check that the delivered row is not out of order. */
+  if( nKey<(int)sizeof(p->aPrevKey) ){
+    if( p->nPrevKey ){
+      int res = keyCompare(p->aPrevKey, p->nPrevKey, pKey, nKey);
+      if( (res<0 && p->bReverse) || (res>0 && p->bReverse==0) ){
+        testPrintError("Returned key out of order at %s:%d\n", 
+            __FILE__, __LINE__
+        );
+      }
+    }
+
+    p->nPrevKey = nKey;
+    memcpy(p->aPrevKey, pKey, MIN(p->nPrevKey, nKey));
+  }
+
+  /* Check that the delivered row is within range. */
+  if( p->pKey1 && (
+      (memcmp(p->pKey1, pKey, MIN(p->nKey1, nKey))>0)
+   || (memcmp(p->pKey1, pKey, MIN(p->nKey1, nKey))==0 && p->nKey1>nKey)
+  )){
+    testPrintError("Returned key too small at %s:%d\n", __FILE__, __LINE__);
+  }
+  if( p->pKey2 && (
+      (memcmp(p->pKey2, pKey, MIN(p->nKey2, nKey))<0)
+   || (memcmp(p->pKey2, pKey, MIN(p->nKey2, nKey))==0 && p->nKey2<nKey)
+  )){
+    testPrintError("Returned key too large at %s:%d\n", __FILE__, __LINE__);
+  }
+
+}
+
+/*
+** Scan the contents of the two databases. Check that they match.
+*/
+void testScanCompare(
+  TestDb *pDb1,                   /* Control (trusted) database */
+  TestDb *pDb2,                   /* Database being tested */
+  int bReverse,
+  void *pKey1, int nKey1, 
+  void *pKey2, int nKey2, 
+  int *pRc
+){
+  static int nCall = 0; nCall++;
+  if( *pRc==0 ){
+    ScanResult res1;
+    ScanResult res2;
+    void *pRes1 = (void *)&res1;
+    void *pRes2 = (void *)&res2;
+
+    memset(&res1, 0, sizeof(ScanResult));
+    memset(&res2, 0, sizeof(ScanResult));
+
+    res1.pDb = pDb1;
+    res1.nKey1 = nKey1; res1.pKey1 = pKey1;
+    res1.nKey2 = nKey2; res1.pKey2 = pKey2;
+    res1.bReverse = bReverse;
+    res2.pDb = pDb2;
+    res2.nKey1 = nKey1; res2.pKey1 = pKey1;
+    res2.nKey2 = nKey2; res2.pKey2 = pKey2;
+    res2.bReverse = bReverse;
+
+    tdb_scan(pDb1, pRes1, bReverse, pKey1, nKey1, pKey2, nKey2, scanCompareCb);
+if( test_scan_debug ) printf("\n\n\n");
+    tdb_scan(pDb2, pRes2, bReverse, pKey1, nKey1, pKey2, nKey2, scanCompareCb);
+if( test_scan_debug ) printf("\n\n\n");
+
+    if( res1.nRow!=res2.nRow 
+     || res1.cksum1!=res2.cksum1 
+     || res1.cksum2!=res2.cksum2
+    ){
+      printf("expected: %d %X %X\n", res1.nRow, res1.cksum1, res1.cksum2);
+      printf("got:      %d %X %X\n", res2.nRow, res2.cksum1, res2.cksum2);
+      testSetError(1);
+      *pRc = 1;
+    }
+  }
+}
+
+void testClose(TestDb **ppDb){
+  tdb_close(*ppDb);
+  *ppDb = 0;
+}
+
+TestDb *testOpen(const char *zSystem, int bClear, int *pRc){
+  TestDb *pDb = 0;
+  if( *pRc==0 ){
+    int rc;
+    rc = tdb_open(zSystem, 0, bClear, &pDb);
+    if( rc!=0 ){
+      testSetError(rc);
+      *pRc = rc;
+    }
+  }
+  return pDb;
+}
+
+void testReopen(TestDb **ppDb, int *pRc){
+  if( *pRc==0 ){
+    const char *zLib;
+    zLib = tdb_library_name(*ppDb);
+    testClose(ppDb);
+    *pRc = tdb_open(zLib, 0, 0, ppDb);
+  }
+}
+
+
+#if 0 /* unused */
+static void testSystemSelect(const char *zSys, int *piSel, int *pRc){
+  if( *pRc==0 ){
+    struct SysName { const char *zName; } *aName;
+    int nSys;
+    int i;
+
+    for(nSys=0; tdb_system_name(nSys); nSys++);
+    aName = malloc(sizeof(struct SysName) * (nSys+1));
+    for(i=0; i<=nSys; i++){
+      aName[i].zName = tdb_system_name(i);
+    }
+
+    *pRc = testArgSelect(aName, "db", zSys, piSel);
+    free(aName);
+  }
+}
+#endif
+
+char *testMallocVPrintf(const char *zFormat, va_list ap){
+  int nByte;
+  va_list copy;
+  char *zRet;
+
+  __va_copy(copy, ap);
+  nByte = vsnprintf(0, 0, zFormat, copy);
+  va_end(copy);
+
+  assert( nByte>=0 );
+  zRet = (char *)testMalloc(nByte+1);
+  vsnprintf(zRet, nByte+1, zFormat, ap);
+  return zRet;
+}
+
+char *testMallocPrintf(const char *zFormat, ...){
+  va_list ap;
+  char *zRet;
+
+  va_start(ap, zFormat);
+  zRet = testMallocVPrintf(zFormat, ap);
+  va_end(ap);
+
+  return zRet;
+}
+
+
+/*
+** A wrapper around malloc(3).
+**
+** This function should be used for all allocations made by test procedures.
+** It has the following properties:
+**
+**   * Test code may assume that allocations may not fail.
+**   * Returned memory is always zeroed.
+**
+** Allocations made using testMalloc() should be freed using testFree().
+*/
+void *testMalloc(int n){
+  u8 *p = (u8*)malloc(n + 8);
+  memset(p, 0, n+8);
+  *(int*)p = n;
+  return (void*)&p[8];
+}
+
+void *testMallocCopy(void *pCopy, int nByte){
+  void *pRet = testMalloc(nByte);
+  memcpy(pRet, pCopy, nByte);
+  return pRet;
+}
+
+void *testRealloc(void *ptr, int n){
+  if( ptr ){
+    u8 *p = (u8*)ptr - 8;
+    int nOrig =  *(int*)p;
+    p = (u8*)realloc(p, n+8);
+    if( nOrig<n ){
+      memset(&p[8+nOrig], 0, n-nOrig);
+    }
+    *(int*)p = n;
+    return (void*)&p[8];
+  }
+  return testMalloc(n);
+}
+
+/*
+** Free an allocation made by an earlier call to testMalloc().
+*/
+void testFree(void *ptr){
+  if( ptr ){
+    u8 *p = (u8*)ptr - 8;
+    memset(p, 0x55, *(int*)p + 8);
+    free(p);
+  }
+}
+
+/*
+** String zPattern contains a glob pattern. Return true if zStr matches 
+** the pattern, or false if it does not.
+*/
+int testGlobMatch(const char *zPattern, const char *zStr){
+  int i = 0;
+  int j = 0;
+
+  while( zPattern[i] ){
+    char p = zPattern[i];
+
+    if( p=='*' || p=='%' ){
+      do {
+        if( testGlobMatch(&zPattern[i+1], &zStr[j]) ) return 1;
+      }while( zStr[j++] );
+      return 0;
+    }
+
+    if( zStr[j]==0 || (p!='?' && p!=zStr[j]) ){
+      /* Match failed. */
+      return 0;
+    }
+
+    j++;
+    i++;
+  }
+
+  return (zPattern[i]==0 && zStr[j]==0);
+}
+
+/* 
+** End of test utilities 
+**************************************************************************/
+
+int do_test(int nArg, char **azArg){
+  int j;
+  int rc;
+  int nFail = 0;
+  const char *zPattern = 0;
+
+  if( nArg>1 ){
+    testPrintError("Usage: test ?PATTERN?\n");
+    return 1;
+  }
+  if( nArg==1 ){
+    zPattern = azArg[0];
+  }
+
+  for(j=0; tdb_system_name(j); j++){
+    rc = 0;
+
+    test_data_1(tdb_system_name(j), zPattern, &rc);
+    test_data_2(tdb_system_name(j), zPattern, &rc);
+    test_data_3(tdb_system_name(j), zPattern, &rc);
+    test_data_4(tdb_system_name(j), zPattern, &rc);
+    test_rollback(tdb_system_name(j), zPattern, &rc);
+    test_mc(tdb_system_name(j), zPattern, &rc);
+    test_mt(tdb_system_name(j), zPattern, &rc);
+
+    if( rc ) nFail++;
+  }
+
+  rc = 0;
+  test_oom(zPattern, &rc);
+  if( rc ) nFail++;
+
+  rc = 0;
+  test_api(zPattern, &rc);
+  if( rc ) nFail++;
+
+  rc = 0;
+  do_crash_test(zPattern, &rc);
+  if( rc ) nFail++;
+
+  rc = 0;
+  do_writer_crash_test(zPattern, &rc);
+  if( rc ) nFail++;
+
+  return (nFail!=0);
+}
+
+static lsm_db *configure_lsm_db(TestDb *pDb){
+  lsm_db *pLsm;
+  pLsm = tdb_lsm(pDb);
+  if( pLsm ){
+    tdb_lsm_config_str(pDb, "mmap=1 autowork=1 automerge=4 worker_automerge=4");
+  }
+  return pLsm;
+}
+
+typedef struct WriteHookEvent WriteHookEvent;
+struct WriteHookEvent {
+  i64 iOff;
+  int nData;
+  int nUs;
+};
+WriteHookEvent prev = {0, 0, 0};
+
+static void flushPrev(FILE *pOut){
+  if( prev.nData ){
+    fprintf(pOut, "w %s %lld %d %d\n", "d", prev.iOff, prev.nData, prev.nUs);
+    prev.nData = 0;
+  }
+}
+
+#if 0 /* unused */
+static void do_speed_write_hook2(
+  void *pCtx,
+  int bLog,
+  i64 iOff,
+  int nData,
+  int nUs
+){
+  FILE *pOut = (FILE *)pCtx;
+  if( bLog ) return;
+
+  if( prev.nData && nData && iOff==prev.iOff+prev.nData ){
+    prev.nData += nData;
+    prev.nUs += nUs;
+  }else{
+    flushPrev(pOut);
+    if( nData==0 ){
+      fprintf(pOut, "s %s 0 0 %d\n", (bLog ? "l" : "d"), nUs);
+    }else{
+      prev.iOff = iOff;
+      prev.nData = nData;
+      prev.nUs = nUs;
+    }
+  }
+}
+#endif
+
+#define ST_REPEAT  0
+#define ST_WRITE   1
+#define ST_PAUSE   2
+#define ST_FETCH   3
+#define ST_SCAN    4
+#define ST_NSCAN   5
+#define ST_KEYSIZE 6
+#define ST_VALSIZE 7
+#define ST_TRANS   8
+
+
+static void print_speed_test_help(){
+  printf(
+"\n"
+"Repeat the following $repeat times:\n"
+"  1. Insert $write key-value pairs. One transaction for each write op.\n"
+"  2. Pause for $pause ms.\n"
+"  3. Perform $fetch queries on the database.\n"
+"\n"
+"  Keys are $keysize bytes in size. Values are $valsize bytes in size\n"
+"  Both keys and values are pseudo-randomly generated\n"
+"\n"
+"Options are:\n"
+"  -repeat  $repeat                 (default value 10)\n"
+"  -write   $write                  (default value 10000)\n"
+"  -pause   $pause                  (default value 0)\n"
+"  -fetch   $fetch                  (default value 0)\n"
+"  -keysize $keysize                (default value 12)\n"
+"  -valsize $valsize                (default value 100)\n"
+"  -system  $system                 (default value \"lsm\")\n"
+"  -trans   $trans                  (default value 0)\n"
+"\n"
+);
+}
+
+int do_speed_test2(int nArg, char **azArg){
+  struct Option {
+    const char *zOpt;
+    int eVal;
+    int iDefault;
+  } aOpt[] = {
+    { "-repeat",  ST_REPEAT,    10},
+    { "-write",   ST_WRITE,  10000},
+    { "-pause",   ST_PAUSE,      0},
+    { "-fetch",   ST_FETCH,      0},
+    { "-scan",    ST_SCAN,       0},
+    { "-nscan",   ST_NSCAN,      0},
+    { "-keysize", ST_KEYSIZE,   12},
+    { "-valsize", ST_VALSIZE,  100},
+    { "-trans",   ST_TRANS,      0},
+    { "-system",  -1,            0},
+    { "help",     -2,            0},
+    {0, 0, 0}
+  };
+  int i;
+  int aParam[9];
+  int rc = 0;
+  int bReadonly = 0;
+  int nContent = 0;
+
+  TestDb *pDb;
+  Datasource *pData;
+  DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 0, 0, 0, 0 };
+  char *zSystem = "";
+  int bLsm = 1;
+  FILE *pLog = 0;
+
+#ifdef NDEBUG
+  /* If NDEBUG is defined, disable the dynamic memory related checks in
+  ** lsmtest_mem.c. They slow things down.  */
+  testMallocUninstall(tdb_lsm_env());
+#endif
+
+  /* Initialize aParam[] with default values. */
+  for(i=0; i<ArraySize(aOpt); i++){
+    if( aOpt[i].zOpt ) aParam[aOpt[i].eVal] = aOpt[i].iDefault;
+  }
+
+  /* Process the command line switches. */
+  for(i=0; i<nArg; i+=2){
+    int iSel;
+    rc = testArgSelect(aOpt, "switch", azArg[i], &iSel);
+    if( rc ){
+      return rc;
+    }
+    if( aOpt[iSel].eVal==-2 ){
+      print_speed_test_help();
+      return 0;
+    }
+    if( i+1==nArg ){
+      testPrintError("option %s requires an argument\n", aOpt[iSel].zOpt);
+      return 1;
+    }
+    if( aOpt[iSel].eVal>=0 ){
+      aParam[aOpt[iSel].eVal] = atoi(azArg[i+1]);
+    }else{
+      zSystem = azArg[i+1];
+      bLsm = 0;
+#if 0
+      for(j=0; zSystem[j]; j++){
+        if( zSystem[j]=='=' ) bLsm = 1;
+      }
+#endif
+    }
+  }
+  
+  printf("#");
+  for(i=0; i<ArraySize(aOpt); i++){
+    if( aOpt[i].zOpt ){
+      if( aOpt[i].eVal>=0 ){
+        printf(" %s=%d", &aOpt[i].zOpt[1], aParam[aOpt[i].eVal]);
+      }else if( aOpt[i].eVal==-1 ){
+        printf(" %s=\"%s\"", &aOpt[i].zOpt[1], zSystem);
+      }
+    }
+  }
+  printf("\n");
+
+  defn.nMinKey = defn.nMaxKey = aParam[ST_KEYSIZE];
+  defn.nMinVal = defn.nMaxVal = aParam[ST_VALSIZE];
+  pData = testDatasourceNew(&defn);
+
+  if( aParam[ST_WRITE]==0 ){
+    bReadonly = 1;
+  }
+
+  if( bLsm ){
+    rc = tdb_lsm_open(zSystem, "testdb.lsm", !bReadonly, &pDb);
+  }else{
+    pDb = testOpen(zSystem, !bReadonly, &rc);
+  }
+  if( rc!=0 ) return rc;
+  if( bReadonly ){
+    nContent = testCountDatabase(pDb);
+  }
+
+#if 0
+  pLog = fopen("/tmp/speed.log", "w");
+  tdb_lsm_write_hook(pDb, do_speed_write_hook2, (void *)pLog);
+#endif
+
+  for(i=0; i<aParam[ST_REPEAT] && rc==0; i++){
+    int msWrite, msFetch;
+    int iFetch;
+    int nWrite = aParam[ST_WRITE];
+
+    if( bReadonly ){
+      msWrite = 0;
+    }else{
+      testTimeInit();
+
+      if( aParam[ST_TRANS] ) testBegin(pDb, 2, &rc);
+      testWriteDatasourceRange(pDb, pData, i*nWrite, nWrite, &rc);
+      if( aParam[ST_TRANS] ) testCommit(pDb, 0, &rc);
+
+      msWrite = testTimeGet();
+      nContent += nWrite;
+    }
+
+    if( aParam[ST_PAUSE] ){
+      if( aParam[ST_PAUSE]/1000 ) sleep(aParam[ST_PAUSE]/1000);
+      if( aParam[ST_PAUSE]%1000 ) usleep(1000 * (aParam[ST_PAUSE]%1000));
+    }
+
+    if( aParam[ST_FETCH] ){
+      testTimeInit();
+      if( aParam[ST_TRANS] ) testBegin(pDb, 1, &rc);
+      for(iFetch=0; iFetch<aParam[ST_FETCH]; iFetch++){
+        int iKey = testPrngValue(i*nWrite+iFetch) % nContent;
+#ifndef NDEBUG
+        testDatasourceFetch(pDb, pData, iKey, &rc);
+#else
+        void *pKey; int nKey;           /* Database key to query for */
+        void *pVal; int nVal;           /* Result of query */
+
+        testDatasourceEntry(pData, iKey, &pKey, &nKey, 0, 0);
+        rc = tdb_fetch(pDb, pKey, nKey, &pVal, &nVal);
+        if( rc==0 && nVal<0 ) rc = 1;
+        if( rc ) break;
+#endif
+      }
+      if( aParam[ST_TRANS] ) testCommit(pDb, 0, &rc);
+      msFetch = testTimeGet();
+    }else{
+      msFetch = 0;
+    }
+
+    if( i==(aParam[ST_REPEAT]-1) ){
+      testTimeInit();
+      testClose(&pDb);
+      msWrite += testTimeGet();
+    }
+
+    printf("%d %d %d\n", i, msWrite, msFetch);
+    fflush(stdout);
+  }
+
+  testClose(&pDb);
+  testDatasourceFree(pData);
+
+  if( pLog ){
+    flushPrev(pLog);
+    fclose(pLog);
+  }
+  return rc;
+}
+
+int do_speed_tests(int nArg, char **azArg){
+
+  struct DbSystem {
+    const char *zLibrary;
+    const char *zColor;
+  } aSys[] = {
+    { "sqlite3",      "black" },
+    { "leveldb",      "blue" },
+    { "lsm",          "red" },
+    { "lsm_mt2",      "orange" },
+    { "lsm_mt3",      "purple" },
+    { "kyotocabinet", "green" },
+    {0, 0}
+  };
+
+  int i;
+  int j;
+  int rc;
+  int nSleep = 0;                 /* ms of rest allowed between INSERT tests */
+  int nRow = 0;                   /* Number of rows to insert into database */
+  int nStep;                      /* Measure INSERT time after this many rows */
+  int nSelStep;                   /* Measure SELECT time after this many rows */
+  int nSelTest;                   /* Number of SELECTs to run for timing */
+  int doReadTest = 1;
+  int doWriteTest = 1;
+
+  int *aTime;                     /* INSERT timing data */
+  int *aWrite;                    /* Writes per nStep inserts */
+  int *aSelTime;                  /* SELECT timing data */
+  int isFirst = 1;
+  int bSleep = 0;
+
+  /* File to write gnuplot script to. */
+  const char *zOut = "lsmtest_speed.gnuplot";
+
+  u32 sys_mask = 0;
+
+  testMallocUninstall(tdb_lsm_env());
+
+  for(i=0; i<nArg; i++){
+    struct Opt { 
+      const char *zOpt; 
+      int isSwitch;
+    } aOpt[] = {
+      { "sqlite3" , 0},
+      { "leveldb" , 0},
+      { "lsm" , 0},
+      { "lsm_mt2" , 0},
+      { "lsm_mt3" , 0},
+      { "kyotocabinet" , 0},
+      { "-rows"     , 1},
+      { "-sleep"    , 2},
+      { "-testmode" , 3},
+      { "-out"      , 4},
+      { 0, 0}
+    };
+    int iSel;
+
+    rc = testArgSelect(aOpt, "argument", azArg[i], &iSel);
+    if( rc ) return rc;
+
+    if( aOpt[iSel].isSwitch ){
+      i++;
+
+      if( i>=nArg ){
+        testPrintError("option %s requires an argument\n", aOpt[iSel].zOpt);
+        return 1;
+      }
+      if( aOpt[iSel].isSwitch==1 ){
+        nRow = atoi(azArg[i]);
+      }
+      if( aOpt[iSel].isSwitch==2 ){
+        nSleep = atoi(azArg[i]);
+      }
+      if( aOpt[iSel].isSwitch==3 ){
+        struct Mode {
+          const char *zMode;
+          int doReadTest;
+          int doWriteTest;
+        } aMode[] = {{"ro", 1, 0} , {"rw", 1, 1}, {"wo", 0, 1}, {0, 0, 0}};
+        int iMode;
+        rc = testArgSelect(aMode, "option", azArg[i], &iMode);
+        if( rc ) return rc;
+        doReadTest = aMode[iMode].doReadTest;
+        doWriteTest = aMode[iMode].doWriteTest;
+      }
+      if( aOpt[iSel].isSwitch==4 ){
+        /* The "-out FILE" switch. This option is used to specify a file to
+        ** write the gnuplot script to. */
+        zOut = azArg[i];
+      }
+    }else{
+      /* A db name */
+      rc = testArgSelect(aOpt, "system", azArg[i], &iSel);
+      if( rc ) return rc;
+      sys_mask |= (1<<iSel);
+    }
+  }
+
+  if( sys_mask==0 ) sys_mask = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3);
+  nRow = MAX(nRow, 100000);
+  nStep = nRow/100;
+  nSelStep = nRow/10;
+  nSelTest = (nSelStep > 100000) ? 100000 : nSelStep;
+
+  aTime = malloc(sizeof(int) * ArraySize(aSys) * nRow/nStep);
+  aWrite = malloc(sizeof(int) * nRow/nStep);
+  aSelTime = malloc(sizeof(int) * ArraySize(aSys) * nRow/nSelStep);
+
+  /* This loop collects the INSERT speed data. */
+  if( doWriteTest ){
+    printf("Writing output to file \"%s\".\n",  zOut);
+
+    for(j=0; aSys[j].zLibrary; j++){
+      FILE *pLog = 0;
+      TestDb *pDb;                  /* Database being tested */
+      lsm_db *pLsm;
+      int iDot = 0;
+  
+      if( ((1<<j)&sys_mask)==0 ) continue;
+      if( bSleep && nSleep ) sqlite3_sleep(nSleep);
+      bSleep = 1;
+
+      testCaseBegin(&rc, 0, "speed.insert.%s", aSys[j].zLibrary);
+
+      rc = tdb_open(aSys[j].zLibrary, 0, 1, &pDb);
+      if( rc ) return rc;
+
+      pLsm = configure_lsm_db(pDb);
+#if 0
+      pLog = fopen("/tmp/speed.log", "w");
+      tdb_lsm_write_hook(pDb, do_speed_write_hook2, (void *)pLog);
+#endif
+  
+      testTimeInit();
+      for(i=0; i<nRow; i+=nStep){
+        int iStep;
+        int nWrite1 = 0, nWrite2 = 0;
+        testCaseProgress(i, nRow, testCaseNDot(), &iDot);
+        if( pLsm ) lsm_info(pLsm, LSM_INFO_NWRITE, &nWrite1);
+        for(iStep=0; iStep<nStep; iStep++){
+          u32 aKey[4];                  /* 16-byte key */
+          u32 aVal[25];                 /* 100 byte value */
+          testPrngArray(i+iStep, aKey, ArraySize(aKey));
+          testPrngArray(i+iStep, aVal, ArraySize(aVal));
+          rc = tdb_write(pDb, aKey, sizeof(aKey), aVal, sizeof(aVal));
+        }
+        aTime[(j*nRow+i)/nStep] = testTimeGet();
+        if( pLsm ) lsm_info(pLsm, LSM_INFO_NWRITE, &nWrite2);
+        aWrite[i/nStep] = nWrite2 - nWrite1;
+      }
+
+      tdb_close(pDb);
+      if( pLog ) fclose(pLog);
+      testCaseFinish(rc);
+    }
+  }
+
+  /* This loop collects the SELECT speed data. */
+  if( doReadTest ){
+    for(j=0; aSys[j].zLibrary; j++){
+      int iDot = 0;
+      TestDb *pDb;                  /* Database being tested */
+
+      if( ((1<<j)&sys_mask)==0 ) continue;
+      if( bSleep && nSleep ) sqlite3_sleep(nSleep);
+      bSleep = 1;
+
+      testCaseBegin(&rc, 0, "speed.select.%s", aSys[j].zLibrary);
+
+      if( doWriteTest ){
+        rc = tdb_open(aSys[j].zLibrary, 0, 1, &pDb);
+        if( rc ) return rc;
+        configure_lsm_db(pDb);
+
+        for(i=0; i<nRow; i+=nSelStep){
+          int iStep;
+          int iSel;
+          testCaseProgress(i, nRow, testCaseNDot(), &iDot);
+          for(iStep=0; iStep<nSelStep; iStep++){
+            u32 aKey[4];                  /* 16-byte key */
+            u32 aVal[25];                 /* 100 byte value */
+            testPrngArray(i+iStep, aKey, ArraySize(aKey));
+            testPrngArray(i+iStep, aVal, ArraySize(aVal));
+            rc = tdb_write(pDb, aKey, sizeof(aKey), aVal, sizeof(aVal));
+          }
+    
+          testTimeInit();
+          for(iSel=0; iSel<nSelTest; iSel++){
+            void *pDummy;
+            int nDummy;
+            u32 iKey;
+            u32 aKey[4];                  /* 16-byte key */
+    
+            iKey = testPrngValue(iSel) % (i+nSelStep);
+            testPrngArray(iKey, aKey, ArraySize(aKey));
+            rc = tdb_fetch(pDb, aKey, sizeof(aKey), &pDummy, &nDummy);
+          }
+          aSelTime[(j*nRow+i)/nSelStep] = testTimeGet();
+          tdb_fetch(pDb, 0, 0, 0, 0);
+        }
+      }else{
+        int t;
+        int iSel;
+
+        rc = tdb_open(aSys[j].zLibrary, 0, 0, &pDb);
+        configure_lsm_db(pDb);
+
+        testTimeInit();
+        for(iSel=0; rc==LSM_OK && iSel<nSelTest; iSel++){
+          void *pDummy;
+          int nDummy;
+          u32 iKey;
+          u32 aKey[4];                  /* 16-byte key */
+#ifndef NDEBUG
+          u32 aVal[25];                 /* 100 byte value */
+#endif
+
+          testCaseProgress(iSel, nSelTest, testCaseNDot(), &iDot);
+    
+          iKey = testPrngValue(iSel) % nRow;
+          testPrngArray(iKey, aKey, ArraySize(aKey));
+          rc = tdb_fetch(pDb, aKey, sizeof(aKey), &pDummy, &nDummy);
+
+#ifndef NDEBUG
+          testPrngArray(iKey, aVal, ArraySize(aVal));
+          assert( nDummy==100 && memcmp(aVal, pDummy, 100)==0 );
+#endif
+        }
+        if( rc!=LSM_OK ) return rc;
+
+        t = testTimeGet();
+        tdb_fetch(pDb, 0, 0, 0, 0);
+
+        printf("%s: %d selects/second\n", 
+            aSys[j].zLibrary, (int)((double)nSelTest*1000.0/t)
+        );
+      }
+
+      tdb_close(pDb);
+      testCaseFinish(rc);
+    }
+  }
+
+
+  if( doWriteTest ){
+    FILE *pOut = fopen(zOut, "w");
+    if( !pOut ){
+      printf("fopen(\"%s\", \"w\"): %s\n", zOut, strerror(errno));
+      return 1;
+    }
+
+    fprintf(pOut, "set xlabel \"Rows Inserted\"\n");
+    fprintf(pOut, "set ylabel \"Inserts per second\"\n");
+    if( doReadTest ){
+      fprintf(pOut, "set y2label \"Selects per second\"\n");
+    }else if( sys_mask==(1<<2) ){
+      fprintf(pOut, "set y2label \"Page writes per insert\"\n");
+    }
+    fprintf(pOut, "set yrange [0:*]\n");
+    fprintf(pOut, "set y2range [0:*]\n");
+    fprintf(pOut, "set xrange [%d:*]\n", MAX(nStep, nRow/20) );
+    fprintf(pOut, "set ytics nomirror\n");
+    fprintf(pOut, "set y2tics nomirror\n");
+    fprintf(pOut, "set key box lw 0.01\n");
+    fprintf(pOut, "plot ");
+  
+    for(j=0; aSys[j].zLibrary; j++){
+      if( (1<<j)&sys_mask ){
+        const char *zLib = aSys[j].zLibrary;
+        fprintf(pOut, "%s\"-\" ti \"%s INSERT\" with lines lc rgb \"%s\" ", 
+            (isFirst?"":", "), zLib, aSys[j].zColor
+        );
+        if( doReadTest ){
+          fprintf(pOut, ", \"-\" ti \"%s SELECT\" "
+                 "axis x1y2 with points lw 3 lc rgb \"%s\""
+              , zLib, aSys[j].zColor
+          );
+        }
+        isFirst = 0;
+      }
+    }
+
+    assert( strcmp(aSys[2].zLibrary, "lsm")==0 );
+    if( sys_mask==(1<<2) && !doReadTest ){
+      fprintf(pOut, ", \"-\" ti \"lsm pages written\" "
+        "axis x1y2 with boxes lw 1 lc rgb \"grey\""
+      );
+    }
+  
+    fprintf(pOut, "\n");
+  
+    for(j=0; aSys[j].zLibrary; j++){
+      if( ((1<<j)&sys_mask)==0 ) continue;
+      fprintf(pOut, "# Rows    Inserts per second\n");
+      for(i=0; i<nRow; i+=nStep){
+        int iTime = aTime[(j*nRow+i)/nStep];
+        int ips = (int)((i+nStep)*1000.0 / (double)iTime);
+        fprintf(pOut, "%d %d\n", i+nStep, ips);
+      }
+      fprintf(pOut, "end\n");
+  
+      if( doReadTest ){
+        fprintf(pOut, "# Rows    Selects per second\n");
+        for(i=0; i<nRow; i+=nSelStep){
+          int sps = (int)(nSelTest*1000.0/(double)aSelTime[(j*nRow+i)/nSelStep]);
+          fprintf(pOut, "%d %d\n", i+nSelStep, sps);
+        }
+        fprintf(pOut, "end\n");
+      }else if( sys_mask==(1<<2) ){
+        for(i=0; i<(nRow/nStep); i++){
+          fprintf(pOut, "%d %f\n", i*nStep, (double)aWrite[i] / (double)nStep);
+        }
+        fprintf(pOut, "end\n");
+      }
+    }
+  
+    fprintf(pOut, "pause -1\n");
+    fclose(pOut);
+  }
+
+  free(aTime);
+  free(aSelTime);
+  free(aWrite);
+  testMallocInstall(tdb_lsm_env());
+  return 0;
+}
+
+/*
+** Usage: lsmtest random ?N?
+**
+** This command prints a sequence of zero or more numbers from the PRNG
+** system to stdout. If the "N" argument is missing, values the first 10
+** values (i=0, i=1, ... i=9) are printed. Otherwise, the first N.
+**
+** This was added to verify that the PRNG values do not change between
+** runs of the lsmtest program.
+*/
+int do_random_tests(int nArg, char **azArg){
+  int i;
+  int nRand;
+  if( nArg==0 ){
+    nRand = 10;
+  }else if( nArg==1 ){
+    nRand = atoi(azArg[0]);
+  }else{
+    testPrintError("Usage: random ?N?\n");
+    return -1;
+  }
+  for(i=0; i<nRand; i++){
+    printf("0x%x\n", testPrngValue(i));
+  }
+  return 0;
+}
+
+static int testFormatSize(char *aBuf, int nBuf, i64 nByte){
+  int res;
+  if( nByte<(1<<10) ){
+    res = snprintf(aBuf, nBuf, "%d byte", (int)nByte);
+  }else if( nByte<(1<<20) ){
+    res = snprintf(aBuf, nBuf, "%dK", (int)(nByte/(1<<10)));
+  }else{
+    res = snprintf(aBuf, nBuf, "%dM", (int)(nByte/(1<<20)));
+  }
+  return res;
+}
+
+static i64 testReadSize(char *z){
+  int n = strlen(z);
+  char c = z[n-1];
+  i64 nMul = 1;
+
+  switch( c ){
+    case 'g': case 'G':
+      nMul = (1<<30);
+      break;
+
+    case 'm': case 'M':
+      nMul = (1<<20);
+      break;
+
+    case 'k': case 'K':
+      nMul = (1<<10);
+      break;
+
+    default:
+      nMul = 1;
+  }
+
+  return nMul * (i64)atoi(z);
+} 
+
+/*
+** Usage: lsmtest writespeed FILESIZE BLOCKSIZE SYNCSIZE
+*/
+static int do_writer_test(int nArg, char **azArg){
+  int nBlock;
+  int nSize;
+  int i;
+  int fd;
+  int ms;
+  char aFilesize[32];
+  char aBlockSize[32];
+
+  char *aPage;
+  int *aOrder;
+  int nSync;
+
+  i64 filesize;
+  i64 blocksize;
+  i64 syncsize;
+  int nPage = 4096;
+
+  /* How long to sleep before running a trial (in ms). */
+#if 0
+  const int nSleep = 10000;
+#endif
+  const int nSleep = 0;
+
+  if( nArg!=3 ){
+    testPrintUsage("FILESIZE BLOCKSIZE SYNCSIZE");
+    return -1;
+  }
+
+  filesize = testReadSize(azArg[0]);
+  blocksize = testReadSize(azArg[1]);
+  syncsize = testReadSize(azArg[2]);
+
+  nBlock = (int)(filesize / blocksize);
+  nSize = (int)blocksize;
+  nSync = (int)(syncsize / blocksize);
+
+  aPage = (char *)malloc(4096);
+  aOrder = (int *)malloc(nBlock * sizeof(int));
+  for(i=0; i<nBlock; i++) aOrder[i] = i;
+  for(i=0; i<(nBlock*25); i++){
+    int tmp;
+    u32 a = testPrngValue(i);
+    u32 b = testPrngValue(a);
+    a = a % nBlock;
+    b = b % nBlock;
+    tmp = aOrder[a];
+    aOrder[a] = aOrder[b];
+    aOrder[b] = tmp;
+  }
+
+  testFormatSize(aFilesize, sizeof(aFilesize), (i64)nBlock * (i64)nSize);
+  testFormatSize(aBlockSize, sizeof(aFilesize), nSize);
+
+  printf("Testing writing a %s file using %s blocks. ", aFilesize, aBlockSize);
+  if( nSync==1 ){
+    printf("Sync after each block.\n");
+  }else{
+    printf("Sync after each %d blocks.\n", nSync);
+  }
+
+  printf("Preparing file... ");
+  fflush(stdout);
+  unlink("writer.out");
+  fd = open("writer.out", O_RDWR|O_CREAT|_O_BINARY, 0664);
+  if( fd<0 ){
+    testPrintError("open(): %d - %s\n", errno, strerror(errno));
+    return -1;
+  }
+  testTimeInit();
+  for(i=0; i<nBlock; i++){
+    int iPg;
+    memset(aPage, i&0xFF, nPage);
+    for(iPg=0; iPg<(nSize/nPage); iPg++){
+      write(fd, aPage, nPage);
+    }
+  }
+  fsync(fd);
+  printf("ok (%d ms)\n", testTimeGet());
+
+  for(i=0; i<5; i++){
+    int j;
+
+    sqlite3_sleep(nSleep);
+    printf("Now writing sequentially...  ");
+    fflush(stdout);
+
+    lseek(fd, 0, SEEK_SET);
+    testTimeInit();
+    for(j=0; j<nBlock; j++){
+      int iPg;
+      if( ((j+1)%nSync)==0 ) fdatasync(fd);
+      memset(aPage, j&0xFF, nPage);
+      for(iPg=0; iPg<(nSize/nPage); iPg++){
+        write(fd, aPage, nPage);
+      }
+    }
+    fdatasync(fd);
+    ms = testTimeGet();
+    printf("%d ms\n", ms);
+    sqlite3_sleep(nSleep);
+    printf("Now in an arbitrary order... ");
+
+    fflush(stdout);
+    testTimeInit();
+    for(j=0; j<nBlock; j++){
+      int iPg;
+      if( ((j+1)%nSync)==0 ) fdatasync(fd);
+      lseek(fd, aOrder[j]*nSize, SEEK_SET);
+      memset(aPage, j&0xFF, nPage);
+      for(iPg=0; iPg<(nSize/nPage); iPg++){
+        write(fd, aPage, nPage);
+      }
+    }
+    fdatasync(fd);
+    ms = testTimeGet();
+    printf("%d ms\n", ms);
+  }
+
+  close(fd);
+  free(aPage);
+  free(aOrder);
+
+  return 0;
+}
+
+static void do_insert_work_hook(lsm_db *db, void *p){
+  char *z = 0;
+  lsm_info(db, LSM_INFO_DB_STRUCTURE, &z);
+  if( z ){
+    printf("%s\n", z);
+    fflush(stdout);
+    lsm_free(lsm_get_env(db), z);
+  }
+
+  unused_parameter(p);
+}
+
+typedef struct InsertWriteHook InsertWriteHook;
+struct InsertWriteHook {
+  FILE *pOut;
+  int bLog;
+  i64 iOff;
+  int nData;
+};
+
+static void flushHook(InsertWriteHook *pHook){
+  if( pHook->nData ){
+    fprintf(pHook->pOut, "write %s %d %d\n", 
+        (pHook->bLog ? "log" : "db"), (int)pHook->iOff, pHook->nData
+    );
+    pHook->nData = 0;
+    fflush(pHook->pOut);
+  }
+}
+
+static void do_insert_write_hook(
+  void *pCtx,
+  int bLog,
+  i64 iOff,
+  int nData,
+  int nUs
+){
+  InsertWriteHook *pHook = (InsertWriteHook *)pCtx;
+  if( bLog ) return;
+
+  if( nData==0 ){
+    flushHook(pHook);
+    fprintf(pHook->pOut, "sync %s\n", (bLog ? "log" : "db"));
+  }else if( pHook->nData 
+         && bLog==pHook->bLog 
+         && iOff==(pHook->iOff+pHook->nData) 
+  ){
+    pHook->nData += nData;
+  }else{
+    flushHook(pHook);
+    pHook->bLog = bLog;
+    pHook->iOff = iOff;
+    pHook->nData = nData;
+  }
+}
+
+static int do_replay(int nArg, char **azArg){
+  char aBuf[4096];
+  FILE *pInput;
+  FILE *pClose = 0;
+  const char *zDb;
+
+  lsm_env *pEnv;
+  lsm_file *pOut;
+  int rc;
+
+  if( nArg!=2 ){
+    testPrintError("Usage: replay WRITELOG FILE\n");
+    return 1;
+  }
+
+  if( strcmp(azArg[0], "-")==0 ){
+    pInput = stdin;
+  }else{
+    pClose = pInput = fopen(azArg[0], "r");
+  }
+  zDb = azArg[1];
+  pEnv = tdb_lsm_env();
+  rc = pEnv->xOpen(pEnv, zDb, 0, &pOut);
+  if( rc!=LSM_OK ) return rc;
+
+  while( feof(pInput)==0 ){
+    char zLine[80];
+    fgets(zLine, sizeof(zLine)-1, pInput);
+    zLine[sizeof(zLine)-1] = '\0';
+
+    if( 0==memcmp("sync db", zLine, 7) ){
+      rc = pEnv->xSync(pOut);
+      if( rc!=0 ) break;
+    }else{
+      int iOff;
+      int nData;
+      int nMatch;
+      nMatch = sscanf(zLine, "write db %d %d", &iOff, &nData);
+      if( nMatch==2 ){
+        int i;
+        for(i=0; i<nData; i+=sizeof(aBuf)){
+          memset(aBuf, i&0xFF, sizeof(aBuf));
+          rc = pEnv->xWrite(pOut, iOff+i, aBuf, sizeof(aBuf));
+          if( rc!=0 ) break;
+        }
+      }
+    }
+  }
+  if( pClose ) fclose(pClose);
+  pEnv->xClose(pOut);
+
+  return rc;
+}
+
+static int do_insert(int nArg, char **azArg){
+  const char *zDb = "lsm";
+  TestDb *pDb = 0;
+  int i;
+  int rc;
+  const int nRow = 1 * 1000 * 1000;
+
+  DatasourceDefn defn = { TEST_DATASOURCE_RANDOM, 8, 15, 80, 150 };
+  Datasource *pData = 0;
+
+  if( nArg>1 ){
+    testPrintError("Usage: insert ?DATABASE?\n");
+    return 1;
+  }
+  if( nArg==1 ){ zDb = azArg[0]; }
+
+  testMallocUninstall(tdb_lsm_env());
+  for(i=0; zDb[i] && zDb[i]!='='; i++);
+  if( zDb[i] ){
+    rc = tdb_lsm_open(zDb, "testdb.lsm", 1, &pDb);
+  }else{
+    rc = tdb_open(zDb, 0, 1, &pDb);
+  }
+
+  if( rc!=0 ){
+    testPrintError("Error opening db \"%s\": %d\n", zDb, rc);
+  }else{
+    InsertWriteHook hook;
+    memset(&hook, 0, sizeof(hook));
+    hook.pOut = fopen("writelog.txt", "w");
+
+    pData = testDatasourceNew(&defn);
+    tdb_lsm_config_work_hook(pDb, do_insert_work_hook, 0);
+    tdb_lsm_write_hook(pDb, do_insert_write_hook, (void *)&hook);
+
+    if( rc==0 ){
+      for(i=0; i<nRow; i++){
+        void *pKey; int nKey;     /* Database key to insert */
+        void *pVal; int nVal;     /* Database value to insert */
+        testDatasourceEntry(pData, i, &pKey, &nKey, &pVal, &nVal);
+        tdb_write(pDb, pKey, nKey, pVal, nVal);
+      }
+    }
+
+    testDatasourceFree(pData);
+    tdb_close(pDb);
+    flushHook(&hook);
+    fclose(hook.pOut);
+  }
+  testMallocInstall(tdb_lsm_env());
+
+  return rc;
+}
+
+static int st_do_show(int a, char **b)      { return do_show(a, b); }
+static int st_do_work(int a, char **b)      { return do_work(a, b); }
+static int st_do_io(int a, char **b)        { return do_io(a, b); }
+
+#ifdef __linux__
+#include <sys/time.h>
+#include <sys/resource.h>
+
+static void lsmtest_rusage_report(void){
+  struct rusage r;
+  memset(&r, 0, sizeof(r));
+
+  getrusage(RUSAGE_SELF, &r);
+  printf("# getrusage: { ru_maxrss %d ru_oublock %d ru_inblock %d }\n", 
+      (int)r.ru_maxrss, (int)r.ru_oublock, (int)r.ru_inblock
+  );
+}
+#else
+static void lsmtest_rusage_report(void){
+  /* no-op */
+}
+#endif
+
+int main(int argc, char **argv){
+  struct TestFunc {
+    const char *zName;
+    int bRusageReport;
+    int (*xFunc)(int, char **);
+  } aTest[] = {
+    {"random",      1, do_random_tests},
+    {"writespeed",  1, do_writer_test},
+    {"io",          1, st_do_io},
+
+    {"insert",      1, do_insert},
+    {"replay",      1, do_replay},
+
+    {"speed",       1, do_speed_tests},
+    {"speed2",      1, do_speed_test2},
+    {"show",        0, st_do_show},
+    {"work",        1, st_do_work},
+    {"test",        1, do_test},
+
+    {0, 0}
+  };
+  int rc;                         /* Return Code */
+  int iFunc;                      /* Index into aTest[] */
+
+  int nLeakAlloc = 0;             /* Allocations leaked by lsm */
+  int nLeakByte = 0;              /* Bytes leaked by lsm */
+
+#ifdef LSM_DEBUG_MEM
+  FILE *pReport = 0;              /* lsm malloc() report file */
+  const char *zReport = "malloc.txt generated";
+#else
+  const char *zReport = "malloc.txt NOT generated";
+#endif
+
+  testMallocInstall(tdb_lsm_env());
+
+  if( argc<2 ){
+    testPrintError("Usage: %s sub-command ?args...?\n", argv[0]);
+    return -1;
+  }
+
+  /* Initialize error reporting */
+  testErrorInit(argc, argv);
+
+  /* Initialize PRNG system */
+  testPrngInit();
+
+  rc = testArgSelect(aTest, "sub-command", argv[1], &iFunc);
+  if( rc==0 ){
+    rc = aTest[iFunc].xFunc(argc-2, &argv[2]);
+  }
+
+#ifdef LSM_DEBUG_MEM
+  pReport = fopen("malloc.txt", "w");
+  testMallocCheck(tdb_lsm_env(), &nLeakAlloc, &nLeakByte, pReport);
+  fclose(pReport);
+#else
+  testMallocCheck(tdb_lsm_env(), &nLeakAlloc, &nLeakByte, 0);
+#endif
+
+  if( nLeakAlloc ){
+    testPrintError("Leaked %d bytes in %d allocations (%s)\n", 
+        nLeakByte, nLeakAlloc, zReport
+    );
+    if( rc==0 ) rc = -1;
+  }
+  testMallocUninstall(tdb_lsm_env());
+
+  if( aTest[iFunc].bRusageReport ){
+    lsmtest_rusage_report();
+  }
+  return rc;
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_mem.c b/ext/lsm1/lsm-test/lsmtest_mem.c
new file mode 100644
index 0000000..4c35e84
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_mem.c
@@ -0,0 +1,409 @@
+
+#include <stdio.h>
+#include <assert.h>
+#include <string.h>
+
+#define ArraySize(x) ((int)(sizeof(x) / sizeof((x)[0])))
+
+#define MIN(x,y) ((x)<(y) ? (x) : (y))
+
+typedef unsigned int  u32;
+typedef unsigned char u8;
+typedef long long int i64;
+typedef unsigned long long int u64;
+
+#if defined(__GLIBC__) && defined(LSM_DEBUG_MEM)
+  extern int backtrace(void**,int);
+  extern void backtrace_symbols_fd(void*const*,int,int);
+# define TM_BACKTRACE 12
+#else
+# define backtrace(A,B) 1
+# define backtrace_symbols_fd(A,B,C)
+#endif
+
+
+typedef struct TmBlockHdr TmBlockHdr;
+typedef struct TmAgg TmAgg;
+typedef struct TmGlobal TmGlobal;
+
+struct TmGlobal {
+  /* Linked list of all currently outstanding allocations. And a table of
+  ** all allocations, past and present, indexed by backtrace() info.  */
+  TmBlockHdr *pFirst;
+#ifdef TM_BACKTRACE
+  TmAgg *aHash[10000];
+#endif
+
+  /* Underlying malloc/realloc/free functions */
+  void *(*xMalloc)(int);          /* underlying malloc(3) function */
+  void *(*xRealloc)(void *, int); /* underlying realloc(3) function */
+  void (*xFree)(void *);          /* underlying free(3) function */
+
+  /* Mutex to protect pFirst and aHash */
+  void (*xEnterMutex)(TmGlobal*); /* Call this to enter the mutex */
+  void (*xLeaveMutex)(TmGlobal*); /* Call this to leave mutex */
+  void (*xDelMutex)(TmGlobal*);   /* Call this to delete mutex */
+  void *pMutex;                   /* Mutex handle */
+
+  void *(*xSaveMalloc)(void *, size_t);
+  void *(*xSaveRealloc)(void *, void *, size_t);
+  void (*xSaveFree)(void *, void *);
+
+  /* OOM injection scheduling. If nCountdown is greater than zero when a 
+  ** malloc attempt is made, it is decremented. If this means nCountdown 
+  ** transitions from 1 to 0, then the allocation fails. If bPersist is true 
+  ** when this happens, nCountdown is then incremented back to 1 (so that the 
+  ** next attempt fails too).  
+  */
+  int nCountdown;
+  int bPersist;
+  int bEnable;
+  void (*xHook)(void *);
+  void *pHookCtx;
+};
+
+struct TmBlockHdr {
+  TmBlockHdr *pNext;
+  TmBlockHdr *pPrev;
+  int nByte;
+#ifdef TM_BACKTRACE
+  TmAgg *pAgg;
+#endif
+  u32 iForeGuard;
+};
+
+#ifdef TM_BACKTRACE
+struct TmAgg {
+  int nAlloc;                     /* Number of allocations at this path */
+  int nByte;                      /* Total number of bytes allocated */
+  int nOutAlloc;                  /* Number of outstanding allocations */
+  int nOutByte;                   /* Number of outstanding bytes */
+  void *aFrame[TM_BACKTRACE];     /* backtrace() output */
+  TmAgg *pNext;                   /* Next object in hash-table collision */
+};
+#endif
+
+#define FOREGUARD 0x80F5E153
+#define REARGUARD 0xE4676B53
+static const u32 rearguard = REARGUARD;
+
+#define ROUND8(x) (((x)+7)&~7)
+
+#define BLOCK_HDR_SIZE (ROUND8( sizeof(TmBlockHdr) ))
+
+static void lsmtest_oom_error(void){
+  static int nErr = 0;
+  nErr++;
+}
+
+static void tmEnterMutex(TmGlobal *pTm){
+  pTm->xEnterMutex(pTm);
+}
+static void tmLeaveMutex(TmGlobal *pTm){
+  pTm->xLeaveMutex(pTm);
+}
+
+static void *tmMalloc(TmGlobal *pTm, int nByte){
+  TmBlockHdr *pNew;               /* New allocation header block */
+  u8 *pUser;                      /* Return value */
+  int nReq;                       /* Total number of bytes requested */
+
+  assert( sizeof(rearguard)==4 );
+  nReq = BLOCK_HDR_SIZE + nByte + 4;
+  pNew = (TmBlockHdr *)pTm->xMalloc(nReq);
+  memset(pNew, 0, sizeof(TmBlockHdr));
+
+  tmEnterMutex(pTm);
+  assert( pTm->nCountdown>=0 );
+  assert( pTm->bPersist==0 || pTm->bPersist==1 );
+
+  if( pTm->bEnable && pTm->nCountdown==1 ){
+    /* Simulate an OOM error. */
+    lsmtest_oom_error();
+    pTm->xFree(pNew);
+    pTm->nCountdown = pTm->bPersist;
+    if( pTm->xHook ) pTm->xHook(pTm->pHookCtx);
+    pUser = 0;
+  }else{
+    if( pTm->bEnable && pTm->nCountdown ) pTm->nCountdown--;
+
+    pNew->iForeGuard = FOREGUARD;
+    pNew->nByte = nByte;
+    pNew->pNext = pTm->pFirst;
+
+    if( pTm->pFirst ){
+      pTm->pFirst->pPrev = pNew;
+    }
+    pTm->pFirst = pNew;
+
+    pUser = &((u8 *)pNew)[BLOCK_HDR_SIZE];
+    memset(pUser, 0x56, nByte);
+    memcpy(&pUser[nByte], &rearguard, 4);
+
+#ifdef TM_BACKTRACE
+    {
+      TmAgg *pAgg;
+      int i;
+      u32 iHash = 0;
+      void *aFrame[TM_BACKTRACE];
+      memset(aFrame, 0, sizeof(aFrame));
+      backtrace(aFrame, TM_BACKTRACE);
+
+      for(i=0; i<ArraySize(aFrame); i++){
+        iHash += (u64)(aFrame[i]) + (iHash<<3);
+      }
+      iHash = iHash % ArraySize(pTm->aHash);
+
+      for(pAgg=pTm->aHash[iHash]; pAgg; pAgg=pAgg->pNext){
+        if( memcmp(pAgg->aFrame, aFrame, sizeof(aFrame))==0 ) break;
+      }
+      if( !pAgg ){
+        pAgg = (TmAgg *)pTm->xMalloc(sizeof(TmAgg));
+        memset(pAgg, 0, sizeof(TmAgg));
+        memcpy(pAgg->aFrame, aFrame, sizeof(aFrame));
+        pAgg->pNext = pTm->aHash[iHash];
+        pTm->aHash[iHash] = pAgg;
+      }
+      pAgg->nAlloc++;
+      pAgg->nByte += nByte;
+      pAgg->nOutAlloc++;
+      pAgg->nOutByte += nByte;
+      pNew->pAgg = pAgg;
+    }
+#endif
+  }
+
+  tmLeaveMutex(pTm);
+  return pUser;
+}
+
+static void tmFree(TmGlobal *pTm, void *p){
+  if( p ){
+    TmBlockHdr *pHdr;
+    u8 *pUser = (u8 *)p;
+
+    tmEnterMutex(pTm);
+    pHdr = (TmBlockHdr *)(pUser - BLOCK_HDR_SIZE);
+    assert( pHdr->iForeGuard==FOREGUARD );
+    assert( 0==memcmp(&pUser[pHdr->nByte], &rearguard, 4) );
+
+    if( pHdr->pPrev ){
+      assert( pHdr->pPrev->pNext==pHdr );
+      pHdr->pPrev->pNext = pHdr->pNext;
+    }else{
+      assert( pHdr==pTm->pFirst );
+      pTm->pFirst = pHdr->pNext;
+    }
+    if( pHdr->pNext ){
+      assert( pHdr->pNext->pPrev==pHdr );
+      pHdr->pNext->pPrev = pHdr->pPrev;
+    }
+
+#ifdef TM_BACKTRACE
+    pHdr->pAgg->nOutAlloc--;
+    pHdr->pAgg->nOutByte -= pHdr->nByte;
+#endif
+
+    tmLeaveMutex(pTm);
+    memset(pUser, 0x58, pHdr->nByte);
+    memset(pHdr, 0x57, sizeof(TmBlockHdr));
+    pTm->xFree(pHdr);
+  }
+}
+
+static void *tmRealloc(TmGlobal *pTm, void *p, int nByte){
+  void *pNew;
+
+  pNew = tmMalloc(pTm, nByte);
+  if( pNew && p ){
+    TmBlockHdr *pHdr;
+    u8 *pUser = (u8 *)p;
+    pHdr = (TmBlockHdr *)(pUser - BLOCK_HDR_SIZE);
+    memcpy(pNew, p, MIN(nByte, pHdr->nByte));
+    tmFree(pTm, p);
+  }
+  return pNew;
+}
+
+static void tmMallocOom(
+  TmGlobal *pTm, 
+  int nCountdown, 
+  int bPersist,
+  void (*xHook)(void *),
+  void *pHookCtx
+){
+  assert( nCountdown>=0 );
+  assert( bPersist==0 || bPersist==1 );
+  pTm->nCountdown = nCountdown;
+  pTm->bPersist = bPersist;
+  pTm->xHook = xHook;
+  pTm->pHookCtx = pHookCtx;
+  pTm->bEnable = 1;
+}
+
+static void tmMallocOomEnable(
+  TmGlobal *pTm, 
+  int bEnable
+){
+  pTm->bEnable = bEnable;
+}
+
+static void tmMallocCheck(
+  TmGlobal *pTm,
+  int *pnLeakAlloc,
+  int *pnLeakByte,
+  FILE *pFile
+){
+  TmBlockHdr *pHdr;
+  int nLeak = 0;
+  int nByte = 0;
+
+  if( pTm==0 ) return;
+
+  for(pHdr=pTm->pFirst; pHdr; pHdr=pHdr->pNext){
+    nLeak++; 
+    nByte += pHdr->nByte;
+  }
+  if( pnLeakAlloc ) *pnLeakAlloc = nLeak;
+  if( pnLeakByte ) *pnLeakByte = nByte;
+
+#ifdef TM_BACKTRACE
+  if( pFile ){
+    int i;
+    fprintf(pFile, "LEAKS\n");
+    for(i=0; i<ArraySize(pTm->aHash); i++){
+      TmAgg *pAgg;
+      for(pAgg=pTm->aHash[i]; pAgg; pAgg=pAgg->pNext){
+        if( pAgg->nOutAlloc ){
+          int j;
+          fprintf(pFile, "%d %d ", pAgg->nOutByte, pAgg->nOutAlloc);
+          for(j=0; j<TM_BACKTRACE; j++){
+            fprintf(pFile, "%p ", pAgg->aFrame[j]);
+          }
+          fprintf(pFile, "\n");
+        }
+      }
+    }
+    fprintf(pFile, "\nALLOCATIONS\n");
+    for(i=0; i<ArraySize(pTm->aHash); i++){
+      TmAgg *pAgg;
+      for(pAgg=pTm->aHash[i]; pAgg; pAgg=pAgg->pNext){
+        int j;
+        fprintf(pFile, "%d %d ", pAgg->nByte, pAgg->nAlloc);
+        for(j=0; j<TM_BACKTRACE; j++) fprintf(pFile, "%p ", pAgg->aFrame[j]);
+        fprintf(pFile, "\n");
+      }
+    }
+  }
+#else
+  (void)pFile;
+#endif
+}
+
+
+#include "lsm.h"
+#include "stdlib.h"
+
+typedef struct LsmMutex LsmMutex;
+struct LsmMutex {
+  lsm_env *pEnv;
+  lsm_mutex *pMutex;
+};
+
+static void tmLsmMutexEnter(TmGlobal *pTm){
+  LsmMutex *p = (LsmMutex *)pTm->pMutex;
+  p->pEnv->xMutexEnter(p->pMutex);
+}
+static void tmLsmMutexLeave(TmGlobal *pTm){
+  LsmMutex *p = (LsmMutex *)(pTm->pMutex);
+  p->pEnv->xMutexLeave(p->pMutex);
+}
+static void tmLsmMutexDel(TmGlobal *pTm){
+  LsmMutex *p = (LsmMutex *)pTm->pMutex;
+  pTm->xFree(p);
+}
+static void *tmLsmMalloc(int n){ return malloc(n); }
+static void tmLsmFree(void *ptr){ free(ptr); }
+static void *tmLsmRealloc(void *ptr, int n){ return realloc(ptr, n); }
+
+static void *tmLsmEnvMalloc(lsm_env *p, size_t n){ 
+  return tmMalloc((TmGlobal *)(p->pMemCtx), n); 
+}
+static void tmLsmEnvFree(lsm_env *p, void *ptr){ 
+  tmFree((TmGlobal *)(p->pMemCtx), ptr); 
+}
+static void *tmLsmEnvRealloc(lsm_env *p, void *ptr, size_t n){ 
+  return tmRealloc((TmGlobal *)(p->pMemCtx), ptr, n);
+}
+
+void testMallocInstall(lsm_env *pEnv){
+  TmGlobal *pGlobal;
+  LsmMutex *pMutex;
+  assert( pEnv->pMemCtx==0 );
+
+  /* Allocate and populate a TmGlobal structure. */
+  pGlobal = (TmGlobal *)tmLsmMalloc(sizeof(TmGlobal));
+  memset(pGlobal, 0, sizeof(TmGlobal));
+  pGlobal->xMalloc = tmLsmMalloc;
+  pGlobal->xRealloc = tmLsmRealloc;
+  pGlobal->xFree = tmLsmFree;
+  pMutex = (LsmMutex *)pGlobal->xMalloc(sizeof(LsmMutex));
+  pMutex->pEnv = pEnv;
+  pEnv->xMutexStatic(pEnv, LSM_MUTEX_HEAP, &pMutex->pMutex);
+  pGlobal->xEnterMutex = tmLsmMutexEnter;
+  pGlobal->xLeaveMutex = tmLsmMutexLeave;
+  pGlobal->xDelMutex = tmLsmMutexDel;
+  pGlobal->pMutex = (void *)pMutex;
+
+  pGlobal->xSaveMalloc = pEnv->xMalloc;
+  pGlobal->xSaveRealloc = pEnv->xRealloc;
+  pGlobal->xSaveFree = pEnv->xFree;
+
+  /* Set up pEnv to the use the new TmGlobal */
+  pEnv->pMemCtx = (void *)pGlobal;
+  pEnv->xMalloc = tmLsmEnvMalloc;
+  pEnv->xRealloc = tmLsmEnvRealloc;
+  pEnv->xFree = tmLsmEnvFree;
+}
+
+void testMallocUninstall(lsm_env *pEnv){
+  TmGlobal *p = (TmGlobal *)pEnv->pMemCtx;
+  pEnv->pMemCtx = 0;
+  if( p ){
+    pEnv->xMalloc = p->xSaveMalloc;
+    pEnv->xRealloc = p->xSaveRealloc;
+    pEnv->xFree = p->xSaveFree;
+    p->xDelMutex(p);
+    tmLsmFree(p);
+  }
+}
+
+void testMallocCheck(
+  lsm_env *pEnv,
+  int *pnLeakAlloc,
+  int *pnLeakByte,
+  FILE *pFile
+){
+  if( pEnv->pMemCtx==0 ){
+    *pnLeakAlloc = 0;
+    *pnLeakByte = 0;
+  }else{
+    tmMallocCheck((TmGlobal *)(pEnv->pMemCtx), pnLeakAlloc, pnLeakByte, pFile);
+  }
+}
+
+void testMallocOom(
+  lsm_env *pEnv, 
+  int nCountdown, 
+  int bPersist,
+  void (*xHook)(void *),
+  void *pHookCtx
+){
+  TmGlobal *pTm = (TmGlobal *)(pEnv->pMemCtx);
+  tmMallocOom(pTm, nCountdown, bPersist, xHook, pHookCtx);
+}
+
+void testMallocOomEnable(lsm_env *pEnv, int bEnable){
+  TmGlobal *pTm = (TmGlobal *)(pEnv->pMemCtx);
+  tmMallocOomEnable(pTm, bEnable);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb.c b/ext/lsm1/lsm-test/lsmtest_tdb.c
new file mode 100644
index 0000000..8f63f64
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb.c
@@ -0,0 +1,846 @@
+
+/*
+** This program attempts to test the correctness of some facets of the 
+** LSM database library. Specifically, that the contents of the database
+** are maintained correctly during a series of inserts and deletes.
+*/
+
+
+#include "lsmtest_tdb.h"
+#include "lsm.h"
+
+#include "lsmtest.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#ifndef _WIN32
+# include <unistd.h>
+#endif
+#include <stdio.h>
+
+
+typedef struct SqlDb SqlDb;
+
+static int error_transaction_function(TestDb *p, int iLevel){ 
+  unused_parameter(p);
+  unused_parameter(iLevel);
+  return -1; 
+}
+
+
+/*************************************************************************
+** Begin wrapper for LevelDB.
+*/
+#ifdef HAVE_LEVELDB
+
+#include <leveldb/c.h>
+
+typedef struct LevelDb LevelDb;
+struct LevelDb {
+  TestDb base;
+  leveldb_t *db;
+  leveldb_options_t *pOpt;
+  leveldb_writeoptions_t *pWriteOpt;
+  leveldb_readoptions_t *pReadOpt;
+
+  char *pVal;
+};
+
+static int test_leveldb_close(TestDb *pTestDb){
+  LevelDb *pDb = (LevelDb *)pTestDb;
+
+  leveldb_close(pDb->db);
+  leveldb_writeoptions_destroy(pDb->pWriteOpt);
+  leveldb_readoptions_destroy(pDb->pReadOpt);
+  leveldb_options_destroy(pDb->pOpt);
+  free(pDb->pVal);
+  free(pDb);
+
+  return 0;
+}
+
+static int test_leveldb_write(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void *pVal, 
+  int nVal
+){
+  LevelDb *pDb = (LevelDb *)pTestDb;
+  char *zErr = 0;
+  leveldb_put(pDb->db, pDb->pWriteOpt, pKey, nKey, pVal, nVal, &zErr);
+  return (zErr!=0);
+}
+
+static int test_leveldb_delete(TestDb *pTestDb, void *pKey, int nKey){
+  LevelDb *pDb = (LevelDb *)pTestDb;
+  char *zErr = 0;
+  leveldb_delete(pDb->db, pDb->pWriteOpt, pKey, nKey, &zErr);
+  return (zErr!=0);
+}
+
+static int test_leveldb_fetch(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void **ppVal, 
+  int *pnVal
+){
+  LevelDb *pDb = (LevelDb *)pTestDb;
+  char *zErr = 0;
+  size_t nVal = 0;
+
+  if( pKey==0 ) return 0;
+  free(pDb->pVal);
+  pDb->pVal = leveldb_get(pDb->db, pDb->pReadOpt, pKey, nKey, &nVal, &zErr);
+  *ppVal = (void *)(pDb->pVal);
+  if( pDb->pVal==0 ){
+    *pnVal = -1;
+  }else{
+    *pnVal = (int)nVal;
+  }
+
+  return (zErr!=0);
+}
+
+static int test_leveldb_scan(
+  TestDb *pTestDb,
+  void *pCtx,
+  int bReverse,
+  void *pKey1, int nKey1,         /* Start of search */
+  void *pKey2, int nKey2,         /* End of search */
+  void (*xCallback)(void *, void *, int , void *, int)
+){
+  LevelDb *pDb = (LevelDb *)pTestDb;
+  leveldb_iterator_t *iter;
+
+  iter = leveldb_create_iterator(pDb->db, pDb->pReadOpt);
+
+  if( bReverse==0 ){
+    if( pKey1 ){
+      leveldb_iter_seek(iter, pKey1, nKey1);
+    }else{
+      leveldb_iter_seek_to_first(iter);
+    }
+  }else{
+    if( pKey2 ){
+      leveldb_iter_seek(iter, pKey2, nKey2);
+
+      if( leveldb_iter_valid(iter)==0 ){
+        leveldb_iter_seek_to_last(iter);
+      }else{
+        const char *k; size_t n;
+        int res;
+        k = leveldb_iter_key(iter, &n);
+        res = memcmp(k, pKey2, MIN(n, nKey2));
+        if( res==0 ) res = n - nKey2;
+        assert( res>=0 );
+        if( res>0 ){
+          leveldb_iter_prev(iter);
+        }
+      }
+    }else{
+      leveldb_iter_seek_to_last(iter);
+    }
+  }
+
+
+  while( leveldb_iter_valid(iter) ){
+    const char *k; size_t n;
+    const char *v; size_t n2;
+    int res;
+
+    k = leveldb_iter_key(iter, &n);
+    if( bReverse==0 && pKey2 ){
+      res = memcmp(k, pKey2, MIN(n, nKey2));
+      if( res==0 ) res = n - nKey2;
+      if( res>0 ) break;
+    }
+    if( bReverse!=0 && pKey1 ){
+      res = memcmp(k, pKey1, MIN(n, nKey1));
+      if( res==0 ) res = n - nKey1;
+      if( res<0 ) break;
+    }
+
+    v = leveldb_iter_value(iter, &n2);
+
+    xCallback(pCtx, (void *)k, n, (void *)v, n2);
+
+    if( bReverse==0 ){
+      leveldb_iter_next(iter);
+    }else{
+      leveldb_iter_prev(iter);
+    }
+  }
+
+  leveldb_iter_destroy(iter);
+  return 0;
+}
+
+static int test_leveldb_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  static const DatabaseMethods LeveldbMethods = {
+    test_leveldb_close,
+    test_leveldb_write,
+    test_leveldb_delete,
+    0,
+    test_leveldb_fetch,
+    test_leveldb_scan,
+    error_transaction_function,
+    error_transaction_function,
+    error_transaction_function
+  };
+
+  LevelDb *pLevelDb;
+  char *zErr = 0;
+
+  if( bClear ){
+    char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
+    system(zCmd);
+    sqlite3_free(zCmd);
+  }
+
+  pLevelDb = (LevelDb *)malloc(sizeof(LevelDb));
+  memset(pLevelDb, 0, sizeof(LevelDb));
+
+  pLevelDb->pOpt = leveldb_options_create();
+  leveldb_options_set_create_if_missing(pLevelDb->pOpt, 1);
+  pLevelDb->pWriteOpt = leveldb_writeoptions_create();
+  pLevelDb->pReadOpt = leveldb_readoptions_create();
+
+  pLevelDb->db = leveldb_open(pLevelDb->pOpt, zFilename, &zErr);
+
+  if( zErr ){
+    test_leveldb_close((TestDb *)pLevelDb);
+    *ppDb = 0;
+    return 1;
+  }
+
+  *ppDb = (TestDb *)pLevelDb;
+  pLevelDb->base.pMethods = &LeveldbMethods;
+  return 0;
+}
+#endif  /* HAVE_LEVELDB */
+/* 
+** End wrapper for LevelDB.
+*************************************************************************/
+
+#ifdef HAVE_KYOTOCABINET
+static int kc_close(TestDb *pTestDb){
+  return test_kc_close(pTestDb);
+}
+
+static int kc_write(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void *pVal, 
+  int nVal
+){
+  return test_kc_write(pTestDb, pKey, nKey, pVal, nVal);
+}
+
+static int kc_delete(TestDb *pTestDb, void *pKey, int nKey){
+  return test_kc_delete(pTestDb, pKey, nKey);
+}
+
+static int kc_delete_range(
+  TestDb *pTestDb, 
+  void *pKey1, int nKey1,
+  void *pKey2, int nKey2
+){
+  return test_kc_delete_range(pTestDb, pKey1, nKey1, pKey2, nKey2);
+}
+
+static int kc_fetch(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void **ppVal, 
+  int *pnVal
+){
+  if( pKey==0 ) return LSM_OK;
+  return test_kc_fetch(pTestDb, pKey, nKey, ppVal, pnVal);
+}
+
+static int kc_scan(
+  TestDb *pTestDb,
+  void *pCtx,
+  int bReverse,
+  void *pFirst, int nFirst,
+  void *pLast, int nLast,
+  void (*xCallback)(void *, void *, int , void *, int)
+){
+  return test_kc_scan(
+      pTestDb, pCtx, bReverse, pFirst, nFirst, pLast, nLast, xCallback
+  );
+}
+
+static int kc_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  static const DatabaseMethods KcdbMethods = {
+    kc_close,
+    kc_write,
+    kc_delete,
+    kc_delete_range,
+    kc_fetch,
+    kc_scan,
+    error_transaction_function,
+    error_transaction_function,
+    error_transaction_function
+  };
+
+  int rc;
+  TestDb *pTestDb = 0;
+
+  rc = test_kc_open(zFilename, bClear, &pTestDb);
+  if( rc!=0 ){
+    *ppDb = 0;
+    return rc;
+  }
+  pTestDb->pMethods = &KcdbMethods;
+  *ppDb = pTestDb;
+  return 0;
+}
+#endif /* HAVE_KYOTOCABINET */
+/* 
+** End wrapper for Kyoto cabinet.
+*************************************************************************/
+
+#ifdef HAVE_MDB
+static int mdb_close(TestDb *pTestDb){
+  return test_mdb_close(pTestDb);
+}
+
+static int mdb_write(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void *pVal, 
+  int nVal
+){
+  return test_mdb_write(pTestDb, pKey, nKey, pVal, nVal);
+}
+
+static int mdb_delete(TestDb *pTestDb, void *pKey, int nKey){
+  return test_mdb_delete(pTestDb, pKey, nKey);
+}
+
+static int mdb_fetch(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void **ppVal, 
+  int *pnVal
+){
+  if( pKey==0 ) return LSM_OK;
+  return test_mdb_fetch(pTestDb, pKey, nKey, ppVal, pnVal);
+}
+
+static int mdb_scan(
+  TestDb *pTestDb,
+  void *pCtx,
+  int bReverse,
+  void *pFirst, int nFirst,
+  void *pLast, int nLast,
+  void (*xCallback)(void *, void *, int , void *, int)
+){
+  return test_mdb_scan(
+      pTestDb, pCtx, bReverse, pFirst, nFirst, pLast, nLast, xCallback
+  );
+}
+
+static int mdb_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  static const DatabaseMethods KcdbMethods = {
+    mdb_close,
+    mdb_write,
+    mdb_delete,
+    0,
+    mdb_fetch,
+    mdb_scan,
+    error_transaction_function,
+    error_transaction_function,
+    error_transaction_function
+  };
+
+  int rc;
+  TestDb *pTestDb = 0;
+
+  rc = test_mdb_open(zSpec, zFilename, bClear, &pTestDb);
+  if( rc!=0 ){
+    *ppDb = 0;
+    return rc;
+  }
+  pTestDb->pMethods = &KcdbMethods;
+  *ppDb = pTestDb;
+  return 0;
+}
+#endif /* HAVE_MDB */
+
+/*************************************************************************
+** Begin wrapper for SQLite.
+*/
+
+/*
+** nOpenTrans:
+**   The number of open nested transactions, in the same sense as used
+**   by the tdb_begin/commit/rollback and SQLite 4 KV interfaces. If this
+**   value is 0, there are no transactions open at all. If it is 1, then
+**   there is a read transaction. If it is 2 or greater, then there are
+**   (nOpenTrans-1) nested write transactions open.
+*/
+struct SqlDb {
+  TestDb base;
+  sqlite3 *db;
+  sqlite3_stmt *pInsert;
+  sqlite3_stmt *pDelete;
+  sqlite3_stmt *pDeleteRange;
+  sqlite3_stmt *pFetch;
+  sqlite3_stmt *apScan[8];
+
+  int nOpenTrans;
+
+  /* Used by sql_fetch() to allocate space for results */
+  int nAlloc;
+  u8 *aAlloc;
+};
+
+static int sql_close(TestDb *pTestDb){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  sqlite3_finalize(pDb->pInsert);
+  sqlite3_finalize(pDb->pDelete);
+  sqlite3_finalize(pDb->pDeleteRange);
+  sqlite3_finalize(pDb->pFetch);
+  sqlite3_finalize(pDb->apScan[0]);
+  sqlite3_finalize(pDb->apScan[1]);
+  sqlite3_finalize(pDb->apScan[2]);
+  sqlite3_finalize(pDb->apScan[3]);
+  sqlite3_finalize(pDb->apScan[4]);
+  sqlite3_finalize(pDb->apScan[5]);
+  sqlite3_finalize(pDb->apScan[6]);
+  sqlite3_finalize(pDb->apScan[7]);
+  sqlite3_close(pDb->db);
+  free((char *)pDb->aAlloc);
+  free((char *)pDb);
+  return SQLITE_OK;
+}
+
+static int sql_write(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void *pVal, 
+  int nVal
+){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  sqlite3_bind_blob(pDb->pInsert, 1, pKey, nKey, SQLITE_STATIC);
+  sqlite3_bind_blob(pDb->pInsert, 2, pVal, nVal, SQLITE_STATIC);
+  sqlite3_step(pDb->pInsert);
+  return sqlite3_reset(pDb->pInsert);
+}
+
+static int sql_delete(TestDb *pTestDb, void *pKey, int nKey){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  sqlite3_bind_blob(pDb->pDelete, 1, pKey, nKey, SQLITE_STATIC);
+  sqlite3_step(pDb->pDelete);
+  return sqlite3_reset(pDb->pDelete);
+}
+
+static int sql_delete_range(
+  TestDb *pTestDb, 
+  void *pKey1, int nKey1,
+  void *pKey2, int nKey2
+){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  sqlite3_bind_blob(pDb->pDeleteRange, 1, pKey1, nKey1, SQLITE_STATIC);
+  sqlite3_bind_blob(pDb->pDeleteRange, 2, pKey2, nKey2, SQLITE_STATIC);
+  sqlite3_step(pDb->pDeleteRange);
+  return sqlite3_reset(pDb->pDeleteRange);
+}
+
+static int sql_fetch(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void **ppVal, 
+  int *pnVal
+){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  int rc;
+
+  sqlite3_reset(pDb->pFetch);
+  if( pKey==0 ){
+    assert( ppVal==0 );
+    assert( pnVal==0 );
+    return LSM_OK;
+  }
+
+  sqlite3_bind_blob(pDb->pFetch, 1, pKey, nKey, SQLITE_STATIC);
+  rc = sqlite3_step(pDb->pFetch);
+  if( rc==SQLITE_ROW ){
+    int nVal = sqlite3_column_bytes(pDb->pFetch, 0);
+    u8 *aVal = (void *)sqlite3_column_blob(pDb->pFetch, 0);
+
+    if( nVal>pDb->nAlloc ){
+      free(pDb->aAlloc);
+      pDb->aAlloc = (u8 *)malloc(nVal*2);
+      pDb->nAlloc = nVal*2;
+    }
+    memcpy(pDb->aAlloc, aVal, nVal);
+    *pnVal = nVal;
+    *ppVal = (void *)pDb->aAlloc;
+  }else{
+    *pnVal = -1;
+    *ppVal = 0;
+  }
+
+  rc = sqlite3_reset(pDb->pFetch);
+  return rc;
+}
+
+static int sql_scan(
+  TestDb *pTestDb,
+  void *pCtx,
+  int bReverse,
+  void *pFirst, int nFirst,
+  void *pLast, int nLast,
+  void (*xCallback)(void *, void *, int , void *, int)
+){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  sqlite3_stmt *pScan;
+
+  assert( bReverse==1 || bReverse==0 );
+  pScan = pDb->apScan[(pFirst==0) + (pLast==0)*2 + bReverse*4];
+
+  if( pFirst ) sqlite3_bind_blob(pScan, 1, pFirst, nFirst, SQLITE_STATIC);
+  if( pLast ) sqlite3_bind_blob(pScan, 2, pLast, nLast, SQLITE_STATIC);
+
+  while( SQLITE_ROW==sqlite3_step(pScan) ){
+    void *pKey; int nKey;
+    void *pVal; int nVal;
+
+    nKey = sqlite3_column_bytes(pScan, 0);
+    pKey = (void *)sqlite3_column_blob(pScan, 0);
+    nVal = sqlite3_column_bytes(pScan, 1);
+    pVal = (void *)sqlite3_column_blob(pScan, 1);
+
+    xCallback(pCtx, pKey, nKey, pVal, nVal);
+  }
+  return sqlite3_reset(pScan);
+}
+
+static int sql_begin(TestDb *pTestDb, int iLevel){
+  int i;
+  SqlDb *pDb = (SqlDb *)pTestDb;
+
+  /* iLevel==0 is a no-op */
+  if( iLevel==0 ) return 0;
+
+  /* If there are no transactions at all open, open a read transaction. */
+  if( pDb->nOpenTrans==0 ){
+    int rc = sqlite3_exec(pDb->db, 
+        "BEGIN; SELECT * FROM sqlite_schema LIMIT 1;" , 0, 0, 0
+    );
+    if( rc!=0 ) return rc;
+    pDb->nOpenTrans = 1;
+  }
+
+  /* Open any required write transactions */
+  for(i=pDb->nOpenTrans; i<iLevel; i++){
+    char *zSql = sqlite3_mprintf("SAVEPOINT x%d", i);
+    int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
+    sqlite3_free(zSql);
+    if( rc!=SQLITE_OK ) return rc;
+  }
+
+  pDb->nOpenTrans = iLevel;
+  return 0;
+}
+
+static int sql_commit(TestDb *pTestDb, int iLevel){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  assert( iLevel>=0 );
+
+  /* Close the read transaction if requested. */
+  if( pDb->nOpenTrans>=1 && iLevel==0 ){
+    int rc = sqlite3_exec(pDb->db, "COMMIT", 0, 0, 0);
+    if( rc!=0 ) return rc;
+    pDb->nOpenTrans = 0;
+  }
+
+  /* Close write transactions as required */
+  if( pDb->nOpenTrans>iLevel ){
+    char *zSql = sqlite3_mprintf("RELEASE x%d", iLevel);
+    int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
+    sqlite3_free(zSql);
+    if( rc!=0 ) return rc;
+  }
+
+  pDb->nOpenTrans = iLevel;
+  return 0;
+}
+
+static int sql_rollback(TestDb *pTestDb, int iLevel){
+  SqlDb *pDb = (SqlDb *)pTestDb;
+  assert( iLevel>=0 );
+
+  if( pDb->nOpenTrans>=1 && iLevel==0 ){
+    /* Close the read transaction if requested. */
+    int rc = sqlite3_exec(pDb->db, "ROLLBACK", 0, 0, 0);
+    if( rc!=0 ) return rc;
+  }else if( pDb->nOpenTrans>1 && iLevel==1 ){
+    /* Or, rollback and close the top-level write transaction */
+    int rc = sqlite3_exec(pDb->db, "ROLLBACK TO x1; RELEASE x1;", 0, 0, 0);
+    if( rc!=0 ) return rc;
+  }else{
+    /* Or, just roll back some nested transactions */
+    char *zSql = sqlite3_mprintf("ROLLBACK TO x%d", iLevel-1);
+    int rc = sqlite3_exec(pDb->db, zSql, 0, 0, 0);
+    sqlite3_free(zSql);
+    if( rc!=0 ) return rc;
+  }
+
+  pDb->nOpenTrans = iLevel;
+  return 0;
+}
+
+static int sql_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  static const DatabaseMethods SqlMethods = {
+    sql_close,
+    sql_write,
+    sql_delete,
+    sql_delete_range,
+    sql_fetch,
+    sql_scan,
+    sql_begin,
+    sql_commit,
+    sql_rollback
+  };
+  const char *zCreate = "CREATE TABLE IF NOT EXISTS t1(k PRIMARY KEY, v)";
+  const char *zInsert = "REPLACE INTO t1 VALUES(?, ?)";
+  const char *zDelete = "DELETE FROM t1 WHERE k = ?";
+  const char *zRange = "DELETE FROM t1 WHERE k>? AND k<?";
+  const char *zFetch  = "SELECT v FROM t1 WHERE k = ?";
+
+  const char *zScan0  = "SELECT * FROM t1 WHERE k BETWEEN ?1 AND ?2 ORDER BY k";
+  const char *zScan1  = "SELECT * FROM t1 WHERE k <= ?2 ORDER BY k";
+  const char *zScan2  = "SELECT * FROM t1 WHERE k >= ?1 ORDER BY k";
+  const char *zScan3  = "SELECT * FROM t1 ORDER BY k";
+
+  const char *zScan4  = 
+    "SELECT * FROM t1 WHERE k BETWEEN ?1 AND ?2 ORDER BY k DESC";
+  const char *zScan5  = "SELECT * FROM t1 WHERE k <= ?2 ORDER BY k DESC";
+  const char *zScan6  = "SELECT * FROM t1 WHERE k >= ?1 ORDER BY k DESC";
+  const char *zScan7  = "SELECT * FROM t1 ORDER BY k DESC";
+
+  int rc;
+  SqlDb *pDb;
+  char *zPragma;
+
+  if( bClear && zFilename && zFilename[0] ){
+    unlink(zFilename);
+  }
+
+  pDb = (SqlDb *)malloc(sizeof(SqlDb));
+  memset(pDb, 0, sizeof(SqlDb));
+  pDb->base.pMethods = &SqlMethods;
+
+  if( 0!=(rc = sqlite3_open(zFilename, &pDb->db))
+   || 0!=(rc = sqlite3_exec(pDb->db, zCreate, 0, 0, 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zInsert, -1, &pDb->pInsert, 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zDelete, -1, &pDb->pDelete, 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zRange, -1, &pDb->pDeleteRange, 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zFetch, -1, &pDb->pFetch, 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan0, -1, &pDb->apScan[0], 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan1, -1, &pDb->apScan[1], 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan2, -1, &pDb->apScan[2], 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan3, -1, &pDb->apScan[3], 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan4, -1, &pDb->apScan[4], 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan5, -1, &pDb->apScan[5], 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan6, -1, &pDb->apScan[6], 0))
+   || 0!=(rc = sqlite3_prepare_v2(pDb->db, zScan7, -1, &pDb->apScan[7], 0))
+  ){
+    *ppDb = 0;
+    sql_close((TestDb *)pDb);
+    return rc;
+  }
+
+  zPragma = sqlite3_mprintf("PRAGMA page_size=%d", TESTDB_DEFAULT_PAGE_SIZE);
+  sqlite3_exec(pDb->db, zPragma, 0, 0, 0);
+  sqlite3_free(zPragma);
+  zPragma = sqlite3_mprintf("PRAGMA cache_size=%d", TESTDB_DEFAULT_CACHE_SIZE);
+  sqlite3_exec(pDb->db, zPragma, 0, 0, 0);
+  sqlite3_free(zPragma);
+
+  /* sqlite3_exec(pDb->db, "PRAGMA locking_mode=EXCLUSIVE", 0, 0, 0); */
+  sqlite3_exec(pDb->db, "PRAGMA synchronous=OFF", 0, 0, 0);
+  sqlite3_exec(pDb->db, "PRAGMA journal_mode=WAL", 0, 0, 0);
+  sqlite3_exec(pDb->db, "PRAGMA wal_autocheckpoint=4096", 0, 0, 0);
+  if( zSpec ){
+    rc = sqlite3_exec(pDb->db, zSpec, 0, 0, 0);
+    if( rc!=SQLITE_OK ){
+      sql_close((TestDb *)pDb);
+      return rc;
+    }
+  }
+
+  *ppDb = (TestDb *)pDb;
+  return 0;
+}
+/* 
+** End wrapper for SQLite.
+*************************************************************************/
+
+/*************************************************************************
+** Begin exported functions.
+*/
+static struct Lib {
+  const char *zName;
+  const char *zDefaultDb;
+  int (*xOpen)(const char *, const char *zFilename, int bClear, TestDb **ppDb);
+} aLib[] = {
+  { "sqlite3",      "testdb.sqlite",    sql_open },
+  { "lsm_small",    "testdb.lsm_small", test_lsm_small_open },
+  { "lsm_lomem",    "testdb.lsm_lomem", test_lsm_lomem_open },
+  { "lsm_lomem2",   "testdb.lsm_lomem2", test_lsm_lomem2_open },
+#ifdef HAVE_ZLIB
+  { "lsm_zip",      "testdb.lsm_zip",   test_lsm_zip_open },
+#endif
+  { "lsm",          "testdb.lsm",       test_lsm_open },
+#ifdef LSM_MUTEX_PTHREADS
+  { "lsm_mt2",      "testdb.lsm_mt2",   test_lsm_mt2 },
+  { "lsm_mt3",      "testdb.lsm_mt3",   test_lsm_mt3 },
+#endif
+#ifdef HAVE_LEVELDB
+  { "leveldb",      "testdb.leveldb",   test_leveldb_open },
+#endif
+#ifdef HAVE_KYOTOCABINET
+  { "kyotocabinet", "testdb.kc",        kc_open },
+#endif
+#ifdef HAVE_MDB
+  { "mdb", "./testdb.mdb",        mdb_open }
+#endif
+};
+
+const char *tdb_system_name(int i){
+  if( i<0 || i>=ArraySize(aLib) ) return 0;
+  return aLib[i].zName;
+}
+
+const char *tdb_default_db(const char *zSys){
+  int i;
+  for(i=0; i<ArraySize(aLib); i++){
+    if( strcmp(aLib[i].zName, zSys)==0 ) return aLib[i].zDefaultDb;
+  }
+  return 0;
+}
+
+int tdb_open(const char *zLib, const char *zDb, int bClear, TestDb **ppDb){
+  int i;
+  int rc = 1;
+  const char *zSpec = 0;
+
+  int nLib = 0;
+  while( zLib[nLib] && zLib[nLib]!=' ' ){
+    nLib++;
+  }
+  zSpec = &zLib[nLib];
+  while( *zSpec==' ' ) zSpec++;
+  if( *zSpec=='\0' ) zSpec = 0;
+
+  for(i=0; i<ArraySize(aLib); i++){
+    if( (int)strlen(aLib[i].zName)==nLib
+        && 0==memcmp(zLib, aLib[i].zName, nLib) ){
+      rc = aLib[i].xOpen(zSpec, (zDb ? zDb : aLib[i].zDefaultDb), bClear, ppDb);
+      if( rc==0 ){
+        (*ppDb)->zLibrary = aLib[i].zName;
+      }
+      break;
+    }
+  }
+
+  if( rc ){
+    /* Failed to find the requested database library. Return an error. */
+    *ppDb = 0;
+  }
+  return rc;
+}
+
+int tdb_close(TestDb *pDb){
+  if( pDb ){
+    return pDb->pMethods->xClose(pDb);
+  }
+  return 0;
+}
+
+int tdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
+  return pDb->pMethods->xWrite(pDb, pKey, nKey, pVal, nVal);
+}
+
+int tdb_delete(TestDb *pDb, void *pKey, int nKey){
+  return pDb->pMethods->xDelete(pDb, pKey, nKey);
+}
+
+int tdb_delete_range(
+    TestDb *pDb, void *pKey1, int nKey1, void *pKey2, int nKey2
+){
+  return pDb->pMethods->xDeleteRange(pDb, pKey1, nKey1, pKey2, nKey2);
+}
+
+int tdb_fetch(TestDb *pDb, void *pKey, int nKey, void **ppVal, int *pnVal){
+  return pDb->pMethods->xFetch(pDb, pKey, nKey, ppVal, pnVal);
+}
+
+int tdb_scan(
+  TestDb *pDb,                    /* Database handle */
+  void *pCtx,                     /* Context pointer to pass to xCallback */
+  int bReverse,                   /* True to scan in reverse order */
+  void *pKey1, int nKey1,         /* Start of search */
+  void *pKey2, int nKey2,         /* End of search */
+  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+){
+  return pDb->pMethods->xScan(
+      pDb, pCtx, bReverse, pKey1, nKey1, pKey2, nKey2, xCallback
+  );
+}
+
+int tdb_begin(TestDb *pDb, int iLevel){
+  return pDb->pMethods->xBegin(pDb, iLevel);
+}
+int tdb_commit(TestDb *pDb, int iLevel){
+  return pDb->pMethods->xCommit(pDb, iLevel);
+}
+int tdb_rollback(TestDb *pDb, int iLevel){
+  return pDb->pMethods->xRollback(pDb, iLevel);
+}
+
+int tdb_transaction_support(TestDb *pDb){
+  return (pDb->pMethods->xBegin != error_transaction_function);
+}
+
+const char *tdb_library_name(TestDb *pDb){
+  return pDb->zLibrary;
+}
+
+/* 
+** End exported functions.
+*************************************************************************/
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb.h b/ext/lsm1/lsm-test/lsmtest_tdb.h
new file mode 100644
index 0000000..c55b6e2
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb.h
@@ -0,0 +1,174 @@
+
+/*
+** This file is the interface to a very simple database library used for
+** testing. The interface is similar to that of the LSM. The main virtue 
+** of this library is that the same API may be used to access a key-value
+** store implemented by LSM, SQLite or another database system. Which 
+** makes it easy to use for correctness and performance tests.
+*/
+
+#ifndef __WRAPPER_H_
+#define __WRAPPER_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "lsm.h"
+
+typedef struct TestDb TestDb;
+
+/*
+** Open a new database connection. The first argument is the name of the
+** database library to use. e.g. something like:
+**
+**     "sqlite3"
+**     "lsm"
+**
+** See function tdb_system_name() for a list of available database systems.
+**
+** The second argument is the name of the database to open (e.g. a filename).
+**
+** If the third parameter is non-zero, then any existing database by the
+** name of zDb is removed before opening a new one. If it is zero, then an
+** existing database may be opened.
+*/
+int tdb_open(const char *zLibrary, const char *zDb, int bClear, TestDb **ppDb);
+
+/*
+** Close a database handle.
+*/
+int tdb_close(TestDb *pDb);
+
+/*
+** Write a new key/value into the database.
+*/
+int tdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal);
+
+/*
+** Delete a key from the database.
+*/
+int tdb_delete(TestDb *pDb, void *pKey, int nKey);
+
+/*
+** Delete a range of keys from the database.
+*/
+int tdb_delete_range(TestDb *, void *pKey1, int nKey1, void *pKey2, int nKey2);
+
+/*
+** Query the database for key (pKey/nKey). If no entry is found, set *ppVal
+** to 0 and *pnVal to -1 before returning. Otherwise, set *ppVal and *pnVal
+** to a pointer to and size of the value associated with (pKey/nKey).
+*/
+int tdb_fetch(TestDb *pDb, void *pKey, int nKey, void **ppVal, int *pnVal);
+
+/*
+** Open and close nested transactions. Currently, these functions only 
+** work for SQLite3 and LSM systems. Use the tdb_transaction_support() 
+** function to determine if a given TestDb handle supports these methods.
+**
+** These functions and the iLevel parameter follow the same conventions as
+** the SQLite 4 transaction interface. Note that this is slightly different
+** from the way LSM does things. As follows:
+**
+** tdb_begin():
+**   A successful call to tdb_begin() with (iLevel>1) guarantees that 
+**   there are at least (iLevel-1) write transactions open. If iLevel==1,
+**   then it guarantees that at least a read-transaction is open. Calling
+**   tdb_begin() with iLevel==0 is a no-op.
+**
+** tdb_commit():
+**   A successful call to tdb_commit() with (iLevel>1) guarantees that 
+**   there are at most (iLevel-1) write transactions open. If iLevel==1,
+**   then it guarantees that there are no write transactions open (although
+**   a read-transaction may remain open).  Calling tdb_commit() with 
+**   iLevel==0 ensures that all transactions, read or write, have been 
+**   closed and committed.
+**
+** tdb_rollback():
+**   This call is similar to tdb_commit(), except that instead of committing
+**   transactions, it reverts them. For example, calling tdb_rollback() with
+**   iLevel==2 ensures that there is at most one write transaction open, and
+**   restores the database to the state that it was in when that transaction
+**   was opened.
+**
+**   In other words, tdb_commit() just closes transactions - tdb_rollback()
+**   closes transactions and then restores the database to the state it
+**   was in before those transactions were even opened.
+*/
+int tdb_begin(TestDb *pDb, int iLevel);
+int tdb_commit(TestDb *pDb, int iLevel);
+int tdb_rollback(TestDb *pDb, int iLevel);
+
+/*
+** Return true if transactions are supported, or false otherwise.
+*/
+int tdb_transaction_support(TestDb *pDb);
+
+/*
+** Return the name of the database library (as passed to tdb_open()) used
+** by the handled passed as the first argument.
+*/
+const char *tdb_library_name(TestDb *pDb);
+
+/*
+** Scan a range of database keys. Invoke the callback function for each
+** key visited.
+*/
+int tdb_scan(
+  TestDb *pDb,                    /* Database handle */
+  void *pCtx,                     /* Context pointer to pass to xCallback */
+  int bReverse,                   /* True to scan in reverse order */
+  void *pKey1, int nKey1,         /* Start of search */
+  void *pKey2, int nKey2,         /* End of search */
+  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+);
+
+const char *tdb_system_name(int i);
+const char *tdb_default_db(const char *zSys);
+
+int tdb_lsm_open(const char *zCfg, const char *zDb, int bClear, TestDb **ppDb);
+
+/*
+** If the TestDb handle passed as an argument is a wrapper around an LSM
+** database, return the LSM handle. Otherwise, if the argument is some other
+** database system, return NULL.
+*/
+lsm_db *tdb_lsm(TestDb *pDb);
+
+/*
+** Return true if the db passed as an argument is a multi-threaded LSM
+** connection.
+*/
+int tdb_lsm_multithread(TestDb *pDb);
+
+/*
+** Return a pointer to the lsm_env object used by all lsm database
+** connections initialized as a copy of the object returned by 
+** lsm_default_env(). It may be modified (e.g. to override functions)
+** if the caller can guarantee that it is not already in use.
+*/
+lsm_env *tdb_lsm_env(void);
+
+/*
+** The following functions only work with LSM database handles. It is
+** illegal to call them with any other type of database handle specified
+** as an argument.
+*/
+void tdb_lsm_enable_log(TestDb *pDb, int bEnable);
+void tdb_lsm_application_crash(TestDb *pDb);
+void tdb_lsm_prepare_system_crash(TestDb *pDb);
+void tdb_lsm_system_crash(TestDb *pDb);
+void tdb_lsm_prepare_sync_crash(TestDb *pDb, int iSync);
+
+
+void tdb_lsm_safety(TestDb *pDb, int eMode);
+void tdb_lsm_config_work_hook(TestDb *pDb, void (*)(lsm_db *, void *), void *);
+void tdb_lsm_write_hook(TestDb *, void(*)(void*,int,lsm_i64,int,int), void*);
+int tdb_lsm_config_str(TestDb *pDb, const char *zStr);
+
+#ifdef __cplusplus
+}  /* End of the 'extern "C"' block */
+#endif
+
+#endif
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb2.cc b/ext/lsm1/lsm-test/lsmtest_tdb2.cc
new file mode 100644
index 0000000..86ebb49
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb2.cc
@@ -0,0 +1,369 @@
+
+
+#include "lsmtest.h"
+#include <stdlib.h>
+
+#ifdef HAVE_KYOTOCABINET
+#include "kcpolydb.h"
+extern "C" {
+  struct KcDb {
+    TestDb base;
+    kyotocabinet::TreeDB* db;
+    char *pVal;
+  };
+}
+
+int test_kc_open(const char *zFilename, int bClear, TestDb **ppDb){
+  KcDb *pKcDb;
+  int ok;
+  int rc = 0;
+
+  if( bClear ){
+    char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
+    system(zCmd);
+    sqlite3_free(zCmd);
+  }
+
+  pKcDb = (KcDb *)malloc(sizeof(KcDb));
+  memset(pKcDb, 0, sizeof(KcDb));
+
+
+  pKcDb->db = new kyotocabinet::TreeDB();
+  pKcDb->db->tune_page(TESTDB_DEFAULT_PAGE_SIZE);
+  pKcDb->db->tune_page_cache(
+      TESTDB_DEFAULT_PAGE_SIZE * TESTDB_DEFAULT_CACHE_SIZE
+  );
+  ok = pKcDb->db->open(zFilename,
+      kyotocabinet::PolyDB::OWRITER | kyotocabinet::PolyDB::OCREATE
+  );
+  if( ok==0 ){
+    free(pKcDb);
+    pKcDb = 0;
+    rc = 1;
+  }
+
+  *ppDb = (TestDb *)pKcDb;
+  return rc;
+}
+
+int test_kc_close(TestDb *pDb){
+  KcDb *pKcDb = (KcDb *)pDb;
+  if( pKcDb->pVal ){
+    delete [] pKcDb->pVal;
+  }
+  pKcDb->db->close();
+  delete pKcDb->db;
+  free(pKcDb);
+  return 0;
+}
+
+int test_kc_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
+  KcDb *pKcDb = (KcDb *)pDb;
+  int ok;
+
+  ok = pKcDb->db->set((const char *)pKey, nKey, (const char *)pVal, nVal);
+  return (ok ? 0 : 1);
+}
+
+int test_kc_delete(TestDb *pDb, void *pKey, int nKey){
+  KcDb *pKcDb = (KcDb *)pDb;
+  int ok;
+
+  ok = pKcDb->db->remove((const char *)pKey, nKey);
+  return (ok ? 0 : 1);
+}
+
+int test_kc_delete_range(
+  TestDb *pDb, 
+  void *pKey1, int nKey1,
+  void *pKey2, int nKey2
+){
+  int res;
+  KcDb *pKcDb = (KcDb *)pDb;
+  kyotocabinet::DB::Cursor* pCur = pKcDb->db->cursor();
+
+  if( pKey1 ){
+    res = pCur->jump((const char *)pKey1, nKey1);
+  }else{
+    res = pCur->jump();
+  }
+
+  while( 1 ){
+    const char *pKey; size_t nKey;
+    const char *pVal; size_t nVal;
+
+    pKey = pCur->get(&nKey, &pVal, &nVal);
+    if( pKey==0 ) break;
+
+#ifndef NDEBUG
+    if( pKey1 ){
+      res = memcmp(pKey, pKey1, MIN((size_t)nKey1, nKey));
+      assert( res>0 || (res==0 && nKey>nKey1) );
+    }
+#endif
+
+    if( pKey2 ){
+      res = memcmp(pKey, pKey2, MIN((size_t)nKey2, nKey));
+      if( res>0 || (res==0 && (size_t)nKey2<nKey) ){
+        delete [] pKey;
+        break;
+      }
+    }
+    pCur->remove();
+    delete [] pKey;
+  }
+
+  delete pCur;
+  return 0;
+}
+
+int test_kc_fetch(
+  TestDb *pDb, 
+  void *pKey, 
+  int nKey, 
+  void **ppVal,
+  int *pnVal
+){
+  KcDb *pKcDb = (KcDb *)pDb;
+  size_t nVal;
+
+  if( pKcDb->pVal ){
+    delete [] pKcDb->pVal;
+    pKcDb->pVal = 0;
+  }
+
+  pKcDb->pVal = pKcDb->db->get((const char *)pKey, nKey, &nVal);
+  if( pKcDb->pVal ){
+    *ppVal = pKcDb->pVal;
+    *pnVal = nVal;
+  }else{
+    *ppVal = 0;
+    *pnVal = -1;
+  }
+
+  return 0;
+}
+
+int test_kc_scan(
+  TestDb *pDb,                    /* Database handle */
+  void *pCtx,                     /* Context pointer to pass to xCallback */
+  int bReverse,                   /* True for a reverse order scan */
+  void *pKey1, int nKey1,         /* Start of search */
+  void *pKey2, int nKey2,         /* End of search */
+  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+){
+  KcDb *pKcDb = (KcDb *)pDb;
+  kyotocabinet::DB::Cursor* pCur = pKcDb->db->cursor();
+  int res;
+
+  if( bReverse==0 ){
+    if( pKey1 ){
+      res = pCur->jump((const char *)pKey1, nKey1);
+    }else{
+      res = pCur->jump();
+    }
+  }else{
+    if( pKey2 ){
+      res = pCur->jump_back((const char *)pKey2, nKey2);
+    }else{
+      res = pCur->jump_back();
+    }
+  }
+
+  while( res ){
+    const char *pKey; size_t nKey;
+    const char *pVal; size_t nVal;
+    pKey = pCur->get(&nKey, &pVal, &nVal);
+
+    if( bReverse==0 && pKey2 ){
+      res = memcmp(pKey, pKey2, MIN((size_t)nKey2, nKey));
+      if( res>0 || (res==0 && (size_t)nKey2<nKey) ){
+        delete [] pKey;
+        break;
+      }
+    }else if( bReverse!=0 && pKey1 ){
+      res = memcmp(pKey, pKey1, MIN((size_t)nKey1, nKey));
+      if( res<0 || (res==0 && (size_t)nKey1>nKey) ){
+        delete [] pKey;
+        break;
+      }
+    }
+
+    xCallback(pCtx, (void *)pKey, (int)nKey, (void *)pVal, (int)nVal);
+    delete [] pKey;
+
+    if( bReverse ){
+      res = pCur->step_back();
+    }else{
+      res = pCur->step();
+    }
+  }
+
+  delete pCur;
+  return 0;
+}
+#endif /* HAVE_KYOTOCABINET */
+
+#ifdef HAVE_MDB 
+#include "lmdb.h"
+
+extern "C" {
+  struct MdbDb {
+    TestDb base;
+    MDB_env *env;
+    MDB_dbi dbi;
+  };
+}
+
+int test_mdb_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  MDB_txn *txn;
+  MdbDb *pMdb;
+  int rc;
+
+  if( bClear ){
+    char *zCmd = sqlite3_mprintf("rm -rf %s\n", zFilename);
+    system(zCmd);
+    sqlite3_free(zCmd);
+  }
+
+  pMdb = (MdbDb *)malloc(sizeof(MdbDb));
+  memset(pMdb, 0, sizeof(MdbDb));
+
+  rc = mdb_env_create(&pMdb->env);
+  if( rc==0 ) rc = mdb_env_set_mapsize(pMdb->env, 1*1024*1024*1024);
+  if( rc==0 ) rc = mdb_env_open(pMdb->env, zFilename, MDB_NOSYNC|MDB_NOSUBDIR, 0600);
+  if( rc==0 ) rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
+  if( rc==0 ){
+    rc = mdb_open(txn, NULL, 0, &pMdb->dbi);
+    mdb_txn_commit(txn);
+  }
+
+  *ppDb = (TestDb *)pMdb;
+  return rc;
+}
+
+int test_mdb_close(TestDb *pDb){
+  MdbDb *pMdb = (MdbDb *)pDb;
+
+  mdb_close(pMdb->env, pMdb->dbi);
+  mdb_env_close(pMdb->env);
+  free(pMdb);
+  return 0;
+}
+
+int test_mdb_write(TestDb *pDb, void *pKey, int nKey, void *pVal, int nVal){
+  int rc;
+  MdbDb *pMdb = (MdbDb *)pDb;
+  MDB_val val;
+  MDB_val key;
+  MDB_txn *txn;
+
+  val.mv_size = nVal; 
+  val.mv_data = pVal;
+  key.mv_size = nKey; 
+  key.mv_data = pKey;
+
+  rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
+  if( rc==0 ){
+    rc = mdb_put(txn, pMdb->dbi, &key, &val, 0);
+    if( rc==0 ){
+      rc = mdb_txn_commit(txn);
+    }else{
+      mdb_txn_abort(txn);
+    }
+  }
+  
+  return rc;
+}
+
+int test_mdb_delete(TestDb *pDb, void *pKey, int nKey){
+  int rc;
+  MdbDb *pMdb = (MdbDb *)pDb;
+  MDB_val key;
+  MDB_txn *txn;
+
+  key.mv_size = nKey; 
+  key.mv_data = pKey;
+  rc = mdb_txn_begin(pMdb->env, NULL, 0, &txn);
+  if( rc==0 ){
+    rc = mdb_del(txn, pMdb->dbi, &key, 0);
+    if( rc==0 ){
+      rc = mdb_txn_commit(txn);
+    }else{
+      mdb_txn_abort(txn);
+    }
+  }
+  
+  return rc;
+}
+
+int test_mdb_fetch(
+  TestDb *pDb, 
+  void *pKey, 
+  int nKey, 
+  void **ppVal,
+  int *pnVal
+){
+  int rc;
+  MdbDb *pMdb = (MdbDb *)pDb;
+  MDB_val key;
+  MDB_txn *txn;
+
+  key.mv_size = nKey;
+  key.mv_data = pKey;
+
+  rc = mdb_txn_begin(pMdb->env, NULL, MDB_RDONLY, &txn);
+  if( rc==0 ){
+    MDB_val val = {0, 0};
+    rc = mdb_get(txn, pMdb->dbi, &key, &val);
+    if( rc==MDB_NOTFOUND ){
+      rc = 0;
+      *ppVal = 0;
+      *pnVal = -1;
+    }else{
+      *ppVal = val.mv_data;
+      *pnVal = val.mv_size;
+    }
+    mdb_txn_commit(txn);
+  }
+
+  return rc;
+}
+
+int test_mdb_scan(
+  TestDb *pDb,                    /* Database handle */
+  void *pCtx,                     /* Context pointer to pass to xCallback */
+  int bReverse,                   /* True for a reverse order scan */
+  void *pKey1, int nKey1,         /* Start of search */
+  void *pKey2, int nKey2,         /* End of search */
+  void (*xCallback)(void *pCtx, void *pKey, int nKey, void *pVal, int nVal)
+){
+  MdbDb *pMdb = (MdbDb *)pDb;
+  int rc;
+  MDB_cursor_op op = bReverse ? MDB_PREV : MDB_NEXT;
+  MDB_txn *txn;
+
+  rc = mdb_txn_begin(pMdb->env, NULL, MDB_RDONLY, &txn);
+  if( rc==0 ){
+    MDB_cursor *csr;
+    MDB_val key = {0, 0};
+    MDB_val val = {0, 0};
+
+    rc = mdb_cursor_open(txn, pMdb->dbi, &csr);
+    if( rc==0 ){
+      while( mdb_cursor_get(csr, &key, &val, op)==0 ){
+        xCallback(pCtx, key.mv_data, key.mv_size, val.mv_data, val.mv_size);
+      }
+      mdb_cursor_close(csr);
+    }
+  }
+
+  return rc;
+}
+
+#endif /* HAVE_MDB */
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb3.c b/ext/lsm1/lsm-test/lsmtest_tdb3.c
new file mode 100644
index 0000000..e29497a
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb3.c
@@ -0,0 +1,1429 @@
+
+#include "lsmtest_tdb.h"
+#include "lsm.h"
+#include "lsmtest.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#ifndef _WIN32
+# include <unistd.h>
+#endif
+#include <stdio.h>
+
+#ifndef _WIN32
+# include <sys/time.h>
+#endif
+
+typedef struct LsmDb LsmDb;
+typedef struct LsmWorker LsmWorker;
+typedef struct LsmFile LsmFile;
+
+#define LSMTEST_DFLT_MT_MAX_CKPT (8*1024)
+#define LSMTEST_DFLT_MT_MIN_CKPT (2*1024)
+
+#ifdef LSM_MUTEX_PTHREADS
+#include <pthread.h>
+
+#define LSMTEST_THREAD_CKPT      1
+#define LSMTEST_THREAD_WORKER    2
+#define LSMTEST_THREAD_WORKER_AC 3
+
+/*
+** There are several different types of worker threads that run in different
+** test configurations, depending on the value of LsmWorker.eType.
+**
+**   1. Checkpointer.
+**   2. Worker with auto-checkpoint.
+**   3. Worker without auto-checkpoint.
+*/
+struct LsmWorker {
+  LsmDb *pDb;                     /* Main database structure */
+  lsm_db *pWorker;                /* Worker database handle */
+  pthread_t worker_thread;        /* Worker thread */
+  pthread_cond_t worker_cond;     /* Condition var the worker waits on */
+  pthread_mutex_t worker_mutex;   /* Mutex used with worker_cond */
+  int bDoWork;                    /* Set to true by client when there is work */
+  int worker_rc;                  /* Store error code here */
+  int eType;                      /* LSMTEST_THREAD_XXX constant */
+  int bBlock;
+};
+#else
+struct LsmWorker { int worker_rc; int bBlock; };
+#endif
+
+static void mt_shutdown(LsmDb *);
+
+lsm_env *tdb_lsm_env(void){
+  static int bInit = 0;
+  static lsm_env env;
+  if( bInit==0 ){
+    memcpy(&env, lsm_default_env(), sizeof(env));
+    bInit = 1;
+  }
+  return &env;
+}
+
+typedef struct FileSector FileSector;
+typedef struct FileData FileData;
+
+struct FileSector {
+  u8 *aOld;                       /* Old data for this sector */
+};
+
+struct FileData {
+  int nSector;                    /* Allocated size of apSector[] array */
+  FileSector *aSector;            /* Array of file sectors */
+};
+
+/*
+** bPrepareCrash:
+**   If non-zero, the file wrappers maintain enough in-memory data to
+**   simulate the effect of a power-failure on the file-system (i.e. that
+**   unsynced sectors may be written, not written, or overwritten with
+**   arbitrary data when the crash occurs).
+**
+** bCrashed:
+**   Set to true after a crash is simulated. Once this variable is true, all
+**   VFS methods other than xClose() return LSM_IOERR as soon as they are
+**   called (without affecting the contents of the file-system).
+**
+** env:
+**   The environment object used by all lsm_db* handles opened by this
+**   object (i.e. LsmDb.db plus any worker connections). Variable env.pVfsCtx
+**   always points to the containing LsmDb structure.
+*/
+struct LsmDb {
+  TestDb base;                    /* Base class - methods table */
+  lsm_env env;                    /* Environment used by connection db */
+  char *zName;                    /* Database file name */
+  lsm_db *db;                     /* LSM database handle */
+
+  lsm_cursor *pCsr;               /* Cursor held open during read transaction */
+  void *pBuf;                     /* Buffer for tdb_fetch() output */
+  int nBuf;                       /* Allocated (not used) size of pBuf */
+
+  /* Crash testing related state */
+  int bCrashed;                   /* True once a crash has occurred */
+  int nAutoCrash;                 /* Number of syncs until a crash */
+  int bPrepareCrash;              /* True to store writes in memory */
+
+  /* Unsynced data (while crash testing) */
+  int szSector;                   /* Assumed size of disk sectors (512B) */
+  FileData aFile[2];              /* Database and log file data */
+
+  /* Other test instrumentation */
+  int bNoRecovery;                /* If true, assume DMS2 is locked */
+
+  /* Work hook redirection */
+  void (*xWork)(lsm_db *, void *);
+  void *pWorkCtx;
+
+  /* IO logging hook */
+  void (*xWriteHook)(void *, int, lsm_i64, int, int);
+  void *pWriteCtx;
+  
+  /* Worker threads (for lsm_mt) */
+  int nMtMinCkpt;
+  int nMtMaxCkpt;
+  int eMode;
+  int nWorker;
+  LsmWorker *aWorker;
+};
+
+#define LSMTEST_MODE_SINGLETHREAD    1
+#define LSMTEST_MODE_BACKGROUND_CKPT 2
+#define LSMTEST_MODE_BACKGROUND_WORK 3
+#define LSMTEST_MODE_BACKGROUND_BOTH 4
+
+/*************************************************************************
+**************************************************************************
+** Begin test VFS code.
+*/
+
+struct LsmFile {
+  lsm_file *pReal;                /* Real underlying file */
+  int bLog;                       /* True for log file. False for db file */
+  LsmDb *pDb;                     /* Database handle that uses this file */
+};
+
+static int testEnvFullpath(
+  lsm_env *pEnv,                  /* Environment for current LsmDb */
+  const char *zFile,              /* Relative path name */
+  char *zOut,                     /* Output buffer */
+  int *pnOut                      /* IN/OUT: Size of output buffer */
+){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  return pRealEnv->xFullpath(pRealEnv, zFile, zOut, pnOut);
+}
+
+static int testEnvOpen(
+  lsm_env *pEnv,                  /* Environment for current LsmDb */
+  const char *zFile,              /* Name of file to open */
+  int flags,
+  lsm_file **ppFile               /* OUT: New file handle object */
+){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmDb *pDb = (LsmDb *)pEnv->pVfsCtx;
+  int rc;                         /* Return Code */
+  LsmFile *pRet;                  /* The new file handle */
+  int nFile;                      /* Length of string zFile in bytes */
+
+  nFile = strlen(zFile);
+  pRet = (LsmFile *)testMalloc(sizeof(LsmFile));
+  pRet->pDb = pDb;
+  pRet->bLog = (nFile > 4 && 0==memcmp("-log", &zFile[nFile-4], 4));
+
+  rc = pRealEnv->xOpen(pRealEnv, zFile, flags, &pRet->pReal);
+  if( rc!=LSM_OK ){
+    testFree(pRet);
+    pRet = 0;
+  }
+
+  *ppFile = (lsm_file *)pRet;
+  return rc;
+}
+
+static int testEnvRead(lsm_file *pFile, lsm_i64 iOff, void *pData, int nData){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+  if( p->pDb->bCrashed ) return LSM_IOERR;
+  return pRealEnv->xRead(p->pReal, iOff, pData, nData);
+}
+
+static int testEnvWrite(lsm_file *pFile, lsm_i64 iOff, void *pData, int nData){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+  LsmDb *pDb = p->pDb;
+
+  if( pDb->bCrashed ) return LSM_IOERR;
+
+  if( pDb->bPrepareCrash ){
+    FileData *pData2 = &pDb->aFile[p->bLog];
+    int iFirst;                 
+    int iLast;
+    int iSector;
+
+    iFirst = (int)(iOff / pDb->szSector);
+    iLast =  (int)((iOff + nData - 1) / pDb->szSector);
+
+    if( pData2->nSector<(iLast+1) ){
+      int nNew = ( ((iLast + 1) + 63) / 64 ) * 64;
+      assert( nNew>iLast );
+      pData2->aSector = (FileSector *)testRealloc(
+          pData2->aSector, nNew*sizeof(FileSector)
+      );
+      memset(&pData2->aSector[pData2->nSector], 
+          0, (nNew - pData2->nSector) * sizeof(FileSector)
+      );
+      pData2->nSector = nNew;
+    }
+
+    for(iSector=iFirst; iSector<=iLast; iSector++){
+      if( pData2->aSector[iSector].aOld==0 ){
+        u8 *aOld = (u8 *)testMalloc(pDb->szSector);
+        pRealEnv->xRead(
+            p->pReal, (lsm_i64)iSector*pDb->szSector, aOld, pDb->szSector
+        );
+        pData2->aSector[iSector].aOld = aOld;
+      }
+    }
+  }
+
+  if( pDb->xWriteHook ){
+    int rc;
+    int nUs;
+    struct timeval t1;
+    struct timeval t2;
+
+    gettimeofday(&t1, 0);
+    assert( nData>0 );
+    rc = pRealEnv->xWrite(p->pReal, iOff, pData, nData);
+    gettimeofday(&t2, 0);
+
+    nUs = (t2.tv_sec - t1.tv_sec) * 1000000 + (t2.tv_usec - t1.tv_usec);
+    pDb->xWriteHook(pDb->pWriteCtx, p->bLog, iOff, nData, nUs);
+    return rc;
+  }
+
+  return pRealEnv->xWrite(p->pReal, iOff, pData, nData);
+}
+
+static void doSystemCrash(LsmDb *pDb);
+
+static int testEnvSync(lsm_file *pFile){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+  LsmDb *pDb = p->pDb;
+  FileData *pData = &pDb->aFile[p->bLog];
+  int i;
+
+  if( pDb->bCrashed ) return LSM_IOERR;
+
+  if( pDb->nAutoCrash ){
+    pDb->nAutoCrash--;
+    if( pDb->nAutoCrash==0 ){
+      doSystemCrash(pDb);
+      pDb->bCrashed = 1;
+      return LSM_IOERR;
+    }
+  }
+
+  if( pDb->bPrepareCrash ){
+    for(i=0; i<pData->nSector; i++){
+      testFree(pData->aSector[i].aOld);
+      pData->aSector[i].aOld = 0;
+    }
+  }
+
+  if( pDb->xWriteHook ){
+    int rc;
+    int nUs;
+    struct timeval t1;
+    struct timeval t2;
+
+    gettimeofday(&t1, 0);
+    rc = pRealEnv->xSync(p->pReal);
+    gettimeofday(&t2, 0);
+
+    nUs = (t2.tv_sec - t1.tv_sec) * 1000000 + (t2.tv_usec - t1.tv_usec);
+    pDb->xWriteHook(pDb->pWriteCtx, p->bLog, 0, 0, nUs);
+    return rc;
+  }
+
+  return pRealEnv->xSync(p->pReal);
+}
+
+static int testEnvTruncate(lsm_file *pFile, lsm_i64 iOff){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+  if( p->pDb->bCrashed ) return LSM_IOERR;
+  return pRealEnv->xTruncate(p->pReal, iOff);
+}
+
+static int testEnvSectorSize(lsm_file *pFile){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+  return pRealEnv->xSectorSize(p->pReal);
+}
+
+static int testEnvRemap(
+  lsm_file *pFile, 
+  lsm_i64 iMin, 
+  void **ppOut,
+  lsm_i64 *pnOut
+){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+  return pRealEnv->xRemap(p->pReal, iMin, ppOut, pnOut);
+}
+
+static int testEnvFileid(
+  lsm_file *pFile, 
+  void *ppOut,
+  int *pnOut
+){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+  return pRealEnv->xFileid(p->pReal, ppOut, pnOut);
+}
+
+static int testEnvClose(lsm_file *pFile){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  LsmFile *p = (LsmFile *)pFile;
+
+  pRealEnv->xClose(p->pReal);
+  testFree(p);
+  return LSM_OK;
+}
+
+static int testEnvUnlink(lsm_env *pEnv, const char *zFile){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  unused_parameter(pEnv);
+  return pRealEnv->xUnlink(pRealEnv, zFile);
+}
+
+static int testEnvLock(lsm_file *pFile, int iLock, int eType){
+  LsmFile *p = (LsmFile *)pFile;
+  lsm_env *pRealEnv = tdb_lsm_env();
+
+  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
+    return LSM_BUSY;
+  }
+  return pRealEnv->xLock(p->pReal, iLock, eType);
+}
+
+static int testEnvTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
+  LsmFile *p = (LsmFile *)pFile;
+  lsm_env *pRealEnv = tdb_lsm_env();
+
+  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
+    return LSM_BUSY;
+  }
+  return pRealEnv->xTestLock(p->pReal, iLock, nLock, eType);
+}
+
+static int testEnvShmMap(lsm_file *pFile, int iRegion, int sz, void **pp){
+  LsmFile *p = (LsmFile *)pFile;
+  lsm_env *pRealEnv = tdb_lsm_env();
+  return pRealEnv->xShmMap(p->pReal, iRegion, sz, pp);
+}
+
+static void testEnvShmBarrier(void){
+}
+
+static int testEnvShmUnmap(lsm_file *pFile, int bDel){
+  LsmFile *p = (LsmFile *)pFile;
+  lsm_env *pRealEnv = tdb_lsm_env();
+  return pRealEnv->xShmUnmap(p->pReal, bDel);
+}
+
+static int testEnvSleep(lsm_env *pEnv, int us){
+  lsm_env *pRealEnv = tdb_lsm_env();
+  return pRealEnv->xSleep(pRealEnv, us);
+}
+
+static void doSystemCrash(LsmDb *pDb){
+  lsm_env *pEnv = tdb_lsm_env();
+  int iFile;
+  int iSeed = pDb->aFile[0].nSector + pDb->aFile[1].nSector;
+
+  char *zFile = pDb->zName;
+  char *zFree = 0;
+
+  for(iFile=0; iFile<2; iFile++){
+    lsm_file *pFile = 0;
+    int i;
+
+    pEnv->xOpen(pEnv, zFile, 0, &pFile);
+    for(i=0; i<pDb->aFile[iFile].nSector; i++){
+      u8 *aOld = pDb->aFile[iFile].aSector[i].aOld;
+      if( aOld ){
+        int iOpt = testPrngValue(iSeed++) % 3;
+        switch( iOpt ){
+          case 0:
+            break;
+
+          case 1:
+            testPrngArray(iSeed++, (u32 *)aOld, pDb->szSector/4);
+            /* Fall-through */
+
+          case 2:
+            pEnv->xWrite(
+                pFile, (lsm_i64)i * pDb->szSector, aOld, pDb->szSector
+            );
+            break;
+        }
+        testFree(aOld);
+        pDb->aFile[iFile].aSector[i].aOld = 0;
+      }
+    }
+    pEnv->xClose(pFile);
+    zFree = zFile = sqlite3_mprintf("%s-log", pDb->zName);
+  }
+
+  sqlite3_free(zFree);
+}
+/*
+** End test VFS code.
+**************************************************************************
+*************************************************************************/
+
+/*************************************************************************
+**************************************************************************
+** Begin test compression hooks.
+*/
+
+#ifdef HAVE_ZLIB
+#include <zlib.h>
+
+static int testZipBound(void *pCtx, int nSrc){
+  return compressBound(nSrc);
+}
+
+static int testZipCompress(
+  void *pCtx,                     /* Context pointer */
+  char *aOut, int *pnOut,         /* OUT: Buffer containing compressed data */
+  const char *aIn, int nIn        /* Buffer containing input data */
+){
+  uLongf n = *pnOut;              /* In/out buffer size for compress() */
+  int rc;                         /* compress() return code */
+ 
+  rc = compress((Bytef*)aOut, &n, (Bytef*)aIn, nIn);
+  *pnOut = n;
+  return (rc==Z_OK ? 0 : LSM_ERROR);
+}
+
+static int testZipUncompress(
+  void *pCtx,                     /* Context pointer */
+  char *aOut, int *pnOut,         /* OUT: Buffer containing uncompressed data */
+  const char *aIn, int nIn        /* Buffer containing input data */
+){
+  uLongf n = *pnOut;              /* In/out buffer size for uncompress() */
+  int rc;                         /* uncompress() return code */
+
+  rc = uncompress((Bytef*)aOut, &n, (Bytef*)aIn, nIn);
+  *pnOut = n;
+  return (rc==Z_OK ? 0 : LSM_ERROR);
+}
+
+static int testConfigureCompression(lsm_db *pDb){
+  static lsm_compress zip = {
+    0,                            /* Context pointer (unused) */
+    1,                            /* Id value */
+    testZipBound,                 /* xBound method */
+    testZipCompress,              /* xCompress method */
+    testZipUncompress             /* xUncompress method */
+  };
+  return lsm_config(pDb, LSM_CONFIG_SET_COMPRESSION, &zip);
+}
+#endif /* ifdef HAVE_ZLIB */
+
+/*
+** End test compression hooks.
+**************************************************************************
+*************************************************************************/
+
+static int test_lsm_close(TestDb *pTestDb){
+  int i;
+  int rc = LSM_OK;
+  LsmDb *pDb = (LsmDb *)pTestDb;
+
+  lsm_csr_close(pDb->pCsr);
+  lsm_close(pDb->db);
+
+  /* If this is a multi-threaded database, wait on the worker threads. */
+  mt_shutdown(pDb);
+  for(i=0; i<pDb->nWorker && rc==LSM_OK; i++){
+    rc = pDb->aWorker[i].worker_rc;
+  }
+
+  for(i=0; i<pDb->aFile[0].nSector; i++){
+    testFree(pDb->aFile[0].aSector[i].aOld);
+  }
+  testFree(pDb->aFile[0].aSector);
+  for(i=0; i<pDb->aFile[1].nSector; i++){
+    testFree(pDb->aFile[1].aSector[i].aOld);
+  }
+  testFree(pDb->aFile[1].aSector);
+
+  memset(pDb, sizeof(LsmDb), 0x11);
+  testFree((char *)pDb->pBuf);
+  testFree((char *)pDb);
+  return rc;
+}
+
+static void mt_signal_worker(LsmDb*, int);
+
+static int waitOnCheckpointer(LsmDb *pDb, lsm_db *db){
+  int nSleep = 0;
+  int nKB;
+  int rc;
+
+  do {
+    nKB = 0;
+    rc = lsm_info(db, LSM_INFO_CHECKPOINT_SIZE, &nKB);
+    if( rc!=LSM_OK || nKB<pDb->nMtMaxCkpt ) break;
+#ifdef LSM_MUTEX_PTHREADS
+    mt_signal_worker(pDb, 
+        (pDb->eMode==LSMTEST_MODE_BACKGROUND_CKPT ? 0 : 1)
+    );
+#endif
+    usleep(5000);
+    nSleep += 5;
+  }while( 1 );
+
+#if 0
+    if( nSleep ) printf("# waitOnCheckpointer(): nSleep=%d\n", nSleep);
+#endif
+
+  return rc;
+}
+
+static int waitOnWorker(LsmDb *pDb){
+  int rc;
+  int nLimit = -1;
+  int nSleep = 0;
+
+  rc = lsm_config(pDb->db, LSM_CONFIG_AUTOFLUSH, &nLimit);
+  do {
+    int nOld, nNew, rc2;
+    rc2 = lsm_info(pDb->db, LSM_INFO_TREE_SIZE, &nOld, &nNew);
+    if( rc2!=LSM_OK ) return rc2;
+    if( nOld==0 || nNew<(nLimit/2) ) break;
+#ifdef LSM_MUTEX_PTHREADS
+    mt_signal_worker(pDb, 0);
+#endif
+    usleep(5000);
+    nSleep += 5;
+  }while( 1 );
+
+#if 0
+  if( nSleep ) printf("# waitOnWorker(): nSleep=%d\n", nSleep);
+#endif
+
+  return rc;
+}
+
+static int test_lsm_write(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void *pVal,
+  int nVal
+){
+  LsmDb *pDb = (LsmDb *)pTestDb;
+  int rc = LSM_OK;
+
+  if( pDb->eMode==LSMTEST_MODE_BACKGROUND_CKPT ){
+    rc = waitOnCheckpointer(pDb, pDb->db);
+  }else if( 
+      pDb->eMode==LSMTEST_MODE_BACKGROUND_WORK
+   || pDb->eMode==LSMTEST_MODE_BACKGROUND_BOTH 
+  ){
+    rc = waitOnWorker(pDb);
+  }
+
+  if( rc==LSM_OK ){
+    rc = lsm_insert(pDb->db, pKey, nKey, pVal, nVal);
+  }
+  return rc;
+}
+
+static int test_lsm_delete(TestDb *pTestDb, void *pKey, int nKey){
+  LsmDb *pDb = (LsmDb *)pTestDb;
+  return lsm_delete(pDb->db, pKey, nKey);
+}
+
+static int test_lsm_delete_range(
+  TestDb *pTestDb, 
+  void *pKey1, int nKey1,
+  void *pKey2, int nKey2
+){
+  LsmDb *pDb = (LsmDb *)pTestDb;
+  return lsm_delete_range(pDb->db, pKey1, nKey1, pKey2, nKey2);
+}
+
+static int test_lsm_fetch(
+  TestDb *pTestDb, 
+  void *pKey, 
+  int nKey, 
+  void **ppVal, 
+  int *pnVal
+){
+  int rc;
+  LsmDb *pDb = (LsmDb *)pTestDb;
+  lsm_cursor *csr;
+
+  if( pKey==0 ) return LSM_OK;
+
+  if( pDb->pCsr==0 ){
+    rc = lsm_csr_open(pDb->db, &csr);
+    if( rc!=LSM_OK ) return rc;
+  }else{
+    csr = pDb->pCsr;
+  }
+
+  rc = lsm_csr_seek(csr, pKey, nKey, LSM_SEEK_EQ);
+  if( rc==LSM_OK ){
+    if( lsm_csr_valid(csr) ){
+      const void *pVal; int nVal;
+      rc = lsm_csr_value(csr, &pVal, &nVal);
+      if( nVal>pDb->nBuf ){
+        testFree(pDb->pBuf);
+        pDb->pBuf = testMalloc(nVal*2);
+        pDb->nBuf = nVal*2;
+      }
+      memcpy(pDb->pBuf, pVal, nVal);
+      *ppVal = pDb->pBuf;
+      *pnVal = nVal;
+    }else{
+      *ppVal = 0;
+      *pnVal = -1;
+    }
+  }
+  if( pDb->pCsr==0 ){
+    lsm_csr_close(csr);
+  }
+  return rc;
+}
+
+static int test_lsm_scan(
+  TestDb *pTestDb,
+  void *pCtx,
+  int bReverse,
+  void *pFirst, int nFirst,
+  void *pLast, int nLast,
+  void (*xCallback)(void *, void *, int , void *, int)
+){
+  LsmDb *pDb = (LsmDb *)pTestDb;
+  lsm_cursor *csr;
+  lsm_cursor *csr2 = 0;
+  int rc;
+
+  if( pDb->pCsr==0 ){
+    rc = lsm_csr_open(pDb->db, &csr);
+    if( rc!=LSM_OK ) return rc;
+  }else{
+    rc = LSM_OK;
+    csr = pDb->pCsr;
+  }
+
+  /* To enhance testing, if both pLast and pFirst are defined, seek the
+  ** cursor to the "end" boundary here. Then the next block seeks it to
+  ** the "start" ready for the scan. The point is to test that cursors
+  ** can be reused.  */
+  if( pLast && pFirst ){
+    if( bReverse ){
+      rc = lsm_csr_seek(csr, pFirst, nFirst, LSM_SEEK_LE);
+    }else{
+      rc = lsm_csr_seek(csr, pLast, nLast, LSM_SEEK_GE);
+    }
+  }
+
+  if( bReverse ){
+    if( pLast ){
+      rc = lsm_csr_seek(csr, pLast, nLast, LSM_SEEK_LE);
+    }else{
+      rc = lsm_csr_last(csr);
+    }
+  }else{
+    if( pFirst ){
+      rc = lsm_csr_seek(csr, pFirst, nFirst, LSM_SEEK_GE);
+    }else{
+      rc = lsm_csr_first(csr);
+    }
+  }
+
+  while( rc==LSM_OK && lsm_csr_valid(csr) ){
+    const void *pKey; int nKey;
+    const void *pVal; int nVal;
+    int cmp;
+
+    lsm_csr_key(csr, &pKey, &nKey);
+    lsm_csr_value(csr, &pVal, &nVal);
+
+    if( bReverse && pFirst ){
+      cmp = memcmp(pFirst, pKey, MIN(nKey, nFirst));
+      if( cmp>0 || (cmp==0 && nFirst>nKey) ) break;
+    }else if( bReverse==0 && pLast ){
+      cmp = memcmp(pLast, pKey, MIN(nKey, nLast));
+      if( cmp<0 || (cmp==0 && nLast<nKey) ) break;
+    }
+
+    xCallback(pCtx, (void *)pKey, nKey, (void *)pVal, nVal);
+
+    if( bReverse ){
+      rc = lsm_csr_prev(csr);
+    }else{
+      rc = lsm_csr_next(csr);
+    }
+  }
+
+  if( pDb->pCsr==0 ){
+    lsm_csr_close(csr);
+  }
+  return rc;
+}
+
+static int test_lsm_begin(TestDb *pTestDb, int iLevel){
+  int rc = LSM_OK;
+  LsmDb *pDb = (LsmDb *)pTestDb;
+
+  /* iLevel==0 is a no-op. */
+  if( iLevel==0 ) return 0;
+
+  if( pDb->pCsr==0 ) rc = lsm_csr_open(pDb->db, &pDb->pCsr);
+  if( rc==LSM_OK && iLevel>1 ){
+    rc = lsm_begin(pDb->db, iLevel-1);
+  }
+
+  return rc;
+}
+static int test_lsm_commit(TestDb *pTestDb, int iLevel){
+  LsmDb *pDb = (LsmDb *)pTestDb;
+
+  /* If iLevel==0, close any open read transaction */
+  if( iLevel==0 && pDb->pCsr ){
+    lsm_csr_close(pDb->pCsr);
+    pDb->pCsr = 0;
+  }
+
+  /* If iLevel==0, close any open read transaction */
+  return lsm_commit(pDb->db, MAX(0, iLevel-1));
+}
+static int test_lsm_rollback(TestDb *pTestDb, int iLevel){
+  LsmDb *pDb = (LsmDb *)pTestDb;
+
+  /* If iLevel==0, close any open read transaction */
+  if( iLevel==0 && pDb->pCsr ){
+    lsm_csr_close(pDb->pCsr);
+    pDb->pCsr = 0;
+  }
+
+  return lsm_rollback(pDb->db, MAX(0, iLevel-1));
+}
+
+/*
+** A log message callback registered with lsm connections. Prints all 
+** messages to stderr.
+*/
+static void xLog(void *pCtx, int rc, const char *z){
+  unused_parameter(rc);
+  /* fprintf(stderr, "lsm: rc=%d \"%s\"\n", rc, z); */
+  if( pCtx ) fprintf(stderr, "%s: ", (char *)pCtx);
+  fprintf(stderr, "%s\n", z);
+  fflush(stderr);
+}
+
+static void xWorkHook(lsm_db *db, void *pArg){
+  LsmDb *p = (LsmDb *)pArg;
+  if( p->xWork ) p->xWork(db, p->pWorkCtx);
+}
+
+#define TEST_NO_RECOVERY -1
+#define TEST_COMPRESSION -3
+
+#define TEST_MT_MODE     -2
+#define TEST_MT_MIN_CKPT -4
+#define TEST_MT_MAX_CKPT -5
+
+
+int test_lsm_config_str(
+  LsmDb *pLsm,
+  lsm_db *db, 
+  int bWorker,
+  const char *zStr,
+  int *pnThread
+){
+  struct CfgParam {
+    const char *zParam;
+    int bWorker;
+    int eParam;
+  } aParam[] = {
+    { "autoflush",        0, LSM_CONFIG_AUTOFLUSH },
+    { "page_size",        0, LSM_CONFIG_PAGE_SIZE },
+    { "block_size",       0, LSM_CONFIG_BLOCK_SIZE },
+    { "safety",           0, LSM_CONFIG_SAFETY },
+    { "autowork",         0, LSM_CONFIG_AUTOWORK },
+    { "autocheckpoint",   0, LSM_CONFIG_AUTOCHECKPOINT },
+    { "mmap",             0, LSM_CONFIG_MMAP },
+    { "use_log",          0, LSM_CONFIG_USE_LOG },
+    { "automerge",        0, LSM_CONFIG_AUTOMERGE },
+    { "max_freelist",     0, LSM_CONFIG_MAX_FREELIST },
+    { "multi_proc",       0, LSM_CONFIG_MULTIPLE_PROCESSES },
+    { "worker_automerge", 1, LSM_CONFIG_AUTOMERGE },
+    { "test_no_recovery", 0, TEST_NO_RECOVERY },
+    { "bg_min_ckpt",      0, TEST_NO_RECOVERY },
+
+    { "mt_mode",          0, TEST_MT_MODE },
+    { "mt_min_ckpt",      0, TEST_MT_MIN_CKPT },
+    { "mt_max_ckpt",      0, TEST_MT_MAX_CKPT },
+
+#ifdef HAVE_ZLIB
+    { "compression",      0, TEST_COMPRESSION },
+#endif
+    { 0, 0 }
+  };
+  const char *z = zStr;
+  int nThread = 1;
+
+  if( zStr==0 ) return 0;
+
+  assert( db );
+  while( z[0] ){
+    const char *zStart;
+
+    /* Skip whitespace */
+    while( *z==' ' ) z++;
+    zStart = z;
+
+    while( *z && *z!='=' ) z++;
+    if( *z ){
+      int eParam;
+      int i;
+      int iVal;
+      int iMul = 1;
+      int rc;
+      char zParam[32];
+      int nParam = z-zStart;
+      if( nParam==0 || nParam>sizeof(zParam)-1 ) goto syntax_error;
+
+      memcpy(zParam, zStart, nParam);
+      zParam[nParam] = '\0';
+      rc = testArgSelect(aParam, "param", zParam, &i);
+      if( rc!=0 ) return rc;
+      eParam = aParam[i].eParam;
+
+      z++;
+      zStart = z;
+      while( *z>='0' && *z<='9' ) z++;
+      if( *z=='k' || *z=='K' ){
+        iMul = 1;
+        z++;
+      }else if( *z=='M' || *z=='M' ){
+        iMul = 1024;
+        z++;
+      }
+      nParam = z-zStart;
+      if( nParam==0 || nParam>sizeof(zParam)-1 ) goto syntax_error;
+      memcpy(zParam, zStart, nParam);
+      zParam[nParam] = '\0';
+      iVal = atoi(zParam) * iMul;
+
+      if( eParam>0 ){
+        if( bWorker || aParam[i].bWorker==0 ){
+          lsm_config(db, eParam, &iVal);
+        }
+      }else{
+        switch( eParam ){
+          case TEST_NO_RECOVERY:
+            if( pLsm ) pLsm->bNoRecovery = iVal;
+            break;
+          case TEST_MT_MODE:
+            if( pLsm ) nThread = iVal;
+            break;
+          case TEST_MT_MIN_CKPT:
+            if( pLsm && iVal>0 ) pLsm->nMtMinCkpt = iVal*1024;
+            break;
+          case TEST_MT_MAX_CKPT:
+            if( pLsm && iVal>0 ) pLsm->nMtMaxCkpt = iVal*1024;
+            break;
+#ifdef HAVE_ZLIB
+          case TEST_COMPRESSION:
+            testConfigureCompression(db);
+            break;
+#endif
+        }
+      }
+    }else if( z!=zStart ){
+      goto syntax_error;
+    }
+  }
+
+  if( pnThread ) *pnThread = nThread;
+  if( pLsm && pLsm->nMtMaxCkpt < pLsm->nMtMinCkpt ){
+    pLsm->nMtMinCkpt = pLsm->nMtMaxCkpt;
+  }
+
+  return 0;
+ syntax_error:
+  testPrintError("syntax error at: \"%s\"\n", z);
+  return 1;
+}
+
+int tdb_lsm_config_str(TestDb *pDb, const char *zStr){
+  int rc = 0;
+  if( tdb_lsm(pDb) ){
+#ifdef LSM_MUTEX_PTHREADS
+    int i;
+#endif
+    LsmDb *pLsm = (LsmDb *)pDb;
+
+    rc = test_lsm_config_str(pLsm, pLsm->db, 0, zStr, 0);
+#ifdef LSM_MUTEX_PTHREADS
+    for(i=0; rc==0 && i<pLsm->nWorker; i++){
+      rc = test_lsm_config_str(0, pLsm->aWorker[i].pWorker, 1, zStr, 0);
+    }
+#endif
+  }
+  return rc;
+}
+
+int tdb_lsm_configure(lsm_db *db, const char *zConfig){
+  return test_lsm_config_str(0, db, 0, zConfig, 0);
+}
+
+static int testLsmStartWorkers(LsmDb *, int, const char *, const char *);
+
+static int testLsmOpen(
+  const char *zCfg,
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  static const DatabaseMethods LsmMethods = {
+    test_lsm_close,
+    test_lsm_write,
+    test_lsm_delete,
+    test_lsm_delete_range,
+    test_lsm_fetch,
+    test_lsm_scan,
+    test_lsm_begin,
+    test_lsm_commit,
+    test_lsm_rollback
+  };
+
+  int rc;
+  int nFilename;
+  LsmDb *pDb;
+
+  /* If the bClear flag is set, delete any existing database. */
+  assert( zFilename);
+  if( bClear ) testDeleteLsmdb(zFilename);
+  nFilename = strlen(zFilename);
+
+  pDb = (LsmDb *)testMalloc(sizeof(LsmDb) + nFilename + 1);
+  memset(pDb, 0, sizeof(LsmDb));
+  pDb->base.pMethods = &LsmMethods;
+  pDb->zName = (char *)&pDb[1];
+  memcpy(pDb->zName, zFilename, nFilename + 1);
+
+  /* Default the sector size used for crash simulation to 512 bytes. 
+  ** Todo: There should be an OS method to obtain this value - just as
+  ** there is in SQLite. For now, LSM assumes that it is smaller than
+  ** the page size (default 4KB).
+  */
+  pDb->szSector = 256;
+
+  /* Default values for the mt_min_ckpt and mt_max_ckpt parameters. */
+  pDb->nMtMinCkpt = LSMTEST_DFLT_MT_MIN_CKPT;
+  pDb->nMtMaxCkpt = LSMTEST_DFLT_MT_MAX_CKPT;
+
+  memcpy(&pDb->env, tdb_lsm_env(), sizeof(lsm_env));
+  pDb->env.pVfsCtx = (void *)pDb;
+  pDb->env.xFullpath = testEnvFullpath;
+  pDb->env.xOpen = testEnvOpen;
+  pDb->env.xRead = testEnvRead;
+  pDb->env.xWrite = testEnvWrite;
+  pDb->env.xTruncate = testEnvTruncate;
+  pDb->env.xSync = testEnvSync;
+  pDb->env.xSectorSize = testEnvSectorSize;
+  pDb->env.xRemap = testEnvRemap;
+  pDb->env.xFileid = testEnvFileid;
+  pDb->env.xClose = testEnvClose;
+  pDb->env.xUnlink = testEnvUnlink;
+  pDb->env.xLock = testEnvLock;
+  pDb->env.xTestLock = testEnvTestLock;
+  pDb->env.xShmBarrier = testEnvShmBarrier;
+  pDb->env.xShmMap = testEnvShmMap;
+  pDb->env.xShmUnmap = testEnvShmUnmap;
+  pDb->env.xSleep = testEnvSleep;
+
+  rc = lsm_new(&pDb->env, &pDb->db);
+  if( rc==LSM_OK ){
+    int nThread = 1;
+    lsm_config_log(pDb->db, xLog, 0);
+    lsm_config_work_hook(pDb->db, xWorkHook, (void *)pDb);
+
+    rc = test_lsm_config_str(pDb, pDb->db, 0, zCfg, &nThread);
+    if( rc==LSM_OK ) rc = lsm_open(pDb->db, zFilename);
+
+    pDb->eMode = nThread;
+#ifdef LSM_MUTEX_PTHREADS
+    if( rc==LSM_OK && nThread>1 ){
+      testLsmStartWorkers(pDb, nThread, zFilename, zCfg);
+    }
+#endif
+
+    if( rc!=LSM_OK ){
+      test_lsm_close((TestDb *)pDb);
+      pDb = 0;
+    }
+  }
+
+  *ppDb = (TestDb *)pDb;
+  return rc;
+}
+
+int test_lsm_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  return testLsmOpen(zSpec, zFilename, bClear, ppDb);
+}
+
+int test_lsm_small_open(
+  const char *zSpec, 
+  const char *zFile, 
+  int bClear, 
+  TestDb **ppDb
+){
+  const char *zCfg = "page_size=256 block_size=64 mmap=1024";
+  return testLsmOpen(zCfg, zFile, bClear, ppDb);
+}
+
+int test_lsm_lomem_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+    /* "max_freelist=4 autocheckpoint=32" */
+  const char *zCfg = 
+    "page_size=256 block_size=64 autoflush=16 "
+    "autocheckpoint=32"
+    "mmap=0 "
+  ;
+  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+int test_lsm_lomem2_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+    /* "max_freelist=4 autocheckpoint=32" */
+  const char *zCfg = 
+    "page_size=512 block_size=64 autoflush=0 mmap=0 "
+  ;
+  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+int test_lsm_zip_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  const char *zCfg = 
+    "page_size=256 block_size=64 autoflush=16 "
+    "autocheckpoint=32 compression=1 mmap=0 "
+  ;
+  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+lsm_db *tdb_lsm(TestDb *pDb){
+  if( pDb->pMethods->xClose==test_lsm_close ){
+    return ((LsmDb *)pDb)->db;
+  }
+  return 0;
+}
+
+int tdb_lsm_multithread(TestDb *pDb){
+  int ret = 0;
+  if( tdb_lsm(pDb) ){
+    ret = ((LsmDb*)pDb)->eMode!=LSMTEST_MODE_SINGLETHREAD;
+  }
+  return ret;
+}
+
+void tdb_lsm_enable_log(TestDb *pDb, int bEnable){
+  lsm_db *db = tdb_lsm(pDb);
+  if( db ){
+    lsm_config_log(db, (bEnable ? xLog : 0), (void *)"client");
+  }
+}
+
+void tdb_lsm_application_crash(TestDb *pDb){
+  if( tdb_lsm(pDb) ){
+    LsmDb *p = (LsmDb *)pDb;
+    p->bCrashed = 1;
+  }
+}
+
+void tdb_lsm_prepare_system_crash(TestDb *pDb){
+  if( tdb_lsm(pDb) ){
+    LsmDb *p = (LsmDb *)pDb;
+    p->bPrepareCrash = 1;
+  }
+}
+
+void tdb_lsm_system_crash(TestDb *pDb){
+  if( tdb_lsm(pDb) ){
+    LsmDb *p = (LsmDb *)pDb;
+    p->bCrashed = 1;
+    doSystemCrash(p);
+  }
+}
+
+void tdb_lsm_safety(TestDb *pDb, int eMode){
+  assert( eMode==LSM_SAFETY_OFF 
+       || eMode==LSM_SAFETY_NORMAL 
+       || eMode==LSM_SAFETY_FULL 
+  );
+  if( tdb_lsm(pDb) ){
+    int iParam = eMode;
+    LsmDb *p = (LsmDb *)pDb;
+    lsm_config(p->db, LSM_CONFIG_SAFETY, &iParam);
+  }
+}
+
+void tdb_lsm_prepare_sync_crash(TestDb *pDb, int iSync){
+  assert( iSync>0 );
+  if( tdb_lsm(pDb) ){
+    LsmDb *p = (LsmDb *)pDb;
+    p->nAutoCrash = iSync;
+    p->bPrepareCrash = 1;
+  }
+}
+
+void tdb_lsm_config_work_hook(
+  TestDb *pDb, 
+  void (*xWork)(lsm_db *, void *), 
+  void *pWorkCtx
+){
+  if( tdb_lsm(pDb) ){
+    LsmDb *p = (LsmDb *)pDb;
+    p->xWork = xWork;
+    p->pWorkCtx = pWorkCtx;
+  }
+}
+
+void tdb_lsm_write_hook(
+  TestDb *pDb, 
+  void (*xWrite)(void *, int, lsm_i64, int, int),
+  void *pWriteCtx
+){
+  if( tdb_lsm(pDb) ){
+    LsmDb *p = (LsmDb *)pDb;
+    p->xWriteHook = xWrite;
+    p->pWriteCtx = pWriteCtx;
+  }
+}
+
+int tdb_lsm_open(const char *zCfg, const char *zDb, int bClear, TestDb **ppDb){
+  return testLsmOpen(zCfg, zDb, bClear, ppDb);
+}
+
+#ifdef LSM_MUTEX_PTHREADS
+
+/*
+** Signal worker thread iWorker that there may be work to do.
+*/
+static void mt_signal_worker(LsmDb *pDb, int iWorker){
+  LsmWorker *p = &pDb->aWorker[iWorker];
+  pthread_mutex_lock(&p->worker_mutex);
+  p->bDoWork = 1;
+  pthread_cond_signal(&p->worker_cond);
+  pthread_mutex_unlock(&p->worker_mutex);
+}
+
+/*
+** This routine is used as the main() for all worker threads.
+*/
+static void *worker_main(void *pArg){
+  LsmWorker *p = (LsmWorker *)pArg;
+  lsm_db *pWorker;                /* Connection to access db through */
+
+  pthread_mutex_lock(&p->worker_mutex);
+  while( (pWorker = p->pWorker) ){
+    int rc = LSM_OK;
+
+    /* Do some work. If an error occurs, exit. */
+
+    pthread_mutex_unlock(&p->worker_mutex);
+    if( p->eType==LSMTEST_THREAD_CKPT ){
+      int nKB = 0;
+      rc = lsm_info(pWorker, LSM_INFO_CHECKPOINT_SIZE, &nKB);
+      if( rc==LSM_OK && nKB>=p->pDb->nMtMinCkpt ){
+        rc = lsm_checkpoint(pWorker, 0);
+      }
+    }else{
+      int nWrite;
+      do {
+
+        if( p->eType==LSMTEST_THREAD_WORKER ){
+          waitOnCheckpointer(p->pDb, pWorker);
+        }
+
+        nWrite = 0;
+        rc = lsm_work(pWorker, 0, 256, &nWrite);
+
+        if( p->eType==LSMTEST_THREAD_WORKER && nWrite ){
+          mt_signal_worker(p->pDb, 1);
+        }
+      }while( nWrite && p->pWorker );
+    }
+    pthread_mutex_lock(&p->worker_mutex);
+
+    if( rc!=LSM_OK && rc!=LSM_BUSY ){
+      p->worker_rc = rc;
+      break;
+    }
+
+    /* The thread will wake up when it is signaled either because another
+    ** thread has created some work for this one or because the connection
+    ** is being closed.  */
+    if( p->pWorker && p->bDoWork==0 ){
+      pthread_cond_wait(&p->worker_cond, &p->worker_mutex);
+    }
+    p->bDoWork = 0;
+  }
+  pthread_mutex_unlock(&p->worker_mutex);
+  
+  return 0;
+}
+
+
+static void mt_stop_worker(LsmDb *pDb, int iWorker){
+  LsmWorker *p = &pDb->aWorker[iWorker];
+  if( p->pWorker ){
+    void *pDummy;
+    lsm_db *pWorker;
+
+    /* Signal the worker to stop */
+    pthread_mutex_lock(&p->worker_mutex);
+    pWorker = p->pWorker;
+    p->pWorker = 0;
+    pthread_cond_signal(&p->worker_cond);
+    pthread_mutex_unlock(&p->worker_mutex);
+
+    /* Join the worker thread. */
+    pthread_join(p->worker_thread, &pDummy);
+
+    /* Free resources allocated in mt_start_worker() */
+    pthread_cond_destroy(&p->worker_cond);
+    pthread_mutex_destroy(&p->worker_mutex);
+    lsm_close(pWorker);
+  }
+}
+
+static void mt_shutdown(LsmDb *pDb){
+  int i;
+  for(i=0; i<pDb->nWorker; i++){
+    mt_stop_worker(pDb, i);
+  }
+}
+
+/*
+** This callback is invoked by LSM when the client database writes to
+** the database file (i.e. to flush the contents of the in-memory tree).
+** This implies there may be work to do on the database, so signal
+** the worker threads.
+*/
+static void mt_client_work_hook(lsm_db *db, void *pArg){
+  LsmDb *pDb = (LsmDb *)pArg;     /* LsmDb database handle */
+
+  /* Invoke the user level work-hook, if any. */
+  if( pDb->xWork ) pDb->xWork(db, pDb->pWorkCtx);
+
+  /* Wake up worker thread 0. */
+  mt_signal_worker(pDb, 0);
+}
+
+static void mt_worker_work_hook(lsm_db *db, void *pArg){
+  LsmDb *pDb = (LsmDb *)pArg;     /* LsmDb database handle */
+
+  /* Invoke the user level work-hook, if any. */
+  if( pDb->xWork ) pDb->xWork(db, pDb->pWorkCtx);
+}
+
+/*
+** Launch worker thread iWorker for database connection pDb.
+*/
+static int mt_start_worker(
+  LsmDb *pDb,                     /* Main database structure */
+  int iWorker,                    /* Worker number to start */
+  const char *zFilename,          /* File name of database to open */
+  const char *zCfg,               /* Connection configuration string */
+  int eType                       /* Type of worker thread */
+){
+  int rc = 0;                     /* Return code */
+  LsmWorker *p;                   /* Object to initialize */
+
+  assert( iWorker<pDb->nWorker );
+  assert( eType==LSMTEST_THREAD_CKPT 
+       || eType==LSMTEST_THREAD_WORKER 
+       || eType==LSMTEST_THREAD_WORKER_AC 
+  );
+
+  p = &pDb->aWorker[iWorker];
+  p->eType = eType;
+  p->pDb = pDb;
+
+  /* Open the worker connection */
+  if( rc==0 ) rc = lsm_new(&pDb->env, &p->pWorker);
+  if( zCfg ){
+    test_lsm_config_str(pDb, p->pWorker, 1, zCfg, 0);
+  }
+  if( rc==0 ) rc = lsm_open(p->pWorker, zFilename);
+  lsm_config_log(p->pWorker, xLog, (void *)"worker");
+
+  /* Configure the work-hook */
+  if( rc==0 ){
+    lsm_config_work_hook(p->pWorker, mt_worker_work_hook, (void *)pDb);
+  }
+
+  if( eType==LSMTEST_THREAD_WORKER ){
+    test_lsm_config_str(0, p->pWorker, 1, "autocheckpoint=0", 0);
+  }
+
+  /* Kick off the worker thread. */
+  if( rc==0 ) rc = pthread_cond_init(&p->worker_cond, 0);
+  if( rc==0 ) rc = pthread_mutex_init(&p->worker_mutex, 0);
+  if( rc==0 ) rc = pthread_create(&p->worker_thread, 0, worker_main, (void *)p);
+
+  return rc;
+}
+
+
+static int testLsmStartWorkers(
+  LsmDb *pDb, int eModel, const char *zFilename, const char *zCfg
+){
+  int rc;
+
+  if( eModel<1 || eModel>4 ) return 1;
+  if( eModel==1 ) return 0;
+
+  /* Configure a work-hook for the client connection. Worker 0 is signalled
+  ** every time the users connection writes to the database.  */
+  lsm_config_work_hook(pDb->db, mt_client_work_hook, (void *)pDb);
+
+  /* Allocate space for two worker connections. They may not both be
+  ** used, but both are allocated.  */
+  pDb->aWorker = (LsmWorker *)testMalloc(sizeof(LsmWorker) * 2);
+  memset(pDb->aWorker, 0, sizeof(LsmWorker) * 2);
+
+  switch( eModel ){
+    case LSMTEST_MODE_BACKGROUND_CKPT:
+      pDb->nWorker = 1;
+      test_lsm_config_str(0, pDb->db, 0, "autocheckpoint=0", 0);
+      rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_CKPT);
+      break;
+
+    case LSMTEST_MODE_BACKGROUND_WORK:
+      pDb->nWorker = 1;
+      test_lsm_config_str(0, pDb->db, 0, "autowork=0", 0);
+      rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_WORKER_AC);
+      break;
+
+    case LSMTEST_MODE_BACKGROUND_BOTH:
+      pDb->nWorker = 2;
+      test_lsm_config_str(0, pDb->db, 0, "autowork=0", 0);
+      rc = mt_start_worker(pDb, 0, zFilename, zCfg, LSMTEST_THREAD_WORKER);
+      if( rc==0 ){
+        rc = mt_start_worker(pDb, 1, zFilename, zCfg, LSMTEST_THREAD_CKPT);
+      }
+      break;
+  }
+
+  return rc;
+}
+
+
+int test_lsm_mt2(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  const char *zCfg = "mt_mode=2";
+  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+int test_lsm_mt3(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  const char *zCfg = "mt_mode=4";
+  return testLsmOpen(zCfg, zFilename, bClear, ppDb);
+}
+
+#else
+static void mt_shutdown(LsmDb *pDb) { 
+  unused_parameter(pDb); 
+}
+int test_lsm_mt(const char *zFilename, int bClear, TestDb **ppDb){
+  unused_parameter(zFilename);
+  unused_parameter(bClear);
+  unused_parameter(ppDb);
+  testPrintError("threads unavailable - recompile with LSM_MUTEX_PTHREADS\n");
+  return 1;
+}
+#endif
diff --git a/ext/lsm1/lsm-test/lsmtest_tdb4.c b/ext/lsm1/lsm-test/lsmtest_tdb4.c
new file mode 100644
index 0000000..1f92928
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_tdb4.c
@@ -0,0 +1,980 @@
+
+/*
+** This file contains the TestDb bt wrapper.
+*/
+
+#include "lsmtest_tdb.h"
+#include "lsmtest.h"
+#include <unistd.h>
+#include "bt.h"
+
+#include <pthread.h>
+
+typedef struct BtDb BtDb;
+typedef struct BtFile BtFile;
+
+/* Background checkpointer interface (see implementations below). */
+typedef struct bt_ckpter bt_ckpter;
+static int bgc_attach(BtDb *pDb, const char*);
+static int bgc_detach(BtDb *pDb);
+
+/*
+** Each database or log file opened by a database handle is wrapped by
+** an object of the following type.
+*/
+struct BtFile {
+  BtDb *pBt;                      /* Database handle that opened this file */
+  bt_env *pVfs;                   /* Underlying VFS */
+  bt_file *pFile;                 /* File handle belonging to underlying VFS */
+  int nSectorSize;                /* Size of sectors in bytes */
+  int nSector;                    /* Allocated size of nSector array */
+  u8 **apSector;                  /* Original sector data */
+};
+
+/*
+** nCrashSync:
+**   If this value is non-zero, then a "crash-test" is running. If
+**   nCrashSync==1, then the crash is simulated during the very next 
+**   call to the xSync() VFS method (on either the db or log file).
+**   If nCrashSync==2, the following call to xSync(), and so on.
+**
+** bCrash:
+**   After a crash is simulated, this variable is set. Any subsequent
+**   attempts to write to a file or modify the file system in any way 
+**   fail once this is set. All the caller can do is close the connection.
+**
+** bFastInsert:
+**   If this variable is set to true, then a BT_CONTROL_FAST_INSERT_OP
+**   control is issued before each callto BtReplace() or BtCsrOpen().
+*/
+struct BtDb {
+  TestDb base;                    /* Base class */
+  bt_db *pBt;                     /* bt database handle */
+  sqlite4_env *pEnv;              /* SQLite environment (for malloc/free) */
+  bt_env *pVfs;                   /* Underlying VFS */
+  int bFastInsert;                /* True to use fast-insert */
+
+  /* Space for bt_fetch() results */
+  u8 *aBuffer;                    /* Space to store results */
+  int nBuffer;                    /* Allocated size of aBuffer[] in bytes */
+  int nRef;
+
+  /* Background checkpointer used by mt connections */
+  bt_ckpter *pCkpter;
+
+  /* Stuff used for crash test simulation */
+  BtFile *apFile[2];              /* Database and log files used by pBt */
+  bt_env env;                     /* Private VFS for this object */
+  int nCrashSync;                 /* Number of syncs until crash (see above) */
+  int bCrash;                     /* True once a crash has been simulated */
+};
+
+static int btVfsFullpath(
+  sqlite4_env *pEnv, 
+  bt_env *pVfs, 
+  const char *z, 
+  char **pzOut
+){
+  BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
+  if( pBt->bCrash ) return SQLITE4_IOERR;
+  return pBt->pVfs->xFullpath(pEnv, pBt->pVfs, z, pzOut);
+}
+
+static int btVfsOpen(
+  sqlite4_env *pEnv, 
+  bt_env *pVfs, 
+  const char *zFile, 
+  int flags, bt_file **ppFile
+){
+  BtFile *p;
+  BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
+  int rc;
+
+  if( pBt->bCrash ) return SQLITE4_IOERR;
+
+  p = (BtFile*)testMalloc(sizeof(BtFile));
+  if( !p ) return SQLITE4_NOMEM;
+  if( flags & BT_OPEN_DATABASE ){
+    pBt->apFile[0] = p;
+  }else if( flags & BT_OPEN_LOG ){
+    pBt->apFile[1] = p;
+  }
+  if( (flags & BT_OPEN_SHARED)==0 ){
+    p->pBt = pBt; 
+  }
+  p->pVfs = pBt->pVfs; 
+
+  rc = pBt->pVfs->xOpen(pEnv, pVfs, zFile, flags, &p->pFile);
+  if( rc!=SQLITE4_OK ){
+    testFree(p);
+    p = 0;
+  }else{
+    pBt->nRef++;
+  }
+
+  *ppFile = (bt_file*)p;
+  return rc;
+}
+
+static int btVfsSize(bt_file *pFile, sqlite4_int64 *piRes){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  return p->pVfs->xSize(p->pFile, piRes);
+}
+
+static int btVfsRead(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  return p->pVfs->xRead(p->pFile, iOff, pBuf, nBuf);
+}
+
+static int btFlushSectors(BtFile *p, int iFile){
+  sqlite4_int64 iSz;
+  int rc;
+  int i;
+  u8 *aTmp = 0;
+
+  rc = p->pBt->pVfs->xSize(p->pFile, &iSz);
+  for(i=0; rc==SQLITE4_OK && i<p->nSector; i++){
+    if( p->pBt->bCrash && p->apSector[i] ){
+
+      /* The system is simulating a crash. There are three choices for
+      ** this sector:
+      **
+      **   1) Leave it as it is (simulating a successful write),
+      **   2) Restore the original data (simulating a lost write),
+      **   3) Populate the disk sector with garbage data.
+      */
+      sqlite4_int64 iSOff = p->nSectorSize*i;
+      int nWrite = MIN(p->nSectorSize, iSz - iSOff);
+
+      if( nWrite ){
+        u8 *aWrite = 0;
+        int iOpt = (testPrngValue(i) % 3) + 1;
+        if( iOpt==1 ){
+          aWrite = p->apSector[i];
+        }else if( iOpt==3 ){
+          if( aTmp==0 ) aTmp = testMalloc(p->nSectorSize);
+          aWrite = aTmp;
+          testPrngArray(i*13, (u32*)aWrite, nWrite/sizeof(u32));
+        }
+
+#if 0
+fprintf(stderr, "handle sector %d of %s with %s\n", i, 
+    iFile==0 ? "db" : "log",
+    iOpt==1 ? "rollback" : iOpt==2 ? "write" : "omit"
+);
+fflush(stderr);
+#endif
+
+        if( aWrite ){
+          rc = p->pBt->pVfs->xWrite(p->pFile, iSOff, aWrite, nWrite);
+        }
+      }
+    }
+    testFree(p->apSector[i]);
+    p->apSector[i] = 0;
+  }
+
+  testFree(aTmp);
+  return rc;
+}
+
+static int btSaveSectors(BtFile *p, sqlite4_int64 iOff, int nBuf){
+  int rc;
+  sqlite4_int64 iSz;              /* Size of file on disk */
+  int iFirst;                     /* First sector affected */
+  int iSector;                    /* Current sector */
+  int iLast;                      /* Last sector affected */
+
+  if( p->nSectorSize==0 ){
+    p->nSectorSize = p->pBt->pVfs->xSectorSize(p->pFile);
+    if( p->nSectorSize<512 ) p->nSectorSize = 512;
+  }
+  iLast = (iOff+nBuf-1) / p->nSectorSize;
+  iFirst = iOff / p->nSectorSize;
+
+  rc = p->pBt->pVfs->xSize(p->pFile, &iSz);
+  for(iSector=iFirst; rc==SQLITE4_OK && iSector<=iLast; iSector++){
+    int nRead;
+    sqlite4_int64 iSOff = iSector * p->nSectorSize;
+    u8 *aBuf = testMalloc(p->nSectorSize);
+    nRead = MIN(p->nSectorSize, (iSz - iSOff));
+    if( nRead>0 ){
+      rc = p->pBt->pVfs->xRead(p->pFile, iSOff, aBuf, nRead);
+    }
+
+    while( rc==SQLITE4_OK && iSector>=p->nSector ){
+      int nNew = p->nSector + 32;
+      u8 **apNew = (u8**)testMalloc(nNew * sizeof(u8*));
+      memcpy(apNew, p->apSector, p->nSector*sizeof(u8*));
+      testFree(p->apSector);
+      p->apSector = apNew;
+      p->nSector = nNew;
+    }
+
+    p->apSector[iSector] = aBuf;
+  }
+
+  return rc;
+}
+
+static int btVfsWrite(bt_file *pFile, sqlite4_int64 iOff, void *pBuf, int nBuf){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  if( p->pBt && p->pBt->nCrashSync ){
+    btSaveSectors(p, iOff, nBuf);
+  }
+  return p->pVfs->xWrite(p->pFile, iOff, pBuf, nBuf);
+}
+
+static int btVfsTruncate(bt_file *pFile, sqlite4_int64 iOff){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  return p->pVfs->xTruncate(p->pFile, iOff);
+}
+
+static int btVfsSync(bt_file *pFile){
+  int rc = SQLITE4_OK;
+  BtFile *p = (BtFile*)pFile;
+  BtDb *pBt = p->pBt;
+
+  if( pBt ){
+    if( pBt->bCrash ) return SQLITE4_IOERR;
+    if( pBt->nCrashSync ){
+      pBt->nCrashSync--;
+      pBt->bCrash = (pBt->nCrashSync==0);
+      if( pBt->bCrash ){
+        btFlushSectors(pBt->apFile[0], 0);
+        btFlushSectors(pBt->apFile[1], 1);
+        rc = SQLITE4_IOERR;
+      }else{
+        btFlushSectors(p, 0);
+      }
+    }
+  }
+
+  if( rc==SQLITE4_OK ){
+    rc = p->pVfs->xSync(p->pFile);
+  }
+  return rc;
+}
+
+static int btVfsSectorSize(bt_file *pFile){
+  BtFile *p = (BtFile*)pFile;
+  return p->pVfs->xSectorSize(p->pFile);
+}
+
+static void btDeref(BtDb *p){
+  p->nRef--;
+  assert( p->nRef>=0 );
+  if( p->nRef<=0 ) testFree(p);
+}
+
+static int btVfsClose(bt_file *pFile){
+  BtFile *p = (BtFile*)pFile;
+  BtDb *pBt = p->pBt;
+  int rc;
+  if( pBt ){
+    btFlushSectors(p, 0);
+    if( p==pBt->apFile[0] ) pBt->apFile[0] = 0;
+    if( p==pBt->apFile[1] ) pBt->apFile[1] = 0;
+  }
+  testFree(p->apSector);
+  rc = p->pVfs->xClose(p->pFile);
+#if 0
+  btDeref(p->pBt);
+#endif
+  testFree(p);
+  return rc;
+}
+
+static int btVfsUnlink(sqlite4_env *pEnv, bt_env *pVfs, const char *zFile){
+  BtDb *pBt = (BtDb*)pVfs->pVfsCtx;
+  if( pBt->bCrash ) return SQLITE4_IOERR;
+  return pBt->pVfs->xUnlink(pEnv, pBt->pVfs, zFile);
+}
+
+static int btVfsLock(bt_file *pFile, int iLock, int eType){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  return p->pVfs->xLock(p->pFile, iLock, eType);
+}
+
+static int btVfsTestLock(bt_file *pFile, int iLock, int nLock, int eType){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  return p->pVfs->xTestLock(p->pFile, iLock, nLock, eType);
+}
+
+static int btVfsShmMap(bt_file *pFile, int iChunk, int sz, void **ppOut){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  return p->pVfs->xShmMap(p->pFile, iChunk, sz, ppOut);
+}
+
+static void btVfsShmBarrier(bt_file *pFile){
+  BtFile *p = (BtFile*)pFile;
+  return p->pVfs->xShmBarrier(p->pFile);
+}
+
+static int btVfsShmUnmap(bt_file *pFile, int bDelete){
+  BtFile *p = (BtFile*)pFile;
+  if( p->pBt && p->pBt->bCrash ) return SQLITE4_IOERR;
+  return p->pVfs->xShmUnmap(p->pFile, bDelete);
+}
+
+static int bt_close(TestDb *pTestDb){
+  BtDb *p = (BtDb*)pTestDb;
+  int rc = sqlite4BtClose(p->pBt);
+  free(p->aBuffer);
+  if( p->apFile[0] ) p->apFile[0]->pBt = 0;
+  if( p->apFile[1] ) p->apFile[1]->pBt = 0;
+  bgc_detach(p);
+  testFree(p);
+  return rc;
+}
+
+static int btMinTransaction(BtDb *p, int iMin, int *piLevel){
+  int iLevel;
+  int rc = SQLITE4_OK;
+
+  iLevel = sqlite4BtTransactionLevel(p->pBt);
+  if( iLevel<iMin ){ 
+    rc = sqlite4BtBegin(p->pBt, iMin); 
+    *piLevel = iLevel;
+  }else{
+    *piLevel = -1;
+  }
+
+  return rc;
+}
+static int btRestoreTransaction(BtDb *p, int iLevel, int rcin){
+  int rc = rcin;
+  if( iLevel>=0 ){
+    if( rc==SQLITE4_OK ){
+      rc = sqlite4BtCommit(p->pBt, iLevel);
+    }else{
+      sqlite4BtRollback(p->pBt, iLevel);
+    }
+    assert( iLevel==sqlite4BtTransactionLevel(p->pBt) );
+  }
+  return rc;
+}
+
+static int bt_write(TestDb *pTestDb, void *pK, int nK, void *pV, int nV){
+  BtDb *p = (BtDb*)pTestDb;
+  int iLevel;
+  int rc;
+
+  rc = btMinTransaction(p, 2, &iLevel);
+  if( rc==SQLITE4_OK ){
+    if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+    rc = sqlite4BtReplace(p->pBt, pK, nK, pV, nV);
+    rc = btRestoreTransaction(p, iLevel, rc);
+  }
+  return rc;
+}
+
+static int bt_delete(TestDb *pTestDb, void *pK, int nK){
+  return bt_write(pTestDb, pK, nK, 0, -1);
+}
+
+static int bt_delete_range(
+  TestDb *pTestDb, 
+  void *pKey1, int nKey1,
+  void *pKey2, int nKey2
+){
+  BtDb *p = (BtDb*)pTestDb;
+  bt_cursor *pCsr = 0;
+  int rc = SQLITE4_OK;
+  int iLevel;
+
+  rc = btMinTransaction(p, 2, &iLevel);
+  if( rc==SQLITE4_OK ){
+    if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+    rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
+  }
+  while( rc==SQLITE4_OK ){
+    const void *pK;
+    int n;
+    int nCmp;
+    int res;
+
+    rc = sqlite4BtCsrSeek(pCsr, pKey1, nKey1, BT_SEEK_GE);
+    if( rc==SQLITE4_INEXACT ) rc = SQLITE4_OK;
+    if( rc!=SQLITE4_OK ) break;
+
+    rc = sqlite4BtCsrKey(pCsr, &pK, &n);
+    if( rc!=SQLITE4_OK ) break;
+
+    nCmp = MIN(n, nKey1);
+    res = memcmp(pKey1, pK, nCmp);
+    assert( res<0 || (res==0 && nKey1<=n) );
+    if( res==0 && nKey1==n ){
+      rc = sqlite4BtCsrNext(pCsr);
+      if( rc!=SQLITE4_OK ) break;
+      rc = sqlite4BtCsrKey(pCsr, &pK, &n);
+      if( rc!=SQLITE4_OK ) break;
+    }
+
+    nCmp = MIN(n, nKey2);
+    res = memcmp(pKey2, pK, nCmp);
+    if( res<0 || (res==0 && nKey2<=n) ) break;
+    
+    rc = sqlite4BtDelete(pCsr);
+  }
+  if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;
+
+  sqlite4BtCsrClose(pCsr);
+
+  rc = btRestoreTransaction(p, iLevel, rc);
+  return rc;
+}
+
+static int bt_fetch(
+  TestDb *pTestDb, 
+  void *pK, int nK, 
+  void **ppVal, int *pnVal
+){
+  BtDb *p = (BtDb*)pTestDb;
+  bt_cursor *pCsr = 0;
+  int iLevel;
+  int rc = SQLITE4_OK;
+
+  iLevel = sqlite4BtTransactionLevel(p->pBt);
+  if( iLevel==0 ){ 
+    rc = sqlite4BtBegin(p->pBt, 1); 
+    if( rc!=SQLITE4_OK ) return rc;
+  }
+
+  if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+  rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
+  if( rc==SQLITE4_OK ){
+    rc = sqlite4BtCsrSeek(pCsr, pK, nK, BT_SEEK_EQ);
+    if( rc==SQLITE4_OK ){
+      const void *pV = 0;
+      int nV = 0;
+      rc = sqlite4BtCsrData(pCsr, 0, -1, &pV, &nV);
+      if( rc==SQLITE4_OK ){
+        if( nV>p->nBuffer ){
+          free(p->aBuffer);
+          p->aBuffer = (u8*)malloc(nV*2);
+          p->nBuffer = nV*2;
+        }
+        memcpy(p->aBuffer, pV, nV);
+        *pnVal = nV;
+        *ppVal = (void*)(p->aBuffer);
+      }
+
+    }else if( rc==SQLITE4_INEXACT || rc==SQLITE4_NOTFOUND ){
+      *ppVal = 0;
+      *pnVal = -1;
+      rc = SQLITE4_OK;
+    }
+    sqlite4BtCsrClose(pCsr);
+  }
+
+  if( iLevel==0 ) sqlite4BtCommit(p->pBt, 0); 
+  return rc;
+}
+
+static int bt_scan(
+  TestDb *pTestDb,
+  void *pCtx,
+  int bReverse,
+  void *pFirst, int nFirst,
+  void *pLast, int nLast,
+  void (*xCallback)(void *, void *, int , void *, int)
+){
+  BtDb *p = (BtDb*)pTestDb;
+  bt_cursor *pCsr = 0;
+  int rc;
+  int iLevel;
+
+  rc = btMinTransaction(p, 1, &iLevel);
+
+  if( rc==SQLITE4_OK ){
+    if( p->bFastInsert ) sqlite4BtControl(p->pBt, BT_CONTROL_FAST_INSERT_OP, 0);
+    rc = sqlite4BtCsrOpen(p->pBt, 0, &pCsr);
+  }
+  if( rc==SQLITE4_OK ){
+    if( bReverse ){
+      if( pLast ){
+        rc = sqlite4BtCsrSeek(pCsr, pLast, nLast, BT_SEEK_LE);
+      }else{
+        rc = sqlite4BtCsrLast(pCsr);
+      }
+    }else{
+      rc = sqlite4BtCsrSeek(pCsr, pFirst, nFirst, BT_SEEK_GE);
+    }
+    if( rc==SQLITE4_INEXACT ) rc = SQLITE4_OK;
+
+    while( rc==SQLITE4_OK ){
+      const void *pK = 0; int nK = 0;
+      const void *pV = 0; int nV = 0;
+
+      rc = sqlite4BtCsrKey(pCsr, &pK, &nK);
+      if( rc==SQLITE4_OK ){
+        rc = sqlite4BtCsrData(pCsr, 0, -1, &pV, &nV);
+      }
+
+      if( rc!=SQLITE4_OK ) break;
+      if( bReverse ){
+        if( pFirst ){
+          int res;
+          int nCmp = MIN(nK, nFirst);
+          res = memcmp(pFirst, pK, nCmp);
+          if( res>0 || (res==0 && nK<nFirst) ) break;
+        }
+      }else{
+        if( pLast ){
+          int res;
+          int nCmp = MIN(nK, nLast);
+          res = memcmp(pLast, pK, nCmp);
+          if( res<0 || (res==0 && nK>nLast) ) break;
+        }
+      }
+
+      xCallback(pCtx, (void*)pK, nK, (void*)pV, nV);
+      if( bReverse ){
+        rc = sqlite4BtCsrPrev(pCsr);
+      }else{
+        rc = sqlite4BtCsrNext(pCsr);
+      }
+    }
+    if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;
+
+    sqlite4BtCsrClose(pCsr);
+  }
+
+  rc = btRestoreTransaction(p, iLevel, rc);
+  return rc;
+}
+
+static int bt_begin(TestDb *pTestDb, int iLvl){
+  BtDb *p = (BtDb*)pTestDb;
+  int rc = sqlite4BtBegin(p->pBt, iLvl);
+  return rc;
+}
+
+static int bt_commit(TestDb *pTestDb, int iLvl){
+  BtDb *p = (BtDb*)pTestDb;
+  int rc = sqlite4BtCommit(p->pBt, iLvl);
+  return rc;
+}
+
+static int bt_rollback(TestDb *pTestDb, int iLvl){
+  BtDb *p = (BtDb*)pTestDb;
+  int rc = sqlite4BtRollback(p->pBt, iLvl);
+  return rc;
+}
+
+static int testParseOption(
+  const char **pzIn,              /* IN/OUT: pointer to next option */
+  const char **pzOpt,             /* OUT: nul-terminated option name */
+  const char **pzArg,             /* OUT: nul-terminated option argument */
+  char *pSpace                    /* Temporary space for output params */
+){
+  const char *p = *pzIn;
+  const char *pStart;
+  int n;
+
+  char *pOut = pSpace;
+
+  while( *p==' ' ) p++;
+  pStart = p;
+  while( *p && *p!='=' ) p++;
+  if( *p==0 ) return 1;
+
+  n = (p - pStart);
+  memcpy(pOut, pStart, n);
+  *pzOpt = pOut;
+  pOut += n;
+  *pOut++ = '\0';
+
+  p++;
+  pStart = p;
+  while( *p && *p!=' ' ) p++;
+  n = (p - pStart);
+
+  memcpy(pOut, pStart, n);
+  *pzArg = pOut;
+  pOut += n;
+  *pOut++ = '\0';
+
+  *pzIn = p;
+  return 0;
+}
+
+static int testParseInt(const char *z, int *piVal){
+  int i = 0;
+  const char *p = z;
+
+  while( *p>='0' && *p<='9' ){
+    i = i*10 + (*p - '0');
+    p++;
+  }
+  if( *p=='K' || *p=='k' ){
+    i = i * 1024;
+    p++;
+  }else if( *p=='M' || *p=='m' ){
+    i = i * 1024 * 1024;
+    p++;
+  }
+
+  if( *p ) return SQLITE4_ERROR;
+  *piVal = i;
+  return SQLITE4_OK;
+}
+
+static int testBtConfigure(BtDb *pDb, const char *zCfg, int *pbMt){
+  int rc = SQLITE4_OK;
+
+  if( zCfg ){
+    struct CfgParam {
+      const char *zParam;
+      int eParam;
+    } aParam[] = {
+      { "safety",         BT_CONTROL_SAFETY },
+      { "autockpt",       BT_CONTROL_AUTOCKPT },
+      { "multiproc",      BT_CONTROL_MULTIPROC },
+      { "blksz",          BT_CONTROL_BLKSZ },
+      { "pagesz",         BT_CONTROL_PAGESZ },
+      { "mt",             -1 },
+      { "fastinsert",     -2 },
+      { 0, 0 }
+    };
+    const char *z = zCfg;
+    int n = strlen(z);
+    char *aSpace;
+    const char *zOpt;
+    const char *zArg;
+
+    aSpace = (char*)testMalloc(n+2);
+    while( rc==SQLITE4_OK && 0==testParseOption(&z, &zOpt, &zArg, aSpace) ){
+      int i;
+      int iVal;
+      rc = testArgSelect(aParam, "param", zOpt, &i);
+      if( rc!=SQLITE4_OK ) break;
+
+      rc = testParseInt(zArg, &iVal);
+      if( rc!=SQLITE4_OK ) break;
+
+      switch( aParam[i].eParam ){
+        case -1:
+          *pbMt = iVal;
+          break;
+        case -2:
+          pDb->bFastInsert = 1;
+          break;
+        default:
+          rc = sqlite4BtControl(pDb->pBt, aParam[i].eParam, (void*)&iVal);
+          break;
+      }
+    }
+    testFree(aSpace);
+  }
+
+  return rc;
+}
+
+
+int test_bt_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+
+  static const DatabaseMethods SqlMethods = {
+    bt_close,
+    bt_write,
+    bt_delete,
+    bt_delete_range,
+    bt_fetch,
+    bt_scan,
+    bt_begin,
+    bt_commit,
+    bt_rollback
+  };
+  BtDb *p = 0;
+  bt_db *pBt = 0;
+  int rc;
+  sqlite4_env *pEnv = sqlite4_env_default();
+
+  if( bClear && zFilename && zFilename[0] ){
+    char *zLog = sqlite3_mprintf("%s-wal", zFilename);
+    unlink(zFilename);
+    unlink(zLog);
+    sqlite3_free(zLog);
+  }
+  
+  rc = sqlite4BtNew(pEnv, 0, &pBt);
+  if( rc==SQLITE4_OK ){
+    int mt = 0;                   /* True for multi-threaded connection */
+
+    p = (BtDb*)testMalloc(sizeof(BtDb));
+    p->base.pMethods = &SqlMethods;
+    p->pBt = pBt;
+    p->pEnv = pEnv;
+    p->nRef = 1;
+
+    p->env.pVfsCtx = (void*)p;
+    p->env.xFullpath = btVfsFullpath;
+    p->env.xOpen = btVfsOpen;
+    p->env.xSize = btVfsSize;
+    p->env.xRead = btVfsRead;
+    p->env.xWrite = btVfsWrite;
+    p->env.xTruncate = btVfsTruncate;
+    p->env.xSync = btVfsSync;
+    p->env.xSectorSize = btVfsSectorSize;
+    p->env.xClose = btVfsClose;
+    p->env.xUnlink = btVfsUnlink;
+    p->env.xLock = btVfsLock;
+    p->env.xTestLock = btVfsTestLock;
+    p->env.xShmMap = btVfsShmMap;
+    p->env.xShmBarrier = btVfsShmBarrier;
+    p->env.xShmUnmap = btVfsShmUnmap;
+
+    sqlite4BtControl(pBt, BT_CONTROL_GETVFS, (void*)&p->pVfs);
+    sqlite4BtControl(pBt, BT_CONTROL_SETVFS, (void*)&p->env);
+
+    rc = testBtConfigure(p, zSpec, &mt);
+    if( rc==SQLITE4_OK ){
+      rc = sqlite4BtOpen(pBt, zFilename);
+    }
+
+    if( rc==SQLITE4_OK && mt ){
+      int nAuto = 0;
+      rc = bgc_attach(p, zSpec);
+      sqlite4BtControl(pBt, BT_CONTROL_AUTOCKPT, (void*)&nAuto);
+    }
+  }
+
+  if( rc!=SQLITE4_OK && p ){
+    bt_close(&p->base);
+  }
+
+  *ppDb = &p->base;
+  return rc;
+}
+
+int test_fbt_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  return test_bt_open("fast=1", zFilename, bClear, ppDb);
+}
+
+int test_fbts_open(
+  const char *zSpec, 
+  const char *zFilename, 
+  int bClear, 
+  TestDb **ppDb
+){
+  return test_bt_open("fast=1 blksz=32K pagesz=512", zFilename, bClear, ppDb);
+}
+
+
+void tdb_bt_prepare_sync_crash(TestDb *pTestDb, int iSync){
+  BtDb *p = (BtDb*)pTestDb;
+  assert( pTestDb->pMethods->xClose==bt_close );
+  assert( p->bCrash==0 );
+  p->nCrashSync = iSync;
+}
+
+bt_db *tdb_bt(TestDb *pDb){
+  if( pDb->pMethods->xClose==bt_close ){
+    return ((BtDb *)pDb)->pBt;
+  }
+  return 0;
+}
+
+/*************************************************************************
+** Beginning of code for background checkpointer.
+*/
+
+struct bt_ckpter {
+  sqlite4_buffer file;            /* File name */
+  sqlite4_buffer spec;            /* Options */
+  int nLogsize;                   /* Minimum log size to checkpoint */
+  int nRef;                       /* Number of clients */
+
+  int bDoWork;                    /* Set by client threads */
+  pthread_t ckpter_thread;        /* Checkpointer thread */
+  pthread_cond_t ckpter_cond;     /* Condition var the ckpter waits on */
+  pthread_mutex_t ckpter_mutex;   /* Mutex used with ckpter_cond */
+
+  bt_ckpter *pNext;               /* Next object in list at gBgc.pCkpter */
+};
+
+static struct GlobalBackgroundCheckpointer {
+  bt_ckpter *pCkpter;             /* Linked list of checkpointers */
+} gBgc;
+
+static void *bgc_main(void *pArg){
+  BtDb *pDb = 0;
+  int rc;
+  int mt;
+  bt_ckpter *pCkpter = (bt_ckpter*)pArg;
+
+  rc = test_bt_open("", (char*)pCkpter->file.p, 0, (TestDb**)&pDb);
+  assert( rc==SQLITE4_OK );
+  rc = testBtConfigure(pDb, (char*)pCkpter->spec.p, &mt);
+
+  while( pCkpter->nRef>0 ){
+    bt_db *db = pDb->pBt;
+    int nLog = 0;
+
+    sqlite4BtBegin(db, 1);
+    sqlite4BtCommit(db, 0);
+    sqlite4BtControl(db, BT_CONTROL_LOGSIZE, (void*)&nLog);
+
+    if( nLog>=pCkpter->nLogsize ){
+      int rc;
+      bt_checkpoint ckpt;
+      memset(&ckpt, 0, sizeof(bt_checkpoint));
+      ckpt.nFrameBuffer = nLog/2;
+      rc = sqlite4BtControl(db, BT_CONTROL_CHECKPOINT, (void*)&ckpt);
+      assert( rc==SQLITE4_OK );
+      sqlite4BtControl(db, BT_CONTROL_LOGSIZE, (void*)&nLog);
+    }
+
+    /* The thread will wake up when it is signaled either because another
+    ** thread has created some work for this one or because the connection
+    ** is being closed.  */
+    pthread_mutex_lock(&pCkpter->ckpter_mutex);
+    if( pCkpter->bDoWork==0 ){
+      pthread_cond_wait(&pCkpter->ckpter_cond, &pCkpter->ckpter_mutex);
+    }
+    pCkpter->bDoWork = 0;
+    pthread_mutex_unlock(&pCkpter->ckpter_mutex);
+  }
+
+  if( pDb ) bt_close((TestDb*)pDb);
+  return 0;
+}
+
+static void bgc_logsize_cb(void *pCtx, int nLogsize){
+  bt_ckpter *p = (bt_ckpter*)pCtx;
+  if( nLogsize>=p->nLogsize ){
+    pthread_mutex_lock(&p->ckpter_mutex);
+    p->bDoWork = 1;
+    pthread_cond_signal(&p->ckpter_cond);
+    pthread_mutex_unlock(&p->ckpter_mutex);
+  }
+}
+
+static int bgc_attach(BtDb *pDb, const char *zSpec){
+  int rc;
+  int n;
+  bt_info info;
+  bt_ckpter *pCkpter;
+
+  /* Figure out the full path to the database opened by handle pDb. */
+  info.eType = BT_INFO_FILENAME;
+  info.pgno = 0;
+  sqlite4_buffer_init(&info.output, 0);
+  rc = sqlite4BtControl(pDb->pBt, BT_CONTROL_INFO, (void*)&info);
+  if( rc!=SQLITE4_OK ) return rc;
+
+  sqlite4_mutex_enter(sqlite4_mutex_alloc(pDb->pEnv, SQLITE4_MUTEX_STATIC_KV));
+
+  /* Search for an existing bt_ckpter object. */
+  n = info.output.n;
+  for(pCkpter=gBgc.pCkpter; pCkpter; pCkpter=pCkpter->pNext){
+    if( n==pCkpter->file.n && 0==memcmp(info.output.p, pCkpter->file.p, n) ){
+      break;
+    }
+  }
+
+  /* Failed to find a suitable checkpointer. Create a new one. */
+  if( pCkpter==0 ){
+    bt_logsizecb cb;
+
+    pCkpter = testMalloc(sizeof(bt_ckpter));
+    memcpy(&pCkpter->file, &info.output, sizeof(sqlite4_buffer));
+    info.output.p = 0;
+    pCkpter->pNext = gBgc.pCkpter;
+    pCkpter->nLogsize = 1000;
+    gBgc.pCkpter = pCkpter;
+    pCkpter->nRef = 1;
+
+    sqlite4_buffer_init(&pCkpter->spec, 0);
+    rc = sqlite4_buffer_set(&pCkpter->spec, zSpec, strlen(zSpec)+1);
+    assert( rc==SQLITE4_OK );
+
+    /* Kick off the checkpointer thread. */
+    if( rc==0 ) rc = pthread_cond_init(&pCkpter->ckpter_cond, 0);
+    if( rc==0 ) rc = pthread_mutex_init(&pCkpter->ckpter_mutex, 0);
+    if( rc==0 ){
+      rc = pthread_create(&pCkpter->ckpter_thread, 0, bgc_main, (void*)pCkpter);
+    }
+    assert( rc==0 ); /* todo: Fix this */
+
+    /* Set up the logsize callback for the client thread */
+    cb.pCtx = (void*)pCkpter;
+    cb.xLogsize = bgc_logsize_cb;
+    sqlite4BtControl(pDb->pBt, BT_CONTROL_LOGSIZECB, (void*)&cb);
+  }else{
+    pCkpter->nRef++;
+  }
+
+  /* Assuming a checkpointer was encountered or effected, attach the 
+  ** connection to it.  */
+  if( pCkpter ){
+    pDb->pCkpter = pCkpter;
+  }
+
+  sqlite4_mutex_leave(sqlite4_mutex_alloc(pDb->pEnv, SQLITE4_MUTEX_STATIC_KV));
+  sqlite4_buffer_clear(&info.output);
+  return rc;
+}
+
+static int bgc_detach(BtDb *pDb){
+  int rc = SQLITE4_OK;
+  bt_ckpter *pCkpter = pDb->pCkpter;
+  if( pCkpter ){
+    int bShutdown = 0;            /* True if this is the last reference */
+
+    sqlite4_mutex_enter(sqlite4_mutex_alloc(pDb->pEnv,SQLITE4_MUTEX_STATIC_KV));
+    pCkpter->nRef--;
+    if( pCkpter->nRef==0 ){
+      bt_ckpter **pp;
+
+      *pp = pCkpter->pNext;
+      for(pp=&gBgc.pCkpter; *pp!=pCkpter; pp=&((*pp)->pNext));
+      bShutdown = 1;
+    }
+    sqlite4_mutex_leave(sqlite4_mutex_alloc(pDb->pEnv,SQLITE4_MUTEX_STATIC_KV));
+
+    if( bShutdown ){
+      void *pDummy;
+
+      /* Signal the checkpointer thread. */
+      pthread_mutex_lock(&pCkpter->ckpter_mutex);
+      pCkpter->bDoWork = 1;
+      pthread_cond_signal(&pCkpter->ckpter_cond);
+      pthread_mutex_unlock(&pCkpter->ckpter_mutex);
+
+      /* Join the checkpointer thread. */
+      pthread_join(pCkpter->ckpter_thread, &pDummy);
+      pthread_cond_destroy(&pCkpter->ckpter_cond);
+      pthread_mutex_destroy(&pCkpter->ckpter_mutex);
+
+      sqlite4_buffer_clear(&pCkpter->file);
+      sqlite4_buffer_clear(&pCkpter->spec);
+      testFree(pCkpter);
+    }
+
+    pDb->pCkpter = 0;
+  }
+  return rc;
+}
+
+/*
+** End of background checkpointer.
+*************************************************************************/
diff --git a/ext/lsm1/lsm-test/lsmtest_util.c b/ext/lsm1/lsm-test/lsmtest_util.c
new file mode 100644
index 0000000..adab8a5
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_util.c
@@ -0,0 +1,223 @@
+
+#include "lsmtest.h"
+#include <stdarg.h>
+#include <stdio.h>
+#include <string.h>
+#ifndef _WIN32
+# include <sys/time.h>
+#endif
+
+/*
+** Global variables used within this module.
+*/
+static struct TestutilGlobal {
+  char **argv;
+  int argc;
+} g = {0, 0};
+
+static struct TestutilRnd {
+  unsigned int aRand1[2048];          /* Bits 0..10 */
+  unsigned int aRand2[2048];          /* Bits 11..21 */
+  unsigned int aRand3[1024];          /* Bits 22..31 */
+} r;
+
+/*************************************************************************
+** The following block is a copy of the implementation of SQLite function
+** sqlite3_randomness. This version has two important differences:
+**
+**   1. It always uses the same seed. So the sequence of random data output
+**      is the same for every run of the program.
+**
+**   2. It is not threadsafe.
+*/
+static struct sqlite3PrngType {
+  unsigned char i, j;             /* State variables */
+  unsigned char s[256];           /* State variables */
+} sqlite3Prng = {
+    0xAF, 0x28,
+  {
+    0x71, 0xF5, 0xB4, 0x6E, 0x80, 0xAB, 0x1D, 0xB8, 
+    0xFB, 0xB7, 0x49, 0xBF, 0xFF, 0x72, 0x2D, 0x14, 
+    0x79, 0x09, 0xE3, 0x78, 0x76, 0xB0, 0x2C, 0x0A, 
+    0x8E, 0x23, 0xEE, 0xDF, 0xE0, 0x9A, 0x2F, 0x67, 
+    0xE1, 0xBE, 0x0E, 0xA7, 0x08, 0x97, 0xEB, 0x77, 
+    0x78, 0xBA, 0x9D, 0xCA, 0x49, 0x4C, 0x60, 0x9A, 
+    0xF6, 0xBD, 0xDA, 0x7F, 0xBC, 0x48, 0x58, 0x52, 
+    0xE5, 0xCD, 0x83, 0x72, 0x23, 0x52, 0xFF, 0x6D, 
+    0xEF, 0x0F, 0x82, 0x29, 0xA0, 0x83, 0x3F, 0x7D, 
+    0xA4, 0x88, 0x31, 0xE7, 0x88, 0x92, 0x3B, 0x9B, 
+    0x3B, 0x2C, 0xC2, 0x4C, 0x71, 0xA2, 0xB0, 0xEA, 
+    0x36, 0xD0, 0x00, 0xF1, 0xD3, 0x39, 0x17, 0x5D, 
+    0x2A, 0x7A, 0xE4, 0xAD, 0xE1, 0x64, 0xCE, 0x0F, 
+    0x9C, 0xD9, 0xF5, 0xED, 0xB0, 0x22, 0x5E, 0x62, 
+    0x97, 0x02, 0xA3, 0x8C, 0x67, 0x80, 0xFC, 0x88, 
+    0x14, 0x0B, 0x15, 0x10, 0x0F, 0xC7, 0x40, 0xD4, 
+    0xF1, 0xF9, 0x0E, 0x1A, 0xCE, 0xB9, 0x1E, 0xA1, 
+    0x72, 0x8E, 0xD7, 0x78, 0x39, 0xCD, 0xF4, 0x5D, 
+    0x2A, 0x59, 0x26, 0x34, 0xF2, 0x73, 0x0B, 0xA0, 
+    0x02, 0x51, 0x2C, 0x03, 0xA3, 0xA7, 0x43, 0x13, 
+    0xE8, 0x98, 0x2B, 0xD2, 0x53, 0xF8, 0xEE, 0x91, 
+    0x7D, 0xE7, 0xE3, 0xDA, 0xD5, 0xBB, 0xC0, 0x92, 
+    0x9D, 0x98, 0x01, 0x2C, 0xF9, 0xB9, 0xA0, 0xEB, 
+    0xCF, 0x32, 0xFA, 0x01, 0x49, 0xA5, 0x1D, 0x9A, 
+    0x76, 0x86, 0x3F, 0x40, 0xD4, 0x89, 0x8F, 0x9C, 
+    0xE2, 0xE3, 0x11, 0x31, 0x37, 0xB2, 0x49, 0x28, 
+    0x35, 0xC0, 0x99, 0xB6, 0xD0, 0xBC, 0x66, 0x35, 
+    0xF7, 0x83, 0x5B, 0xD7, 0x37, 0x1A, 0x2B, 0x18, 
+    0xA6, 0xFF, 0x8D, 0x7C, 0x81, 0xA8, 0xFC, 0x9E, 
+    0xC4, 0xEC, 0x80, 0xD0, 0x98, 0xA7, 0x76, 0xCC, 
+    0x9C, 0x2F, 0x7B, 0xFF, 0x8E, 0x0E, 0xBB, 0x90, 
+    0xAE, 0x13, 0x06, 0xF5, 0x1C, 0x4E, 0x52, 0xF7
+  }
+};
+
+/* Generate and return single random byte */
+static unsigned char randomByte(void){
+  unsigned char t;
+  sqlite3Prng.i++;
+  t = sqlite3Prng.s[sqlite3Prng.i];
+  sqlite3Prng.j += t;
+  sqlite3Prng.s[sqlite3Prng.i] = sqlite3Prng.s[sqlite3Prng.j];
+  sqlite3Prng.s[sqlite3Prng.j] = t;
+  t += sqlite3Prng.s[sqlite3Prng.i];
+  return sqlite3Prng.s[t];
+}
+
+/*
+** Return N random bytes.
+*/
+static void randomBlob(int nBuf, unsigned char *zBuf){
+  int i;
+  for(i=0; i<nBuf; i++){
+    zBuf[i] = randomByte();
+  }
+}
+/*
+** End of code copied from SQLite.
+*************************************************************************/
+
+
+int testPrngInit(void){
+  randomBlob(sizeof(r.aRand1), (unsigned char *)r.aRand1);
+  randomBlob(sizeof(r.aRand2), (unsigned char *)r.aRand2);
+  randomBlob(sizeof(r.aRand3), (unsigned char *)r.aRand3);
+  return 0;
+}
+
+unsigned int testPrngValue(unsigned int iVal){
+  return
+    r.aRand1[iVal & 0x000007FF] ^
+    r.aRand2[(iVal>>11) & 0x000007FF] ^
+    r.aRand3[(iVal>>22) & 0x000003FF]
+  ;
+}
+
+void testPrngArray(unsigned int iVal, unsigned int *aOut, int nOut){
+  int i;
+  for(i=0; i<nOut; i++){
+    aOut[i] = testPrngValue(iVal+i);
+  }
+}
+
+void testPrngString(unsigned int iVal, char *aOut, int nOut){
+  int i;
+  for(i=0; i<(nOut-1); i++){
+    aOut[i] = 'a' + (testPrngValue(iVal+i) % 26);
+  }
+  aOut[i] = '\0';
+}
+
+void testErrorInit(int argc, char **argv){
+  g.argc = argc;
+  g.argv = argv;
+}
+
+void testPrintError(const char *zFormat, ...){
+  va_list ap;
+  va_start(ap, zFormat);
+  vfprintf(stderr, zFormat, ap);
+  va_end(ap);
+}
+
+void testPrintFUsage(const char *zFormat, ...){
+  va_list ap;
+  va_start(ap, zFormat);
+  fprintf(stderr, "Usage: %s %s ", g.argv[0], g.argv[1]);
+  vfprintf(stderr, zFormat, ap);
+  fprintf(stderr, "\n");
+  va_end(ap);
+}
+
+void testPrintUsage(const char *zArgs){
+  testPrintError("Usage: %s %s %s\n", g.argv[0], g.argv[1], zArgs);
+}
+
+
+static void argError(void *aData, const char *zType, int sz, const char *zArg){
+  struct Entry { const char *zName; };
+  struct Entry *pEntry;
+  const char *zPrev = 0;
+
+  testPrintError("unrecognized %s \"%s\": must be ", zType, zArg);
+  for(pEntry=(struct Entry *)aData; 
+      pEntry->zName; 
+      pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
+  ){
+    if( zPrev ){ testPrintError("%s, ", zPrev); }
+    zPrev = pEntry->zName;
+  }
+  testPrintError("or %s\n", zPrev);
+}
+
+int testArgSelectX(
+  void *aData, 
+  const char *zType, 
+  int sz, 
+  const char *zArg, 
+  int *piOut
+){
+  struct Entry { const char *zName; };
+  struct Entry *pEntry;
+  int nArg = strlen(zArg);
+
+  int i = 0;
+  int iOut = -1;
+  int nOut = 0;
+
+  for(pEntry=(struct Entry *)aData; 
+      pEntry->zName; 
+      pEntry=(struct Entry *)&((unsigned char *)pEntry)[sz]
+  ){
+    int nName = strlen(pEntry->zName);
+    if( nArg<=nName && memcmp(pEntry->zName, zArg, nArg)==0 ){
+      iOut = i;
+      if( nName==nArg ){
+        nOut = 1;
+        break;
+      }
+      nOut++;
+    }
+    i++;
+  }
+
+  if( nOut!=1 ){
+    argError(aData, zType, sz, zArg);
+  }else{
+    *piOut = iOut;
+  }
+  return (nOut!=1);
+}
+
+struct timeval zero_time;
+
+void testTimeInit(void){
+  gettimeofday(&zero_time, 0);
+}
+
+int testTimeGet(void){
+  struct timeval now;
+  gettimeofday(&now, 0);
+  return
+    (((int)now.tv_sec - (int)zero_time.tv_sec)*1000) +
+    (((int)now.tv_usec - (int)zero_time.tv_usec)/1000);
+}
diff --git a/ext/lsm1/lsm-test/lsmtest_win32.c b/ext/lsm1/lsm-test/lsmtest_win32.c
new file mode 100644
index 0000000..9472723
--- /dev/null
+++ b/ext/lsm1/lsm-test/lsmtest_win32.c
@@ -0,0 +1,30 @@
+
+#include "lsmtest.h"
+
+#ifdef _WIN32
+
+#define TICKS_PER_SECOND      (10000000)
+#define TICKS_PER_MICROSECOND (10)
+#define TICKS_UNIX_EPOCH      (116444736000000000LL)
+
+int win32GetTimeOfDay(
+  struct timeval *tp,
+  void *tzp
+){
+  FILETIME fileTime;
+  ULONGLONG ticks;
+  ULONGLONG unixTicks;
+
+  unused_parameter(tzp);
+  memset(&fileTime, 0, sizeof(FILETIME));
+  GetSystemTimeAsFileTime(&fileTime);
+  ticks = (ULONGLONG)fileTime.dwHighDateTime << 32;
+  ticks |= (ULONGLONG)fileTime.dwLowDateTime;
+  unixTicks = ticks - TICKS_UNIX_EPOCH;
+  tp->tv_sec = (long)(unixTicks / TICKS_PER_SECOND);
+  unixTicks -= ((ULONGLONG)tp->tv_sec * TICKS_PER_SECOND);
+  tp->tv_usec = (long)(unixTicks / TICKS_PER_MICROSECOND);
+
+  return 0;
+}
+#endif
diff --git a/ext/lsm1/lsm.h b/ext/lsm1/lsm.h
new file mode 100644
index 0000000..48701c4
--- /dev/null
+++ b/ext/lsm1/lsm.h
@@ -0,0 +1,684 @@
+/*
+** 2011-08-10
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** This file defines the LSM API.
+*/
+#ifndef _LSM_H
+#define _LSM_H
+#include <stddef.h>
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+** Opaque handle types.
+*/
+typedef struct lsm_compress lsm_compress;   /* Compression library functions */
+typedef struct lsm_compress_factory lsm_compress_factory;
+typedef struct lsm_cursor lsm_cursor;       /* Database cursor handle */
+typedef struct lsm_db lsm_db;               /* Database connection handle */
+typedef struct lsm_env lsm_env;             /* Runtime environment */
+typedef struct lsm_file lsm_file;           /* OS file handle */
+typedef struct lsm_mutex lsm_mutex;         /* Mutex handle */
+
+/* 64-bit integer type used for file offsets. */
+typedef long long int lsm_i64;              /* 64-bit signed integer type */
+
+/* Candidate values for the 3rd argument to lsm_env.xLock() */
+#define LSM_LOCK_UNLOCK 0
+#define LSM_LOCK_SHARED 1
+#define LSM_LOCK_EXCL   2
+
+/* Flags for lsm_env.xOpen() */
+#define LSM_OPEN_READONLY 0x0001
+
+/*
+** CAPI: Database Runtime Environment
+**
+** Run-time environment used by LSM
+*/
+struct lsm_env {
+  int nByte;                 /* Size of this structure in bytes */
+  int iVersion;              /* Version number of this structure (1) */
+  /****** file i/o ***********************************************/
+  void *pVfsCtx;
+  int (*xFullpath)(lsm_env*, const char *, char *, int *);
+  int (*xOpen)(lsm_env*, const char *, int flags, lsm_file **);
+  int (*xRead)(lsm_file *, lsm_i64, void *, int);
+  int (*xWrite)(lsm_file *, lsm_i64, void *, int);
+  int (*xTruncate)(lsm_file *, lsm_i64);
+  int (*xSync)(lsm_file *);
+  int (*xSectorSize)(lsm_file *);
+  int (*xRemap)(lsm_file *, lsm_i64, void **, lsm_i64*);
+  int (*xFileid)(lsm_file *, void *pBuf, int *pnBuf);
+  int (*xClose)(lsm_file *);
+  int (*xUnlink)(lsm_env*, const char *);
+  int (*xLock)(lsm_file*, int, int);
+  int (*xTestLock)(lsm_file*, int, int, int);
+  int (*xShmMap)(lsm_file*, int, int, void **);
+  void (*xShmBarrier)(void);
+  int (*xShmUnmap)(lsm_file*, int);
+  /****** memory allocation ****************************************/
+  void *pMemCtx;
+  void *(*xMalloc)(lsm_env*, size_t);            /* malloc(3) function */
+  void *(*xRealloc)(lsm_env*, void *, size_t);   /* realloc(3) function */
+  void (*xFree)(lsm_env*, void *);               /* free(3) function */
+  size_t (*xSize)(lsm_env*, void *);             /* xSize function */
+  /****** mutexes ****************************************************/
+  void *pMutexCtx;
+  int (*xMutexStatic)(lsm_env*,int,lsm_mutex**); /* Obtain a static mutex */
+  int (*xMutexNew)(lsm_env*, lsm_mutex**);       /* Get a new dynamic mutex */
+  void (*xMutexDel)(lsm_mutex *);           /* Delete an allocated mutex */
+  void (*xMutexEnter)(lsm_mutex *);         /* Grab a mutex */
+  int (*xMutexTry)(lsm_mutex *);            /* Attempt to obtain a mutex */
+  void (*xMutexLeave)(lsm_mutex *);         /* Leave a mutex */
+  int (*xMutexHeld)(lsm_mutex *);           /* Return true if mutex is held */
+  int (*xMutexNotHeld)(lsm_mutex *);        /* Return true if mutex not held */
+  /****** other ****************************************************/
+  int (*xSleep)(lsm_env*, int microseconds);
+
+  /* New fields may be added in future releases, in which case the
+  ** iVersion value will increase. */
+};
+
+/* 
+** Values that may be passed as the second argument to xMutexStatic. 
+*/
+#define LSM_MUTEX_GLOBAL 1
+#define LSM_MUTEX_HEAP   2
+
+/*
+** CAPI: LSM Error Codes
+*/
+#define LSM_OK         0
+#define LSM_ERROR      1
+#define LSM_BUSY       5
+#define LSM_NOMEM      7
+#define LSM_READONLY   8
+#define LSM_IOERR     10
+#define LSM_CORRUPT   11
+#define LSM_FULL      13
+#define LSM_CANTOPEN  14
+#define LSM_PROTOCOL  15
+#define LSM_MISUSE    21
+
+#define LSM_MISMATCH  50
+
+
+#define LSM_IOERR_NOENT (LSM_IOERR | (1<<8))
+
+/* 
+** CAPI: Creating and Destroying Database Connection Handles
+**
+** Open and close a database connection handle.
+*/
+int lsm_new(lsm_env*, lsm_db **ppDb);
+int lsm_close(lsm_db *pDb);
+
+/* 
+** CAPI: Connecting to a Database
+*/
+int lsm_open(lsm_db *pDb, const char *zFilename);
+
+/*
+** CAPI: Obtaining pointers to database environments
+**
+** Return a pointer to the environment used by the database connection 
+** passed as the first argument. Assuming the argument is valid, this 
+** function always returns a valid environment pointer - it cannot fail.
+*/
+lsm_env *lsm_get_env(lsm_db *pDb);
+
+/*
+** The lsm_default_env() function returns a pointer to the default LSM
+** environment for the current platform.
+*/
+lsm_env *lsm_default_env(void);
+
+
+/*
+** CAPI: Configuring a database connection.
+**
+** The lsm_config() function is used to configure a database connection.
+*/
+int lsm_config(lsm_db *, int, ...);
+
+/*
+** The following values may be passed as the second argument to lsm_config().
+**
+** LSM_CONFIG_AUTOFLUSH:
+**   A read/write integer parameter. 
+**
+**   This value determines the amount of data allowed to accumulate in a
+**   live in-memory tree before it is marked as old. After committing a
+**   transaction, a connection checks if the size of the live in-memory tree,
+**   including data structure overhead, is greater than the value of this
+**   option in KB. If it is, and there is not already an old in-memory tree,
+**   the live in-memory tree is marked as old.
+**
+**   The maximum allowable value is 1048576 (1GB). There is no minimum 
+**   value. If this parameter is set to zero, then an attempt is made to
+**   mark the live in-memory tree as old after each transaction is committed.
+**
+**   The default value is 1024 (1MB).
+**
+** LSM_CONFIG_PAGE_SIZE:
+**   A read/write integer parameter. This parameter may only be set before
+**   lsm_open() has been called.
+**
+** LSM_CONFIG_BLOCK_SIZE:
+**   A read/write integer parameter. 
+**
+**   This parameter may only be set before lsm_open() has been called. It
+**   must be set to a power of two between 64 and 65536, inclusive (block 
+**   sizes between 64KB and 64MB).
+**
+**   If the connection creates a new database, the block size of the new
+**   database is set to the value of this option in KB. After lsm_open()
+**   has been called, querying this parameter returns the actual block
+**   size of the opened database.
+**
+**   The default value is 1024 (1MB blocks).
+**
+** LSM_CONFIG_SAFETY:
+**   A read/write integer parameter. Valid values are 0, 1 (the default) 
+**   and 2. This parameter determines how robust the database is in the
+**   face of a system crash (e.g. a power failure or operating system 
+**   crash). As follows:
+**
+**     0 (off):    No robustness. A system crash may corrupt the database.
+**
+**     1 (normal): Some robustness. A system crash may not corrupt the
+**                 database file, but recently committed transactions may
+**                 be lost following recovery.
+**
+**     2 (full):   Full robustness. A system crash may not corrupt the
+**                 database file. Following recovery the database file
+**                 contains all successfully committed transactions.
+**
+** LSM_CONFIG_AUTOWORK:
+**   A read/write integer parameter.
+**
+** LSM_CONFIG_AUTOCHECKPOINT:
+**   A read/write integer parameter.
+**
+**   If this option is set to non-zero value N, then a checkpoint is
+**   automatically attempted after each N KB of data have been written to 
+**   the database file.
+**
+**   The amount of uncheckpointed data already written to the database file
+**   is a global parameter. After performing database work (writing to the
+**   database file), the process checks if the total amount of uncheckpointed 
+**   data exceeds the value of this paramter. If so, a checkpoint is performed.
+**   This means that this option may cause the connection to perform a 
+**   checkpoint even if the current connection has itself written very little
+**   data into the database file.
+**
+**   The default value is 2048 (checkpoint every 2MB).
+**
+** LSM_CONFIG_MMAP:
+**   A read/write integer parameter. If this value is set to 0, then the 
+**   database file is accessed using ordinary read/write IO functions. Or,
+**   if it is set to 1, then the database file is memory mapped and accessed
+**   that way. If this parameter is set to any value N greater than 1, then
+**   up to the first N KB of the file are memory mapped, and any remainder
+**   accessed using read/write IO.
+**
+**   The default value is 1 on 64-bit platforms and 32768 on 32-bit platforms.
+**   
+**
+** LSM_CONFIG_USE_LOG:
+**   A read/write boolean parameter. True (the default) to use the log
+**   file normally. False otherwise.
+**
+** LSM_CONFIG_AUTOMERGE:
+**   A read/write integer parameter. The minimum number of segments to
+**   merge together at a time. Default value 4.
+**
+** LSM_CONFIG_MAX_FREELIST:
+**   A read/write integer parameter. The maximum number of free-list 
+**   entries that are stored in a database checkpoint (the others are
+**   stored elsewhere in the database).
+**
+**   There is no reason for an application to configure or query this
+**   parameter. It is only present because configuring a small value
+**   makes certain parts of the lsm code easier to test.
+**
+** LSM_CONFIG_MULTIPLE_PROCESSES:
+**   A read/write boolean parameter. This parameter may only be set before
+**   lsm_open() has been called. If true, the library uses shared-memory
+**   and posix advisory locks to co-ordinate access by clients from within
+**   multiple processes. Otherwise, if false, all database clients must be 
+**   located in the same process. The default value is true.
+**
+** LSM_CONFIG_SET_COMPRESSION:
+**   Set the compression methods used to compress and decompress database
+**   content. The argument to this option should be a pointer to a structure
+**   of type lsm_compress. The lsm_config() method takes a copy of the 
+**   structures contents.
+**
+**   This option may only be used before lsm_open() is called. Invoking it
+**   after lsm_open() has been called results in an LSM_MISUSE error.
+**
+** LSM_CONFIG_GET_COMPRESSION:
+**   Query the compression methods used to compress and decompress database
+**   content.
+**
+** LSM_CONFIG_SET_COMPRESSION_FACTORY:
+**   Configure a factory method to be invoked in case of an LSM_MISMATCH
+**   error.
+**
+** LSM_CONFIG_READONLY:
+**   A read/write boolean parameter. This parameter may only be set before
+**   lsm_open() is called.
+*/
+#define LSM_CONFIG_AUTOFLUSH                1
+#define LSM_CONFIG_PAGE_SIZE                2
+#define LSM_CONFIG_SAFETY                   3
+#define LSM_CONFIG_BLOCK_SIZE               4
+#define LSM_CONFIG_AUTOWORK                 5
+#define LSM_CONFIG_MMAP                     7
+#define LSM_CONFIG_USE_LOG                  8
+#define LSM_CONFIG_AUTOMERGE                9
+#define LSM_CONFIG_MAX_FREELIST            10
+#define LSM_CONFIG_MULTIPLE_PROCESSES      11
+#define LSM_CONFIG_AUTOCHECKPOINT          12
+#define LSM_CONFIG_SET_COMPRESSION         13
+#define LSM_CONFIG_GET_COMPRESSION         14
+#define LSM_CONFIG_SET_COMPRESSION_FACTORY 15
+#define LSM_CONFIG_READONLY                16
+
+#define LSM_SAFETY_OFF    0
+#define LSM_SAFETY_NORMAL 1
+#define LSM_SAFETY_FULL   2
+
+/*
+** CAPI: Compression and/or Encryption Hooks
+*/
+struct lsm_compress {
+  void *pCtx;
+  unsigned int iId;
+  int (*xBound)(void *, int nSrc);
+  int (*xCompress)(void *, char *, int *, const char *, int);
+  int (*xUncompress)(void *, char *, int *, const char *, int);
+  void (*xFree)(void *pCtx);
+};
+
+struct lsm_compress_factory {
+  void *pCtx;
+  int (*xFactory)(void *, lsm_db *, unsigned int);
+  void (*xFree)(void *pCtx);
+};
+
+#define LSM_COMPRESSION_EMPTY 0
+#define LSM_COMPRESSION_NONE  1
+
+/*
+** CAPI: Allocating and Freeing Memory
+**
+** Invoke the memory allocation functions that belong to environment
+** pEnv. Or the system defaults if no memory allocation functions have 
+** been registered.
+*/
+void *lsm_malloc(lsm_env*, size_t);
+void *lsm_realloc(lsm_env*, void *, size_t);
+void lsm_free(lsm_env*, void *);
+
+/*
+** CAPI: Querying a Connection For Operational Data
+**
+** Query a database connection for operational statistics or data.
+*/
+int lsm_info(lsm_db *, int, ...);
+
+int lsm_get_user_version(lsm_db *, unsigned int *);
+int lsm_set_user_version(lsm_db *, unsigned int);
+
+/*
+** The following values may be passed as the second argument to lsm_info().
+**
+** LSM_INFO_NWRITE:
+**   The third parameter should be of type (int *). The location pointed
+**   to by the third parameter is set to the number of 4KB pages written to
+**   the database file during the lifetime of this connection. 
+**
+** LSM_INFO_NREAD:
+**   The third parameter should be of type (int *). The location pointed
+**   to by the third parameter is set to the number of 4KB pages read from
+**   the database file during the lifetime of this connection.
+**
+** LSM_INFO_DB_STRUCTURE:
+**   The third argument should be of type (char **). The location pointed
+**   to is populated with a pointer to a nul-terminated string containing
+**   the string representation of a Tcl data-structure reflecting the 
+**   current structure of the database file. Specifically, the current state
+**   of the worker snapshot. The returned string should be eventually freed 
+**   by the caller using lsm_free().
+**
+**   The returned list contains one element for each level in the database,
+**   in order from most to least recent. Each element contains a 
+**   single element for each segment comprising the corresponding level,
+**   starting with the lhs segment, then each of the rhs segments (if any)
+**   in order from most to least recent.
+**
+**   Each segment element is itself a list of 4 integer values, as follows:
+**
+**   <ol><li> First page of segment
+**       <li> Last page of segment
+**       <li> Root page of segment (if applicable)
+**       <li> Total number of pages in segment
+**   </ol>
+**
+** LSM_INFO_ARRAY_STRUCTURE:
+**   There should be two arguments passed following this option (i.e. a 
+**   total of four arguments passed to lsm_info()). The first argument 
+**   should be the page number of the first page in a database array 
+**   (perhaps obtained from an earlier INFO_DB_STRUCTURE call). The second 
+**   trailing argument should be of type (char **). The location pointed 
+**   to is populated with a pointer to a nul-terminated string that must 
+**   be eventually freed using lsm_free() by the caller.
+**
+**   The output string contains the text representation of a Tcl list of
+**   integers. Each pair of integers represent a range of pages used by
+**   the identified array. For example, if the array occupies database
+**   pages 993 to 1024, then pages 2048 to 2777, then the returned string
+**   will be "993 1024 2048 2777".
+**
+**   If the specified integer argument does not correspond to the first
+**   page of any database array, LSM_ERROR is returned and the output
+**   pointer is set to a NULL value.
+**
+** LSM_INFO_LOG_STRUCTURE:
+**   The third argument should be of type (char **). The location pointed
+**   to is populated with a pointer to a nul-terminated string containing
+**   the string representation of a Tcl data-structure. The returned 
+**   string should be eventually freed by the caller using lsm_free().
+**
+**   The Tcl structure returned is a list of six integers that describe
+**   the current structure of the log file.
+**
+** LSM_INFO_ARRAY_PAGES:
+**
+** LSM_INFO_PAGE_ASCII_DUMP:
+**   As with LSM_INFO_ARRAY_STRUCTURE, there should be two arguments passed
+**   with calls that specify this option - an integer page number and a
+**   (char **) used to return a nul-terminated string that must be later
+**   freed using lsm_free(). In this case the output string is populated
+**   with a human-readable description of the page content.
+**
+**   If the page cannot be decoded, it is not an error. In this case the
+**   human-readable output message will report the systems failure to 
+**   interpret the page data.
+**
+** LSM_INFO_PAGE_HEX_DUMP:
+**   This argument is similar to PAGE_ASCII_DUMP, except that keys and
+**   values are represented using hexadecimal notation instead of ascii.
+**
+** LSM_INFO_FREELIST:
+**   The third argument should be of type (char **). The location pointed
+**   to is populated with a pointer to a nul-terminated string containing
+**   the string representation of a Tcl data-structure. The returned 
+**   string should be eventually freed by the caller using lsm_free().
+**
+**   The Tcl structure returned is a list containing one element for each
+**   free block in the database. The element itself consists of two 
+**   integers - the block number and the id of the snapshot that freed it.
+**
+** LSM_INFO_CHECKPOINT_SIZE:
+**   The third argument should be of type (int *). The location pointed to
+**   by this argument is populated with the number of KB written to the
+**   database file since the most recent checkpoint.
+**
+** LSM_INFO_TREE_SIZE:
+**   If this value is passed as the second argument to an lsm_info() call, it
+**   should be followed by two arguments of type (int *) (for a total of four
+**   arguments).
+**
+**   At any time, there are either one or two tree structures held in shared
+**   memory that new database clients will access (there may also be additional
+**   tree structures being used by older clients - this API does not provide
+**   information on them). One tree structure - the current tree - is used to
+**   accumulate new data written to the database. The other tree structure -
+**   the old tree - is a read-only tree holding older data and may be flushed 
+**   to disk at any time.
+** 
+**   Assuming no error occurs, the location pointed to by the first of the two
+**   (int *) arguments is set to the size of the old in-memory tree in KB.
+**   The second is set to the size of the current, or live in-memory tree.
+**
+** LSM_INFO_COMPRESSION_ID:
+**   This value should be followed by a single argument of type 
+**   (unsigned int *). If successful, the location pointed to is populated 
+**   with the database compression id before returning.
+*/
+#define LSM_INFO_NWRITE           1
+#define LSM_INFO_NREAD            2
+#define LSM_INFO_DB_STRUCTURE     3
+#define LSM_INFO_LOG_STRUCTURE    4
+#define LSM_INFO_ARRAY_STRUCTURE  5
+#define LSM_INFO_PAGE_ASCII_DUMP  6
+#define LSM_INFO_PAGE_HEX_DUMP    7
+#define LSM_INFO_FREELIST         8
+#define LSM_INFO_ARRAY_PAGES      9
+#define LSM_INFO_CHECKPOINT_SIZE 10
+#define LSM_INFO_TREE_SIZE       11
+#define LSM_INFO_FREELIST_SIZE   12
+#define LSM_INFO_COMPRESSION_ID  13
+
+
+/* 
+** CAPI: Opening and Closing Write Transactions
+**
+** These functions are used to open and close transactions and nested 
+** sub-transactions.
+**
+** The lsm_begin() function is used to open transactions and sub-transactions. 
+** A successful call to lsm_begin() ensures that there are at least iLevel 
+** nested transactions open. To open a top-level transaction, pass iLevel=1. 
+** To open a sub-transaction within the top-level transaction, iLevel=2. 
+** Passing iLevel=0 is a no-op.
+**
+** lsm_commit() is used to commit transactions and sub-transactions. A
+** successful call to lsm_commit() ensures that there are at most iLevel 
+** nested transactions open. To commit a top-level transaction, pass iLevel=0. 
+** To commit all sub-transactions inside the main transaction, pass iLevel=1.
+**
+** Function lsm_rollback() is used to roll back transactions and
+** sub-transactions. A successful call to lsm_rollback() restores the database 
+** to the state it was in when the iLevel'th nested sub-transaction (if any) 
+** was first opened. And then closes transactions to ensure that there are 
+** at most iLevel nested transactions open. Passing iLevel=0 rolls back and 
+** closes the top-level transaction. iLevel=1 also rolls back the top-level 
+** transaction, but leaves it open. iLevel=2 rolls back the sub-transaction 
+** nested directly inside the top-level transaction (and leaves it open).
+*/
+int lsm_begin(lsm_db *pDb, int iLevel);
+int lsm_commit(lsm_db *pDb, int iLevel);
+int lsm_rollback(lsm_db *pDb, int iLevel);
+
+/* 
+** CAPI: Writing to a Database
+**
+** Write a new value into the database. If a value with a duplicate key 
+** already exists it is replaced.
+*/
+int lsm_insert(lsm_db*, const void *pKey, int nKey, const void *pVal, int nVal);
+
+/*
+** Delete a value from the database. No error is returned if the specified
+** key value does not exist in the database.
+*/
+int lsm_delete(lsm_db *, const void *pKey, int nKey);
+
+/*
+** Delete all database entries with keys that are greater than (pKey1/nKey1) 
+** and smaller than (pKey2/nKey2). Note that keys (pKey1/nKey1) and
+** (pKey2/nKey2) themselves, if they exist in the database, are not deleted.
+**
+** Return LSM_OK if successful, or an LSM error code otherwise.
+*/
+int lsm_delete_range(lsm_db *, 
+    const void *pKey1, int nKey1, const void *pKey2, int nKey2
+);
+
+/*
+** CAPI: Explicit Database Work and Checkpointing
+**
+** This function is called by a thread to work on the database structure.
+*/
+int lsm_work(lsm_db *pDb, int nMerge, int nKB, int *pnWrite);
+
+int lsm_flush(lsm_db *pDb);
+
+/*
+** Attempt to checkpoint the current database snapshot. Return an LSM
+** error code if an error occurs or LSM_OK otherwise.
+**
+** If the current snapshot has already been checkpointed, calling this 
+** function is a no-op. In this case if pnKB is not NULL, *pnKB is
+** set to 0. Or, if the current snapshot is successfully checkpointed
+** by this function and pbKB is not NULL, *pnKB is set to the number
+** of bytes written to the database file since the previous checkpoint
+** (the same measure as returned by the LSM_INFO_CHECKPOINT_SIZE query).
+*/
+int lsm_checkpoint(lsm_db *pDb, int *pnKB);
+
+/*
+** CAPI: Opening and Closing Database Cursors
+**
+** Open and close a database cursor.
+*/
+int lsm_csr_open(lsm_db *pDb, lsm_cursor **ppCsr);
+int lsm_csr_close(lsm_cursor *pCsr);
+
+/* 
+** CAPI: Positioning Database Cursors
+**
+** If the fourth parameter is LSM_SEEK_EQ, LSM_SEEK_GE or LSM_SEEK_LE,
+** this function searches the database for an entry with key (pKey/nKey). 
+** If an error occurs, an LSM error code is returned. Otherwise, LSM_OK.
+**
+** If no error occurs and the requested key is present in the database, the
+** cursor is left pointing to the entry with the specified key. Or, if the 
+** specified key is not present in the database the state of the cursor 
+** depends on the value passed as the final parameter, as follows:
+**
+** LSM_SEEK_EQ:
+**   The cursor is left at EOF (invalidated). A call to lsm_csr_valid()
+**   returns non-zero.
+**
+** LSM_SEEK_LE:
+**   The cursor is left pointing to the largest key in the database that
+**   is smaller than (pKey/nKey). If the database contains no keys smaller
+**   than (pKey/nKey), the cursor is left at EOF.
+**
+** LSM_SEEK_GE:
+**   The cursor is left pointing to the smallest key in the database that
+**   is larger than (pKey/nKey). If the database contains no keys larger
+**   than (pKey/nKey), the cursor is left at EOF.
+**
+** If the fourth parameter is LSM_SEEK_LEFAST, this function searches the
+** database in a similar manner to LSM_SEEK_LE, with two differences:
+**
+** <ol><li>Even if a key can be found (the cursor is not left at EOF), the
+** lsm_csr_value() function may not be used (attempts to do so return
+** LSM_MISUSE).
+**
+** <li>The key that the cursor is left pointing to may be one that has 
+** been recently deleted from the database. In this case it is
+** guaranteed that the returned key is larger than any key currently 
+** in the database that is less than or equal to (pKey/nKey).
+** </ol>
+**
+** LSM_SEEK_LEFAST requests are intended to be used to allocate database
+** keys.
+*/
+int lsm_csr_seek(lsm_cursor *pCsr, const void *pKey, int nKey, int eSeek);
+
+int lsm_csr_first(lsm_cursor *pCsr);
+int lsm_csr_last(lsm_cursor *pCsr);
+
+/*
+** Advance the specified cursor to the next or previous key in the database.
+** Return LSM_OK if successful, or an LSM error code otherwise.
+**
+** Functions lsm_csr_seek(), lsm_csr_first() and lsm_csr_last() are "seek"
+** functions. Whether or not lsm_csr_next and lsm_csr_prev may be called
+** successfully also depends on the most recent seek function called on
+** the cursor. Specifically:
+**
+** <ul>
+** <li> At least one seek function must have been called on the cursor.
+** <li> To call lsm_csr_next(), the most recent call to a seek function must
+** have been either lsm_csr_first() or a call to lsm_csr_seek() specifying
+** LSM_SEEK_GE.
+** <li> To call lsm_csr_prev(), the most recent call to a seek function must
+** have been either lsm_csr_last() or a call to lsm_csr_seek() specifying
+** LSM_SEEK_LE.
+** </ul>
+**
+** Otherwise, if the above conditions are not met when lsm_csr_next or 
+** lsm_csr_prev is called, LSM_MISUSE is returned and the cursor position
+** remains unchanged.
+*/
+int lsm_csr_next(lsm_cursor *pCsr);
+int lsm_csr_prev(lsm_cursor *pCsr);
+
+/*
+** Values that may be passed as the fourth argument to lsm_csr_seek().
+*/
+#define LSM_SEEK_LEFAST   -2
+#define LSM_SEEK_LE       -1
+#define LSM_SEEK_EQ        0
+#define LSM_SEEK_GE        1
+
+/* 
+** CAPI: Extracting Data From Database Cursors
+**
+** Retrieve data from a database cursor.
+*/
+int lsm_csr_valid(lsm_cursor *pCsr);
+int lsm_csr_key(lsm_cursor *pCsr, const void **ppKey, int *pnKey);
+int lsm_csr_value(lsm_cursor *pCsr, const void **ppVal, int *pnVal);
+
+/*
+** If no error occurs, this function compares the database key passed via
+** the pKey/nKey arguments with the key that the cursor passed as the first
+** argument currently points to. If the cursors key is less than, equal to
+** or greater than pKey/nKey, *piRes is set to less than, equal to or greater
+** than zero before returning. LSM_OK is returned in this case.
+**
+** Or, if an error occurs, an LSM error code is returned and the final 
+** value of *piRes is undefined. If the cursor does not point to a valid
+** key when this function is called, LSM_MISUSE is returned.
+*/
+int lsm_csr_cmp(lsm_cursor *pCsr, const void *pKey, int nKey, int *piRes);
+
+/*
+** CAPI: Change these!!
+**
+** Configure a callback to which debugging and other messages should 
+** be directed. Only useful for debugging lsm.
+*/
+void lsm_config_log(lsm_db *, void (*)(void *, int, const char *), void *);
+
+/*
+** Configure a callback that is invoked if the database connection ever
+** writes to the database file.
+*/
+void lsm_config_work_hook(lsm_db *, void (*)(lsm_db *, void *), void *);
+
+/* ENDOFAPI */
+#ifdef __cplusplus
+}  /* End of the 'extern "C"' block */
+#endif
+#endif /* ifndef _LSM_H */
diff --git a/ext/lsm1/lsmInt.h b/ext/lsm1/lsmInt.h
new file mode 100644
index 0000000..0f822e4
--- /dev/null
+++ b/ext/lsm1/lsmInt.h
@@ -0,0 +1,993 @@
+/*
+** 2011-08-18
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+** Internal structure definitions for the LSM module.
+*/
+#ifndef _LSM_INT_H
+#define _LSM_INT_H
+
+#include "lsm.h"
+#include <assert.h>
+#include <string.h>
+
+#include <stdarg.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <ctype.h>
+
+#ifdef _WIN32
+# ifdef _MSC_VER
+#  define snprintf _snprintf
+# endif
+#else
+# include <unistd.h>
+#endif
+
+#ifdef NDEBUG
+# ifdef LSM_DEBUG_EXPENSIVE
+#  undef LSM_DEBUG_EXPENSIVE
+# endif
+# ifdef LSM_DEBUG
+#  undef LSM_DEBUG
+# endif
+#else
+# ifndef LSM_DEBUG
+#  define LSM_DEBUG
+# endif
+#endif
+
+/*
+** Default values for various data structure parameters. These may be
+** overridden by calls to lsm_config().
+*/
+#define LSM_DFLT_PAGE_SIZE          (4 * 1024)
+#define LSM_DFLT_BLOCK_SIZE         (1 * 1024 * 1024)
+#define LSM_DFLT_AUTOFLUSH          (1 * 1024 * 1024)
+#define LSM_DFLT_AUTOCHECKPOINT     (i64)(2 * 1024 * 1024)
+#define LSM_DFLT_AUTOWORK           1
+#define LSM_DFLT_LOG_SIZE           (128*1024)
+#define LSM_DFLT_AUTOMERGE          4
+#define LSM_DFLT_SAFETY             LSM_SAFETY_NORMAL
+#define LSM_DFLT_MMAP               (LSM_IS_64_BIT ? 1 : 32768)
+#define LSM_DFLT_MULTIPLE_PROCESSES 1
+#define LSM_DFLT_USE_LOG            1
+
+/* Initial values for log file checksums. These are only used if the 
+** database file does not contain a valid checkpoint.  */
+#define LSM_CKSUM0_INIT 42
+#define LSM_CKSUM1_INIT 42
+
+/* "mmap" mode is currently only used in environments with 64-bit address 
+** spaces. The following macro is used to test for this.  */
+#define LSM_IS_64_BIT (sizeof(void*)==8)
+
+#define LSM_AUTOWORK_QUANT 32
+
+typedef struct Database Database;
+typedef struct DbLog DbLog;
+typedef struct FileSystem FileSystem;
+typedef struct Freelist Freelist;
+typedef struct FreelistEntry FreelistEntry;
+typedef struct Level Level;
+typedef struct LogMark LogMark;
+typedef struct LogRegion LogRegion;
+typedef struct LogWriter LogWriter;
+typedef struct LsmString LsmString;
+typedef struct Mempool Mempool;
+typedef struct Merge Merge;
+typedef struct MergeInput MergeInput;
+typedef struct MetaPage MetaPage;
+typedef struct MultiCursor MultiCursor;
+typedef struct Page Page;
+typedef struct Redirect Redirect;
+typedef struct Segment Segment;
+typedef struct SegmentMerger SegmentMerger;
+typedef struct ShmChunk ShmChunk;
+typedef struct ShmHeader ShmHeader;
+typedef struct ShmReader ShmReader;
+typedef struct Snapshot Snapshot;
+typedef struct TransMark TransMark;
+typedef struct Tree Tree;
+typedef struct TreeCursor TreeCursor;
+typedef struct TreeHeader TreeHeader;
+typedef struct TreeMark TreeMark;
+typedef struct TreeRoot TreeRoot;
+
+#ifndef _SQLITEINT_H_
+typedef unsigned char u8;
+typedef unsigned short int u16;
+typedef unsigned int u32;
+typedef lsm_i64 i64;
+typedef unsigned long long int u64;
+#endif
+
+/* A page number is a 64-bit integer. */
+typedef i64 LsmPgno;
+
+#ifdef LSM_DEBUG
+int lsmErrorBkpt(int);
+#else
+# define lsmErrorBkpt(x) (x)
+#endif
+
+#define LSM_PROTOCOL_BKPT lsmErrorBkpt(LSM_PROTOCOL)
+#define LSM_IOERR_BKPT    lsmErrorBkpt(LSM_IOERR)
+#define LSM_NOMEM_BKPT    lsmErrorBkpt(LSM_NOMEM)
+#define LSM_CORRUPT_BKPT  lsmErrorBkpt(LSM_CORRUPT)
+#define LSM_MISUSE_BKPT   lsmErrorBkpt(LSM_MISUSE)
+
+#define unused_parameter(x) (void)(x)
+#define array_size(x) (sizeof(x)/sizeof(x[0]))
+
+
+/* The size of each shared-memory chunk */
+#define LSM_SHM_CHUNK_SIZE (32*1024)
+
+/* The number of bytes reserved at the start of each shm chunk for MM. */
+#define LSM_SHM_CHUNK_HDR  (sizeof(ShmChunk))
+
+/* The number of available read locks. */
+#define LSM_LOCK_NREADER   6
+
+/* The number of available read-write client locks. */
+#define LSM_LOCK_NRWCLIENT   16
+
+/* Lock definitions. 
+*/
+#define LSM_LOCK_DMS1         1   /* Serialize connect/disconnect ops */
+#define LSM_LOCK_DMS2         2   /* Read-write connections */
+#define LSM_LOCK_DMS3         3   /* Read-only connections */
+#define LSM_LOCK_WRITER       4
+#define LSM_LOCK_WORKER       5
+#define LSM_LOCK_CHECKPOINTER 6
+#define LSM_LOCK_ROTRANS      7
+#define LSM_LOCK_READER(i)    ((i) + LSM_LOCK_ROTRANS + 1)
+#define LSM_LOCK_RWCLIENT(i)  ((i) + LSM_LOCK_READER(LSM_LOCK_NREADER))
+
+#define LSM_N_LOCK LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT)
+
+/*
+** Meta-page size and usable size.
+*/
+#define LSM_META_PAGE_SIZE 4096
+
+#define LSM_META_RW_PAGE_SIZE (LSM_META_PAGE_SIZE - LSM_N_LOCK)
+
+/*
+** Hard limit on the number of free-list entries that may be stored in 
+** a checkpoint (the remainder are stored as a system record in the LSM).
+** See also LSM_CONFIG_MAX_FREELIST.
+*/
+#define LSM_MAX_FREELIST_ENTRIES 24
+
+#define LSM_MAX_BLOCK_REDIRECTS 16
+
+#define LSM_ATTEMPTS_BEFORE_PROTOCOL 10000
+
+
+/*
+** Each entry stored in the LSM (or in-memory tree structure) has an
+** associated mask of the following flags.
+*/
+#define LSM_START_DELETE 0x01     /* Start of open-ended delete range */
+#define LSM_END_DELETE   0x02     /* End of open-ended delete range */
+#define LSM_POINT_DELETE 0x04     /* Delete this key */
+#define LSM_INSERT       0x08     /* Insert this key and value */
+#define LSM_SEPARATOR    0x10     /* True if entry is separator key only */
+#define LSM_SYSTEMKEY    0x20     /* True if entry is a system key (FREELIST) */
+
+#define LSM_CONTIGUOUS   0x40     /* Used in lsm_tree.c */
+
+/*
+** A string that can grow by appending.
+*/
+struct LsmString {
+  lsm_env *pEnv;              /* Run-time environment */
+  int n;                      /* Size of string.  -1 indicates error */
+  int nAlloc;                 /* Space allocated for z[] */
+  char *z;                    /* The string content */
+};
+
+typedef struct LsmFile LsmFile;
+struct LsmFile {
+  lsm_file *pFile;
+  LsmFile *pNext;
+};
+
+/*
+** An instance of the following type is used to store an ordered list of
+** u32 values. 
+**
+** Note: This is a place-holder implementation. It should be replaced by
+** a version that avoids making a single large allocation when the array
+** contains a large number of values. For this reason, the internals of 
+** this object should only manipulated by the intArrayXXX() functions in 
+** lsm_tree.c.
+*/
+typedef struct IntArray IntArray;
+struct IntArray {
+  int nAlloc;
+  int nArray;
+  u32 *aArray;
+};
+
+struct Redirect {
+  int n;                          /* Number of redirects */
+  struct RedirectEntry {
+    int iFrom;
+    int iTo;
+  } *a;
+};
+
+/*
+** An instance of this structure represents a point in the history of the
+** tree structure to roll back to. Refer to comments in lsm_tree.c for 
+** details.
+*/
+struct TreeMark {
+  u32 iRoot;                      /* Offset of root node in shm file */
+  u32 nHeight;                    /* Current height of tree structure */
+  u32 iWrite;                     /* Write offset in shm file */
+  u32 nChunk;                     /* Number of chunks in shared-memory file */
+  u32 iFirst;                     /* First chunk in linked list */
+  u32 iNextShmid;                 /* Next id to allocate */
+  int iRollback;                  /* Index in lsm->rollback to revert to */
+};
+
+/*
+** An instance of this structure represents a point in the database log.
+*/
+struct LogMark {
+  i64 iOff;                       /* Offset into log (see lsm_log.c) */
+  int nBuf;                       /* Size of in-memory buffer here */
+  u8 aBuf[8];                     /* Bytes of content in aBuf[] */
+  u32 cksum0;                     /* Checksum 0 at offset (iOff-nBuf) */
+  u32 cksum1;                     /* Checksum 1 at offset (iOff-nBuf) */
+};
+
+struct TransMark {
+  TreeMark tree;
+  LogMark log;
+};
+
+/*
+** A structure that defines the start and end offsets of a region in the
+** log file. The size of the region in bytes is (iEnd - iStart), so if
+** iEnd==iStart the region is zero bytes in size.
+*/
+struct LogRegion {
+  i64 iStart;                     /* Start of region in log file */
+  i64 iEnd;                       /* End of region in log file */
+};
+
+struct DbLog {
+  u32 cksum0;                     /* Checksum 0 at offset iOff */
+  u32 cksum1;                     /* Checksum 1 at offset iOff */
+  i64 iSnapshotId;                /* Log space has been reclaimed to this ss */
+  LogRegion aRegion[3];           /* Log file regions (see docs in lsm_log.c) */
+};
+
+struct TreeRoot {
+  u32 iRoot;
+  u32 nHeight;
+  u32 nByte;                      /* Total size of this tree in bytes */
+  u32 iTransId;
+};
+
+/*
+** Tree header structure. 
+*/
+struct TreeHeader {
+  u32 iUsedShmid;                 /* Id of first shm chunk used by this tree */
+  u32 iNextShmid;                 /* Shm-id of next chunk allocated */
+  u32 iFirst;                     /* Chunk number of smallest shm-id */
+  u32 nChunk;                     /* Number of chunks in shared-memory file */
+  TreeRoot root;                  /* Root and height of current tree */
+  u32 iWrite;                     /* Write offset in shm file */
+  TreeRoot oldroot;               /* Root and height of the previous tree */
+  u32 iOldShmid;                  /* Last shm-id used by previous tree */
+  u32 iUsrVersion;                /* get/set_user_version() value */
+  i64 iOldLog;                    /* Log offset associated with old tree */
+  u32 oldcksum0;
+  u32 oldcksum1;
+  DbLog log;                      /* Current layout of log file */ 
+  u32 aCksum[2];                  /* Checksums 1 and 2. */
+};
+
+/*
+** Database handle structure.
+**
+** mLock:
+**   A bitmask representing the locks currently held by the connection.
+**   An LSM database supports N distinct locks, where N is some number less
+**   than or equal to 32. Locks are numbered starting from 1 (see the 
+**   definitions for LSM_LOCK_WRITER and co.).
+**
+**   The least significant 32-bits in mLock represent EXCLUSIVE locks. The
+**   most significant are SHARED locks. So, if a connection holds a SHARED
+**   lock on lock region iLock, then the following is true:
+**
+**       (mLock & ((iLock+32-1) << 1))
+**
+**   Or for an EXCLUSIVE lock:
+**
+**       (mLock & ((iLock-1) << 1))
+** 
+** pCsr:
+**   Points to the head of a linked list that contains all currently open
+**   cursors. Once this list becomes empty, the user has no outstanding
+**   cursors and the database handle can be successfully closed.
+**
+** pCsrCache:
+**   This list contains cursor objects that have been closed using
+**   lsm_csr_close(). Each time a cursor is closed, it is shifted from 
+**   the pCsr list to this list. When a new cursor is opened, this list
+**   is inspected to see if there exists a cursor object that can be
+**   reused. This is an optimization only.
+*/
+struct lsm_db {
+
+  /* Database handle configuration */
+  lsm_env *pEnv;                            /* runtime environment */
+  int (*xCmp)(void *, int, void *, int);    /* Compare function */
+
+  /* Values configured by calls to lsm_config */
+  int eSafety;                    /* LSM_SAFETY_OFF, NORMAL or FULL */
+  int bAutowork;                  /* Configured by LSM_CONFIG_AUTOWORK */
+  int nTreeLimit;                 /* Configured by LSM_CONFIG_AUTOFLUSH */
+  int nMerge;                     /* Configured by LSM_CONFIG_AUTOMERGE */
+  int bUseLog;                    /* Configured by LSM_CONFIG_USE_LOG */
+  int nDfltPgsz;                  /* Configured by LSM_CONFIG_PAGE_SIZE */
+  int nDfltBlksz;                 /* Configured by LSM_CONFIG_BLOCK_SIZE */
+  int nMaxFreelist;               /* Configured by LSM_CONFIG_MAX_FREELIST */
+  int iMmap;                      /* Configured by LSM_CONFIG_MMAP */
+  i64 nAutockpt;                  /* Configured by LSM_CONFIG_AUTOCHECKPOINT */
+  int bMultiProc;                 /* Configured by L_C_MULTIPLE_PROCESSES */
+  int bReadonly;                  /* Configured by LSM_CONFIG_READONLY */
+  lsm_compress compress;          /* Compression callbacks */
+  lsm_compress_factory factory;   /* Compression callback factory */
+
+  /* Sub-system handles */
+  FileSystem *pFS;                /* On-disk portion of database */
+  Database *pDatabase;            /* Database shared data */
+
+  int iRwclient;                  /* Read-write client lock held (-1 == none) */
+
+  /* Client transaction context */
+  Snapshot *pClient;              /* Client snapshot */
+  int iReader;                    /* Read lock held (-1 == unlocked) */
+  int bRoTrans;                   /* True if a read-only db trans is open */
+  MultiCursor *pCsr;              /* List of all open cursors */
+  LogWriter *pLogWriter;          /* Context for writing to the log file */
+  int nTransOpen;                 /* Number of opened write transactions */
+  int nTransAlloc;                /* Allocated size of aTrans[] array */
+  TransMark *aTrans;              /* Array of marks for transaction rollback */
+  IntArray rollback;              /* List of tree-nodes to roll back */
+  int bDiscardOld;                /* True if lsmTreeDiscardOld() was called */
+
+  MultiCursor *pCsrCache;         /* List of all closed cursors */
+
+  /* Worker context */
+  Snapshot *pWorker;              /* Worker snapshot (or NULL) */
+  Freelist *pFreelist;            /* See sortedNewToplevel() */
+  int bUseFreelist;               /* True to use pFreelist */
+  int bIncrMerge;                 /* True if currently doing a merge */
+
+  int bInFactory;                 /* True if within factory.xFactory() */
+
+  /* Debugging message callback */
+  void (*xLog)(void *, int, const char *);
+  void *pLogCtx;
+
+  /* Work done notification callback */
+  void (*xWork)(lsm_db *, void *);
+  void *pWorkCtx;
+
+  u64 mLock;                      /* Mask of current locks. See lsmShmLock(). */
+  lsm_db *pNext;                  /* Next connection to same database */
+
+  int nShm;                       /* Size of apShm[] array */
+  void **apShm;                   /* Shared memory chunks */
+  ShmHeader *pShmhdr;             /* Live shared-memory header */
+  TreeHeader treehdr;             /* Local copy of tree-header */
+  u32 aSnapshot[LSM_META_PAGE_SIZE / sizeof(u32)];
+};
+
+struct Segment {
+  LsmPgno iFirst;                  /* First page of this run */
+  LsmPgno iLastPg;                 /* Last page of this run */
+  LsmPgno iRoot;                   /* Root page number (if any) */
+  int nSize;                       /* Size of this run in pages */
+
+  Redirect *pRedirect;             /* Block redirects (or NULL) */
+};
+
+/*
+** iSplitTopic/pSplitKey/nSplitKey:
+**   If nRight>0, this buffer contains a copy of the largest key that has
+**   already been written to the left-hand-side of the level.
+*/
+struct Level {
+  Segment lhs;                    /* Left-hand (main) segment */
+  int nRight;                     /* Size of apRight[] array */
+  Segment *aRhs;                  /* Old segments being merged into this */
+  int iSplitTopic;                /* Split key topic (if nRight>0) */
+  void *pSplitKey;                /* Pointer to split-key (if nRight>0) */
+  int nSplitKey;                  /* Number of bytes in split-key */
+
+  u16 iAge;                       /* Number of times data has been written */
+  u16 flags;                      /* Mask of LEVEL_XXX bits */
+  Merge *pMerge;                  /* Merge operation currently underway */
+  Level *pNext;                   /* Next level in tree */
+};
+
+/*
+** The Level.flags field is set to a combination of the following bits.
+**
+** LEVEL_FREELIST_ONLY:
+**   Set if the level consists entirely of free-list entries. 
+**
+** LEVEL_INCOMPLETE:
+**   This is set while a new toplevel level is being constructed. It is
+**   never set for any level other than a new toplevel.
+*/
+#define LEVEL_FREELIST_ONLY      0x0001
+#define LEVEL_INCOMPLETE         0x0002
+
+
+/*
+** A structure describing an ongoing merge. There is an instance of this
+** structure for every Level currently undergoing a merge in the worker
+** snapshot.
+**
+** It is assumed that code that uses an instance of this structure has
+** access to the associated Level struct.
+**
+** iOutputOff:
+**   The byte offset to write to next within the last page of the 
+**   output segment.
+*/
+struct MergeInput {
+  LsmPgno iPg;                    /* Page on which next input is stored */
+  int iCell;                      /* Cell containing next input to merge */
+};
+struct Merge {
+  int nInput;                     /* Number of input runs being merged */
+  MergeInput *aInput;             /* Array nInput entries in size */
+  MergeInput splitkey;            /* Location in file of current splitkey */
+  int nSkip;                      /* Number of separators entries to skip */
+  int iOutputOff;                 /* Write offset on output page */
+  LsmPgno iCurrentPtr;            /* Current pointer value */
+};
+
+/* 
+** The first argument to this macro is a pointer to a Segment structure.
+** Returns true if the structure instance indicates that the separators
+** array is valid.
+*/
+#define segmentHasSeparators(pSegment) ((pSegment)->sep.iFirst>0)
+
+/*
+** The values that accompany the lock held by a database reader.
+*/
+struct ShmReader {
+  u32 iTreeId;
+  i64 iLsmId;
+};
+
+/*
+** An instance of this structure is stored in the first shared-memory
+** page. The shared-memory header.
+**
+** bWriter:
+**   Immediately after opening a write transaction taking the WRITER lock, 
+**   each writer client sets this flag. It is cleared right before the 
+**   WRITER lock is relinquished. If a subsequent writer finds that this
+**   flag is already set when a write transaction is opened, this indicates
+**   that a previous writer failed mid-transaction.
+**
+** iMetaPage:
+**   If the database file does not contain a valid, synced, checkpoint, this
+**   value is set to 0. Otherwise, it is set to the meta-page number that
+**   contains the most recently written checkpoint (either 1 or 2).
+**
+** hdr1, hdr2:
+**   The two copies of the in-memory tree header. Two copies are required
+**   in case a writer fails while updating one of them.
+*/
+struct ShmHeader {
+  u32 aSnap1[LSM_META_PAGE_SIZE / 4];
+  u32 aSnap2[LSM_META_PAGE_SIZE / 4];
+  u32 bWriter;
+  u32 iMetaPage;
+  TreeHeader hdr1;
+  TreeHeader hdr2;
+  ShmReader aReader[LSM_LOCK_NREADER];
+};
+
+/*
+** An instance of this structure is stored at the start of each shared-memory
+** chunk except the first (which is the header chunk - see above).
+*/
+struct ShmChunk {
+  u32 iShmid;
+  u32 iNext;
+};
+
+/*
+** Maximum number of shared-memory chunks allowed in the *-shm file. Since
+** each shared-memory chunk is 32KB in size, this is a theoretical limit only.
+*/
+#define LSM_MAX_SHMCHUNKS  (1<<30)
+
+/* Return true if shm-sequence "a" is larger than or equal to "b" */
+#define shm_sequence_ge(a, b) (((u32)a-(u32)b) < LSM_MAX_SHMCHUNKS)
+
+#define LSM_APPLIST_SZ 4
+
+/*
+** An instance of the following structure stores the in-memory part of
+** the current free block list. This structure is to the free block list
+** as the in-memory tree is to the users database content. The contents 
+** of the free block list is found by merging the in-memory components 
+** with those stored in the LSM, just as the contents of the database is
+** found by merging the in-memory tree with the user data entries in the
+** LSM.
+**
+** Each FreelistEntry structure in the array represents either an insert
+** or delete operation on the free-list. For deletes, the FreelistEntry.iId
+** field is set to -1. For inserts, it is set to zero or greater. 
+**
+** The array of FreelistEntry structures is always sorted in order of
+** block number (ascending).
+**
+** When the in-memory free block list is written into the LSM, each insert
+** operation is written separately. The entry key is the bitwise inverse
+** of the block number as a 32-bit big-endian integer. This is done so that
+** the entries in the LSM are sorted in descending order of block id. 
+** The associated value is the snapshot id, formated as a varint.
+*/
+struct Freelist {
+  FreelistEntry *aEntry;          /* Free list entries */
+  int nEntry;                     /* Number of valid slots in aEntry[] */
+  int nAlloc;                     /* Allocated size of aEntry[] */
+};
+struct FreelistEntry {
+  u32 iBlk;                       /* Block number */
+  i64 iId;                        /* Largest snapshot id to use this block */
+};
+
+/*
+** A snapshot of a database. A snapshot contains all the information required
+** to read or write a database file on disk. See the description of struct
+** Database below for futher details.
+*/
+struct Snapshot {
+  Database *pDatabase;            /* Database this snapshot belongs to */
+  u32 iCmpId;                     /* Id of compression scheme */
+  Level *pLevel;                  /* Pointer to level 0 of snapshot (or NULL) */
+  i64 iId;                        /* Snapshot id */
+  i64 iLogOff;                    /* Log file offset */
+  Redirect redirect;              /* Block redirection array */
+
+  /* Used by worker snapshots only */
+  int nBlock;                        /* Number of blocks in database file */
+  LsmPgno aiAppend[LSM_APPLIST_SZ];  /* Append point list */
+  Freelist freelist;                 /* Free block list */
+  u32 nWrite;                        /* Total number of pages written to disk */
+};
+#define LSM_INITIAL_SNAPSHOT_ID 11
+
+/*
+** Functions from file "lsm_ckpt.c".
+*/
+int lsmCheckpointWrite(lsm_db *, u32 *);
+int lsmCheckpointLevels(lsm_db *, int, void **, int *);
+int lsmCheckpointLoadLevels(lsm_db *pDb, void *pVal, int nVal);
+
+int lsmCheckpointRecover(lsm_db *);
+int lsmCheckpointDeserialize(lsm_db *, int, u32 *, Snapshot **);
+
+int lsmCheckpointLoadWorker(lsm_db *pDb);
+int lsmCheckpointStore(lsm_db *pDb, int);
+
+int lsmCheckpointLoad(lsm_db *pDb, int *);
+int lsmCheckpointLoadOk(lsm_db *pDb, int);
+int lsmCheckpointClientCacheOk(lsm_db *);
+
+u32 lsmCheckpointNBlock(u32 *);
+i64 lsmCheckpointId(u32 *, int);
+u32 lsmCheckpointNWrite(u32 *, int);
+i64 lsmCheckpointLogOffset(u32 *);
+int lsmCheckpointPgsz(u32 *);
+int lsmCheckpointBlksz(u32 *);
+void lsmCheckpointLogoffset(u32 *aCkpt, DbLog *pLog);
+void lsmCheckpointZeroLogoffset(lsm_db *);
+
+int lsmCheckpointSaveWorker(lsm_db *pDb, int);
+int lsmDatabaseFull(lsm_db *pDb);
+int lsmCheckpointSynced(lsm_db *pDb, i64 *piId, i64 *piLog, u32 *pnWrite);
+
+int lsmCheckpointSize(lsm_db *db, int *pnByte);
+
+int lsmInfoCompressionId(lsm_db *db, u32 *piCmpId);
+
+/* 
+** Functions from file "lsm_tree.c".
+*/
+int lsmTreeNew(lsm_env *, int (*)(void *, int, void *, int), Tree **ppTree);
+void lsmTreeRelease(lsm_env *, Tree *);
+int lsmTreeInit(lsm_db *);
+int lsmTreeRepair(lsm_db *);
+
+void lsmTreeMakeOld(lsm_db *pDb);
+void lsmTreeDiscardOld(lsm_db *pDb);
+int lsmTreeHasOld(lsm_db *pDb);
+
+int lsmTreeSize(lsm_db *);
+int lsmTreeEndTransaction(lsm_db *pDb, int bCommit);
+int lsmTreeLoadHeader(lsm_db *pDb, int *);
+int lsmTreeLoadHeaderOk(lsm_db *, int);
+
+int lsmTreeInsert(lsm_db *pDb, void *pKey, int nKey, void *pVal, int nVal);
+int lsmTreeDelete(lsm_db *db, void *pKey1, int nKey1, void *pKey2, int nKey2);
+void lsmTreeRollback(lsm_db *pDb, TreeMark *pMark);
+void lsmTreeMark(lsm_db *pDb, TreeMark *pMark);
+
+int lsmTreeCursorNew(lsm_db *pDb, int, TreeCursor **);
+void lsmTreeCursorDestroy(TreeCursor *);
+
+int lsmTreeCursorSeek(TreeCursor *pCsr, void *pKey, int nKey, int *pRes);
+int lsmTreeCursorNext(TreeCursor *pCsr);
+int lsmTreeCursorPrev(TreeCursor *pCsr);
+int lsmTreeCursorEnd(TreeCursor *pCsr, int bLast);
+void lsmTreeCursorReset(TreeCursor *pCsr);
+int lsmTreeCursorKey(TreeCursor *pCsr, int *pFlags, void **ppKey, int *pnKey);
+int lsmTreeCursorFlags(TreeCursor *pCsr);
+int lsmTreeCursorValue(TreeCursor *pCsr, void **ppVal, int *pnVal);
+int lsmTreeCursorValid(TreeCursor *pCsr);
+int lsmTreeCursorSave(TreeCursor *pCsr);
+
+void lsmFlagsToString(int flags, char *zFlags);
+
+/* 
+** Functions from file "mem.c".
+*/
+void *lsmMalloc(lsm_env*, size_t);
+void lsmFree(lsm_env*, void *);
+void *lsmRealloc(lsm_env*, void *, size_t);
+void *lsmReallocOrFree(lsm_env*, void *, size_t);
+void *lsmReallocOrFreeRc(lsm_env *, void *, size_t, int *);
+
+void *lsmMallocZeroRc(lsm_env*, size_t, int *);
+void *lsmMallocRc(lsm_env*, size_t, int *);
+
+void *lsmMallocZero(lsm_env *pEnv, size_t);
+char *lsmMallocStrdup(lsm_env *pEnv, const char *);
+
+/* 
+** Functions from file "lsm_mutex.c".
+*/
+int lsmMutexStatic(lsm_env*, int, lsm_mutex **);
+int lsmMutexNew(lsm_env*, lsm_mutex **);
+void lsmMutexDel(lsm_env*, lsm_mutex *);
+void lsmMutexEnter(lsm_env*, lsm_mutex *);
+int lsmMutexTry(lsm_env*, lsm_mutex *);
+void lsmMutexLeave(lsm_env*, lsm_mutex *);
+
+#ifndef NDEBUG
+int lsmMutexHeld(lsm_env *, lsm_mutex *);
+int lsmMutexNotHeld(lsm_env *, lsm_mutex *);
+#endif
+
+/**************************************************************************
+** Start of functions from "lsm_file.c".
+*/
+int lsmFsOpen(lsm_db *, const char *, int);
+int lsmFsOpenLog(lsm_db *, int *);
+void lsmFsCloseLog(lsm_db *);
+void lsmFsClose(FileSystem *);
+
+int lsmFsUnmap(FileSystem *);
+
+int lsmFsConfigure(lsm_db *db);
+
+int lsmFsBlockSize(FileSystem *);
+void lsmFsSetBlockSize(FileSystem *, int);
+int lsmFsMoveBlock(FileSystem *pFS, Segment *pSeg, int iTo, int iFrom);
+
+int lsmFsPageSize(FileSystem *);
+void lsmFsSetPageSize(FileSystem *, int);
+
+int lsmFsFileid(lsm_db *pDb, void **ppId, int *pnId);
+
+/* Creating, populating, gobbling and deleting sorted runs. */
+void lsmFsGobble(lsm_db *, Segment *, LsmPgno *, int);
+int lsmFsSortedDelete(FileSystem *, Snapshot *, int, Segment *);
+int lsmFsSortedFinish(FileSystem *, Segment *);
+int lsmFsSortedAppend(FileSystem *, Snapshot *, Level *, int, Page **);
+int lsmFsSortedPadding(FileSystem *, Snapshot *, Segment *);
+
+/* Functions to retrieve the lsm_env pointer from a FileSystem or Page object */
+lsm_env *lsmFsEnv(FileSystem *);
+lsm_env *lsmPageEnv(Page *);
+FileSystem *lsmPageFS(Page *);
+
+int lsmFsSectorSize(FileSystem *);
+
+void lsmSortedSplitkey(lsm_db *, Level *, int *);
+
+/* Reading sorted run content. */
+int lsmFsDbPageLast(FileSystem *pFS, Segment *pSeg, Page **ppPg);
+int lsmFsDbPageGet(FileSystem *, Segment *, LsmPgno, Page **);
+int lsmFsDbPageNext(Segment *, Page *, int eDir, Page **);
+
+u8 *lsmFsPageData(Page *, int *);
+int lsmFsPageRelease(Page *);
+int lsmFsPagePersist(Page *);
+void lsmFsPageRef(Page *);
+LsmPgno lsmFsPageNumber(Page *);
+
+int lsmFsNRead(FileSystem *);
+int lsmFsNWrite(FileSystem *);
+
+int lsmFsMetaPageGet(FileSystem *, int, int, MetaPage **);
+int lsmFsMetaPageRelease(MetaPage *);
+u8 *lsmFsMetaPageData(MetaPage *, int *);
+
+#ifdef LSM_DEBUG
+int lsmFsDbPageIsLast(Segment *pSeg, Page *pPg);
+int lsmFsIntegrityCheck(lsm_db *);
+#endif
+
+LsmPgno lsmFsRedirectPage(FileSystem *, Redirect *, LsmPgno);
+
+int lsmFsPageWritable(Page *);
+
+/* Functions to read, write and sync the log file. */
+int lsmFsWriteLog(FileSystem *pFS, i64 iOff, LsmString *pStr);
+int lsmFsSyncLog(FileSystem *pFS);
+int lsmFsReadLog(FileSystem *pFS, i64 iOff, int nRead, LsmString *pStr);
+int lsmFsTruncateLog(FileSystem *pFS, i64 nByte);
+int lsmFsTruncateDb(FileSystem *pFS, i64 nByte);
+int lsmFsCloseAndDeleteLog(FileSystem *pFS);
+
+LsmFile *lsmFsDeferClose(FileSystem *pFS);
+
+/* And to sync the db file */
+int lsmFsSyncDb(FileSystem *, int);
+
+void lsmFsFlushWaiting(FileSystem *, int *);
+
+/* Used by lsm_info(ARRAY_STRUCTURE) and lsm_config(MMAP) */
+int lsmInfoArrayStructure(lsm_db *pDb, int bBlock, LsmPgno iFirst, char **pz);
+int lsmInfoArrayPages(lsm_db *pDb, LsmPgno iFirst, char **pzOut);
+int lsmConfigMmap(lsm_db *pDb, int *piParam);
+
+int lsmEnvOpen(lsm_env *, const char *, int, lsm_file **);
+int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile);
+int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock);
+int lsmEnvTestLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int nLock, int);
+
+int lsmEnvShmMap(lsm_env *, lsm_file *, int, int, void **); 
+void lsmEnvShmBarrier(lsm_env *);
+void lsmEnvShmUnmap(lsm_env *, lsm_file *, int);
+
+void lsmEnvSleep(lsm_env *, int);
+
+int lsmFsReadSyncedId(lsm_db *db, int, i64 *piVal);
+
+int lsmFsSegmentContainsPg(FileSystem *pFS, Segment *, LsmPgno, int *);
+
+void lsmFsPurgeCache(FileSystem *);
+
+/*
+** End of functions from "lsm_file.c".
+**************************************************************************/
+
+/* 
+** Functions from file "lsm_sorted.c".
+*/
+int lsmInfoPageDump(lsm_db *, LsmPgno, int, char **);
+void lsmSortedCleanup(lsm_db *);
+int lsmSortedAutoWork(lsm_db *, int nUnit);
+
+int lsmSortedWalkFreelist(lsm_db *, int, int (*)(void *, int, i64), void *);
+
+int lsmSaveWorker(lsm_db *, int);
+
+int lsmFlushTreeToDisk(lsm_db *pDb);
+
+void lsmSortedRemap(lsm_db *pDb);
+
+void lsmSortedFreeLevel(lsm_env *pEnv, Level *);
+
+int lsmSortedAdvanceAll(lsm_db *pDb);
+
+int lsmSortedLoadMerge(lsm_db *, Level *, u32 *, int *);
+int lsmSortedLoadFreelist(lsm_db *pDb, void **, int *);
+
+void *lsmSortedSplitKey(Level *pLevel, int *pnByte);
+
+void lsmSortedSaveTreeCursors(lsm_db *);
+
+int lsmMCursorNew(lsm_db *, MultiCursor **);
+void lsmMCursorClose(MultiCursor *, int);
+int lsmMCursorSeek(MultiCursor *, int, void *, int , int);
+int lsmMCursorFirst(MultiCursor *);
+int lsmMCursorPrev(MultiCursor *);
+int lsmMCursorLast(MultiCursor *);
+int lsmMCursorValid(MultiCursor *);
+int lsmMCursorNext(MultiCursor *);
+int lsmMCursorKey(MultiCursor *, void **, int *);
+int lsmMCursorValue(MultiCursor *, void **, int *);
+int lsmMCursorType(MultiCursor *, int *);
+lsm_db *lsmMCursorDb(MultiCursor *);
+void lsmMCursorFreeCache(lsm_db *);
+
+int lsmSaveCursors(lsm_db *pDb);
+int lsmRestoreCursors(lsm_db *pDb);
+
+void lsmSortedDumpStructure(lsm_db *pDb, Snapshot *, int, int, const char *);
+void lsmFsDumpBlocklists(lsm_db *);
+
+void lsmSortedExpandBtreePage(Page *pPg, int nOrig);
+
+void lsmPutU32(u8 *, u32);
+u32 lsmGetU32(u8 *);
+u64 lsmGetU64(u8 *);
+
+/*
+** Functions from "lsm_varint.c".
+*/
+int lsmVarintPut32(u8 *, int);
+int lsmVarintGet32(u8 *, int *);
+int lsmVarintPut64(u8 *aData, i64 iVal);
+int lsmVarintGet64(const u8 *aData, i64 *piVal);
+
+int lsmVarintLen32(int);
+int lsmVarintSize(u8 c);
+
+/* 
+** Functions from file "main.c".
+*/
+void lsmLogMessage(lsm_db *, int, const char *, ...);
+int lsmInfoFreelist(lsm_db *pDb, char **pzOut);
+
+/*
+** Functions from file "lsm_log.c".
+*/
+int lsmLogBegin(lsm_db *pDb);
+int lsmLogWrite(lsm_db *, int, void *, int, void *, int);
+int lsmLogCommit(lsm_db *);
+void lsmLogEnd(lsm_db *pDb, int bCommit);
+void lsmLogTell(lsm_db *, LogMark *);
+void lsmLogSeek(lsm_db *, LogMark *);
+void lsmLogClose(lsm_db *);
+
+int lsmLogRecover(lsm_db *);
+int lsmInfoLogStructure(lsm_db *pDb, char **pzVal);
+
+/* Valid values for the second argument to lsmLogWrite(). */
+#define LSM_WRITE        0x06
+#define LSM_DELETE       0x08
+#define LSM_DRANGE       0x0A
+
+/**************************************************************************
+** Functions from file "lsm_shared.c".
+*/
+
+int lsmDbDatabaseConnect(lsm_db*, const char *);
+void lsmDbDatabaseRelease(lsm_db *);
+
+int lsmBeginReadTrans(lsm_db *);
+int lsmBeginWriteTrans(lsm_db *);
+int lsmBeginFlush(lsm_db *);
+
+int lsmDetectRoTrans(lsm_db *db, int *);
+int lsmBeginRoTrans(lsm_db *db);
+
+int lsmBeginWork(lsm_db *);
+void lsmFinishWork(lsm_db *, int, int *);
+
+int lsmFinishRecovery(lsm_db *);
+void lsmFinishReadTrans(lsm_db *);
+int lsmFinishWriteTrans(lsm_db *, int);
+int lsmFinishFlush(lsm_db *, int);
+
+int lsmSnapshotSetFreelist(lsm_db *, int *, int);
+
+Snapshot *lsmDbSnapshotClient(lsm_db *);
+Snapshot *lsmDbSnapshotWorker(lsm_db *);
+
+void lsmSnapshotSetCkptid(Snapshot *, i64);
+
+Level *lsmDbSnapshotLevel(Snapshot *);
+void lsmDbSnapshotSetLevel(Snapshot *, Level *);
+
+void lsmDbRecoveryComplete(lsm_db *, int);
+
+int lsmBlockAllocate(lsm_db *, int, int *);
+int lsmBlockFree(lsm_db *, int);
+int lsmBlockRefree(lsm_db *, int);
+
+void lsmFreelistDeltaBegin(lsm_db *);
+void lsmFreelistDeltaEnd(lsm_db *);
+int lsmFreelistDelta(lsm_db *pDb);
+
+DbLog *lsmDatabaseLog(lsm_db *pDb);
+
+#ifdef LSM_DEBUG
+  int lsmHoldingClientMutex(lsm_db *pDb);
+  int lsmShmAssertLock(lsm_db *db, int iLock, int eOp);
+  int lsmShmAssertWorker(lsm_db *db);
+#endif
+
+void lsmFreeSnapshot(lsm_env *, Snapshot *);
+
+
+/* Candidate values for the 3rd argument to lsmShmLock() */
+#define LSM_LOCK_UNLOCK 0
+#define LSM_LOCK_SHARED 1
+#define LSM_LOCK_EXCL   2
+
+int lsmShmCacheChunks(lsm_db *db, int nChunk);
+int lsmShmLock(lsm_db *db, int iLock, int eOp, int bBlock);
+int lsmShmTestLock(lsm_db *db, int iLock, int nLock, int eOp);
+void lsmShmBarrier(lsm_db *db);
+
+#ifdef LSM_DEBUG
+void lsmShmHasLock(lsm_db *db, int iLock, int eOp);
+#else
+# define lsmShmHasLock(x,y,z)
+#endif
+
+int lsmReadlock(lsm_db *, i64 iLsm, u32 iShmMin, u32 iShmMax);
+
+int lsmLsmInUse(lsm_db *db, i64 iLsmId, int *pbInUse);
+int lsmTreeInUse(lsm_db *db, u32 iLsmId, int *pbInUse);
+int lsmFreelistAppend(lsm_env *pEnv, Freelist *p, int iBlk, i64 iId);
+
+int lsmDbMultiProc(lsm_db *);
+void lsmDbDeferredClose(lsm_db *, lsm_file *, LsmFile *);
+LsmFile *lsmDbRecycleFd(lsm_db *);
+
+int lsmWalkFreelist(lsm_db *, int, int (*)(void *, int, i64), void *);
+
+int lsmCheckCompressionId(lsm_db *, u32);
+
+
+/**************************************************************************
+** functions in lsm_str.c
+*/
+void lsmStringInit(LsmString*, lsm_env *pEnv);
+int lsmStringExtend(LsmString*, int);
+int lsmStringAppend(LsmString*, const char *, int);
+void lsmStringVAppendf(LsmString*, const char *zFormat, va_list, va_list);
+void lsmStringAppendf(LsmString*, const char *zFormat, ...);
+void lsmStringClear(LsmString*);
+char *lsmMallocPrintf(lsm_env*, const char*, ...);
+int lsmStringBinAppend(LsmString *pStr, const u8 *a, int n);
+
+int lsmStrlen(const char *zName);
+
+
+
+/* 
+** Round up a number to the next larger multiple of 8.  This is used
+** to force 8-byte alignment on 64-bit architectures.
+*/
+#define ROUND8(x)     (((x)+7)&~7)
+
+#define LSM_MIN(x,y) ((x)>(y) ? (y) : (x))
+#define LSM_MAX(x,y) ((x)>(y) ? (x) : (y))
+
+#endif
diff --git a/ext/lsm1/lsm_ckpt.c b/ext/lsm1/lsm_ckpt.c
new file mode 100644
index 0000000..ba92a82
--- /dev/null
+++ b/ext/lsm1/lsm_ckpt.c
@@ -0,0 +1,1239 @@
+/*
+** 2011-09-11
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** This file contains code to read and write checkpoints.
+**
+** A checkpoint represents the database layout at a single point in time.
+** It includes a log offset. When an existing database is opened, the
+** current state is determined by reading the newest checkpoint and updating
+** it with all committed transactions from the log that follow the specified
+** offset.
+*/
+#include "lsmInt.h"
+
+/*
+** CHECKPOINT BLOB FORMAT:
+**
+** A checkpoint blob is a series of unsigned 32-bit integers stored in
+** big-endian byte order. As follows:
+**
+**   Checkpoint header (see the CKPT_HDR_XXX #defines):
+**
+**     1. The checkpoint id MSW.
+**     2. The checkpoint id LSW.
+**     3. The number of integer values in the entire checkpoint, including 
+**        the two checksum values.
+**     4. The compression scheme id.
+**     5. The total number of blocks in the database.
+**     6. The block size.
+**     7. The number of levels.
+**     8. The nominal database page size.
+**     9. The number of pages (in total) written to the database file.
+**
+**   Log pointer:
+**
+**     1. The log offset MSW.
+**     2. The log offset LSW.
+**     3. Log checksum 0.
+**     4. Log checksum 1.
+**
+**     Note that the "log offset" is not the literal byte offset. Instead,
+**     it is the byte offset multiplied by 2, with least significant bit
+**     toggled each time the log pointer value is changed. This is to make
+**     sure that this field changes each time the log pointer is updated,
+**     even if the log file itself is disabled. See lsmTreeMakeOld().
+**
+**     See ckptExportLog() and ckptImportLog().
+**
+**   Append points:
+**
+**     8 integers (4 * 64-bit page numbers). See ckptExportAppendlist().
+**
+**   For each level in the database, a level record. Formatted as follows:
+**
+**     0. Age of the level (least significant 16-bits). And flags mask (most
+**        significant 16-bits).
+**     1. The number of right-hand segments (nRight, possibly 0),
+**     2. Segment record for left-hand segment (8 integers defined below),
+**     3. Segment record for each right-hand segment (8 integers defined below),
+**     4. If nRight>0, The number of segments involved in the merge
+**     5. if nRight>0, Current nSkip value (see Merge structure defn.),
+**     6. For each segment in the merge:
+**        5a. Page number of next cell to read during merge (this field
+**            is 64-bits - 2 integers)
+**        5b. Cell number of next cell to read during merge
+**     7. Page containing current split-key (64-bits - 2 integers).
+**     8. Cell within page containing current split-key.
+**     9. Current pointer value (64-bits - 2 integers).
+**
+**   The block redirect array:
+**
+**     1. Number of redirections (maximum LSM_MAX_BLOCK_REDIRECTS).
+**     2. For each redirection:
+**        a. "from" block number
+**        b. "to" block number
+**
+**   The in-memory freelist entries. Each entry is either an insert or a
+**   delete. The in-memory freelist is to the free-block-list as the
+**   in-memory tree is to the users database content.
+**
+**     1. Number of free-list entries stored in checkpoint header.
+**     2. Number of free blocks (in total).
+**     3. Total number of blocks freed during database lifetime.
+**     4. For each entry:
+**        2a. Block number of free block.
+**        2b. A 64-bit integer (MSW followed by LSW). -1 for a delete entry,
+**            or the associated checkpoint id for an insert.
+**
+**   The checksum:
+**
+**     1. Checksum value 1.
+**     2. Checksum value 2.
+**
+** In the above, a segment record consists of the following four 64-bit 
+** fields (converted to 2 * u32 by storing the MSW followed by LSW):
+**
+**     1. First page of array,
+**     2. Last page of array,
+**     3. Root page of array (or 0),
+**     4. Size of array in pages.
+*/
+
+/*
+** LARGE NUMBERS OF LEVEL RECORDS:
+**
+** A limit on the number of rhs segments that may be present in the database
+** file. Defining this limit ensures that all level records fit within
+** the 4096 byte limit for checkpoint blobs.
+**
+** The number of right-hand-side segments in a database is counted as 
+** follows:
+**
+**   * For each level in the database not undergoing a merge, add 1.
+**
+**   * For each level in the database that is undergoing a merge, add 
+**     the number of segments on the rhs of the level.
+**
+** A level record not undergoing a merge is 10 integers. A level record 
+** with nRhs rhs segments and (nRhs+1) input segments (i.e. including the 
+** separators from the next level) is (11*nRhs+20) integers. The maximum
+** per right-hand-side level is therefore 21 integers. So the maximum
+** size of all level records in a checkpoint is 21*40=820 integers.
+**
+** TODO: Before pointer values were changed from 32 to 64 bits, the above
+** used to come to 420 bytes - leaving significant space for a free-list
+** prefix. No more. To fix this, reduce the size of the level records in
+** a db snapshot, and improve management of the free-list tail in 
+** lsm_sorted.c. 
+*/
+#define LSM_MAX_RHS_SEGMENTS 40
+
+/*
+** LARGE NUMBERS OF FREELIST ENTRIES:
+**
+** There is also a limit (LSM_MAX_FREELIST_ENTRIES - defined in lsmInt.h)
+** on the number of free-list entries stored in a checkpoint. Since each 
+** free-list entry consists of 3 integers, the maximum free-list size is 
+** 3*100=300 integers. Combined with the limit on rhs segments defined
+** above, this ensures that a checkpoint always fits within a 4096 byte
+** meta page.
+**
+** If the database contains more than 100 free blocks, the "overflow" flag
+** in the checkpoint header is set and the remainder are stored in the
+** system FREELIST entry in the LSM (along with user data). The value
+** accompanying the FREELIST key in the LSM is, like a checkpoint, an array
+** of 32-bit big-endian integers. As follows:
+**
+**     For each entry:
+**       a. Block number of free block.
+**       b. MSW of associated checkpoint id.
+**       c. LSW of associated checkpoint id.
+**
+** The number of entries is not required - it is implied by the size of the
+** value blob containing the integer array.
+**
+** Note that the limit defined by LSM_MAX_FREELIST_ENTRIES is a hard limit.
+** The actual value used may be configured using LSM_CONFIG_MAX_FREELIST.
+*/
+
+/*
+** The argument to this macro must be of type u32. On a little-endian
+** architecture, it returns the u32 value that results from interpreting
+** the 4 bytes as a big-endian value. On a big-endian architecture, it
+** returns the value that would be produced by intepreting the 4 bytes
+** of the input value as a little-endian integer.
+*/
+#define BYTESWAP32(x) ( \
+   (((x)&0x000000FF)<<24) + (((x)&0x0000FF00)<<8)  \
+ + (((x)&0x00FF0000)>>8)  + (((x)&0xFF000000)>>24) \
+)
+
+static const int one = 1;
+#define LSM_LITTLE_ENDIAN (*(u8 *)(&one))
+
+/* Sizes, in integers, of various parts of the checkpoint. */
+#define CKPT_HDR_SIZE         9
+#define CKPT_LOGPTR_SIZE      4
+#define CKPT_APPENDLIST_SIZE  (LSM_APPLIST_SZ * 2)
+
+/* A #define to describe each integer in the checkpoint header. */
+#define CKPT_HDR_ID_MSW   0
+#define CKPT_HDR_ID_LSW   1
+#define CKPT_HDR_NCKPT    2
+#define CKPT_HDR_CMPID    3
+#define CKPT_HDR_NBLOCK   4
+#define CKPT_HDR_BLKSZ    5
+#define CKPT_HDR_NLEVEL   6
+#define CKPT_HDR_PGSZ     7
+#define CKPT_HDR_NWRITE   8
+
+#define CKPT_HDR_LO_MSW     9
+#define CKPT_HDR_LO_LSW    10
+#define CKPT_HDR_LO_CKSUM1 11
+#define CKPT_HDR_LO_CKSUM2 12
+
+typedef struct CkptBuffer CkptBuffer;
+
+/*
+** Dynamic buffer used to accumulate data for a checkpoint.
+*/
+struct CkptBuffer {
+  lsm_env *pEnv;
+  int nAlloc;
+  u32 *aCkpt;
+};
+
+/*
+** Calculate the checksum of the checkpoint specified by arguments aCkpt and
+** nCkpt. Store the checksum in *piCksum1 and *piCksum2 before returning.
+**
+** The value of the nCkpt parameter includes the two checksum values at
+** the end of the checkpoint. They are not used as inputs to the checksum 
+** calculation. The checksum is based on the array of (nCkpt-2) integers
+** at aCkpt[].
+*/
+static void ckptChecksum(u32 *aCkpt, u32 nCkpt, u32 *piCksum1, u32 *piCksum2){
+  u32 i;
+  u32 cksum1 = 1;
+  u32 cksum2 = 2;
+
+  if( nCkpt % 2 ){
+    cksum1 += aCkpt[nCkpt-3] & 0x0000FFFF;
+    cksum2 += aCkpt[nCkpt-3] & 0xFFFF0000;
+  }
+
+  for(i=0; (i+3)<nCkpt; i+=2){
+    cksum1 += cksum2 + aCkpt[i];
+    cksum2 += cksum1 + aCkpt[i+1];
+  }
+
+  *piCksum1 = cksum1;
+  *piCksum2 = cksum2;
+}
+
+/*
+** Set integer iIdx of the checkpoint accumulating in buffer *p to iVal.
+*/
+static void ckptSetValue(CkptBuffer *p, int iIdx, u32 iVal, int *pRc){
+  if( *pRc ) return;
+  if( iIdx>=p->nAlloc ){
+    int nNew = LSM_MAX(8, iIdx*2);
+    p->aCkpt = (u32 *)lsmReallocOrFree(p->pEnv, p->aCkpt, nNew*sizeof(u32));
+    if( !p->aCkpt ){
+      *pRc = LSM_NOMEM_BKPT;
+      return;
+    }
+    p->nAlloc = nNew;
+  }
+  p->aCkpt[iIdx] = iVal;
+}
+
+/*
+** Argument aInt points to an array nInt elements in size. Switch the 
+** endian-ness of each element of the array.
+*/
+static void ckptChangeEndianness(u32 *aInt, int nInt){
+  if( LSM_LITTLE_ENDIAN ){
+    int i;
+    for(i=0; i<nInt; i++) aInt[i] = BYTESWAP32(aInt[i]);
+  }
+}
+
+/*
+** Object *p contains a checkpoint in native byte-order. The checkpoint is
+** nCkpt integers in size, not including any checksum. This function sets
+** the two checksum elements of the checkpoint accordingly.
+*/
+static void ckptAddChecksum(CkptBuffer *p, int nCkpt, int *pRc){
+  if( *pRc==LSM_OK ){
+    u32 aCksum[2] = {0, 0};
+    ckptChecksum(p->aCkpt, nCkpt+2, &aCksum[0], &aCksum[1]);
+    ckptSetValue(p, nCkpt, aCksum[0], pRc);
+    ckptSetValue(p, nCkpt+1, aCksum[1], pRc);
+  }
+}
+
+static void ckptAppend64(CkptBuffer *p, int *piOut, i64 iVal, int *pRc){
+  int iOut = *piOut;
+  ckptSetValue(p, iOut++, (iVal >> 32) & 0xFFFFFFFF, pRc);
+  ckptSetValue(p, iOut++, (iVal & 0xFFFFFFFF), pRc);
+  *piOut = iOut;
+}
+
+static i64 ckptRead64(u32 *a){
+  return (((i64)a[0]) << 32) + (i64)a[1];
+}
+
+static i64 ckptGobble64(u32 *a, int *piIn){
+  int iIn = *piIn;
+  *piIn += 2;
+  return ckptRead64(&a[iIn]);
+}
+
+
+/*
+** Append a 6-value segment record corresponding to pSeg to the checkpoint 
+** buffer passed as the third argument.
+*/
+static void ckptExportSegment(
+  Segment *pSeg, 
+  CkptBuffer *p, 
+  int *piOut, 
+  int *pRc
+){
+  ckptAppend64(p, piOut, pSeg->iFirst, pRc);
+  ckptAppend64(p, piOut, pSeg->iLastPg, pRc);
+  ckptAppend64(p, piOut, pSeg->iRoot, pRc);
+  ckptAppend64(p, piOut, pSeg->nSize, pRc);
+}
+
+static void ckptExportLevel(
+  Level *pLevel,                  /* Level object to serialize */
+  CkptBuffer *p,                  /* Append new level record to this ckpt */
+  int *piOut,                     /* IN/OUT: Size of checkpoint so far */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int iOut = *piOut;
+  Merge *pMerge;
+
+  pMerge = pLevel->pMerge;
+  ckptSetValue(p, iOut++, (u32)pLevel->iAge + (u32)(pLevel->flags<<16), pRc);
+  ckptSetValue(p, iOut++, pLevel->nRight, pRc);
+  ckptExportSegment(&pLevel->lhs, p, &iOut, pRc);
+
+  assert( (pLevel->nRight>0)==(pMerge!=0) );
+  if( pMerge ){
+    int i;
+    for(i=0; i<pLevel->nRight; i++){
+      ckptExportSegment(&pLevel->aRhs[i], p, &iOut, pRc);
+    }
+    assert( pMerge->nInput==pLevel->nRight 
+         || pMerge->nInput==pLevel->nRight+1 
+    );
+    ckptSetValue(p, iOut++, pMerge->nInput, pRc);
+    ckptSetValue(p, iOut++, pMerge->nSkip, pRc);
+    for(i=0; i<pMerge->nInput; i++){
+      ckptAppend64(p, &iOut, pMerge->aInput[i].iPg, pRc);
+      ckptSetValue(p, iOut++, pMerge->aInput[i].iCell, pRc);
+    }
+    ckptAppend64(p, &iOut, pMerge->splitkey.iPg, pRc);
+    ckptSetValue(p, iOut++, pMerge->splitkey.iCell, pRc);
+    ckptAppend64(p, &iOut, pMerge->iCurrentPtr, pRc);
+  }
+
+  *piOut = iOut;
+}
+
+/*
+** Populate the log offset fields of the checkpoint buffer. 4 values.
+*/
+static void ckptExportLog(
+  lsm_db *pDb, 
+  int bFlush,
+  CkptBuffer *p, 
+  int *piOut, 
+  int *pRc
+){
+  int iOut = *piOut;
+
+  assert( iOut==CKPT_HDR_LO_MSW );
+
+  if( bFlush ){
+    i64 iOff = pDb->treehdr.iOldLog;
+    ckptAppend64(p, &iOut, iOff, pRc);
+    ckptSetValue(p, iOut++, pDb->treehdr.oldcksum0, pRc);
+    ckptSetValue(p, iOut++, pDb->treehdr.oldcksum1, pRc);
+  }else{
+    for(; iOut<=CKPT_HDR_LO_CKSUM2; iOut++){
+      ckptSetValue(p, iOut, pDb->pShmhdr->aSnap2[iOut], pRc);
+    }
+  }
+
+  assert( *pRc || iOut==CKPT_HDR_LO_CKSUM2+1 );
+  *piOut = iOut;
+}
+
+static void ckptExportAppendlist(
+  lsm_db *db,                     /* Database connection */
+  CkptBuffer *p,                  /* Checkpoint buffer to write to */
+  int *piOut,                     /* IN/OUT: Offset within checkpoint buffer */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  int i;
+  LsmPgno *aiAppend = db->pWorker->aiAppend;
+
+  for(i=0; i<LSM_APPLIST_SZ; i++){
+    ckptAppend64(p, piOut, aiAppend[i], pRc);
+  }
+};
+
+static int ckptExportSnapshot( 
+  lsm_db *pDb,                    /* Connection handle */
+  int bLog,                       /* True to update log-offset fields */
+  i64 iId,                        /* Checkpoint id */
+  int bCksum,                     /* If true, include checksums */
+  void **ppCkpt,                  /* OUT: Buffer containing checkpoint */
+  int *pnCkpt                     /* OUT: Size of checkpoint in bytes */
+){
+  int rc = LSM_OK;                /* Return Code */
+  FileSystem *pFS = pDb->pFS;     /* File system object */
+  Snapshot *pSnap = pDb->pWorker; /* Worker snapshot */
+  int nLevel = 0;                 /* Number of levels in checkpoint */
+  int iLevel;                     /* Used to count out nLevel levels */
+  int iOut = 0;                   /* Current offset in aCkpt[] */
+  Level *pLevel;                  /* Level iterator */
+  int i;                          /* Iterator used while serializing freelist */
+  CkptBuffer ckpt;
+
+  /* Initialize the output buffer */
+  memset(&ckpt, 0, sizeof(CkptBuffer));
+  ckpt.pEnv = pDb->pEnv;
+  iOut = CKPT_HDR_SIZE;
+
+  /* Write the log offset into the checkpoint. */
+  ckptExportLog(pDb, bLog, &ckpt, &iOut, &rc);
+
+  /* Write the append-point list */
+  ckptExportAppendlist(pDb, &ckpt, &iOut, &rc);
+
+  /* Figure out how many levels will be written to the checkpoint. */
+  for(pLevel=lsmDbSnapshotLevel(pSnap); pLevel; pLevel=pLevel->pNext) nLevel++;
+
+  /* Serialize nLevel levels. */
+  iLevel = 0;
+  for(pLevel=lsmDbSnapshotLevel(pSnap); iLevel<nLevel; pLevel=pLevel->pNext){
+    ckptExportLevel(pLevel, &ckpt, &iOut, &rc);
+    iLevel++;
+  }
+
+  /* Write the block-redirect list */
+  ckptSetValue(&ckpt, iOut++, pSnap->redirect.n, &rc);
+  for(i=0; i<pSnap->redirect.n; i++){
+    ckptSetValue(&ckpt, iOut++, pSnap->redirect.a[i].iFrom, &rc);
+    ckptSetValue(&ckpt, iOut++, pSnap->redirect.a[i].iTo, &rc);
+  }
+
+  /* Write the freelist */
+  assert( pSnap->freelist.nEntry<=pDb->nMaxFreelist );
+  if( rc==LSM_OK ){
+    int nFree = pSnap->freelist.nEntry;
+    ckptSetValue(&ckpt, iOut++, nFree, &rc);
+    for(i=0; i<nFree; i++){
+      FreelistEntry *p = &pSnap->freelist.aEntry[i];
+      ckptSetValue(&ckpt, iOut++, p->iBlk, &rc);
+      ckptSetValue(&ckpt, iOut++, (p->iId >> 32) & 0xFFFFFFFF, &rc);
+      ckptSetValue(&ckpt, iOut++, p->iId & 0xFFFFFFFF, &rc);
+    }
+  }
+
+  /* Write the checkpoint header */
+  assert( iId>=0 );
+  assert( pSnap->iCmpId==pDb->compress.iId
+       || pSnap->iCmpId==LSM_COMPRESSION_EMPTY 
+  );
+  ckptSetValue(&ckpt, CKPT_HDR_ID_MSW, (u32)(iId>>32), &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_ID_LSW, (u32)(iId&0xFFFFFFFF), &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_NCKPT, iOut+2, &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_CMPID, pDb->compress.iId, &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_NBLOCK, pSnap->nBlock, &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_BLKSZ, lsmFsBlockSize(pFS), &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_NLEVEL, nLevel, &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_PGSZ, lsmFsPageSize(pFS), &rc);
+  ckptSetValue(&ckpt, CKPT_HDR_NWRITE, pSnap->nWrite, &rc);
+
+  if( bCksum ){
+    ckptAddChecksum(&ckpt, iOut, &rc);
+  }else{
+    ckptSetValue(&ckpt, iOut, 0, &rc);
+    ckptSetValue(&ckpt, iOut+1, 0, &rc);
+  }
+  iOut += 2;
+  assert( iOut<=1024 );
+
+#ifdef LSM_LOG_FREELIST
+  lsmLogMessage(pDb, rc, 
+      "ckptExportSnapshot(): id=%lld freelist: %d", iId, pSnap->freelist.nEntry
+  );
+  for(i=0; i<pSnap->freelist.nEntry; i++){
+  lsmLogMessage(pDb, rc, 
+      "ckptExportSnapshot(): iBlk=%d id=%lld", 
+      pSnap->freelist.aEntry[i].iBlk,
+      pSnap->freelist.aEntry[i].iId
+  );
+  }
+#endif
+
+  *ppCkpt = (void *)ckpt.aCkpt;
+  if( pnCkpt ) *pnCkpt = sizeof(u32)*iOut;
+  return rc;
+}
+
+
+/*
+** Helper function for ckptImport().
+*/
+static void ckptNewSegment(
+  u32 *aIn,
+  int *piIn,
+  Segment *pSegment               /* Populate this structure */
+){
+  assert( pSegment->iFirst==0 && pSegment->iLastPg==0 );
+  assert( pSegment->nSize==0 && pSegment->iRoot==0 );
+  pSegment->iFirst = ckptGobble64(aIn, piIn);
+  pSegment->iLastPg = ckptGobble64(aIn, piIn);
+  pSegment->iRoot = ckptGobble64(aIn, piIn);
+  pSegment->nSize = (int)ckptGobble64(aIn, piIn);
+  assert( pSegment->iFirst );
+}
+
+static int ckptSetupMerge(lsm_db *pDb, u32 *aInt, int *piIn, Level *pLevel){
+  Merge *pMerge;                  /* Allocated Merge object */
+  int nInput;                     /* Number of input segments in merge */
+  int iIn = *piIn;                /* Next value to read from aInt[] */
+  int i;                          /* Iterator variable */
+  int nByte;                      /* Number of bytes to allocate */
+
+  /* Allocate the Merge object. If malloc() fails, return LSM_NOMEM. */
+  nInput = (int)aInt[iIn++];
+  nByte = sizeof(Merge) + sizeof(MergeInput) * nInput;
+  pMerge = (Merge *)lsmMallocZero(pDb->pEnv, nByte);
+  if( !pMerge ) return LSM_NOMEM_BKPT;
+  pLevel->pMerge = pMerge;
+
+  /* Populate the Merge object. */
+  pMerge->aInput = (MergeInput *)&pMerge[1];
+  pMerge->nInput = nInput;
+  pMerge->iOutputOff = -1;
+  pMerge->nSkip = (int)aInt[iIn++];
+  for(i=0; i<nInput; i++){
+    pMerge->aInput[i].iPg = ckptGobble64(aInt, &iIn);
+    pMerge->aInput[i].iCell = (int)aInt[iIn++];
+  }
+  pMerge->splitkey.iPg = ckptGobble64(aInt, &iIn);
+  pMerge->splitkey.iCell = (int)aInt[iIn++];
+  pMerge->iCurrentPtr = ckptGobble64(aInt, &iIn);
+
+  /* Set *piIn and return LSM_OK. */
+  *piIn = iIn;
+  return LSM_OK;
+}
+
+
+static int ckptLoadLevels(
+  lsm_db *pDb,
+  u32 *aIn, 
+  int *piIn, 
+  int nLevel,
+  Level **ppLevel
+){
+  int i;
+  int rc = LSM_OK;
+  Level *pRet = 0;
+  Level **ppNext;
+  int iIn = *piIn;
+
+  ppNext = &pRet;
+  for(i=0; rc==LSM_OK && i<nLevel; i++){
+    int iRight;
+    Level *pLevel;
+
+    /* Allocate space for the Level structure and Level.apRight[] array */
+    pLevel = (Level *)lsmMallocZeroRc(pDb->pEnv, sizeof(Level), &rc);
+    if( rc==LSM_OK ){
+      pLevel->iAge = (u16)(aIn[iIn] & 0x0000FFFF);
+      pLevel->flags = (u16)((aIn[iIn]>>16) & 0x0000FFFF);
+      iIn++;
+      pLevel->nRight = aIn[iIn++];
+      if( pLevel->nRight ){
+        int nByte = sizeof(Segment) * pLevel->nRight;
+        pLevel->aRhs = (Segment *)lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
+      }
+      if( rc==LSM_OK ){
+        *ppNext = pLevel;
+        ppNext = &pLevel->pNext;
+
+        /* Allocate the main segment */
+        ckptNewSegment(aIn, &iIn, &pLevel->lhs);
+
+        /* Allocate each of the right-hand segments, if any */
+        for(iRight=0; iRight<pLevel->nRight; iRight++){
+          ckptNewSegment(aIn, &iIn, &pLevel->aRhs[iRight]);
+        }
+
+        /* Set up the Merge object, if required */
+        if( pLevel->nRight>0 ){
+          rc = ckptSetupMerge(pDb, aIn, &iIn, pLevel);
+        }
+      }
+    }
+  }
+
+  if( rc!=LSM_OK ){
+    /* An OOM must have occurred. Free any level structures allocated and
+    ** return the error to the caller. */
+    lsmSortedFreeLevel(pDb->pEnv, pRet);
+    pRet = 0;
+  }
+
+  *ppLevel = pRet;
+  *piIn = iIn;
+  return rc;
+}
+
+
+int lsmCheckpointLoadLevels(lsm_db *pDb, void *pVal, int nVal){
+  int rc = LSM_OK;
+  if( nVal>0 ){
+    u32 *aIn;
+
+    aIn = lsmMallocRc(pDb->pEnv, nVal, &rc);
+    if( aIn ){
+      Level *pLevel = 0;
+      Level *pParent;
+
+      int nIn;
+      int nLevel;
+      int iIn = 1;
+      memcpy(aIn, pVal, nVal);
+      nIn = nVal / sizeof(u32);
+
+      ckptChangeEndianness(aIn, nIn);
+      nLevel = aIn[0];
+      rc = ckptLoadLevels(pDb, aIn, &iIn, nLevel, &pLevel);
+      lsmFree(pDb->pEnv, aIn);
+      assert( rc==LSM_OK || pLevel==0 );
+      if( rc==LSM_OK ){
+        pParent = lsmDbSnapshotLevel(pDb->pWorker);
+        assert( pParent );
+        while( pParent->pNext ) pParent = pParent->pNext;
+        pParent->pNext = pLevel;
+      }
+    }
+  }
+
+  return rc;
+}
+
+/*
+** Return the data for the LEVELS record.
+**
+** The size of the checkpoint that can be stored in the database header
+** must not exceed 1024 32-bit integers. Normally, it does not. However,
+** if it does, part of the checkpoint must be stored in the LSM. This
+** routine returns that part.
+*/
+int lsmCheckpointLevels(
+  lsm_db *pDb,                    /* Database handle */
+  int nLevel,                     /* Number of levels to write to blob */
+  void **paVal,                   /* OUT: Pointer to LEVELS blob */
+  int *pnVal                      /* OUT: Size of LEVELS blob in bytes */
+){
+  Level *p;                       /* Used to iterate through levels */
+  int nAll= 0;
+  int rc;
+  int i;
+  int iOut;
+  CkptBuffer ckpt;
+  assert( nLevel>0 );
+
+  for(p=lsmDbSnapshotLevel(pDb->pWorker); p; p=p->pNext) nAll++;
+
+  assert( nAll>nLevel );
+  nAll -= nLevel;
+  for(p=lsmDbSnapshotLevel(pDb->pWorker); p && nAll>0; p=p->pNext) nAll--;
+
+  memset(&ckpt, 0, sizeof(CkptBuffer));
+  ckpt.pEnv = pDb->pEnv;
+
+  ckptSetValue(&ckpt, 0, nLevel, &rc);
+  iOut = 1;
+  for(i=0; rc==LSM_OK && i<nLevel; i++){
+    ckptExportLevel(p, &ckpt, &iOut, &rc);
+    p = p->pNext;
+  }
+  assert( rc!=LSM_OK || p==0 );
+
+  if( rc==LSM_OK ){
+    ckptChangeEndianness(ckpt.aCkpt, iOut);
+    *paVal = (void *)ckpt.aCkpt;
+    *pnVal = iOut * sizeof(u32);
+  }else{
+    *pnVal = 0;
+    *paVal = 0;
+  }
+
+  return rc;
+}
+
+/*
+** Read the checkpoint id from meta-page pPg.
+*/
+static i64 ckptLoadId(MetaPage *pPg){
+  i64 ret = 0;
+  if( pPg ){
+    int nData;
+    u8 *aData = lsmFsMetaPageData(pPg, &nData);
+    ret = (((i64)lsmGetU32(&aData[CKPT_HDR_ID_MSW*4])) << 32) + 
+          ((i64)lsmGetU32(&aData[CKPT_HDR_ID_LSW*4]));
+  }
+  return ret;
+}
+
+/*
+** Return true if the buffer passed as an argument contains a valid
+** checkpoint.
+*/
+static int ckptChecksumOk(u32 *aCkpt){
+  u32 nCkpt = aCkpt[CKPT_HDR_NCKPT];
+  u32 cksum1;
+  u32 cksum2;
+
+  if( nCkpt<CKPT_HDR_NCKPT || nCkpt>(LSM_META_RW_PAGE_SIZE)/sizeof(u32) ){
+    return 0;
+  }
+  ckptChecksum(aCkpt, nCkpt, &cksum1, &cksum2);
+  return (cksum1==aCkpt[nCkpt-2] && cksum2==aCkpt[nCkpt-1]);
+}
+
+/*
+** Attempt to load a checkpoint from meta page iMeta.
+**
+** This function is a no-op if *pRc is set to any value other than LSM_OK
+** when it is called. If an error occurs, *pRc is set to an LSM error code
+** before returning.
+**
+** If no error occurs and the checkpoint is successfully loaded, copy it to
+** ShmHeader.aSnap1[] and ShmHeader.aSnap2[], and set ShmHeader.iMetaPage 
+** to indicate its origin. In this case return 1. Or, if the checkpoint 
+** cannot be loaded (because the checksum does not compute), return 0.
+*/
+static int ckptTryLoad(lsm_db *pDb, MetaPage *pPg, u32 iMeta, int *pRc){
+  int bLoaded = 0;                /* Return value */
+  if( *pRc==LSM_OK ){
+    int rc = LSM_OK;              /* Error code */
+    u32 *aCkpt = 0;               /* Pointer to buffer containing checkpoint */
+    u32 nCkpt;                    /* Number of elements in aCkpt[] */
+    int nData;                    /* Bytes of data in aData[] */
+    u8 *aData;                    /* Meta page data */
+   
+    aData = lsmFsMetaPageData(pPg, &nData);
+    nCkpt = (u32)lsmGetU32(&aData[CKPT_HDR_NCKPT*sizeof(u32)]);
+    if( nCkpt<=nData/sizeof(u32) && nCkpt>CKPT_HDR_NCKPT ){
+      aCkpt = (u32 *)lsmMallocRc(pDb->pEnv, nCkpt*sizeof(u32), &rc);
+    }
+    if( aCkpt ){
+      memcpy(aCkpt, aData, nCkpt*sizeof(u32));
+      ckptChangeEndianness(aCkpt, nCkpt);
+      if( ckptChecksumOk(aCkpt) ){
+        ShmHeader *pShm = pDb->pShmhdr;
+        memcpy(pShm->aSnap1, aCkpt, nCkpt*sizeof(u32));
+        memcpy(pShm->aSnap2, aCkpt, nCkpt*sizeof(u32));
+        memcpy(pDb->aSnapshot, aCkpt, nCkpt*sizeof(u32));
+        pShm->iMetaPage = iMeta;
+        bLoaded = 1;
+      }
+    }
+
+    lsmFree(pDb->pEnv, aCkpt);
+    *pRc = rc;
+  }
+  return bLoaded;
+}
+
+/*
+** Initialize the shared-memory header with an empty snapshot. This function
+** is called when no valid snapshot can be found in the database header.
+*/
+static void ckptLoadEmpty(lsm_db *pDb){
+  u32 aCkpt[] = {
+    0,                       /* CKPT_HDR_ID_MSW */
+    10,                      /* CKPT_HDR_ID_LSW */
+    0,                       /* CKPT_HDR_NCKPT */
+    LSM_COMPRESSION_EMPTY,   /* CKPT_HDR_CMPID */
+    0,                       /* CKPT_HDR_NBLOCK */
+    0,                       /* CKPT_HDR_BLKSZ */
+    0,                       /* CKPT_HDR_NLEVEL */
+    0,                       /* CKPT_HDR_PGSZ */
+    0,                       /* CKPT_HDR_NWRITE */
+    0, 0, 1234, 5678,        /* The log pointer and initial checksum */
+    0,0,0,0, 0,0,0,0,        /* The append list */
+    0,                       /* The redirected block list */
+    0,                       /* The free block list */
+    0, 0                     /* Space for checksum values */
+  };
+  u32 nCkpt = array_size(aCkpt);
+  ShmHeader *pShm = pDb->pShmhdr;
+
+  aCkpt[CKPT_HDR_NCKPT] = nCkpt;
+  aCkpt[CKPT_HDR_BLKSZ] = pDb->nDfltBlksz;
+  aCkpt[CKPT_HDR_PGSZ] = pDb->nDfltPgsz;
+  ckptChecksum(aCkpt, array_size(aCkpt), &aCkpt[nCkpt-2], &aCkpt[nCkpt-1]);
+
+  memcpy(pShm->aSnap1, aCkpt, nCkpt*sizeof(u32));
+  memcpy(pShm->aSnap2, aCkpt, nCkpt*sizeof(u32));
+  memcpy(pDb->aSnapshot, aCkpt, nCkpt*sizeof(u32));
+}
+
+/*
+** This function is called as part of database recovery to initialize the
+** ShmHeader.aSnap1[] and ShmHeader.aSnap2[] snapshots.
+*/
+int lsmCheckpointRecover(lsm_db *pDb){
+  int rc = LSM_OK;                /* Return Code */
+  i64 iId1;                       /* Id of checkpoint on meta-page 1 */
+  i64 iId2;                       /* Id of checkpoint on meta-page 2 */
+  int bLoaded = 0;                /* True once checkpoint has been loaded */
+  int cmp;                        /* True if (iId2>iId1) */
+  MetaPage *apPg[2] = {0, 0};     /* Meta-pages 1 and 2 */
+
+  rc = lsmFsMetaPageGet(pDb->pFS, 0, 1, &apPg[0]);
+  if( rc==LSM_OK ) rc = lsmFsMetaPageGet(pDb->pFS, 0, 2, &apPg[1]);
+
+  iId1 = ckptLoadId(apPg[0]);
+  iId2 = ckptLoadId(apPg[1]);
+  cmp = (iId2 > iId1);
+  bLoaded = ckptTryLoad(pDb, apPg[cmp?1:0], (cmp?2:1), &rc);
+  if( bLoaded==0 ){
+    bLoaded = ckptTryLoad(pDb, apPg[cmp?0:1], (cmp?1:2), &rc);
+  }
+
+  /* The database does not contain a valid checkpoint. Initialize the shared
+  ** memory header with an empty checkpoint.  */
+  if( bLoaded==0 ){
+    ckptLoadEmpty(pDb);
+  }
+
+  lsmFsMetaPageRelease(apPg[0]);
+  lsmFsMetaPageRelease(apPg[1]);
+
+  return rc;
+}
+
+/* 
+** Store the snapshot in pDb->aSnapshot[] in meta-page iMeta.
+*/
+int lsmCheckpointStore(lsm_db *pDb, int iMeta){
+  MetaPage *pPg = 0;
+  int rc;
+
+  assert( iMeta==1 || iMeta==2 );
+  rc = lsmFsMetaPageGet(pDb->pFS, 1, iMeta, &pPg);
+  if( rc==LSM_OK ){
+    u8 *aData;
+    int nData;
+    int nCkpt;
+
+    nCkpt = (int)pDb->aSnapshot[CKPT_HDR_NCKPT];
+    aData = lsmFsMetaPageData(pPg, &nData);
+    memcpy(aData, pDb->aSnapshot, nCkpt*sizeof(u32));
+    ckptChangeEndianness((u32 *)aData, nCkpt);
+    rc = lsmFsMetaPageRelease(pPg);
+  }
+      
+  return rc;
+}
+
+/*
+** Copy the current client snapshot from shared-memory to pDb->aSnapshot[].
+*/
+int lsmCheckpointLoad(lsm_db *pDb, int *piRead){
+  int nRem = LSM_ATTEMPTS_BEFORE_PROTOCOL;
+  ShmHeader *pShm = pDb->pShmhdr;
+  while( (nRem--)>0 ){
+    int nInt;
+
+    nInt = pShm->aSnap1[CKPT_HDR_NCKPT];
+    if( nInt<=(LSM_META_RW_PAGE_SIZE / sizeof(u32)) ){
+      memcpy(pDb->aSnapshot, pShm->aSnap1, nInt*sizeof(u32));
+      if( ckptChecksumOk(pDb->aSnapshot) ){
+        if( piRead ) *piRead = 1;
+        return LSM_OK;
+      }
+    }
+
+    nInt = pShm->aSnap2[CKPT_HDR_NCKPT];
+    if( nInt<=(LSM_META_RW_PAGE_SIZE / sizeof(u32)) ){
+      memcpy(pDb->aSnapshot, pShm->aSnap2, nInt*sizeof(u32));
+      if( ckptChecksumOk(pDb->aSnapshot) ){
+        if( piRead ) *piRead = 2;
+        return LSM_OK;
+      }
+    }
+
+    lsmShmBarrier(pDb);
+  }
+  return LSM_PROTOCOL_BKPT;
+}
+
+int lsmInfoCompressionId(lsm_db *db, u32 *piCmpId){
+  int rc;
+
+  assert( db->pClient==0 && db->pWorker==0 );
+  rc = lsmCheckpointLoad(db, 0);
+  if( rc==LSM_OK ){
+    *piCmpId = db->aSnapshot[CKPT_HDR_CMPID];
+  }
+
+  return rc;
+}
+
+int lsmCheckpointLoadOk(lsm_db *pDb, int iSnap){
+  u32 *aShm;
+  assert( iSnap==1 || iSnap==2 );
+  aShm = (iSnap==1) ? pDb->pShmhdr->aSnap1 : pDb->pShmhdr->aSnap2;
+  return (lsmCheckpointId(pDb->aSnapshot, 0)==lsmCheckpointId(aShm, 0) );
+}
+
+int lsmCheckpointClientCacheOk(lsm_db *pDb){
+  return ( pDb->pClient 
+        && pDb->pClient->iId==lsmCheckpointId(pDb->aSnapshot, 0)
+        && pDb->pClient->iId==lsmCheckpointId(pDb->pShmhdr->aSnap1, 0)
+        && pDb->pClient->iId==lsmCheckpointId(pDb->pShmhdr->aSnap2, 0)
+  );
+}
+
+int lsmCheckpointLoadWorker(lsm_db *pDb){
+  int rc;
+  ShmHeader *pShm = pDb->pShmhdr;
+  int nInt1;
+  int nInt2;
+
+  /* Must be holding the WORKER lock to do this. Or DMS2. */
+  assert( 
+      lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL) 
+   || lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) 
+  );
+
+  /* Check that the two snapshots match. If not, repair them. */
+  nInt1 = pShm->aSnap1[CKPT_HDR_NCKPT];
+  nInt2 = pShm->aSnap2[CKPT_HDR_NCKPT];
+  if( nInt1!=nInt2 || memcmp(pShm->aSnap1, pShm->aSnap2, nInt2*sizeof(u32)) ){
+    if( ckptChecksumOk(pShm->aSnap1) ){
+      memcpy(pShm->aSnap2, pShm->aSnap1, sizeof(u32)*nInt1);
+    }else if( ckptChecksumOk(pShm->aSnap2) ){
+      memcpy(pShm->aSnap1, pShm->aSnap2, sizeof(u32)*nInt2);
+    }else{
+      return LSM_PROTOCOL_BKPT;
+    }
+  }
+
+  rc = lsmCheckpointDeserialize(pDb, 1, pShm->aSnap1, &pDb->pWorker);
+  if( pDb->pWorker ) pDb->pWorker->pDatabase = pDb->pDatabase;
+
+  if( rc==LSM_OK ){
+    rc = lsmCheckCompressionId(pDb, pDb->pWorker->iCmpId);
+  }
+
+#if 0
+  assert( rc!=LSM_OK || lsmFsIntegrityCheck(pDb) );
+#endif
+  return rc;
+}
+
+int lsmCheckpointDeserialize(
+  lsm_db *pDb, 
+  int bInclFreelist,              /* If true, deserialize free-list */
+  u32 *aCkpt, 
+  Snapshot **ppSnap
+){
+  int rc = LSM_OK;
+  Snapshot *pNew;
+
+  pNew = (Snapshot *)lsmMallocZeroRc(pDb->pEnv, sizeof(Snapshot), &rc);
+  if( rc==LSM_OK ){
+    Level *pLvl;
+    int nFree;
+    int i;
+    int nLevel = (int)aCkpt[CKPT_HDR_NLEVEL];
+    int iIn = CKPT_HDR_SIZE + CKPT_APPENDLIST_SIZE + CKPT_LOGPTR_SIZE;
+
+    pNew->iId = lsmCheckpointId(aCkpt, 0);
+    pNew->nBlock = aCkpt[CKPT_HDR_NBLOCK];
+    pNew->nWrite = aCkpt[CKPT_HDR_NWRITE];
+    rc = ckptLoadLevels(pDb, aCkpt, &iIn, nLevel, &pNew->pLevel);
+    pNew->iLogOff = lsmCheckpointLogOffset(aCkpt);
+    pNew->iCmpId = aCkpt[CKPT_HDR_CMPID];
+
+    /* Make a copy of the append-list */
+    for(i=0; i<LSM_APPLIST_SZ; i++){
+      u32 *a = &aCkpt[CKPT_HDR_SIZE + CKPT_LOGPTR_SIZE + i*2];
+      pNew->aiAppend[i] = ckptRead64(a);
+    }
+
+    /* Read the block-redirect list */
+    pNew->redirect.n = aCkpt[iIn++];
+    if( pNew->redirect.n ){
+      pNew->redirect.a = lsmMallocZeroRc(pDb->pEnv, 
+          (sizeof(struct RedirectEntry) * LSM_MAX_BLOCK_REDIRECTS), &rc
+      );
+      if( rc==LSM_OK ){
+        for(i=0; i<pNew->redirect.n; i++){
+          pNew->redirect.a[i].iFrom = aCkpt[iIn++];
+          pNew->redirect.a[i].iTo = aCkpt[iIn++];
+        }
+      }
+      for(pLvl=pNew->pLevel; pLvl->pNext; pLvl=pLvl->pNext);
+      if( pLvl->nRight ){
+        pLvl->aRhs[pLvl->nRight-1].pRedirect = &pNew->redirect;
+      }else{
+        pLvl->lhs.pRedirect = &pNew->redirect;
+      }
+    }
+
+    /* Copy the free-list */
+    if( rc==LSM_OK && bInclFreelist ){
+      nFree = aCkpt[iIn++];
+      if( nFree ){
+        pNew->freelist.aEntry = (FreelistEntry *)lsmMallocZeroRc(
+            pDb->pEnv, sizeof(FreelistEntry)*nFree, &rc
+        );
+        if( rc==LSM_OK ){
+          int j;
+          for(j=0; j<nFree; j++){
+            FreelistEntry *p = &pNew->freelist.aEntry[j];
+            p->iBlk = aCkpt[iIn++];
+            p->iId = ((i64)(aCkpt[iIn])<<32) + aCkpt[iIn+1];
+            iIn += 2;
+          }
+          pNew->freelist.nEntry = pNew->freelist.nAlloc = nFree;
+        }
+      }
+    }
+  }
+
+  if( rc!=LSM_OK ){
+    lsmFreeSnapshot(pDb->pEnv, pNew);
+    pNew = 0;
+  }
+
+  *ppSnap = pNew;
+  return rc;
+}
+
+/*
+** Connection pDb must be the worker connection in order to call this
+** function. It returns true if the database already contains the maximum
+** number of levels or false otherwise.
+**
+** This is used when flushing the in-memory tree to disk. If the database
+** is already full, then the caller should invoke lsm_work() or similar
+** until it is not full before creating a new level by flushing the in-memory
+** tree to disk. Limiting the number of levels in the database ensures that
+** the records describing them always fit within the checkpoint blob.
+*/
+int lsmDatabaseFull(lsm_db *pDb){
+  Level *p;
+  int nRhs = 0;
+
+  assert( lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL) );
+  assert( pDb->pWorker );
+
+  for(p=pDb->pWorker->pLevel; p; p=p->pNext){
+    nRhs += (p->nRight ? p->nRight : 1);
+  }
+
+  return (nRhs >= LSM_MAX_RHS_SEGMENTS);
+}
+
+/*
+** The connection passed as the only argument is currently the worker
+** connection. Some work has been performed on the database by the connection,
+** but no new snapshot has been written into shared memory.
+**
+** This function updates the shared-memory worker and client snapshots with
+** the new snapshot produced by the work performed by pDb.
+**
+** If successful, LSM_OK is returned. Otherwise, if an error occurs, an LSM
+** error code is returned.
+*/
+int lsmCheckpointSaveWorker(lsm_db *pDb, int bFlush){
+  Snapshot *pSnap = pDb->pWorker;
+  ShmHeader *pShm = pDb->pShmhdr;
+  void *p = 0;
+  int n = 0;
+  int rc;
+
+  pSnap->iId++;
+  rc = ckptExportSnapshot(pDb, bFlush, pSnap->iId, 1, &p, &n);
+  if( rc!=LSM_OK ) return rc;
+  assert( ckptChecksumOk((u32 *)p) );
+
+  assert( n<=LSM_META_RW_PAGE_SIZE );
+  memcpy(pShm->aSnap2, p, n);
+  lsmShmBarrier(pDb);
+  memcpy(pShm->aSnap1, p, n);
+  lsmFree(pDb->pEnv, p);
+
+  /* assert( lsmFsIntegrityCheck(pDb) ); */
+  return LSM_OK;
+}
+
+/*
+** This function is used to determine the snapshot-id of the most recently
+** checkpointed snapshot. Variable ShmHeader.iMetaPage indicates which of
+** the two meta-pages said snapshot resides on (if any). 
+**
+** If successful, this function loads the snapshot from the meta-page, 
+** verifies its checksum and sets *piId to the snapshot-id before returning
+** LSM_OK. Or, if the checksum attempt fails, *piId is set to zero and
+** LSM_OK returned. If an error occurs, an LSM error code is returned and
+** the final value of *piId is undefined.
+*/
+int lsmCheckpointSynced(lsm_db *pDb, i64 *piId, i64 *piLog, u32 *pnWrite){
+  int rc = LSM_OK;
+  MetaPage *pPg;
+  u32 iMeta;
+
+  iMeta = pDb->pShmhdr->iMetaPage;
+  if( iMeta==1 || iMeta==2 ){
+    rc = lsmFsMetaPageGet(pDb->pFS, 0, iMeta, &pPg);
+    if( rc==LSM_OK ){
+      int nCkpt;
+      int nData;
+      u8 *aData; 
+
+      aData = lsmFsMetaPageData(pPg, &nData);
+      assert( nData==LSM_META_RW_PAGE_SIZE );
+      nCkpt = lsmGetU32(&aData[CKPT_HDR_NCKPT*sizeof(u32)]);
+      if( nCkpt<(LSM_META_RW_PAGE_SIZE/sizeof(u32)) ){
+        u32 *aCopy = lsmMallocRc(pDb->pEnv, sizeof(u32) * nCkpt, &rc);
+        if( aCopy ){
+          memcpy(aCopy, aData, nCkpt*sizeof(u32));
+          ckptChangeEndianness(aCopy, nCkpt);
+          if( ckptChecksumOk(aCopy) ){
+            if( piId ) *piId = lsmCheckpointId(aCopy, 0);
+            if( piLog ) *piLog = (lsmCheckpointLogOffset(aCopy) >> 1);
+            if( pnWrite ) *pnWrite = aCopy[CKPT_HDR_NWRITE];
+          }
+          lsmFree(pDb->pEnv, aCopy);
+        }
+      }
+      lsmFsMetaPageRelease(pPg);
+    }
+  }
+
+  if( (iMeta!=1 && iMeta!=2) || rc!=LSM_OK || pDb->pShmhdr->iMetaPage!=iMeta ){
+    if( piId ) *piId = 0;
+    if( piLog ) *piLog = 0;
+    if( pnWrite ) *pnWrite = 0;
+  }
+  return rc;
+}
+
+/*
+** Return the checkpoint-id of the checkpoint array passed as the first
+** argument to this function. If the second argument is true, then assume
+** that the checkpoint is made up of 32-bit big-endian integers. If it
+** is false, assume that the integers are in machine byte order.
+*/
+i64 lsmCheckpointId(u32 *aCkpt, int bDisk){
+  i64 iId;
+  if( bDisk ){
+    u8 *aData = (u8 *)aCkpt;
+    iId = (((i64)lsmGetU32(&aData[CKPT_HDR_ID_MSW*4])) << 32);
+    iId += ((i64)lsmGetU32(&aData[CKPT_HDR_ID_LSW*4]));
+  }else{
+    iId = ((i64)aCkpt[CKPT_HDR_ID_MSW] << 32) + (i64)aCkpt[CKPT_HDR_ID_LSW];
+  }
+  return iId;
+}
+
+u32 lsmCheckpointNBlock(u32 *aCkpt){
+  return aCkpt[CKPT_HDR_NBLOCK];
+}
+
+u32 lsmCheckpointNWrite(u32 *aCkpt, int bDisk){
+  if( bDisk ){
+    return lsmGetU32((u8 *)&aCkpt[CKPT_HDR_NWRITE]);
+  }else{
+    return aCkpt[CKPT_HDR_NWRITE];
+  }
+}
+
+i64 lsmCheckpointLogOffset(u32 *aCkpt){
+  return ((i64)aCkpt[CKPT_HDR_LO_MSW] << 32) + (i64)aCkpt[CKPT_HDR_LO_LSW];
+}
+
+int lsmCheckpointPgsz(u32 *aCkpt){ return (int)aCkpt[CKPT_HDR_PGSZ]; }
+
+int lsmCheckpointBlksz(u32 *aCkpt){ return (int)aCkpt[CKPT_HDR_BLKSZ]; }
+
+void lsmCheckpointLogoffset(
+  u32 *aCkpt,
+  DbLog *pLog
+){ 
+  pLog->aRegion[2].iStart = (lsmCheckpointLogOffset(aCkpt) >> 1);
+
+  pLog->cksum0 = aCkpt[CKPT_HDR_LO_CKSUM1];
+  pLog->cksum1 = aCkpt[CKPT_HDR_LO_CKSUM2];
+  pLog->iSnapshotId = lsmCheckpointId(aCkpt, 0);
+}
+
+void lsmCheckpointZeroLogoffset(lsm_db *pDb){
+  u32 nCkpt;
+
+  nCkpt = pDb->aSnapshot[CKPT_HDR_NCKPT];
+  assert( nCkpt>CKPT_HDR_NCKPT );
+  assert( nCkpt==pDb->pShmhdr->aSnap1[CKPT_HDR_NCKPT] );
+  assert( 0==memcmp(pDb->aSnapshot, pDb->pShmhdr->aSnap1, nCkpt*sizeof(u32)) );
+  assert( 0==memcmp(pDb->aSnapshot, pDb->pShmhdr->aSnap2, nCkpt*sizeof(u32)) );
+
+  pDb->aSnapshot[CKPT_HDR_LO_MSW] = 0;
+  pDb->aSnapshot[CKPT_HDR_LO_LSW] = 0;
+  ckptChecksum(pDb->aSnapshot, nCkpt, 
+      &pDb->aSnapshot[nCkpt-2], &pDb->aSnapshot[nCkpt-1]
+  );
+
+  memcpy(pDb->pShmhdr->aSnap1, pDb->aSnapshot, nCkpt*sizeof(u32));
+  memcpy(pDb->pShmhdr->aSnap2, pDb->aSnapshot, nCkpt*sizeof(u32));
+}
+
+/*
+** Set the output variable to the number of KB of data written into the
+** database file since the most recent checkpoint.
+*/
+int lsmCheckpointSize(lsm_db *db, int *pnKB){
+  int rc = LSM_OK;
+  u32 nSynced;
+
+  /* Set nSynced to the number of pages that had been written when the 
+  ** database was last checkpointed. */
+  rc = lsmCheckpointSynced(db, 0, 0, &nSynced);
+
+  if( rc==LSM_OK ){
+    u32 nPgsz = db->pShmhdr->aSnap1[CKPT_HDR_PGSZ];
+    u32 nWrite = db->pShmhdr->aSnap1[CKPT_HDR_NWRITE];
+    *pnKB = (int)(( ((i64)(nWrite - nSynced) * nPgsz) + 1023) / 1024);
+  }
+
+  return rc;
+}
diff --git a/ext/lsm1/lsm_file.c b/ext/lsm1/lsm_file.c
new file mode 100644
index 0000000..1dcdd05
--- /dev/null
+++ b/ext/lsm1/lsm_file.c
@@ -0,0 +1,3312 @@
+/*
+** 2011-08-26
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+** 
+** NORMAL DATABASE FILE FORMAT
+**
+** The following database file format concepts are used by the code in
+** this file to read and write the database file.
+**
+** Pages:
+**
+**   A database file is divided into pages. The first 8KB of the file consists
+**   of two 4KB meta-pages. The meta-page size is not configurable. The 
+**   remainder of the file is made up of database pages. The default database
+**   page size is 4KB. Database pages are aligned to page-size boundaries,
+**   so if the database page size is larger than 8KB there is a gap between
+**   the end of the meta pages and the start of the database pages.
+**
+**   Database pages are numbered based on their position in the file. Page N
+**   begins at byte offset ((N-1)*pgsz). This means that page 1 does not 
+**   exist - since it would always overlap with the meta pages. If the 
+**   page-size is (say) 512 bytes, then the first usable page in the database
+**   is page 33.
+**
+**   It is assumed that the first two meta pages and the data that follows
+**   them are located on different disk sectors. So that if a power failure 
+**   while writing to a meta page there is no risk of damage to the other
+**   meta page or any other part of the database file. TODO: This may need
+**   to be revisited.
+**
+** Blocks:
+**
+**   The database file is also divided into blocks. The default block size is
+**   1MB. When writing to the database file, an attempt is made to write data
+**   in contiguous block-sized chunks.
+**
+**   The first and last page on each block are special in that they are 4 
+**   bytes smaller than all other pages. This is because the last four bytes 
+**   of space on the first and last pages of each block are reserved for
+**   pointers to other blocks (i.e. a 32-bit block number).
+**
+** Runs:
+**
+**   A run is a sequence of pages that the upper layer uses to store a 
+**   sorted array of database keys (and accompanying data - values, FC 
+**   pointers and so on). Given a page within a run, it is possible to
+**   navigate to the next page in the run as follows:
+**
+**     a) if the current page is not the last in a block, the next page 
+**        in the run is located immediately after the current page, OR
+**
+**     b) if the current page is the last page in a block, the next page 
+**        in the run is the first page on the block identified by the
+**        block pointer stored in the last 4 bytes of the current block.
+**
+**   It is possible to navigate to the previous page in a similar fashion,
+**   using the block pointer embedded in the last 4 bytes of the first page
+**   of each block as required.
+**
+**   The upper layer is responsible for identifying by page number the 
+**   first and last page of any run that it needs to navigate - there are
+**   no "end-of-run" markers stored or identified by this layer. This is
+**   necessary as clients reading different database snapshots may access 
+**   different subsets of a run.
+**
+** THE LOG FILE 
+**
+** This file opens and closes the log file. But it does not contain any
+** logic related to the log file format. Instead, it exports the following
+** functions that are used by the code in lsm_log.c to read and write the
+** log file:
+**
+**     lsmFsOpenLog
+**     lsmFsWriteLog
+**     lsmFsSyncLog
+**     lsmFsReadLog
+**     lsmFsTruncateLog
+**     lsmFsCloseAndDeleteLog
+**
+** COMPRESSED DATABASE FILE FORMAT
+**
+** The compressed database file format is very similar to the normal format.
+** The file still begins with two 4KB meta-pages (which are never compressed).
+** It is still divided into blocks.
+**
+** The first and last four bytes of each block are reserved for 32-bit 
+** pointer values. Similar to the way four bytes are carved from the end of 
+** the first and last page of each block in uncompressed databases. From
+** the point of view of the upper layer, all pages are the same size - this
+** is different from the uncompressed format where the first and last pages
+** on each block are 4 bytes smaller than the others.
+**
+** Pages are stored in variable length compressed form, as follows:
+**
+**     * 3-byte size field containing the size of the compressed page image
+**       in bytes. The most significant bit of each byte of the size field
+**       is always set. The remaining 7 bits are used to store a 21-bit
+**       integer value (in big-endian order - the first byte in the field
+**       contains the most significant 7 bits). Since the maximum allowed 
+**       size of a compressed page image is (2^17 - 1) bytes, there are
+**       actually 4 unused bits in the size field.
+**
+**       In other words, if the size of the compressed page image is nSz,
+**       the header can be serialized as follows:
+**
+**         u8 aHdr[3]
+**         aHdr[0] = 0x80 | (u8)(nSz >> 14);
+**         aHdr[1] = 0x80 | (u8)(nSz >>  7);
+**         aHdr[2] = 0x80 | (u8)(nSz >>  0);
+**
+**     * Compressed page image.
+**
+**     * A second copy of the 3-byte record header.
+**
+** A page number is a byte offset into the database file. So the smallest
+** possible page number is 8192 (immediately after the two meta-pages).
+** The first and root page of a segment are identified by a page number
+** corresponding to the byte offset of the first byte in the corresponding
+** page record. The last page of a segment is identified by the byte offset
+** of the last byte in its record.
+**
+** Unlike uncompressed pages, compressed page records may span blocks.
+**
+** Sometimes, in order to avoid touching sectors that contain synced data
+** when writing, it is necessary to insert unused space between compressed
+** page records. This can be done as follows:
+**
+**     * For less than 6 bytes of empty space, the first and last byte
+**       of the free space contain the total number of free bytes. For
+**       example:
+**
+**         Block of 4 free bytes: 0x04 0x?? 0x?? 0x04
+**         Block of 2 free bytes: 0x02 0x02
+**         A single free byte:    0x01
+**
+**     * For 6 or more bytes of empty space, a record similar to a 
+**       compressed page record is added to the segment. A padding record
+**       is distinguished from a compressed page record by the most 
+**       significant bit of the second byte of the size field, which is
+**       cleared instead of set. 
+*/
+#include "lsmInt.h"
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+
+/*
+** File-system object. Each database connection allocates a single instance
+** of the following structure. It is used for all access to the database and
+** log files.
+**
+** The database file may be accessed via two methods - using mmap() or using
+** read() and write() calls. In the general case both methods are used - a
+** prefix of the file is mapped into memory and the remainder accessed using
+** read() and write(). This is helpful when accessing very large files (or
+** files that may grow very large during the lifetime of a database
+** connection) on systems with 32-bit address spaces. However, it also requires
+** that this object manage two distinct types of Page objects simultaneously -
+** those that carry pointers to the mapped file and those that carry arrays
+** populated by read() calls.
+**
+** pFree:
+**   The head of a singly-linked list that containing currently unused Page 
+**   structures suitable for use as mmap-page handles. Connected by the
+**   Page.pFreeNext pointers.
+**
+** pMapped:
+**   The head of a singly-linked list that contains all pages that currently
+**   carry pointers to the mapped region. This is used if the region is
+**   every remapped - the pointers carried by existing pages can be adjusted
+**   to account for the remapping. Connected by the Page.pMappedNext pointers.
+**
+** pWaiting:
+**   When the upper layer wishes to append a new b-tree page to a segment,
+**   it allocates a Page object that carries a malloc'd block of memory -
+**   regardless of the mmap-related configuration. The page is not assigned
+**   a page number at first. When the upper layer has finished constructing
+**   the page contents, it calls lsmFsPagePersist() to assign a page number
+**   to it. At this point it is likely that N pages have been written to the
+**   segment, the (N+1)th page is still outstanding and the b-tree page is
+**   assigned page number (N+2). To avoid writing page (N+2) before page 
+**   (N+1), the recently completed b-tree page is held in the singly linked
+**   list headed by pWaiting until page (N+1) has been written. 
+**
+**   Function lsmFsFlushWaiting() is responsible for eventually writing 
+**   waiting pages to disk.
+**
+** apHash/nHash:
+**   Hash table used to store all Page objects that carry malloc'd arrays,
+**   except those b-tree pages that have not yet been assigned page numbers.
+**   Once they have been assigned page numbers - they are added to this
+**   hash table.
+**
+**   Hash table overflow chains are connected using the Page.pHashNext
+**   pointers.
+**
+** pLruFirst, pLruLast:
+**   The first and last entries in a doubly-linked list of pages. This
+**   list contains all pages with malloc'd data that are present in the
+**   hash table and have a ref-count of zero.
+*/
+struct FileSystem {
+  lsm_db *pDb;                    /* Database handle that owns this object */
+  lsm_env *pEnv;                  /* Environment pointer */
+  char *zDb;                      /* Database file name */
+  char *zLog;                     /* Database file name */
+  int nMetasize;                  /* Size of meta pages in bytes */
+  int nMetaRwSize;                /* Read/written size of meta pages in bytes */
+  int nPagesize;                  /* Database page-size in bytes */
+  int nBlocksize;                 /* Database block-size in bytes */
+
+  /* r/w file descriptors for both files. */
+  LsmFile *pLsmFile;              /* Used after lsm_close() to link into list */
+  lsm_file *fdDb;                 /* Database file */
+  lsm_file *fdLog;                /* Log file */
+  int szSector;                   /* Database file sector size */
+
+  /* If this is a compressed database, a pointer to the compression methods.
+  ** For an uncompressed database, a NULL pointer.  */
+  lsm_compress *pCompress;
+  u8 *aIBuffer;                   /* Buffer to compress to */
+  u8 *aOBuffer;                   /* Buffer to uncompress from */
+  int nBuffer;                    /* Allocated size of above buffers in bytes */
+
+  /* mmap() page related things */
+  i64 nMapLimit;                  /* Maximum bytes of file to map */
+  void *pMap;                     /* Current mapping of database file */
+  i64 nMap;                       /* Bytes mapped at pMap */
+  Page *pFree;                    /* Unused Page structures */
+  Page *pMapped;                  /* List of Page structs that point to pMap */
+
+  /* Page cache parameters for non-mmap() pages */
+  int nCacheMax;                  /* Configured cache size (in pages) */
+  int nCacheAlloc;                /* Current cache size (in pages) */
+  Page *pLruFirst;                /* Head of the LRU list */
+  Page *pLruLast;                 /* Tail of the LRU list */
+  int nHash;                      /* Number of hash slots in hash table */
+  Page **apHash;                  /* nHash Hash slots */
+  Page *pWaiting;                 /* b-tree pages waiting to be written */
+
+  /* Statistics */
+  int nOut;                       /* Number of outstanding pages */
+  int nWrite;                     /* Total number of pages written */
+  int nRead;                      /* Total number of pages read */
+};
+
+/*
+** Database page handle.
+**
+** pSeg:
+**   When lsmFsSortedAppend() is called on a compressed database, the new
+**   page is not assigned a page number or location in the database file
+**   immediately. Instead, these are assigned by the lsmFsPagePersist() call
+**   right before it writes the compressed page image to disk.
+**
+**   The lsmFsSortedAppend() function sets the pSeg pointer to point to the
+**   segment that the new page will be a part of. It is unset by
+**   lsmFsPagePersist() after the page is written to disk.
+*/
+struct Page {
+  u8 *aData;                      /* Buffer containing page data */
+  int nData;                      /* Bytes of usable data at aData[] */
+  LsmPgno iPg;                    /* Page number */
+  int nRef;                       /* Number of outstanding references */
+  int flags;                      /* Combination of PAGE_XXX flags */
+  Page *pHashNext;                /* Next page in hash table slot */
+  Page *pLruNext;                 /* Next page in LRU list */
+  Page *pLruPrev;                 /* Previous page in LRU list */
+  FileSystem *pFS;                /* File system that owns this page */
+
+  /* Only used in compressed database mode: */
+  int nCompress;                  /* Compressed size (or 0 for uncomp. db) */
+  int nCompressPrev;              /* Compressed size of prev page */
+  Segment *pSeg;                  /* Segment this page will be written to */
+
+  /* Pointers for singly linked lists */
+  Page *pWaitingNext;             /* Next page in FileSystem.pWaiting list */
+  Page *pFreeNext;                /* Next page in FileSystem.pFree list */
+  Page *pMappedNext;              /* Next page in FileSystem.pMapped list */
+};
+
+/*
+** Meta-data page handle. There are two meta-data pages at the start of
+** the database file, each FileSystem.nMetasize bytes in size.
+*/
+struct MetaPage {
+  int iPg;                        /* Either 1 or 2 */
+  int bWrite;                     /* Write back to db file on release */
+  u8 *aData;                      /* Pointer to buffer */
+  FileSystem *pFS;                /* FileSystem that owns this page */
+};
+
+/* 
+** Values for LsmPage.flags 
+*/
+#define PAGE_DIRTY   0x00000001   /* Set if page is dirty */
+#define PAGE_FREE    0x00000002   /* Set if Page.aData requires lsmFree() */
+#define PAGE_HASPREV 0x00000004   /* Set if page is first on uncomp. block */
+
+/*
+** Number of pgsz byte pages omitted from the start of block 1. The start
+** of block 1 contains two 4096 byte meta pages (8192 bytes in total).
+*/
+#define BLOCK1_HDR_SIZE(pgsz)  LSM_MAX(1, 8192/(pgsz))
+
+/*
+** If NDEBUG is not defined, set a breakpoint in function lsmIoerrBkpt()
+** to catch IO errors (any error returned by a VFS method). 
+*/
+#ifndef NDEBUG
+static void lsmIoerrBkpt(void){
+  static int nErr = 0;
+  nErr++;
+}
+static int IOERR_WRAPPER(int rc){
+  if( rc!=LSM_OK ) lsmIoerrBkpt();
+  return rc;
+}
+#else
+# define IOERR_WRAPPER(rc) (rc)
+#endif
+
+#ifdef NDEBUG
+# define assert_lists_are_ok(x)
+#else
+static Page *fsPageFindInHash(FileSystem *pFS, LsmPgno iPg, int *piHash);
+
+static void assert_lists_are_ok(FileSystem *pFS){
+#if 0
+  Page *p;
+
+  assert( pFS->nMapLimit>=0 );
+
+  /* Check that all pages in the LRU list have nRef==0, pointers to buffers
+  ** in heap memory, and corresponding entries in the hash table.  */
+  for(p=pFS->pLruFirst; p; p=p->pLruNext){
+    assert( p==pFS->pLruFirst || p->pLruPrev!=0 );
+    assert( p==pFS->pLruLast || p->pLruNext!=0 );
+    assert( p->pLruPrev==0 || p->pLruPrev->pLruNext==p );
+    assert( p->pLruNext==0 || p->pLruNext->pLruPrev==p );
+    assert( p->nRef==0 );
+    assert( p->flags & PAGE_FREE );
+    assert( p==fsPageFindInHash(pFS, p->iPg, 0) );
+  }
+#endif
+}
+#endif
+
+/*
+** Wrappers around the VFS methods of the lsm_env object:
+**
+**     lsmEnvOpen()
+**     lsmEnvRead()
+**     lsmEnvWrite()
+**     lsmEnvSync()
+**     lsmEnvSectorSize()
+**     lsmEnvClose()
+**     lsmEnvTruncate()
+**     lsmEnvUnlink()
+**     lsmEnvRemap()
+*/
+int lsmEnvOpen(lsm_env *pEnv, const char *zFile, int flags, lsm_file **ppNew){
+  return pEnv->xOpen(pEnv, zFile, flags, ppNew);
+}
+
+static int lsmEnvRead(
+  lsm_env *pEnv, 
+  lsm_file *pFile, 
+  lsm_i64 iOff, 
+  void *pRead, 
+  int nRead
+){
+  return IOERR_WRAPPER( pEnv->xRead(pFile, iOff, pRead, nRead) );
+}
+
+static int lsmEnvWrite(
+  lsm_env *pEnv, 
+  lsm_file *pFile, 
+  lsm_i64 iOff, 
+  const void *pWrite, 
+  int nWrite
+){
+  return IOERR_WRAPPER( pEnv->xWrite(pFile, iOff, (void *)pWrite, nWrite) );
+}
+
+static int lsmEnvSync(lsm_env *pEnv, lsm_file *pFile){
+  return IOERR_WRAPPER( pEnv->xSync(pFile) );
+}
+
+static int lsmEnvSectorSize(lsm_env *pEnv, lsm_file *pFile){
+  return pEnv->xSectorSize(pFile);
+}
+
+int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile){
+  return IOERR_WRAPPER( pEnv->xClose(pFile) );
+}
+
+static int lsmEnvTruncate(lsm_env *pEnv, lsm_file *pFile, lsm_i64 nByte){
+  return IOERR_WRAPPER( pEnv->xTruncate(pFile, nByte) );
+}
+
+static int lsmEnvUnlink(lsm_env *pEnv, const char *zDel){
+  return IOERR_WRAPPER( pEnv->xUnlink(pEnv, zDel) );
+}
+
+static int lsmEnvRemap(
+  lsm_env *pEnv, 
+  lsm_file *pFile, 
+  i64 szMin,
+  void **ppMap,
+  i64 *pszMap
+){
+  return pEnv->xRemap(pFile, szMin, ppMap, pszMap);
+}
+
+int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock){
+  if( pFile==0 ) return LSM_OK;
+  return pEnv->xLock(pFile, iLock, eLock);
+}
+
+int lsmEnvTestLock(
+  lsm_env *pEnv, 
+  lsm_file *pFile, 
+  int iLock, 
+  int nLock, 
+  int eLock
+){
+  return pEnv->xTestLock(pFile, iLock, nLock, eLock);
+}
+
+int lsmEnvShmMap(
+  lsm_env *pEnv, 
+  lsm_file *pFile, 
+  int iChunk, 
+  int sz, 
+  void **ppOut
+){
+  return pEnv->xShmMap(pFile, iChunk, sz, ppOut);
+}
+
+void lsmEnvShmBarrier(lsm_env *pEnv){
+  pEnv->xShmBarrier();
+}
+
+void lsmEnvShmUnmap(lsm_env *pEnv, lsm_file *pFile, int bDel){
+  pEnv->xShmUnmap(pFile, bDel);
+}
+
+void lsmEnvSleep(lsm_env *pEnv, int nUs){
+  pEnv->xSleep(pEnv, nUs);
+}
+
+
+/*
+** Write the contents of string buffer pStr into the log file, starting at
+** offset iOff.
+*/
+int lsmFsWriteLog(FileSystem *pFS, i64 iOff, LsmString *pStr){
+  assert( pFS->fdLog );
+  return lsmEnvWrite(pFS->pEnv, pFS->fdLog, iOff, pStr->z, pStr->n);
+}
+
+/*
+** fsync() the log file.
+*/
+int lsmFsSyncLog(FileSystem *pFS){
+  assert( pFS->fdLog );
+  return lsmEnvSync(pFS->pEnv, pFS->fdLog);
+}
+
+/*
+** Read nRead bytes of data starting at offset iOff of the log file. Append
+** the results to string buffer pStr.
+*/
+int lsmFsReadLog(FileSystem *pFS, i64 iOff, int nRead, LsmString *pStr){
+  int rc;                         /* Return code */
+  assert( pFS->fdLog );
+  rc = lsmStringExtend(pStr, nRead);
+  if( rc==LSM_OK ){
+    rc = lsmEnvRead(pFS->pEnv, pFS->fdLog, iOff, &pStr->z[pStr->n], nRead);
+    pStr->n += nRead;
+  }
+  return rc;
+}
+
+/*
+** Truncate the log file to nByte bytes in size.
+*/
+int lsmFsTruncateLog(FileSystem *pFS, i64 nByte){
+  if( pFS->fdLog==0 ) return LSM_OK;
+  return lsmEnvTruncate(pFS->pEnv, pFS->fdLog, nByte);
+}
+
+/*
+** Truncate the db file to nByte bytes in size.
+*/
+int lsmFsTruncateDb(FileSystem *pFS, i64 nByte){
+  if( pFS->fdDb==0 ) return LSM_OK;
+  return lsmEnvTruncate(pFS->pEnv, pFS->fdDb, nByte);
+}
+
+/*
+** Close the log file. Then delete it from the file-system. This function
+** is called during database shutdown only.
+*/
+int lsmFsCloseAndDeleteLog(FileSystem *pFS){
+  char *zDel;
+
+  if( pFS->fdLog ){
+    lsmEnvClose(pFS->pEnv, pFS->fdLog );
+    pFS->fdLog = 0;
+  }
+
+  zDel = lsmMallocPrintf(pFS->pEnv, "%s-log", pFS->zDb);
+  if( zDel ){
+    lsmEnvUnlink(pFS->pEnv, zDel);
+    lsmFree(pFS->pEnv, zDel);
+  }
+  return LSM_OK;
+}
+
+/*
+** Return true if page iReal of the database should be accessed using mmap.
+** False otherwise.
+*/
+static int fsMmapPage(FileSystem *pFS, LsmPgno iReal){
+  return ((i64)iReal*pFS->nPagesize <= pFS->nMapLimit);
+}
+
+/*
+** Given that there are currently nHash slots in the hash table, return 
+** the hash key for file iFile, page iPg.
+*/
+static int fsHashKey(int nHash, LsmPgno iPg){
+  return (iPg % nHash);
+}
+
+/*
+** This is a helper function for lsmFsOpen(). It opens a single file on
+** disk (either the database or log file).
+*/
+static lsm_file *fsOpenFile(
+  FileSystem *pFS,                /* File system object */
+  int bReadonly,                  /* True to open this file read-only */
+  int bLog,                       /* True for log, false for db */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  lsm_file *pFile = 0;
+  if( *pRc==LSM_OK ){
+    int flags = (bReadonly ? LSM_OPEN_READONLY : 0);
+    const char *zPath = (bLog ? pFS->zLog : pFS->zDb);
+
+    *pRc = lsmEnvOpen(pFS->pEnv, zPath, flags, &pFile);
+  }
+  return pFile;
+}
+
+/*
+** If it is not already open, this function opens the log file. It returns
+** LSM_OK if successful (or if the log file was already open) or an LSM
+** error code otherwise.
+**
+** The log file must be opened before any of the following may be called:
+**
+**     lsmFsWriteLog
+**     lsmFsSyncLog
+**     lsmFsReadLog
+*/
+int lsmFsOpenLog(lsm_db *db, int *pbOpen){
+  int rc = LSM_OK;
+  FileSystem *pFS = db->pFS;
+
+  if( 0==pFS->fdLog ){ 
+    pFS->fdLog = fsOpenFile(pFS, db->bReadonly, 1, &rc); 
+
+    if( rc==LSM_IOERR_NOENT && db->bReadonly ){
+      rc = LSM_OK;
+    }
+  }
+
+  if( pbOpen ) *pbOpen = (pFS->fdLog!=0);
+  return rc;
+}
+
+/*
+** Close the log file, if it is open.
+*/
+void lsmFsCloseLog(lsm_db *db){
+  FileSystem *pFS = db->pFS;
+  if( pFS->fdLog ){
+    lsmEnvClose(pFS->pEnv, pFS->fdLog);
+    pFS->fdLog = 0;
+  }
+}
+
+/*
+** Open a connection to a database stored within the file-system.
+**
+** If parameter bReadonly is true, then open a read-only file-descriptor
+** on the database file. It is possible that bReadonly will be false even
+** if the user requested that pDb be opened read-only. This is because the
+** file-descriptor may later on be recycled by a read-write connection.
+** If the db file can be opened for read-write access, it always is. Parameter
+** bReadonly is only ever true if it has already been determined that the
+** db can only be opened for read-only access.
+**
+** Return LSM_OK if successful or an lsm error code otherwise.
+*/
+int lsmFsOpen(
+  lsm_db *pDb,                    /* Database connection to open fd for */
+  const char *zDb,                /* Full path to database file */
+  int bReadonly                   /* True to open db file read-only */
+){
+  FileSystem *pFS;
+  int rc = LSM_OK;
+  int nDb = strlen(zDb);
+  int nByte;
+
+  assert( pDb->pFS==0 );
+  assert( pDb->pWorker==0 && pDb->pClient==0 );
+
+  nByte = sizeof(FileSystem) + nDb+1 + nDb+4+1;
+  pFS = (FileSystem *)lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
+  if( pFS ){
+    LsmFile *pLsmFile;
+    pFS->zDb = (char *)&pFS[1];
+    pFS->zLog = &pFS->zDb[nDb+1];
+    pFS->nPagesize = LSM_DFLT_PAGE_SIZE;
+    pFS->nBlocksize = LSM_DFLT_BLOCK_SIZE;
+    pFS->nMetasize = LSM_META_PAGE_SIZE;
+    pFS->nMetaRwSize = LSM_META_RW_PAGE_SIZE;
+    pFS->pDb = pDb;
+    pFS->pEnv = pDb->pEnv;
+
+    /* Make a copy of the database and log file names. */
+    memcpy(pFS->zDb, zDb, nDb+1);
+    memcpy(pFS->zLog, zDb, nDb);
+    memcpy(&pFS->zLog[nDb], "-log", 5);
+
+    /* Allocate the hash-table here. At some point, it should be changed
+    ** so that it can grow dynamicly. */
+    pFS->nCacheMax = 2048*1024 / pFS->nPagesize;
+    pFS->nHash = 4096;
+    pFS->apHash = lsmMallocZeroRc(pDb->pEnv, sizeof(Page *) * pFS->nHash, &rc);
+
+    /* Open the database file */
+    pLsmFile = lsmDbRecycleFd(pDb);
+    if( pLsmFile ){
+      pFS->pLsmFile = pLsmFile;
+      pFS->fdDb = pLsmFile->pFile;
+      memset(pLsmFile, 0, sizeof(LsmFile));
+    }else{
+      pFS->pLsmFile = lsmMallocZeroRc(pDb->pEnv, sizeof(LsmFile), &rc);
+      if( rc==LSM_OK ){
+        pFS->fdDb = fsOpenFile(pFS, bReadonly, 0, &rc);
+      }
+    }
+
+    if( rc!=LSM_OK ){
+      lsmFsClose(pFS);
+      pFS = 0;
+    }else{
+      pFS->szSector = lsmEnvSectorSize(pFS->pEnv, pFS->fdDb);
+    }
+  }
+
+  pDb->pFS = pFS;
+  return rc;
+}
+
+/*
+** Configure the file-system object according to the current values of
+** the LSM_CONFIG_MMAP and LSM_CONFIG_SET_COMPRESSION options.
+*/
+int lsmFsConfigure(lsm_db *db){
+  FileSystem *pFS = db->pFS;
+  if( pFS ){
+    lsm_env *pEnv = pFS->pEnv;
+    Page *pPg;
+
+    assert( pFS->nOut==0 );
+    assert( pFS->pWaiting==0 );
+    assert( pFS->pMapped==0 );
+
+    /* Reset any compression/decompression buffers already allocated */
+    lsmFree(pEnv, pFS->aIBuffer);
+    lsmFree(pEnv, pFS->aOBuffer);
+    pFS->nBuffer = 0;
+
+    /* Unmap the file, if it is currently mapped */
+    if( pFS->pMap ){
+      lsmEnvRemap(pEnv, pFS->fdDb, -1, &pFS->pMap, &pFS->nMap);
+      pFS->nMapLimit = 0;
+    }
+
+    /* Free all allocated page structures */
+    pPg = pFS->pLruFirst;
+    while( pPg ){
+      Page *pNext = pPg->pLruNext;
+      assert( pPg->flags & PAGE_FREE );
+      lsmFree(pEnv, pPg->aData);
+      lsmFree(pEnv, pPg);
+      pPg = pNext;
+    }
+
+    pPg = pFS->pFree;
+    while( pPg ){
+      Page *pNext = pPg->pFreeNext;
+      lsmFree(pEnv, pPg);
+      pPg = pNext;
+    }
+
+    /* Zero pointers that point to deleted page objects */
+    pFS->nCacheAlloc = 0;
+    pFS->pLruFirst = 0;
+    pFS->pLruLast = 0;
+    pFS->pFree = 0;
+    if( pFS->apHash ){
+      memset(pFS->apHash, 0, pFS->nHash*sizeof(pFS->apHash[0]));
+    }
+
+    /* Configure the FileSystem object */
+    if( db->compress.xCompress ){
+      pFS->pCompress = &db->compress;
+      pFS->nMapLimit = 0;
+    }else{
+      pFS->pCompress = 0;
+      if( db->iMmap==1 ){
+        /* Unlimited */
+        pFS->nMapLimit = (i64)1 << 60;
+      }else{
+        /* iMmap is a limit in KB. Set nMapLimit to the same value in bytes. */
+        pFS->nMapLimit = (i64)db->iMmap * 1024;
+      }
+    }
+  }
+
+  return LSM_OK;
+}
+
+/*
+** Close and destroy a FileSystem object.
+*/
+void lsmFsClose(FileSystem *pFS){
+  if( pFS ){
+    Page *pPg;
+    lsm_env *pEnv = pFS->pEnv;
+
+    assert( pFS->nOut==0 );
+    pPg = pFS->pLruFirst;
+    while( pPg ){
+      Page *pNext = pPg->pLruNext;
+      if( pPg->flags & PAGE_FREE ) lsmFree(pEnv, pPg->aData);
+      lsmFree(pEnv, pPg);
+      pPg = pNext;
+    }
+
+    pPg = pFS->pFree;
+    while( pPg ){
+      Page *pNext = pPg->pFreeNext;
+      if( pPg->flags & PAGE_FREE ) lsmFree(pEnv, pPg->aData);
+      lsmFree(pEnv, pPg);
+      pPg = pNext;
+    }
+
+    if( pFS->fdDb ) lsmEnvClose(pFS->pEnv, pFS->fdDb );
+    if( pFS->fdLog ) lsmEnvClose(pFS->pEnv, pFS->fdLog );
+    lsmFree(pEnv, pFS->pLsmFile);
+    lsmFree(pEnv, pFS->apHash);
+    lsmFree(pEnv, pFS->aIBuffer);
+    lsmFree(pEnv, pFS->aOBuffer);
+    lsmFree(pEnv, pFS);
+  }
+}
+
+/*
+** This function is called when closing a database handle (i.e. lsm_close()) 
+** if there exist other connections to the same database within this process.
+** In that case the file-descriptor open on the database file is not closed
+** when the FileSystem object is destroyed, as this would cause any POSIX
+** locks held by the other connections to be silently dropped (see "man close"
+** for details). Instead, the file-descriptor is stored in a list by the
+** lsm_shared.c module until it is either closed or reused.
+**
+** This function returns a pointer to an object that can be linked into
+** the list described above. The returned object now 'owns' the database
+** file descriptr, so that when the FileSystem object is destroyed, it
+** will not be closed. 
+**
+** This function may be called at most once in the life-time of a 
+** FileSystem object. The results of any operations involving the database 
+** file descriptor are undefined once this function has been called.
+**
+** None of this is necessary on non-POSIX systems. But we do it anyway in
+** the name of using as similar code as possible on all platforms.
+*/
+LsmFile *lsmFsDeferClose(FileSystem *pFS){
+  LsmFile *p = pFS->pLsmFile;
+  assert( p->pNext==0 );
+  p->pFile = pFS->fdDb;
+  pFS->fdDb = 0;
+  pFS->pLsmFile = 0;
+  return p;
+}
+
+/*
+** Allocate a buffer and populate it with the output of the xFileid() 
+** method of the database file handle. If successful, set *ppId to point 
+** to the buffer and *pnId to the number of bytes in the buffer and return
+** LSM_OK. Otherwise, set *ppId and *pnId to zero and return an LSM
+** error code.
+*/
+int lsmFsFileid(lsm_db *pDb, void **ppId, int *pnId){
+  lsm_env *pEnv = pDb->pEnv;
+  FileSystem *pFS = pDb->pFS;
+  int rc;
+  int nId = 0;
+  void *pId;
+
+  rc = pEnv->xFileid(pFS->fdDb, 0, &nId);
+  pId = lsmMallocZeroRc(pEnv, nId, &rc);
+  if( rc==LSM_OK ) rc = pEnv->xFileid(pFS->fdDb, pId, &nId);
+
+  if( rc!=LSM_OK ){
+    lsmFree(pEnv, pId);
+    pId = 0;
+    nId = 0;
+  }
+
+  *ppId = pId;
+  *pnId = nId;
+  return rc;
+}
+
+/*
+** Return the nominal page-size used by this file-system. Actual pages
+** may be smaller or larger than this value.
+*/
+int lsmFsPageSize(FileSystem *pFS){
+  return pFS->nPagesize;
+}
+
+/*
+** Return the block-size used by this file-system.
+*/
+int lsmFsBlockSize(FileSystem *pFS){
+  return pFS->nBlocksize;
+}
+
+/*
+** Configure the nominal page-size used by this file-system. Actual 
+** pages may be smaller or larger than this value.
+*/
+void lsmFsSetPageSize(FileSystem *pFS, int nPgsz){
+  pFS->nPagesize = nPgsz;
+  pFS->nCacheMax = 2048*1024 / pFS->nPagesize;
+}
+
+/*
+** Configure the block-size used by this file-system. 
+*/
+void lsmFsSetBlockSize(FileSystem *pFS, int nBlocksize){
+  pFS->nBlocksize = nBlocksize;
+}
+
+/*
+** Return the page number of the first page on block iBlock. Blocks are
+** numbered starting from 1.
+**
+** For a compressed database, page numbers are byte offsets. The first
+** page on each block is the byte offset immediately following the 4-byte
+** "previous block" pointer at the start of each block.
+*/
+static LsmPgno fsFirstPageOnBlock(FileSystem *pFS, int iBlock){
+  LsmPgno iPg;
+  if( pFS->pCompress ){
+    if( iBlock==1 ){
+      iPg = pFS->nMetasize * 2 + 4;
+    }else{
+      iPg = pFS->nBlocksize * (LsmPgno)(iBlock-1) + 4;
+    }
+  }else{
+    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
+    if( iBlock==1 ){
+      iPg = 1 + ((pFS->nMetasize*2 + pFS->nPagesize - 1) / pFS->nPagesize);
+    }else{
+      iPg = 1 + (iBlock-1) * nPagePerBlock;
+    }
+  }
+  return iPg;
+}
+
+/*
+** Return the page number of the last page on block iBlock. Blocks are
+** numbered starting from 1.
+**
+** For a compressed database, page numbers are byte offsets. The first
+** page on each block is the byte offset of the byte immediately before 
+** the 4-byte "next block" pointer at the end of each block.
+*/
+static LsmPgno fsLastPageOnBlock(FileSystem *pFS, int iBlock){
+  if( pFS->pCompress ){
+    return pFS->nBlocksize * (LsmPgno)iBlock - 1 - 4;
+  }else{
+    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
+    return iBlock * nPagePerBlock;
+  }
+}
+
+/*
+** Return the block number of the block that page iPg is located on. 
+** Blocks are numbered starting from 1.
+*/
+static int fsPageToBlock(FileSystem *pFS, LsmPgno iPg){
+  if( pFS->pCompress ){
+    return (int)((iPg / pFS->nBlocksize) + 1);
+  }else{
+    return (int)(1 + ((iPg-1) / (pFS->nBlocksize / pFS->nPagesize)));
+  }
+}
+
+/*
+** Return true if page iPg is the last page on its block.
+**
+** This function is only called in non-compressed database mode.
+*/
+static int fsIsLast(FileSystem *pFS, LsmPgno iPg){
+  const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
+  assert( !pFS->pCompress );
+  return ( iPg && (iPg % nPagePerBlock)==0 );
+}
+
+/*
+** Return true if page iPg is the first page on its block.
+**
+** This function is only called in non-compressed database mode.
+*/
+static int fsIsFirst(FileSystem *pFS, LsmPgno iPg){
+  const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
+  assert( !pFS->pCompress );
+  return ( (iPg % nPagePerBlock)==1
+        || (iPg<nPagePerBlock && iPg==fsFirstPageOnBlock(pFS, 1))
+  );
+}
+
+/*
+** Given a page reference, return a pointer to the buffer containing the 
+** pages contents. If parameter pnData is not NULL, set *pnData to the size
+** of the buffer in bytes before returning.
+*/
+u8 *lsmFsPageData(Page *pPage, int *pnData){
+  if( pnData ){
+    *pnData = pPage->nData;
+  }
+  return pPage->aData;
+}
+
+/*
+** Return the page number of a page.
+*/
+LsmPgno lsmFsPageNumber(Page *pPage){
+  /* assert( (pPage->flags & PAGE_DIRTY)==0 ); */
+  return pPage ? pPage->iPg : 0;
+}
+
+/*
+** Page pPg is currently part of the LRU list belonging to pFS. Remove
+** it from the list. pPg->pLruNext and pPg->pLruPrev are cleared by this
+** operation.
+*/
+static void fsPageRemoveFromLru(FileSystem *pFS, Page *pPg){
+  assert( pPg->pLruNext || pPg==pFS->pLruLast );
+  assert( pPg->pLruPrev || pPg==pFS->pLruFirst );
+  if( pPg->pLruNext ){
+    pPg->pLruNext->pLruPrev = pPg->pLruPrev;
+  }else{
+    pFS->pLruLast = pPg->pLruPrev;
+  }
+  if( pPg->pLruPrev ){
+    pPg->pLruPrev->pLruNext = pPg->pLruNext;
+  }else{
+    pFS->pLruFirst = pPg->pLruNext;
+  }
+  pPg->pLruPrev = 0;
+  pPg->pLruNext = 0;
+}
+
+/*
+** Page pPg is not currently part of the LRU list belonging to pFS. Add it.
+*/
+static void fsPageAddToLru(FileSystem *pFS, Page *pPg){
+  assert( pPg->pLruNext==0 && pPg->pLruPrev==0 );
+  pPg->pLruPrev = pFS->pLruLast;
+  if( pPg->pLruPrev ){
+    pPg->pLruPrev->pLruNext = pPg;
+  }else{
+    pFS->pLruFirst = pPg;
+  }
+  pFS->pLruLast = pPg;
+}
+
+/*
+** Page pPg is currently stored in the apHash/nHash hash table. Remove it.
+*/
+static void fsPageRemoveFromHash(FileSystem *pFS, Page *pPg){
+  int iHash;
+  Page **pp;
+
+  iHash = fsHashKey(pFS->nHash, pPg->iPg);
+  for(pp=&pFS->apHash[iHash]; *pp!=pPg; pp=&(*pp)->pHashNext);
+  *pp = pPg->pHashNext;
+  pPg->pHashNext = 0;
+}
+
+/*
+** Free a Page object allocated by fsPageBuffer().
+*/
+static void fsPageBufferFree(Page *pPg){
+  pPg->pFS->nCacheAlloc--;
+  lsmFree(pPg->pFS->pEnv, pPg->aData);
+  lsmFree(pPg->pFS->pEnv, pPg);
+}
+
+
+/*
+** Purge the cache of all non-mmap pages with nRef==0.
+*/
+void lsmFsPurgeCache(FileSystem *pFS){
+  Page *pPg;
+
+  pPg = pFS->pLruFirst;
+  while( pPg ){
+    Page *pNext = pPg->pLruNext;
+    assert( pPg->flags & PAGE_FREE );
+    fsPageRemoveFromHash(pFS, pPg);
+    fsPageBufferFree(pPg);
+    pPg = pNext;
+  }
+  pFS->pLruFirst = 0;
+  pFS->pLruLast = 0;
+
+  assert( pFS->nCacheAlloc<=pFS->nOut && pFS->nCacheAlloc>=0 );
+}
+
+/*
+** Search the hash-table for page iPg. If an entry is round, return a pointer
+** to it. Otherwise, return NULL.
+**
+** Either way, if argument piHash is not NULL set *piHash to the hash slot
+** number that page iPg would be stored in before returning.
+*/
+static Page *fsPageFindInHash(FileSystem *pFS, LsmPgno iPg, int *piHash){
+  Page *p;                        /* Return value */
+  int iHash = fsHashKey(pFS->nHash, iPg);
+
+  if( piHash ) *piHash = iHash;
+  for(p=pFS->apHash[iHash]; p; p=p->pHashNext){
+    if( p->iPg==iPg) break;
+  }
+  return p;
+}
+
+/*
+** Allocate and return a non-mmap Page object. If there are already 
+** nCacheMax such Page objects outstanding, try to recycle an existing 
+** Page instead.
+*/
+static int fsPageBuffer(
+  FileSystem *pFS, 
+  Page **ppOut
+){
+  int rc = LSM_OK;
+  Page *pPage = 0;
+  if( pFS->pLruFirst==0 || pFS->nCacheAlloc<pFS->nCacheMax ){
+    /* Allocate a new Page object */
+    pPage = lsmMallocZero(pFS->pEnv, sizeof(Page));
+    if( !pPage ){
+      rc = LSM_NOMEM_BKPT;
+    }else{
+      pPage->aData = (u8 *)lsmMalloc(pFS->pEnv, pFS->nPagesize);
+      if( !pPage->aData ){
+        lsmFree(pFS->pEnv, pPage);
+        rc = LSM_NOMEM_BKPT;
+        pPage = 0;
+      }else{
+        pFS->nCacheAlloc++;
+      }
+    }
+  }else{
+    /* Reuse an existing Page object */
+    u8 *aData;
+    pPage = pFS->pLruFirst;
+    aData = pPage->aData;
+    fsPageRemoveFromLru(pFS, pPage);
+    fsPageRemoveFromHash(pFS, pPage);
+
+    memset(pPage, 0, sizeof(Page));
+    pPage->aData = aData;
+  }
+
+  if( pPage ){
+    pPage->flags = PAGE_FREE;
+  }
+  *ppOut = pPage;
+  return rc;
+}
+
+/*
+** Assuming *pRc is initially LSM_OK, attempt to ensure that the 
+** memory-mapped region is at least iSz bytes in size. If it is not already,
+** iSz bytes in size, extend it and update the pointers associated with any
+** outstanding Page objects.
+**
+** If *pRc is not LSM_OK when this function is called, it is a no-op. 
+** Otherwise, *pRc is set to an lsm error code if an error occurs, or
+** left unmodified otherwise.
+**
+** This function is never called in compressed database mode.
+*/
+static void fsGrowMapping(
+  FileSystem *pFS,                /* File system object */
+  i64 iSz,                        /* Minimum size to extend mapping to */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  assert( pFS->pCompress==0 );
+  assert( PAGE_HASPREV==4 );
+
+  if( *pRc==LSM_OK && iSz>pFS->nMap ){
+    int rc;
+    u8 *aOld = pFS->pMap;
+    rc = lsmEnvRemap(pFS->pEnv, pFS->fdDb, iSz, &pFS->pMap, &pFS->nMap);
+    if( rc==LSM_OK && pFS->pMap!=aOld ){
+      Page *pFix;
+      i64 iOff = (u8 *)pFS->pMap - aOld;
+      for(pFix=pFS->pMapped; pFix; pFix=pFix->pMappedNext){
+        pFix->aData += iOff;
+      }
+      lsmSortedRemap(pFS->pDb);
+    }
+    *pRc = rc;
+  }
+}
+
+/*
+** If it is mapped, unmap the database file.
+*/
+int lsmFsUnmap(FileSystem *pFS){
+  int rc = LSM_OK;
+  if( pFS ){
+    rc = lsmEnvRemap(pFS->pEnv, pFS->fdDb, -1, &pFS->pMap, &pFS->nMap);
+  }
+  return rc;
+}
+
+/*
+** fsync() the database file.
+*/
+int lsmFsSyncDb(FileSystem *pFS, int nBlock){
+  return lsmEnvSync(pFS->pEnv, pFS->fdDb);
+}
+
+/*
+** If block iBlk has been redirected according to the redirections in the
+** object passed as the first argument, return the destination block to
+** which it is redirected. Otherwise, return a copy of iBlk.
+*/
+static int fsRedirectBlock(Redirect *p, int iBlk){
+  if( p ){
+    int i;
+    for(i=0; i<p->n; i++){
+      if( iBlk==p->a[i].iFrom ) return p->a[i].iTo;
+    }
+  }
+  assert( iBlk!=0 );
+  return iBlk;
+}
+
+/*
+** If page iPg has been redirected according to the redirections in the
+** object passed as the second argument, return the destination page to
+** which it is redirected. Otherwise, return a copy of iPg.
+*/
+LsmPgno lsmFsRedirectPage(FileSystem *pFS, Redirect *pRedir, LsmPgno iPg){
+  LsmPgno iReal = iPg;
+
+  if( pRedir ){
+    const int nPagePerBlock = (
+        pFS->pCompress ? pFS->nBlocksize : (pFS->nBlocksize / pFS->nPagesize)
+    );
+    int iBlk = fsPageToBlock(pFS, iPg);
+    int i;
+    for(i=0; i<pRedir->n; i++){
+      int iFrom = pRedir->a[i].iFrom;
+      if( iFrom>iBlk ) break;
+      if( iFrom==iBlk ){
+        int iTo = pRedir->a[i].iTo;
+        iReal = iPg - (LsmPgno)(iFrom - iTo) * nPagePerBlock;
+        if( iTo==1 ){
+          iReal += (fsFirstPageOnBlock(pFS, 1)-1);
+        }
+        break;
+      }
+    }
+  }
+
+  assert( iReal!=0 );
+  return iReal;
+}
+
+/* Required by the circular fsBlockNext<->fsPageGet dependency. */
+static int fsPageGet(FileSystem *, Segment *, LsmPgno, int, Page **, int *);
+
+/*
+** Parameter iBlock is a database file block. This function reads the value 
+** stored in the blocks "next block" pointer and stores it in *piNext.
+** LSM_OK is returned if everything is successful, or an LSM error code
+** otherwise.
+*/
+static int fsBlockNext(
+  FileSystem *pFS,                /* File-system object handle */
+  Segment *pSeg,                  /* Use this segment for block redirects */
+  int iBlock,                     /* Read field from this block */
+  int *piNext                     /* OUT: Next block in linked list */
+){
+  int rc;
+  int iRead;                      /* Read block from here */
+  
+  if( pSeg ){
+    iRead = fsRedirectBlock(pSeg->pRedirect, iBlock);
+  }else{
+    iRead = iBlock;
+  }
+
+  assert( pFS->nMapLimit==0 || pFS->pCompress==0 );
+  if( pFS->pCompress ){
+    i64 iOff;                     /* File offset to read data from */
+    u8 aNext[4];                  /* 4-byte pointer read from db file */
+
+    iOff = (i64)iRead * pFS->nBlocksize - sizeof(aNext);
+    rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aNext, sizeof(aNext));
+    if( rc==LSM_OK ){
+      *piNext = (int)lsmGetU32(aNext);
+    }
+  }else{
+    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
+    Page *pLast;
+    rc = fsPageGet(pFS, 0, iRead*nPagePerBlock, 0, &pLast, 0);
+    if( rc==LSM_OK ){
+      *piNext = lsmGetU32(&pLast->aData[pFS->nPagesize-4]);
+      lsmFsPageRelease(pLast);
+    }
+  }
+
+  if( pSeg ){
+    *piNext = fsRedirectBlock(pSeg->pRedirect, *piNext);
+  }
+  return rc;
+}
+
+/*
+** Return the page number of the last page on the same block as page iPg.
+*/
+LsmPgno fsLastPageOnPagesBlock(FileSystem *pFS, LsmPgno iPg){
+  return fsLastPageOnBlock(pFS, fsPageToBlock(pFS, iPg));
+}
+
+/*
+** Read nData bytes of data from offset iOff of the database file into
+** buffer aData. If this means reading past the end of a block, follow
+** the block pointer to the next block and continue reading.
+**
+** Offset iOff is an absolute offset - not subject to any block redirection.
+** However any block pointer followed is. Use pSeg->pRedirect in this case.
+**
+** This function is only called in compressed database mode.
+*/
+static int fsReadData(
+  FileSystem *pFS,                /* File-system handle */
+  Segment *pSeg,                  /* Block redirection */
+  i64 iOff,                       /* Read data from this offset */
+  u8 *aData,                      /* Buffer to read data into */
+  int nData                       /* Number of bytes to read */
+){
+  i64 iEob;                       /* End of block */
+  int nRead;
+  int rc;
+
+  assert( pFS->pCompress );
+
+  iEob = fsLastPageOnPagesBlock(pFS, iOff) + 1;
+  nRead = (int)LSM_MIN(iEob - iOff, nData);
+
+  rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aData, nRead);
+  if( rc==LSM_OK && nRead!=nData ){
+    int iBlk;
+
+    rc = fsBlockNext(pFS, pSeg, fsPageToBlock(pFS, iOff), &iBlk);
+    if( rc==LSM_OK ){
+      i64 iOff2 = fsFirstPageOnBlock(pFS, iBlk);
+      rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff2, &aData[nRead], nData-nRead);
+    }
+  }
+
+  return rc;
+}
+
+/*
+** Parameter iBlock is a database file block. This function reads the value 
+** stored in the blocks "previous block" pointer and stores it in *piPrev.
+** LSM_OK is returned if everything is successful, or an LSM error code
+** otherwise.
+*/
+static int fsBlockPrev(
+  FileSystem *pFS,                /* File-system object handle */
+  Segment *pSeg,                  /* Use this segment for block redirects */
+  int iBlock,                     /* Read field from this block */
+  int *piPrev                     /* OUT: Previous block in linked list */
+){
+  int rc = LSM_OK;                /* Return code */
+
+  assert( pFS->nMapLimit==0 || pFS->pCompress==0 );
+  assert( iBlock>0 );
+
+  if( pFS->pCompress ){
+    i64 iOff = fsFirstPageOnBlock(pFS, iBlock) - 4;
+    u8 aPrev[4];                  /* 4-byte pointer read from db file */
+    rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aPrev, sizeof(aPrev));
+    if( rc==LSM_OK ){
+      Redirect *pRedir = (pSeg ? pSeg->pRedirect : 0);
+      *piPrev = fsRedirectBlock(pRedir, (int)lsmGetU32(aPrev));
+    }
+  }else{
+    assert( 0 );
+  }
+  return rc;
+}
+
+/*
+** Encode and decode routines for record size fields.
+*/
+static void putRecordSize(u8 *aBuf, int nByte, int bFree){
+  aBuf[0] = (u8)(nByte >> 14) | 0x80;
+  aBuf[1] = ((u8)(nByte >>  7) & 0x7F) | (bFree ? 0x00 : 0x80);
+  aBuf[2] = (u8)nByte | 0x80;
+}
+static int getRecordSize(u8 *aBuf, int *pbFree){
+  int nByte;
+  nByte  = (aBuf[0] & 0x7F) << 14;
+  nByte += (aBuf[1] & 0x7F) << 7;
+  nByte += (aBuf[2] & 0x7F);
+  *pbFree = !(aBuf[1] & 0x80);
+  return nByte;
+}
+
+/*
+** Subtract iSub from database file offset iOff and set *piRes to the
+** result. If doing so means passing the start of a block, follow the
+** block pointer stored in the first 4 bytes of the block.
+**
+** Offset iOff is an absolute offset - not subject to any block redirection.
+** However any block pointer followed is. Use pSeg->pRedirect in this case.
+**
+** Return LSM_OK if successful or an lsm error code if an error occurs.
+*/
+static int fsSubtractOffset(
+  FileSystem *pFS, 
+  Segment *pSeg,
+  i64 iOff, 
+  int iSub, 
+  i64 *piRes
+){
+  i64 iStart;
+  int iBlk = 0;
+  int rc;
+
+  assert( pFS->pCompress );
+
+  iStart = fsFirstPageOnBlock(pFS, fsPageToBlock(pFS, iOff));
+  if( (iOff-iSub)>=iStart ){
+    *piRes = (iOff-iSub);
+    return LSM_OK;
+  }
+
+  rc = fsBlockPrev(pFS, pSeg, fsPageToBlock(pFS, iOff), &iBlk);
+  *piRes = fsLastPageOnBlock(pFS, iBlk) - iSub + (iOff - iStart + 1);
+  return rc;
+}
+
+/*
+** Add iAdd to database file offset iOff and set *piRes to the
+** result. If doing so means passing the end of a block, follow the
+** block pointer stored in the last 4 bytes of the block.
+**
+** Offset iOff is an absolute offset - not subject to any block redirection.
+** However any block pointer followed is. Use pSeg->pRedirect in this case.
+**
+** Return LSM_OK if successful or an lsm error code if an error occurs.
+*/
+static int fsAddOffset(
+  FileSystem *pFS, 
+  Segment *pSeg,
+  i64 iOff, 
+  int iAdd, 
+  i64 *piRes
+){
+  i64 iEob;
+  int iBlk;
+  int rc;
+
+  assert( pFS->pCompress );
+
+  iEob = fsLastPageOnPagesBlock(pFS, iOff);
+  if( (iOff+iAdd)<=iEob ){
+    *piRes = (iOff+iAdd);
+    return LSM_OK;
+  }
+
+  rc = fsBlockNext(pFS, pSeg, fsPageToBlock(pFS, iOff), &iBlk);
+  *piRes = fsFirstPageOnBlock(pFS, iBlk) + iAdd - (iEob - iOff + 1);
+  return rc;
+}
+
+/*
+** If it is not already allocated, allocate either the FileSystem.aOBuffer (if
+** bWrite is true) or the FileSystem.aIBuffer (if bWrite is false). Return
+** LSM_OK if successful if the attempt to allocate memory fails.
+*/
+static int fsAllocateBuffer(FileSystem *pFS, int bWrite){
+  u8 **pp;                        /* Pointer to either aIBuffer or aOBuffer */
+
+  assert( pFS->pCompress );
+
+  /* If neither buffer has been allocated, figure out how large they
+  ** should be. Store this value in FileSystem.nBuffer.  */
+  if( pFS->nBuffer==0 ){
+    assert( pFS->aIBuffer==0 && pFS->aOBuffer==0 );
+    pFS->nBuffer = pFS->pCompress->xBound(pFS->pCompress->pCtx, pFS->nPagesize);
+    if( pFS->nBuffer<(pFS->szSector+6) ){
+      pFS->nBuffer = pFS->szSector+6;
+    }
+  }
+
+  pp = (bWrite ? &pFS->aOBuffer : &pFS->aIBuffer);
+  if( *pp==0 ){
+    *pp = lsmMalloc(pFS->pEnv, LSM_MAX(pFS->nBuffer, pFS->nPagesize));
+    if( *pp==0 ) return LSM_NOMEM_BKPT;
+  }
+
+  return LSM_OK;
+}
+
+/*
+** This function is only called in compressed database mode. It reads and
+** uncompresses the compressed data for page pPg from the database and
+** populates the pPg->aData[] buffer and pPg->nCompress field.
+**
+** It is possible that instead of a page record, there is free space
+** at offset pPg->iPgno. In this case no data is read from the file, but
+** output variable *pnSpace is set to the total number of free bytes.
+**
+** LSM_OK is returned if successful, or an LSM error code otherwise.
+*/
+static int fsReadPagedata(
+  FileSystem *pFS,                /* File-system handle */
+  Segment *pSeg,                  /* pPg is part of this segment */
+  Page *pPg,                      /* Page to read and uncompress data for */
+  int *pnSpace                    /* OUT: Total bytes of free space */
+){
+  lsm_compress *p = pFS->pCompress;
+  i64 iOff = pPg->iPg;
+  u8 aSz[3];
+  int rc;
+
+  assert( p && pPg->nCompress==0 );
+
+  if( fsAllocateBuffer(pFS, 0) ) return LSM_NOMEM;
+
+  rc = fsReadData(pFS, pSeg, iOff, aSz, sizeof(aSz));
+
+  if( rc==LSM_OK ){
+    int bFree;
+    if( aSz[0] & 0x80 ){
+      pPg->nCompress = (int)getRecordSize(aSz, &bFree);
+    }else{
+      pPg->nCompress = (int)aSz[0] - sizeof(aSz)*2;
+      bFree = 1;
+    }
+    if( bFree ){
+      if( pnSpace ){
+        *pnSpace = pPg->nCompress + sizeof(aSz)*2;
+      }else{
+        rc = LSM_CORRUPT_BKPT;
+      }
+    }else{
+      rc = fsAddOffset(pFS, pSeg, iOff, 3, &iOff);
+      if( rc==LSM_OK ){
+        if( pPg->nCompress>pFS->nBuffer ){
+          rc = LSM_CORRUPT_BKPT;
+        }else{
+          rc = fsReadData(pFS, pSeg, iOff, pFS->aIBuffer, pPg->nCompress);
+        }
+        if( rc==LSM_OK ){
+          int n = pFS->nPagesize;
+          rc = p->xUncompress(p->pCtx, 
+              (char *)pPg->aData, &n, 
+              (const char *)pFS->aIBuffer, pPg->nCompress
+          );
+          if( rc==LSM_OK && n!=pPg->pFS->nPagesize ){
+            rc = LSM_CORRUPT_BKPT;
+          }
+        }
+      }
+    }
+  }
+  return rc;
+}
+
+/*
+** Return a handle for a database page.
+**
+** If this file-system object is accessing a compressed database it may be
+** that there is no page record at database file offset iPg. Instead, there
+** may be a free space record. In this case, set *ppPg to NULL and *pnSpace
+** to the total number of free bytes before returning.
+**
+** If no error occurs, LSM_OK is returned. Otherwise, an lsm error code.
+*/
+static int fsPageGet(
+  FileSystem *pFS,                /* File-system handle */
+  Segment *pSeg,                  /* Block redirection to use (or NULL) */
+  LsmPgno iPg,                    /* Page id */
+  int noContent,                  /* True to not load content from disk */
+  Page **ppPg,                    /* OUT: New page handle */
+  int *pnSpace                    /* OUT: Bytes of free space */
+){
+  Page *p;
+  int iHash;
+  int rc = LSM_OK;
+
+  /* In most cases iReal is the same as iPg. Except, if pSeg->pRedirect is 
+  ** not NULL, and the block containing iPg has been redirected, then iReal
+  ** is the page number after redirection.  */
+  LsmPgno iReal = lsmFsRedirectPage(pFS, (pSeg ? pSeg->pRedirect : 0), iPg);
+
+  assert_lists_are_ok(pFS);
+  assert( iPg>=fsFirstPageOnBlock(pFS, 1) );
+  assert( iReal>=fsFirstPageOnBlock(pFS, 1) );
+  *ppPg = 0;
+
+  /* Search the hash-table for the page */
+  p = fsPageFindInHash(pFS, iReal, &iHash);
+
+  if( p ){
+    assert( p->flags & PAGE_FREE );
+    if( p->nRef==0 ) fsPageRemoveFromLru(pFS, p);
+  }else{
+
+    if( fsMmapPage(pFS, iReal) ){
+      i64 iEnd = (i64)iReal * pFS->nPagesize;
+      fsGrowMapping(pFS, iEnd, &rc);
+      if( rc!=LSM_OK ) return rc;
+
+      if( pFS->pFree ){
+        p = pFS->pFree;
+        pFS->pFree = p->pFreeNext;
+        assert( p->nRef==0 );
+      }else{
+        p = lsmMallocZeroRc(pFS->pEnv, sizeof(Page), &rc);
+        if( rc ) return rc;
+        p->pFS = pFS;
+      }
+      p->aData = &((u8 *)pFS->pMap)[pFS->nPagesize * (iReal-1)];
+      p->iPg = iReal;
+
+      /* This page now carries a pointer to the mapping. Link it in to
+      ** the FileSystem.pMapped list.  */
+      assert( p->pMappedNext==0 );
+      p->pMappedNext = pFS->pMapped;
+      pFS->pMapped = p;
+
+      assert( pFS->pCompress==0 );
+      assert( (p->flags & PAGE_FREE)==0 );
+    }else{
+      rc = fsPageBuffer(pFS, &p);
+      if( rc==LSM_OK ){
+        int nSpace = 0;
+        p->iPg = iReal;
+        p->nRef = 0;
+        p->pFS = pFS;
+        assert( p->flags==0 || p->flags==PAGE_FREE );
+
+#ifdef LSM_DEBUG
+        memset(p->aData, 0x56, pFS->nPagesize);
+#endif
+        assert( p->pLruNext==0 && p->pLruPrev==0 );
+        if( noContent==0 ){
+          if( pFS->pCompress ){
+            rc = fsReadPagedata(pFS, pSeg, p, &nSpace);
+          }else{
+            int nByte = pFS->nPagesize;
+            i64 iOff = (i64)(iReal-1) * pFS->nPagesize;
+            rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, p->aData, nByte);
+          }
+          pFS->nRead++;
+        }
+
+        /* If the xRead() call was successful (or not attempted), link the
+        ** page into the page-cache hash-table. Otherwise, if it failed,
+        ** free the buffer. */
+        if( rc==LSM_OK && nSpace==0 ){
+          p->pHashNext = pFS->apHash[iHash];
+          pFS->apHash[iHash] = p;
+        }else{
+          fsPageBufferFree(p);
+          p = 0;
+          if( pnSpace ) *pnSpace = nSpace;
+        }
+      }
+    }
+
+    assert( (rc==LSM_OK && (p || (pnSpace && *pnSpace)))
+         || (rc!=LSM_OK && p==0) 
+    );
+  }
+
+  if( rc==LSM_OK && p ){
+    if( pFS->pCompress==0 && (fsIsLast(pFS, iReal) || fsIsFirst(pFS, iReal)) ){
+      p->nData = pFS->nPagesize - 4;
+      if( fsIsFirst(pFS, iReal) && p->nRef==0 ){
+        p->aData += 4;
+        p->flags |= PAGE_HASPREV;
+      }
+    }else{
+      p->nData = pFS->nPagesize;
+    }
+    pFS->nOut += (p->nRef==0);
+    p->nRef++;
+  }
+  *ppPg = p;
+  return rc;
+}
+
+/*
+** Read the 64-bit checkpoint id of the checkpoint currently stored on meta
+** page iMeta of the database file. If no error occurs, store the id value
+** in *piVal and return LSM_OK. Otherwise, return an LSM error code and leave
+** *piVal unmodified.
+**
+** If a checkpointer connection is currently updating meta-page iMeta, or an
+** earlier checkpointer crashed while doing so, the value read into *piVal
+** may be garbage. It is the callers responsibility to deal with this.
+*/
+int lsmFsReadSyncedId(lsm_db *db, int iMeta, i64 *piVal){
+  FileSystem *pFS = db->pFS;
+  int rc = LSM_OK;
+
+  assert( iMeta==1 || iMeta==2 );
+  if( pFS->nMapLimit>0 ){
+    fsGrowMapping(pFS, iMeta*LSM_META_PAGE_SIZE, &rc);
+    if( rc==LSM_OK ){
+      *piVal = (i64)lsmGetU64(&((u8 *)pFS->pMap)[(iMeta-1)*LSM_META_PAGE_SIZE]);
+    }
+  }else{
+    MetaPage *pMeta = 0;
+    rc = lsmFsMetaPageGet(pFS, 0, iMeta, &pMeta);
+    if( rc==LSM_OK ){
+      *piVal = (i64)lsmGetU64(pMeta->aData);
+      lsmFsMetaPageRelease(pMeta);
+    }
+  }
+
+  return rc;
+}
+
+
+/*
+** Return true if the first or last page of segment pRun falls between iFirst
+** and iLast, inclusive, and pRun is not equal to pIgnore.
+*/
+static int fsRunEndsBetween(
+  Segment *pRun, 
+  Segment *pIgnore, 
+  LsmPgno iFirst, 
+  LsmPgno iLast
+){
+  return (pRun!=pIgnore && (
+        (pRun->iFirst>=iFirst && pRun->iFirst<=iLast)
+     || (pRun->iLastPg>=iFirst && pRun->iLastPg<=iLast)
+  ));
+}
+
+/*
+** Return true if level pLevel contains a segment other than pIgnore for
+** which the first or last page is between iFirst and iLast, inclusive.
+*/
+static int fsLevelEndsBetween(
+  Level *pLevel, 
+  Segment *pIgnore, 
+  LsmPgno iFirst, 
+  LsmPgno iLast
+){
+  int i;
+
+  if( fsRunEndsBetween(&pLevel->lhs, pIgnore, iFirst, iLast) ){
+    return 1;
+  }
+  for(i=0; i<pLevel->nRight; i++){
+    if( fsRunEndsBetween(&pLevel->aRhs[i], pIgnore, iFirst, iLast) ){
+      return 1;
+    }
+  }
+
+  return 0;
+}
+
+/*
+** Block iBlk is no longer in use by segment pIgnore. If it is not in use
+** by any other segment, move it to the free block list.
+*/
+static int fsFreeBlock(
+  FileSystem *pFS,                /* File system object */
+  Snapshot *pSnapshot,            /* Worker snapshot */
+  Segment *pIgnore,               /* Ignore this run when searching */
+  int iBlk                        /* Block number of block to free */
+){
+  int rc = LSM_OK;                /* Return code */
+  LsmPgno iFirst;                 /* First page on block iBlk */
+  LsmPgno iLast;                  /* Last page on block iBlk */
+  Level *pLevel;                  /* Used to iterate through levels */
+
+  int iIn;                        /* Used to iterate through append points */
+  int iOut = 0;                   /* Used to output append points */
+  LsmPgno *aApp = pSnapshot->aiAppend;
+
+  iFirst = fsFirstPageOnBlock(pFS, iBlk);
+  iLast = fsLastPageOnBlock(pFS, iBlk);
+
+  /* Check if any other run in the snapshot has a start or end page 
+  ** within this block. If there is such a run, return early. */
+  for(pLevel=lsmDbSnapshotLevel(pSnapshot); pLevel; pLevel=pLevel->pNext){
+    if( fsLevelEndsBetween(pLevel, pIgnore, iFirst, iLast) ){
+      return LSM_OK;
+    }
+  }
+
+  /* Remove any entries that lie on this block from the append-list. */
+  for(iIn=0; iIn<LSM_APPLIST_SZ; iIn++){
+    if( aApp[iIn]<iFirst || aApp[iIn]>iLast ){
+      aApp[iOut++] = aApp[iIn];
+    }
+  }
+  while( iOut<LSM_APPLIST_SZ ) aApp[iOut++] = 0;
+
+  if( rc==LSM_OK ){
+    rc = lsmBlockFree(pFS->pDb, iBlk);
+  }
+  return rc;
+}
+
+/*
+** Delete or otherwise recycle the blocks currently occupied by run pDel.
+*/
+int lsmFsSortedDelete(
+  FileSystem *pFS, 
+  Snapshot *pSnapshot,
+  int bZero,                      /* True to zero the Segment structure */
+  Segment *pDel
+){
+  if( pDel->iFirst ){
+    int rc = LSM_OK;
+
+    int iBlk;
+    int iLastBlk;
+
+    iBlk = fsPageToBlock(pFS, pDel->iFirst);
+    iLastBlk = fsPageToBlock(pFS, pDel->iLastPg);
+
+    /* Mark all blocks currently used by this sorted run as free */
+    while( iBlk && rc==LSM_OK ){
+      int iNext = 0;
+      if( iBlk!=iLastBlk ){
+        rc = fsBlockNext(pFS, pDel, iBlk, &iNext);
+      }else if( bZero==0 && pDel->iLastPg!=fsLastPageOnBlock(pFS, iLastBlk) ){
+        break;
+      }
+      rc = fsFreeBlock(pFS, pSnapshot, pDel, iBlk);
+      iBlk = iNext;
+    }
+
+    if( pDel->pRedirect ){
+      assert( pDel->pRedirect==&pSnapshot->redirect );
+      pSnapshot->redirect.n = 0;
+    }
+
+    if( bZero ) memset(pDel, 0, sizeof(Segment));
+  }
+  return LSM_OK;
+}
+
+/*
+** aPgno is an array containing nPgno page numbers. Return the smallest page
+** number from the array that falls on block iBlk. Or, if none of the pages
+** in aPgno[] fall on block iBlk, return 0.
+*/
+static LsmPgno firstOnBlock(
+  FileSystem *pFS, 
+  int iBlk, 
+  LsmPgno *aPgno, 
+  int nPgno
+){
+  LsmPgno iRet = 0;
+  int i;
+  for(i=0; i<nPgno; i++){
+    LsmPgno iPg = aPgno[i];
+    if( fsPageToBlock(pFS, iPg)==iBlk && (iRet==0 || iPg<iRet) ){
+      iRet = iPg;
+    }
+  }
+  return iRet;
+}
+
+#ifndef NDEBUG
+/*
+** Return true if page iPg, which is a part of segment p, lies on
+** a redirected block. 
+*/
+static int fsPageRedirects(FileSystem *pFS, Segment *p, LsmPgno iPg){
+  return (iPg!=0 && iPg!=lsmFsRedirectPage(pFS, p->pRedirect, iPg));
+}
+
+/*
+** Return true if the second argument is not NULL and any of the first
+** last or root pages lie on a redirected block. 
+*/
+static int fsSegmentRedirects(FileSystem *pFS, Segment *p){
+  return (p && (
+      fsPageRedirects(pFS, p, p->iFirst)
+   || fsPageRedirects(pFS, p, p->iRoot)
+   || fsPageRedirects(pFS, p, p->iLastPg)
+  ));
+}
+#endif
+
+/*
+** Argument aPgno is an array of nPgno page numbers. All pages belong to
+** the segment pRun. This function gobbles from the start of the run to the
+** first page that appears in aPgno[] (i.e. so that the aPgno[] entry is
+** the new first page of the run).
+*/
+void lsmFsGobble(
+  lsm_db *pDb,
+  Segment *pRun, 
+  LsmPgno *aPgno,
+  int nPgno
+){
+  int rc = LSM_OK;
+  FileSystem *pFS = pDb->pFS;
+  Snapshot *pSnapshot = pDb->pWorker;
+  int iBlk;
+
+  assert( pRun->nSize>0 );
+  assert( 0==fsSegmentRedirects(pFS, pRun) );
+  assert( nPgno>0 && 0==fsPageRedirects(pFS, pRun, aPgno[0]) );
+
+  iBlk = fsPageToBlock(pFS, pRun->iFirst);
+  pRun->nSize += (int)(pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));
+
+  while( rc==LSM_OK ){
+    int iNext = 0;
+    LsmPgno iFirst = firstOnBlock(pFS, iBlk, aPgno, nPgno);
+    if( iFirst ){
+      pRun->iFirst = iFirst;
+      break;
+    }
+    rc = fsBlockNext(pFS, pRun, iBlk, &iNext);
+    if( rc==LSM_OK ) rc = fsFreeBlock(pFS, pSnapshot, pRun, iBlk);
+    pRun->nSize -= (int)(
+        1 + fsLastPageOnBlock(pFS, iBlk) - fsFirstPageOnBlock(pFS, iBlk)
+    );
+    iBlk = iNext;
+  }
+
+  pRun->nSize -= (int)(pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));
+  assert( pRun->nSize>0 );
+}
+
+/*
+** This function is only used in compressed database mode.
+**
+** Argument iPg is the page number (byte offset) of a page within segment
+** pSeg. The page record, including all headers, is nByte bytes in size.
+** Before returning, set *piNext to the page number of the next page in
+** the segment, or to zero if iPg is the last.
+**
+** In other words, do:
+**
+**   *piNext = iPg + nByte;
+**
+** But take block overflow and redirection into account.
+*/
+static int fsNextPageOffset(
+  FileSystem *pFS,                /* File system object */
+  Segment *pSeg,                  /* Segment to move within */
+  LsmPgno iPg,                    /* Offset of current page */
+  int nByte,                      /* Size of current page including headers */
+  LsmPgno *piNext                 /* OUT: Offset of next page. Or zero (EOF) */
+){
+  LsmPgno iNext;
+  int rc;
+
+  assert( pFS->pCompress );
+
+  rc = fsAddOffset(pFS, pSeg, iPg, nByte-1, &iNext);
+  if( pSeg && iNext==pSeg->iLastPg ){
+    iNext = 0;
+  }else if( rc==LSM_OK ){
+    rc = fsAddOffset(pFS, pSeg, iNext, 1, &iNext);
+  }
+
+  *piNext = iNext;
+  return rc;
+}
+
+/*
+** This function is only used in compressed database mode.
+**
+** Argument iPg is the page number of a pagethat appears in segment pSeg.
+** This function determines the page number of the previous page in the
+** same run. *piPrev is set to the previous page number before returning.
+**
+** LSM_OK is returned if no error occurs. Otherwise, an lsm error code.
+** If any value other than LSM_OK is returned, then the final value of
+** *piPrev is undefined.
+*/
+static int fsGetPageBefore(
+  FileSystem *pFS, 
+  Segment *pSeg, 
+  LsmPgno iPg, 
+  LsmPgno *piPrev
+){
+  u8 aSz[3];
+  int rc;
+  i64 iRead;
+
+  assert( pFS->pCompress );
+
+  rc = fsSubtractOffset(pFS, pSeg, iPg, sizeof(aSz), &iRead);
+  if( rc==LSM_OK ) rc = fsReadData(pFS, pSeg, iRead, aSz, sizeof(aSz));
+
+  if( rc==LSM_OK ){
+    int bFree;
+    int nSz;
+    if( aSz[2] & 0x80 ){
+      nSz = getRecordSize(aSz, &bFree) + sizeof(aSz)*2;
+    }else{
+      nSz = (int)(aSz[2] & 0x7F);
+      bFree = 1;
+    }
+    rc = fsSubtractOffset(pFS, pSeg, iPg, nSz, piPrev);
+  }
+
+  return rc;
+}
+
+/*
+** The first argument to this function is a valid reference to a database
+** file page that is part of a sorted run. If parameter eDir is -1, this 
+** function attempts to locate and load the previous page in the same run. 
+** Or, if eDir is +1, it attempts to find the next page in the same run.
+** The results of passing an eDir value other than positive or negative one
+** are undefined.
+**
+** If parameter pRun is not NULL then it must point to the run that page
+** pPg belongs to. In this case, if pPg is the first or last page of the
+** run, and the request is for the previous or next page, respectively,
+** *ppNext is set to NULL before returning LSM_OK. If pRun is NULL, then it
+** is assumed that the next or previous page, as requested, exists.
+**
+** If the previous/next page does exist and is successfully loaded, *ppNext
+** is set to point to it and LSM_OK is returned. Otherwise, if an error 
+** occurs, *ppNext is set to NULL and and lsm error code returned.
+**
+** Page references returned by this function should be released by the 
+** caller using lsmFsPageRelease().
+*/
+int lsmFsDbPageNext(Segment *pRun, Page *pPg, int eDir, Page **ppNext){
+  int rc = LSM_OK;
+  FileSystem *pFS = pPg->pFS;
+  LsmPgno iPg = pPg->iPg;
+
+  assert( 0==fsSegmentRedirects(pFS, pRun) );
+  if( pFS->pCompress ){
+    int nSpace = pPg->nCompress + 2*3;
+
+    do {
+      if( eDir>0 ){
+        rc = fsNextPageOffset(pFS, pRun, iPg, nSpace, &iPg);
+      }else{
+        if( iPg==pRun->iFirst ){
+          iPg = 0;
+        }else{
+          rc = fsGetPageBefore(pFS, pRun, iPg, &iPg);
+        }
+      }
+
+      nSpace = 0;
+      if( iPg!=0 ){
+        rc = fsPageGet(pFS, pRun, iPg, 0, ppNext, &nSpace);
+        assert( (*ppNext==0)==(rc!=LSM_OK || nSpace>0) );
+      }else{
+        *ppNext = 0;
+      }
+    }while( nSpace>0 && rc==LSM_OK );
+
+  }else{
+    Redirect *pRedir = pRun ? pRun->pRedirect : 0;
+    assert( eDir==1 || eDir==-1 );
+    if( eDir<0 ){
+      if( pRun && iPg==pRun->iFirst ){
+        *ppNext = 0;
+        return LSM_OK;
+      }else if( fsIsFirst(pFS, iPg) ){
+        assert( pPg->flags & PAGE_HASPREV );
+        iPg = fsLastPageOnBlock(pFS, lsmGetU32(&pPg->aData[-4]));
+      }else{
+        iPg--;
+      }
+    }else{
+      if( pRun ){
+        if( iPg==pRun->iLastPg ){
+          *ppNext = 0;
+          return LSM_OK;
+        }
+      }
+
+      if( fsIsLast(pFS, iPg) ){
+        int iBlk = fsRedirectBlock(
+            pRedir, lsmGetU32(&pPg->aData[pFS->nPagesize-4])
+        );
+        iPg = fsFirstPageOnBlock(pFS, iBlk);
+      }else{
+        iPg++;
+      }
+    }
+    rc = fsPageGet(pFS, pRun, iPg, 0, ppNext, 0);
+  }
+
+  return rc;
+}
+
+/*
+** This function is called when creating a new segment to determine if the
+** first part of it can be written following an existing segment on an
+** already allocated block. If it is possible, the page number of the first
+** page to use for the new segment is returned. Otherwise zero.
+**
+** If argument pLvl is not NULL, then this function will not attempt to
+** start the new segment immediately following any segment that is part
+** of the right-hand-side of pLvl.
+*/
+static LsmPgno findAppendPoint(FileSystem *pFS, Level *pLvl){
+  int i;
+  LsmPgno *aiAppend = pFS->pDb->pWorker->aiAppend;
+  LsmPgno iRet = 0;
+
+  for(i=LSM_APPLIST_SZ-1; iRet==0 && i>=0; i--){
+    if( (iRet = aiAppend[i]) ){
+      if( pLvl ){
+        int iBlk = fsPageToBlock(pFS, iRet);
+        int j;
+        for(j=0; iRet && j<pLvl->nRight; j++){
+          if( fsPageToBlock(pFS, pLvl->aRhs[j].iLastPg)==iBlk ){
+            iRet = 0;
+          }
+        }
+      }
+      if( iRet ) aiAppend[i] = 0;
+    }
+  }
+  return iRet;
+}
+
+/*
+** Append a page to the left-hand-side of pLvl. Set the ref-count to 1 and
+** return a pointer to it. The page is writable until either 
+** lsmFsPagePersist() is called on it or the ref-count drops to zero.
+*/
+int lsmFsSortedAppend(
+  FileSystem *pFS, 
+  Snapshot *pSnapshot,
+  Level *pLvl,
+  int bDefer,
+  Page **ppOut
+){
+  int rc = LSM_OK;
+  Page *pPg = 0;
+  LsmPgno iApp = 0;
+  LsmPgno iNext = 0;
+  Segment *p = &pLvl->lhs;
+  LsmPgno iPrev = p->iLastPg;
+
+  *ppOut = 0;
+  assert( p->pRedirect==0 );
+
+  if( pFS->pCompress || bDefer ){
+    /* In compressed database mode the page is not assigned a page number
+    ** or location in the database file at this point. This will be done
+    ** by the lsmFsPagePersist() call.  */
+    rc = fsPageBuffer(pFS, &pPg);
+    if( rc==LSM_OK ){
+      pPg->pFS = pFS;
+      pPg->pSeg = p;
+      pPg->iPg = 0;
+      pPg->flags |= PAGE_DIRTY;
+      pPg->nData = pFS->nPagesize;
+      assert( pPg->aData );
+      if( pFS->pCompress==0 ) pPg->nData -= 4;
+
+      pPg->nRef = 1;
+      pFS->nOut++;
+    }
+  }else{
+    if( iPrev==0 ){
+      iApp = findAppendPoint(pFS, pLvl);
+    }else if( fsIsLast(pFS, iPrev) ){
+      int iNext2;
+      rc = fsBlockNext(pFS, 0, fsPageToBlock(pFS, iPrev), &iNext2);
+      if( rc!=LSM_OK ) return rc;
+      iApp = fsFirstPageOnBlock(pFS, iNext2);
+    }else{
+      iApp = iPrev + 1;
+    }
+
+    /* If this is the first page allocated, or if the page allocated is the
+    ** last in the block, also allocate the next block here.  */
+    if( iApp==0 || fsIsLast(pFS, iApp) ){
+      int iNew;                     /* New block number */
+
+      rc = lsmBlockAllocate(pFS->pDb, 0, &iNew);
+      if( rc!=LSM_OK ) return rc;
+      if( iApp==0 ){
+        iApp = fsFirstPageOnBlock(pFS, iNew);
+      }else{
+        iNext = fsFirstPageOnBlock(pFS, iNew);
+      }
+    }
+
+    /* Grab the new page. */
+    pPg = 0;
+    rc = fsPageGet(pFS, 0, iApp, 1, &pPg, 0);
+    assert( rc==LSM_OK || pPg==0 );
+
+    /* If this is the first or last page of a block, fill in the pointer 
+     ** value at the end of the new page. */
+    if( rc==LSM_OK ){
+      p->nSize++;
+      p->iLastPg = iApp;
+      if( p->iFirst==0 ) p->iFirst = iApp;
+      pPg->flags |= PAGE_DIRTY;
+
+      if( fsIsLast(pFS, iApp) ){
+        lsmPutU32(&pPg->aData[pFS->nPagesize-4], fsPageToBlock(pFS, iNext));
+      }else if( fsIsFirst(pFS, iApp) ){
+        lsmPutU32(&pPg->aData[-4], fsPageToBlock(pFS, iPrev));
+      }
+    }
+  }
+
+  *ppOut = pPg;
+  return rc;
+}
+
+/*
+** Mark the segment passed as the second argument as finished. Once a segment
+** is marked as finished it is not possible to append any further pages to 
+** it.
+**
+** Return LSM_OK if successful or an lsm error code if an error occurs.
+*/
+int lsmFsSortedFinish(FileSystem *pFS, Segment *p){
+  int rc = LSM_OK;
+  if( p && p->iLastPg ){
+    assert( p->pRedirect==0 );
+
+    /* Check if the last page of this run happens to be the last of a block.
+    ** If it is, then an extra block has already been allocated for this run.
+    ** Shift this extra block back to the free-block list. 
+    **
+    ** Otherwise, add the first free page in the last block used by the run
+    ** to the lAppend list.
+    */
+    if( fsLastPageOnPagesBlock(pFS, p->iLastPg)!=p->iLastPg ){
+      int i;
+      LsmPgno *aiAppend = pFS->pDb->pWorker->aiAppend;
+      for(i=0; i<LSM_APPLIST_SZ; i++){
+        if( aiAppend[i]==0 ){
+          aiAppend[i] = p->iLastPg+1;
+          break;
+        }
+      }
+    }else if( pFS->pCompress==0 ){
+      Page *pLast;
+      rc = fsPageGet(pFS, 0, p->iLastPg, 0, &pLast, 0);
+      if( rc==LSM_OK ){
+        int iBlk = (int)lsmGetU32(&pLast->aData[pFS->nPagesize-4]);
+        lsmBlockRefree(pFS->pDb, iBlk);
+        lsmFsPageRelease(pLast);
+      }
+    }else{
+      int iBlk = 0;
+      rc = fsBlockNext(pFS, p, fsPageToBlock(pFS, p->iLastPg), &iBlk);
+      if( rc==LSM_OK ){
+        lsmBlockRefree(pFS->pDb, iBlk);
+      }
+    }
+  }
+  return rc;
+}
+
+/*
+** Obtain a reference to page number iPg.
+**
+** Return LSM_OK if successful, or an lsm error code if an error occurs.
+*/
+int lsmFsDbPageGet(FileSystem *pFS, Segment *pSeg, LsmPgno iPg, Page **ppPg){
+  return fsPageGet(pFS, pSeg, iPg, 0, ppPg, 0);
+}
+
+/*
+** Obtain a reference to the last page in the segment passed as the 
+** second argument.
+**
+** Return LSM_OK if successful, or an lsm error code if an error occurs.
+*/
+int lsmFsDbPageLast(FileSystem *pFS, Segment *pSeg, Page **ppPg){
+  int rc;
+  LsmPgno iPg = pSeg->iLastPg;
+  if( pFS->pCompress ){
+    int nSpace;
+    iPg++;
+    do {
+      nSpace = 0;
+      rc = fsGetPageBefore(pFS, pSeg, iPg, &iPg);
+      if( rc==LSM_OK ){
+        rc = fsPageGet(pFS, pSeg, iPg, 0, ppPg, &nSpace);
+      }
+    }while( rc==LSM_OK && nSpace>0 );
+
+  }else{
+    rc = fsPageGet(pFS, pSeg, iPg, 0, ppPg, 0);
+  }
+  return rc;
+}
+
+/*
+** Return a reference to meta-page iPg. If successful, LSM_OK is returned
+** and *ppPg populated with the new page reference. The reference should
+** be released by the caller using lsmFsPageRelease().
+**
+** Otherwise, if an error occurs, *ppPg is set to NULL and an LSM error 
+** code is returned.
+*/
+int lsmFsMetaPageGet(
+  FileSystem *pFS,                /* File-system connection */
+  int bWrite,                     /* True for write access, false for read */
+  int iPg,                        /* Either 1 or 2 */
+  MetaPage **ppPg                 /* OUT: Pointer to MetaPage object */
+){
+  int rc = LSM_OK;
+  MetaPage *pPg;
+  assert( iPg==1 || iPg==2 );
+
+  pPg = lsmMallocZeroRc(pFS->pEnv, sizeof(Page), &rc);
+
+  if( pPg ){
+    i64 iOff = (iPg-1) * pFS->nMetasize;
+    if( pFS->nMapLimit>0 ){
+      fsGrowMapping(pFS, 2*pFS->nMetasize, &rc);
+      pPg->aData = (u8 *)(pFS->pMap) + iOff;
+    }else{
+      pPg->aData = lsmMallocRc(pFS->pEnv, pFS->nMetasize, &rc);
+      if( rc==LSM_OK && bWrite==0 ){
+        rc = lsmEnvRead(
+            pFS->pEnv, pFS->fdDb, iOff, pPg->aData, pFS->nMetaRwSize
+        );
+      }
+#ifndef NDEBUG
+      /* pPg->aData causes an uninitialized access via a downstreadm write().
+         After discussion on this list, this memory should not, for performance
+         reasons, be memset. However, tracking down "real" misuse is more
+         difficult with this "false" positive, so it is set when NDEBUG.
+      */
+      else if( rc==LSM_OK ){
+        memset( pPg->aData, 0x77, pFS->nMetasize );
+      }
+#endif
+    }
+
+    if( rc!=LSM_OK ){
+      if( pFS->nMapLimit==0 ) lsmFree(pFS->pEnv, pPg->aData);
+      lsmFree(pFS->pEnv, pPg);
+      pPg = 0;
+    }else{
+      pPg->iPg = iPg;
+      pPg->bWrite = bWrite;
+      pPg->pFS = pFS;
+    }
+  }
+
+  *ppPg = pPg;
+  return rc;
+}
+
+/*
+** Release a meta-page reference obtained via a call to lsmFsMetaPageGet().
+*/
+int lsmFsMetaPageRelease(MetaPage *pPg){
+  int rc = LSM_OK;
+  if( pPg ){
+    FileSystem *pFS = pPg->pFS;
+
+    if( pFS->nMapLimit==0 ){
+      if( pPg->bWrite ){
+        i64 iOff = (pPg->iPg==2 ? pFS->nMetasize : 0);
+        int nWrite = pFS->nMetaRwSize;
+        rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iOff, pPg->aData, nWrite);
+      }
+      lsmFree(pFS->pEnv, pPg->aData);
+    }
+
+    lsmFree(pFS->pEnv, pPg);
+  }
+  return rc;
+}
+
+/*
+** Return a pointer to a buffer containing the data associated with the
+** meta-page passed as the first argument. If parameter pnData is not NULL,
+** set *pnData to the size of the meta-page in bytes before returning.
+*/
+u8 *lsmFsMetaPageData(MetaPage *pPg, int *pnData){
+  if( pnData ) *pnData = pPg->pFS->nMetaRwSize;
+  return pPg->aData;
+}
+
+/*
+** Return true if page is currently writable. This is used in assert() 
+** statements only.
+*/
+#ifndef NDEBUG
+int lsmFsPageWritable(Page *pPg){
+  return (pPg->flags & PAGE_DIRTY) ? 1 : 0;
+}
+#endif
+
+/*
+** This is called when block iFrom is being redirected to iTo. If page 
+** number (*piPg) lies on block iFrom, then calculate the equivalent
+** page on block iTo and set *piPg to this value before returning.
+*/
+static void fsMovePage(
+  FileSystem *pFS,                /* File system object */
+  int iTo,                        /* Destination block */
+  int iFrom,                      /* Source block */
+  LsmPgno *piPg                   /* IN/OUT: Page number */
+){
+  LsmPgno iPg = *piPg;
+  if( iFrom==fsPageToBlock(pFS, iPg) ){
+    const int nPagePerBlock = (
+        pFS->pCompress ? pFS ->nBlocksize : (pFS->nBlocksize / pFS->nPagesize)
+    );
+    *piPg = iPg - (LsmPgno)(iFrom - iTo) * nPagePerBlock;
+  }
+}
+
+/*
+** Copy the contents of block iFrom to block iTo. 
+**
+** It is safe to assume that there are no outstanding references to pages 
+** on block iTo. And that block iFrom is not currently being written. In
+** other words, the data can be read and written directly.
+*/
+int lsmFsMoveBlock(FileSystem *pFS, Segment *pSeg, int iTo, int iFrom){
+  Snapshot *p = pFS->pDb->pWorker;
+  int rc = LSM_OK;
+  int i;
+  i64 nMap;
+
+  i64 iFromOff = (i64)(iFrom-1) * pFS->nBlocksize;
+  i64 iToOff = (i64)(iTo-1) * pFS->nBlocksize;
+  
+  assert( iTo!=1 );
+  assert( iFrom>iTo );
+
+  /* Grow the mapping as required. */
+  nMap = LSM_MIN(pFS->nMapLimit, (i64)iFrom * pFS->nBlocksize);
+  fsGrowMapping(pFS, nMap, &rc);
+
+  if( rc==LSM_OK ){
+    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
+    int nSz = pFS->nPagesize;
+    u8 *aBuf = 0;
+    u8 *aData = 0;
+
+    for(i=0; rc==LSM_OK && i<nPagePerBlock; i++){
+      i64 iOff = iFromOff + i*nSz;
+
+      /* Set aData to point to a buffer containing the from page */
+      if( (iOff+nSz)<=pFS->nMapLimit ){
+        u8 *aMap = (u8 *)(pFS->pMap);
+        aData = &aMap[iOff];
+      }else{
+        if( aBuf==0 ){
+          aBuf = (u8 *)lsmMallocRc(pFS->pEnv, nSz, &rc);
+          if( aBuf==0 ) break;
+        }
+        aData = aBuf;
+        rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aData, nSz);
+      }
+
+      /* Copy aData to the to page */
+      if( rc==LSM_OK ){
+        iOff = iToOff + i*nSz;
+        if( (iOff+nSz)<=pFS->nMapLimit ){
+          u8 *aMap = (u8 *)(pFS->pMap);
+          memcpy(&aMap[iOff], aData, nSz);
+        }else{
+          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iOff, aData, nSz);
+        }
+      }
+    }
+    lsmFree(pFS->pEnv, aBuf);
+    lsmFsPurgeCache(pFS);
+  }
+
+  /* Update append-point list if necessary */
+  for(i=0; i<LSM_APPLIST_SZ; i++){
+    fsMovePage(pFS, iTo, iFrom, &p->aiAppend[i]);
+  }
+
+  /* Update the Segment structure itself */
+  fsMovePage(pFS, iTo, iFrom, &pSeg->iFirst);
+  fsMovePage(pFS, iTo, iFrom, &pSeg->iLastPg);
+  fsMovePage(pFS, iTo, iFrom, &pSeg->iRoot);
+
+  return rc;
+}
+
+/*
+** Append raw data to a segment. Return the database file offset that the
+** data is written to (this may be used as the page number if the data
+** being appended is a new page record).
+**
+** This function is only used in compressed database mode.
+*/
+static LsmPgno fsAppendData(
+  FileSystem *pFS,                /* File-system handle */
+  Segment *pSeg,                  /* Segment to append to */
+  const u8 *aData,                /* Buffer containing data to write */
+  int nData,                      /* Size of buffer aData[] in bytes */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  LsmPgno iRet = 0;
+  int rc = *pRc;
+  assert( pFS->pCompress );
+  if( rc==LSM_OK ){
+    int nRem = 0;
+    int nWrite = 0;
+    LsmPgno iLastOnBlock;
+    LsmPgno iApp = pSeg->iLastPg+1;
+
+    /* If this is the first data written into the segment, find an append-point
+    ** or allocate a new block.  */
+    if( iApp==1 ){
+      pSeg->iFirst = iApp = findAppendPoint(pFS, 0);
+      if( iApp==0 ){
+        int iBlk;
+        rc = lsmBlockAllocate(pFS->pDb, 0, &iBlk);
+        pSeg->iFirst = iApp = fsFirstPageOnBlock(pFS, iBlk);
+      }
+    }
+    iRet = iApp;
+
+    /* Write as much data as is possible at iApp (usually all of it). */
+    iLastOnBlock = fsLastPageOnPagesBlock(pFS, iApp);
+    if( rc==LSM_OK ){
+      int nSpace = (int)(iLastOnBlock - iApp + 1);
+      nWrite = LSM_MIN(nData, nSpace);
+      nRem = nData - nWrite;
+      assert( nWrite>=0 );
+      if( nWrite!=0 ){
+        rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, aData, nWrite);
+      }
+      iApp += nWrite;
+    }
+
+    /* If required, allocate a new block and write the rest of the data
+    ** into it. Set the next and previous block pointers to link the new
+    ** block to the old.  */
+    assert( nRem<=0 || (iApp-1)==iLastOnBlock );
+    if( rc==LSM_OK && (iApp-1)==iLastOnBlock ){
+      u8 aPtr[4];                 /* Space to serialize a u32 */
+      int iBlk;                   /* New block number */
+
+      if( nWrite>0 ){
+        /* Allocate a new block. */
+        rc = lsmBlockAllocate(pFS->pDb, 0, &iBlk);
+
+        /* Set the "next" pointer on the old block */
+        if( rc==LSM_OK ){
+          assert( iApp==(fsPageToBlock(pFS, iApp)*pFS->nBlocksize)-4 );
+          lsmPutU32(aPtr, iBlk);
+          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, aPtr, sizeof(aPtr));
+        }
+
+        /* Set the "prev" pointer on the new block */
+        if( rc==LSM_OK ){
+          LsmPgno iWrite;
+          lsmPutU32(aPtr, fsPageToBlock(pFS, iApp));
+          iWrite = fsFirstPageOnBlock(pFS, iBlk);
+          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iWrite-4, aPtr, sizeof(aPtr));
+          if( nRem>0 ) iApp = iWrite;
+        }
+      }else{
+        /* The next block is already allocated. */
+        assert( nRem>0 );
+        assert( pSeg->pRedirect==0 );
+        rc = fsBlockNext(pFS, 0, fsPageToBlock(pFS, iApp), &iBlk);
+        iRet = iApp = fsFirstPageOnBlock(pFS, iBlk);
+      }
+
+      /* Write the remaining data into the new block */
+      if( rc==LSM_OK && nRem>0 ){
+        rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, &aData[nWrite], nRem);
+        iApp += nRem;
+      }
+    }
+
+    pSeg->iLastPg = iApp-1;
+    *pRc = rc;
+  }
+
+  return iRet;
+}
+
+/*
+** This function is only called in compressed database mode. It 
+** compresses the contents of page pPg and writes the result to the 
+** buffer at pFS->aOBuffer. The size of the compressed data is stored in
+** pPg->nCompress.
+**
+** If buffer pFS->aOBuffer[] has not been allocated then this function
+** allocates it. If this fails, LSM_NOMEM is returned. Otherwise, LSM_OK.
+*/
+static int fsCompressIntoBuffer(FileSystem *pFS, Page *pPg){
+  lsm_compress *p = pFS->pCompress;
+
+  if( fsAllocateBuffer(pFS, 1) ) return LSM_NOMEM;
+  assert( pPg->nData==pFS->nPagesize );
+
+  pPg->nCompress = pFS->nBuffer;
+  return p->xCompress(p->pCtx, 
+      (char *)pFS->aOBuffer, &pPg->nCompress, 
+      (const char *)pPg->aData, pPg->nData
+  );
+}
+
+/*
+** Append a new page to segment pSeg. Set output variable *piNew to the
+** page number of the new page before returning.
+**
+** If the new page is the last on its block, then the 'next' block that
+** will be used by the segment is allocated here too. In this case output
+** variable *piNext is set to the block number of the next block.
+**
+** If the new page is the first on its block but not the first in the
+** entire segment, set output variable *piPrev to the block number of
+** the previous block in the segment.
+**
+** LSM_OK is returned if successful, or an lsm error code otherwise. If
+** any value other than LSM_OK is returned, then the final value of all
+** output variables is undefined.
+*/
+static int fsAppendPage(
+  FileSystem *pFS, 
+  Segment *pSeg,
+  LsmPgno *piNew,
+  int *piPrev,
+  int *piNext
+){
+  LsmPgno iPrev = pSeg->iLastPg;
+  int rc;
+  assert( iPrev!=0 );
+
+  *piPrev = 0;
+  *piNext = 0;
+
+  if( fsIsLast(pFS, iPrev) ){
+    /* Grab the first page on the next block (which has already be
+    ** allocated). In this case set *piPrev to tell the caller to set
+    ** the "previous block" pointer in the first 4 bytes of the page.
+    */
+    int iNext;
+    int iBlk = fsPageToBlock(pFS, iPrev);
+    assert( pSeg->pRedirect==0 );
+    rc = fsBlockNext(pFS, 0, iBlk, &iNext);
+    if( rc!=LSM_OK ) return rc;
+    *piNew = fsFirstPageOnBlock(pFS, iNext);
+    *piPrev = iBlk;
+  }else{
+    *piNew = iPrev+1;
+    if( fsIsLast(pFS, *piNew) ){
+      /* Allocate the next block here. */
+      int iBlk;
+      rc = lsmBlockAllocate(pFS->pDb, 0, &iBlk);
+      if( rc!=LSM_OK ) return rc;
+      *piNext = iBlk;
+    }
+  }
+
+  pSeg->nSize++;
+  pSeg->iLastPg = *piNew;
+  return LSM_OK;
+}
+
+/*
+** Flush all pages in the FileSystem.pWaiting list to disk.
+*/
+void lsmFsFlushWaiting(FileSystem *pFS, int *pRc){
+  int rc = *pRc;
+  Page *pPg;
+
+  pPg = pFS->pWaiting;
+  pFS->pWaiting = 0;
+
+  while( pPg ){
+    Page *pNext = pPg->pWaitingNext;
+    if( rc==LSM_OK ) rc = lsmFsPagePersist(pPg);
+    assert( pPg->nRef==1 );
+    lsmFsPageRelease(pPg);
+    pPg = pNext;
+  }
+  *pRc = rc;
+}
+
+/*
+** If there exists a hash-table entry associated with page iPg, remove it.
+*/
+static void fsRemoveHashEntry(FileSystem *pFS, LsmPgno iPg){
+  Page *p;
+  int iHash = fsHashKey(pFS->nHash, iPg);
+
+  for(p=pFS->apHash[iHash]; p && p->iPg!=iPg; p=p->pHashNext);
+
+  if( p ){
+    assert( p->nRef==0 || (p->flags & PAGE_FREE)==0 );
+    fsPageRemoveFromHash(pFS, p);
+    p->iPg = 0;
+    iHash = fsHashKey(pFS->nHash, 0);
+    p->pHashNext = pFS->apHash[iHash];
+    pFS->apHash[iHash] = p;
+  }
+}
+
+/*
+** If the page passed as an argument is dirty, update the database file
+** (or mapping of the database file) with its current contents and mark
+** the page as clean.
+**
+** Return LSM_OK if the operation is a success, or an LSM error code
+** otherwise.
+*/
+int lsmFsPagePersist(Page *pPg){
+  int rc = LSM_OK;
+  if( pPg && (pPg->flags & PAGE_DIRTY) ){
+    FileSystem *pFS = pPg->pFS;
+
+    if( pFS->pCompress ){
+      int iHash;                  /* Hash key of assigned page number */
+      u8 aSz[3];                  /* pPg->nCompress as a 24-bit big-endian */
+      assert( pPg->pSeg && pPg->iPg==0 && pPg->nCompress==0 );
+
+      /* Compress the page image. */
+      rc = fsCompressIntoBuffer(pFS, pPg);
+
+      /* Serialize the compressed size into buffer aSz[] */
+      putRecordSize(aSz, pPg->nCompress, 0);
+
+      /* Write the serialized page record into the database file. */
+      pPg->iPg = fsAppendData(pFS, pPg->pSeg, aSz, sizeof(aSz), &rc);
+      fsAppendData(pFS, pPg->pSeg, pFS->aOBuffer, pPg->nCompress, &rc);
+      fsAppendData(pFS, pPg->pSeg, aSz, sizeof(aSz), &rc);
+
+      /* Now that it has a page number, insert the page into the hash table */
+      iHash = fsHashKey(pFS->nHash, pPg->iPg);
+      pPg->pHashNext = pFS->apHash[iHash];
+      pFS->apHash[iHash] = pPg;
+
+      pPg->pSeg->nSize += (sizeof(aSz) * 2) + pPg->nCompress;
+
+      pPg->flags &= ~PAGE_DIRTY;
+      pFS->nWrite++;
+    }else{
+
+      if( pPg->iPg==0 ){
+        /* No page number has been assigned yet. This occurs with pages used
+        ** in the b-tree hierarchy. They were not assigned page numbers when
+        ** they were created as doing so would cause this call to
+        ** lsmFsPagePersist() to write an out-of-order page. Instead a page 
+        ** number is assigned here so that the page data will be appended
+        ** to the current segment.
+        */
+        Page **pp;
+        int iPrev = 0;
+        int iNext = 0;
+        int iHash;
+
+        assert( pPg->pSeg->iFirst );
+        assert( pPg->flags & PAGE_FREE );
+        assert( (pPg->flags & PAGE_HASPREV)==0 );
+        assert( pPg->nData==pFS->nPagesize-4 );
+
+        rc = fsAppendPage(pFS, pPg->pSeg, &pPg->iPg, &iPrev, &iNext);
+        if( rc!=LSM_OK ) return rc;
+
+        assert( pPg->flags & PAGE_FREE );
+        iHash = fsHashKey(pFS->nHash, pPg->iPg);
+        fsRemoveHashEntry(pFS, pPg->iPg);
+        pPg->pHashNext = pFS->apHash[iHash];
+        pFS->apHash[iHash] = pPg;
+        assert( pPg->pHashNext==0 || pPg->pHashNext->iPg!=pPg->iPg );
+
+        if( iPrev ){
+          assert( iNext==0 );
+          memmove(&pPg->aData[4], pPg->aData, pPg->nData);
+          lsmPutU32(pPg->aData, iPrev);
+          pPg->flags |= PAGE_HASPREV;
+          pPg->aData += 4;
+        }else if( iNext ){
+          assert( iPrev==0 );
+          lsmPutU32(&pPg->aData[pPg->nData], iNext);
+        }else{
+          int nData = pPg->nData;
+          pPg->nData += 4;
+          lsmSortedExpandBtreePage(pPg, nData);
+        }
+
+        pPg->nRef++;
+        for(pp=&pFS->pWaiting; *pp; pp=&(*pp)->pWaitingNext);
+        *pp = pPg;
+        assert( pPg->pWaitingNext==0 );
+
+      }else{
+        i64 iOff;                   /* Offset to write within database file */
+
+        iOff = (i64)pFS->nPagesize * (i64)(pPg->iPg-1);
+        if( fsMmapPage(pFS, pPg->iPg)==0 ){
+          u8 *aData = pPg->aData - (pPg->flags & PAGE_HASPREV);
+          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iOff, aData, pFS->nPagesize);
+        }else if( pPg->flags & PAGE_FREE ){
+          fsGrowMapping(pFS, iOff + pFS->nPagesize, &rc);
+          if( rc==LSM_OK ){
+            u8 *aTo = &((u8 *)(pFS->pMap))[iOff];
+            u8 *aFrom = pPg->aData - (pPg->flags & PAGE_HASPREV);
+            memcpy(aTo, aFrom, pFS->nPagesize);
+            lsmFree(pFS->pEnv, aFrom);
+            pFS->nCacheAlloc--;
+            pPg->aData = aTo + (pPg->flags & PAGE_HASPREV);
+            pPg->flags &= ~PAGE_FREE;
+            fsPageRemoveFromHash(pFS, pPg);
+            pPg->pMappedNext = pFS->pMapped;
+            pFS->pMapped = pPg;
+          }
+        }
+
+        lsmFsFlushWaiting(pFS, &rc);
+        pPg->flags &= ~PAGE_DIRTY;
+        pFS->nWrite++;
+      }
+    }
+  }
+
+  return rc;
+}
+
+/*
+** For non-compressed databases, this function is a no-op. For compressed
+** databases, it adds a padding record to the segment passed as the third
+** argument.
+**
+** The size of the padding records is selected so that the last byte 
+** written is the last byte of a disk sector. This means that if a 
+** snapshot is taken and checkpointed, subsequent worker processes will
+** not write to any sector that contains checkpointed data.
+*/
+int lsmFsSortedPadding(
+  FileSystem *pFS, 
+  Snapshot *pSnapshot,
+  Segment *pSeg
+){
+  int rc = LSM_OK;
+  if( pFS->pCompress && pSeg->iFirst ){
+    LsmPgno iLast2;
+    LsmPgno iLast = pSeg->iLastPg;  /* Current last page of segment */
+    int nPad;                       /* Bytes of padding required */
+    u8 aSz[3];
+
+    iLast2 = (1 + iLast/pFS->szSector) * pFS->szSector - 1;
+    assert( fsPageToBlock(pFS, iLast)==fsPageToBlock(pFS, iLast2) );
+    nPad = (int)(iLast2 - iLast);
+
+    if( iLast2>fsLastPageOnPagesBlock(pFS, iLast) ){
+      nPad -= 4;
+    }
+    assert( nPad>=0 );
+
+    if( nPad>=6 ){
+      pSeg->nSize += nPad;
+      nPad -= 6;
+      putRecordSize(aSz, nPad, 1);
+      fsAppendData(pFS, pSeg, aSz, sizeof(aSz), &rc);
+      memset(pFS->aOBuffer, 0, nPad);
+      fsAppendData(pFS, pSeg, pFS->aOBuffer, nPad, &rc);
+      fsAppendData(pFS, pSeg, aSz, sizeof(aSz), &rc);
+    }else if( nPad>0 ){
+      u8 aBuf[5] = {0,0,0,0,0};
+      aBuf[0] = (u8)nPad;
+      aBuf[nPad-1] = (u8)nPad;
+      fsAppendData(pFS, pSeg, aBuf, nPad, &rc);
+    }
+
+    assert( rc!=LSM_OK 
+        || pSeg->iLastPg==fsLastPageOnPagesBlock(pFS, pSeg->iLastPg)
+        || ((pSeg->iLastPg + 1) % pFS->szSector)==0
+    );
+  }
+
+  return rc;
+}
+
+
+/*
+** Increment the reference count on the page object passed as the first
+** argument.
+*/
+void lsmFsPageRef(Page *pPg){
+  if( pPg ){
+    pPg->nRef++;
+  }
+}
+
+/*
+** Release a page-reference obtained using fsPageGet().
+*/
+int lsmFsPageRelease(Page *pPg){
+  int rc = LSM_OK;
+  if( pPg ){
+    assert( pPg->nRef>0 );
+    pPg->nRef--;
+    if( pPg->nRef==0 ){
+      FileSystem *pFS = pPg->pFS;
+      rc = lsmFsPagePersist(pPg);
+      pFS->nOut--;
+
+      assert( pPg->pFS->pCompress 
+           || fsIsFirst(pPg->pFS, pPg->iPg)==0 
+           || (pPg->flags & PAGE_HASPREV)
+      );
+      pPg->aData -= (pPg->flags & PAGE_HASPREV);
+      pPg->flags &= ~PAGE_HASPREV;
+
+      if( (pPg->flags & PAGE_FREE)==0 ){
+        /* Removed from mapped list */
+        Page **pp;
+        for(pp=&pFS->pMapped; (*pp)!=pPg; pp=&(*pp)->pMappedNext);
+        *pp = pPg->pMappedNext;
+        pPg->pMappedNext = 0;
+
+        /* Add to free list */
+        pPg->pFreeNext = pFS->pFree;
+        pFS->pFree = pPg;
+      }else{
+        fsPageAddToLru(pFS, pPg);
+      }
+    }
+  }
+
+  return rc;
+}
+
+/*
+** Return the total number of pages read from the database file.
+*/
+int lsmFsNRead(FileSystem *pFS){ return pFS->nRead; }
+
+/*
+** Return the total number of pages written to the database file.
+*/
+int lsmFsNWrite(FileSystem *pFS){ return pFS->nWrite; }
+
+/*
+** Return a copy of the environment pointer used by the file-system object.
+*/
+lsm_env *lsmFsEnv(FileSystem *pFS){ 
+  return pFS->pEnv; 
+}
+
+/*
+** Return a copy of the environment pointer used by the file-system object
+** to which this page belongs.
+*/
+lsm_env *lsmPageEnv(Page *pPg) { 
+  return pPg->pFS->pEnv; 
+}
+
+/*
+** Return a pointer to the file-system object associated with the Page
+** passed as the only argument.
+*/
+FileSystem *lsmPageFS(Page *pPg){
+  return pPg->pFS;
+}
+
+/*
+** Return the sector-size as reported by the log file handle.
+*/
+int lsmFsSectorSize(FileSystem *pFS){
+  return pFS->szSector;
+}
+
+/*
+** Helper function for lsmInfoArrayStructure().
+*/
+static Segment *startsWith(Segment *pRun, LsmPgno iFirst){
+  return (iFirst==pRun->iFirst) ? pRun : 0;
+}
+
+/*
+** Return the segment that starts with page iFirst, if any. If no such segment
+** can be found, return NULL.
+*/
+static Segment *findSegment(Snapshot *pWorker, LsmPgno iFirst){
+  Level *pLvl;                    /* Used to iterate through db levels */
+  Segment *pSeg = 0;              /* Pointer to segment to return */
+
+  for(pLvl=lsmDbSnapshotLevel(pWorker); pLvl && pSeg==0; pLvl=pLvl->pNext){
+    if( 0==(pSeg = startsWith(&pLvl->lhs, iFirst)) ){
+      int i;
+      for(i=0; i<pLvl->nRight; i++){
+        if( (pSeg = startsWith(&pLvl->aRhs[i], iFirst)) ) break;
+      }
+    }
+  }
+
+  return pSeg;
+}
+
+/*
+** This function implements the lsm_info(LSM_INFO_ARRAY_STRUCTURE) request.
+** If successful, *pzOut is set to point to a nul-terminated string 
+** containing the array structure and LSM_OK is returned. The caller should
+** eventually free the string using lsmFree().
+**
+** If an error occurs, *pzOut is set to NULL and an LSM error code returned.
+*/
+int lsmInfoArrayStructure(
+  lsm_db *pDb, 
+  int bBlock,                     /* True for block numbers only */
+  LsmPgno iFirst,
+  char **pzOut
+){
+  int rc = LSM_OK;
+  Snapshot *pWorker;              /* Worker snapshot */
+  Segment *pArray = 0;            /* Array to report on */
+  int bUnlock = 0;
+
+  *pzOut = 0;
+  if( iFirst==0 ) return LSM_ERROR;
+
+  /* Obtain the worker snapshot */
+  pWorker = pDb->pWorker;
+  if( !pWorker ){
+    rc = lsmBeginWork(pDb);
+    if( rc!=LSM_OK ) return rc;
+    pWorker = pDb->pWorker;
+    bUnlock = 1;
+  }
+
+  /* Search for the array that starts on page iFirst */
+  pArray = findSegment(pWorker, iFirst);
+
+  if( pArray==0 ){
+    /* Could not find the requested array. This is an error. */
+    rc = LSM_ERROR;
+  }else{
+    FileSystem *pFS = pDb->pFS;
+    LsmString str;
+    int iBlk;
+    int iLastBlk;
+   
+    iBlk = fsPageToBlock(pFS, pArray->iFirst);
+    iLastBlk = fsPageToBlock(pFS, pArray->iLastPg);
+
+    lsmStringInit(&str, pDb->pEnv);
+    if( bBlock ){
+      lsmStringAppendf(&str, "%d", iBlk);
+      while( iBlk!=iLastBlk ){
+        fsBlockNext(pFS, pArray, iBlk, &iBlk);
+        lsmStringAppendf(&str, " %d", iBlk);
+      }
+    }else{
+      lsmStringAppendf(&str, "%d", pArray->iFirst);
+      while( iBlk!=iLastBlk ){
+        lsmStringAppendf(&str, " %d", fsLastPageOnBlock(pFS, iBlk));
+        fsBlockNext(pFS, pArray, iBlk, &iBlk);
+        lsmStringAppendf(&str, " %d", fsFirstPageOnBlock(pFS, iBlk));
+      }
+      lsmStringAppendf(&str, " %d", pArray->iLastPg);
+    }
+
+    *pzOut = str.z;
+  }
+
+  if( bUnlock ){
+    int rcwork = LSM_BUSY;
+    lsmFinishWork(pDb, 0, &rcwork);
+  }
+  return rc;
+}
+
+int lsmFsSegmentContainsPg(
+  FileSystem *pFS, 
+  Segment *pSeg, 
+  LsmPgno iPg, 
+  int *pbRes
+){
+  Redirect *pRedir = pSeg->pRedirect;
+  int rc = LSM_OK;
+  int iBlk;
+  int iLastBlk;
+  int iPgBlock;                   /* Block containing page iPg */
+
+  iPgBlock = fsPageToBlock(pFS, pSeg->iFirst);
+  iBlk = fsRedirectBlock(pRedir, fsPageToBlock(pFS, pSeg->iFirst));
+  iLastBlk = fsRedirectBlock(pRedir, fsPageToBlock(pFS, pSeg->iLastPg));
+
+  while( iBlk!=iLastBlk && iBlk!=iPgBlock && rc==LSM_OK ){
+    rc = fsBlockNext(pFS, pSeg, iBlk, &iBlk);
+  }
+
+  *pbRes = (iBlk==iPgBlock);
+  return rc;
+}
+
+/*
+** This function implements the lsm_info(LSM_INFO_ARRAY_PAGES) request.
+** If successful, *pzOut is set to point to a nul-terminated string 
+** containing the array structure and LSM_OK is returned. The caller should
+** eventually free the string using lsmFree().
+**
+** If an error occurs, *pzOut is set to NULL and an LSM error code returned.
+*/
+int lsmInfoArrayPages(lsm_db *pDb, LsmPgno iFirst, char **pzOut){
+  int rc = LSM_OK;
+  Snapshot *pWorker;              /* Worker snapshot */
+  Segment *pSeg = 0;              /* Array to report on */
+  int bUnlock = 0;
+
+  *pzOut = 0;
+  if( iFirst==0 ) return LSM_ERROR;
+
+  /* Obtain the worker snapshot */
+  pWorker = pDb->pWorker;
+  if( !pWorker ){
+    rc = lsmBeginWork(pDb);
+    if( rc!=LSM_OK ) return rc;
+    pWorker = pDb->pWorker;
+    bUnlock = 1;
+  }
+
+  /* Search for the array that starts on page iFirst */
+  pSeg = findSegment(pWorker, iFirst);
+
+  if( pSeg==0 ){
+    /* Could not find the requested array. This is an error. */
+    rc = LSM_ERROR;
+  }else{
+    Page *pPg = 0;
+    FileSystem *pFS = pDb->pFS;
+    LsmString str;
+
+    lsmStringInit(&str, pDb->pEnv);
+    rc = lsmFsDbPageGet(pFS, pSeg, iFirst, &pPg);
+    while( rc==LSM_OK && pPg ){
+      Page *pNext = 0;
+      lsmStringAppendf(&str, " %lld", lsmFsPageNumber(pPg));
+      rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
+      lsmFsPageRelease(pPg);
+      pPg = pNext;
+    }
+
+    if( rc!=LSM_OK ){
+      lsmFree(pDb->pEnv, str.z);
+    }else{
+      *pzOut = str.z;
+    }
+  }
+
+  if( bUnlock ){
+    int rcwork = LSM_BUSY;
+    lsmFinishWork(pDb, 0, &rcwork);
+  }
+  return rc;
+}
+
+/*
+** The following macros are used by the integrity-check code. Associated with
+** each block in the database is an 8-bit bit mask (the entry in the aUsed[]
+** array). As the integrity-check meanders through the database, it sets the
+** following bits to indicate how each block is used.
+**
+** INTEGRITY_CHECK_FIRST_PG:
+**   First page of block is in use by sorted run.
+**
+** INTEGRITY_CHECK_LAST_PG:
+**   Last page of block is in use by sorted run.
+**
+** INTEGRITY_CHECK_USED:
+**   At least one page of the block is in use by a sorted run.
+**
+** INTEGRITY_CHECK_FREE:
+**   The free block list contains an entry corresponding to this block.
+*/
+#define INTEGRITY_CHECK_FIRST_PG 0x01
+#define INTEGRITY_CHECK_LAST_PG  0x02
+#define INTEGRITY_CHECK_USED     0x04
+#define INTEGRITY_CHECK_FREE     0x08
+
+/*
+** Helper function for lsmFsIntegrityCheck()
+*/
+static void checkBlocks(
+  FileSystem *pFS, 
+  Segment *pSeg,
+  int bExtra,                     /* If true, count the "next" block if any */
+  int nUsed,
+  u8 *aUsed
+){
+  if( pSeg ){
+    if( pSeg && pSeg->nSize>0 ){
+      int rc;
+      int iBlk;                   /* Current block (during iteration) */
+      int iLastBlk;               /* Last block of segment */
+      int iFirstBlk;              /* First block of segment */
+      int bLastIsLastOnBlock;     /* True iLast is the last on its block */
+
+      assert( 0==fsSegmentRedirects(pFS, pSeg) );
+      iBlk = iFirstBlk = fsPageToBlock(pFS, pSeg->iFirst);
+      iLastBlk = fsPageToBlock(pFS, pSeg->iLastPg);
+
+      bLastIsLastOnBlock = (fsLastPageOnBlock(pFS, iLastBlk)==pSeg->iLastPg);
+      assert( iBlk>0 );
+
+      do {
+        /* iBlk is a part of this sorted run. */
+        aUsed[iBlk-1] |= INTEGRITY_CHECK_USED;
+
+        /* If the first page of this block is also part of the segment,
+        ** set the flag to indicate that the first page of iBlk is in use.  
+        */
+        if( fsFirstPageOnBlock(pFS, iBlk)==pSeg->iFirst || iBlk!=iFirstBlk ){
+          assert( (aUsed[iBlk-1] & INTEGRITY_CHECK_FIRST_PG)==0 );
+          aUsed[iBlk-1] |= INTEGRITY_CHECK_FIRST_PG;
+        }
+
+        /* Unless the sorted run finishes before the last page on this block, 
+        ** the last page of this block is also in use.  */
+        if( iBlk!=iLastBlk || bLastIsLastOnBlock ){
+          assert( (aUsed[iBlk-1] & INTEGRITY_CHECK_LAST_PG)==0 );
+          aUsed[iBlk-1] |= INTEGRITY_CHECK_LAST_PG;
+        }
+
+        /* Special case. The sorted run being scanned is the output run of
+        ** a level currently undergoing an incremental merge. The sorted
+        ** run ends on the last page of iBlk, but the next block has already
+        ** been allocated. So mark it as in use as well.  */
+        if( iBlk==iLastBlk && bLastIsLastOnBlock && bExtra ){
+          int iExtra = 0;
+          rc = fsBlockNext(pFS, pSeg, iBlk, &iExtra);
+          assert( rc==LSM_OK );
+
+          assert( aUsed[iExtra-1]==0 );
+          aUsed[iExtra-1] |= INTEGRITY_CHECK_USED;
+          aUsed[iExtra-1] |= INTEGRITY_CHECK_FIRST_PG;
+          aUsed[iExtra-1] |= INTEGRITY_CHECK_LAST_PG;
+        }
+
+        /* Move on to the next block in the sorted run. Or set iBlk to zero
+        ** in order to break out of the loop if this was the last block in
+        ** the run.  */
+        if( iBlk==iLastBlk ){
+          iBlk = 0;
+        }else{
+          rc = fsBlockNext(pFS, pSeg, iBlk, &iBlk);
+          assert( rc==LSM_OK );
+        }
+      }while( iBlk );
+    }
+  }
+}
+
+typedef struct CheckFreelistCtx CheckFreelistCtx;
+struct CheckFreelistCtx {
+  u8 *aUsed;
+  int nBlock;
+};
+static int checkFreelistCb(void *pCtx, int iBlk, i64 iSnapshot){
+  CheckFreelistCtx *p = (CheckFreelistCtx *)pCtx;
+
+  assert( iBlk>=1 );
+  assert( iBlk<=p->nBlock );
+  assert( p->aUsed[iBlk-1]==0 );
+  p->aUsed[iBlk-1] = INTEGRITY_CHECK_FREE;
+  return 0;
+}
+
+/*
+** This function checks that all blocks in the database file are accounted
+** for. For each block, exactly one of the following must be true:
+**
+**   + the block is part of a sorted run, or
+**   + the block is on the free-block list
+**
+** This function also checks that there are no references to blocks with
+** out-of-range block numbers.
+**
+** If no errors are found, non-zero is returned. If an error is found, an
+** assert() fails.
+*/
+int lsmFsIntegrityCheck(lsm_db *pDb){
+  CheckFreelistCtx ctx;
+  FileSystem *pFS = pDb->pFS;
+  int i;
+  int rc;
+  Freelist freelist = {0, 0, 0};
+  u8 *aUsed;
+  Level *pLevel;
+  Snapshot *pWorker = pDb->pWorker;
+  int nBlock = pWorker->nBlock;
+
+#if 0 
+  static int nCall = 0;
+  nCall++;
+  printf("%d calls\n", nCall);
+#endif
+
+  aUsed = lsmMallocZero(pDb->pEnv, nBlock);
+  if( aUsed==0 ){
+    /* Malloc has failed. Since this function is only called within debug
+    ** builds, this probably means the user is running an OOM injection test.
+    ** Regardless, it will not be possible to run the integrity-check at this
+    ** time, so assume the database is Ok and return non-zero. */
+    return 1;
+  }
+
+  for(pLevel=pWorker->pLevel; pLevel; pLevel=pLevel->pNext){
+    int j;
+    checkBlocks(pFS, &pLevel->lhs, (pLevel->nRight!=0), nBlock, aUsed);
+    for(j=0; j<pLevel->nRight; j++){
+      checkBlocks(pFS, &pLevel->aRhs[j], 0, nBlock, aUsed);
+    }
+  }
+
+  /* Mark all blocks in the free-list as used */
+  ctx.aUsed = aUsed;
+  ctx.nBlock = nBlock;
+  rc = lsmWalkFreelist(pDb, 0, checkFreelistCb, (void *)&ctx);
+
+  if( rc==LSM_OK ){
+    for(i=0; i<nBlock; i++) assert( aUsed[i]!=0 );
+  }
+
+  lsmFree(pDb->pEnv, aUsed);
+  lsmFree(pDb->pEnv, freelist.aEntry);
+
+  return 1;
+}
+
+#ifndef NDEBUG
+/*
+** Return true if pPg happens to be the last page in segment pSeg. Or false
+** otherwise. This function is only invoked as part of assert() conditions.
+*/
+int lsmFsDbPageIsLast(Segment *pSeg, Page *pPg){
+  if( pPg->pFS->pCompress ){
+    LsmPgno iNext = 0;
+    int rc;
+    rc = fsNextPageOffset(pPg->pFS, pSeg, pPg->iPg, pPg->nCompress+6, &iNext);
+    return (rc!=LSM_OK || iNext==0);
+  }
+  return (pPg->iPg==pSeg->iLastPg);
+}
+#endif
diff --git a/ext/lsm1/lsm_log.c b/ext/lsm1/lsm_log.c
new file mode 100644
index 0000000..a66e40b
--- /dev/null
+++ b/ext/lsm1/lsm_log.c
@@ -0,0 +1,1156 @@
+/*
+** 2011-08-13
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** This file contains the implementation of LSM database logging. Logging
+** has one purpose in LSM - to make transactions durable.
+**
+** When data is written to an LSM database, it is initially stored in an
+** in-memory tree structure. Since this structure is in volatile memory,
+** if a power failure or application crash occurs it may be lost. To
+** prevent loss of data in this case, each time a record is written to the
+** in-memory tree an equivalent record is appended to the log on disk.
+** If a power failure or application crash does occur, data can be recovered
+** by reading the log.
+**
+** A log file consists of the following types of records representing data
+** written into the database:
+**
+**   LOG_WRITE:  A key-value pair written to the database.
+**   LOG_DELETE: A delete key issued to the database.
+**   LOG_COMMIT: A transaction commit.
+**
+** And the following types of records for ancillary purposes..
+**
+**   LOG_EOF:    A record indicating the end of a log file.
+**   LOG_PAD1:   A single byte padding record.
+**   LOG_PAD2:   An N byte padding record (N>1).
+**   LOG_JUMP:   A pointer to another offset within the log file.
+**
+** Each transaction written to the log contains one or more LOG_WRITE and/or
+** LOG_DELETE records, followed by a LOG_COMMIT record. The LOG_COMMIT record
+** contains an 8-byte checksum based on all previous data written to the
+** log file.
+**
+** LOG CHECKSUMS & RECOVERY
+**
+**   Checksums are found in two types of log records: LOG_COMMIT and
+**   LOG_CKSUM records. In order to recover content from a log, a client
+**   reads each record from the start of the log, calculating a checksum as
+**   it does. Each time a LOG_COMMIT or LOG_CKSUM is encountered, the 
+**   recovery process verifies that the checksum stored in the log 
+**   matches the calculated checksum. If it does not, the recovery process
+**   can stop reading the log.
+**
+**   If a recovery process reads records (other than COMMIT or CKSUM) 
+**   consisting of at least LSM_CKSUM_MAXDATA bytes, then the next record in
+**   the log must be either a LOG_CKSUM or LOG_COMMIT record. If it is
+**   not, the recovery process also stops reading the log.
+**
+**   To recover the log file, it must be read twice. The first time to 
+**   determine the location of the last valid commit record. And the second
+**   time to load data into the in-memory tree.
+**
+**   Todo: Surely there is a better way...
+**
+** LOG WRAPPING
+**
+**   If the log file were never deleted or wrapped, it would be possible to
+**   read it from start to end each time is required recovery (i.e each time
+**   the number of database clients changes from 0 to 1). Effectively reading
+**   the entire history of the database each time. This would quickly become 
+**   inefficient. Additionally, since the log file would grow without bound,
+**   it wastes storage space.
+**
+**   Instead, part of each checkpoint written into the database file contains 
+**   a log offset (and other information required to read the log starting at
+**   at this offset) at which to begin recovery. Offset $O.
+**
+**   Once a checkpoint has been written and synced into the database file, it
+**   is guaranteed that no recovery process will need to read any data before
+**   offset $O of the log file. It is therefore safe to begin overwriting
+**   any data that occurs before offset $O.
+**
+**   This implementation separates the log into three regions mapped into
+**   the log file - regions 0, 1 and 2. During recovery, regions are read
+**   in ascending order (i.e. 0, then 1, then 2). Each region is zero or
+**   more bytes in size.
+**
+**     |---1---|..|--0--|.|--2--|....
+**
+**   New records are always appended to the end of region 2.
+**
+**   Initially (when it is empty), all three regions are zero bytes in size.
+**   Each of them are located at the beginning of the file. As records are
+**   added to the log, region 2 grows, so that the log consists of a zero
+**   byte region 1, followed by a zero byte region 0, followed by an N byte
+**   region 2. After one or more checkpoints have been written to disk, 
+**   the start point of region 2 is moved to $O. For example:
+**
+**     A) ||.........|--2--|....
+**   
+**   (both regions 0 and 1 are 0 bytes in size at offset 0).
+**
+**   Eventually, the log wraps around to write new records into the start.
+**   At this point, region 2 is renamed to region 0. Region 0 is renamed
+**   to region 2. After appending a few records to the new region 2, the
+**   log file looks like this:
+**
+**     B) ||--2--|...|--0--|....
+**
+**   (region 1 is still 0 bytes in size, located at offset 0).
+**
+**   Any checkpoints made at this point may reduce the size of region 0.
+**   However, if they do not, and region 2 expands so that it is about to
+**   overwrite the start of region 0, then region 2 is renamed to region 1,
+**   and a new region 2 created at the end of the file following the existing
+**   region 0.
+**
+**     C) |---1---|..|--0--|.|-2-|
+**
+**   In this state records are appended to region 2 until checkpoints have
+**   contracted regions 0 AND 1 UNTil they are both zero bytes in size. They 
+**   are then shifted to the start of the log file, leaving the system in 
+**   the equivalent of state A above.
+**
+**   Alternatively, state B may transition directly to state A if the size
+**   of region 0 is reduced to zero bytes before region 2 threatens to 
+**   encroach upon it.
+**
+** LOG_PAD1 & LOG_PAD2 RECORDS
+**
+**   PAD1 and PAD2 records may appear in a log file at any point. They allow
+**   a process writing the log file align the beginning of transactions with 
+**   the beginning of disk sectors, which increases robustness.
+**
+** RECORD FORMATS:
+**
+**   LOG_EOF:    * A single 0x00 byte.
+**
+**   LOG_PAD1:   * A single 0x01 byte.
+**
+**   LOG_PAD2:   * A single 0x02 byte, followed by
+**               * The number of unused bytes (N) as a varint,
+**               * An N byte block of unused space.
+**
+**   LOG_COMMIT: * A single 0x03 byte.
+**               * An 8-byte checksum.
+**
+**   LOG_JUMP:   * A single 0x04 byte.
+**               * Absolute file offset to jump to, encoded as a varint.
+**
+**   LOG_WRITE:  * A single 0x06 or 0x07 byte, 
+**               * The number of bytes in the key, encoded as a varint, 
+**               * The number of bytes in the value, encoded as a varint, 
+**               * If the first byte was 0x07, an 8 byte checksum.
+**               * The key data,
+**               * The value data.
+**
+**   LOG_DELETE: * A single 0x08 or 0x09 byte, 
+**               * The number of bytes in the key, encoded as a varint, 
+**               * If the first byte was 0x09, an 8 byte checksum.
+**               * The key data.
+**
+**   Varints are as described in lsm_varint.c (SQLite 4 format).
+**
+** CHECKSUMS:
+**
+**   The checksum is calculated using two 32-bit unsigned integers, s0 and
+**   s1. The initial value for both is 42. It is updated each time a record
+**   is written into the log file by treating the encoded (binary) record as 
+**   an array of 32-bit little-endian integers. Then, if x[] is the integer
+**   array, updating the checksum accumulators as follows:
+**
+**     for i from 0 to n-1 step 2:
+**       s0 += x[i] + s1;
+**       s1 += x[i+1] + s0;
+**     endfor
+**
+**   If the record is not an even multiple of 8-bytes in size it is padded
+**   with zeroes to make it so before the checksum is updated.
+**
+**   The checksum stored in a COMMIT, WRITE or DELETE is based on all bytes
+**   up to the start of the 8-byte checksum itself, including the COMMIT,
+**   WRITE or DELETE fields that appear before the checksum in the record.
+**
+** VARINT FORMAT
+**
+** See lsm_varint.c.
+*/
+
+#ifndef _LSM_INT_H
+# include "lsmInt.h"
+#endif
+
+/* Log record types */
+#define LSM_LOG_EOF          0x00
+#define LSM_LOG_PAD1         0x01
+#define LSM_LOG_PAD2         0x02
+#define LSM_LOG_COMMIT       0x03
+#define LSM_LOG_JUMP         0x04
+
+#define LSM_LOG_WRITE        0x06
+#define LSM_LOG_WRITE_CKSUM  0x07
+
+#define LSM_LOG_DELETE       0x08
+#define LSM_LOG_DELETE_CKSUM 0x09
+
+#define LSM_LOG_DRANGE       0x0A
+#define LSM_LOG_DRANGE_CKSUM 0x0B
+
+/* Require a checksum every 32KB. */
+#define LSM_CKSUM_MAXDATA (32*1024)
+
+/* Do not wrap a log file smaller than this in bytes. */
+#define LSM_MIN_LOGWRAP      (128*1024)
+
+/*
+** szSector:
+**   Commit records must be aligned to end on szSector boundaries. If
+**   the safety-mode is set to NORMAL or OFF, this value is 1. Otherwise,
+**   if the safety-mode is set to FULL, it is the size of the file-system
+**   sectors as reported by lsmFsSectorSize().
+*/
+struct LogWriter {
+  u32 cksum0;                     /* Checksum 0 at offset iOff */
+  u32 cksum1;                     /* Checksum 1 at offset iOff */
+  int iCksumBuf;                  /* Bytes of buf that have been checksummed */
+  i64 iOff;                       /* Offset at start of buffer buf */
+  int szSector;                   /* Sector size for this transaction */
+  LogRegion jump;                 /* Avoid writing to this region */
+  i64 iRegion1End;                /* End of first region written by trans */
+  i64 iRegion2Start;              /* Start of second regions written by trans */
+  LsmString buf;                  /* Buffer containing data not yet written */
+};
+
+/*
+** Return the result of interpreting the first 4 bytes in buffer aIn as 
+** a 32-bit unsigned little-endian integer.
+*/
+static u32 getU32le(u8 *aIn){
+  return ((u32)aIn[3] << 24) 
+       + ((u32)aIn[2] << 16) 
+       + ((u32)aIn[1] << 8) 
+       + ((u32)aIn[0]);
+}
+
+
+/*
+** This function is the same as logCksum(), except that pointer "a" need
+** not be aligned to an 8-byte boundary or padded with zero bytes. This
+** version is slower, but sometimes more convenient to use.
+*/
+static void logCksumUnaligned(
+  char *z,                        /* Input buffer */
+  int n,                          /* Size of input buffer in bytes */
+  u32 *pCksum0,                   /* IN/OUT: Checksum value 1 */
+  u32 *pCksum1                    /* IN/OUT: Checksum value 2 */
+){
+  u8 *a = (u8 *)z;
+  u32 cksum0 = *pCksum0;
+  u32 cksum1 = *pCksum1;
+  int nIn = (n/8) * 8;
+  int i;
+
+  assert( n>0 );
+  for(i=0; i<nIn; i+=8){
+    cksum0 += getU32le(&a[i]) + cksum1;
+    cksum1 += getU32le(&a[i+4]) + cksum0;
+  }
+
+  if( nIn!=n ){
+    u8 aBuf[8] = {0, 0, 0, 0, 0, 0, 0, 0};
+    assert( (n-nIn)<8 && n>nIn );
+    memcpy(aBuf, &a[nIn], n-nIn);
+    cksum0 += getU32le(aBuf) + cksum1;
+    cksum1 += getU32le(&aBuf[4]) + cksum0;
+  }
+
+  *pCksum0 = cksum0;
+  *pCksum1 = cksum1;
+}
+
+/*
+** Update pLog->cksum0 and pLog->cksum1 so that the first nBuf bytes in the 
+** write buffer (pLog->buf) are included in the checksum.
+*/
+static void logUpdateCksum(LogWriter *pLog, int nBuf){
+  assert( (pLog->iCksumBuf % 8)==0 );
+  assert( pLog->iCksumBuf<=nBuf );
+  assert( (nBuf % 8)==0 || nBuf==pLog->buf.n );
+  if( nBuf>pLog->iCksumBuf ){
+    logCksumUnaligned(
+        &pLog->buf.z[pLog->iCksumBuf], nBuf-pLog->iCksumBuf, 
+        &pLog->cksum0, &pLog->cksum1
+    );
+  }
+  pLog->iCksumBuf = nBuf;
+}
+
+static i64 firstByteOnSector(LogWriter *pLog, i64 iOff){
+  return (iOff / pLog->szSector) * pLog->szSector;
+}
+static i64 lastByteOnSector(LogWriter *pLog, i64 iOff){
+  return firstByteOnSector(pLog, iOff) + pLog->szSector - 1;
+}
+
+/*
+** If possible, reclaim log file space. Log file space is reclaimed after
+** a snapshot that points to the same data in the database file is synced
+** into the db header.
+*/
+static int logReclaimSpace(lsm_db *pDb){
+  int rc;
+  int iMeta;
+  int bRotrans;                   /* True if there exists some ro-trans */
+
+  /* Test if there exists some other connection with a read-only transaction
+  ** open. If there does, then log file space may not be reclaimed.  */
+  rc = lsmDetectRoTrans(pDb, &bRotrans);
+  if( rc!=LSM_OK || bRotrans ) return rc;
+
+  iMeta = (int)pDb->pShmhdr->iMetaPage;
+  if( iMeta==1 || iMeta==2 ){
+    DbLog *pLog = &pDb->treehdr.log;
+    i64 iSyncedId;
+
+    /* Read the snapshot-id of the snapshot stored on meta-page iMeta. Note
+    ** that in theory, the value read is untrustworthy (due to a race 
+    ** condition - see comments above lsmFsReadSyncedId()). So it is only 
+    ** ever used to conclude that no log space can be reclaimed. If it seems
+    ** to indicate that it may be possible to reclaim log space, a
+    ** second call to lsmCheckpointSynced() (which does return trustworthy
+    ** values) is made below to confirm.  */
+    rc = lsmFsReadSyncedId(pDb, iMeta, &iSyncedId);
+
+    if( rc==LSM_OK && pLog->iSnapshotId!=iSyncedId ){
+      i64 iSnapshotId = 0;
+      i64 iOff = 0;
+      rc = lsmCheckpointSynced(pDb, &iSnapshotId, &iOff, 0);
+      if( rc==LSM_OK && pLog->iSnapshotId<iSnapshotId ){
+        int iRegion;
+        for(iRegion=0; iRegion<3; iRegion++){
+          LogRegion *p = &pLog->aRegion[iRegion];
+          if( iOff>=p->iStart && iOff<=p->iEnd ) break;
+          p->iStart = 0;
+          p->iEnd = 0;
+        }
+        assert( iRegion<3 );
+        pLog->aRegion[iRegion].iStart = iOff;
+        pLog->iSnapshotId = iSnapshotId;
+      }
+    }
+  }
+  return rc;
+}
+
+/*
+** This function is called when a write-transaction is first opened. It
+** is assumed that the caller is holding the client-mutex when it is 
+** called.
+**
+** Before returning, this function allocates the LogWriter object that
+** will be used to write to the log file during the write transaction.
+** LSM_OK is returned if no error occurs, otherwise an LSM error code.
+*/
+int lsmLogBegin(lsm_db *pDb){
+  int rc = LSM_OK;
+  LogWriter *pNew;
+  LogRegion *aReg;
+
+  if( pDb->bUseLog==0 ) return LSM_OK;
+
+  /* If the log file has not yet been opened, open it now. Also allocate
+  ** the LogWriter structure, if it has not already been allocated.  */
+  rc = lsmFsOpenLog(pDb, 0);
+  if( pDb->pLogWriter==0 ){
+    pNew = lsmMallocZeroRc(pDb->pEnv, sizeof(LogWriter), &rc);
+    if( pNew ){
+      lsmStringInit(&pNew->buf, pDb->pEnv);
+      rc = lsmStringExtend(&pNew->buf, 2);
+    }
+    pDb->pLogWriter = pNew;
+  }else{
+    pNew = pDb->pLogWriter;
+    assert( (u8 *)(&pNew[1])==(u8 *)(&((&pNew->buf)[1])) );
+    memset(pNew, 0, ((u8 *)&pNew->buf) - (u8 *)pNew);
+    pNew->buf.n = 0;
+  }
+
+  if( rc==LSM_OK ){
+    /* The following call detects whether or not a new snapshot has been 
+    ** synced into the database file. If so, it updates the contents of
+    ** the pDb->treehdr.log structure to reclaim any space in the log
+    ** file that is no longer required. 
+    **
+    ** TODO: Calling this every transaction is overkill. And since the 
+    ** call has to read and checksum a snapshot from the database file,
+    ** it is expensive. It would be better to figure out a way so that
+    ** this is only called occasionally - say for every 32KB written to 
+    ** the log file.
+    */
+    rc = logReclaimSpace(pDb);
+  }
+  if( rc!=LSM_OK ){
+    lsmLogClose(pDb);
+    return rc;
+  }
+
+  /* Set the effective sector-size for this transaction. Sectors are assumed
+  ** to be one byte in size if the safety-mode is OFF or NORMAL, or as
+  ** reported by lsmFsSectorSize if it is FULL.  */
+  if( pDb->eSafety==LSM_SAFETY_FULL ){
+    pNew->szSector = lsmFsSectorSize(pDb->pFS);
+    assert( pNew->szSector>0 );
+  }else{
+    pNew->szSector = 1;
+  }
+
+  /* There are now three scenarios:
+  **
+  **   1) Regions 0 and 1 are both zero bytes in size and region 2 begins
+  **      at a file offset greater than LSM_MIN_LOGWRAP. In this case, wrap
+  **      around to the start and write data into the start of the log file. 
+  **
+  **   2) Region 1 is zero bytes in size and region 2 occurs earlier in the 
+  **      file than region 0. In this case, append data to region 2, but
+  **      remember to jump over region 1 if required.
+  **
+  **   3) Region 2 is the last in the file. Append to it.
+  */
+  aReg = &pDb->treehdr.log.aRegion[0];
+
+  assert( aReg[0].iEnd==0 || aReg[0].iEnd>aReg[0].iStart );
+  assert( aReg[1].iEnd==0 || aReg[1].iEnd>aReg[1].iStart );
+
+  pNew->cksum0 = pDb->treehdr.log.cksum0;
+  pNew->cksum1 = pDb->treehdr.log.cksum1;
+
+  if( aReg[0].iEnd==0 && aReg[1].iEnd==0 && aReg[2].iStart>=LSM_MIN_LOGWRAP ){
+    /* Case 1. Wrap around to the start of the file. Write an LSM_LOG_JUMP 
+    ** into the log file in this case. Pad it out to 8 bytes using a PAD2
+    ** record so that the checksums can be updated immediately.  */
+    u8 aJump[] = { 
+      LSM_LOG_PAD2, 0x04, 0x00, 0x00, 0x00, 0x00, LSM_LOG_JUMP, 0x00 
+    };
+
+    lsmStringBinAppend(&pNew->buf, aJump, sizeof(aJump));
+    logUpdateCksum(pNew, pNew->buf.n);
+    rc = lsmFsWriteLog(pDb->pFS, aReg[2].iEnd, &pNew->buf);
+    pNew->iCksumBuf = pNew->buf.n = 0;
+
+    aReg[2].iEnd += 8;
+    pNew->jump = aReg[0] = aReg[2];
+    aReg[2].iStart = aReg[2].iEnd = 0;
+  }else if( aReg[1].iEnd==0 && aReg[2].iEnd<aReg[0].iEnd ){
+    /* Case 2. */
+    pNew->iOff = aReg[2].iEnd;
+    pNew->jump = aReg[0];
+  }else{
+    /* Case 3. */
+    assert( aReg[2].iStart>=aReg[0].iEnd && aReg[2].iStart>=aReg[1].iEnd );
+    pNew->iOff = aReg[2].iEnd;
+  }
+
+  if( pNew->jump.iStart ){
+    i64 iRound;
+    assert( pNew->jump.iStart>pNew->iOff );
+
+    iRound = firstByteOnSector(pNew, pNew->jump.iStart);
+    if( iRound>pNew->iOff ) pNew->jump.iStart = iRound;
+    pNew->jump.iEnd = lastByteOnSector(pNew, pNew->jump.iEnd);
+  }
+
+  assert( pDb->pLogWriter==pNew );
+  return rc;
+}
+
+/*
+** This function is called when a write-transaction is being closed.
+** Parameter bCommit is true if the transaction is being committed,
+** or false otherwise. The caller must hold the client-mutex to call
+** this function.
+**
+** A call to this function deletes the LogWriter object allocated by
+** lsmLogBegin(). If the transaction is being committed, the shared state
+** in *pLog is updated before returning.
+*/
+void lsmLogEnd(lsm_db *pDb, int bCommit){
+  DbLog *pLog;
+  LogWriter *p;
+  p = pDb->pLogWriter;
+
+  if( p==0 ) return;
+  pLog = &pDb->treehdr.log;
+
+  if( bCommit ){
+    pLog->aRegion[2].iEnd = p->iOff;
+    pLog->cksum0 = p->cksum0;
+    pLog->cksum1 = p->cksum1;
+    if( p->iRegion1End ){
+      /* This happens when the transaction had to jump over some other
+      ** part of the log.  */
+      assert( pLog->aRegion[1].iEnd==0 );
+      assert( pLog->aRegion[2].iStart<p->iRegion1End );
+      pLog->aRegion[1].iStart = pLog->aRegion[2].iStart;
+      pLog->aRegion[1].iEnd = p->iRegion1End;
+      pLog->aRegion[2].iStart = p->iRegion2Start;
+    }
+  }
+}
+
+static int jumpIfRequired(
+  lsm_db *pDb,
+  LogWriter *pLog,
+  int nReq,
+  int *pbJump
+){
+  /* Determine if it is necessary to add an LSM_LOG_JUMP to jump over the
+  ** jump region before writing the LSM_LOG_WRITE or DELETE record. This
+  ** is necessary if there is insufficient room between the current offset
+  ** and the jump region to fit the new WRITE/DELETE record and the largest
+  ** possible JUMP record with up to 7 bytes of padding (a total of 17 
+  ** bytes).  */
+  if( (pLog->jump.iStart > (pLog->iOff + pLog->buf.n))
+   && (pLog->jump.iStart < (pLog->iOff + pLog->buf.n + (nReq + 17))) 
+  ){
+    int rc;                       /* Return code */
+    i64 iJump;                    /* Offset to jump to */
+    u8 aJump[10];                 /* Encoded jump record */
+    int nJump;                    /* Valid bytes in aJump[] */
+    int nPad;                     /* Bytes of padding required */
+
+    /* Serialize the JUMP record */
+    iJump = pLog->jump.iEnd+1;
+    aJump[0] = LSM_LOG_JUMP;
+    nJump = 1 + lsmVarintPut64(&aJump[1], iJump);
+
+    /* Adding padding to the contents of the buffer so that it will be a 
+    ** multiple of 8 bytes in size after the JUMP record is appended. This
+    ** is not strictly required, it just makes the keeping the running 
+    ** checksum up to date in this file a little simpler.  */
+    nPad = (pLog->buf.n + nJump) % 8;
+    if( nPad ){
+      u8 aPad[7] = {0,0,0,0,0,0,0};
+      nPad = 8-nPad;
+      if( nPad==1 ){
+        aPad[0] = LSM_LOG_PAD1;
+      }else{
+        aPad[0] = LSM_LOG_PAD2;
+        aPad[1] = (u8)(nPad-2);
+      }
+      rc = lsmStringBinAppend(&pLog->buf, aPad, nPad);
+      if( rc!=LSM_OK ) return rc;
+    }
+
+    /* Append the JUMP record to the buffer. Then flush the buffer to disk
+    ** and update the checksums. The next write to the log file (assuming
+    ** there is no transaction rollback) will be to offset iJump (just past
+    ** the jump region).  */
+    rc = lsmStringBinAppend(&pLog->buf, aJump, nJump);
+    if( rc!=LSM_OK ) return rc;
+    assert( (pLog->buf.n % 8)==0 );
+    rc = lsmFsWriteLog(pDb->pFS, pLog->iOff, &pLog->buf);
+    if( rc!=LSM_OK ) return rc;
+    logUpdateCksum(pLog, pLog->buf.n);
+    pLog->iRegion1End = (pLog->iOff + pLog->buf.n);
+    pLog->iRegion2Start = iJump;
+    pLog->iOff = iJump;
+    pLog->iCksumBuf = pLog->buf.n = 0;
+    if( pbJump ) *pbJump = 1;
+  }
+
+  return LSM_OK;
+}
+
+static int logCksumAndFlush(lsm_db *pDb){
+  int rc;                         /* Return code */
+  LogWriter *pLog = pDb->pLogWriter;
+
+  /* Calculate the checksum value. Append it to the buffer. */
+  logUpdateCksum(pLog, pLog->buf.n);
+  lsmPutU32((u8 *)&pLog->buf.z[pLog->buf.n], pLog->cksum0);
+  pLog->buf.n += 4;
+  lsmPutU32((u8 *)&pLog->buf.z[pLog->buf.n], pLog->cksum1);
+  pLog->buf.n += 4;
+
+  /* Write the contents of the buffer to disk. */
+  rc = lsmFsWriteLog(pDb->pFS, pLog->iOff, &pLog->buf);
+  pLog->iOff += pLog->buf.n;
+  pLog->iCksumBuf = pLog->buf.n = 0;
+
+  return rc;
+}
+
+/*
+** Write the contents of the log-buffer to disk. Then write either a CKSUM
+** or COMMIT record, depending on the value of parameter eType.
+*/
+static int logFlush(lsm_db *pDb, int eType){
+  int rc;
+  int nReq;
+  LogWriter *pLog = pDb->pLogWriter;
+  
+  assert( eType==LSM_LOG_COMMIT );
+  assert( pLog );
+
+  /* Commit record is always 9 bytes in size. */
+  nReq = 9;
+  if( eType==LSM_LOG_COMMIT && pLog->szSector>1 ) nReq += pLog->szSector + 17;
+  rc = jumpIfRequired(pDb, pLog, nReq, 0);
+
+  /* If this is a COMMIT, add padding to the log so that the COMMIT record
+  ** is aligned against the end of a disk sector. In other words, add padding
+  ** so that the first byte following the COMMIT record lies on a different
+  ** sector.  */
+  if( eType==LSM_LOG_COMMIT && pLog->szSector>1 ){
+    int nPad;                     /* Bytes of padding to add */
+
+    /* Determine the value of nPad. */
+    nPad = ((pLog->iOff + pLog->buf.n + 9) % pLog->szSector);
+    if( nPad ) nPad = pLog->szSector - nPad;
+    rc = lsmStringExtend(&pLog->buf, nPad);
+    if( rc!=LSM_OK ) return rc;
+
+    while( nPad ){
+      if( nPad==1 ){
+        pLog->buf.z[pLog->buf.n++] = LSM_LOG_PAD1;
+        nPad = 0;
+      }else{
+        int n = LSM_MIN(200, nPad-2);
+        pLog->buf.z[pLog->buf.n++] = LSM_LOG_PAD2;
+        pLog->buf.z[pLog->buf.n++] = (char)n;
+        nPad -= 2;
+        memset(&pLog->buf.z[pLog->buf.n], 0x2B, n);
+        pLog->buf.n += n;
+        nPad -= n;
+      }
+    }
+  }
+
+  /* Make sure there is room in the log-buffer to add the CKSUM or COMMIT
+  ** record. Then add the first byte of it.  */
+  rc = lsmStringExtend(&pLog->buf, 9);
+  if( rc!=LSM_OK ) return rc;
+  pLog->buf.z[pLog->buf.n++] = (char)eType;
+  memset(&pLog->buf.z[pLog->buf.n], 0, 8);
+
+  rc = logCksumAndFlush(pDb);
+
+  /* If this is a commit and synchronous=full, sync the log to disk. */
+  if( rc==LSM_OK && eType==LSM_LOG_COMMIT && pDb->eSafety==LSM_SAFETY_FULL ){
+    rc = lsmFsSyncLog(pDb->pFS);
+  }
+  return rc;
+}
+
+/*
+** Append an LSM_LOG_WRITE (if nVal>=0) or LSM_LOG_DELETE (if nVal<0) 
+** record to the database log.
+*/
+int lsmLogWrite(
+  lsm_db *pDb,                    /* Database handle */
+  int eType,
+  void *pKey, int nKey,           /* Database key to write to log */
+  void *pVal, int nVal            /* Database value (or nVal<0) to write */
+){
+  int rc = LSM_OK;
+  LogWriter *pLog;                /* Log object to write to */
+  int nReq;                       /* Bytes of space required in log */
+  int bCksum = 0;                 /* True to embed a checksum in this record */
+
+  assert( eType==LSM_WRITE || eType==LSM_DELETE || eType==LSM_DRANGE );
+  assert( LSM_LOG_WRITE==LSM_WRITE );
+  assert( LSM_LOG_DELETE==LSM_DELETE );
+  assert( LSM_LOG_DRANGE==LSM_DRANGE );
+  assert( (eType==LSM_LOG_DELETE)==(nVal<0) );
+
+  if( pDb->bUseLog==0 ) return LSM_OK;
+  pLog = pDb->pLogWriter;
+
+  /* Determine how many bytes of space are required, assuming that a checksum
+  ** will be embedded in this record (even though it may not be).  */
+  nReq = 1 + lsmVarintLen32(nKey) + 8 + nKey;
+  if( eType!=LSM_LOG_DELETE ) nReq += lsmVarintLen32(nVal) + nVal;
+
+  /* Jump over the jump region if required. Set bCksum to true to tell the
+  ** code below to include a checksum in the record if either (a) writing
+  ** this record would mean that more than LSM_CKSUM_MAXDATA bytes of data
+  ** have been written to the log since the last checksum, or (b) the jump
+  ** is taken.  */
+  rc = jumpIfRequired(pDb, pLog, nReq, &bCksum);
+  if( (pLog->buf.n+nReq) > LSM_CKSUM_MAXDATA ) bCksum = 1;
+
+  if( rc==LSM_OK ){
+    rc = lsmStringExtend(&pLog->buf, nReq);
+  }
+  if( rc==LSM_OK ){
+    u8 *a = (u8 *)&pLog->buf.z[pLog->buf.n];
+    
+    /* Write the record header - the type byte followed by either 1 (for
+    ** DELETE) or 2 (for WRITE) varints.  */
+    assert( LSM_LOG_WRITE_CKSUM == (LSM_LOG_WRITE | 0x0001) );
+    assert( LSM_LOG_DELETE_CKSUM == (LSM_LOG_DELETE | 0x0001) );
+    assert( LSM_LOG_DRANGE_CKSUM == (LSM_LOG_DRANGE | 0x0001) );
+    *(a++) = (u8)eType | (u8)bCksum;
+    a += lsmVarintPut32(a, nKey);
+    if( eType!=LSM_LOG_DELETE ) a += lsmVarintPut32(a, nVal);
+
+    if( bCksum ){
+      pLog->buf.n = (a - (u8 *)pLog->buf.z);
+      rc = logCksumAndFlush(pDb);
+      a = (u8 *)&pLog->buf.z[pLog->buf.n];
+    }
+
+    memcpy(a, pKey, nKey);
+    a += nKey;
+    if( eType!=LSM_LOG_DELETE ){
+      memcpy(a, pVal, nVal);
+      a += nVal;
+    }
+    pLog->buf.n = a - (u8 *)pLog->buf.z;
+    assert( pLog->buf.n<=pLog->buf.nAlloc );
+  }
+
+  return rc;
+}
+
+/*
+** Append an LSM_LOG_COMMIT record to the database log.
+*/
+int lsmLogCommit(lsm_db *pDb){
+  if( pDb->bUseLog==0 ) return LSM_OK;
+  return logFlush(pDb, LSM_LOG_COMMIT);
+}
+
+/*
+** Store the current offset and other checksum related information in the
+** structure *pMark. Later, *pMark can be passed to lsmLogSeek() to "rewind"
+** the LogWriter object to the current log file offset. This is used when
+** rolling back savepoint transactions.
+*/
+void lsmLogTell(
+  lsm_db *pDb,                    /* Database handle */
+  LogMark *pMark                  /* Populate this object with current offset */
+){
+  LogWriter *pLog;
+  int nCksum;
+
+  if( pDb->bUseLog==0 ) return;
+  pLog = pDb->pLogWriter;
+  nCksum = pLog->buf.n & 0xFFFFFFF8;
+  logUpdateCksum(pLog, nCksum);
+  assert( pLog->iCksumBuf==nCksum );
+  pMark->nBuf = pLog->buf.n - nCksum;
+  memcpy(pMark->aBuf, &pLog->buf.z[nCksum], pMark->nBuf);
+
+  pMark->iOff = pLog->iOff + pLog->buf.n;
+  pMark->cksum0 = pLog->cksum0;
+  pMark->cksum1 = pLog->cksum1;
+}
+
+/*
+** Seek (rewind) back to the log file offset stored by an ealier call to
+** lsmLogTell() in *pMark.
+*/
+void lsmLogSeek(
+  lsm_db *pDb,                    /* Database handle */
+  LogMark *pMark                  /* Object containing log offset to seek to */
+){
+  LogWriter *pLog;
+
+  if( pDb->bUseLog==0 ) return;
+  pLog = pDb->pLogWriter;
+
+  assert( pMark->iOff<=pLog->iOff+pLog->buf.n );
+  if( (pMark->iOff & 0xFFFFFFF8)>=pLog->iOff ){
+    pLog->buf.n = (int)(pMark->iOff - pLog->iOff);
+    pLog->iCksumBuf = (pLog->buf.n & 0xFFFFFFF8);
+  }else{
+    pLog->buf.n = pMark->nBuf;
+    memcpy(pLog->buf.z, pMark->aBuf, pMark->nBuf);
+    pLog->iCksumBuf = 0;
+    pLog->iOff = pMark->iOff - pMark->nBuf;
+  }
+  pLog->cksum0 = pMark->cksum0;
+  pLog->cksum1 = pMark->cksum1;
+
+  if( pMark->iOff > pLog->iRegion1End ) pLog->iRegion1End = 0;
+  if( pMark->iOff > pLog->iRegion2Start ) pLog->iRegion2Start = 0;
+}
+
+/*
+** This function does the work for an lsm_info(LOG_STRUCTURE) request.
+*/
+int lsmInfoLogStructure(lsm_db *pDb, char **pzVal){
+  int rc = LSM_OK;
+  char *zVal = 0;
+
+  /* If there is no read or write transaction open, read the latest 
+  ** tree-header from shared-memory to report on. If necessary, update
+  ** it based on the contents of the database header.  
+  **
+  ** No locks are taken here - these are passive read operations only.
+  */
+  if( pDb->pCsr==0 && pDb->nTransOpen==0 ){
+    rc = lsmTreeLoadHeader(pDb, 0);
+    if( rc==LSM_OK ) rc = logReclaimSpace(pDb);
+  }
+
+  if( rc==LSM_OK ){
+    DbLog *pLog = &pDb->treehdr.log;
+    zVal = lsmMallocPrintf(pDb->pEnv, 
+        "%d %d %d %d %d %d", 
+        (int)pLog->aRegion[0].iStart, (int)pLog->aRegion[0].iEnd,
+        (int)pLog->aRegion[1].iStart, (int)pLog->aRegion[1].iEnd,
+        (int)pLog->aRegion[2].iStart, (int)pLog->aRegion[2].iEnd
+    );
+    if( !zVal ) rc = LSM_NOMEM_BKPT;
+  }
+
+  *pzVal = zVal;
+  return rc;
+}
+
+/*************************************************************************
+** Begin code for log recovery.
+*/
+
+typedef struct LogReader LogReader;
+struct LogReader {
+  FileSystem *pFS;                /* File system to read from */
+  i64 iOff;                       /* File offset at end of buf content */
+  int iBuf;                       /* Current read offset in buf */
+  LsmString buf;                  /* Buffer containing file content */
+
+  int iCksumBuf;                  /* Offset in buf corresponding to cksum[01] */
+  u32 cksum0;                     /* Checksum 0 at offset iCksumBuf */
+  u32 cksum1;                     /* Checksum 1 at offset iCksumBuf */
+};
+
+static void logReaderBlob(
+  LogReader *p,                   /* Log reader object */
+  LsmString *pBuf,                /* Dynamic storage, if required */
+  int nBlob,                      /* Number of bytes to read */
+  u8 **ppBlob,                    /* OUT: Pointer to blob read */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  static const int LOG_READ_SIZE = 512;
+  int rc = *pRc;                  /* Return code */
+  int nReq = nBlob;               /* Bytes required */
+
+  while( rc==LSM_OK && nReq>0 ){
+    int nAvail;                   /* Bytes of data available in p->buf */
+    if( p->buf.n==p->iBuf ){
+      int nCksum;                 /* Total bytes requiring checksum */
+      int nCarry = 0;             /* Total bytes requiring checksum */
+
+      nCksum = p->iBuf - p->iCksumBuf;
+      if( nCksum>0 ){
+        nCarry = nCksum % 8;
+        nCksum = ((nCksum / 8) * 8);
+        if( nCksum>0 ){
+          logCksumUnaligned(
+              &p->buf.z[p->iCksumBuf], nCksum, &p->cksum0, &p->cksum1
+          );
+        }
+      }
+      if( nCarry>0 ) memcpy(p->buf.z, &p->buf.z[p->iBuf-nCarry], nCarry);
+      p->buf.n = nCarry;
+      p->iBuf = nCarry;
+
+      rc = lsmFsReadLog(p->pFS, p->iOff, LOG_READ_SIZE, &p->buf);
+      if( rc!=LSM_OK ) break;
+      p->iCksumBuf = 0;
+      p->iOff += LOG_READ_SIZE;
+    }
+
+    nAvail = p->buf.n - p->iBuf;
+    if( ppBlob && nReq==nBlob && nBlob<=nAvail ){
+      *ppBlob = (u8 *)&p->buf.z[p->iBuf];
+      p->iBuf += nBlob;
+      nReq = 0;
+    }else{
+      int nCopy = LSM_MIN(nAvail, nReq);
+      if( nBlob==nReq ){
+        pBuf->n = 0;
+      }
+      rc = lsmStringBinAppend(pBuf, (u8 *)&p->buf.z[p->iBuf], nCopy);
+      nReq -= nCopy;
+      p->iBuf += nCopy;
+      if( nReq==0 && ppBlob ){
+        *ppBlob = (u8*)pBuf->z;
+      }
+    }
+  }
+
+  *pRc = rc;
+}
+
+static void logReaderVarint(
+  LogReader *p, 
+  LsmString *pBuf,
+  int *piVal,                     /* OUT: Value read from log */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  if( *pRc==LSM_OK ){
+    u8 *aVarint;
+    if( p->buf.n==p->iBuf ){
+      logReaderBlob(p, 0, 10, &aVarint, pRc);
+      if( LSM_OK==*pRc ) p->iBuf -= (10 - lsmVarintGet32(aVarint, piVal));
+    }else{
+      logReaderBlob(p, pBuf, lsmVarintSize(p->buf.z[p->iBuf]), &aVarint, pRc);
+      if( LSM_OK==*pRc ) lsmVarintGet32(aVarint, piVal);
+    }
+  }
+}
+
+static void logReaderByte(LogReader *p, u8 *pByte, int *pRc){
+  u8 *pPtr = 0;
+  logReaderBlob(p, 0, 1, &pPtr, pRc);
+  if( pPtr ) *pByte = *pPtr;
+}
+
+static void logReaderCksum(LogReader *p, LsmString *pBuf, int *pbEof, int *pRc){
+  if( *pRc==LSM_OK ){
+    u8 *pPtr = 0;
+    u32 cksum0, cksum1;
+    int nCksum = p->iBuf - p->iCksumBuf;
+
+    /* Update in-memory (expected) checksums */
+    assert( nCksum>=0 );
+    logCksumUnaligned(&p->buf.z[p->iCksumBuf], nCksum, &p->cksum0, &p->cksum1);
+    p->iCksumBuf = p->iBuf + 8;
+    logReaderBlob(p, pBuf, 8, &pPtr, pRc);
+    assert( pPtr || *pRc );
+
+    /* Read the checksums from the log file. Set *pbEof if they do not match. */
+    if( pPtr ){
+      cksum0 = lsmGetU32(pPtr);
+      cksum1 = lsmGetU32(&pPtr[4]);
+      *pbEof = (cksum0!=p->cksum0 || cksum1!=p->cksum1);
+      p->iCksumBuf = p->iBuf;
+    }
+  }
+}
+
+static void logReaderInit(
+  lsm_db *pDb,                    /* Database handle */
+  DbLog *pLog,                    /* Log object associated with pDb */
+  int bInitBuf,                   /* True if p->buf is uninitialized */
+  LogReader *p                    /* Initialize this LogReader object */
+){
+  p->pFS = pDb->pFS;
+  p->iOff = pLog->aRegion[2].iStart;
+  p->cksum0 = pLog->cksum0;
+  p->cksum1 = pLog->cksum1;
+  if( bInitBuf ){ lsmStringInit(&p->buf, pDb->pEnv); }
+  p->buf.n = 0;
+  p->iCksumBuf = 0;
+  p->iBuf = 0;
+}
+
+/*
+** This function is called after reading the header of a LOG_DELETE or
+** LOG_WRITE record. Parameter nByte is the total size of the key and
+** value that follow the header just read. Return true if the size and
+** position of the record indicate that it should contain a checksum.
+*/
+static int logRequireCksum(LogReader *p, int nByte){
+  return ((p->iBuf + nByte - p->iCksumBuf) > LSM_CKSUM_MAXDATA);
+}
+
+/*
+** Recover the contents of the log file.
+*/
+int lsmLogRecover(lsm_db *pDb){
+  LsmString buf1;                 /* Key buffer */
+  LsmString buf2;                 /* Value buffer */
+  LogReader reader;               /* Log reader object */
+  int rc = LSM_OK;                /* Return code */
+  int nCommit = 0;                /* Number of transactions to recover */
+  int iPass;
+  int nJump = 0;                  /* Number of LSM_LOG_JUMP records in pass 0 */
+  DbLog *pLog;
+  int bOpen;
+
+  rc = lsmFsOpenLog(pDb, &bOpen);
+  if( rc!=LSM_OK ) return rc;
+
+  rc = lsmTreeInit(pDb);
+  if( rc!=LSM_OK ) return rc;
+
+  pLog = &pDb->treehdr.log;
+  lsmCheckpointLogoffset(pDb->pShmhdr->aSnap2, pLog);
+
+  logReaderInit(pDb, pLog, 1, &reader);
+  lsmStringInit(&buf1, pDb->pEnv);
+  lsmStringInit(&buf2, pDb->pEnv);
+
+  /* The outer for() loop runs at most twice. The first iteration is to 
+  ** count the number of committed transactions in the log. The second 
+  ** iterates through those transactions and updates the in-memory tree 
+  ** structure with their contents.  */
+  if( bOpen ){
+    for(iPass=0; iPass<2 && rc==LSM_OK; iPass++){
+      int bEof = 0;
+
+      while( rc==LSM_OK && !bEof ){
+        u8 eType = 0;
+        logReaderByte(&reader, &eType, &rc);
+
+        switch( eType ){
+          case LSM_LOG_PAD1:
+            break;
+
+          case LSM_LOG_PAD2: {
+            int nPad;
+            logReaderVarint(&reader, &buf1, &nPad, &rc);
+            logReaderBlob(&reader, &buf1, nPad, 0, &rc);
+            break;
+          }
+
+          case LSM_LOG_DRANGE:
+          case LSM_LOG_DRANGE_CKSUM:
+          case LSM_LOG_WRITE:
+          case LSM_LOG_WRITE_CKSUM: {
+            int nKey;
+            int nVal;
+            u8 *aVal;
+            logReaderVarint(&reader, &buf1, &nKey, &rc);
+            logReaderVarint(&reader, &buf2, &nVal, &rc);
+
+            if( eType==LSM_LOG_WRITE_CKSUM || eType==LSM_LOG_DRANGE_CKSUM ){
+              logReaderCksum(&reader, &buf1, &bEof, &rc);
+            }else{
+              bEof = logRequireCksum(&reader, nKey+nVal);
+            }
+            if( bEof ) break;
+
+            logReaderBlob(&reader, &buf1, nKey, 0, &rc);
+            logReaderBlob(&reader, &buf2, nVal, &aVal, &rc);
+            if( iPass==1 && rc==LSM_OK ){ 
+              if( eType==LSM_LOG_WRITE || eType==LSM_LOG_WRITE_CKSUM ){
+                rc = lsmTreeInsert(pDb, (u8 *)buf1.z, nKey, aVal, nVal);
+              }else{
+                rc = lsmTreeDelete(pDb, (u8 *)buf1.z, nKey, aVal, nVal);
+              }
+            }
+            break;
+          }
+
+          case LSM_LOG_DELETE:
+          case LSM_LOG_DELETE_CKSUM: {
+            int nKey; u8 *aKey;
+            logReaderVarint(&reader, &buf1, &nKey, &rc);
+
+            if( eType==LSM_LOG_DELETE_CKSUM ){
+              logReaderCksum(&reader, &buf1, &bEof, &rc);
+            }else{
+              bEof = logRequireCksum(&reader, nKey);
+            }
+            if( bEof ) break;
+
+            logReaderBlob(&reader, &buf1, nKey, &aKey, &rc);
+            if( iPass==1 && rc==LSM_OK ){ 
+              rc = lsmTreeInsert(pDb, aKey, nKey, NULL, -1);
+            }
+            break;
+          }
+
+          case LSM_LOG_COMMIT:
+            logReaderCksum(&reader, &buf1, &bEof, &rc);
+            if( bEof==0 ){
+              nCommit++;
+              assert( nCommit>0 || iPass==1 );
+              if( nCommit==0 ) bEof = 1;
+            }
+            break;
+
+          case LSM_LOG_JUMP: {
+            int iOff = 0;
+            logReaderVarint(&reader, &buf1, &iOff, &rc);
+            if( rc==LSM_OK ){
+              if( iPass==1 ){
+                if( pLog->aRegion[2].iStart==0 ){
+                  assert( pLog->aRegion[1].iStart==0 );
+                  pLog->aRegion[1].iEnd = reader.iOff;
+                }else{
+                  assert( pLog->aRegion[0].iStart==0 );
+                  pLog->aRegion[0].iStart = pLog->aRegion[2].iStart;
+                  pLog->aRegion[0].iEnd = reader.iOff-reader.buf.n+reader.iBuf;
+                }
+                pLog->aRegion[2].iStart = iOff;
+              }else{
+                if( (nJump++)==2 ){
+                  bEof = 1;
+                }
+              }
+
+              reader.iOff = iOff;
+              reader.buf.n = reader.iBuf;
+            }
+            break;
+          }
+
+          default:
+            /* Including LSM_LOG_EOF */
+            bEof = 1;
+            break;
+        }
+      }
+
+      if( rc==LSM_OK && iPass==0 ){
+        if( nCommit==0 ){
+          if( pLog->aRegion[2].iStart==0 ){
+            iPass = 1;
+          }else{
+            pLog->aRegion[2].iStart = 0;
+            iPass = -1;
+            lsmCheckpointZeroLogoffset(pDb);
+          }
+        }
+        logReaderInit(pDb, pLog, 0, &reader);
+        nCommit = nCommit * -1;
+      }
+    }
+  }
+
+  /* Initialize DbLog object */
+  if( rc==LSM_OK ){
+    pLog->aRegion[2].iEnd = reader.iOff - reader.buf.n + reader.iBuf;
+    pLog->cksum0 = reader.cksum0;
+    pLog->cksum1 = reader.cksum1;
+  }
+
+  if( rc==LSM_OK ){
+    rc = lsmFinishRecovery(pDb);
+  }else{
+    lsmFinishRecovery(pDb);
+  }
+
+  if( pDb->bRoTrans ){
+    lsmFsCloseLog(pDb);
+  }
+
+  lsmStringClear(&buf1);
+  lsmStringClear(&buf2);
+  lsmStringClear(&reader.buf);
+  return rc;
+}
+
+void lsmLogClose(lsm_db *db){
+  if( db->pLogWriter ){
+    lsmFree(db->pEnv, db->pLogWriter->buf.z);
+    lsmFree(db->pEnv, db->pLogWriter);
+    db->pLogWriter = 0;
+  }
+}
diff --git a/ext/lsm1/lsm_main.c b/ext/lsm1/lsm_main.c
new file mode 100644
index 0000000..a9c48e0
--- /dev/null
+++ b/ext/lsm1/lsm_main.c
@@ -0,0 +1,1008 @@
+/*
+** 2011-08-18
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** The main interface to the LSM module.
+*/
+#include "lsmInt.h"
+
+
+#ifdef LSM_DEBUG
+/*
+** This function returns a copy of its only argument.
+**
+** When the library is built with LSM_DEBUG defined, this function is called
+** whenever an error code is generated (not propagated - generated). So
+** if the library is mysteriously returning (say) LSM_IOERR, a breakpoint
+** may be set in this function to determine why.
+*/
+int lsmErrorBkpt(int rc){
+  /* Set breakpoint here! */
+  return rc;
+}
+
+/*
+** This function contains various assert() statements that test that the
+** lsm_db structure passed as an argument is internally consistent.
+*/
+static void assert_db_state(lsm_db *pDb){
+
+  /* If there is at least one cursor or a write transaction open, the database
+  ** handle must be holding a pointer to a client snapshot. And the reverse 
+  ** - if there are no open cursors and no write transactions then there must 
+  ** not be a client snapshot.  */
+  
+  assert( (pDb->pCsr!=0||pDb->nTransOpen>0)==(pDb->iReader>=0||pDb->bRoTrans) );
+
+  assert( (pDb->iReader<0 && pDb->bRoTrans==0) || pDb->pClient!=0 );
+
+  assert( pDb->nTransOpen>=0 );
+}
+#else
+# define assert_db_state(x) 
+#endif
+
+/*
+** The default key-compare function.
+*/
+static int xCmp(void *p1, int n1, void *p2, int n2){
+  int res;
+  res = memcmp(p1, p2, LSM_MIN(n1, n2));
+  if( res==0 ) res = (n1-n2);
+  return res;
+}
+
+static void xLog(void *pCtx, int rc, const char *z){
+  (void)(rc);
+  (void)(pCtx);
+  fprintf(stderr, "%s\n", z);
+  fflush(stderr);
+}
+
+/*
+** Allocate a new db handle.
+*/
+int lsm_new(lsm_env *pEnv, lsm_db **ppDb){
+  lsm_db *pDb;
+
+  /* If the user did not provide an environment, use the default. */
+  if( pEnv==0 ) pEnv = lsm_default_env();
+  assert( pEnv );
+
+  /* Allocate the new database handle */
+  *ppDb = pDb = (lsm_db *)lsmMallocZero(pEnv, sizeof(lsm_db));
+  if( pDb==0 ) return LSM_NOMEM_BKPT;
+
+  /* Initialize the new object */
+  pDb->pEnv = pEnv;
+  pDb->nTreeLimit = LSM_DFLT_AUTOFLUSH;
+  pDb->nAutockpt = LSM_DFLT_AUTOCHECKPOINT;
+  pDb->bAutowork = LSM_DFLT_AUTOWORK;
+  pDb->eSafety = LSM_DFLT_SAFETY;
+  pDb->xCmp = xCmp;
+  pDb->nDfltPgsz = LSM_DFLT_PAGE_SIZE;
+  pDb->nDfltBlksz = LSM_DFLT_BLOCK_SIZE;
+  pDb->nMerge = LSM_DFLT_AUTOMERGE;
+  pDb->nMaxFreelist = LSM_MAX_FREELIST_ENTRIES;
+  pDb->bUseLog = LSM_DFLT_USE_LOG;
+  pDb->iReader = -1;
+  pDb->iRwclient = -1;
+  pDb->bMultiProc = LSM_DFLT_MULTIPLE_PROCESSES;
+  pDb->iMmap = LSM_DFLT_MMAP;
+  pDb->xLog = xLog;
+  pDb->compress.iId = LSM_COMPRESSION_NONE;
+  return LSM_OK;
+}
+
+lsm_env *lsm_get_env(lsm_db *pDb){
+  assert( pDb->pEnv );
+  return pDb->pEnv;
+}
+
+/*
+** If database handle pDb is currently holding a client snapshot, but does
+** not have any open cursors or write transactions, release it.
+*/
+static void dbReleaseClientSnapshot(lsm_db *pDb){
+  if( pDb->nTransOpen==0 && pDb->pCsr==0 ){
+    lsmFinishReadTrans(pDb);
+  }
+}
+
+static int getFullpathname(
+  lsm_env *pEnv, 
+  const char *zRel,
+  char **pzAbs
+){
+  int nAlloc = 0;
+  char *zAlloc = 0;
+  int nReq = 0;
+  int rc;
+
+  do{
+    nAlloc = nReq;
+    rc = pEnv->xFullpath(pEnv, zRel, zAlloc, &nReq);
+    if( nReq>nAlloc ){
+      zAlloc = lsmReallocOrFreeRc(pEnv, zAlloc, nReq, &rc);
+    }
+  }while( nReq>nAlloc && rc==LSM_OK );
+
+  if( rc!=LSM_OK ){
+    lsmFree(pEnv, zAlloc);
+    zAlloc = 0;
+  }
+  *pzAbs = zAlloc;
+  return rc;
+}
+
+/*
+** Check that the bits in the db->mLock mask are consistent with the
+** value stored in db->iRwclient. An assert shall fail otherwise.
+*/
+static void assertRwclientLockValue(lsm_db *db){
+#ifndef NDEBUG
+  u64 msk;                        /* Mask of mLock bits for RWCLIENT locks */
+  u64 rwclient = 0;               /* Bit corresponding to db->iRwclient */
+
+  if( db->iRwclient>=0 ){
+    rwclient = ((u64)1 << (LSM_LOCK_RWCLIENT(db->iRwclient)-1));
+  }
+  msk  = ((u64)1 << (LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT)-1)) - 1;
+  msk -= (((u64)1 << (LSM_LOCK_RWCLIENT(0)-1)) - 1);
+
+  assert( (db->mLock & msk)==rwclient );
+#endif
+}
+
+/*
+** Open a new connection to database zFilename.
+*/
+int lsm_open(lsm_db *pDb, const char *zFilename){
+  int rc;
+
+  if( pDb->pDatabase ){
+    rc = LSM_MISUSE;
+  }else{
+    char *zFull;
+
+    /* Translate the possibly relative pathname supplied by the user into
+    ** an absolute pathname. This is required because the supplied path
+    ** is used (either directly or with "-log" appended to it) for more 
+    ** than one purpose - to open both the database and log files, and 
+    ** perhaps to unlink the log file during disconnection. An absolute
+    ** path is required to ensure that the correct files are operated
+    ** on even if the application changes the cwd.  */
+    rc = getFullpathname(pDb->pEnv, zFilename, &zFull);
+    assert( rc==LSM_OK || zFull==0 );
+
+    /* Connect to the database. */
+    if( rc==LSM_OK ){
+      rc = lsmDbDatabaseConnect(pDb, zFull);
+    }
+
+    if( pDb->bReadonly==0 ){
+      /* Configure the file-system connection with the page-size and block-size
+      ** of this database. Even if the database file is zero bytes in size
+      ** on disk, these values have been set in shared-memory by now, and so 
+      ** are guaranteed not to change during the lifetime of this connection.  
+      */
+      if( rc==LSM_OK && LSM_OK==(rc = lsmCheckpointLoad(pDb, 0)) ){
+        lsmFsSetPageSize(pDb->pFS, lsmCheckpointPgsz(pDb->aSnapshot));
+        lsmFsSetBlockSize(pDb->pFS, lsmCheckpointBlksz(pDb->aSnapshot));
+      }
+    }
+
+    lsmFree(pDb->pEnv, zFull);
+    assertRwclientLockValue(pDb);
+  }
+
+  assert( pDb->bReadonly==0 || pDb->bReadonly==1 );
+  assert( rc!=LSM_OK || (pDb->pShmhdr==0)==(pDb->bReadonly==1) );
+
+  return rc;
+}
+
+int lsm_close(lsm_db *pDb){
+  int rc = LSM_OK;
+  if( pDb ){
+    assert_db_state(pDb);
+    if( pDb->pCsr || pDb->nTransOpen ){
+      rc = LSM_MISUSE_BKPT;
+    }else{
+      lsmMCursorFreeCache(pDb);
+      lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
+      pDb->pClient = 0;
+
+      assertRwclientLockValue(pDb);
+
+      lsmDbDatabaseRelease(pDb);
+      lsmLogClose(pDb);
+      lsmFsClose(pDb->pFS);
+      /* assert( pDb->mLock==0 ); */
+      
+      /* Invoke any destructors registered for the compression or 
+      ** compression factory callbacks.  */
+      if( pDb->factory.xFree ) pDb->factory.xFree(pDb->factory.pCtx);
+      if( pDb->compress.xFree ) pDb->compress.xFree(pDb->compress.pCtx);
+
+      lsmFree(pDb->pEnv, pDb->rollback.aArray);
+      lsmFree(pDb->pEnv, pDb->aTrans);
+      lsmFree(pDb->pEnv, pDb->apShm);
+      lsmFree(pDb->pEnv, pDb);
+    }
+  }
+  return rc;
+}
+
+int lsm_config(lsm_db *pDb, int eParam, ...){
+  int rc = LSM_OK;
+  va_list ap;
+  va_start(ap, eParam);
+
+  switch( eParam ){
+    case LSM_CONFIG_AUTOFLUSH: {
+      /* This parameter is read and written in KB. But all internal 
+      ** processing is done in bytes.  */
+      int *piVal = va_arg(ap, int *);
+      int iVal = *piVal;
+      if( iVal>=0 && iVal<=(1024*1024) ){
+        pDb->nTreeLimit = iVal*1024;
+      }
+      *piVal = (pDb->nTreeLimit / 1024);
+      break;
+    }
+
+    case LSM_CONFIG_AUTOWORK: {
+      int *piVal = va_arg(ap, int *);
+      if( *piVal>=0 ){
+        pDb->bAutowork = *piVal;
+      }
+      *piVal = pDb->bAutowork;
+      break;
+    }
+
+    case LSM_CONFIG_AUTOCHECKPOINT: {
+      /* This parameter is read and written in KB. But all internal processing
+      ** (including the lsm_db.nAutockpt variable) is done in bytes.  */
+      int *piVal = va_arg(ap, int *);
+      if( *piVal>=0 ){
+        int iVal = *piVal;
+        pDb->nAutockpt = (i64)iVal * 1024;
+      }
+      *piVal = (int)(pDb->nAutockpt / 1024);
+      break;
+    }
+
+    case LSM_CONFIG_PAGE_SIZE: {
+      int *piVal = va_arg(ap, int *);
+      if( pDb->pDatabase ){
+        /* If lsm_open() has been called, this is a read-only parameter. 
+        ** Set the output variable to the page-size according to the 
+        ** FileSystem object.  */
+        *piVal = lsmFsPageSize(pDb->pFS);
+      }else{
+        if( *piVal>=256 && *piVal<=65536 && ((*piVal-1) & *piVal)==0 ){
+          pDb->nDfltPgsz = *piVal;
+        }else{
+          *piVal = pDb->nDfltPgsz;
+        }
+      }
+      break;
+    }
+
+    case LSM_CONFIG_BLOCK_SIZE: {
+      /* This parameter is read and written in KB. But all internal 
+      ** processing is done in bytes.  */
+      int *piVal = va_arg(ap, int *);
+      if( pDb->pDatabase ){
+        /* If lsm_open() has been called, this is a read-only parameter. 
+        ** Set the output variable to the block-size in KB according to the 
+        ** FileSystem object.  */
+        *piVal = lsmFsBlockSize(pDb->pFS) / 1024;
+      }else{
+        int iVal = *piVal;
+        if( iVal>=64 && iVal<=65536 && ((iVal-1) & iVal)==0 ){
+          pDb->nDfltBlksz = iVal * 1024;
+        }else{
+          *piVal = pDb->nDfltBlksz / 1024;
+        }
+      }
+      break;
+    }
+
+    case LSM_CONFIG_SAFETY: {
+      int *piVal = va_arg(ap, int *);
+      if( *piVal>=0 && *piVal<=2 ){
+        pDb->eSafety = *piVal;
+      }
+      *piVal = pDb->eSafety;
+      break;
+    }
+
+    case LSM_CONFIG_MMAP: {
+      int *piVal = va_arg(ap, int *);
+      if( pDb->iReader<0 && *piVal>=0 ){
+        pDb->iMmap = *piVal;
+        rc = lsmFsConfigure(pDb);
+      }
+      *piVal = pDb->iMmap;
+      break;
+    }
+
+    case LSM_CONFIG_USE_LOG: {
+      int *piVal = va_arg(ap, int *);
+      if( pDb->nTransOpen==0 && (*piVal==0 || *piVal==1) ){
+        pDb->bUseLog = *piVal;
+      }
+      *piVal = pDb->bUseLog;
+      break;
+    }
+
+    case LSM_CONFIG_AUTOMERGE: {
+      int *piVal = va_arg(ap, int *);
+      if( *piVal>1 ) pDb->nMerge = *piVal;
+      *piVal = pDb->nMerge;
+      break;
+    }
+
+    case LSM_CONFIG_MAX_FREELIST: {
+      int *piVal = va_arg(ap, int *);
+      if( *piVal>=2 && *piVal<=LSM_MAX_FREELIST_ENTRIES ){
+        pDb->nMaxFreelist = *piVal;
+      }
+      *piVal = pDb->nMaxFreelist;
+      break;
+    }
+
+    case LSM_CONFIG_MULTIPLE_PROCESSES: {
+      int *piVal = va_arg(ap, int *);
+      if( pDb->pDatabase ){
+        /* If lsm_open() has been called, this is a read-only parameter. 
+        ** Set the output variable to true if this connection is currently
+        ** in multi-process mode.  */
+        *piVal = lsmDbMultiProc(pDb);
+      }else{
+        pDb->bMultiProc = *piVal = (*piVal!=0);
+      }
+      break;
+    }
+
+    case LSM_CONFIG_READONLY: {
+      int *piVal = va_arg(ap, int *);
+      /* If lsm_open() has been called, this is a read-only parameter. */
+      if( pDb->pDatabase==0 && *piVal>=0 ){
+        pDb->bReadonly = *piVal = (*piVal!=0);
+      }
+      *piVal = pDb->bReadonly;
+      break;
+    }
+
+    case LSM_CONFIG_SET_COMPRESSION: {
+      lsm_compress *p = va_arg(ap, lsm_compress *);
+      if( pDb->iReader>=0 && pDb->bInFactory==0 ){
+        /* May not change compression schemes with an open transaction */
+        rc = LSM_MISUSE_BKPT;
+      }else{
+        if( pDb->compress.xFree ){
+          /* Invoke any destructor belonging to the current compression. */
+          pDb->compress.xFree(pDb->compress.pCtx);
+        }
+        if( p->xBound==0 ){
+          memset(&pDb->compress, 0, sizeof(lsm_compress));
+          pDb->compress.iId = LSM_COMPRESSION_NONE;
+        }else{
+          memcpy(&pDb->compress, p, sizeof(lsm_compress));
+        }
+        rc = lsmFsConfigure(pDb);
+      }
+      break;
+    }
+
+    case LSM_CONFIG_SET_COMPRESSION_FACTORY: {
+      lsm_compress_factory *p = va_arg(ap, lsm_compress_factory *);
+      if( pDb->factory.xFree ){
+        /* Invoke any destructor belonging to the current factory. */
+        pDb->factory.xFree(pDb->factory.pCtx);
+      }
+      memcpy(&pDb->factory, p, sizeof(lsm_compress_factory));
+      break;
+    }
+
+    case LSM_CONFIG_GET_COMPRESSION: {
+      lsm_compress *p = va_arg(ap, lsm_compress *);
+      memcpy(p, &pDb->compress, sizeof(lsm_compress));
+      break;
+    }
+
+    default:
+      rc = LSM_MISUSE;
+      break;
+  }
+
+  va_end(ap);
+  return rc;
+}
+
+void lsmAppendSegmentList(LsmString *pStr, char *zPre, Segment *pSeg){
+  lsmStringAppendf(pStr, "%s{%d %d %d %d}", zPre, 
+        pSeg->iFirst, pSeg->iLastPg, pSeg->iRoot, pSeg->nSize
+  );
+}
+
+static int infoGetWorker(lsm_db *pDb, Snapshot **pp, int *pbUnlock){
+  int rc = LSM_OK;
+
+  assert( *pbUnlock==0 );
+  if( !pDb->pWorker ){
+    rc = lsmBeginWork(pDb);
+    if( rc!=LSM_OK ) return rc;
+    *pbUnlock = 1;
+  }
+  if( pp ) *pp = pDb->pWorker;
+  return rc;
+}
+
+static void infoFreeWorker(lsm_db *pDb, int bUnlock){
+  if( bUnlock ){
+    int rcdummy = LSM_BUSY;
+    lsmFinishWork(pDb, 0, &rcdummy);
+  }
+}
+
+int lsmStructList(
+  lsm_db *pDb,                    /* Database handle */
+  char **pzOut                    /* OUT: Nul-terminated string (tcl list) */
+){
+  Level *pTopLevel = 0;           /* Top level of snapshot to report on */
+  int rc = LSM_OK;
+  Level *p;
+  LsmString s;
+  Snapshot *pWorker;              /* Worker snapshot */
+  int bUnlock = 0;
+
+  /* Obtain the worker snapshot */
+  rc = infoGetWorker(pDb, &pWorker, &bUnlock);
+  if( rc!=LSM_OK ) return rc;
+
+  /* Format the contents of the snapshot as text */
+  pTopLevel = lsmDbSnapshotLevel(pWorker);
+  lsmStringInit(&s, pDb->pEnv);
+  for(p=pTopLevel; rc==LSM_OK && p; p=p->pNext){
+    int i;
+    lsmStringAppendf(&s, "%s{%d", (s.n ? " " : ""), (int)p->iAge);
+    lsmAppendSegmentList(&s, " ", &p->lhs);
+    for(i=0; rc==LSM_OK && i<p->nRight; i++){
+      lsmAppendSegmentList(&s, " ", &p->aRhs[i]);
+    }
+    lsmStringAppend(&s, "}", 1);
+  }
+  rc = s.n>=0 ? LSM_OK : LSM_NOMEM;
+
+  /* Release the snapshot and return */
+  infoFreeWorker(pDb, bUnlock);
+  *pzOut = s.z;
+  return rc;
+}
+
+static int infoFreelistCb(void *pCtx, int iBlk, i64 iSnapshot){
+  LsmString *pStr = (LsmString *)pCtx;
+  lsmStringAppendf(pStr, "%s{%d %lld}", (pStr->n?" ":""), iBlk, iSnapshot);
+  return 0;
+}
+
+int lsmInfoFreelist(lsm_db *pDb, char **pzOut){
+  Snapshot *pWorker;              /* Worker snapshot */
+  int bUnlock = 0;
+  LsmString s;
+  int rc;
+
+  /* Obtain the worker snapshot */
+  rc = infoGetWorker(pDb, &pWorker, &bUnlock);
+  if( rc!=LSM_OK ) return rc;
+
+  lsmStringInit(&s, pDb->pEnv);
+  rc = lsmWalkFreelist(pDb, 0, infoFreelistCb, &s);
+  if( rc!=LSM_OK ){
+    lsmFree(pDb->pEnv, s.z);
+  }else{
+    *pzOut = s.z;
+  }
+
+  /* Release the snapshot and return */
+  infoFreeWorker(pDb, bUnlock);
+  return rc;
+}
+
+static int infoTreeSize(lsm_db *db, int *pnOldKB, int *pnNewKB){
+  ShmHeader *pShm = db->pShmhdr;
+  TreeHeader *p = &pShm->hdr1;
+
+  /* The following code suffers from two race conditions, as it accesses and
+  ** trusts the contents of shared memory without verifying checksums:
+  **
+  **   * The two values read - TreeHeader.root.nByte and oldroot.nByte - are 
+  **     32-bit fields. It is assumed that reading from one of these
+  **     is atomic - that it is not possible to read a partially written
+  **     garbage value. However the two values may be mutually inconsistent. 
+  **
+  **   * TreeHeader.iLogOff is a 64-bit value. And lsmCheckpointLogOffset()
+  **     reads a 64-bit value from a snapshot stored in shared memory. It
+  **     is assumed that in each case it is possible to read a partially
+  **     written garbage value. If this occurs, then the value returned
+  **     for the size of the "old" tree may reflect the size of an "old"
+  **     tree that was recently flushed to disk.
+  **
+  ** Given the context in which this function is called (as a result of an
+  ** lsm_info(LSM_INFO_TREE_SIZE) request), neither of these are considered to
+  ** be problems.
+  */
+  *pnNewKB = ((int)p->root.nByte + 1023) / 1024;
+  if( p->iOldShmid ){
+    if( p->iOldLog==lsmCheckpointLogOffset(pShm->aSnap1) ){
+      *pnOldKB = 0;
+    }else{
+      *pnOldKB = ((int)p->oldroot.nByte + 1023) / 1024;
+    }
+  }else{
+    *pnOldKB = 0;
+  }
+
+  return LSM_OK;
+}
+
+int lsm_info(lsm_db *pDb, int eParam, ...){
+  int rc = LSM_OK;
+  va_list ap;
+  va_start(ap, eParam);
+
+  switch( eParam ){
+    case LSM_INFO_NWRITE: {
+      int *piVal = va_arg(ap, int *);
+      *piVal = lsmFsNWrite(pDb->pFS);
+      break;
+    }
+
+    case LSM_INFO_NREAD: {
+      int *piVal = va_arg(ap, int *);
+      *piVal = lsmFsNRead(pDb->pFS);
+      break;
+    }
+
+    case LSM_INFO_DB_STRUCTURE: {
+      char **pzVal = va_arg(ap, char **);
+      rc = lsmStructList(pDb, pzVal);
+      break;
+    }
+
+    case LSM_INFO_ARRAY_STRUCTURE: {
+      LsmPgno pgno = va_arg(ap, LsmPgno);
+      char **pzVal = va_arg(ap, char **);
+      rc = lsmInfoArrayStructure(pDb, 0, pgno, pzVal);
+      break;
+    }
+
+    case LSM_INFO_ARRAY_PAGES: {
+      LsmPgno pgno = va_arg(ap, LsmPgno);
+      char **pzVal = va_arg(ap, char **);
+      rc = lsmInfoArrayPages(pDb, pgno, pzVal);
+      break;
+    }
+
+    case LSM_INFO_PAGE_HEX_DUMP:
+    case LSM_INFO_PAGE_ASCII_DUMP: {
+      LsmPgno pgno = va_arg(ap, LsmPgno);
+      char **pzVal = va_arg(ap, char **);
+      int bUnlock = 0;
+      rc = infoGetWorker(pDb, 0, &bUnlock);
+      if( rc==LSM_OK ){
+        int bHex = (eParam==LSM_INFO_PAGE_HEX_DUMP);
+        rc = lsmInfoPageDump(pDb, pgno, bHex, pzVal);
+      }
+      infoFreeWorker(pDb, bUnlock);
+      break;
+    }
+
+    case LSM_INFO_LOG_STRUCTURE: {
+      char **pzVal = va_arg(ap, char **);
+      rc = lsmInfoLogStructure(pDb, pzVal);
+      break;
+    }
+
+    case LSM_INFO_FREELIST: {
+      char **pzVal = va_arg(ap, char **);
+      rc = lsmInfoFreelist(pDb, pzVal);
+      break;
+    }
+
+    case LSM_INFO_CHECKPOINT_SIZE: {
+      int *pnKB = va_arg(ap, int *);
+      rc = lsmCheckpointSize(pDb, pnKB);
+      break;
+    }
+
+    case LSM_INFO_TREE_SIZE: {
+      int *pnOld = va_arg(ap, int *);
+      int *pnNew = va_arg(ap, int *);
+      rc = infoTreeSize(pDb, pnOld, pnNew);
+      break;
+    }
+
+    case LSM_INFO_COMPRESSION_ID: {
+      unsigned int *piOut = va_arg(ap, unsigned int *);
+      if( pDb->pClient ){
+        *piOut = pDb->pClient->iCmpId;
+      }else{
+        rc = lsmInfoCompressionId(pDb, piOut);
+      }
+      break;
+    }
+
+    default:
+      rc = LSM_MISUSE;
+      break;
+  }
+
+  va_end(ap);
+  return rc;
+}
+
+static int doWriteOp(
+  lsm_db *pDb,
+  int bDeleteRange,
+  const void *pKey, int nKey,     /* Key to write or delete */
+  const void *pVal, int nVal      /* Value to write. Or nVal==-1 for a delete */
+){
+  int rc = LSM_OK;                /* Return code */
+  int bCommit = 0;                /* True to commit before returning */
+
+  if( pDb->nTransOpen==0 ){
+    bCommit = 1;
+    rc = lsm_begin(pDb, 1);
+  }
+
+  if( rc==LSM_OK ){
+    int eType = (bDeleteRange ? LSM_DRANGE : (nVal>=0?LSM_WRITE:LSM_DELETE));
+    rc = lsmLogWrite(pDb, eType, (void *)pKey, nKey, (void *)pVal, nVal);
+  }
+
+  lsmSortedSaveTreeCursors(pDb);
+
+  if( rc==LSM_OK ){
+    int pgsz = lsmFsPageSize(pDb->pFS);
+    int nQuant = LSM_AUTOWORK_QUANT * pgsz;
+    int nBefore;
+    int nAfter;
+    int nDiff;
+
+    if( nQuant>pDb->nTreeLimit ){
+      nQuant = LSM_MAX(pDb->nTreeLimit, pgsz);
+    }
+
+    nBefore = lsmTreeSize(pDb);
+    if( bDeleteRange ){
+      rc = lsmTreeDelete(pDb, (void *)pKey, nKey, (void *)pVal, nVal);
+    }else{
+      rc = lsmTreeInsert(pDb, (void *)pKey, nKey, (void *)pVal, nVal);
+    }
+
+    nAfter = lsmTreeSize(pDb);
+    nDiff = (nAfter/nQuant) - (nBefore/nQuant);
+    if( rc==LSM_OK && pDb->bAutowork && nDiff!=0 ){
+      rc = lsmSortedAutoWork(pDb, nDiff * LSM_AUTOWORK_QUANT);
+    }
+  }
+
+  /* If a transaction was opened at the start of this function, commit it. 
+  ** Or, if an error has occurred, roll it back.  */
+  if( bCommit ){
+    if( rc==LSM_OK ){
+      rc = lsm_commit(pDb, 0);
+    }else{
+      lsm_rollback(pDb, 0);
+    }
+  }
+
+  return rc;
+}
+
+/* 
+** Write a new value into the database.
+*/
+int lsm_insert(
+  lsm_db *db,                     /* Database connection */
+  const void *pKey, int nKey,     /* Key to write or delete */
+  const void *pVal, int nVal      /* Value to write. Or nVal==-1 for a delete */
+){
+  return doWriteOp(db, 0, pKey, nKey, pVal, nVal);
+}
+
+/*
+** Delete a value from the database. 
+*/
+int lsm_delete(lsm_db *db, const void *pKey, int nKey){
+  return doWriteOp(db, 0, pKey, nKey, 0, -1);
+}
+
+/*
+** Delete a range of database keys.
+*/
+int lsm_delete_range(
+  lsm_db *db,                     /* Database handle */
+  const void *pKey1, int nKey1,   /* Lower bound of range to delete */
+  const void *pKey2, int nKey2    /* Upper bound of range to delete */
+){
+  int rc = LSM_OK;
+  if( db->xCmp((void *)pKey1, nKey1, (void *)pKey2, nKey2)<0 ){
+    rc = doWriteOp(db, 1, pKey1, nKey1, pKey2, nKey2);
+  }
+  return rc;
+}
+
+/*
+** Open a new cursor handle. 
+**
+** If there are currently no other open cursor handles, and no open write
+** transaction, open a read transaction here.
+*/
+int lsm_csr_open(lsm_db *pDb, lsm_cursor **ppCsr){
+  int rc = LSM_OK;                /* Return code */
+  MultiCursor *pCsr = 0;          /* New cursor object */
+
+  /* Open a read transaction if one is not already open. */
+  assert_db_state(pDb);
+
+  if( pDb->pShmhdr==0 ){
+    assert( pDb->bReadonly );
+    rc = lsmBeginRoTrans(pDb);
+  }else if( pDb->iReader<0 ){
+    rc = lsmBeginReadTrans(pDb);
+  }
+
+  /* Allocate the multi-cursor. */
+  if( rc==LSM_OK ){
+    rc = lsmMCursorNew(pDb, &pCsr);
+  }
+
+  /* If an error has occured, set the output to NULL and delete any partially
+  ** allocated cursor. If this means there are no open cursors, release the
+  ** client snapshot.  */
+  if( rc!=LSM_OK ){
+    lsmMCursorClose(pCsr, 0);
+    dbReleaseClientSnapshot(pDb);
+  }
+
+  assert_db_state(pDb);
+  *ppCsr = (lsm_cursor *)pCsr;
+  return rc;
+}
+
+/*
+** Close a cursor opened using lsm_csr_open().
+*/
+int lsm_csr_close(lsm_cursor *p){
+  if( p ){
+    lsm_db *pDb = lsmMCursorDb((MultiCursor *)p);
+    assert_db_state(pDb);
+    lsmMCursorClose((MultiCursor *)p, 1);
+    dbReleaseClientSnapshot(pDb);
+    assert_db_state(pDb);
+  }
+  return LSM_OK;
+}
+
+/*
+** Attempt to seek the cursor to the database entry specified by pKey/nKey.
+** If an error occurs (e.g. an OOM or IO error), return an LSM error code.
+** Otherwise, return LSM_OK.
+*/
+int lsm_csr_seek(lsm_cursor *pCsr, const void *pKey, int nKey, int eSeek){
+  return lsmMCursorSeek((MultiCursor *)pCsr, 0, (void *)pKey, nKey, eSeek);
+}
+
+int lsm_csr_next(lsm_cursor *pCsr){
+  return lsmMCursorNext((MultiCursor *)pCsr);
+}
+
+int lsm_csr_prev(lsm_cursor *pCsr){
+  return lsmMCursorPrev((MultiCursor *)pCsr);
+}
+
+int lsm_csr_first(lsm_cursor *pCsr){
+  return lsmMCursorFirst((MultiCursor *)pCsr);
+}
+
+int lsm_csr_last(lsm_cursor *pCsr){
+  return lsmMCursorLast((MultiCursor *)pCsr);
+}
+
+int lsm_csr_valid(lsm_cursor *pCsr){
+  return lsmMCursorValid((MultiCursor *)pCsr);
+}
+
+int lsm_csr_key(lsm_cursor *pCsr, const void **ppKey, int *pnKey){
+  return lsmMCursorKey((MultiCursor *)pCsr, (void **)ppKey, pnKey);
+}
+
+int lsm_csr_value(lsm_cursor *pCsr, const void **ppVal, int *pnVal){
+  return lsmMCursorValue((MultiCursor *)pCsr, (void **)ppVal, pnVal);
+}
+
+void lsm_config_log(
+  lsm_db *pDb, 
+  void (*xLog)(void *, int, const char *), 
+  void *pCtx
+){
+  pDb->xLog = xLog;
+  pDb->pLogCtx = pCtx;
+}
+
+void lsm_config_work_hook(
+  lsm_db *pDb, 
+  void (*xWork)(lsm_db *, void *), 
+  void *pCtx
+){
+  pDb->xWork = xWork;
+  pDb->pWorkCtx = pCtx;
+}
+
+void lsmLogMessage(lsm_db *pDb, int rc, const char *zFormat, ...){
+  if( pDb->xLog ){
+    LsmString s;
+    va_list ap, ap2;
+    lsmStringInit(&s, pDb->pEnv);
+    va_start(ap, zFormat);
+    va_start(ap2, zFormat);
+    lsmStringVAppendf(&s, zFormat, ap, ap2);
+    va_end(ap);
+    va_end(ap2);
+    pDb->xLog(pDb->pLogCtx, rc, s.z);
+    lsmStringClear(&s);
+  }
+}
+
+int lsm_begin(lsm_db *pDb, int iLevel){
+  int rc;
+
+  assert_db_state( pDb );
+  rc = (pDb->bReadonly ? LSM_READONLY : LSM_OK);
+
+  /* A value less than zero means open one more transaction. */
+  if( iLevel<0 ) iLevel = pDb->nTransOpen + 1;
+  if( iLevel>pDb->nTransOpen ){
+    int i;
+
+    /* Extend the pDb->aTrans[] array if required. */
+    if( rc==LSM_OK && pDb->nTransAlloc<iLevel ){
+      TransMark *aNew;            /* New allocation */
+      int nByte = sizeof(TransMark) * (iLevel+1);
+      aNew = (TransMark *)lsmRealloc(pDb->pEnv, pDb->aTrans, nByte);
+      if( !aNew ){
+        rc = LSM_NOMEM;
+      }else{
+        nByte = sizeof(TransMark) * (iLevel+1 - pDb->nTransAlloc);
+        memset(&aNew[pDb->nTransAlloc], 0, nByte);
+        pDb->nTransAlloc = iLevel+1;
+        pDb->aTrans = aNew;
+      }
+    }
+
+    if( rc==LSM_OK && pDb->nTransOpen==0 ){
+      rc = lsmBeginWriteTrans(pDb);
+    }
+
+    if( rc==LSM_OK ){
+      for(i=pDb->nTransOpen; i<iLevel; i++){
+        lsmTreeMark(pDb, &pDb->aTrans[i].tree);
+        lsmLogTell(pDb, &pDb->aTrans[i].log);
+      }
+      pDb->nTransOpen = iLevel;
+    }
+  }
+
+  return rc;
+}
+
+int lsm_commit(lsm_db *pDb, int iLevel){
+  int rc = LSM_OK;
+
+  assert_db_state( pDb );
+
+  /* A value less than zero means close the innermost nested transaction. */
+  if( iLevel<0 ) iLevel = LSM_MAX(0, pDb->nTransOpen - 1);
+
+  if( iLevel<pDb->nTransOpen ){
+    if( iLevel==0 ){
+      int rc2;
+      /* Commit the transaction to disk. */
+      if( rc==LSM_OK ) rc = lsmLogCommit(pDb);
+      if( rc==LSM_OK && pDb->eSafety==LSM_SAFETY_FULL ){
+        rc = lsmFsSyncLog(pDb->pFS);
+      }
+      rc2 = lsmFinishWriteTrans(pDb, (rc==LSM_OK));
+      if( rc==LSM_OK ) rc = rc2;
+    }
+    pDb->nTransOpen = iLevel;
+  }
+  dbReleaseClientSnapshot(pDb);
+  return rc;
+}
+
+int lsm_rollback(lsm_db *pDb, int iLevel){
+  int rc = LSM_OK;
+  assert_db_state( pDb );
+
+  if( pDb->nTransOpen ){
+    /* A value less than zero means close the innermost nested transaction. */
+    if( iLevel<0 ) iLevel = LSM_MAX(0, pDb->nTransOpen - 1);
+
+    if( iLevel<=pDb->nTransOpen ){
+      TransMark *pMark = &pDb->aTrans[(iLevel==0 ? 0 : iLevel-1)];
+      lsmTreeRollback(pDb, &pMark->tree);
+      if( iLevel ) lsmLogSeek(pDb, &pMark->log);
+      pDb->nTransOpen = iLevel;
+    }
+
+    if( pDb->nTransOpen==0 ){
+      lsmFinishWriteTrans(pDb, 0);
+    }
+    dbReleaseClientSnapshot(pDb);
+  }
+
+  return rc;
+}
+
+int lsm_get_user_version(lsm_db *pDb, unsigned int *piUsr){
+  int rc = LSM_OK;                /* Return code */
+
+  /* Open a read transaction if one is not already open. */
+  assert_db_state(pDb);
+  if( pDb->pShmhdr==0 ){
+    assert( pDb->bReadonly );
+    rc = lsmBeginRoTrans(pDb);
+  }else if( pDb->iReader<0 ){
+    rc = lsmBeginReadTrans(pDb);
+  }
+
+  /* Allocate the multi-cursor. */
+  if( rc==LSM_OK ){
+    *piUsr = pDb->treehdr.iUsrVersion;
+  }
+
+  dbReleaseClientSnapshot(pDb);
+  assert_db_state(pDb);
+  return rc;
+}
+
+int lsm_set_user_version(lsm_db *pDb, unsigned int iUsr){
+  int rc = LSM_OK;                /* Return code */
+  int bCommit = 0;                /* True to commit before returning */
+
+  if( pDb->nTransOpen==0 ){
+    bCommit = 1;
+    rc = lsm_begin(pDb, 1);
+  }
+
+  if( rc==LSM_OK ){
+    pDb->treehdr.iUsrVersion = iUsr;
+  }
+
+  /* If a transaction was opened at the start of this function, commit it. 
+  ** Or, if an error has occurred, roll it back.  */
+  if( bCommit ){
+    if( rc==LSM_OK ){
+      rc = lsm_commit(pDb, 0);
+    }else{
+      lsm_rollback(pDb, 0);
+    }
+  }
+
+  return rc;
+}
diff --git a/ext/lsm1/lsm_mem.c b/ext/lsm1/lsm_mem.c
new file mode 100644
index 0000000..13dd9fe
--- /dev/null
+++ b/ext/lsm1/lsm_mem.c
@@ -0,0 +1,104 @@
+/*
+** 2011-08-18
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** Helper routines for memory allocation.
+*/
+#include "lsmInt.h"
+
+/*
+** The following routines are called internally by LSM sub-routines. In
+** this case a valid environment pointer must be supplied.
+*/
+void *lsmMalloc(lsm_env *pEnv, size_t N){
+  assert( pEnv );
+  return pEnv->xMalloc(pEnv, N);
+}
+void lsmFree(lsm_env *pEnv, void *p){
+  assert( pEnv );
+  pEnv->xFree(pEnv, p);
+}
+void *lsmRealloc(lsm_env *pEnv, void *p, size_t N){
+  assert( pEnv );
+  return pEnv->xRealloc(pEnv, p, N);
+}
+
+/*
+** Core memory allocation routines for LSM.
+*/
+void *lsm_malloc(lsm_env *pEnv, size_t N){
+  return lsmMalloc(pEnv ? pEnv : lsm_default_env(), N);
+}
+void lsm_free(lsm_env *pEnv, void *p){
+  lsmFree(pEnv ? pEnv : lsm_default_env(), p);
+}
+void *lsm_realloc(lsm_env *pEnv, void *p, size_t N){
+  return lsmRealloc(pEnv ? pEnv : lsm_default_env(), p, N);
+}
+
+void *lsmMallocZero(lsm_env *pEnv, size_t N){
+  void *pRet;
+  assert( pEnv );
+  pRet = lsmMalloc(pEnv, N);
+  if( pRet ) memset(pRet, 0, N);
+  return pRet;
+}
+
+void *lsmMallocRc(lsm_env *pEnv, size_t N, int *pRc){
+  void *pRet = 0;
+  if( *pRc==LSM_OK ){
+    pRet = lsmMalloc(pEnv, N);
+    if( pRet==0 ){
+      *pRc = LSM_NOMEM_BKPT;
+    }
+  }
+  return pRet;
+}
+
+void *lsmMallocZeroRc(lsm_env *pEnv, size_t N, int *pRc){
+  void *pRet = 0;
+  if( *pRc==LSM_OK ){
+    pRet = lsmMallocZero(pEnv, N);
+    if( pRet==0 ){
+      *pRc = LSM_NOMEM_BKPT;
+    }
+  }
+  return pRet;
+}
+
+void *lsmReallocOrFree(lsm_env *pEnv, void *p, size_t N){
+  void *pNew;
+  pNew = lsm_realloc(pEnv, p, N);
+  if( !pNew ) lsm_free(pEnv, p);
+  return pNew;
+}
+
+void *lsmReallocOrFreeRc(lsm_env *pEnv, void *p, size_t N, int *pRc){
+  void *pRet = 0;
+  if( *pRc ){
+    lsmFree(pEnv, p);
+  }else{
+    pRet = lsmReallocOrFree(pEnv, p, N);
+    if( !pRet ) *pRc = LSM_NOMEM_BKPT;
+  }
+  return pRet;
+}
+
+char *lsmMallocStrdup(lsm_env *pEnv, const char *zIn){
+  int nByte;
+  char *zRet;
+  nByte = strlen(zIn);
+  zRet = lsmMalloc(pEnv, nByte+1);
+  if( zRet ){
+    memcpy(zRet, zIn, nByte+1);
+  }
+  return zRet;
+}
diff --git a/ext/lsm1/lsm_mutex.c b/ext/lsm1/lsm_mutex.c
new file mode 100644
index 0000000..cb99b2a
--- /dev/null
+++ b/ext/lsm1/lsm_mutex.c
@@ -0,0 +1,88 @@
+/*
+** 2012-01-30
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** Mutex functions for LSM.
+*/
+#include "lsmInt.h"
+
+/*
+** Allocate a new mutex.
+*/
+int lsmMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
+  return pEnv->xMutexNew(pEnv, ppNew);
+}
+
+/*
+** Return a handle for one of the static mutexes.
+*/
+int lsmMutexStatic(lsm_env *pEnv, int iMutex, lsm_mutex **ppStatic){
+  return pEnv->xMutexStatic(pEnv, iMutex, ppStatic);
+}
+
+/*
+** Free a mutex allocated by lsmMutexNew().
+*/
+void lsmMutexDel(lsm_env *pEnv, lsm_mutex *pMutex){
+  if( pMutex ) pEnv->xMutexDel(pMutex);
+}
+
+/*
+** Enter a mutex.
+*/
+void lsmMutexEnter(lsm_env *pEnv, lsm_mutex *pMutex){
+  pEnv->xMutexEnter(pMutex);
+}
+
+/*
+** Attempt to enter a mutex, but do not block. If successful, return zero.
+** Otherwise, if the mutex is already held by some other thread and is not
+** entered, return non zero.
+**
+** Each successful call to this function must be matched by a call to
+** lsmMutexLeave().
+*/
+int lsmMutexTry(lsm_env *pEnv, lsm_mutex *pMutex){
+  return pEnv->xMutexTry(pMutex);
+}
+
+/*
+** Leave a mutex.
+*/
+void lsmMutexLeave(lsm_env *pEnv, lsm_mutex *pMutex){
+  pEnv->xMutexLeave(pMutex);
+}
+
+#ifndef NDEBUG
+/*
+** Return non-zero if the mutex passed as the second argument is held
+** by the calling thread, or zero otherwise. If the implementation is not 
+** able to tell if the mutex is held by the caller, it should return
+** non-zero.
+**
+** This function is only used as part of assert() statements.
+*/
+int lsmMutexHeld(lsm_env *pEnv, lsm_mutex *pMutex){
+  return pEnv->xMutexHeld ? pEnv->xMutexHeld(pMutex) : 1;
+}
+
+/*
+** Return non-zero if the mutex passed as the second argument is not 
+** held by the calling thread, or zero otherwise. If the implementation 
+** is not able to tell if the mutex is held by the caller, it should 
+** return non-zero.
+**
+** This function is only used as part of assert() statements.
+*/
+int lsmMutexNotHeld(lsm_env *pEnv, lsm_mutex *pMutex){
+  return pEnv->xMutexNotHeld ? pEnv->xMutexNotHeld(pMutex) : 1;
+}
+#endif
diff --git a/ext/lsm1/lsm_shared.c b/ext/lsm1/lsm_shared.c
new file mode 100644
index 0000000..2fdacf1
--- /dev/null
+++ b/ext/lsm1/lsm_shared.c
@@ -0,0 +1,1976 @@
+/*
+** 2012-01-23
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** Utilities used to help multiple LSM clients to coexist within the
+** same process space.
+*/
+#include "lsmInt.h"
+
+/*
+** Global data. All global variables used by code in this file are grouped
+** into the following structure instance.
+**
+** pDatabase:
+**   Linked list of all Database objects allocated within this process.
+**   This list may not be traversed without holding the global mutex (see
+**   functions enterGlobalMutex() and leaveGlobalMutex()).
+*/
+static struct SharedData {
+  Database *pDatabase;            /* Linked list of all Database objects */
+} gShared;
+
+/*
+** Database structure. There is one such structure for each distinct 
+** database accessed by this process. They are stored in the singly linked 
+** list starting at global variable gShared.pDatabase. Database objects are 
+** reference counted. Once the number of connections to the associated
+** database drops to zero, they are removed from the linked list and deleted.
+**
+** pFile:
+**   In multi-process mode, this file descriptor is used to obtain locks 
+**   and to access shared-memory. In single process mode, its only job is
+**   to hold the exclusive lock on the file.
+**   
+*/
+struct Database {
+  /* Protected by the global mutex (enterGlobalMutex/leaveGlobalMutex): */
+  char *zName;                    /* Canonical path to database file */
+  int nName;                      /* strlen(zName) */
+  int nDbRef;                     /* Number of associated lsm_db handles */
+  Database *pDbNext;              /* Next Database structure in global list */
+
+  /* Protected by the local mutex (pClientMutex) */
+  int bReadonly;                  /* True if Database.pFile is read-only */
+  int bMultiProc;                 /* True if running in multi-process mode */
+  lsm_file *pFile;                /* Used for locks/shm in multi-proc mode */
+  LsmFile *pLsmFile;              /* List of deferred closes */
+  lsm_mutex *pClientMutex;        /* Protects the apShmChunk[] and pConn */
+  int nShmChunk;                  /* Number of entries in apShmChunk[] array */
+  void **apShmChunk;              /* Array of "shared" memory regions */
+  lsm_db *pConn;                  /* List of connections to this db. */
+};
+
+/*
+** Functions to enter and leave the global mutex. This mutex is used
+** to protect the global linked-list headed at gShared.pDatabase.
+*/
+static int enterGlobalMutex(lsm_env *pEnv){
+  lsm_mutex *p;
+  int rc = lsmMutexStatic(pEnv, LSM_MUTEX_GLOBAL, &p);
+  if( rc==LSM_OK ) lsmMutexEnter(pEnv, p);
+  return rc;
+}
+static void leaveGlobalMutex(lsm_env *pEnv){
+  lsm_mutex *p;
+  lsmMutexStatic(pEnv, LSM_MUTEX_GLOBAL, &p);
+  lsmMutexLeave(pEnv, p);
+}
+
+#ifdef LSM_DEBUG
+static int holdingGlobalMutex(lsm_env *pEnv){
+  lsm_mutex *p;
+  lsmMutexStatic(pEnv, LSM_MUTEX_GLOBAL, &p);
+  return lsmMutexHeld(pEnv, p);
+}
+#endif
+
+#if 0
+static void assertNotInFreelist(Freelist *p, int iBlk){
+  int i; 
+  for(i=0; i<p->nEntry; i++){
+    assert( p->aEntry[i].iBlk!=iBlk );
+  }
+}
+#else
+# define assertNotInFreelist(x,y)
+#endif
+
+/*
+** Append an entry to the free-list. If (iId==-1), this is a delete.
+*/
+int freelistAppend(lsm_db *db, u32 iBlk, i64 iId){
+  lsm_env *pEnv = db->pEnv;
+  Freelist *p;
+  int i; 
+
+  assert( iId==-1 || iId>=0 );
+  p = db->bUseFreelist ? db->pFreelist : &db->pWorker->freelist;
+
+  /* Extend the space allocated for the freelist, if required */
+  assert( p->nAlloc>=p->nEntry );
+  if( p->nAlloc==p->nEntry ){
+    int nNew; 
+    int nByte; 
+    FreelistEntry *aNew;
+
+    nNew = (p->nAlloc==0 ? 4 : p->nAlloc*2);
+    nByte = sizeof(FreelistEntry) * nNew;
+    aNew = (FreelistEntry *)lsmRealloc(pEnv, p->aEntry, nByte);
+    if( !aNew ) return LSM_NOMEM_BKPT;
+    p->nAlloc = nNew;
+    p->aEntry = aNew;
+  }
+
+  for(i=0; i<p->nEntry; i++){
+    assert( i==0 || p->aEntry[i].iBlk > p->aEntry[i-1].iBlk );
+    if( p->aEntry[i].iBlk>=iBlk ) break;
+  }
+
+  if( i<p->nEntry && p->aEntry[i].iBlk==iBlk ){
+    /* Clobber an existing entry */
+    p->aEntry[i].iId = iId;
+  }else{
+    /* Insert a new entry into the list */
+    int nByte = sizeof(FreelistEntry)*(p->nEntry-i);
+    memmove(&p->aEntry[i+1], &p->aEntry[i], nByte);
+    p->aEntry[i].iBlk = iBlk;
+    p->aEntry[i].iId = iId;
+    p->nEntry++;
+  }
+
+  return LSM_OK;
+}
+
+/*
+** This function frees all resources held by the Database structure passed
+** as the only argument.
+*/
+static void freeDatabase(lsm_env *pEnv, Database *p){
+  assert( holdingGlobalMutex(pEnv) );
+  if( p ){
+    /* Free the mutexes */
+    lsmMutexDel(pEnv, p->pClientMutex);
+
+    if( p->pFile ){
+      lsmEnvClose(pEnv, p->pFile);
+    }
+
+    /* Free the array of shm pointers */
+    lsmFree(pEnv, p->apShmChunk);
+
+    /* Free the memory allocated for the Database struct itself */
+    lsmFree(pEnv, p);
+  }
+}
+
+typedef struct DbTruncateCtx DbTruncateCtx;
+struct DbTruncateCtx {
+  int nBlock;
+  i64 iInUse;
+};
+
+static int dbTruncateCb(void *pCtx, int iBlk, i64 iSnapshot){
+  DbTruncateCtx *p = (DbTruncateCtx *)pCtx;
+  if( iBlk!=p->nBlock || (p->iInUse>=0 && iSnapshot>=p->iInUse) ) return 1;
+  p->nBlock--;
+  return 0;
+}
+
+static int dbTruncate(lsm_db *pDb, i64 iInUse){
+  int rc = LSM_OK;
+#if 0
+  int i;
+  DbTruncateCtx ctx;
+
+  assert( pDb->pWorker );
+  ctx.nBlock = pDb->pWorker->nBlock;
+  ctx.iInUse = iInUse;
+
+  rc = lsmWalkFreelist(pDb, 1, dbTruncateCb, (void *)&ctx);
+  for(i=ctx.nBlock+1; rc==LSM_OK && i<=pDb->pWorker->nBlock; i++){
+    rc = freelistAppend(pDb, i, -1);
+  }
+
+  if( rc==LSM_OK ){
+#ifdef LSM_LOG_FREELIST
+    if( ctx.nBlock!=pDb->pWorker->nBlock ){
+      lsmLogMessage(pDb, 0, 
+          "dbTruncate(): truncated db to %d blocks",ctx.nBlock
+      );
+    }
+#endif
+    pDb->pWorker->nBlock = ctx.nBlock;
+  }
+#endif
+  return rc;
+}
+
+
+/*
+** This function is called during database shutdown (when the number of
+** connections drops from one to zero). It truncates the database file
+** to as small a size as possible without truncating away any blocks that
+** contain data.
+*/
+static int dbTruncateFile(lsm_db *pDb){
+  int rc;
+
+  assert( pDb->pWorker==0 );
+  assert( lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) );
+  rc = lsmCheckpointLoadWorker(pDb);
+
+  if( rc==LSM_OK ){
+    DbTruncateCtx ctx;
+
+    /* Walk the database free-block-list in reverse order. Set ctx.nBlock
+    ** to the block number of the last block in the database that actually
+    ** contains data. */
+    ctx.nBlock = pDb->pWorker->nBlock;
+    ctx.iInUse = -1;
+    rc = lsmWalkFreelist(pDb, 1, dbTruncateCb, (void *)&ctx);
+
+    /* If the last block that contains data is not already the last block in
+    ** the database file, truncate the database file so that it is. */
+    if( rc==LSM_OK ){
+      rc = lsmFsTruncateDb(
+          pDb->pFS, (i64)ctx.nBlock*lsmFsBlockSize(pDb->pFS)
+      );
+    }
+  }
+
+  lsmFreeSnapshot(pDb->pEnv, pDb->pWorker);
+  pDb->pWorker = 0;
+  return rc;
+}
+
+static void doDbDisconnect(lsm_db *pDb){
+  int rc;
+
+  if( pDb->bReadonly ){
+    lsmShmLock(pDb, LSM_LOCK_DMS3, LSM_LOCK_UNLOCK, 0);
+  }else{
+    /* Block for an exclusive lock on DMS1. This lock serializes all calls
+    ** to doDbConnect() and doDbDisconnect() across all processes.  */
+    rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
+    if( rc==LSM_OK ){
+
+      lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);
+
+      /* Try an exclusive lock on DMS2. If successful, this is the last
+      ** connection to the database. In this case flush the contents of the
+      ** in-memory tree to disk and write a checkpoint.  */
+      rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 1, LSM_LOCK_EXCL);
+      if( rc==LSM_OK ){
+        rc = lsmShmTestLock(pDb, LSM_LOCK_CHECKPOINTER, 1, LSM_LOCK_EXCL);
+      }
+      if( rc==LSM_OK ){
+        int bReadonly = 0;        /* True if there exist read-only conns. */
+
+        /* Flush the in-memory tree, if required. If there is data to flush,
+        ** this will create a new client snapshot in Database.pClient. The
+        ** checkpoint (serialization) of this snapshot may be written to disk
+        ** by the following block.  
+        **
+        ** There is no need to take a WRITER lock here. That there are no 
+        ** other locks on DMS2 guarantees that there are no other read-write
+        ** connections at this time (and the lock on DMS1 guarantees that
+        ** no new ones may appear).
+        */
+        rc = lsmTreeLoadHeader(pDb, 0);
+        if( rc==LSM_OK && (lsmTreeHasOld(pDb) || lsmTreeSize(pDb)>0) ){
+          rc = lsmFlushTreeToDisk(pDb);
+        }
+
+        /* Now check if there are any read-only connections. If there are,
+        ** then do not truncate the db file or unlink the shared-memory 
+        ** region.  */
+        if( rc==LSM_OK ){
+          rc = lsmShmTestLock(pDb, LSM_LOCK_DMS3, 1, LSM_LOCK_EXCL);
+          if( rc==LSM_BUSY ){
+            bReadonly = 1;
+            rc = LSM_OK;
+          }
+        }
+
+        /* Write a checkpoint to disk. */
+        if( rc==LSM_OK ){
+          rc = lsmCheckpointWrite(pDb, 0);
+        }
+
+        /* If the checkpoint was written successfully, delete the log file
+        ** and, if possible, truncate the database file.  */
+        if( rc==LSM_OK ){
+          int bRotrans = 0;
+          Database *p = pDb->pDatabase;
+
+          /* The log file may only be deleted if there are no clients 
+          ** read-only clients running rotrans transactions.  */
+          rc = lsmDetectRoTrans(pDb, &bRotrans);
+          if( rc==LSM_OK && bRotrans==0 ){
+            lsmFsCloseAndDeleteLog(pDb->pFS);
+          }
+
+          /* The database may only be truncated if there exist no read-only
+          ** clients - either connected or running rotrans transactions. */
+          if( bReadonly==0 && bRotrans==0 ){
+            lsmFsUnmap(pDb->pFS);
+            dbTruncateFile(pDb);
+            if( p->pFile && p->bMultiProc ){
+              lsmEnvShmUnmap(pDb->pEnv, p->pFile, 1);
+            }
+          }
+        }
+      }
+    }
+
+    if( pDb->iRwclient>=0 ){
+      lsmShmLock(pDb, LSM_LOCK_RWCLIENT(pDb->iRwclient), LSM_LOCK_UNLOCK, 0);
+      pDb->iRwclient = -1;
+    }
+
+    lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
+  }
+  pDb->pShmhdr = 0;
+}
+
+static int doDbConnect(lsm_db *pDb){
+  const int nUsMax = 100000;      /* Max value for nUs */
+  int nUs = 1000;                 /* us to wait between DMS1 attempts */
+  int rc;
+
+  /* Obtain a pointer to the shared-memory header */
+  assert( pDb->pShmhdr==0 );
+  assert( pDb->bReadonly==0 );
+
+  /* Block for an exclusive lock on DMS1. This lock serializes all calls
+  ** to doDbConnect() and doDbDisconnect() across all processes.  */
+  while( 1 ){
+    rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
+    if( rc!=LSM_BUSY ) break;
+    lsmEnvSleep(pDb->pEnv, nUs);
+    nUs = nUs * 2;
+    if( nUs>nUsMax ) nUs = nUsMax;
+  }
+  if( rc==LSM_OK ){
+    rc = lsmShmCacheChunks(pDb, 1);
+  }
+  if( rc!=LSM_OK ) return rc;
+  pDb->pShmhdr = (ShmHeader *)pDb->apShm[0];
+
+  /* Try an exclusive lock on DMS2/DMS3. If successful, this is the first 
+  ** and only connection to the database. In this case initialize the 
+  ** shared-memory and run log file recovery.  */
+  assert( LSM_LOCK_DMS3==1+LSM_LOCK_DMS2 );
+  rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 2, LSM_LOCK_EXCL);
+  if( rc==LSM_OK ){
+    memset(pDb->pShmhdr, 0, sizeof(ShmHeader));
+    rc = lsmCheckpointRecover(pDb);
+    if( rc==LSM_OK ){
+      rc = lsmLogRecover(pDb);
+    }
+    if( rc==LSM_OK ){
+      ShmHeader *pShm = pDb->pShmhdr;
+      pShm->aReader[0].iLsmId = lsmCheckpointId(pShm->aSnap1, 0);
+      pShm->aReader[0].iTreeId = pDb->treehdr.iUsedShmid;
+    }
+  }else if( rc==LSM_BUSY ){
+    rc = LSM_OK;
+  }
+
+  /* Take a shared lock on DMS2. In multi-process mode this lock "cannot" 
+  ** fail, as connections may only hold an exclusive lock on DMS2 if they 
+  ** first hold an exclusive lock on DMS1. And this connection is currently 
+  ** holding the exclusive lock on DSM1. 
+  **
+  ** However, if some other connection has the database open in single-process
+  ** mode, this operation will fail. In this case, return the error to the
+  ** caller - the attempt to connect to the db has failed.
+  */
+  if( rc==LSM_OK ){
+    rc = lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_SHARED, 0);
+  }
+
+  /* If anything went wrong, unlock DMS2. Otherwise, try to take an exclusive
+  ** lock on one of the LSM_LOCK_RWCLIENT() locks. Unlock DMS1 in any case. */
+  if( rc!=LSM_OK ){
+    pDb->pShmhdr = 0;
+  }else{
+    int i;
+    for(i=0; i<LSM_LOCK_NRWCLIENT; i++){
+      int rc2 = lsmShmLock(pDb, LSM_LOCK_RWCLIENT(i), LSM_LOCK_EXCL, 0);
+      if( rc2==LSM_OK ) pDb->iRwclient = i;
+      if( rc2!=LSM_BUSY ){
+        rc = rc2;
+        break;
+      }
+    }
+  }
+  lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
+
+  return rc;
+}
+
+static int dbOpenSharedFd(lsm_env *pEnv, Database *p, int bRoOk){
+  int rc;
+
+  rc = lsmEnvOpen(pEnv, p->zName, 0, &p->pFile);
+  if( rc==LSM_IOERR && bRoOk ){
+    rc = lsmEnvOpen(pEnv, p->zName, LSM_OPEN_READONLY, &p->pFile);
+    p->bReadonly = 1;
+  }
+
+  return rc;
+}
+
+/*
+** Return a reference to the shared Database handle for the database 
+** identified by canonical path zName. If this is the first connection to
+** the named database, a new Database object is allocated. Otherwise, a
+** pointer to an existing object is returned.
+**
+** If successful, *ppDatabase is set to point to the shared Database 
+** structure and LSM_OK returned. Otherwise, *ppDatabase is set to NULL
+** and and LSM error code returned.
+**
+** Each successful call to this function should be (eventually) matched
+** by a call to lsmDbDatabaseRelease().
+*/
+int lsmDbDatabaseConnect(
+  lsm_db *pDb,                    /* Database handle */
+  const char *zName               /* Full-path to db file */
+){
+  lsm_env *pEnv = pDb->pEnv;
+  int rc;                         /* Return code */
+  Database *p = 0;                /* Pointer returned via *ppDatabase */
+  int nName = lsmStrlen(zName);
+
+  assert( pDb->pDatabase==0 );
+  rc = enterGlobalMutex(pEnv);
+  if( rc==LSM_OK ){
+
+    /* Search the global list for an existing object. TODO: Need something
+    ** better than the memcmp() below to figure out if a given Database
+    ** object represents the requested file.  */
+    for(p=gShared.pDatabase; p; p=p->pDbNext){
+      if( nName==p->nName && 0==memcmp(zName, p->zName, nName) ) break;
+    }
+
+    /* If no suitable Database object was found, allocate a new one. */
+    if( p==0 ){
+      p = (Database *)lsmMallocZeroRc(pEnv, sizeof(Database)+nName+1, &rc);
+
+      /* If the allocation was successful, fill in other fields and
+      ** allocate the client mutex. */ 
+      if( rc==LSM_OK ){
+        p->bMultiProc = pDb->bMultiProc;
+        p->zName = (char *)&p[1];
+        p->nName = nName;
+        memcpy((void *)p->zName, zName, nName+1);
+        rc = lsmMutexNew(pEnv, &p->pClientMutex);
+      }
+
+      /* If nothing has gone wrong so far, open the shared fd. And if that
+      ** succeeds and this connection requested single-process mode, 
+      ** attempt to take the exclusive lock on DMS2.  */
+      if( rc==LSM_OK ){
+        int bReadonly = (pDb->bReadonly && pDb->bMultiProc);
+        rc = dbOpenSharedFd(pDb->pEnv, p, bReadonly);
+      }
+
+      if( rc==LSM_OK && p->bMultiProc==0 ){
+        /* Hold an exclusive lock DMS1 while grabbing DMS2. This ensures
+        ** that any ongoing call to doDbDisconnect() (even one in another
+        ** process) is finished before proceeding.  */
+        assert( p->bReadonly==0 );
+        rc = lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS1, LSM_LOCK_EXCL);
+        if( rc==LSM_OK ){
+          rc = lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS2, LSM_LOCK_EXCL);
+          lsmEnvLock(pDb->pEnv, p->pFile, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK);
+        }
+      }
+
+      if( rc==LSM_OK ){
+        p->pDbNext = gShared.pDatabase;
+        gShared.pDatabase = p;
+      }else{
+        freeDatabase(pEnv, p);
+        p = 0;
+      }
+    }
+
+    if( p ){
+      p->nDbRef++;
+    }
+    leaveGlobalMutex(pEnv);
+
+    if( p ){
+      lsmMutexEnter(pDb->pEnv, p->pClientMutex);
+      pDb->pNext = p->pConn;
+      p->pConn = pDb;
+      lsmMutexLeave(pDb->pEnv, p->pClientMutex);
+    }
+  }
+
+  pDb->pDatabase = p;
+  if( rc==LSM_OK ){
+    assert( p );
+    rc = lsmFsOpen(pDb, zName, p->bReadonly);
+  }
+
+  /* If the db handle is read-write, then connect to the system now. Run
+  ** recovery as necessary. Or, if this is a read-only database handle,
+  ** defer attempting to connect to the system until a read-transaction
+  ** is opened.  */
+  if( rc==LSM_OK ){
+    rc = lsmFsConfigure(pDb);
+  }
+  if( rc==LSM_OK && pDb->bReadonly==0 ){
+    rc = doDbConnect(pDb);
+  }
+
+  return rc;
+}
+
+static void dbDeferClose(lsm_db *pDb){
+  if( pDb->pFS ){
+    LsmFile *pLsmFile;
+    Database *p = pDb->pDatabase;
+    pLsmFile = lsmFsDeferClose(pDb->pFS);
+    pLsmFile->pNext = p->pLsmFile;
+    p->pLsmFile = pLsmFile;
+  }
+}
+
+LsmFile *lsmDbRecycleFd(lsm_db *db){
+  LsmFile *pRet;
+  Database *p = db->pDatabase;
+  lsmMutexEnter(db->pEnv, p->pClientMutex);
+  if( (pRet = p->pLsmFile)!=0 ){
+    p->pLsmFile = pRet->pNext;
+  }
+  lsmMutexLeave(db->pEnv, p->pClientMutex);
+  return pRet;
+}
+
+/*
+** Release a reference to a Database object obtained from 
+** lsmDbDatabaseConnect(). There should be exactly one call to this function 
+** for each successful call to Find().
+*/
+void lsmDbDatabaseRelease(lsm_db *pDb){
+  Database *p = pDb->pDatabase;
+  if( p ){
+    lsm_db **ppDb;
+
+    if( pDb->pShmhdr ){
+      doDbDisconnect(pDb);
+    }
+
+    lsmFsUnmap(pDb->pFS);
+    lsmMutexEnter(pDb->pEnv, p->pClientMutex);
+    for(ppDb=&p->pConn; *ppDb!=pDb; ppDb=&((*ppDb)->pNext));
+    *ppDb = pDb->pNext;
+    dbDeferClose(pDb);
+    lsmMutexLeave(pDb->pEnv, p->pClientMutex);
+
+    enterGlobalMutex(pDb->pEnv);
+    p->nDbRef--;
+    if( p->nDbRef==0 ){
+      LsmFile *pIter;
+      LsmFile *pNext;
+      Database **pp;
+
+      /* Remove the Database structure from the linked list. */
+      for(pp=&gShared.pDatabase; *pp!=p; pp=&((*pp)->pDbNext));
+      *pp = p->pDbNext;
+
+      /* If they were allocated from the heap, free the shared memory chunks */
+      if( p->bMultiProc==0 ){
+        int i;
+        for(i=0; i<p->nShmChunk; i++){
+          lsmFree(pDb->pEnv, p->apShmChunk[i]);
+        }
+      }
+
+      /* Close any outstanding file descriptors */
+      for(pIter=p->pLsmFile; pIter; pIter=pNext){
+        pNext = pIter->pNext;
+        lsmEnvClose(pDb->pEnv, pIter->pFile);
+        lsmFree(pDb->pEnv, pIter);
+      }
+      freeDatabase(pDb->pEnv, p);
+    }
+    leaveGlobalMutex(pDb->pEnv);
+  }
+}
+
+Level *lsmDbSnapshotLevel(Snapshot *pSnapshot){
+  return pSnapshot->pLevel;
+}
+
+void lsmDbSnapshotSetLevel(Snapshot *pSnap, Level *pLevel){
+  pSnap->pLevel = pLevel;
+}
+
+/* TODO: Shuffle things around to get rid of this */
+static int firstSnapshotInUse(lsm_db *, i64 *);
+
+/* 
+** Context object used by the lsmWalkFreelist() utility. 
+*/
+typedef struct WalkFreelistCtx WalkFreelistCtx;
+struct WalkFreelistCtx {
+  lsm_db *pDb;
+  int bReverse;
+  Freelist *pFreelist;
+  int iFree;
+  int (*xUsr)(void *, int, i64);  /* User callback function */
+  void *pUsrctx;                  /* User callback context */
+  int bDone;                      /* Set to true after xUsr() returns true */
+};
+
+/* 
+** Callback used by lsmWalkFreelist().
+*/
+static int walkFreelistCb(void *pCtx, int iBlk, i64 iSnapshot){
+  WalkFreelistCtx *p = (WalkFreelistCtx *)pCtx;
+  const int iDir = (p->bReverse ? -1 : 1);
+  Freelist *pFree = p->pFreelist;
+
+  assert( p->bDone==0 );
+  assert( iBlk>=0 );
+  if( pFree ){
+    while( (p->iFree < pFree->nEntry) && p->iFree>=0 ){
+      FreelistEntry *pEntry = &pFree->aEntry[p->iFree];
+      if( (p->bReverse==0 && pEntry->iBlk>(u32)iBlk)
+       || (p->bReverse!=0 && pEntry->iBlk<(u32)iBlk)
+      ){
+        break;
+      }else{
+        p->iFree += iDir;
+        if( pEntry->iId>=0 
+            && p->xUsr(p->pUsrctx, pEntry->iBlk, pEntry->iId) 
+          ){
+          p->bDone = 1;
+          return 1;
+        }
+        if( pEntry->iBlk==(u32)iBlk ) return 0;
+      }
+    }
+  }
+
+  if( p->xUsr(p->pUsrctx, iBlk, iSnapshot) ){
+    p->bDone = 1;
+    return 1;
+  }
+  return 0;
+}
+
+/*
+** The database handle passed as the first argument must be the worker
+** connection. This function iterates through the contents of the current
+** free block list, invoking the supplied callback once for each list
+** element.
+**
+** The difference between this function and lsmSortedWalkFreelist() is
+** that lsmSortedWalkFreelist() only considers those free-list elements
+** stored within the LSM. This function also merges in any in-memory 
+** elements.
+*/
+int lsmWalkFreelist(
+  lsm_db *pDb,                    /* Database handle (must be worker) */
+  int bReverse,                   /* True to iterate from largest to smallest */
+  int (*x)(void *, int, i64),     /* Callback function */
+  void *pCtx                      /* First argument to pass to callback */
+){
+  const int iDir = (bReverse ? -1 : 1);
+  int rc;
+  int iCtx;
+
+  WalkFreelistCtx ctx[2];
+
+  ctx[0].pDb = pDb;
+  ctx[0].bReverse = bReverse;
+  ctx[0].pFreelist = &pDb->pWorker->freelist;
+  if( ctx[0].pFreelist && bReverse ){
+    ctx[0].iFree = ctx[0].pFreelist->nEntry-1;
+  }else{
+    ctx[0].iFree = 0;
+  }
+  ctx[0].xUsr = walkFreelistCb;
+  ctx[0].pUsrctx = (void *)&ctx[1];
+  ctx[0].bDone = 0;
+
+  ctx[1].pDb = pDb;
+  ctx[1].bReverse = bReverse;
+  ctx[1].pFreelist = pDb->pFreelist;
+  if( ctx[1].pFreelist && bReverse ){
+    ctx[1].iFree = ctx[1].pFreelist->nEntry-1;
+  }else{
+    ctx[1].iFree = 0;
+  }
+  ctx[1].xUsr = x;
+  ctx[1].pUsrctx = pCtx;
+  ctx[1].bDone = 0;
+
+  rc = lsmSortedWalkFreelist(pDb, bReverse, walkFreelistCb, (void *)&ctx[0]);
+
+  if( ctx[0].bDone==0 ){
+    for(iCtx=0; iCtx<2; iCtx++){
+      int i;
+      WalkFreelistCtx *p = &ctx[iCtx];
+      for(i=p->iFree; 
+          p->pFreelist && rc==LSM_OK && i<p->pFreelist->nEntry && i>=0;
+          i += iDir
+         ){
+        FreelistEntry *pEntry = &p->pFreelist->aEntry[i];
+        if( pEntry->iId>=0 && p->xUsr(p->pUsrctx, pEntry->iBlk, pEntry->iId) ){
+          return LSM_OK;
+        }
+      }
+    }
+  }
+
+  return rc;
+}
+
+
+typedef struct FindFreeblockCtx FindFreeblockCtx;
+struct FindFreeblockCtx {
+  i64 iInUse;
+  int iRet;
+  int bNotOne;
+};
+
+static int findFreeblockCb(void *pCtx, int iBlk, i64 iSnapshot){
+  FindFreeblockCtx *p = (FindFreeblockCtx *)pCtx;
+  if( iSnapshot<p->iInUse && (iBlk!=1 || p->bNotOne==0) ){
+    p->iRet = iBlk;
+    return 1;
+  }
+  return 0;
+}
+
+static int findFreeblock(lsm_db *pDb, i64 iInUse, int bNotOne, int *piRet){
+  int rc;                         /* Return code */
+  FindFreeblockCtx ctx;           /* Context object */
+
+  ctx.iInUse = iInUse;
+  ctx.iRet = 0;
+  ctx.bNotOne = bNotOne;
+  rc = lsmWalkFreelist(pDb, 0, findFreeblockCb, (void *)&ctx);
+  *piRet = ctx.iRet;
+
+  return rc;
+}
+
+/*
+** Allocate a new database file block to write data to, either by extending
+** the database file or by recycling a free-list entry. The worker snapshot 
+** must be held in order to call this function.
+**
+** If successful, *piBlk is set to the block number allocated and LSM_OK is
+** returned. Otherwise, *piBlk is zeroed and an lsm error code returned.
+*/
+int lsmBlockAllocate(lsm_db *pDb, int iBefore, int *piBlk){
+  Snapshot *p = pDb->pWorker;
+  int iRet = 0;                   /* Block number of allocated block */
+  int rc = LSM_OK;
+  i64 iInUse = 0;                 /* Snapshot id still in use */
+  i64 iSynced = 0;                /* Snapshot id synced to disk */
+
+  assert( p );
+
+#ifdef LSM_LOG_FREELIST
+  {
+    static int nCall = 0;
+    char *zFree = 0;
+    nCall++;
+    rc = lsmInfoFreelist(pDb, &zFree);
+    if( rc!=LSM_OK ) return rc;
+    lsmLogMessage(pDb, 0, "lsmBlockAllocate(): %d freelist: %s", nCall, zFree);
+    lsmFree(pDb->pEnv, zFree);
+  }
+#endif
+
+  /* Set iInUse to the smallest snapshot id that is either:
+  **
+  **   * Currently in use by a database client,
+  **   * May be used by a database client in the future, or
+  **   * Is the most recently checkpointed snapshot (i.e. the one that will
+  **     be used following recovery if a failure occurs at this point).
+  */
+  rc = lsmCheckpointSynced(pDb, &iSynced, 0, 0);
+  if( rc==LSM_OK && iSynced==0 ) iSynced = p->iId;
+  iInUse = iSynced;
+  if( rc==LSM_OK && pDb->iReader>=0 ){
+    assert( pDb->pClient );
+    iInUse = LSM_MIN(iInUse, pDb->pClient->iId);
+  }
+  if( rc==LSM_OK ) rc = firstSnapshotInUse(pDb, &iInUse);
+
+#ifdef LSM_LOG_FREELIST
+  {
+    lsmLogMessage(pDb, 0, "lsmBlockAllocate(): "
+        "snapshot-in-use: %lld (iSynced=%lld) (client-id=%lld)", 
+        iInUse, iSynced, (pDb->iReader>=0 ? pDb->pClient->iId : 0)
+    );
+  }
+#endif
+
+
+  /* Unless there exists a read-only transaction (which prevents us from
+  ** recycling any blocks regardless, query the free block list for a 
+  ** suitable block to reuse. 
+  **
+  ** It might seem more natural to check for a read-only transaction at
+  ** the start of this function. However, it is better do wait until after
+  ** the call to lsmCheckpointSynced() to do so.
+  */
+  if( rc==LSM_OK ){
+    int bRotrans;
+    rc = lsmDetectRoTrans(pDb, &bRotrans);
+
+    if( rc==LSM_OK && bRotrans==0 ){
+      rc = findFreeblock(pDb, iInUse, (iBefore>0), &iRet);
+    }
+  }
+
+  if( iBefore>0 && (iRet<=0 || iRet>=iBefore) ){
+    iRet = 0;
+
+  }else if( rc==LSM_OK ){
+    /* If a block was found in the free block list, use it and remove it from 
+    ** the list. Otherwise, if no suitable block was found, allocate one from
+    ** the end of the file.  */
+    if( iRet>0 ){
+#ifdef LSM_LOG_FREELIST
+      lsmLogMessage(pDb, 0, 
+          "reusing block %d (snapshot-in-use=%lld)", iRet, iInUse);
+#endif
+      rc = freelistAppend(pDb, iRet, -1);
+      if( rc==LSM_OK ){
+        rc = dbTruncate(pDb, iInUse);
+      }
+    }else{
+      iRet = ++(p->nBlock);
+#ifdef LSM_LOG_FREELIST
+      lsmLogMessage(pDb, 0, "extending file to %d blocks", iRet);
+#endif
+    }
+  }
+
+  assert( iBefore>0 || iRet>0 || rc!=LSM_OK );
+  *piBlk = iRet;
+  return rc;
+}
+
+/*
+** Free a database block. The worker snapshot must be held in order to call 
+** this function.
+**
+** If successful, LSM_OK is returned. Otherwise, an lsm error code (e.g. 
+** LSM_NOMEM).
+*/
+int lsmBlockFree(lsm_db *pDb, int iBlk){
+  Snapshot *p = pDb->pWorker;
+  assert( lsmShmAssertWorker(pDb) );
+
+#ifdef LSM_LOG_FREELIST
+  lsmLogMessage(pDb, LSM_OK, "lsmBlockFree(): Free block %d", iBlk);
+#endif
+
+  return freelistAppend(pDb, iBlk, p->iId);
+}
+
+/*
+** Refree a database block. The worker snapshot must be held in order to call 
+** this function.
+**
+** Refreeing is required when a block is allocated using lsmBlockAllocate()
+** but then not used. This function is used to push the block back onto
+** the freelist. Refreeing a block is different from freeing is, as a refreed
+** block may be reused immediately. Whereas a freed block can not be reused 
+** until (at least) after the next checkpoint.
+*/
+int lsmBlockRefree(lsm_db *pDb, int iBlk){
+  int rc = LSM_OK;                /* Return code */
+
+#ifdef LSM_LOG_FREELIST
+  lsmLogMessage(pDb, LSM_OK, "lsmBlockRefree(): Refree block %d", iBlk);
+#endif
+
+  rc = freelistAppend(pDb, iBlk, 0);
+  return rc;
+}
+
+/*
+** If required, copy a database checkpoint from shared memory into the
+** database itself.
+**
+** The WORKER lock must not be held when this is called. This is because
+** this function may indirectly call fsync(). And the WORKER lock should
+** not be held that long (in case it is required by a client flushing an
+** in-memory tree to disk).
+*/
+int lsmCheckpointWrite(lsm_db *pDb, u32 *pnWrite){
+  int rc;                         /* Return Code */
+  u32 nWrite = 0;
+
+  assert( pDb->pWorker==0 );
+  assert( 1 || pDb->pClient==0 );
+  assert( lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK) );
+
+  rc = lsmShmLock(pDb, LSM_LOCK_CHECKPOINTER, LSM_LOCK_EXCL, 0);
+  if( rc!=LSM_OK ) return rc;
+
+  rc = lsmCheckpointLoad(pDb, 0);
+  if( rc==LSM_OK ){
+    int nBlock = lsmCheckpointNBlock(pDb->aSnapshot);
+    ShmHeader *pShm = pDb->pShmhdr;
+    int bDone = 0;                /* True if checkpoint is already stored */
+
+    /* Check if this checkpoint has already been written to the database
+    ** file. If so, set variable bDone to true.  */
+    if( pShm->iMetaPage ){
+      MetaPage *pPg;              /* Meta page */
+      u8 *aData;                  /* Meta-page data buffer */
+      int nData;                  /* Size of aData[] in bytes */
+      i64 iCkpt;                  /* Id of checkpoint just loaded */
+      i64 iDisk = 0;              /* Id of checkpoint already stored in db */
+      iCkpt = lsmCheckpointId(pDb->aSnapshot, 0);
+      rc = lsmFsMetaPageGet(pDb->pFS, 0, pShm->iMetaPage, &pPg);
+      if( rc==LSM_OK ){
+        aData = lsmFsMetaPageData(pPg, &nData);
+        iDisk = lsmCheckpointId((u32 *)aData, 1);
+        nWrite = lsmCheckpointNWrite((u32 *)aData, 1);
+        lsmFsMetaPageRelease(pPg);
+      }
+      bDone = (iDisk>=iCkpt);
+    }
+
+    if( rc==LSM_OK && bDone==0 ){
+      int iMeta = (pShm->iMetaPage % 2) + 1;
+      if( pDb->eSafety!=LSM_SAFETY_OFF ){
+        rc = lsmFsSyncDb(pDb->pFS, nBlock);
+      }
+      if( rc==LSM_OK ) rc = lsmCheckpointStore(pDb, iMeta);
+      if( rc==LSM_OK && pDb->eSafety!=LSM_SAFETY_OFF){
+        rc = lsmFsSyncDb(pDb->pFS, 0);
+      }
+      if( rc==LSM_OK ){
+        pShm->iMetaPage = iMeta;
+        nWrite = lsmCheckpointNWrite(pDb->aSnapshot, 0) - nWrite;
+      }
+#ifdef LSM_LOG_WORK
+      lsmLogMessage(pDb, 0, "finish checkpoint %d", 
+          (int)lsmCheckpointId(pDb->aSnapshot, 0)
+      );
+#endif
+    }
+  }
+
+  lsmShmLock(pDb, LSM_LOCK_CHECKPOINTER, LSM_LOCK_UNLOCK, 0);
+  if( pnWrite && rc==LSM_OK ) *pnWrite = nWrite;
+  return rc;
+}
+
+int lsmBeginWork(lsm_db *pDb){
+  int rc;
+
+  /* Attempt to take the WORKER lock */
+  rc = lsmShmLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL, 0);
+
+  /* Deserialize the current worker snapshot */
+  if( rc==LSM_OK ){
+    rc = lsmCheckpointLoadWorker(pDb);
+  }
+  return rc;
+}
+
+void lsmFreeSnapshot(lsm_env *pEnv, Snapshot *p){
+  if( p ){
+    lsmSortedFreeLevel(pEnv, p->pLevel);
+    lsmFree(pEnv, p->freelist.aEntry);
+    lsmFree(pEnv, p->redirect.a);
+    lsmFree(pEnv, p);
+  }
+}
+
+/*
+** Attempt to populate one of the read-lock slots to contain lock values
+** iLsm/iShm. Or, if such a slot exists already, this function is a no-op.
+**
+** It is not an error if no slot can be populated because the write-lock
+** cannot be obtained. If any other error occurs, return an LSM error code.
+** Otherwise, LSM_OK.
+**
+** This function is called at various points to try to ensure that there
+** always exists at least one read-lock slot that can be used by a read-only
+** client. And so that, in the usual case, there is an "exact match" available
+** whenever a read transaction is opened by any client. At present this
+** function is called when:
+**
+**    * A write transaction that called lsmTreeDiscardOld() is committed, and
+**    * Whenever the working snapshot is updated (i.e. lsmFinishWork()).
+*/
+static int dbSetReadLock(lsm_db *db, i64 iLsm, u32 iShm){
+  int rc = LSM_OK;
+  ShmHeader *pShm = db->pShmhdr;
+  int i;
+
+  /* Check if there is already a slot containing the required values. */
+  for(i=0; i<LSM_LOCK_NREADER; i++){
+    ShmReader *p = &pShm->aReader[i];
+    if( p->iLsmId==iLsm && p->iTreeId==iShm ) return LSM_OK;
+  }
+
+  /* Iterate through all read-lock slots, attempting to take a write-lock
+  ** on each of them. If a write-lock succeeds, populate the locked slot
+  ** with the required values and break out of the loop.  */
+  for(i=0; rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
+    rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
+    if( rc==LSM_BUSY ){
+      rc = LSM_OK;
+    }else{
+      ShmReader *p = &pShm->aReader[i];
+      p->iLsmId = iLsm;
+      p->iTreeId = iShm;
+      lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_UNLOCK, 0);
+      break;
+    }
+  }
+
+  return rc;
+}
+
+/*
+** Release the read-lock currently held by connection db.
+*/
+int dbReleaseReadlock(lsm_db *db){
+  int rc = LSM_OK;
+  if( db->iReader>=0 ){
+    rc = lsmShmLock(db, LSM_LOCK_READER(db->iReader), LSM_LOCK_UNLOCK, 0);
+    db->iReader = -1;
+  }
+  db->bRoTrans = 0;
+  return rc;
+}
+
+
+/*
+** Argument bFlush is true if the contents of the in-memory tree has just
+** been flushed to disk. The significance of this is that once the snapshot
+** created to hold the updated state of the database is synced to disk, log
+** file space can be recycled.
+*/
+void lsmFinishWork(lsm_db *pDb, int bFlush, int *pRc){
+  int rc = *pRc;
+  assert( rc!=0 || pDb->pWorker );
+  if( pDb->pWorker ){
+    /* If no error has occurred, serialize the worker snapshot and write
+    ** it to shared memory.  */
+    if( rc==LSM_OK ){
+      rc = lsmSaveWorker(pDb, bFlush);
+    }
+
+    /* Assuming no error has occurred, update a read lock slot with the
+    ** new snapshot id (see comments above function dbSetReadLock()).  */
+    if( rc==LSM_OK ){
+      if( pDb->iReader<0 ){
+        rc = lsmTreeLoadHeader(pDb, 0);
+      }
+      if( rc==LSM_OK ){
+        rc = dbSetReadLock(pDb, pDb->pWorker->iId, pDb->treehdr.iUsedShmid);
+      }
+    }
+
+    /* Free the snapshot object. */
+    lsmFreeSnapshot(pDb->pEnv, pDb->pWorker);
+    pDb->pWorker = 0;
+  }
+
+  lsmShmLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK, 0);
+  *pRc = rc;
+}
+
+/*
+** Called when recovery is finished.
+*/
+int lsmFinishRecovery(lsm_db *pDb){
+  lsmTreeEndTransaction(pDb, 1);
+  return LSM_OK;
+}
+
+/*
+** Check if the currently configured compression functions
+** (LSM_CONFIG_SET_COMPRESSION) are compatible with a database that has its
+** compression id set to iReq. Compression routines are compatible if iReq
+** is zero (indicating the database is empty), or if it is equal to the 
+** compression id of the configured compression routines.
+**
+** If the check shows that the current compression are incompatible and there
+** is a compression factory registered, give it a chance to install new
+** compression routines.
+**
+** If, after any registered factory is invoked, the compression functions
+** are still incompatible, return LSM_MISMATCH. Otherwise, LSM_OK.
+*/
+int lsmCheckCompressionId(lsm_db *pDb, u32 iReq){
+  if( iReq!=LSM_COMPRESSION_EMPTY && pDb->compress.iId!=iReq ){
+    if( pDb->factory.xFactory ){
+      pDb->bInFactory = 1;
+      pDb->factory.xFactory(pDb->factory.pCtx, pDb, iReq);
+      pDb->bInFactory = 0;
+    }
+    if( pDb->compress.iId!=iReq ){
+      /* Incompatible */
+      return LSM_MISMATCH;
+    }
+  }
+  /* Compatible */
+  return LSM_OK;
+}
+
+/*
+** Begin a read transaction. This function is a no-op if the connection
+** passed as the only argument already has an open read transaction.
+*/
+int lsmBeginReadTrans(lsm_db *pDb){
+  const int MAX_READLOCK_ATTEMPTS = 10;
+  const int nMaxAttempt = (pDb->bRoTrans ? 1 : MAX_READLOCK_ATTEMPTS);
+
+  int rc = LSM_OK;                /* Return code */
+  int iAttempt = 0;
+
+  assert( pDb->pWorker==0 );
+
+  while( rc==LSM_OK && pDb->iReader<0 && (iAttempt++)<nMaxAttempt ){
+    int iTreehdr = 0;
+    int iSnap = 0;
+    assert( pDb->pCsr==0 && pDb->nTransOpen==0 );
+
+    /* Load the in-memory tree header. */
+    rc = lsmTreeLoadHeader(pDb, &iTreehdr);
+
+    /* Load the database snapshot */
+    if( rc==LSM_OK ){
+      if( lsmCheckpointClientCacheOk(pDb)==0 ){
+        lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
+        pDb->pClient = 0;
+        lsmMCursorFreeCache(pDb);
+        lsmFsPurgeCache(pDb->pFS);
+        rc = lsmCheckpointLoad(pDb, &iSnap);
+      }else{
+        iSnap = 1;
+      }
+    }
+
+    /* Take a read-lock on the tree and snapshot just loaded. Then check
+    ** that the shared-memory still contains the same values. If so, proceed.
+    ** Otherwise, relinquish the read-lock and retry the whole procedure
+    ** (starting with loading the in-memory tree header).  */
+    if( rc==LSM_OK ){
+      u32 iShmMax = pDb->treehdr.iUsedShmid;
+      u32 iShmMin = pDb->treehdr.iNextShmid+1-LSM_MAX_SHMCHUNKS;
+      rc = lsmReadlock(
+          pDb, lsmCheckpointId(pDb->aSnapshot, 0), iShmMin, iShmMax
+      );
+      if( rc==LSM_OK ){
+        if( lsmTreeLoadHeaderOk(pDb, iTreehdr)
+         && lsmCheckpointLoadOk(pDb, iSnap)
+        ){
+          /* Read lock has been successfully obtained. Deserialize the 
+          ** checkpoint just loaded. TODO: This will be removed after 
+          ** lsm_sorted.c is changed to work directly from the serialized
+          ** version of the snapshot.  */
+          if( pDb->pClient==0 ){
+            rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot,&pDb->pClient);
+          }
+          assert( (rc==LSM_OK)==(pDb->pClient!=0) );
+          assert( pDb->iReader>=0 );
+
+          /* Check that the client has the right compression hooks loaded.
+          ** If not, set rc to LSM_MISMATCH.  */
+          if( rc==LSM_OK ){
+            rc = lsmCheckCompressionId(pDb, pDb->pClient->iCmpId);
+          }
+        }else{
+          rc = dbReleaseReadlock(pDb);
+        }
+      }
+
+      if( rc==LSM_BUSY ){
+        rc = LSM_OK;
+      }
+    }
+#if 0
+if( rc==LSM_OK && pDb->pClient ){
+  fprintf(stderr, 
+      "reading %p: snapshot:%d used-shmid:%d trans-id:%d iOldShmid=%d\n",
+      (void *)pDb,
+      (int)pDb->pClient->iId, (int)pDb->treehdr.iUsedShmid, 
+      (int)pDb->treehdr.root.iTransId,
+      (int)pDb->treehdr.iOldShmid
+  );
+}
+#endif
+  }
+
+  if( rc==LSM_OK ){
+    rc = lsmShmCacheChunks(pDb, pDb->treehdr.nChunk);
+  }
+  if( rc!=LSM_OK ){
+    dbReleaseReadlock(pDb);
+  }
+  if( pDb->pClient==0 && rc==LSM_OK ) rc = LSM_BUSY;
+  return rc;
+}
+
+/*
+** This function is used by a read-write connection to determine if there
+** are currently one or more read-only transactions open on the database
+** (in this context a read-only transaction is one opened by a read-only
+** connection on a non-live database).
+**
+** If no error occurs, LSM_OK is returned and *pbExists is set to true if
+** some other connection has a read-only transaction open, or false 
+** otherwise. If an error occurs an LSM error code is returned and the final
+** value of *pbExist is undefined.
+*/
+int lsmDetectRoTrans(lsm_db *db, int *pbExist){
+  int rc;
+
+  /* Only a read-write connection may use this function. */
+  assert( db->bReadonly==0 );
+
+  rc = lsmShmTestLock(db, LSM_LOCK_ROTRANS, 1, LSM_LOCK_EXCL);
+  if( rc==LSM_BUSY ){
+    *pbExist = 1;
+    rc = LSM_OK;
+  }else{
+    *pbExist = 0;
+  }
+
+  return rc;
+}
+
+/*
+** db is a read-only database handle in the disconnected state. This function
+** attempts to open a read-transaction on the database. This may involve
+** connecting to the database system (opening shared memory etc.).
+*/
+int lsmBeginRoTrans(lsm_db *db){
+  int rc = LSM_OK;
+
+  assert( db->bReadonly && db->pShmhdr==0 );
+  assert( db->iReader<0 );
+
+  if( db->bRoTrans==0 ){
+
+    /* Attempt a shared-lock on DMS1. */
+    rc = lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_SHARED, 0);
+    if( rc!=LSM_OK ) return rc;
+
+    rc = lsmShmTestLock(
+        db, LSM_LOCK_RWCLIENT(0), LSM_LOCK_NREADER, LSM_LOCK_SHARED
+    );
+    if( rc==LSM_OK ){
+      /* System is not live. Take a SHARED lock on the ROTRANS byte and
+      ** release DMS1. Locking ROTRANS tells all read-write clients that they
+      ** may not recycle any disk space from within the database or log files,
+      ** as a read-only client may be using it.  */
+      rc = lsmShmLock(db, LSM_LOCK_ROTRANS, LSM_LOCK_SHARED, 0);
+      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
+
+      if( rc==LSM_OK ){
+        db->bRoTrans = 1;
+        rc = lsmShmCacheChunks(db, 1);
+        if( rc==LSM_OK ){
+          db->pShmhdr = (ShmHeader *)db->apShm[0];
+          memset(db->pShmhdr, 0, sizeof(ShmHeader));
+          rc = lsmCheckpointRecover(db);
+          if( rc==LSM_OK ){
+            rc = lsmLogRecover(db);
+          }
+        }
+      }
+    }else if( rc==LSM_BUSY ){
+      /* System is live! */
+      rc = lsmShmLock(db, LSM_LOCK_DMS3, LSM_LOCK_SHARED, 0);
+      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
+      if( rc==LSM_OK ){
+        rc = lsmShmCacheChunks(db, 1);
+        if( rc==LSM_OK ){
+          db->pShmhdr = (ShmHeader *)db->apShm[0];
+        }
+      }
+    }
+
+    if( rc==LSM_OK ){
+      rc = lsmBeginReadTrans(db);
+    }
+  }
+
+  return rc;
+}
+
+/*
+** Close the currently open read transaction.
+*/
+void lsmFinishReadTrans(lsm_db *pDb){
+
+  /* Worker connections should not be closing read transactions. And
+  ** read transactions should only be closed after all cursors and write
+  ** transactions have been closed. Finally pClient should be non-NULL
+  ** only iff pDb->iReader>=0.  */
+  assert( pDb->pWorker==0 );
+  assert( pDb->pCsr==0 && pDb->nTransOpen==0 );
+
+  if( pDb->bRoTrans ){
+    int i;
+    for(i=0; i<pDb->nShm; i++){
+      lsmFree(pDb->pEnv, pDb->apShm[i]);
+    }
+    lsmFree(pDb->pEnv, pDb->apShm);
+    pDb->apShm = 0;
+    pDb->nShm = 0;
+    pDb->pShmhdr = 0;
+
+    lsmShmLock(pDb, LSM_LOCK_ROTRANS, LSM_LOCK_UNLOCK, 0);
+  }
+  dbReleaseReadlock(pDb);
+}
+
+/*
+** Open a write transaction.
+*/
+int lsmBeginWriteTrans(lsm_db *pDb){
+  int rc = LSM_OK;                /* Return code */
+  ShmHeader *pShm = pDb->pShmhdr; /* Shared memory header */
+
+  assert( pDb->nTransOpen==0 );
+  assert( pDb->bDiscardOld==0 );
+  assert( pDb->bReadonly==0 );
+
+  /* If there is no read-transaction open, open one now. */
+  if( pDb->iReader<0 ){
+    rc = lsmBeginReadTrans(pDb);
+  }
+
+  /* Attempt to take the WRITER lock */
+  if( rc==LSM_OK ){
+    rc = lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_EXCL, 0);
+  }
+
+  /* If the previous writer failed mid-transaction, run emergency rollback. */
+  if( rc==LSM_OK && pShm->bWriter ){
+    rc = lsmTreeRepair(pDb);
+    if( rc==LSM_OK ) pShm->bWriter = 0;
+  }
+
+  /* Check that this connection is currently reading from the most recent
+  ** version of the database. If not, return LSM_BUSY.  */
+  if( rc==LSM_OK && memcmp(&pShm->hdr1, &pDb->treehdr, sizeof(TreeHeader)) ){
+    rc = LSM_BUSY;
+  }
+
+  if( rc==LSM_OK ){
+    rc = lsmLogBegin(pDb);
+  }
+
+  /* If everything was successful, set the "transaction-in-progress" flag
+  ** and return LSM_OK. Otherwise, if some error occurred, relinquish the 
+  ** WRITER lock and return an error code.  */
+  if( rc==LSM_OK ){
+    TreeHeader *p = &pDb->treehdr;
+    pShm->bWriter = 1;
+    p->root.iTransId++;
+    if( lsmTreeHasOld(pDb) && p->iOldLog==pDb->pClient->iLogOff ){
+      lsmTreeDiscardOld(pDb);
+      pDb->bDiscardOld = 1;
+    }
+  }else{
+    lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_UNLOCK, 0);
+    if( pDb->pCsr==0 ) lsmFinishReadTrans(pDb);
+  }
+  return rc;
+}
+
+/*
+** End the current write transaction. The connection is left with an open
+** read transaction. It is an error to call this if there is no open write 
+** transaction.
+**
+** If the transaction was committed, then a commit record has already been
+** written into the log file when this function is called. Or, if the
+** transaction was rolled back, both the log file and in-memory tree 
+** structure have already been restored. In either case, this function 
+** merely releases locks and other resources held by the write-transaction.
+**
+** LSM_OK is returned if successful, or an LSM error code otherwise.
+*/
+int lsmFinishWriteTrans(lsm_db *pDb, int bCommit){
+  int rc = LSM_OK;
+  int bFlush = 0;
+
+  lsmLogEnd(pDb, bCommit);
+  if( rc==LSM_OK && bCommit && lsmTreeSize(pDb)>pDb->nTreeLimit ){
+    bFlush = 1;
+    lsmTreeMakeOld(pDb);
+  }
+  lsmTreeEndTransaction(pDb, bCommit);
+
+  if( rc==LSM_OK ){
+    if( bFlush && pDb->bAutowork ){
+      rc = lsmSortedAutoWork(pDb, 1);
+    }else if( bCommit && pDb->bDiscardOld ){
+      rc = dbSetReadLock(pDb, pDb->pClient->iId, pDb->treehdr.iUsedShmid);
+    }
+  }
+  pDb->bDiscardOld = 0;
+  lsmShmLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_UNLOCK, 0);
+
+  if( bFlush && pDb->bAutowork==0 && pDb->xWork ){
+    pDb->xWork(pDb, pDb->pWorkCtx);
+  }
+  return rc;
+}
+
+
+/*
+** Return non-zero if the caller is holding the client mutex.
+*/
+#ifdef LSM_DEBUG
+int lsmHoldingClientMutex(lsm_db *pDb){
+  return lsmMutexHeld(pDb->pEnv, pDb->pDatabase->pClientMutex);
+}
+#endif
+
+static int slotIsUsable(ShmReader *p, i64 iLsm, u32 iShmMin, u32 iShmMax){
+  return( 
+      p->iLsmId && p->iLsmId<=iLsm 
+      && shm_sequence_ge(iShmMax, p->iTreeId)
+      && shm_sequence_ge(p->iTreeId, iShmMin)
+  );
+}
+
+/*
+** Obtain a read-lock on database version identified by the combination
+** of snapshot iLsm and tree iTree. Return LSM_OK if successful, or
+** an LSM error code otherwise.
+*/
+int lsmReadlock(lsm_db *db, i64 iLsm, u32 iShmMin, u32 iShmMax){
+  int rc = LSM_OK;
+  ShmHeader *pShm = db->pShmhdr;
+  int i;
+
+  assert( db->iReader<0 );
+  assert( shm_sequence_ge(iShmMax, iShmMin) );
+
+  /* This is a no-op if the read-only transaction flag is set. */
+  if( db->bRoTrans ){
+    db->iReader = 0;
+    return LSM_OK;
+  }
+
+  /* Search for an exact match. */
+  for(i=0; db->iReader<0 && rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
+    ShmReader *p = &pShm->aReader[i];
+    if( p->iLsmId==iLsm && p->iTreeId==iShmMax ){
+      rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_SHARED, 0);
+      if( rc==LSM_OK && p->iLsmId==iLsm && p->iTreeId==iShmMax ){
+        db->iReader = i;
+      }else if( rc==LSM_BUSY ){
+        rc = LSM_OK;
+      }
+    }
+  }
+
+  /* Try to obtain a write-lock on each slot, in order. If successful, set
+  ** the slot values to iLsm/iTree.  */
+  for(i=0; db->iReader<0 && rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
+    rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
+    if( rc==LSM_BUSY ){
+      rc = LSM_OK;
+    }else{
+      ShmReader *p = &pShm->aReader[i];
+      p->iLsmId = iLsm;
+      p->iTreeId = iShmMax;
+      rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_SHARED, 0);
+      assert( rc!=LSM_BUSY );
+      if( rc==LSM_OK ) db->iReader = i;
+    }
+  }
+
+  /* Search for any usable slot */
+  for(i=0; db->iReader<0 && rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
+    ShmReader *p = &pShm->aReader[i];
+    if( slotIsUsable(p, iLsm, iShmMin, iShmMax) ){
+      rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_SHARED, 0);
+      if( rc==LSM_OK && slotIsUsable(p, iLsm, iShmMin, iShmMax) ){
+        db->iReader = i;
+      }else if( rc==LSM_BUSY ){
+        rc = LSM_OK;
+      }
+    }
+  }
+
+  if( rc==LSM_OK && db->iReader<0 ){
+    rc = LSM_BUSY;
+  }
+  return rc;
+}
+
+/*
+** This is used to check if there exists a read-lock locking a particular
+** version of either the in-memory tree or database file. 
+**
+** If iLsmId is non-zero, then it is a snapshot id. If there exists a 
+** read-lock using this snapshot or newer, set *pbInUse to true. Or,
+** if there is no such read-lock, set it to false.
+**
+** Or, if iLsmId is zero, then iShmid is a shared-memory sequence id.
+** Search for a read-lock using this sequence id or newer. etc.
+*/
+static int isInUse(lsm_db *db, i64 iLsmId, u32 iShmid, int *pbInUse){
+  ShmHeader *pShm = db->pShmhdr;
+  int i;
+  int rc = LSM_OK;
+
+  for(i=0; rc==LSM_OK && i<LSM_LOCK_NREADER; i++){
+    ShmReader *p = &pShm->aReader[i];
+    if( p->iLsmId ){
+      if( (iLsmId!=0 && p->iLsmId!=0 && iLsmId>=p->iLsmId) 
+       || (iLsmId==0 && shm_sequence_ge(p->iTreeId, iShmid))
+      ){
+        rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
+        if( rc==LSM_OK ){
+          p->iLsmId = 0;
+          lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_UNLOCK, 0);
+        }
+      }
+    }
+  }
+
+  if( rc==LSM_BUSY ){
+    *pbInUse = 1;
+    return LSM_OK;
+  }
+  *pbInUse = 0;
+  return rc;
+}
+
+/*
+** This function is called by worker connections to determine the smallest
+** snapshot id that is currently in use by a database client. The worker
+** connection uses this result to determine whether or not it is safe to
+** recycle a database block.
+*/
+static int firstSnapshotInUse(
+  lsm_db *db,                     /* Database handle */
+  i64 *piInUse                    /* IN/OUT: Smallest snapshot id in use */
+){
+  ShmHeader *pShm = db->pShmhdr;
+  i64 iInUse = *piInUse;
+  int i;
+
+  assert( iInUse>0 );
+  for(i=0; i<LSM_LOCK_NREADER; i++){
+    ShmReader *p = &pShm->aReader[i];
+    if( p->iLsmId ){
+      i64 iThis = p->iLsmId;
+      if( iThis!=0 && iInUse>iThis ){
+        int rc = lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_EXCL, 0);
+        if( rc==LSM_OK ){
+          p->iLsmId = 0;
+          lsmShmLock(db, LSM_LOCK_READER(i), LSM_LOCK_UNLOCK, 0);
+        }else if( rc==LSM_BUSY ){
+          iInUse = iThis;
+        }else{
+          /* Some error other than LSM_BUSY. Return the error code to
+          ** the caller in this case.  */
+          return rc;
+        }
+      }
+    }
+  }
+
+  *piInUse = iInUse;
+  return LSM_OK;
+}
+
+int lsmTreeInUse(lsm_db *db, u32 iShmid, int *pbInUse){
+  if( db->treehdr.iUsedShmid==iShmid ){
+    *pbInUse = 1;
+    return LSM_OK;
+  }
+  return isInUse(db, 0, iShmid, pbInUse);
+}
+
+int lsmLsmInUse(lsm_db *db, i64 iLsmId, int *pbInUse){
+  if( db->pClient && db->pClient->iId<=iLsmId ){
+    *pbInUse = 1;
+    return LSM_OK;
+  }
+  return isInUse(db, iLsmId, 0, pbInUse);
+}
+
+/*
+** This function may only be called after a successful call to
+** lsmDbDatabaseConnect(). It returns true if the connection is in
+** multi-process mode, or false otherwise.
+*/
+int lsmDbMultiProc(lsm_db *pDb){
+  return pDb->pDatabase && pDb->pDatabase->bMultiProc;
+}
+
+
+/*************************************************************************
+**************************************************************************
+**************************************************************************
+**************************************************************************
+**************************************************************************
+*************************************************************************/
+
+/*
+** Ensure that database connection db has cached pointers to at least the 
+** first nChunk chunks of shared memory.
+*/
+int lsmShmCacheChunks(lsm_db *db, int nChunk){
+  int rc = LSM_OK;
+  if( nChunk>db->nShm ){
+    static const int NINCR = 16;
+    Database *p = db->pDatabase;
+    lsm_env *pEnv = db->pEnv;
+    int nAlloc;
+    int i;
+
+    /* Ensure that the db->apShm[] array is large enough. If an attempt to
+    ** allocate memory fails, return LSM_NOMEM immediately. The apShm[] array
+    ** is always extended in multiples of 16 entries - so the actual allocated
+    ** size can be inferred from nShm.  */ 
+    nAlloc = ((db->nShm + NINCR - 1) / NINCR) * NINCR;
+    while( nChunk>=nAlloc ){
+      void **apShm;
+      nAlloc += NINCR;
+      apShm = lsmRealloc(pEnv, db->apShm, sizeof(void*)*nAlloc);
+      if( !apShm ) return LSM_NOMEM_BKPT;
+      db->apShm = apShm;
+    }
+
+    if( db->bRoTrans ){
+      for(i=db->nShm; rc==LSM_OK && i<nChunk; i++){
+        db->apShm[i] = lsmMallocZeroRc(pEnv, LSM_SHM_CHUNK_SIZE, &rc);
+        db->nShm++;
+      }
+
+    }else{
+
+      /* Enter the client mutex */
+      lsmMutexEnter(pEnv, p->pClientMutex);
+
+      /* Extend the Database objects apShmChunk[] array if necessary. Using the
+       ** same pattern as for the lsm_db.apShm[] array above.  */
+      nAlloc = ((p->nShmChunk + NINCR - 1) / NINCR) * NINCR;
+      while( nChunk>=nAlloc ){
+        void **apShm;
+        nAlloc +=  NINCR;
+        apShm = lsmRealloc(pEnv, p->apShmChunk, sizeof(void*)*nAlloc);
+        if( !apShm ){
+          rc = LSM_NOMEM_BKPT;
+          break;
+        }
+        p->apShmChunk = apShm;
+      }
+
+      for(i=db->nShm; rc==LSM_OK && i<nChunk; i++){
+        if( i>=p->nShmChunk ){
+          void *pChunk = 0;
+          if( p->bMultiProc==0 ){
+            /* Single process mode */
+            pChunk = lsmMallocZeroRc(pEnv, LSM_SHM_CHUNK_SIZE, &rc);
+          }else{
+            /* Multi-process mode */
+            rc = lsmEnvShmMap(pEnv, p->pFile, i, LSM_SHM_CHUNK_SIZE, &pChunk);
+          }
+          if( rc==LSM_OK ){
+            p->apShmChunk[i] = pChunk;
+            p->nShmChunk++;
+          }
+        }
+        if( rc==LSM_OK ){
+          db->apShm[i] = p->apShmChunk[i];
+          db->nShm++;
+        }
+      }
+
+      /* Release the client mutex */
+      lsmMutexLeave(pEnv, p->pClientMutex);
+    }
+  }
+
+  return rc;
+}
+
+static int lockSharedFile(lsm_env *pEnv, Database *p, int iLock, int eOp){
+  int rc = LSM_OK;
+  if( p->bMultiProc ){
+    rc = lsmEnvLock(pEnv, p->pFile, iLock, eOp);
+  }
+  return rc;
+}
+
+/*
+** Test if it would be possible for connection db to obtain a lock of type
+** eType on the nLock locks starting at iLock. If so, return LSM_OK. If it
+** would not be possible to obtain the lock due to a lock held by another
+** connection, return LSM_BUSY. If an IO or other error occurs (i.e. in the 
+** lsm_env.xTestLock function), return some other LSM error code.
+**
+** Note that this function never actually locks the database - it merely
+** queries the system to see if there exists a lock that would prevent
+** it from doing so.
+*/
+int lsmShmTestLock(
+  lsm_db *db,
+  int iLock,
+  int nLock,
+  int eOp
+){
+  int rc = LSM_OK;
+  lsm_db *pIter;
+  Database *p = db->pDatabase;
+  int i;
+  u64 mask = 0;
+
+  for(i=iLock; i<(iLock+nLock); i++){
+    mask |= ((u64)1 << (iLock-1));
+    if( eOp==LSM_LOCK_EXCL ) mask |= ((u64)1 << (iLock+32-1));
+  }
+
+  lsmMutexEnter(db->pEnv, p->pClientMutex);
+  for(pIter=p->pConn; pIter; pIter=pIter->pNext){
+    if( pIter!=db && (pIter->mLock & mask) ){
+      assert( pIter!=db );
+      break;
+    }
+  }
+
+  if( pIter ){
+    rc = LSM_BUSY;
+  }else if( p->bMultiProc ){
+    rc = lsmEnvTestLock(db->pEnv, p->pFile, iLock, nLock, eOp);
+  }
+
+  lsmMutexLeave(db->pEnv, p->pClientMutex);
+  return rc;
+}
+
+/*
+** Attempt to obtain the lock identified by the iLock and bExcl parameters.
+** If successful, return LSM_OK. If the lock cannot be obtained because 
+** there exists some other conflicting lock, return LSM_BUSY. If some other
+** error occurs, return an LSM error code.
+**
+** Parameter iLock must be one of LSM_LOCK_WRITER, WORKER or CHECKPOINTER,
+** or else a value returned by the LSM_LOCK_READER macro.
+*/
+int lsmShmLock(
+  lsm_db *db, 
+  int iLock,
+  int eOp,                        /* One of LSM_LOCK_UNLOCK, SHARED or EXCL */
+  int bBlock                      /* True for a blocking lock */
+){
+  lsm_db *pIter;
+  const u64 me = ((u64)1 << (iLock-1));
+  const u64 ms = ((u64)1 << (iLock+32-1));
+  int rc = LSM_OK;
+  Database *p = db->pDatabase;
+
+  assert( eOp!=LSM_LOCK_EXCL || p->bReadonly==0 );
+  assert( iLock>=1 && iLock<=LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT-1) );
+  assert( LSM_LOCK_RWCLIENT(LSM_LOCK_NRWCLIENT-1)<=32 );
+  assert( eOp==LSM_LOCK_UNLOCK || eOp==LSM_LOCK_SHARED || eOp==LSM_LOCK_EXCL );
+
+  /* Check for a no-op. Proceed only if this is not one of those. */
+  if( (eOp==LSM_LOCK_UNLOCK && (db->mLock & (me|ms))!=0)
+   || (eOp==LSM_LOCK_SHARED && (db->mLock & (me|ms))!=ms)
+   || (eOp==LSM_LOCK_EXCL   && (db->mLock & me)==0)
+  ){
+    int nExcl = 0;                /* Number of connections holding EXCLUSIVE */
+    int nShared = 0;              /* Number of connections holding SHARED */
+    lsmMutexEnter(db->pEnv, p->pClientMutex);
+
+    /* Figure out the locks currently held by this process on iLock, not
+    ** including any held by connection db.  */
+    for(pIter=p->pConn; pIter; pIter=pIter->pNext){
+      assert( (pIter->mLock & me)==0 || (pIter->mLock & ms)!=0 );
+      if( pIter!=db ){
+        if( pIter->mLock & me ){
+          nExcl++;
+        }else if( pIter->mLock & ms ){
+          nShared++;
+        }
+      }
+    }
+    assert( nExcl==0 || nExcl==1 );
+    assert( nExcl==0 || nShared==0 );
+    assert( nExcl==0 || (db->mLock & (me|ms))==0 );
+
+    switch( eOp ){
+      case LSM_LOCK_UNLOCK:
+        if( nShared==0 ){
+          lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_UNLOCK);
+        }
+        db->mLock &= ~(me|ms);
+        break;
+
+      case LSM_LOCK_SHARED:
+        if( nExcl ){
+          rc = LSM_BUSY;
+        }else{
+          if( nShared==0 ){
+            rc = lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_SHARED);
+          }
+          if( rc==LSM_OK ){
+            db->mLock |= ms;
+            db->mLock &= ~me;
+          }
+        }
+        break;
+
+      default:
+        assert( eOp==LSM_LOCK_EXCL );
+        if( nExcl || nShared ){
+          rc = LSM_BUSY;
+        }else{
+          rc = lockSharedFile(db->pEnv, p, iLock, LSM_LOCK_EXCL);
+          if( rc==LSM_OK ){
+            db->mLock |= (me|ms);
+          }
+        }
+        break;
+    }
+
+    lsmMutexLeave(db->pEnv, p->pClientMutex);
+  }
+
+  return rc;
+}
+
+#ifdef LSM_DEBUG
+
+int shmLockType(lsm_db *db, int iLock){
+  const u64 me = ((u64)1 << (iLock-1));
+  const u64 ms = ((u64)1 << (iLock+32-1));
+
+  if( db->mLock & me ) return LSM_LOCK_EXCL;
+  if( db->mLock & ms ) return LSM_LOCK_SHARED;
+  return LSM_LOCK_UNLOCK;
+}
+
+/*
+** The arguments passed to this function are similar to those passed to
+** the lsmShmLock() function. However, instead of obtaining a new lock 
+** this function returns true if the specified connection already holds 
+** (or does not hold) such a lock, depending on the value of eOp. As
+** follows:
+**
+**   (eOp==LSM_LOCK_UNLOCK) -> true if db has no lock on iLock
+**   (eOp==LSM_LOCK_SHARED) -> true if db has at least a SHARED lock on iLock.
+**   (eOp==LSM_LOCK_EXCL)   -> true if db has an EXCLUSIVE lock on iLock.
+*/
+int lsmShmAssertLock(lsm_db *db, int iLock, int eOp){
+  int ret = 0;
+  int eHave;
+
+  assert( iLock>=1 && iLock<=LSM_LOCK_READER(LSM_LOCK_NREADER-1) );
+  assert( iLock<=16 );
+  assert( eOp==LSM_LOCK_UNLOCK || eOp==LSM_LOCK_SHARED || eOp==LSM_LOCK_EXCL );
+
+  eHave = shmLockType(db, iLock);
+
+  switch( eOp ){
+    case LSM_LOCK_UNLOCK:
+      ret = (eHave==LSM_LOCK_UNLOCK);
+      break;
+    case LSM_LOCK_SHARED:
+      ret = (eHave!=LSM_LOCK_UNLOCK);
+      break;
+    case LSM_LOCK_EXCL:
+      ret = (eHave==LSM_LOCK_EXCL);
+      break;
+    default:
+      assert( !"bad eOp value passed to lsmShmAssertLock()" );
+      break;
+  }
+
+  return ret;
+}
+
+int lsmShmAssertWorker(lsm_db *db){
+  return lsmShmAssertLock(db, LSM_LOCK_WORKER, LSM_LOCK_EXCL) && db->pWorker;
+}
+
+/*
+** This function does not contribute to library functionality, and is not
+** included in release builds. It is intended to be called from within
+** an interactive debugger.
+**
+** When called, this function prints a single line of human readable output
+** to stdout describing the locks currently held by the connection. For 
+** example:
+**
+**     (gdb) call print_db_locks(pDb)
+**     (shared on dms2) (exclusive on writer) 
+*/
+void print_db_locks(lsm_db *db){
+  int iLock;
+  for(iLock=0; iLock<16; iLock++){
+    int bOne = 0;
+    const char *azLock[] = {0, "shared", "exclusive"};
+    const char *azName[] = {
+      0, "dms1", "dms2", "writer", "worker", "checkpointer",
+      "reader0", "reader1", "reader2", "reader3", "reader4", "reader5"
+    };
+    int eHave = shmLockType(db, iLock);
+    if( azLock[eHave] ){
+      printf("%s(%s on %s)", (bOne?" ":""), azLock[eHave], azName[iLock]);
+      bOne = 1;
+    }
+  }
+  printf("\n");
+}
+void print_all_db_locks(lsm_db *db){
+  lsm_db *p;
+  for(p=db->pDatabase->pConn; p; p=p->pNext){
+    printf("%s connection %p ", ((p==db)?"*":""), p);
+    print_db_locks(p);
+  }
+}
+#endif
+
+void lsmShmBarrier(lsm_db *db){
+  lsmEnvShmBarrier(db->pEnv);
+}
+
+int lsm_checkpoint(lsm_db *pDb, int *pnKB){
+  int rc;                         /* Return code */
+  u32 nWrite = 0;                 /* Number of pages checkpointed */
+
+  /* Attempt the checkpoint. If successful, nWrite is set to the number of
+  ** pages written between this and the previous checkpoint.  */
+  rc = lsmCheckpointWrite(pDb, &nWrite);
+
+  /* If required, calculate the output variable (KB of data checkpointed). 
+  ** Set it to zero if an error occured.  */
+  if( pnKB ){
+    int nKB = 0;
+    if( rc==LSM_OK && nWrite ){
+      nKB = (((i64)nWrite * lsmFsPageSize(pDb->pFS)) + 1023) / 1024;
+    }
+    *pnKB = nKB;
+  }
+
+  return rc;
+}
diff --git a/ext/lsm1/lsm_sorted.c b/ext/lsm1/lsm_sorted.c
new file mode 100644
index 0000000..4a24e4b
--- /dev/null
+++ b/ext/lsm1/lsm_sorted.c
@@ -0,0 +1,6192 @@
+/*
+** 2011-08-14
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** PAGE FORMAT:
+**
+**   The maximum page size is 65536 bytes.
+**
+**   Since all records are equal to or larger than 2 bytes in size, and 
+**   some space within the page is consumed by the page footer, there must
+**   be less than 2^15 records on each page.
+**
+**   Each page ends with a footer that describes the pages contents. This
+**   footer serves as similar purpose to the page header in an SQLite database.
+**   A footer is used instead of a header because it makes it easier to
+**   populate a new page based on a sorted list of key/value pairs.
+**
+**   The footer consists of the following values (starting at the end of
+**   the page and continuing backwards towards the start). All values are
+**   stored as unsigned big-endian integers.
+**
+**     * Number of records on page (2 bytes).
+**     * Flags field (2 bytes).
+**     * Left-hand pointer value (8 bytes).
+**     * The starting offset of each record (2 bytes per record).
+**
+**   Records may span pages. Unless it happens to be an exact fit, the part
+**   of the final record that starts on page X that does not fit on page X
+**   is stored at the start of page (X+1). This means there may be pages where
+**   (N==0). And on most pages the first record that starts on the page will
+**   not start at byte offset 0. For example:
+**
+**      aaaaa bbbbb ccc <footer>    cc eeeee fffff g <footer>    gggg....
+**
+** RECORD FORMAT:
+** 
+**   The first byte of the record is a flags byte. It is a combination
+**   of the following flags (defined in lsmInt.h):
+**
+**       LSM_START_DELETE
+**       LSM_END_DELETE 
+**       LSM_POINT_DELETE
+**       LSM_INSERT    
+**       LSM_SEPARATOR
+**       LSM_SYSTEMKEY
+**
+**   Immediately following the type byte is a pointer to the smallest key 
+**   in the next file that is larger than the key in the current record. The 
+**   pointer is encoded as a varint. When added to the 32-bit page number 
+**   stored in the footer, it is the page number of the page that contains the
+**   smallest key in the next sorted file that is larger than this key. 
+**
+**   Next is the number of bytes in the key, encoded as a varint.
+**
+**   If the LSM_INSERT flag is set, the number of bytes in the value, as
+**   a varint, is next.
+**
+**   Finally, the blob of data containing the key, and for LSM_INSERT
+**   records, the value as well.
+*/
+
+#ifndef _LSM_INT_H
+# include "lsmInt.h"
+#endif
+
+#define LSM_LOG_STRUCTURE 0
+#define LSM_LOG_DATA      0
+
+/*
+** Macros to help decode record types.
+*/
+#define rtTopic(eType)       ((eType) & LSM_SYSTEMKEY)
+#define rtIsDelete(eType)    (((eType) & 0x0F)==LSM_POINT_DELETE)
+
+#define rtIsSeparator(eType) (((eType) & LSM_SEPARATOR)!=0)
+#define rtIsWrite(eType)     (((eType) & LSM_INSERT)!=0)
+#define rtIsSystem(eType)    (((eType) & LSM_SYSTEMKEY)!=0)
+
+/*
+** The following macros are used to access a page footer.
+*/
+#define SEGMENT_NRECORD_OFFSET(pgsz)        ((pgsz) - 2)
+#define SEGMENT_FLAGS_OFFSET(pgsz)          ((pgsz) - 2 - 2)
+#define SEGMENT_POINTER_OFFSET(pgsz)        ((pgsz) - 2 - 2 - 8)
+#define SEGMENT_CELLPTR_OFFSET(pgsz, iCell) ((pgsz) - 2 - 2 - 8 - 2 - (iCell)*2)
+
+#define SEGMENT_EOF(pgsz, nEntry) SEGMENT_CELLPTR_OFFSET(pgsz, nEntry-1)
+
+#define SEGMENT_BTREE_FLAG     0x0001
+#define PGFTR_SKIP_NEXT_FLAG   0x0002
+#define PGFTR_SKIP_THIS_FLAG   0x0004
+
+
+#ifndef LSM_SEGMENTPTR_FREE_THRESHOLD
+# define LSM_SEGMENTPTR_FREE_THRESHOLD 1024
+#endif
+
+typedef struct SegmentPtr SegmentPtr;
+typedef struct LsmBlob LsmBlob;
+
+struct LsmBlob {
+  lsm_env *pEnv;
+  void *pData;
+  int nData;
+  int nAlloc;
+};
+
+/*
+** A SegmentPtr object may be used for one of two purposes:
+**
+**   * To iterate and/or seek within a single Segment (the combination of a 
+**     main run and an optional sorted run).
+**
+**   * To iterate through the separators array of a segment.
+*/
+struct SegmentPtr {
+  Level *pLevel;                /* Level object segment is part of */
+  Segment *pSeg;                /* Segment to access */
+
+  /* Current page. See segmentPtrLoadPage(). */
+  Page *pPg;                    /* Current page */
+  u16 flags;                    /* Copy of page flags field */
+  int nCell;                    /* Number of cells on pPg */
+  LsmPgno iPtr;                 /* Base cascade pointer */
+
+  /* Current cell. See segmentPtrLoadCell() */
+  int iCell;                    /* Current record within page pPg */
+  int eType;                    /* Type of current record */
+  LsmPgno iPgPtr;               /* Cascade pointer offset */
+  void *pKey; int nKey;         /* Key associated with current record */
+  void *pVal; int nVal;         /* Current record value (eType==WRITE only) */
+
+  /* Blobs used to allocate buffers for pKey and pVal as required */
+  LsmBlob blob1;
+  LsmBlob blob2;
+};
+
+/*
+** Used to iterate through the keys stored in a b-tree hierarchy from start
+** to finish. Only First() and Next() operations are required.
+**
+**   btreeCursorNew()
+**   btreeCursorFirst()
+**   btreeCursorNext()
+**   btreeCursorFree()
+**   btreeCursorPosition()
+**   btreeCursorRestore()
+*/
+typedef struct BtreePg BtreePg;
+typedef struct BtreeCursor BtreeCursor;
+struct BtreePg {
+  Page *pPage;
+  int iCell;
+};
+struct BtreeCursor {
+  Segment *pSeg;                  /* Iterate through this segments btree */
+  FileSystem *pFS;                /* File system to read pages from */
+  int nDepth;                     /* Allocated size of aPg[] */
+  int iPg;                        /* Current entry in aPg[]. -1 -> EOF. */
+  BtreePg *aPg;                   /* Pages from root to current location */
+
+  /* Cache of current entry. pKey==0 for EOF. */
+  void *pKey;
+  int nKey;
+  int eType;
+  LsmPgno iPtr;
+
+  /* Storage for key, if not local */
+  LsmBlob blob;
+};
+
+
+/*
+** A cursor used for merged searches or iterations through up to one
+** Tree structure and any number of sorted files.
+**
+**   lsmMCursorNew()
+**   lsmMCursorSeek()
+**   lsmMCursorNext()
+**   lsmMCursorPrev()
+**   lsmMCursorFirst()
+**   lsmMCursorLast()
+**   lsmMCursorKey()
+**   lsmMCursorValue()
+**   lsmMCursorValid()
+**
+** iFree:
+**   This variable is only used by cursors providing input data for a
+**   new top-level segment. Such cursors only ever iterate forwards, not
+**   backwards.
+*/
+struct MultiCursor {
+  lsm_db *pDb;                    /* Connection that owns this cursor */
+  MultiCursor *pNext;             /* Next cursor owned by connection pDb */
+  int flags;                      /* Mask of CURSOR_XXX flags */
+
+  int eType;                      /* Cache of current key type */
+  LsmBlob key;                    /* Cache of current key (or NULL) */
+  LsmBlob val;                    /* Cache of current value */
+
+  /* All the component cursors: */
+  TreeCursor *apTreeCsr[2];       /* Up to two tree cursors */
+  int iFree;                      /* Next element of free-list (-ve for eof) */
+  SegmentPtr *aPtr;               /* Array of segment pointers */
+  int nPtr;                       /* Size of array aPtr[] */
+  BtreeCursor *pBtCsr;            /* b-tree cursor (db writes only) */
+
+  /* Comparison results */
+  int nTree;                      /* Size of aTree[] array */
+  int *aTree;                     /* Array of comparison results */
+
+  /* Used by cursors flushing the in-memory tree only */
+  void *pSystemVal;               /* Pointer to buffer to free */
+
+  /* Used by worker cursors only */
+  LsmPgno *pPrevMergePtr;
+};
+
+/*
+** The following constants are used to assign integers to each component
+** cursor of a multi-cursor.
+*/
+#define CURSOR_DATA_TREE0     0   /* Current tree cursor (apTreeCsr[0]) */
+#define CURSOR_DATA_TREE1     1   /* The "old" tree, if any (apTreeCsr[1]) */
+#define CURSOR_DATA_SYSTEM    2   /* Free-list entries (new-toplevel only) */
+#define CURSOR_DATA_SEGMENT   3   /* First segment pointer (aPtr[0]) */
+
+/*
+** CURSOR_IGNORE_DELETE
+**   If set, this cursor will not visit SORTED_DELETE keys.
+**
+** CURSOR_FLUSH_FREELIST
+**   This cursor is being used to create a new toplevel. It should also 
+**   iterate through the contents of the in-memory free block list.
+**
+** CURSOR_IGNORE_SYSTEM
+**   If set, this cursor ignores system keys.
+**
+** CURSOR_NEXT_OK
+**   Set if it is Ok to call lsm_csr_next().
+**
+** CURSOR_PREV_OK
+**   Set if it is Ok to call lsm_csr_prev().
+**
+** CURSOR_READ_SEPARATORS
+**   Set if this cursor should visit the separator keys in segment 
+**   aPtr[nPtr-1].
+**
+** CURSOR_SEEK_EQ
+**   Cursor has undergone a successful lsm_csr_seek(LSM_SEEK_EQ) operation.
+**   The key and value are stored in MultiCursor.key and MultiCursor.val
+**   respectively.
+*/
+#define CURSOR_IGNORE_DELETE    0x00000001
+#define CURSOR_FLUSH_FREELIST   0x00000002
+#define CURSOR_IGNORE_SYSTEM    0x00000010
+#define CURSOR_NEXT_OK          0x00000020
+#define CURSOR_PREV_OK          0x00000040
+#define CURSOR_READ_SEPARATORS  0x00000080
+#define CURSOR_SEEK_EQ          0x00000100
+
+typedef struct MergeWorker MergeWorker;
+typedef struct Hierarchy Hierarchy;
+
+struct Hierarchy {
+  Page **apHier;
+  int nHier;
+};
+
+/*
+** aSave:
+**   When mergeWorkerNextPage() is called to advance to the next page in
+**   the output segment, if the bStore flag for an element of aSave[] is
+**   true, it is cleared and the corresponding iPgno value is set to the 
+**   page number of the page just completed.
+**
+**   aSave[0] is used to record the pointer value to be pushed into the
+**   b-tree hierarchy. aSave[1] is used to save the page number of the
+**   page containing the indirect key most recently written to the b-tree.
+**   see mergeWorkerPushHierarchy() for details.
+*/
+struct MergeWorker {
+  lsm_db *pDb;                    /* Database handle */
+  Level *pLevel;                  /* Worker snapshot Level being merged */
+  MultiCursor *pCsr;              /* Cursor to read new segment contents from */
+  int bFlush;                     /* True if this is an in-memory tree flush */
+  Hierarchy hier;                 /* B-tree hierarchy under construction */
+  Page *pPage;                    /* Current output page */
+  int nWork;                      /* Number of calls to mergeWorkerNextPage() */
+  LsmPgno *aGobble;               /* Gobble point for each input segment */
+
+  LsmPgno iIndirect;
+  struct SavedPgno {
+    LsmPgno iPgno;
+    int bStore;
+  } aSave[2];
+};
+
+#ifdef LSM_DEBUG_EXPENSIVE
+static int assertPointersOk(lsm_db *, Segment *, Segment *, int);
+static int assertBtreeOk(lsm_db *, Segment *);
+static void assertRunInOrder(lsm_db *pDb, Segment *pSeg);
+#else
+#define assertRunInOrder(x,y)
+#define assertBtreeOk(x,y)
+#endif
+
+
+struct FilePage { u8 *aData; int nData; };
+static u8 *fsPageData(Page *pPg, int *pnData){
+  *pnData = ((struct FilePage *)(pPg))->nData;
+  return ((struct FilePage *)(pPg))->aData;
+}
+/*UNUSED static u8 *fsPageDataPtr(Page *pPg){
+  return ((struct FilePage *)(pPg))->aData;
+}*/
+
+/*
+** Write nVal as a 16-bit unsigned big-endian integer into buffer aOut.
+*/
+void lsmPutU16(u8 *aOut, u16 nVal){
+  aOut[0] = (u8)((nVal>>8) & 0xFF);
+  aOut[1] = (u8)(nVal & 0xFF);
+}
+
+void lsmPutU32(u8 *aOut, u32 nVal){
+  aOut[0] = (u8)((nVal>>24) & 0xFF);
+  aOut[1] = (u8)((nVal>>16) & 0xFF);
+  aOut[2] = (u8)((nVal>> 8) & 0xFF);
+  aOut[3] = (u8)((nVal    ) & 0xFF);
+}
+
+int lsmGetU16(u8 *aOut){
+  return (aOut[0] << 8) + aOut[1];
+}
+
+u32 lsmGetU32(u8 *aOut){
+  return ((u32)aOut[0] << 24) 
+       + ((u32)aOut[1] << 16) 
+       + ((u32)aOut[2] << 8) 
+       + ((u32)aOut[3]);
+}
+
+u64 lsmGetU64(u8 *aOut){
+  return ((u64)aOut[0] << 56) 
+       + ((u64)aOut[1] << 48) 
+       + ((u64)aOut[2] << 40) 
+       + ((u64)aOut[3] << 32) 
+       + ((u64)aOut[4] << 24)
+       + ((u32)aOut[5] << 16) 
+       + ((u32)aOut[6] << 8) 
+       + ((u32)aOut[7]);
+}
+
+void lsmPutU64(u8 *aOut, u64 nVal){
+  aOut[0] = (u8)((nVal>>56) & 0xFF);
+  aOut[1] = (u8)((nVal>>48) & 0xFF);
+  aOut[2] = (u8)((nVal>>40) & 0xFF);
+  aOut[3] = (u8)((nVal>>32) & 0xFF);
+  aOut[4] = (u8)((nVal>>24) & 0xFF);
+  aOut[5] = (u8)((nVal>>16) & 0xFF);
+  aOut[6] = (u8)((nVal>> 8) & 0xFF);
+  aOut[7] = (u8)((nVal    ) & 0xFF);
+}
+
+static int sortedBlobGrow(lsm_env *pEnv, LsmBlob *pBlob, int nData){
+  assert( pBlob->pEnv==pEnv || (pBlob->pEnv==0 && pBlob->pData==0) );
+  if( pBlob->nAlloc<nData ){
+    pBlob->pData = lsmReallocOrFree(pEnv, pBlob->pData, nData);
+    if( !pBlob->pData ) return LSM_NOMEM_BKPT;
+    pBlob->nAlloc = nData;
+    pBlob->pEnv = pEnv;
+  }
+  return LSM_OK;
+}
+
+static int sortedBlobSet(lsm_env *pEnv, LsmBlob *pBlob, void *pData, int nData){
+  if( sortedBlobGrow(pEnv, pBlob, nData) ) return LSM_NOMEM;
+  memcpy(pBlob->pData, pData, nData);
+  pBlob->nData = nData;
+  return LSM_OK;
+}
+
+#if 0
+static int sortedBlobCopy(LsmBlob *pDest, LsmBlob *pSrc){
+  return sortedBlobSet(pDest, pSrc->pData, pSrc->nData);
+}
+#endif
+
+static void sortedBlobFree(LsmBlob *pBlob){
+  assert( pBlob->pEnv || pBlob->pData==0 );
+  if( pBlob->pData ) lsmFree(pBlob->pEnv, pBlob->pData);
+  memset(pBlob, 0, sizeof(LsmBlob));
+}
+
+static int sortedReadData(
+  Segment *pSeg,
+  Page *pPg,
+  int iOff,
+  int nByte,
+  void **ppData,
+  LsmBlob *pBlob
+){
+  int rc = LSM_OK;
+  int iEnd;
+  int nData;
+  int nCell;
+  u8 *aData;
+
+  aData = fsPageData(pPg, &nData);
+  nCell = lsmGetU16(&aData[SEGMENT_NRECORD_OFFSET(nData)]);
+  iEnd = SEGMENT_EOF(nData, nCell);
+  assert( iEnd>0 && iEnd<nData );
+
+  if( iOff+nByte<=iEnd ){
+    *ppData = (void *)&aData[iOff];
+  }else{
+    int nRem = nByte;
+    int i = iOff;
+    u8 *aDest;
+
+    /* Make sure the blob is big enough to store the value being loaded. */
+    rc = sortedBlobGrow(lsmPageEnv(pPg), pBlob, nByte);
+    if( rc!=LSM_OK ) return rc;
+    pBlob->nData = nByte;
+    aDest = (u8 *)pBlob->pData;
+    *ppData = pBlob->pData;
+
+    /* Increment the pointer pages ref-count. */
+    lsmFsPageRef(pPg);
+
+    while( rc==LSM_OK ){
+      Page *pNext;
+      int flags;
+
+      /* Copy data from pPg into the output buffer. */
+      int nCopy = LSM_MIN(nRem, iEnd-i);
+      if( nCopy>0 ){
+        memcpy(&aDest[nByte-nRem], &aData[i], nCopy);
+        nRem -= nCopy;
+        i += nCopy;
+        assert( nRem==0 || i==iEnd );
+      }
+      assert( nRem>=0 );
+      if( nRem==0 ) break;
+      i -= iEnd;
+
+      /* Grab the next page in the segment */
+
+      do {
+        rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
+        if( rc==LSM_OK && pNext==0 ){
+          rc = LSM_CORRUPT_BKPT;
+        }
+        if( rc ) break;
+        lsmFsPageRelease(pPg);
+        pPg = pNext;
+        aData = fsPageData(pPg, &nData);
+        flags = lsmGetU16(&aData[SEGMENT_FLAGS_OFFSET(nData)]);
+      }while( flags&SEGMENT_BTREE_FLAG );
+
+      iEnd = SEGMENT_EOF(nData, lsmGetU16(&aData[nData-2]));
+      assert( iEnd>0 && iEnd<nData );
+    }
+
+    lsmFsPageRelease(pPg);
+  }
+
+  return rc;
+}
+
+static int pageGetNRec(u8 *aData, int nData){
+  return (int)lsmGetU16(&aData[SEGMENT_NRECORD_OFFSET(nData)]);
+}
+
+static LsmPgno pageGetPtr(u8 *aData, int nData){
+  return (LsmPgno)lsmGetU64(&aData[SEGMENT_POINTER_OFFSET(nData)]);
+}
+
+static int pageGetFlags(u8 *aData, int nData){
+  return (int)lsmGetU16(&aData[SEGMENT_FLAGS_OFFSET(nData)]);
+}
+
+static u8 *pageGetCell(u8 *aData, int nData, int iCell){
+  return &aData[lsmGetU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, iCell)])];
+}
+
+/*
+** Return the number of cells on page pPg.
+*/
+static int pageObjGetNRec(Page *pPg){
+  int nData;
+  u8 *aData = lsmFsPageData(pPg, &nData);
+  return pageGetNRec(aData, nData);
+}
+
+/*
+** Return the decoded (possibly relative) pointer value stored in cell 
+** iCell from page aData/nData.
+*/
+static LsmPgno pageGetRecordPtr(u8 *aData, int nData, int iCell){
+  LsmPgno iRet;                   /* Return value */
+  u8 *aCell;                      /* Pointer to cell iCell */
+
+  assert( iCell<pageGetNRec(aData, nData) && iCell>=0 );
+  aCell = pageGetCell(aData, nData, iCell);
+  lsmVarintGet64(&aCell[1], &iRet);
+  return iRet;
+}
+
+static u8 *pageGetKey(
+  Segment *pSeg,                  /* Segment pPg belongs to */
+  Page *pPg,                      /* Page to read from */
+  int iCell,                      /* Index of cell on page to read */
+  int *piTopic,                   /* OUT: Topic associated with this key */
+  int *pnKey,                     /* OUT: Size of key in bytes */
+  LsmBlob *pBlob                  /* If required, use this for dynamic memory */
+){
+  u8 *pKey;
+  int nDummy;
+  int eType;
+  u8 *aData;
+  int nData;
+
+  aData = fsPageData(pPg, &nData);
+
+  assert( !(pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG) );
+  assert( iCell<pageGetNRec(aData, nData) );
+
+  pKey = pageGetCell(aData, nData, iCell);
+  eType = *pKey++;
+  pKey += lsmVarintGet32(pKey, &nDummy);
+  pKey += lsmVarintGet32(pKey, pnKey);
+  if( rtIsWrite(eType) ){
+    pKey += lsmVarintGet32(pKey, &nDummy);
+  }
+  *piTopic = rtTopic(eType);
+
+  sortedReadData(pSeg, pPg, pKey-aData, *pnKey, (void **)&pKey, pBlob);
+  return pKey;
+}
+
+static int pageGetKeyCopy(
+  lsm_env *pEnv,                  /* Environment handle */
+  Segment *pSeg,                  /* Segment pPg belongs to */
+  Page *pPg,                      /* Page to read from */
+  int iCell,                      /* Index of cell on page to read */
+  int *piTopic,                   /* OUT: Topic associated with this key */
+  LsmBlob *pBlob                  /* If required, use this for dynamic memory */
+){
+  int rc = LSM_OK;
+  int nKey;
+  u8 *aKey;
+
+  aKey = pageGetKey(pSeg, pPg, iCell, piTopic, &nKey, pBlob);
+  assert( (void *)aKey!=pBlob->pData || nKey==pBlob->nData );
+  if( (void *)aKey!=pBlob->pData ){
+    rc = sortedBlobSet(pEnv, pBlob, aKey, nKey);
+  }
+
+  return rc;
+}
+
+static LsmPgno pageGetBtreeRef(Page *pPg, int iKey){
+  LsmPgno iRef;
+  u8 *aData;
+  int nData;
+  u8 *aCell;
+
+  aData = fsPageData(pPg, &nData);
+  aCell = pageGetCell(aData, nData, iKey);
+  assert( aCell[0]==0 );
+  aCell++;
+  aCell += lsmVarintGet64(aCell, &iRef);
+  lsmVarintGet64(aCell, &iRef);
+  assert( iRef>0 );
+  return iRef;
+}
+
+#define GETVARINT64(a, i) (((i)=((u8*)(a))[0])<=240?1:lsmVarintGet64((a), &(i)))
+#define GETVARINT32(a, i) (((i)=((u8*)(a))[0])<=240?1:lsmVarintGet32((a), &(i)))
+
+static int pageGetBtreeKey(
+  Segment *pSeg,                  /* Segment page pPg belongs to */
+  Page *pPg,
+  int iKey, 
+  LsmPgno *piPtr, 
+  int *piTopic, 
+  void **ppKey,
+  int *pnKey,
+  LsmBlob *pBlob
+){
+  u8 *aData;
+  int nData;
+  u8 *aCell;
+  int eType;
+
+  aData = fsPageData(pPg, &nData);
+  assert( SEGMENT_BTREE_FLAG & pageGetFlags(aData, nData) );
+  assert( iKey>=0 && iKey<pageGetNRec(aData, nData) );
+
+  aCell = pageGetCell(aData, nData, iKey);
+  eType = *aCell++;
+  aCell += GETVARINT64(aCell, *piPtr);
+
+  if( eType==0 ){
+    int rc;
+    LsmPgno iRef;               /* Page number of referenced page */
+    Page *pRef;
+    aCell += GETVARINT64(aCell, iRef);
+    rc = lsmFsDbPageGet(lsmPageFS(pPg), pSeg, iRef, &pRef);
+    if( rc!=LSM_OK ) return rc;
+    pageGetKeyCopy(lsmPageEnv(pPg), pSeg, pRef, 0, &eType, pBlob);
+    lsmFsPageRelease(pRef);
+    *ppKey = pBlob->pData;
+    *pnKey = pBlob->nData;
+  }else{
+    aCell += GETVARINT32(aCell, *pnKey);
+    *ppKey = aCell;
+  }
+  if( piTopic ) *piTopic = rtTopic(eType);
+
+  return LSM_OK;
+}
+
+static int btreeCursorLoadKey(BtreeCursor *pCsr){
+  int rc = LSM_OK;
+  if( pCsr->iPg<0 ){
+    pCsr->pKey = 0;
+    pCsr->nKey = 0;
+    pCsr->eType = 0;
+  }else{
+    LsmPgno dummy;
+    int iPg = pCsr->iPg;
+    int iCell = pCsr->aPg[iPg].iCell;
+    while( iCell<0 && (--iPg)>=0 ){
+      iCell = pCsr->aPg[iPg].iCell-1;
+    }
+    if( iPg<0 || iCell<0 ) return LSM_CORRUPT_BKPT;
+
+    rc = pageGetBtreeKey(
+        pCsr->pSeg,
+        pCsr->aPg[iPg].pPage, iCell,
+        &dummy, &pCsr->eType, &pCsr->pKey, &pCsr->nKey, &pCsr->blob
+    );
+    pCsr->eType |= LSM_SEPARATOR;
+  }
+
+  return rc;
+}
+
+static int btreeCursorPtr(u8 *aData, int nData, int iCell){
+  int nCell;
+
+  nCell = pageGetNRec(aData, nData);
+  if( iCell>=nCell ){
+    return (int)pageGetPtr(aData, nData);
+  }
+  return (int)pageGetRecordPtr(aData, nData, iCell);
+}
+
+static int btreeCursorNext(BtreeCursor *pCsr){
+  int rc = LSM_OK;
+
+  BtreePg *pPg = &pCsr->aPg[pCsr->iPg];
+  int nCell; 
+  u8 *aData;
+  int nData;
+
+  assert( pCsr->iPg>=0 );
+  assert( pCsr->iPg==pCsr->nDepth-1 );
+
+  aData = fsPageData(pPg->pPage, &nData);
+  nCell = pageGetNRec(aData, nData);
+  assert( pPg->iCell<=nCell );
+  pPg->iCell++;
+  if( pPg->iCell==nCell ){
+    LsmPgno iLoad;
+
+    /* Up to parent. */
+    lsmFsPageRelease(pPg->pPage);
+    pPg->pPage = 0;
+    pCsr->iPg--;
+    while( pCsr->iPg>=0 ){
+      pPg = &pCsr->aPg[pCsr->iPg];
+      aData = fsPageData(pPg->pPage, &nData);
+      if( pPg->iCell<pageGetNRec(aData, nData) ) break;
+      lsmFsPageRelease(pPg->pPage);
+      pCsr->iPg--;
+    }
+
+    /* Read the key */
+    rc = btreeCursorLoadKey(pCsr);
+
+    /* Unless the cursor is at EOF, descend to cell -1 (yes, negative one) of 
+    ** the left-most most descendent. */
+    if( pCsr->iPg>=0 ){
+      pCsr->aPg[pCsr->iPg].iCell++;
+
+      iLoad = btreeCursorPtr(aData, nData, pPg->iCell);
+      do {
+        Page *pLoad;
+        pCsr->iPg++;
+        rc = lsmFsDbPageGet(pCsr->pFS, pCsr->pSeg, iLoad, &pLoad);
+        pCsr->aPg[pCsr->iPg].pPage = pLoad;
+        pCsr->aPg[pCsr->iPg].iCell = 0;
+        if( rc==LSM_OK ){
+          if( pCsr->iPg==(pCsr->nDepth-1) ) break;
+          aData = fsPageData(pLoad, &nData);
+          iLoad = btreeCursorPtr(aData, nData, 0);
+        }
+      }while( rc==LSM_OK && pCsr->iPg<(pCsr->nDepth-1) );
+      pCsr->aPg[pCsr->iPg].iCell = -1;
+    }
+
+  }else{
+    rc = btreeCursorLoadKey(pCsr);
+  }
+
+  if( rc==LSM_OK && pCsr->iPg>=0 ){
+    aData = fsPageData(pCsr->aPg[pCsr->iPg].pPage, &nData);
+    pCsr->iPtr = btreeCursorPtr(aData, nData, pCsr->aPg[pCsr->iPg].iCell+1);
+  }
+
+  return rc;
+}
+
+static void btreeCursorFree(BtreeCursor *pCsr){
+  if( pCsr ){
+    int i;
+    lsm_env *pEnv = lsmFsEnv(pCsr->pFS);
+    for(i=0; i<=pCsr->iPg; i++){
+      lsmFsPageRelease(pCsr->aPg[i].pPage);
+    }
+    sortedBlobFree(&pCsr->blob);
+    lsmFree(pEnv, pCsr->aPg);
+    lsmFree(pEnv, pCsr);
+  }
+}
+
+static int btreeCursorFirst(BtreeCursor *pCsr){
+  int rc;
+
+  Page *pPg = 0;
+  FileSystem *pFS = pCsr->pFS;
+  int iPg = (int)pCsr->pSeg->iRoot;
+
+  do {
+    rc = lsmFsDbPageGet(pFS, pCsr->pSeg, iPg, &pPg);
+    assert( (rc==LSM_OK)==(pPg!=0) );
+    if( rc==LSM_OK ){
+      u8 *aData;
+      int nData;
+      int flags;
+
+      aData = fsPageData(pPg, &nData);
+      flags = pageGetFlags(aData, nData);
+      if( (flags & SEGMENT_BTREE_FLAG)==0 ) break;
+
+      if( (pCsr->nDepth % 8)==0 ){
+        int nNew = pCsr->nDepth + 8;
+        pCsr->aPg = (BtreePg *)lsmReallocOrFreeRc(
+            lsmFsEnv(pFS), pCsr->aPg, sizeof(BtreePg) * nNew, &rc
+        );
+        if( rc==LSM_OK ){
+          memset(&pCsr->aPg[pCsr->nDepth], 0, sizeof(BtreePg) * 8);
+        }
+      }
+
+      if( rc==LSM_OK ){
+        assert( pCsr->aPg[pCsr->nDepth].iCell==0 );
+        pCsr->aPg[pCsr->nDepth].pPage = pPg;
+        pCsr->nDepth++;
+        iPg = (int)pageGetRecordPtr(aData, nData, 0);
+      }
+    }
+  }while( rc==LSM_OK );
+  lsmFsPageRelease(pPg);
+  pCsr->iPg = pCsr->nDepth-1;
+
+  if( rc==LSM_OK && pCsr->nDepth ){
+    pCsr->aPg[pCsr->iPg].iCell = -1;
+    rc = btreeCursorNext(pCsr);
+  }
+
+  return rc;
+}
+
+static void btreeCursorPosition(BtreeCursor *pCsr, MergeInput *p){
+  if( pCsr->iPg>=0 ){
+    p->iPg = lsmFsPageNumber(pCsr->aPg[pCsr->iPg].pPage);
+    p->iCell = ((pCsr->aPg[pCsr->iPg].iCell + 1) << 8) + pCsr->nDepth;
+  }else{
+    p->iPg = 0;
+    p->iCell = 0;
+  }
+}
+
+static void btreeCursorSplitkey(BtreeCursor *pCsr, MergeInput *p){
+  int iCell = pCsr->aPg[pCsr->iPg].iCell;
+  if( iCell>=0 ){
+    p->iCell = iCell;
+    p->iPg = lsmFsPageNumber(pCsr->aPg[pCsr->iPg].pPage);
+  }else{
+    int i;
+    for(i=pCsr->iPg-1; i>=0; i--){
+      if( pCsr->aPg[i].iCell>0 ) break;
+    }
+    assert( i>=0 );
+    p->iCell = pCsr->aPg[i].iCell-1;
+    p->iPg = lsmFsPageNumber(pCsr->aPg[i].pPage);
+  }
+}
+
+static int sortedKeyCompare(
+  int (*xCmp)(void *, int, void *, int),
+  int iLhsTopic, void *pLhsKey, int nLhsKey,
+  int iRhsTopic, void *pRhsKey, int nRhsKey
+){
+  int res = iLhsTopic - iRhsTopic;
+  if( res==0 ){
+    res = xCmp(pLhsKey, nLhsKey, pRhsKey, nRhsKey);
+  }
+  return res;
+}
+
+static int btreeCursorRestore(
+  BtreeCursor *pCsr, 
+  int (*xCmp)(void *, int, void *, int),
+  MergeInput *p
+){
+  int rc = LSM_OK;
+
+  if( p->iPg ){
+    lsm_env *pEnv = lsmFsEnv(pCsr->pFS);
+    int iCell;                    /* Current cell number on leaf page */
+    LsmPgno iLeaf;                /* Page number of current leaf page */
+    int nDepth;                   /* Depth of b-tree structure */
+    Segment *pSeg = pCsr->pSeg;
+
+    /* Decode the MergeInput structure */
+    iLeaf = p->iPg;
+    nDepth = (p->iCell & 0x00FF);
+    iCell = (p->iCell >> 8) - 1;
+
+    /* Allocate the BtreeCursor.aPg[] array */
+    assert( pCsr->aPg==0 );
+    pCsr->aPg = (BtreePg *)lsmMallocZeroRc(pEnv, sizeof(BtreePg) * nDepth, &rc);
+
+    /* Populate the last entry of the aPg[] array */
+    if( rc==LSM_OK ){
+      Page **pp = &pCsr->aPg[nDepth-1].pPage;
+      pCsr->iPg = nDepth-1;
+      pCsr->nDepth = nDepth;
+      pCsr->aPg[pCsr->iPg].iCell = iCell;
+      rc = lsmFsDbPageGet(pCsr->pFS, pSeg, iLeaf, pp);
+    }
+
+    /* Populate any other aPg[] array entries */
+    if( rc==LSM_OK && nDepth>1 ){
+      LsmBlob blob = {0,0,0};
+      void *pSeek;
+      int nSeek;
+      int iTopicSeek;
+      int iPg = 0;
+      int iLoad = (int)pSeg->iRoot;
+      Page *pPg = pCsr->aPg[nDepth-1].pPage;
+ 
+      if( pageObjGetNRec(pPg)==0 ){
+        /* This can happen when pPg is the right-most leaf in the b-tree.
+        ** In this case, set the iTopicSeek/pSeek/nSeek key to a value
+        ** greater than any real key.  */
+        assert( iCell==-1 );
+        iTopicSeek = 1000;
+        pSeek = 0;
+        nSeek = 0;
+      }else{
+        LsmPgno dummy;
+        rc = pageGetBtreeKey(pSeg, pPg,
+            0, &dummy, &iTopicSeek, &pSeek, &nSeek, &pCsr->blob
+        );
+      }
+
+      do {
+        Page *pPg2;
+        rc = lsmFsDbPageGet(pCsr->pFS, pSeg, iLoad, &pPg2);
+        assert( rc==LSM_OK || pPg2==0 );
+        if( rc==LSM_OK ){
+          u8 *aData;                  /* Buffer containing page data */
+          int nData;                  /* Size of aData[] in bytes */
+          int iMin;
+          int iMax;
+          int iCell2;
+
+          aData = fsPageData(pPg2, &nData);
+          assert( (pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG) );
+
+          iLoad = (int)pageGetPtr(aData, nData);
+          iCell2 = pageGetNRec(aData, nData); 
+          iMax = iCell2-1;
+          iMin = 0;
+
+          while( iMax>=iMin ){
+            int iTry = (iMin+iMax)/2;
+            void *pKey; int nKey;         /* Key for cell iTry */
+            int iTopic;                   /* Topic for key pKeyT/nKeyT */
+            LsmPgno iPtr;                 /* Pointer for cell iTry */
+            int res;                      /* (pSeek - pKeyT) */
+
+            rc = pageGetBtreeKey(
+                pSeg, pPg2, iTry, &iPtr, &iTopic, &pKey, &nKey, &blob
+            );
+            if( rc!=LSM_OK ) break;
+
+            res = sortedKeyCompare(
+                xCmp, iTopicSeek, pSeek, nSeek, iTopic, pKey, nKey
+            );
+            assert( res!=0 );
+
+            if( res<0 ){
+              iLoad = (int)iPtr;
+              iCell2 = iTry;
+              iMax = iTry-1;
+            }else{
+              iMin = iTry+1;
+            }
+          }
+
+          pCsr->aPg[iPg].pPage = pPg2;
+          pCsr->aPg[iPg].iCell = iCell2;
+          iPg++;
+          assert( iPg!=nDepth-1 
+               || lsmFsRedirectPage(pCsr->pFS, pSeg->pRedirect, iLoad)==iLeaf
+          );
+        }
+      }while( rc==LSM_OK && iPg<(nDepth-1) );
+      sortedBlobFree(&blob);
+    }
+
+    /* Load the current key and pointer */
+    if( rc==LSM_OK ){
+      BtreePg *pBtreePg;
+      u8 *aData;
+      int nData;
+
+      pBtreePg = &pCsr->aPg[pCsr->iPg];
+      aData = fsPageData(pBtreePg->pPage, &nData);
+      pCsr->iPtr = btreeCursorPtr(aData, nData, pBtreePg->iCell+1);
+      if( pBtreePg->iCell<0 ){
+        LsmPgno dummy;
+        int i;
+        for(i=pCsr->iPg-1; i>=0; i--){
+          if( pCsr->aPg[i].iCell>0 ) break;
+        }
+        assert( i>=0 );
+        rc = pageGetBtreeKey(pSeg,
+            pCsr->aPg[i].pPage, pCsr->aPg[i].iCell-1,
+            &dummy, &pCsr->eType, &pCsr->pKey, &pCsr->nKey, &pCsr->blob
+        );
+        pCsr->eType |= LSM_SEPARATOR;
+
+      }else{
+        rc = btreeCursorLoadKey(pCsr);
+      }
+    }
+  }
+  return rc;
+}
+
+static int btreeCursorNew(
+  lsm_db *pDb,
+  Segment *pSeg,
+  BtreeCursor **ppCsr
+){
+  int rc = LSM_OK;
+  BtreeCursor *pCsr;
+  
+  assert( pSeg->iRoot );
+  pCsr = lsmMallocZeroRc(pDb->pEnv, sizeof(BtreeCursor), &rc);
+  if( pCsr ){
+    pCsr->pFS = pDb->pFS;
+    pCsr->pSeg = pSeg;
+    pCsr->iPg = -1;
+  }
+
+  *ppCsr = pCsr;
+  return rc;
+}
+
+static void segmentPtrSetPage(SegmentPtr *pPtr, Page *pNext){
+  lsmFsPageRelease(pPtr->pPg);
+  if( pNext ){
+    int nData;
+    u8 *aData = fsPageData(pNext, &nData);
+    pPtr->nCell = pageGetNRec(aData, nData);
+    pPtr->flags = (u16)pageGetFlags(aData, nData);
+    pPtr->iPtr = pageGetPtr(aData, nData);
+  }
+  pPtr->pPg = pNext;
+}
+
+/*
+** Load a new page into the SegmentPtr object pPtr.
+*/
+static int segmentPtrLoadPage(
+  FileSystem *pFS,
+  SegmentPtr *pPtr,              /* Load page into this SegmentPtr object */
+  int iNew                       /* Page number of new page */
+){
+  Page *pPg = 0;                 /* The new page */
+  int rc;                        /* Return Code */
+
+  rc = lsmFsDbPageGet(pFS, pPtr->pSeg, iNew, &pPg);
+  assert( rc==LSM_OK || pPg==0 );
+  segmentPtrSetPage(pPtr, pPg);
+
+  return rc;
+}
+
+static int segmentPtrReadData(
+  SegmentPtr *pPtr,
+  int iOff,
+  int nByte,
+  void **ppData,
+  LsmBlob *pBlob
+){
+  return sortedReadData(pPtr->pSeg, pPtr->pPg, iOff, nByte, ppData, pBlob);
+}
+
+static int segmentPtrNextPage(
+  SegmentPtr *pPtr,              /* Load page into this SegmentPtr object */
+  int eDir                       /* +1 for next(), -1 for prev() */
+){
+  Page *pNext;                   /* New page to load */
+  int rc;                        /* Return code */
+
+  assert( eDir==1 || eDir==-1 );
+  assert( pPtr->pPg );
+  assert( pPtr->pSeg || eDir>0 );
+
+  rc = lsmFsDbPageNext(pPtr->pSeg, pPtr->pPg, eDir, &pNext);
+  assert( rc==LSM_OK || pNext==0 );
+  segmentPtrSetPage(pPtr, pNext);
+  return rc;
+}
+
+static int segmentPtrLoadCell(
+  SegmentPtr *pPtr,              /* Load page into this SegmentPtr object */
+  int iNew                       /* Cell number of new cell */
+){
+  int rc = LSM_OK;
+  if( pPtr->pPg ){
+    u8 *aData;                    /* Pointer to page data buffer */
+    int iOff;                     /* Offset in aData[] to read from */
+    int nPgsz;                    /* Size of page (aData[]) in bytes */
+
+    assert( iNew<pPtr->nCell );
+    pPtr->iCell = iNew;
+    aData = fsPageData(pPtr->pPg, &nPgsz);
+    iOff = lsmGetU16(&aData[SEGMENT_CELLPTR_OFFSET(nPgsz, pPtr->iCell)]);
+    pPtr->eType = aData[iOff];
+    iOff++;
+    iOff += GETVARINT64(&aData[iOff], pPtr->iPgPtr);
+    iOff += GETVARINT32(&aData[iOff], pPtr->nKey);
+    if( rtIsWrite(pPtr->eType) ){
+      iOff += GETVARINT32(&aData[iOff], pPtr->nVal);
+    }
+    assert( pPtr->nKey>=0 );
+
+    rc = segmentPtrReadData(
+        pPtr, iOff, pPtr->nKey, &pPtr->pKey, &pPtr->blob1
+    );
+    if( rc==LSM_OK && rtIsWrite(pPtr->eType) ){
+      rc = segmentPtrReadData(
+          pPtr, iOff+pPtr->nKey, pPtr->nVal, &pPtr->pVal, &pPtr->blob2
+      );
+    }else{
+      pPtr->nVal = 0;
+      pPtr->pVal = 0;
+    }
+  }
+
+  return rc;
+}
+
+
+static Segment *sortedSplitkeySegment(Level *pLevel){
+  Merge *pMerge = pLevel->pMerge;
+  MergeInput *p = &pMerge->splitkey;
+  Segment *pSeg;
+  int i;
+
+  for(i=0; i<pMerge->nInput; i++){
+    if( p->iPg==pMerge->aInput[i].iPg ) break;
+  }
+  if( pMerge->nInput==(pLevel->nRight+1) && i>=(pMerge->nInput-1) ){
+    pSeg = &pLevel->pNext->lhs;
+  }else{
+    pSeg = &pLevel->aRhs[i];
+  }
+
+  return pSeg;
+}
+
+static void sortedSplitkey(lsm_db *pDb, Level *pLevel, int *pRc){
+  Segment *pSeg;
+  Page *pPg = 0;
+  lsm_env *pEnv = pDb->pEnv;      /* Environment handle */
+  int rc = *pRc;
+  Merge *pMerge = pLevel->pMerge;
+
+  pSeg = sortedSplitkeySegment(pLevel);
+  if( rc==LSM_OK ){
+    rc = lsmFsDbPageGet(pDb->pFS, pSeg, pMerge->splitkey.iPg, &pPg);
+  }
+  if( rc==LSM_OK ){
+    int iTopic;
+    LsmBlob blob = {0, 0, 0, 0};
+    u8 *aData;
+    int nData;
+  
+    aData = lsmFsPageData(pPg, &nData);
+    if( pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG ){
+      void *pKey;
+      int nKey;
+      LsmPgno dummy;
+      rc = pageGetBtreeKey(pSeg,
+          pPg, pMerge->splitkey.iCell, &dummy, &iTopic, &pKey, &nKey, &blob
+      );
+      if( rc==LSM_OK && blob.pData!=pKey ){
+        rc = sortedBlobSet(pEnv, &blob, pKey, nKey);
+      }
+    }else{
+      rc = pageGetKeyCopy(
+          pEnv, pSeg, pPg, pMerge->splitkey.iCell, &iTopic, &blob
+      );
+    }
+
+    pLevel->iSplitTopic = iTopic;
+    pLevel->pSplitKey = blob.pData;
+    pLevel->nSplitKey = blob.nData;
+    lsmFsPageRelease(pPg);
+  }
+
+  *pRc = rc;
+}
+
+/*
+** Reset a segment cursor. Also free its buffers if they are nThreshold
+** bytes or larger in size.
+*/
+static void segmentPtrReset(SegmentPtr *pPtr, int nThreshold){
+  lsmFsPageRelease(pPtr->pPg);
+  pPtr->pPg = 0;
+  pPtr->nCell = 0;
+  pPtr->pKey = 0;
+  pPtr->nKey = 0;
+  pPtr->pVal = 0;
+  pPtr->nVal = 0;
+  pPtr->eType = 0;
+  pPtr->iCell = 0;
+  if( pPtr->blob1.nAlloc>=nThreshold ) sortedBlobFree(&pPtr->blob1);
+  if( pPtr->blob2.nAlloc>=nThreshold ) sortedBlobFree(&pPtr->blob2);
+}
+
+static int segmentPtrIgnoreSeparators(MultiCursor *pCsr, SegmentPtr *pPtr){
+  return (pCsr->flags & CURSOR_READ_SEPARATORS)==0
+      || (pPtr!=&pCsr->aPtr[pCsr->nPtr-1]);
+}
+
+static int segmentPtrAdvance(
+  MultiCursor *pCsr, 
+  SegmentPtr *pPtr,
+  int bReverse
+){
+  int eDir = (bReverse ? -1 : 1);
+  Level *pLvl = pPtr->pLevel;
+  do {
+    int rc;
+    int iCell;                    /* Number of new cell in page */
+    int svFlags = 0;              /* SegmentPtr.eType before advance */
+
+    iCell = pPtr->iCell + eDir;
+    assert( pPtr->pPg );
+    assert( iCell<=pPtr->nCell && iCell>=-1 );
+
+    if( bReverse && pPtr->pSeg!=&pPtr->pLevel->lhs ){
+      svFlags = pPtr->eType;
+      assert( svFlags );
+    }
+
+    if( iCell>=pPtr->nCell || iCell<0 ){
+      do {
+        rc = segmentPtrNextPage(pPtr, eDir); 
+      }while( rc==LSM_OK 
+           && pPtr->pPg 
+           && (pPtr->nCell==0 || (pPtr->flags & SEGMENT_BTREE_FLAG) ) 
+      );
+      if( rc!=LSM_OK ) return rc;
+      iCell = bReverse ? (pPtr->nCell-1) : 0;
+    }
+    rc = segmentPtrLoadCell(pPtr, iCell);
+    if( rc!=LSM_OK ) return rc;
+
+    if( svFlags && pPtr->pPg ){
+      int res = sortedKeyCompare(pCsr->pDb->xCmp,
+          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
+          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
+      );
+      if( res<0 ) segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
+    }
+
+    if( pPtr->pPg==0 && (svFlags & LSM_END_DELETE) ){
+      Segment *pSeg = pPtr->pSeg;
+      rc = lsmFsDbPageGet(pCsr->pDb->pFS, pSeg, pSeg->iFirst, &pPtr->pPg);
+      if( rc!=LSM_OK ) return rc;
+      pPtr->eType = LSM_START_DELETE | LSM_POINT_DELETE;
+      pPtr->eType |= (pLvl->iSplitTopic ? LSM_SYSTEMKEY : 0);
+      pPtr->pKey = pLvl->pSplitKey;
+      pPtr->nKey = pLvl->nSplitKey;
+    }
+
+  }while( pCsr 
+       && pPtr->pPg 
+       && segmentPtrIgnoreSeparators(pCsr, pPtr)
+       && rtIsSeparator(pPtr->eType)
+  );
+
+  return LSM_OK;
+}
+
+static void segmentPtrEndPage(
+  FileSystem *pFS, 
+  SegmentPtr *pPtr, 
+  int bLast, 
+  int *pRc
+){
+  if( *pRc==LSM_OK ){
+    Segment *pSeg = pPtr->pSeg;
+    Page *pNew = 0;
+    if( bLast ){
+      *pRc = lsmFsDbPageLast(pFS, pSeg, &pNew);
+    }else{
+      *pRc = lsmFsDbPageGet(pFS, pSeg, pSeg->iFirst, &pNew);
+    }
+    segmentPtrSetPage(pPtr, pNew);
+  }
+}
+
+
+/*
+** Try to move the segment pointer passed as the second argument so that it
+** points at either the first (bLast==0) or last (bLast==1) cell in the valid
+** region of the segment defined by pPtr->iFirst and pPtr->iLast.
+**
+** Return LSM_OK if successful or an lsm error code if something goes
+** wrong (IO error, OOM etc.).
+*/
+static int segmentPtrEnd(MultiCursor *pCsr, SegmentPtr *pPtr, int bLast){
+  Level *pLvl = pPtr->pLevel;
+  int rc = LSM_OK;
+  FileSystem *pFS = pCsr->pDb->pFS;
+  int bIgnore;
+
+  segmentPtrEndPage(pFS, pPtr, bLast, &rc);
+  while( rc==LSM_OK && pPtr->pPg 
+      && (pPtr->nCell==0 || (pPtr->flags & SEGMENT_BTREE_FLAG))
+  ){
+    rc = segmentPtrNextPage(pPtr, (bLast ? -1 : 1));
+  }
+
+  if( rc==LSM_OK && pPtr->pPg ){
+    rc = segmentPtrLoadCell(pPtr, bLast ? (pPtr->nCell-1) : 0);
+    if( rc==LSM_OK && bLast && pPtr->pSeg!=&pLvl->lhs ){
+      int res = sortedKeyCompare(pCsr->pDb->xCmp,
+          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
+          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
+      );
+      if( res<0 ) segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
+    }
+  }
+  
+  bIgnore = segmentPtrIgnoreSeparators(pCsr, pPtr);
+  if( rc==LSM_OK && pPtr->pPg && bIgnore && rtIsSeparator(pPtr->eType) ){
+    rc = segmentPtrAdvance(pCsr, pPtr, bLast);
+  }
+
+#if 0
+  if( bLast && rc==LSM_OK && pPtr->pPg
+   && pPtr->pSeg==&pLvl->lhs 
+   && pLvl->nRight && (pPtr->eType & LSM_START_DELETE)
+  ){
+    pPtr->iCell++;
+    pPtr->eType = LSM_END_DELETE | (pLvl->iSplitTopic);
+    pPtr->pKey = pLvl->pSplitKey;
+    pPtr->nKey = pLvl->nSplitKey;
+    pPtr->pVal = 0;
+    pPtr->nVal = 0;
+  }
+#endif
+
+  return rc;
+}
+
+static void segmentPtrKey(SegmentPtr *pPtr, void **ppKey, int *pnKey){
+  assert( pPtr->pPg );
+  *ppKey = pPtr->pKey;
+  *pnKey = pPtr->nKey;
+}
+
+#if 0 /* NOT USED */
+static char *keyToString(lsm_env *pEnv, void *pKey, int nKey){
+  int i;
+  u8 *aKey = (u8 *)pKey;
+  char *zRet = (char *)lsmMalloc(pEnv, nKey+1);
+
+  for(i=0; i<nKey; i++){
+    zRet[i] = (char)(isalnum(aKey[i]) ? aKey[i] : '.');
+  }
+  zRet[nKey] = '\0';
+  return zRet;
+}
+#endif
+
+#if 0 /* NOT USED */
+/*
+** Check that the page that pPtr currently has loaded is the correct page
+** to search for key (pKey/nKey). If it is, return 1. Otherwise, an assert
+** fails and this function does not return.
+*/
+static int assertKeyLocation(
+  MultiCursor *pCsr, 
+  SegmentPtr *pPtr, 
+  void *pKey, int nKey
+){
+  lsm_env *pEnv = lsmFsEnv(pCsr->pDb->pFS);
+  LsmBlob blob = {0, 0, 0};
+  int eDir;
+  int iTopic = 0;                 /* TODO: Fix me */
+
+  for(eDir=-1; eDir<=1; eDir+=2){
+    Page *pTest = pPtr->pPg;
+
+    lsmFsPageRef(pTest);
+    while( pTest ){
+      Segment *pSeg = pPtr->pSeg;
+      Page *pNext;
+
+      int rc = lsmFsDbPageNext(pSeg, pTest, eDir, &pNext);
+      lsmFsPageRelease(pTest);
+      if( rc ) return 1;
+      pTest = pNext;
+
+      if( pTest ){
+        int nData;
+        u8 *aData = fsPageData(pTest, &nData);
+        int nCell = pageGetNRec(aData, nData);
+        int flags = pageGetFlags(aData, nData);
+        if( nCell && 0==(flags&SEGMENT_BTREE_FLAG) ){
+          int nPgKey;
+          int iPgTopic;
+          u8 *pPgKey;
+          int res;
+          int iCell;
+
+          iCell = ((eDir < 0) ? (nCell-1) : 0);
+          pPgKey = pageGetKey(pSeg, pTest, iCell, &iPgTopic, &nPgKey, &blob);
+          res = iTopic - iPgTopic;
+          if( res==0 ) res = pCsr->pDb->xCmp(pKey, nKey, pPgKey, nPgKey);
+          if( (eDir==1 && res>0) || (eDir==-1 && res<0) ){
+            /* Taking this branch means something has gone wrong. */
+            char *zMsg = lsmMallocPrintf(pEnv, "Key \"%s\" is not on page %d", 
+                keyToString(pEnv, pKey, nKey), lsmFsPageNumber(pPtr->pPg)
+            );
+            fprintf(stderr, "%s\n", zMsg);
+            assert( !"assertKeyLocation() failed" );
+          }
+          lsmFsPageRelease(pTest);
+          pTest = 0;
+        }
+      }
+    }
+  }
+
+  sortedBlobFree(&blob);
+  return 1;
+}
+#endif
+
+#ifndef NDEBUG
+static int assertSeekResult(
+  MultiCursor *pCsr,
+  SegmentPtr *pPtr,
+  int iTopic,
+  void *pKey,
+  int nKey,
+  int eSeek
+){
+  if( pPtr->pPg ){
+    int res;
+    res = sortedKeyCompare(pCsr->pDb->xCmp, iTopic, pKey, nKey,
+        rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey
+    );
+
+    if( eSeek==LSM_SEEK_EQ ) return (res==0);
+    if( eSeek==LSM_SEEK_LE ) return (res>=0);
+    if( eSeek==LSM_SEEK_GE ) return (res<=0);
+  }
+
+  return 1;
+}
+#endif
+
+static int segmentPtrSearchOversized(
+  MultiCursor *pCsr,              /* Cursor context */
+  SegmentPtr *pPtr,               /* Pointer to seek */
+  int iTopic,                     /* Topic of key to search for */
+  void *pKey, int nKey            /* Key to seek to */
+){
+  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
+  int rc = LSM_OK;
+
+  /* If the OVERSIZED flag is set, then there is no pointer in the
+  ** upper level to the next page in the segment that contains at least
+  ** one key. So compare the largest key on the current page with the
+  ** key being sought (pKey/nKey). If (pKey/nKey) is larger, advance
+  ** to the next page in the segment that contains at least one key. 
+  */
+  while( rc==LSM_OK && (pPtr->flags & PGFTR_SKIP_NEXT_FLAG) ){
+    u8 *pLastKey;
+    int nLastKey;
+    int iLastTopic;
+    int res;                      /* Result of comparison */
+    Page *pNext;
+
+    /* Load the last key on the current page. */
+    pLastKey = pageGetKey(pPtr->pSeg,
+        pPtr->pPg, pPtr->nCell-1, &iLastTopic, &nLastKey, &pPtr->blob1
+    );
+
+    /* If the loaded key is >= than (pKey/nKey), break out of the loop.
+    ** If (pKey/nKey) is present in this array, it must be on the current 
+    ** page.  */
+    res = sortedKeyCompare(
+        xCmp, iLastTopic, pLastKey, nLastKey, iTopic, pKey, nKey
+    );
+    if( res>=0 ) break;
+
+    /* Advance to the next page that contains at least one key. */
+    pNext = pPtr->pPg;
+    lsmFsPageRef(pNext);
+    while( 1 ){
+      Page *pLoad;
+      u8 *aData; int nData;
+
+      rc = lsmFsDbPageNext(pPtr->pSeg, pNext, 1, &pLoad);
+      lsmFsPageRelease(pNext);
+      pNext = pLoad;
+      if( pNext==0 ) break;
+
+      assert( rc==LSM_OK );
+      aData = lsmFsPageData(pNext, &nData);
+      if( (pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG)==0
+       && pageGetNRec(aData, nData)>0
+      ){
+        break;
+      }
+    }
+    if( pNext==0 ) break;
+    segmentPtrSetPage(pPtr, pNext);
+
+    /* This should probably be an LSM_CORRUPT error. */
+    assert( rc!=LSM_OK || (pPtr->flags & PGFTR_SKIP_THIS_FLAG) );
+  }
+
+  return rc;
+}
+
+static int ptrFwdPointer(
+  Page *pPage,
+  int iCell,
+  Segment *pSeg,
+  LsmPgno *piPtr,
+  int *pbFound
+){
+  Page *pPg = pPage;
+  int iFirst = iCell;
+  int rc = LSM_OK;
+
+  do {
+    Page *pNext = 0;
+    u8 *aData;
+    int nData;
+
+    aData = lsmFsPageData(pPg, &nData);
+    if( (pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG)==0 ){
+      int i;
+      int nCell = pageGetNRec(aData, nData);
+      for(i=iFirst; i<nCell; i++){
+        u8 eType = *pageGetCell(aData, nData, i);
+        if( (eType & LSM_START_DELETE)==0 ){
+          *pbFound = 1;
+          *piPtr = pageGetRecordPtr(aData, nData, i) + pageGetPtr(aData, nData);
+          lsmFsPageRelease(pPg);
+          return LSM_OK;
+        }
+      }
+    }
+
+    rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
+    lsmFsPageRelease(pPg);
+    pPg = pNext;
+    iFirst = 0;
+  }while( pPg && rc==LSM_OK );
+  lsmFsPageRelease(pPg);
+
+  *pbFound = 0;
+  return rc;
+}
+
+static int sortedRhsFirst(MultiCursor *pCsr, Level *pLvl, SegmentPtr *pPtr){
+  int rc;
+  rc = segmentPtrEnd(pCsr, pPtr, 0);
+  while( pPtr->pPg && rc==LSM_OK ){
+    int res = sortedKeyCompare(pCsr->pDb->xCmp,
+        pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey,
+        rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey
+    );
+    if( res<=0 ) break;
+    rc = segmentPtrAdvance(pCsr, pPtr, 0);
+  }
+  return rc;
+}
+
+
+/*
+** This function is called as part of a SEEK_GE op on a multi-cursor if the 
+** FC pointer read from segment *pPtr comes from an entry with the 
+** LSM_START_DELETE flag set. In this case the pointer value cannot be 
+** trusted. Instead, the pointer that should be followed is that associated
+** with the next entry in *pPtr that does not have LSM_START_DELETE set.
+**
+** Why the pointers can't be trusted:
+**
+**
+**
+** TODO: This is a stop-gap solution:
+** 
+**   At the moment, this function is called from within segmentPtrSeek(), 
+**   as part of the initial lsmMCursorSeek() call. However, consider a 
+**   database where the following has occurred:
+**
+**      1. A range delete removes keys 1..9999 using a range delete.
+**      2. Keys 1 through 9999 are reinserted.
+**      3. The levels containing the ops in 1. and 2. above are merged. Call
+**         this level N. Level N contains FC pointers to level N+1.
+**
+**   Then, if the user attempts to query for (key>=2 LIMIT 10), the 
+**   lsmMCursorSeek() call will iterate through 9998 entries searching for a 
+**   pointer down to the level N+1 that is never actually used. It would be
+**   much better if the multi-cursor could do this lazily - only seek to the
+**   level (N+1) page after the user has moved the cursor on level N passed
+**   the big range-delete.
+*/
+static int segmentPtrFwdPointer(
+  MultiCursor *pCsr,              /* Multi-cursor pPtr belongs to */
+  SegmentPtr *pPtr,               /* Segment-pointer to extract FC ptr from */
+  LsmPgno *piPtr                  /* OUT: FC pointer value */
+){
+  Level *pLvl = pPtr->pLevel;
+  Level *pNext = pLvl->pNext;
+  Page *pPg = pPtr->pPg;
+  int rc;
+  int bFound;
+  LsmPgno iOut = 0;
+
+  if( pPtr->pSeg==&pLvl->lhs || pPtr->pSeg==&pLvl->aRhs[pLvl->nRight-1] ){
+    if( pNext==0 
+        || (pNext->nRight==0 && pNext->lhs.iRoot)
+        || (pNext->nRight!=0 && pNext->aRhs[0].iRoot)
+      ){
+      /* Do nothing. The pointer will not be used anyway. */
+      return LSM_OK;
+    }
+  }else{
+    if( pPtr[1].pSeg->iRoot ){
+      return LSM_OK;
+    }
+  }
+
+  /* Search for a pointer within the current segment. */
+  lsmFsPageRef(pPg);
+  rc = ptrFwdPointer(pPg, pPtr->iCell, pPtr->pSeg, &iOut, &bFound);
+
+  if( rc==LSM_OK && bFound==0 ){
+    /* This case happens when pPtr points to the left-hand-side of a segment
+    ** currently undergoing an incremental merge. In this case, jump to the
+    ** oldest segment in the right-hand-side of the same level and continue
+    ** searching. But - do not consider any keys smaller than the levels
+    ** split-key. */
+    SegmentPtr ptr;
+
+    if( pPtr->pLevel->nRight==0 || pPtr->pSeg!=&pPtr->pLevel->lhs ){
+      return LSM_CORRUPT_BKPT;
+    }
+
+    memset(&ptr, 0, sizeof(SegmentPtr));
+    ptr.pLevel = pPtr->pLevel;
+    ptr.pSeg = &ptr.pLevel->aRhs[ptr.pLevel->nRight-1];
+    rc = sortedRhsFirst(pCsr, ptr.pLevel, &ptr);
+    if( rc==LSM_OK ){
+      rc = ptrFwdPointer(ptr.pPg, ptr.iCell, ptr.pSeg, &iOut, &bFound);
+      ptr.pPg = 0;
+    }
+    segmentPtrReset(&ptr, 0);
+  }
+
+  *piPtr = iOut;
+  return rc;
+}
+
+static int segmentPtrSeek(
+  MultiCursor *pCsr,              /* Cursor context */
+  SegmentPtr *pPtr,               /* Pointer to seek */
+  int iTopic,                     /* Key topic to seek to */
+  void *pKey, int nKey,           /* Key to seek to */
+  int eSeek,                      /* Search bias - see above */
+  int *piPtr,                     /* OUT: FC pointer */
+  int *pbStop
+){
+  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
+  int res = 0;                        /* Result of comparison operation */
+  int rc = LSM_OK;
+  int iMin;
+  int iMax;
+  LsmPgno iPtrOut = 0;
+
+  /* If the current page contains an oversized entry, then there are no
+  ** pointers to one or more of the subsequent pages in the sorted run.
+  ** The following call ensures that the segment-ptr points to the correct 
+  ** page in this case.  */
+  rc = segmentPtrSearchOversized(pCsr, pPtr, iTopic, pKey, nKey);
+  iPtrOut = pPtr->iPtr;
+
+  /* Assert that this page is the right page of this segment for the key
+  ** that we are searching for. Do this by loading page (iPg-1) and testing
+  ** that pKey/nKey is greater than all keys on that page, and then by 
+  ** loading (iPg+1) and testing that pKey/nKey is smaller than all
+  ** the keys it houses.  
+  **
+  ** TODO: With range-deletes in the tree, the test described above may fail.
+  */
+#if 0
+  assert( assertKeyLocation(pCsr, pPtr, pKey, nKey) );
+#endif
+
+  assert( pPtr->nCell>0 
+       || pPtr->pSeg->nSize==1 
+       || lsmFsDbPageIsLast(pPtr->pSeg, pPtr->pPg)
+  );
+  if( pPtr->nCell==0 ){
+    segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
+  }else{
+    iMin = 0;
+    iMax = pPtr->nCell-1;
+
+    while( 1 ){
+      int iTry = (iMin+iMax)/2;
+      void *pKeyT; int nKeyT;       /* Key for cell iTry */
+      int iTopicT;
+
+      assert( iTry<iMax || iMin==iMax );
+
+      rc = segmentPtrLoadCell(pPtr, iTry);
+      if( rc!=LSM_OK ) break;
+
+      segmentPtrKey(pPtr, &pKeyT, &nKeyT);
+      iTopicT = rtTopic(pPtr->eType);
+
+      res = sortedKeyCompare(xCmp, iTopicT, pKeyT, nKeyT, iTopic, pKey, nKey);
+      if( res<=0 ){
+        iPtrOut = pPtr->iPtr + pPtr->iPgPtr;
+      }
+
+      if( res==0 || iMin==iMax ){
+        break;
+      }else if( res>0 ){
+        iMax = LSM_MAX(iTry-1, iMin);
+      }else{
+        iMin = iTry+1;
+      }
+    }
+
+    if( rc==LSM_OK ){
+      assert( res==0 || (iMin==iMax && iMin>=0 && iMin<pPtr->nCell) );
+      if( res ){
+        rc = segmentPtrLoadCell(pPtr, iMin);
+      }
+      assert( rc!=LSM_OK || res>0 || iPtrOut==(pPtr->iPtr + pPtr->iPgPtr) );
+
+      if( rc==LSM_OK ){
+        switch( eSeek ){
+          case LSM_SEEK_EQ: {
+            int eType = pPtr->eType;
+            if( (res<0 && (eType & LSM_START_DELETE))
+             || (res>0 && (eType & LSM_END_DELETE))
+             || (res==0 && (eType & LSM_POINT_DELETE))
+            ){
+              *pbStop = 1;
+            }else if( res==0 && (eType & LSM_INSERT) ){
+              lsm_env *pEnv = pCsr->pDb->pEnv;
+              *pbStop = 1;
+              pCsr->eType = pPtr->eType;
+              rc = sortedBlobSet(pEnv, &pCsr->key, pPtr->pKey, pPtr->nKey);
+              if( rc==LSM_OK ){
+                rc = sortedBlobSet(pEnv, &pCsr->val, pPtr->pVal, pPtr->nVal);
+              }
+              pCsr->flags |= CURSOR_SEEK_EQ;
+            }
+            segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
+            break;
+          }
+          case LSM_SEEK_LE:
+            if( res>0 ) rc = segmentPtrAdvance(pCsr, pPtr, 1);
+            break;
+          case LSM_SEEK_GE: {
+            /* Figure out if we need to 'skip' the pointer forward or not */
+            if( (res<=0 && (pPtr->eType & LSM_START_DELETE)) 
+             || (res>0  && (pPtr->eType & LSM_END_DELETE)) 
+            ){
+              rc = segmentPtrFwdPointer(pCsr, pPtr, &iPtrOut);
+            }
+            if( res<0 && rc==LSM_OK ){
+              rc = segmentPtrAdvance(pCsr, pPtr, 0);
+            }
+            break;
+          }
+        }
+      }
+    }
+
+    /* If the cursor seek has found a separator key, and this cursor is
+    ** supposed to ignore separators keys, advance to the next entry.  */
+    if( rc==LSM_OK && pPtr->pPg
+     && segmentPtrIgnoreSeparators(pCsr, pPtr) 
+     && rtIsSeparator(pPtr->eType)
+    ){
+      assert( eSeek!=LSM_SEEK_EQ );
+      rc = segmentPtrAdvance(pCsr, pPtr, eSeek==LSM_SEEK_LE);
+    }
+  }
+
+  assert( rc!=LSM_OK || assertSeekResult(pCsr,pPtr,iTopic,pKey,nKey,eSeek) );
+  *piPtr = (int)iPtrOut;
+  return rc;
+}
+
+static int seekInBtree(
+  MultiCursor *pCsr,              /* Multi-cursor object */
+  Segment *pSeg,                  /* Seek within this segment */
+  int iTopic,
+  void *pKey, int nKey,           /* Key to seek to */
+  LsmPgno *aPg,                   /* OUT: Page numbers */
+  Page **ppPg                     /* OUT: Leaf (sorted-run) page reference */
+){
+  int i = 0;
+  int rc;
+  int iPg;
+  Page *pPg = 0;
+  LsmBlob blob = {0, 0, 0};
+
+  iPg = (int)pSeg->iRoot;
+  do {
+    LsmPgno *piFirst = 0;
+    if( aPg ){
+      aPg[i++] = iPg;
+      piFirst = &aPg[i];
+    }
+
+    rc = lsmFsDbPageGet(pCsr->pDb->pFS, pSeg, iPg, &pPg);
+    assert( rc==LSM_OK || pPg==0 );
+    if( rc==LSM_OK ){
+      u8 *aData;                  /* Buffer containing page data */
+      int nData;                  /* Size of aData[] in bytes */
+      int iMin;
+      int iMax;
+      int nRec;
+      int flags;
+
+      aData = fsPageData(pPg, &nData);
+      flags = pageGetFlags(aData, nData);
+      if( (flags & SEGMENT_BTREE_FLAG)==0 ) break;
+
+      iPg = (int)pageGetPtr(aData, nData);
+      nRec = pageGetNRec(aData, nData);
+
+      iMin = 0;
+      iMax = nRec-1;
+      while( iMax>=iMin ){
+        int iTry = (iMin+iMax)/2;
+        void *pKeyT; int nKeyT;       /* Key for cell iTry */
+        int iTopicT;                  /* Topic for key pKeyT/nKeyT */
+        LsmPgno iPtr;                 /* Pointer associated with cell iTry */
+        int res;                      /* (pKey - pKeyT) */
+
+        rc = pageGetBtreeKey(
+            pSeg, pPg, iTry, &iPtr, &iTopicT, &pKeyT, &nKeyT, &blob
+        );
+        if( rc!=LSM_OK ) break;
+        if( piFirst && pKeyT==blob.pData ){
+          *piFirst = pageGetBtreeRef(pPg, iTry);
+          piFirst = 0;
+          i++;
+        }
+
+        res = sortedKeyCompare(
+            pCsr->pDb->xCmp, iTopic, pKey, nKey, iTopicT, pKeyT, nKeyT
+        );
+        if( res<0 ){
+          iPg = (int)iPtr;
+          iMax = iTry-1;
+        }else{
+          iMin = iTry+1;
+        }
+      }
+      lsmFsPageRelease(pPg);
+      pPg = 0;
+    }
+  }while( rc==LSM_OK );
+
+  sortedBlobFree(&blob);
+  assert( (rc==LSM_OK)==(pPg!=0) );
+  if( ppPg ){
+    *ppPg = pPg;
+  }else{
+    lsmFsPageRelease(pPg);
+  }
+  return rc;
+}
+
+static int seekInSegment(
+  MultiCursor *pCsr, 
+  SegmentPtr *pPtr,
+  int iTopic,
+  void *pKey, int nKey,
+  int iPg,                        /* Page to search */
+  int eSeek,                      /* Search bias - see above */
+  int *piPtr,                     /* OUT: FC pointer */
+  int *pbStop                     /* OUT: Stop search flag */
+){
+  int iPtr = iPg;
+  int rc = LSM_OK;
+
+  if( pPtr->pSeg->iRoot ){
+    Page *pPg;
+    assert( pPtr->pSeg->iRoot!=0 );
+    rc = seekInBtree(pCsr, pPtr->pSeg, iTopic, pKey, nKey, 0, &pPg);
+    if( rc==LSM_OK ) segmentPtrSetPage(pPtr, pPg);
+  }else{
+    if( iPtr==0 ){
+      iPtr = (int)pPtr->pSeg->iFirst;
+    }
+    if( rc==LSM_OK ){
+      rc = segmentPtrLoadPage(pCsr->pDb->pFS, pPtr, iPtr);
+    }
+  }
+
+  if( rc==LSM_OK ){
+    rc = segmentPtrSeek(pCsr, pPtr, iTopic, pKey, nKey, eSeek, piPtr, pbStop);
+  }
+  return rc;
+}
+
+/*
+** Seek each segment pointer in the array of (pLvl->nRight+1) at aPtr[].
+**
+** pbStop:
+**   This parameter is only significant if parameter eSeek is set to
+**   LSM_SEEK_EQ. In this case, it is set to true before returning if
+**   the seek operation is finished. This can happen in two ways:
+**   
+**     a) A key matching (pKey/nKey) is found, or
+**     b) A point-delete or range-delete deleting the key is found.
+**
+**   In case (a), the multi-cursor CURSOR_SEEK_EQ flag is set and the pCsr->key
+**   and pCsr->val blobs populated before returning.
+*/
+static int seekInLevel(
+  MultiCursor *pCsr,              /* Sorted cursor object to seek */
+  SegmentPtr *aPtr,               /* Pointer to array of (nRhs+1) SPs */
+  int eSeek,                      /* Search bias - see above */
+  int iTopic,                     /* Key topic to search for */
+  void *pKey, int nKey,           /* Key to search for */
+  LsmPgno *piPgno,                /* IN/OUT: fraction cascade pointer (or 0) */
+  int *pbStop                     /* OUT: See above */
+){
+  Level *pLvl = aPtr[0].pLevel;   /* Level to seek within */
+  int rc = LSM_OK;                /* Return code */
+  int iOut = 0;                   /* Pointer to return to caller */
+  int res = -1;                   /* Result of xCmp(pKey, split) */
+  int nRhs = pLvl->nRight;        /* Number of right-hand-side segments */
+  int bStop = 0;
+
+  /* If this is a composite level (one currently undergoing an incremental
+  ** merge), figure out if the search key is larger or smaller than the
+  ** levels split-key.  */
+  if( nRhs ){
+    res = sortedKeyCompare(pCsr->pDb->xCmp, iTopic, pKey, nKey, 
+        pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
+    );
+  }
+
+  /* If (res<0), then key pKey/nKey is smaller than the split-key (or this
+  ** is not a composite level and there is no split-key). Search the 
+  ** left-hand-side of the level in this case.  */
+  if( res<0 ){
+    int i;
+    int iPtr = 0;
+    if( nRhs==0 ) iPtr = (int)*piPgno;
+
+    rc = seekInSegment(
+        pCsr, &aPtr[0], iTopic, pKey, nKey, iPtr, eSeek, &iOut, &bStop
+    );
+    if( rc==LSM_OK && nRhs>0 && eSeek==LSM_SEEK_GE && aPtr[0].pPg==0 ){
+      res = 0;
+    }
+    for(i=1; i<=nRhs; i++){
+      segmentPtrReset(&aPtr[i], LSM_SEGMENTPTR_FREE_THRESHOLD);
+    }
+  }
+  
+  if( res>=0 ){
+    int bHit = 0;                 /* True if at least one rhs is not EOF */
+    int iPtr = (int)*piPgno;
+    int i;
+    segmentPtrReset(&aPtr[0], LSM_SEGMENTPTR_FREE_THRESHOLD);
+    for(i=1; rc==LSM_OK && i<=nRhs && bStop==0; i++){
+      SegmentPtr *pPtr = &aPtr[i];
+      iOut = 0;
+      rc = seekInSegment(
+          pCsr, pPtr, iTopic, pKey, nKey, iPtr, eSeek, &iOut, &bStop
+      );
+      iPtr = iOut;
+
+      /* If the segment-pointer has settled on a key that is smaller than
+      ** the splitkey, invalidate the segment-pointer.  */
+      if( pPtr->pPg ){
+        res = sortedKeyCompare(pCsr->pDb->xCmp, 
+            rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey, 
+            pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
+        );
+        if( res<0 ){
+          if( pPtr->eType & LSM_START_DELETE ){
+            pPtr->eType &= ~LSM_INSERT;
+            pPtr->pKey = pLvl->pSplitKey;
+            pPtr->nKey = pLvl->nSplitKey;
+            pPtr->pVal = 0;
+            pPtr->nVal = 0;
+          }else{
+            segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
+          }
+        }
+      }
+
+      if( aPtr[i].pKey ) bHit = 1;
+    }
+
+    if( rc==LSM_OK && eSeek==LSM_SEEK_LE && bHit==0 ){
+      rc = segmentPtrEnd(pCsr, &aPtr[0], 1);
+    }
+  }
+
+  assert( eSeek==LSM_SEEK_EQ || bStop==0 );
+  *piPgno = iOut;
+  *pbStop = bStop;
+  return rc;
+}
+
+static void multiCursorGetKey(
+  MultiCursor *pCsr, 
+  int iKey,
+  int *peType,                    /* OUT: Key type (SORTED_WRITE etc.) */
+  void **ppKey,                   /* OUT: Pointer to buffer containing key */
+  int *pnKey                      /* OUT: Size of *ppKey in bytes */
+){
+  int nKey = 0;
+  void *pKey = 0;
+  int eType = 0;
+
+  switch( iKey ){
+    case CURSOR_DATA_TREE0:
+    case CURSOR_DATA_TREE1: {
+      TreeCursor *pTreeCsr = pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0];
+      if( lsmTreeCursorValid(pTreeCsr) ){
+        lsmTreeCursorKey(pTreeCsr, &eType, &pKey, &nKey);
+      }
+      break;
+    }
+
+    case CURSOR_DATA_SYSTEM: {
+      Snapshot *pWorker = pCsr->pDb->pWorker;
+      if( pWorker && (pCsr->flags & CURSOR_FLUSH_FREELIST) ){
+        int nEntry = pWorker->freelist.nEntry;
+        if( pCsr->iFree < (nEntry*2) ){
+          FreelistEntry *aEntry = pWorker->freelist.aEntry;
+          int i = nEntry - 1 - (pCsr->iFree / 2);
+          u32 iKey2 = 0;
+
+          if( (pCsr->iFree % 2) ){
+            eType = LSM_END_DELETE|LSM_SYSTEMKEY;
+            iKey2 = aEntry[i].iBlk-1;
+          }else if( aEntry[i].iId>=0 ){
+            eType = LSM_INSERT|LSM_SYSTEMKEY;
+            iKey2 = aEntry[i].iBlk;
+
+            /* If the in-memory entry immediately before this one was a
+             ** DELETE, and the block number is one greater than the current
+             ** block number, mark this entry as an "end-delete-range". */
+            if( i<(nEntry-1) && aEntry[i+1].iBlk==iKey2+1 && aEntry[i+1].iId<0 ){
+              eType |= LSM_END_DELETE;
+            }
+
+          }else{
+            eType = LSM_START_DELETE|LSM_SYSTEMKEY;
+            iKey2 = aEntry[i].iBlk + 1;
+          }
+
+          /* If the in-memory entry immediately after this one is a
+          ** DELETE, and the block number is one less than the current
+          ** key, mark this entry as an "start-delete-range".  */
+          if( i>0 && aEntry[i-1].iBlk==iKey2-1 && aEntry[i-1].iId<0 ){
+            eType |= LSM_START_DELETE;
+          }
+
+          pKey = pCsr->pSystemVal;
+          nKey = 4;
+          lsmPutU32(pKey, ~iKey2);
+        }
+      }
+      break;
+    }
+
+    default: {
+      int iPtr = iKey - CURSOR_DATA_SEGMENT;
+      assert( iPtr>=0 );
+      if( iPtr==pCsr->nPtr ){
+        if( pCsr->pBtCsr ){
+          pKey = pCsr->pBtCsr->pKey;
+          nKey = pCsr->pBtCsr->nKey;
+          eType = pCsr->pBtCsr->eType;
+        }
+      }else if( iPtr<pCsr->nPtr ){
+        SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
+        if( pPtr->pPg ){
+          pKey = pPtr->pKey;
+          nKey = pPtr->nKey;
+          eType = pPtr->eType;
+        }
+      }
+      break;
+    }
+  }
+
+  if( peType ) *peType = eType;
+  if( pnKey ) *pnKey = nKey;
+  if( ppKey ) *ppKey = pKey;
+}
+
+static int sortedDbKeyCompare(
+  MultiCursor *pCsr,
+  int iLhsFlags, void *pLhsKey, int nLhsKey,
+  int iRhsFlags, void *pRhsKey, int nRhsKey
+){
+  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
+  int res;
+
+  /* Compare the keys, including the system flag. */
+  res = sortedKeyCompare(xCmp, 
+    rtTopic(iLhsFlags), pLhsKey, nLhsKey,
+    rtTopic(iRhsFlags), pRhsKey, nRhsKey
+  );
+
+  /* If a key has the LSM_START_DELETE flag set, but not the LSM_INSERT or
+  ** LSM_POINT_DELETE flags, it is considered a delta larger. This prevents
+  ** the beginning of an open-ended set from masking a database entry or
+  ** delete at a lower level.  */
+  if( res==0 && (pCsr->flags & CURSOR_IGNORE_DELETE) ){
+    const int m = LSM_POINT_DELETE|LSM_INSERT|LSM_END_DELETE |LSM_START_DELETE;
+    int iDel1 = 0;
+    int iDel2 = 0;
+
+    if( LSM_START_DELETE==(iLhsFlags & m) ) iDel1 = +1;
+    if( LSM_END_DELETE  ==(iLhsFlags & m) ) iDel1 = -1;
+    if( LSM_START_DELETE==(iRhsFlags & m) ) iDel2 = +1;
+    if( LSM_END_DELETE  ==(iRhsFlags & m) ) iDel2 = -1;
+
+    res = (iDel1 - iDel2);
+  }
+
+  return res;
+}
+
+static void multiCursorDoCompare(MultiCursor *pCsr, int iOut, int bReverse){
+  int i1;
+  int i2;
+  int iRes;
+  void *pKey1; int nKey1; int eType1;
+  void *pKey2; int nKey2; int eType2;
+  const int mul = (bReverse ? -1 : 1);
+
+  assert( pCsr->aTree && iOut<pCsr->nTree );
+  if( iOut>=(pCsr->nTree/2) ){
+    i1 = (iOut - pCsr->nTree/2) * 2;
+    i2 = i1 + 1;
+  }else{
+    i1 = pCsr->aTree[iOut*2];
+    i2 = pCsr->aTree[iOut*2+1];
+  }
+
+  multiCursorGetKey(pCsr, i1, &eType1, &pKey1, &nKey1);
+  multiCursorGetKey(pCsr, i2, &eType2, &pKey2, &nKey2);
+
+  if( pKey1==0 ){
+    iRes = i2;
+  }else if( pKey2==0 ){
+    iRes = i1;
+  }else{
+    int res;
+
+    /* Compare the keys */
+    res = sortedDbKeyCompare(pCsr,
+        eType1, pKey1, nKey1, eType2, pKey2, nKey2
+    );
+
+    res = res * mul;
+    if( res==0 ){
+      /* The two keys are identical. Normally, this means that the key from
+      ** the newer run clobbers the old. However, if the newer key is a
+      ** separator key, or a range-delete-boundary only, do not allow it
+      ** to clobber an older entry.  */
+      int nc1 = (eType1 & (LSM_INSERT|LSM_POINT_DELETE))==0;
+      int nc2 = (eType2 & (LSM_INSERT|LSM_POINT_DELETE))==0;
+      iRes = (nc1 > nc2) ? i2 : i1;
+    }else if( res<0 ){
+      iRes = i1;
+    }else{
+      iRes = i2;
+    }
+  }
+
+  pCsr->aTree[iOut] = iRes;
+}
+
+/*
+** This function advances segment pointer iPtr belonging to multi-cursor
+** pCsr forward (bReverse==0) or backward (bReverse!=0).
+**
+** If the segment pointer points to a segment that is part of a composite
+** level, then the following special case is handled.
+**
+**   * If iPtr is the lhs of a composite level, and the cursor is being
+**     advanced forwards, and segment iPtr is at EOF, move all pointers
+**     that correspond to rhs segments of the same level to the first
+**     key in their respective data.
+*/
+static int segmentCursorAdvance(
+  MultiCursor *pCsr, 
+  int iPtr,
+  int bReverse
+){
+  int rc;
+  SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
+  Level *pLvl = pPtr->pLevel;
+  int bComposite;                 /* True if pPtr is part of composite level */
+
+  /* Advance the segment-pointer object. */
+  rc = segmentPtrAdvance(pCsr, pPtr, bReverse);
+  if( rc!=LSM_OK ) return rc;
+
+  bComposite = (pLvl->nRight>0 && pCsr->nPtr>pLvl->nRight);
+  if( bComposite && pPtr->pPg==0 ){
+    int bFix = 0;
+    if( (bReverse==0)==(pPtr->pSeg==&pLvl->lhs) ){
+      int i;
+      if( bReverse ){
+        SegmentPtr *pLhs = &pCsr->aPtr[iPtr - 1 - (pPtr->pSeg - pLvl->aRhs)];
+        for(i=0; i<pLvl->nRight; i++){
+          if( pLhs[i+1].pPg ) break;
+        }
+        if( i==pLvl->nRight ){
+          bFix = 1;
+          rc = segmentPtrEnd(pCsr, pLhs, 1);
+        }
+      }else{
+        bFix = 1;
+        for(i=0; rc==LSM_OK && i<pLvl->nRight; i++){
+          rc = sortedRhsFirst(pCsr, pLvl, &pCsr->aPtr[iPtr+1+i]);
+        }
+      }
+    }
+
+    if( bFix ){
+      int i;
+      for(i=pCsr->nTree-1; i>0; i--){
+        multiCursorDoCompare(pCsr, i, bReverse);
+      }
+    }
+  }
+
+#if 0
+  if( bComposite && pPtr->pSeg==&pLvl->lhs       /* lhs of composite level */
+   && bReverse==0                                /* csr advanced forwards */
+   && pPtr->pPg==0                               /* segment at EOF */
+  ){
+    int i;
+    for(i=0; rc==LSM_OK && i<pLvl->nRight; i++){
+      rc = sortedRhsFirst(pCsr, pLvl, &pCsr->aPtr[iPtr+1+i]);
+    }
+    for(i=pCsr->nTree-1; i>0; i--){
+      multiCursorDoCompare(pCsr, i, 0);
+    }
+  }
+#endif
+
+  return rc;
+}
+
+static void mcursorFreeComponents(MultiCursor *pCsr){
+  int i;
+  lsm_env *pEnv = pCsr->pDb->pEnv;
+
+  /* Close the tree cursor, if any. */
+  lsmTreeCursorDestroy(pCsr->apTreeCsr[0]);
+  lsmTreeCursorDestroy(pCsr->apTreeCsr[1]);
+
+  /* Reset the segment pointers */
+  for(i=0; i<pCsr->nPtr; i++){
+    segmentPtrReset(&pCsr->aPtr[i], 0);
+  }
+
+  /* And the b-tree cursor, if any */
+  btreeCursorFree(pCsr->pBtCsr);
+
+  /* Free allocations */
+  lsmFree(pEnv, pCsr->aPtr);
+  lsmFree(pEnv, pCsr->aTree);
+  lsmFree(pEnv, pCsr->pSystemVal);
+
+  /* Zero fields */
+  pCsr->nPtr = 0;
+  pCsr->aPtr = 0;
+  pCsr->nTree = 0;
+  pCsr->aTree = 0;
+  pCsr->pSystemVal = 0;
+  pCsr->apTreeCsr[0] = 0;
+  pCsr->apTreeCsr[1] = 0;
+  pCsr->pBtCsr = 0;
+}
+
+void lsmMCursorFreeCache(lsm_db *pDb){
+  MultiCursor *p;
+  MultiCursor *pNext;
+  for(p=pDb->pCsrCache; p; p=pNext){
+    pNext = p->pNext;
+    lsmMCursorClose(p, 0);
+  }
+  pDb->pCsrCache = 0;
+}
+
+/*
+** Close the cursor passed as the first argument.
+**
+** If the bCache parameter is true, then shift the cursor to the pCsrCache
+** list for possible reuse instead of actually deleting it.
+*/
+void lsmMCursorClose(MultiCursor *pCsr, int bCache){
+  if( pCsr ){
+    lsm_db *pDb = pCsr->pDb;
+    MultiCursor **pp;             /* Iterator variable */
+
+    /* The cursor may or may not be currently part of the linked list 
+    ** starting at lsm_db.pCsr. If it is, extract it.  */
+    for(pp=&pDb->pCsr; *pp; pp=&((*pp)->pNext)){
+      if( *pp==pCsr ){
+        *pp = pCsr->pNext;
+        break;
+      }
+    }
+
+    if( bCache ){
+      int i;                      /* Used to iterate through segment-pointers */
+
+      /* Release any page references held by this cursor. */
+      assert( !pCsr->pBtCsr );
+      for(i=0; i<pCsr->nPtr; i++){
+        SegmentPtr *pPtr = &pCsr->aPtr[i];
+        lsmFsPageRelease(pPtr->pPg);
+        pPtr->pPg = 0;
+      }
+
+      /* Reset the tree cursors */
+      lsmTreeCursorReset(pCsr->apTreeCsr[0]);
+      lsmTreeCursorReset(pCsr->apTreeCsr[1]);
+
+      /* Add the cursor to the pCsrCache list */
+      pCsr->pNext = pDb->pCsrCache;
+      pDb->pCsrCache = pCsr;
+    }else{
+      /* Free the allocation used to cache the current key, if any. */
+      sortedBlobFree(&pCsr->key);
+      sortedBlobFree(&pCsr->val);
+
+      /* Free the component cursors */
+      mcursorFreeComponents(pCsr);
+
+      /* Free the cursor structure itself */
+      lsmFree(pDb->pEnv, pCsr);
+    }
+  }
+}
+
+#define TREE_NONE 0
+#define TREE_OLD  1
+#define TREE_BOTH 2
+
+/*
+** Parameter eTree is one of TREE_OLD or TREE_BOTH.
+*/
+static int multiCursorAddTree(MultiCursor *pCsr, Snapshot *pSnap, int eTree){
+  int rc = LSM_OK;
+  lsm_db *db = pCsr->pDb;
+
+  /* Add a tree cursor on the 'old' tree, if it exists. */
+  if( eTree!=TREE_NONE 
+   && lsmTreeHasOld(db) 
+   && db->treehdr.iOldLog!=pSnap->iLogOff 
+  ){
+    rc = lsmTreeCursorNew(db, 1, &pCsr->apTreeCsr[1]);
+  }
+
+  /* Add a tree cursor on the 'current' tree, if required. */
+  if( rc==LSM_OK && eTree==TREE_BOTH ){
+    rc = lsmTreeCursorNew(db, 0, &pCsr->apTreeCsr[0]);
+  }
+
+  return rc;
+}
+
+static int multiCursorAddRhs(MultiCursor *pCsr, Level *pLvl){
+  int i;
+  int nRhs = pLvl->nRight;
+
+  assert( pLvl->nRight>0 );
+  assert( pCsr->aPtr==0 );
+  pCsr->aPtr = lsmMallocZero(pCsr->pDb->pEnv, sizeof(SegmentPtr) * nRhs);
+  if( !pCsr->aPtr ) return LSM_NOMEM_BKPT;
+  pCsr->nPtr = nRhs;
+
+  for(i=0; i<nRhs; i++){
+    pCsr->aPtr[i].pSeg = &pLvl->aRhs[i];
+    pCsr->aPtr[i].pLevel = pLvl;
+  }
+
+  return LSM_OK;
+}
+
+static void multiCursorAddOne(MultiCursor *pCsr, Level *pLvl, int *pRc){
+  if( *pRc==LSM_OK ){
+    int iPtr = pCsr->nPtr;
+    int i;
+    pCsr->aPtr[iPtr].pLevel = pLvl;
+    pCsr->aPtr[iPtr].pSeg = &pLvl->lhs;
+    iPtr++;
+    for(i=0; i<pLvl->nRight; i++){
+      pCsr->aPtr[iPtr].pLevel = pLvl;
+      pCsr->aPtr[iPtr].pSeg = &pLvl->aRhs[i];
+      iPtr++;
+    }
+
+    if( pLvl->nRight && pLvl->pSplitKey==0 ){
+      sortedSplitkey(pCsr->pDb, pLvl, pRc);
+    }
+    pCsr->nPtr = iPtr;
+  }
+}
+
+static int multiCursorAddAll(MultiCursor *pCsr, Snapshot *pSnap){
+  Level *pLvl;
+  int nPtr = 0;
+  int rc = LSM_OK;
+
+  for(pLvl=pSnap->pLevel; pLvl; pLvl=pLvl->pNext){
+    /* If the LEVEL_INCOMPLETE flag is set, then this function is being
+    ** called (indirectly) from within a sortedNewToplevel() call to
+    ** construct pLvl. In this case ignore pLvl - this cursor is going to
+    ** be used to retrieve a freelist entry from the LSM, and the partially
+    ** complete level may confuse it.  */
+    if( pLvl->flags & LEVEL_INCOMPLETE ) continue;
+    nPtr += (1 + pLvl->nRight);
+  }
+
+  assert( pCsr->aPtr==0 );
+  pCsr->aPtr = lsmMallocZeroRc(pCsr->pDb->pEnv, sizeof(SegmentPtr) * nPtr, &rc);
+
+  for(pLvl=pSnap->pLevel; pLvl; pLvl=pLvl->pNext){
+    if( (pLvl->flags & LEVEL_INCOMPLETE)==0 ){
+      multiCursorAddOne(pCsr, pLvl, &rc);
+    }
+  }
+
+  return rc;
+}
+
+static int multiCursorInit(MultiCursor *pCsr, Snapshot *pSnap){
+  int rc;
+  rc = multiCursorAddAll(pCsr, pSnap);
+  if( rc==LSM_OK ){
+    rc = multiCursorAddTree(pCsr, pSnap, TREE_BOTH);
+  }
+  pCsr->flags |= (CURSOR_IGNORE_SYSTEM | CURSOR_IGNORE_DELETE);
+  return rc;
+}
+
+static MultiCursor *multiCursorNew(lsm_db *db, int *pRc){
+  MultiCursor *pCsr;
+  pCsr = (MultiCursor *)lsmMallocZeroRc(db->pEnv, sizeof(MultiCursor), pRc);
+  if( pCsr ){
+    pCsr->pNext = db->pCsr;
+    db->pCsr = pCsr;
+    pCsr->pDb = db;
+  }
+  return pCsr;
+}
+
+
+void lsmSortedRemap(lsm_db *pDb){
+  MultiCursor *pCsr;
+  for(pCsr=pDb->pCsr; pCsr; pCsr=pCsr->pNext){
+    int iPtr;
+    if( pCsr->pBtCsr ){
+      btreeCursorLoadKey(pCsr->pBtCsr);
+    }
+    for(iPtr=0; iPtr<pCsr->nPtr; iPtr++){
+      segmentPtrLoadCell(&pCsr->aPtr[iPtr], pCsr->aPtr[iPtr].iCell);
+    }
+  }
+}
+
+static void multiCursorReadSeparators(MultiCursor *pCsr){
+  if( pCsr->nPtr>0 ){
+    pCsr->flags |= CURSOR_READ_SEPARATORS;
+  }
+}
+
+/*
+** Have this cursor skip over SORTED_DELETE entries.
+*/
+static void multiCursorIgnoreDelete(MultiCursor *pCsr){
+  if( pCsr ) pCsr->flags |= CURSOR_IGNORE_DELETE;
+}
+
+/*
+** If the free-block list is not empty, then have this cursor visit a key
+** with (a) the system bit set, and (b) the key "FREELIST" and (c) a value 
+** blob containing the serialized free-block list.
+*/
+static int multiCursorVisitFreelist(MultiCursor *pCsr){
+  int rc = LSM_OK;
+  pCsr->flags |= CURSOR_FLUSH_FREELIST;
+  pCsr->pSystemVal = lsmMallocRc(pCsr->pDb->pEnv, 4 + 8, &rc);
+  return rc;
+}
+
+/*
+** Allocate and return a new database cursor.
+**
+** This method should only be called to allocate user cursors. As it may
+** recycle a cursor from lsm_db.pCsrCache.
+*/
+int lsmMCursorNew(
+  lsm_db *pDb,                    /* Database handle */
+  MultiCursor **ppCsr             /* OUT: Allocated cursor */
+){
+  MultiCursor *pCsr = 0;
+  int rc = LSM_OK;
+
+  if( pDb->pCsrCache ){
+    int bOld;                     /* True if there is an old in-memory tree */
+
+    /* Remove a cursor from the pCsrCache list and add it to the open list. */
+    pCsr = pDb->pCsrCache;
+    pDb->pCsrCache = pCsr->pNext;
+    pCsr->pNext = pDb->pCsr;
+    pDb->pCsr = pCsr;
+
+    /* The cursor can almost be used as is, except that the old in-memory
+    ** tree cursor may be present and not required, or required and not
+    ** present. Fix this if required.  */
+    bOld = (lsmTreeHasOld(pDb) && pDb->treehdr.iOldLog!=pDb->pClient->iLogOff);
+    if( !bOld && pCsr->apTreeCsr[1] ){
+      lsmTreeCursorDestroy(pCsr->apTreeCsr[1]);
+      pCsr->apTreeCsr[1] = 0;
+    }else if( bOld && !pCsr->apTreeCsr[1] ){
+      rc = lsmTreeCursorNew(pDb, 1, &pCsr->apTreeCsr[1]);
+    }
+
+    pCsr->flags = (CURSOR_IGNORE_SYSTEM | CURSOR_IGNORE_DELETE);
+
+  }else{
+    pCsr = multiCursorNew(pDb, &rc);
+    if( rc==LSM_OK ) rc = multiCursorInit(pCsr, pDb->pClient);
+  }
+
+  if( rc!=LSM_OK ){
+    lsmMCursorClose(pCsr, 0);
+    pCsr = 0;
+  }
+  assert( (rc==LSM_OK)==(pCsr!=0) );
+  *ppCsr = pCsr;
+  return rc;
+}
+
+static int multiCursorGetVal(
+  MultiCursor *pCsr, 
+  int iVal, 
+  void **ppVal, 
+  int *pnVal
+){
+  int rc = LSM_OK;
+
+  *ppVal = 0;
+  *pnVal = 0;
+
+  switch( iVal ){
+    case CURSOR_DATA_TREE0:
+    case CURSOR_DATA_TREE1: {
+      TreeCursor *pTreeCsr = pCsr->apTreeCsr[iVal-CURSOR_DATA_TREE0];
+      if( lsmTreeCursorValid(pTreeCsr) ){
+        lsmTreeCursorValue(pTreeCsr, ppVal, pnVal);
+      }else{
+        *ppVal = 0;
+        *pnVal = 0;
+      }
+      break;
+    }
+
+    case CURSOR_DATA_SYSTEM: {
+      Snapshot *pWorker = pCsr->pDb->pWorker;
+      if( pWorker 
+       && (pCsr->iFree % 2)==0
+       && pCsr->iFree < (pWorker->freelist.nEntry*2)
+      ){
+        int iEntry = pWorker->freelist.nEntry - 1 - (pCsr->iFree / 2);
+        u8 *aVal = &((u8 *)(pCsr->pSystemVal))[4];
+        lsmPutU64(aVal, pWorker->freelist.aEntry[iEntry].iId);
+        *ppVal = aVal;
+        *pnVal = 8;
+      }
+      break;
+    }
+
+    default: {
+      int iPtr = iVal-CURSOR_DATA_SEGMENT;
+      if( iPtr<pCsr->nPtr ){
+        SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
+        if( pPtr->pPg ){
+          *ppVal = pPtr->pVal;
+          *pnVal = pPtr->nVal;
+        }
+      }
+    }
+  }
+
+  assert( rc==LSM_OK || (*ppVal==0 && *pnVal==0) );
+  return rc;
+}
+
+static int multiCursorAdvance(MultiCursor *pCsr, int bReverse);
+
+/*
+** This function is called by worker connections to walk the part of the
+** free-list stored within the LSM data structure.
+*/
+int lsmSortedWalkFreelist(
+  lsm_db *pDb,                    /* Database handle */
+  int bReverse,                   /* True to iterate from largest to smallest */
+  int (*x)(void *, int, i64),     /* Callback function */
+  void *pCtx                      /* First argument to pass to callback */
+){
+  MultiCursor *pCsr;              /* Cursor used to read db */
+  int rc = LSM_OK;                /* Return Code */
+  Snapshot *pSnap = 0;
+
+  assert( pDb->pWorker );
+  if( pDb->bIncrMerge ){
+    rc = lsmCheckpointDeserialize(pDb, 0, pDb->pShmhdr->aSnap1, &pSnap);
+    if( rc!=LSM_OK ) return rc;
+  }else{
+    pSnap = pDb->pWorker;
+  }
+
+  pCsr = multiCursorNew(pDb, &rc);
+  if( pCsr ){
+    rc = multiCursorAddAll(pCsr, pSnap);
+    pCsr->flags |= CURSOR_IGNORE_DELETE;
+  }
+  
+  if( rc==LSM_OK ){
+    if( bReverse==0 ){
+      rc = lsmMCursorLast(pCsr);
+    }else{
+      rc = lsmMCursorSeek(pCsr, 1, "", 0, LSM_SEEK_GE);
+    }
+
+    while( rc==LSM_OK && lsmMCursorValid(pCsr) && rtIsSystem(pCsr->eType) ){
+      void *pKey; int nKey;
+      void *pVal = 0; int nVal = 0;
+
+      rc = lsmMCursorKey(pCsr, &pKey, &nKey);
+      if( rc==LSM_OK ) rc = lsmMCursorValue(pCsr, &pVal, &nVal);
+      if( rc==LSM_OK && (nKey!=4 || nVal!=8) ) rc = LSM_CORRUPT_BKPT;
+
+      if( rc==LSM_OK ){
+        int iBlk;
+        i64 iSnap;
+        iBlk = (int)(~(lsmGetU32((u8 *)pKey)));
+        iSnap = (i64)lsmGetU64((u8 *)pVal);
+        if( x(pCtx, iBlk, iSnap) ) break;
+        rc = multiCursorAdvance(pCsr, !bReverse);
+      }
+    }
+  }
+
+  lsmMCursorClose(pCsr, 0);
+  if( pSnap!=pDb->pWorker ){
+    lsmFreeSnapshot(pDb->pEnv, pSnap);
+  }
+
+  return rc;
+}
+
+int lsmSortedLoadFreelist(
+  lsm_db *pDb,                    /* Database handle (must be worker) */
+  void **ppVal,                   /* OUT: Blob containing LSM free-list */
+  int *pnVal                      /* OUT: Size of *ppVal blob in bytes */
+){
+  MultiCursor *pCsr;              /* Cursor used to retreive free-list */
+  int rc = LSM_OK;                /* Return Code */
+
+  assert( pDb->pWorker );
+  assert( *ppVal==0 && *pnVal==0 );
+
+  pCsr = multiCursorNew(pDb, &rc);
+  if( pCsr ){
+    rc = multiCursorAddAll(pCsr, pDb->pWorker);
+    pCsr->flags |= CURSOR_IGNORE_DELETE;
+  }
+  
+  if( rc==LSM_OK ){
+    rc = lsmMCursorLast(pCsr);
+    if( rc==LSM_OK 
+     && rtIsWrite(pCsr->eType) && rtIsSystem(pCsr->eType)
+     && pCsr->key.nData==8 
+     && 0==memcmp(pCsr->key.pData, "FREELIST", 8)
+    ){
+      void *pVal; int nVal;         /* Value read from database */
+      rc = lsmMCursorValue(pCsr, &pVal, &nVal);
+      if( rc==LSM_OK ){
+        *ppVal = lsmMallocRc(pDb->pEnv, nVal, &rc);
+        if( *ppVal ){
+          memcpy(*ppVal, pVal, nVal);
+          *pnVal = nVal;
+        }
+      }
+    }
+
+    lsmMCursorClose(pCsr, 0);
+  }
+
+  return rc;
+}
+
+static int multiCursorAllocTree(MultiCursor *pCsr){
+  int rc = LSM_OK;
+  if( pCsr->aTree==0 ){
+    int nByte;                    /* Bytes of space to allocate */
+    int nMin;                     /* Total number of cursors being merged */
+
+    nMin = CURSOR_DATA_SEGMENT + pCsr->nPtr + (pCsr->pBtCsr!=0);
+    pCsr->nTree = 2;
+    while( pCsr->nTree<nMin ){
+      pCsr->nTree = pCsr->nTree*2;
+    }
+
+    nByte = sizeof(int)*pCsr->nTree*2;
+    pCsr->aTree = (int *)lsmMallocZeroRc(pCsr->pDb->pEnv, nByte, &rc);
+  }
+  return rc;
+}
+
+static void multiCursorCacheKey(MultiCursor *pCsr, int *pRc){
+  if( *pRc==LSM_OK ){
+    void *pKey;
+    int nKey;
+    multiCursorGetKey(pCsr, pCsr->aTree[1], &pCsr->eType, &pKey, &nKey);
+    *pRc = sortedBlobSet(pCsr->pDb->pEnv, &pCsr->key, pKey, nKey);
+  }
+}
+
+#ifdef LSM_DEBUG_EXPENSIVE
+static void assertCursorTree(MultiCursor *pCsr){
+  int bRev = !!(pCsr->flags & CURSOR_PREV_OK);
+  int *aSave = pCsr->aTree;
+  int nSave = pCsr->nTree;
+  int rc;
+
+  pCsr->aTree = 0;
+  pCsr->nTree = 0;
+  rc = multiCursorAllocTree(pCsr);
+  if( rc==LSM_OK ){
+    int i;
+    for(i=pCsr->nTree-1; i>0; i--){
+      multiCursorDoCompare(pCsr, i, bRev);
+    }
+
+    assert( nSave==pCsr->nTree 
+        && 0==memcmp(aSave, pCsr->aTree, sizeof(int)*nSave)
+    );
+
+    lsmFree(pCsr->pDb->pEnv, pCsr->aTree);
+  }
+
+  pCsr->aTree = aSave;
+  pCsr->nTree = nSave;
+}
+#else
+# define assertCursorTree(x)
+#endif
+
+static int mcursorLocationOk(MultiCursor *pCsr, int bDeleteOk){
+  int eType = pCsr->eType;
+  int iKey;
+  int i;
+  int rdmask;
+  
+  assert( pCsr->flags & (CURSOR_NEXT_OK|CURSOR_PREV_OK) );
+  assertCursorTree(pCsr);
+
+  rdmask = (pCsr->flags & CURSOR_NEXT_OK) ? LSM_END_DELETE : LSM_START_DELETE;
+
+  /* If the cursor does not currently point to an actual database key (i.e.
+  ** it points to a delete key, or the start or end of a range-delete), and
+  ** the CURSOR_IGNORE_DELETE flag is set, skip past this entry.  */
+  if( (pCsr->flags & CURSOR_IGNORE_DELETE) && bDeleteOk==0 ){
+    if( (eType & LSM_INSERT)==0 ) return 0;
+  }
+
+  /* If the cursor points to a system key (free-list entry), and the
+  ** CURSOR_IGNORE_SYSTEM flag is set, skip thie entry.  */
+  if( (pCsr->flags & CURSOR_IGNORE_SYSTEM) && rtTopic(eType)!=0 ){
+    return 0;
+  }
+
+#ifndef NDEBUG
+  /* This block fires assert() statements to check one of the assumptions
+  ** in the comment below - that if the lhs sub-cursor of a level undergoing
+  ** a merge is valid, then all the rhs sub-cursors must be at EOF. 
+  **
+  ** Also assert that all rhs sub-cursors are either at EOF or point to
+  ** a key that is not less than the level split-key.  */
+  for(i=0; i<pCsr->nPtr; i++){
+    SegmentPtr *pPtr = &pCsr->aPtr[i];
+    Level *pLvl = pPtr->pLevel;
+    if( pLvl->nRight && pPtr->pPg ){
+      if( pPtr->pSeg==&pLvl->lhs ){
+        int j;
+        for(j=0; j<pLvl->nRight; j++) assert( pPtr[j+1].pPg==0 );
+      }else{
+        int res = sortedKeyCompare(pCsr->pDb->xCmp, 
+            rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
+            pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
+        );
+        assert( res>=0 );
+      }
+    }
+  }
+#endif
+
+  /* Now check if this key has already been deleted by a range-delete. If 
+  ** so, skip past it.
+  **
+  ** Assume, for the moment, that the tree contains no levels currently 
+  ** undergoing incremental merge, and that this cursor is iterating forwards
+  ** through the database keys. The cursor currently points to a key in
+  ** level L. This key has already been deleted if any of the sub-cursors
+  ** that point to levels newer than L (or to the in-memory tree) point to
+  ** a key greater than the current key with the LSM_END_DELETE flag set.
+  **
+  ** Or, if the cursor is iterating backwards through data keys, if any
+  ** such sub-cursor points to a key smaller than the current key with the
+  ** LSM_START_DELETE flag set.
+  **
+  ** Why it works with levels undergoing a merge too:
+  **
+  ** When a cursor iterates forwards, the sub-cursors for the rhs of a 
+  ** level are only activated once the lhs reaches EOF. So when iterating
+  ** forwards, the keys visited are the same as if the level was completely
+  ** merged.
+  **
+  ** If the cursor is iterating backwards, then the lhs sub-cursor is not 
+  ** initialized until the last of the rhs sub-cursors has reached EOF.
+  ** Additionally, if the START_DELETE flag is set on the last entry (in
+  ** reverse order - so the entry with the smallest key) of a rhs sub-cursor,
+  ** then a pseudo-key equal to the levels split-key with the END_DELETE
+  ** flag set is visited by the sub-cursor.
+  */ 
+  iKey = pCsr->aTree[1];
+  for(i=0; i<iKey; i++){
+    int csrflags;
+    multiCursorGetKey(pCsr, i, &csrflags, 0, 0);
+    if( (rdmask & csrflags) ){
+      const int SD_ED = (LSM_START_DELETE|LSM_END_DELETE);
+      if( (csrflags & SD_ED)==SD_ED 
+       || (pCsr->flags & CURSOR_IGNORE_DELETE)==0
+      ){
+        void *pKey; int nKey;
+        multiCursorGetKey(pCsr, i, 0, &pKey, &nKey);
+        if( 0==sortedKeyCompare(pCsr->pDb->xCmp,
+              rtTopic(eType), pCsr->key.pData, pCsr->key.nData,
+              rtTopic(csrflags), pKey, nKey
+        )){
+          continue;
+        }
+      }
+      return 0;
+    }
+  }
+
+  /* The current cursor position is one this cursor should visit. Return 1. */
+  return 1;
+}
+
+static int multiCursorSetupTree(MultiCursor *pCsr, int bRev){
+  int rc;
+
+  rc = multiCursorAllocTree(pCsr);
+  if( rc==LSM_OK ){
+    int i;
+    for(i=pCsr->nTree-1; i>0; i--){
+      multiCursorDoCompare(pCsr, i, bRev);
+    }
+  }
+
+  assertCursorTree(pCsr);
+  multiCursorCacheKey(pCsr, &rc);
+
+  if( rc==LSM_OK && mcursorLocationOk(pCsr, 0)==0 ){
+    rc = multiCursorAdvance(pCsr, bRev);
+  }
+  return rc;
+}
+
+
+static int multiCursorEnd(MultiCursor *pCsr, int bLast){
+  int rc = LSM_OK;
+  int i;
+
+  pCsr->flags &= ~(CURSOR_NEXT_OK | CURSOR_PREV_OK | CURSOR_SEEK_EQ);
+  pCsr->flags |= (bLast ? CURSOR_PREV_OK : CURSOR_NEXT_OK);
+  pCsr->iFree = 0;
+
+  /* Position the two in-memory tree cursors */
+  for(i=0; rc==LSM_OK && i<2; i++){
+    if( pCsr->apTreeCsr[i] ){
+      rc = lsmTreeCursorEnd(pCsr->apTreeCsr[i], bLast);
+    }
+  }
+
+  for(i=0; rc==LSM_OK && i<pCsr->nPtr; i++){
+    SegmentPtr *pPtr = &pCsr->aPtr[i];
+    Level *pLvl = pPtr->pLevel;
+    int iRhs;
+    int bHit = 0;
+
+    if( bLast ){
+      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
+        rc = segmentPtrEnd(pCsr, &pPtr[iRhs+1], 1);
+        if( pPtr[iRhs+1].pPg ) bHit = 1;
+      }
+      if( bHit==0 && rc==LSM_OK ){
+        rc = segmentPtrEnd(pCsr, pPtr, 1);
+      }else{
+        segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
+      }
+    }else{
+      int bLhs = (pPtr->pSeg==&pLvl->lhs);
+      assert( pPtr->pSeg==&pLvl->lhs || pPtr->pSeg==&pLvl->aRhs[0] );
+
+      if( bLhs ){
+        rc = segmentPtrEnd(pCsr, pPtr, 0);
+        if( pPtr->pKey ) bHit = 1;
+      }
+      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
+        if( bHit ){
+          segmentPtrReset(&pPtr[iRhs+1], LSM_SEGMENTPTR_FREE_THRESHOLD);
+        }else{
+          rc = sortedRhsFirst(pCsr, pLvl, &pPtr[iRhs+bLhs]);
+        }
+      }
+    }
+    i += pLvl->nRight;
+  }
+
+  /* And the b-tree cursor, if applicable */
+  if( rc==LSM_OK && pCsr->pBtCsr ){
+    assert( bLast==0 );
+    rc = btreeCursorFirst(pCsr->pBtCsr);
+  }
+
+  if( rc==LSM_OK ){
+    rc = multiCursorSetupTree(pCsr, bLast);
+  }
+  
+  return rc;
+}
+
+
+int mcursorSave(MultiCursor *pCsr){
+  int rc = LSM_OK;
+  if( pCsr->aTree ){
+    int iTree = pCsr->aTree[1];
+    if( iTree==CURSOR_DATA_TREE0 || iTree==CURSOR_DATA_TREE1 ){
+      multiCursorCacheKey(pCsr, &rc);
+    }
+  }
+  mcursorFreeComponents(pCsr);
+  return rc;
+}
+
+int mcursorRestore(lsm_db *pDb, MultiCursor *pCsr){
+  int rc;
+  rc = multiCursorInit(pCsr, pDb->pClient);
+  if( rc==LSM_OK && pCsr->key.pData ){
+    rc = lsmMCursorSeek(pCsr, 
+         rtTopic(pCsr->eType), pCsr->key.pData, pCsr->key.nData, +1
+    );
+  }
+  return rc;
+}
+
+int lsmSaveCursors(lsm_db *pDb){
+  int rc = LSM_OK;
+  MultiCursor *pCsr;
+
+  for(pCsr=pDb->pCsr; rc==LSM_OK && pCsr; pCsr=pCsr->pNext){
+    rc = mcursorSave(pCsr);
+  }
+  return rc;
+}
+
+int lsmRestoreCursors(lsm_db *pDb){
+  int rc = LSM_OK;
+  MultiCursor *pCsr;
+
+  for(pCsr=pDb->pCsr; rc==LSM_OK && pCsr; pCsr=pCsr->pNext){
+    rc = mcursorRestore(pDb, pCsr);
+  }
+  return rc;
+}
+
+int lsmMCursorFirst(MultiCursor *pCsr){
+  return multiCursorEnd(pCsr, 0);
+}
+
+int lsmMCursorLast(MultiCursor *pCsr){
+  return multiCursorEnd(pCsr, 1);
+}
+
+lsm_db *lsmMCursorDb(MultiCursor *pCsr){
+  return pCsr->pDb;
+}
+
+void lsmMCursorReset(MultiCursor *pCsr){
+  int i;
+  lsmTreeCursorReset(pCsr->apTreeCsr[0]);
+  lsmTreeCursorReset(pCsr->apTreeCsr[1]);
+  for(i=0; i<pCsr->nPtr; i++){
+    segmentPtrReset(&pCsr->aPtr[i], LSM_SEGMENTPTR_FREE_THRESHOLD);
+  }
+  pCsr->key.nData = 0;
+}
+
+static int treeCursorSeek(
+  MultiCursor *pCsr,
+  TreeCursor *pTreeCsr, 
+  void *pKey, int nKey, 
+  int eSeek,
+  int *pbStop
+){
+  int rc = LSM_OK;
+  if( pTreeCsr ){
+    int res = 0;
+    lsmTreeCursorSeek(pTreeCsr, pKey, nKey, &res);
+    switch( eSeek ){
+      case LSM_SEEK_EQ: {
+        int eType = lsmTreeCursorFlags(pTreeCsr);
+        if( (res<0 && (eType & LSM_START_DELETE))
+         || (res>0 && (eType & LSM_END_DELETE))
+         || (res==0 && (eType & LSM_POINT_DELETE))
+        ){
+          *pbStop = 1;
+        }else if( res==0 && (eType & LSM_INSERT) ){
+          lsm_env *pEnv = pCsr->pDb->pEnv;
+          void *p; int n;         /* Key/value from tree-cursor */
+          *pbStop = 1;
+          pCsr->flags |= CURSOR_SEEK_EQ;
+          rc = lsmTreeCursorKey(pTreeCsr, &pCsr->eType, &p, &n);
+          if( rc==LSM_OK ) rc = sortedBlobSet(pEnv, &pCsr->key, p, n);
+          if( rc==LSM_OK ) rc = lsmTreeCursorValue(pTreeCsr, &p, &n);
+          if( rc==LSM_OK ) rc = sortedBlobSet(pEnv, &pCsr->val, p, n);
+        }
+        lsmTreeCursorReset(pTreeCsr);
+        break;
+      }
+      case LSM_SEEK_GE:
+        if( res<0 && lsmTreeCursorValid(pTreeCsr) ){
+          lsmTreeCursorNext(pTreeCsr);
+        }
+        break;
+      default:
+        if( res>0 ){
+          assert( lsmTreeCursorValid(pTreeCsr) );
+          lsmTreeCursorPrev(pTreeCsr);
+        }
+        break;
+    }
+  }
+  return rc;
+}
+
+
+/*
+** Seek the cursor.
+*/
+int lsmMCursorSeek(
+  MultiCursor *pCsr, 
+  int iTopic, 
+  void *pKey, int nKey, 
+  int eSeek
+){
+  int eESeek = eSeek;             /* Effective eSeek parameter */
+  int bStop = 0;                  /* Set to true to halt search operation */
+  int rc = LSM_OK;                /* Return code */
+  int iPtr = 0;                   /* Used to iterate through pCsr->aPtr[] */
+  LsmPgno iPgno = 0;              /* FC pointer value */
+
+  assert( pCsr->apTreeCsr[0]==0 || iTopic==0 );
+  assert( pCsr->apTreeCsr[1]==0 || iTopic==0 );
+
+  if( eESeek==LSM_SEEK_LEFAST ) eESeek = LSM_SEEK_LE;
+
+  assert( eESeek==LSM_SEEK_EQ || eESeek==LSM_SEEK_LE || eESeek==LSM_SEEK_GE );
+  assert( (pCsr->flags & CURSOR_FLUSH_FREELIST)==0 );
+  assert( pCsr->nPtr==0 || pCsr->aPtr[0].pLevel );
+
+  pCsr->flags &= ~(CURSOR_NEXT_OK | CURSOR_PREV_OK | CURSOR_SEEK_EQ);
+  rc = treeCursorSeek(pCsr, pCsr->apTreeCsr[0], pKey, nKey, eESeek, &bStop);
+  if( rc==LSM_OK && bStop==0 ){
+    rc = treeCursorSeek(pCsr, pCsr->apTreeCsr[1], pKey, nKey, eESeek, &bStop);
+  }
+
+  /* Seek all segment pointers. */
+  for(iPtr=0; iPtr<pCsr->nPtr && rc==LSM_OK && bStop==0; iPtr++){
+    SegmentPtr *pPtr = &pCsr->aPtr[iPtr];
+    assert( pPtr->pSeg==&pPtr->pLevel->lhs );
+    rc = seekInLevel(pCsr, pPtr, eESeek, iTopic, pKey, nKey, &iPgno, &bStop);
+    iPtr += pPtr->pLevel->nRight;
+  }
+
+  if( eSeek!=LSM_SEEK_EQ ){
+    if( rc==LSM_OK ){
+      rc = multiCursorAllocTree(pCsr);
+    }
+    if( rc==LSM_OK ){
+      int i;
+      for(i=pCsr->nTree-1; i>0; i--){
+        multiCursorDoCompare(pCsr, i, eESeek==LSM_SEEK_LE);
+      }
+      if( eSeek==LSM_SEEK_GE ) pCsr->flags |= CURSOR_NEXT_OK;
+      if( eSeek==LSM_SEEK_LE ) pCsr->flags |= CURSOR_PREV_OK;
+    }
+
+    multiCursorCacheKey(pCsr, &rc);
+    if( rc==LSM_OK && eSeek!=LSM_SEEK_LEFAST && 0==mcursorLocationOk(pCsr, 0) ){
+      switch( eESeek ){
+        case LSM_SEEK_EQ:
+          lsmMCursorReset(pCsr);
+          break;
+        case LSM_SEEK_GE:
+          rc = lsmMCursorNext(pCsr);
+          break;
+        default:
+          rc = lsmMCursorPrev(pCsr);
+          break;
+      }
+    }
+  }
+
+  return rc;
+}
+
+int lsmMCursorValid(MultiCursor *pCsr){
+  int res = 0;
+  if( pCsr->flags & CURSOR_SEEK_EQ ){
+    res = 1;
+  }else if( pCsr->aTree ){
+    int iKey = pCsr->aTree[1];
+    if( iKey==CURSOR_DATA_TREE0 || iKey==CURSOR_DATA_TREE1 ){
+      res = lsmTreeCursorValid(pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0]);
+    }else{
+      void *pKey; 
+      multiCursorGetKey(pCsr, iKey, 0, &pKey, 0);
+      res = pKey!=0;
+    }
+  }
+  return res;
+}
+
+static int mcursorAdvanceOk(
+  MultiCursor *pCsr, 
+  int bReverse,
+  int *pRc
+){
+  void *pNew;                     /* Pointer to buffer containing new key */
+  int nNew;                       /* Size of buffer pNew in bytes */
+  int eNewType;                   /* Type of new record */
+
+  if( *pRc ) return 1;
+
+  /* Check the current key value. If it is not greater than (if bReverse==0)
+  ** or less than (if bReverse!=0) the key currently cached in pCsr->key, 
+  ** then the cursor has not yet been successfully advanced.  
+  */
+  multiCursorGetKey(pCsr, pCsr->aTree[1], &eNewType, &pNew, &nNew);
+  if( pNew ){
+    int typemask = (pCsr->flags & CURSOR_IGNORE_DELETE) ? ~(0) : LSM_SYSTEMKEY;
+    int res = sortedDbKeyCompare(pCsr,
+      eNewType & typemask, pNew, nNew, 
+      pCsr->eType & typemask, pCsr->key.pData, pCsr->key.nData
+    );
+
+    if( (bReverse==0 && res<=0) || (bReverse!=0 && res>=0) ){
+      return 0;
+    }
+
+    multiCursorCacheKey(pCsr, pRc);
+    assert( pCsr->eType==eNewType );
+
+    /* If this cursor is configured to skip deleted keys, and the current
+    ** cursor points to a SORTED_DELETE entry, then the cursor has not been 
+    ** successfully advanced.  
+    **
+    ** Similarly, if the cursor is configured to skip system keys and the
+    ** current cursor points to a system key, it has not yet been advanced.
+    */
+    if( *pRc==LSM_OK && 0==mcursorLocationOk(pCsr, 0) ) return 0;
+  }
+  return 1;
+}
+
+static void flCsrAdvance(MultiCursor *pCsr){
+  assert( pCsr->flags & CURSOR_FLUSH_FREELIST );
+  if( pCsr->iFree % 2 ){
+    pCsr->iFree++;
+  }else{
+    int nEntry = pCsr->pDb->pWorker->freelist.nEntry;
+    FreelistEntry *aEntry = pCsr->pDb->pWorker->freelist.aEntry;
+
+    int i = nEntry - 1 - (pCsr->iFree / 2);
+
+    /* If the current entry is a delete and the "end-delete" key will not
+    ** be attached to the next entry, increment iFree by 1 only. */
+    if( aEntry[i].iId<0 ){
+      while( 1 ){
+        if( i==0 || aEntry[i-1].iBlk!=aEntry[i].iBlk-1 ){
+          pCsr->iFree--;
+          break;
+        }
+        if( aEntry[i-1].iId>=0 ) break;
+        pCsr->iFree += 2;
+        i--;
+      }
+    }
+    pCsr->iFree += 2;
+  }
+}
+
+static int multiCursorAdvance(MultiCursor *pCsr, int bReverse){
+  int rc = LSM_OK;                /* Return Code */
+  if( lsmMCursorValid(pCsr) ){
+    do {
+      int iKey = pCsr->aTree[1];
+
+      assertCursorTree(pCsr);
+
+      /* If this multi-cursor is advancing forwards, and the sub-cursor
+      ** being advanced is the one that separator keys may be being read
+      ** from, record the current absolute pointer value.  */
+      if( pCsr->pPrevMergePtr ){
+        if( iKey==(CURSOR_DATA_SEGMENT+pCsr->nPtr) ){
+          assert( pCsr->pBtCsr );
+          *pCsr->pPrevMergePtr = pCsr->pBtCsr->iPtr;
+        }else if( pCsr->pBtCsr==0 && pCsr->nPtr>0
+               && iKey==(CURSOR_DATA_SEGMENT+pCsr->nPtr-1) 
+        ){
+          SegmentPtr *pPtr = &pCsr->aPtr[iKey-CURSOR_DATA_SEGMENT];
+          *pCsr->pPrevMergePtr = pPtr->iPtr+pPtr->iPgPtr;
+        }
+      }
+
+      if( iKey==CURSOR_DATA_TREE0 || iKey==CURSOR_DATA_TREE1 ){
+        TreeCursor *pTreeCsr = pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0];
+        if( bReverse ){
+          rc = lsmTreeCursorPrev(pTreeCsr);
+        }else{
+          rc = lsmTreeCursorNext(pTreeCsr);
+        }
+      }else if( iKey==CURSOR_DATA_SYSTEM ){
+        assert( pCsr->flags & CURSOR_FLUSH_FREELIST );
+        assert( bReverse==0 );
+        flCsrAdvance(pCsr);
+      }else if( iKey==(CURSOR_DATA_SEGMENT+pCsr->nPtr) ){
+        assert( bReverse==0 && pCsr->pBtCsr );
+        rc = btreeCursorNext(pCsr->pBtCsr);
+      }else{
+        rc = segmentCursorAdvance(pCsr, iKey-CURSOR_DATA_SEGMENT, bReverse);
+      }
+      if( rc==LSM_OK ){
+        int i;
+        for(i=(iKey+pCsr->nTree)/2; i>0; i=i/2){
+          multiCursorDoCompare(pCsr, i, bReverse);
+        }
+        assertCursorTree(pCsr);
+      }
+    }while( mcursorAdvanceOk(pCsr, bReverse, &rc)==0 );
+  }
+  return rc;
+}
+
+int lsmMCursorNext(MultiCursor *pCsr){
+  if( (pCsr->flags & CURSOR_NEXT_OK)==0 ) return LSM_MISUSE_BKPT;
+  return multiCursorAdvance(pCsr, 0);
+}
+
+int lsmMCursorPrev(MultiCursor *pCsr){
+  if( (pCsr->flags & CURSOR_PREV_OK)==0 ) return LSM_MISUSE_BKPT;
+  return multiCursorAdvance(pCsr, 1);
+}
+
+int lsmMCursorKey(MultiCursor *pCsr, void **ppKey, int *pnKey){
+  if( (pCsr->flags & CURSOR_SEEK_EQ) || pCsr->aTree==0 ){
+    *pnKey = pCsr->key.nData;
+    *ppKey = pCsr->key.pData;
+  }else{
+    int iKey = pCsr->aTree[1];
+
+    if( iKey==CURSOR_DATA_TREE0 || iKey==CURSOR_DATA_TREE1 ){
+      TreeCursor *pTreeCsr = pCsr->apTreeCsr[iKey-CURSOR_DATA_TREE0];
+      lsmTreeCursorKey(pTreeCsr, 0, ppKey, pnKey);
+    }else{
+      int nKey;
+
+#ifndef NDEBUG
+      void *pKey;
+      int eType;
+      multiCursorGetKey(pCsr, iKey, &eType, &pKey, &nKey);
+      assert( eType==pCsr->eType );
+      assert( nKey==pCsr->key.nData );
+      assert( memcmp(pKey, pCsr->key.pData, nKey)==0 );
+#endif
+
+      nKey = pCsr->key.nData;
+      if( nKey==0 ){
+        *ppKey = 0;
+      }else{
+        *ppKey = pCsr->key.pData;
+      }
+      *pnKey = nKey; 
+    }
+  }
+  return LSM_OK;
+}
+
+/*
+** Compare the current key that cursor csr points to with pKey/nKey. Set
+** *piRes to the result and return LSM_OK.
+*/
+int lsm_csr_cmp(lsm_cursor *csr, const void *pKey, int nKey, int *piRes){
+  MultiCursor *pCsr = (MultiCursor *)csr;
+  void *pCsrkey; int nCsrkey;
+  int rc;
+  rc = lsmMCursorKey(pCsr, &pCsrkey, &nCsrkey);
+  if( rc==LSM_OK ){
+    int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
+    *piRes = sortedKeyCompare(xCmp, 0, pCsrkey, nCsrkey, 0, (void *)pKey, nKey);
+  }
+  return rc;
+}
+
+int lsmMCursorValue(MultiCursor *pCsr, void **ppVal, int *pnVal){
+  void *pVal;
+  int nVal;
+  int rc;
+  if( (pCsr->flags & CURSOR_SEEK_EQ) || pCsr->aTree==0 ){
+    rc = LSM_OK;
+    nVal = pCsr->val.nData;
+    pVal = pCsr->val.pData;
+  }else{
+
+    assert( pCsr->aTree );
+    assert( mcursorLocationOk(pCsr, (pCsr->flags & CURSOR_IGNORE_DELETE)) );
+
+    rc = multiCursorGetVal(pCsr, pCsr->aTree[1], &pVal, &nVal);
+    if( pVal && rc==LSM_OK ){
+      rc = sortedBlobSet(pCsr->pDb->pEnv, &pCsr->val, pVal, nVal);
+      pVal = pCsr->val.pData;
+    }
+
+    if( rc!=LSM_OK ){
+      pVal = 0;
+      nVal = 0;
+    }
+  }
+  *ppVal = pVal;
+  *pnVal = nVal;
+  return rc;
+}
+
+int lsmMCursorType(MultiCursor *pCsr, int *peType){
+  assert( pCsr->aTree );
+  multiCursorGetKey(pCsr, pCsr->aTree[1], peType, 0, 0);
+  return LSM_OK;
+}
+
+/*
+** Buffer aData[], size nData, is assumed to contain a valid b-tree 
+** hierarchy page image. Return the offset in aData[] of the next free
+** byte in the data area (where a new cell may be written if there is
+** space).
+*/
+static int mergeWorkerPageOffset(u8 *aData, int nData){
+  int nRec;
+  int iOff;
+  int nKey;
+  int eType;
+
+  nRec = lsmGetU16(&aData[SEGMENT_NRECORD_OFFSET(nData)]);
+  iOff = lsmGetU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec-1)]);
+  eType = aData[iOff++];
+  assert( eType==0 
+       || eType==(LSM_SYSTEMKEY|LSM_SEPARATOR) 
+       || eType==(LSM_SEPARATOR)
+  );
+
+  iOff += lsmVarintGet32(&aData[iOff], &nKey);
+  iOff += lsmVarintGet32(&aData[iOff], &nKey);
+
+  return iOff + (eType ? nKey : 0);
+}
+
+/*
+** Following a checkpoint operation, database pages that are part of the
+** checkpointed state of the LSM are deemed read-only. This includes the
+** right-most page of the b-tree hierarchy of any separators array under
+** construction, and all pages between it and the b-tree root, inclusive.
+** This is a problem, as when further pages are appended to the separators
+** array, entries must be added to the indicated b-tree hierarchy pages.
+**
+** This function copies all such b-tree pages to new locations, so that
+** they can be modified as required.
+**
+** The complication is that not all database pages are the same size - due
+** to the way the file.c module works some (the first and last in each block)
+** are 4 bytes smaller than the others.
+*/
+static int mergeWorkerMoveHierarchy(
+  MergeWorker *pMW,               /* Merge worker */
+  int bSep                        /* True for separators run */
+){
+  lsm_db *pDb = pMW->pDb;         /* Database handle */
+  int rc = LSM_OK;                /* Return code */
+  int i;
+  Page **apHier = pMW->hier.apHier;
+  int nHier = pMW->hier.nHier;
+
+  for(i=0; rc==LSM_OK && i<nHier; i++){
+    Page *pNew = 0;
+    rc = lsmFsSortedAppend(pDb->pFS, pDb->pWorker, pMW->pLevel, 1, &pNew);
+    assert( rc==LSM_OK );
+
+    if( rc==LSM_OK ){
+      u8 *a1; int n1;
+      u8 *a2; int n2;
+
+      a1 = fsPageData(pNew, &n1);
+      a2 = fsPageData(apHier[i], &n2);
+
+      assert( n1==n2 || n1+4==n2 );
+
+      if( n1==n2 ){
+        memcpy(a1, a2, n2);
+      }else{
+        int nEntry = pageGetNRec(a2, n2);
+        int iEof1 = SEGMENT_EOF(n1, nEntry);
+        int iEof2 = SEGMENT_EOF(n2, nEntry);
+
+        memcpy(a1, a2, iEof2 - 4);
+        memcpy(&a1[iEof1], &a2[iEof2], n2 - iEof2);
+      }
+
+      lsmFsPageRelease(apHier[i]);
+      apHier[i] = pNew;
+
+#if 0
+      assert( n1==n2 || n1+4==n2 || n2+4==n1 );
+      if( n1>=n2 ){
+        /* If n1 (size of the new page) is equal to or greater than n2 (the
+        ** size of the old page), then copy the data into the new page. If
+        ** n1==n2, this could be done with a single memcpy(). However, 
+        ** since sometimes n1>n2, the page content and footer must be copied 
+        ** separately. */
+        int nEntry = pageGetNRec(a2, n2);
+        int iEof1 = SEGMENT_EOF(n1, nEntry);
+        int iEof2 = SEGMENT_EOF(n2, nEntry);
+        memcpy(a1, a2, iEof2);
+        memcpy(&a1[iEof1], &a2[iEof2], n2 - iEof2);
+        lsmFsPageRelease(apHier[i]);
+        apHier[i] = pNew;
+      }else{
+        lsmPutU16(&a1[SEGMENT_FLAGS_OFFSET(n1)], SEGMENT_BTREE_FLAG);
+        lsmPutU16(&a1[SEGMENT_NRECORD_OFFSET(n1)], 0);
+        lsmPutU64(&a1[SEGMENT_POINTER_OFFSET(n1)], 0);
+        i = i - 1;
+        lsmFsPageRelease(pNew);
+      }
+#endif
+    }
+  }
+
+#ifdef LSM_DEBUG
+  if( rc==LSM_OK ){
+    for(i=0; i<nHier; i++) assert( lsmFsPageWritable(apHier[i]) );
+  }
+#endif
+
+  return rc;
+}
+
+/*
+** Allocate and populate the MergeWorker.apHier[] array.
+*/
+static int mergeWorkerLoadHierarchy(MergeWorker *pMW){
+  int rc = LSM_OK;
+  Segment *pSeg;
+  Hierarchy *p;
+ 
+  pSeg = &pMW->pLevel->lhs;
+  p = &pMW->hier;
+
+  if( p->apHier==0 && pSeg->iRoot!=0 ){
+    FileSystem *pFS = pMW->pDb->pFS;
+    lsm_env *pEnv = pMW->pDb->pEnv;
+    Page **apHier = 0;
+    int nHier = 0;
+    int iPg = (int)pSeg->iRoot;
+
+    do {
+      Page *pPg = 0;
+      u8 *aData;
+      int nData;
+      int flags;
+
+      rc = lsmFsDbPageGet(pFS, pSeg, iPg, &pPg);
+      if( rc!=LSM_OK ) break;
+
+      aData = fsPageData(pPg, &nData);
+      flags = pageGetFlags(aData, nData);
+      if( flags&SEGMENT_BTREE_FLAG ){
+        Page **apNew = (Page **)lsmRealloc(
+            pEnv, apHier, sizeof(Page *)*(nHier+1)
+        );
+        if( apNew==0 ){
+          rc = LSM_NOMEM_BKPT;
+          break;
+        }
+        apHier = apNew;
+        memmove(&apHier[1], &apHier[0], sizeof(Page *) * nHier);
+        nHier++;
+
+        apHier[0] = pPg;
+        iPg = (int)pageGetPtr(aData, nData);
+      }else{
+        lsmFsPageRelease(pPg);
+        break;
+      }
+    }while( 1 );
+
+    if( rc==LSM_OK ){
+      u8 *aData;
+      int nData;
+      aData = fsPageData(apHier[0], &nData);
+      pMW->aSave[0].iPgno = pageGetPtr(aData, nData);
+      p->nHier = nHier;
+      p->apHier = apHier;
+      rc = mergeWorkerMoveHierarchy(pMW, 0);
+    }else{
+      int i;
+      for(i=0; i<nHier; i++){
+        lsmFsPageRelease(apHier[i]);
+      }
+      lsmFree(pEnv, apHier);
+    }
+  }
+
+  return rc;
+}
+
+/*
+** B-tree pages use almost the same format as regular pages. The 
+** differences are:
+**
+**   1. The record format is (usually, see below) as follows:
+**
+**         + Type byte (always SORTED_SEPARATOR or SORTED_SYSTEM_SEPARATOR),
+**         + Absolute pointer value (varint),
+**         + Number of bytes in key (varint),
+**         + LsmBlob containing key data.
+**
+**   2. All pointer values are stored as absolute values (not offsets 
+**      relative to the footer pointer value).
+**
+**   3. Each pointer that is part of a record points to a page that 
+**      contains keys smaller than the records key (note: not "equal to or
+**      smaller than - smaller than").
+**
+**   4. The pointer in the page footer of a b-tree page points to a page
+**      that contains keys equal to or larger than the largest key on the
+**      b-tree page.
+**
+** The reason for having the page footer pointer point to the right-child
+** (instead of the left) is that doing things this way makes the 
+** mergeWorkerMoveHierarchy() operation less complicated (since the pointers 
+** that need to be updated are all stored as fixed-size integers within the 
+** page footer, not varints in page records).
+**
+** Records may not span b-tree pages. If this function is called to add a
+** record larger than (page-size / 4) bytes, then a pointer to the indexed
+** array page that contains the main record is added to the b-tree instead.
+** In this case the record format is:
+**
+**         + 0x00 byte (1 byte) 
+**         + Absolute pointer value (varint),
+**         + Absolute page number of page containing key (varint).
+**
+** See function seekInBtree() for the code that traverses b-tree pages.
+*/
+
+static int mergeWorkerBtreeWrite(
+  MergeWorker *pMW,
+  u8 eType,
+  LsmPgno iPtr,
+  LsmPgno iKeyPg,
+  void *pKey,
+  int nKey
+){
+  Hierarchy *p = &pMW->hier;
+  lsm_db *pDb = pMW->pDb;         /* Database handle */
+  int rc = LSM_OK;                /* Return Code */
+  int iLevel;                     /* Level of b-tree hierachy to write to */
+  int nData;                      /* Size of aData[] in bytes */
+  u8 *aData;                      /* Page data for level iLevel */
+  int iOff;                       /* Offset on b-tree page to write record to */
+  int nRec;                       /* Initial number of records on b-tree page */
+
+  /* iKeyPg should be zero for an ordinary b-tree key, or non-zero for an
+  ** indirect key. The flags byte for an indirect key is 0x00.  */
+  assert( (eType==0)==(iKeyPg!=0) );
+
+  /* The MergeWorker.apHier[] array contains the right-most leaf of the b-tree
+  ** hierarchy, the root node, and all nodes that lie on the path between.
+  ** apHier[0] is the right-most leaf and apHier[pMW->nHier-1] is the current
+  ** root page.
+  **
+  ** This loop searches for a node with enough space to store the key on,
+  ** starting with the leaf and iterating up towards the root. When the loop
+  ** exits, the key may be written to apHier[iLevel].  */
+  for(iLevel=0; iLevel<=p->nHier; iLevel++){
+    int nByte;                    /* Number of free bytes required */
+
+    if( iLevel==p->nHier ){
+      /* Extend the array and allocate a new root page. */
+      Page **aNew;
+      aNew = (Page **)lsmRealloc(
+          pMW->pDb->pEnv, p->apHier, sizeof(Page *)*(p->nHier+1)
+      );
+      if( !aNew ){
+        return LSM_NOMEM_BKPT;
+      }
+      p->apHier = aNew;
+    }else{
+      Page *pOld;
+      int nFree;
+
+      /* If the key will fit on this page, break out of the loop here.
+      ** The new entry will be written to page apHier[iLevel]. */
+      pOld = p->apHier[iLevel];
+      assert( lsmFsPageWritable(pOld) );
+      aData = fsPageData(pOld, &nData);
+      if( eType==0 ){
+        nByte = 2 + 1 + lsmVarintLen32((int)iPtr) + lsmVarintLen32((int)iKeyPg);
+      }else{
+        nByte = 2 + 1 + lsmVarintLen32((int)iPtr) + lsmVarintLen32(nKey) + nKey;
+      }
+      nRec = pageGetNRec(aData, nData);
+      nFree = SEGMENT_EOF(nData, nRec) - mergeWorkerPageOffset(aData, nData);
+      if( nByte<=nFree ) break;
+
+      /* Otherwise, this page is full. Set the right-hand-child pointer
+      ** to iPtr and release it.  */
+      lsmPutU64(&aData[SEGMENT_POINTER_OFFSET(nData)], iPtr);
+      assert( lsmFsPageNumber(pOld)==0 );
+      rc = lsmFsPagePersist(pOld);
+      if( rc==LSM_OK ){
+        iPtr = lsmFsPageNumber(pOld);
+        lsmFsPageRelease(pOld);
+      }
+    }
+
+    /* Allocate a new page for apHier[iLevel]. */
+    p->apHier[iLevel] = 0;
+    if( rc==LSM_OK ){
+      rc = lsmFsSortedAppend(
+          pDb->pFS, pDb->pWorker, pMW->pLevel, 1, &p->apHier[iLevel]
+      );
+    }
+    if( rc!=LSM_OK ) return rc;
+
+    aData = fsPageData(p->apHier[iLevel], &nData);
+    memset(aData, 0, nData);
+    lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], SEGMENT_BTREE_FLAG);
+    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], 0);
+
+    if( iLevel==p->nHier ){
+      p->nHier++;
+      break;
+    }
+  }
+
+  /* Write the key into page apHier[iLevel]. */
+  aData = fsPageData(p->apHier[iLevel], &nData);
+  iOff = mergeWorkerPageOffset(aData, nData);
+  nRec = pageGetNRec(aData, nData);
+  lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], (u16)iOff);
+  lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], (u16)(nRec+1));
+  if( eType==0 ){
+    aData[iOff++] = 0x00;
+    iOff += lsmVarintPut32(&aData[iOff], (int)iPtr);
+    iOff += lsmVarintPut32(&aData[iOff], (int)iKeyPg);
+  }else{
+    aData[iOff++] = eType;
+    iOff += lsmVarintPut32(&aData[iOff], (int)iPtr);
+    iOff += lsmVarintPut32(&aData[iOff], nKey);
+    memcpy(&aData[iOff], pKey, nKey);
+  }
+
+  return rc;
+}
+
+static int mergeWorkerBtreeIndirect(MergeWorker *pMW){
+  int rc = LSM_OK;
+  if( pMW->iIndirect ){
+    LsmPgno iKeyPg = pMW->aSave[1].iPgno;
+    rc = mergeWorkerBtreeWrite(pMW, 0, pMW->iIndirect, iKeyPg, 0, 0);
+    pMW->iIndirect = 0;
+  }
+  return rc;
+}
+
+/*
+** Append the database key (iTopic/pKey/nKey) to the b-tree under 
+** construction. This key has not yet been written to a segment page.
+** The pointer that will accompany the new key in the b-tree - that
+** points to the completed segment page that contains keys smaller than
+** (pKey/nKey) is currently stored in pMW->aSave[0].iPgno.
+*/
+static int mergeWorkerPushHierarchy(
+  MergeWorker *pMW,               /* Merge worker object */
+  int iTopic,                     /* Topic value for this key */
+  void *pKey,                     /* Pointer to key buffer */
+  int nKey                        /* Size of pKey buffer in bytes */
+){
+  int rc = LSM_OK;                /* Return Code */
+  LsmPgno iPtr;                   /* Pointer value to accompany pKey/nKey */
+
+  assert( pMW->aSave[0].bStore==0 );
+  assert( pMW->aSave[1].bStore==0 );
+  rc = mergeWorkerBtreeIndirect(pMW);
+
+  /* Obtain the absolute pointer value to store along with the key in the
+  ** page body. This pointer points to a page that contains keys that are
+  ** smaller than pKey/nKey.  */
+  iPtr = pMW->aSave[0].iPgno;
+  assert( iPtr!=0 );
+
+  /* Determine if the indirect format should be used. */
+  if( (nKey*4 > lsmFsPageSize(pMW->pDb->pFS)) ){
+    pMW->iIndirect = iPtr;
+    pMW->aSave[1].bStore = 1;
+  }else{
+    rc = mergeWorkerBtreeWrite(
+        pMW, (u8)(iTopic | LSM_SEPARATOR), iPtr, 0, pKey, nKey
+    );
+  }
+
+  /* Ensure that the SortedRun.iRoot field is correct. */
+  return rc;
+}
+
+static int mergeWorkerFinishHierarchy(
+  MergeWorker *pMW                /* Merge worker object */
+){
+  int i;                          /* Used to loop through apHier[] */
+  int rc = LSM_OK;                /* Return code */
+  LsmPgno iPtr;                   /* New right-hand-child pointer value */
+
+  iPtr = pMW->aSave[0].iPgno;
+  for(i=0; i<pMW->hier.nHier && rc==LSM_OK; i++){
+    Page *pPg = pMW->hier.apHier[i];
+    int nData;                    /* Size of aData[] in bytes */
+    u8 *aData;                    /* Page data for pPg */
+
+    aData = fsPageData(pPg, &nData);
+    lsmPutU64(&aData[SEGMENT_POINTER_OFFSET(nData)], iPtr);
+
+    rc = lsmFsPagePersist(pPg);
+    iPtr = lsmFsPageNumber(pPg);
+    lsmFsPageRelease(pPg);
+  }
+
+  if( pMW->hier.nHier ){
+    pMW->pLevel->lhs.iRoot = iPtr;
+    lsmFree(pMW->pDb->pEnv, pMW->hier.apHier);
+    pMW->hier.apHier = 0;
+    pMW->hier.nHier = 0;
+  }
+
+  return rc;
+}
+
+static int mergeWorkerAddPadding(
+  MergeWorker *pMW                /* Merge worker object */
+){
+  FileSystem *pFS = pMW->pDb->pFS;
+  return lsmFsSortedPadding(pFS, pMW->pDb->pWorker, &pMW->pLevel->lhs);
+}
+
+/*
+** Release all page references currently held by the merge-worker passed
+** as the only argument. Unless an error has occurred, all pages have
+** already been released.
+*/
+static void mergeWorkerReleaseAll(MergeWorker *pMW){
+  int i;
+  lsmFsPageRelease(pMW->pPage);
+  pMW->pPage = 0;
+
+  for(i=0; i<pMW->hier.nHier; i++){
+    lsmFsPageRelease(pMW->hier.apHier[i]);
+    pMW->hier.apHier[i] = 0;
+  }
+  lsmFree(pMW->pDb->pEnv, pMW->hier.apHier);
+  pMW->hier.apHier = 0;
+  pMW->hier.nHier = 0;
+}
+
+static int keyszToSkip(FileSystem *pFS, int nKey){
+  int nPgsz;                /* Nominal database page size */
+  nPgsz = lsmFsPageSize(pFS);
+  return LSM_MIN(((nKey * 4) / nPgsz), 3);
+}
+
+/*
+** Release the reference to the current output page of merge-worker *pMW
+** (reference pMW->pPage). Set the page number values in aSave[] as 
+** required (see comments above struct MergeWorker for details).
+*/
+static int mergeWorkerPersistAndRelease(MergeWorker *pMW){
+  int rc;
+  int i;
+
+  assert( pMW->pPage || (pMW->aSave[0].bStore==0 && pMW->aSave[1].bStore==0) );
+
+  /* Persist the page */
+  rc = lsmFsPagePersist(pMW->pPage);
+
+  /* If required, save the page number. */
+  for(i=0; i<2; i++){
+    if( pMW->aSave[i].bStore ){
+      pMW->aSave[i].iPgno = lsmFsPageNumber(pMW->pPage);
+      pMW->aSave[i].bStore = 0;
+    }
+  }
+
+  /* Release the completed output page. */
+  lsmFsPageRelease(pMW->pPage);
+  pMW->pPage = 0;
+  return rc;
+}
+
+/*
+** Advance to the next page of an output run being populated by merge-worker
+** pMW. The footer of the new page is initialized to indicate that it contains
+** zero records. The flags field is cleared. The page footer pointer field
+** is set to iFPtr.
+**
+** If successful, LSM_OK is returned. Otherwise, an error code.
+*/
+static int mergeWorkerNextPage(
+  MergeWorker *pMW,               /* Merge worker object to append page to */
+  LsmPgno iFPtr                   /* Pointer value for footer of new page */
+){
+  int rc = LSM_OK;                /* Return code */
+  Page *pNext = 0;                /* New page appended to run */
+  lsm_db *pDb = pMW->pDb;         /* Database handle */
+
+  rc = lsmFsSortedAppend(pDb->pFS, pDb->pWorker, pMW->pLevel, 0, &pNext);
+  assert( rc || pMW->pLevel->lhs.iFirst>0 || pMW->pDb->compress.xCompress );
+
+  if( rc==LSM_OK ){
+    u8 *aData;                    /* Data buffer belonging to page pNext */
+    int nData;                    /* Size of aData[] in bytes */
+
+    rc = mergeWorkerPersistAndRelease(pMW);
+
+    pMW->pPage = pNext;
+    pMW->pLevel->pMerge->iOutputOff = 0;
+    aData = fsPageData(pNext, &nData);
+    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], 0);
+    lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], 0);
+    lsmPutU64(&aData[SEGMENT_POINTER_OFFSET(nData)], iFPtr);
+    pMW->nWork++;
+  }
+
+  return rc;
+}
+
+/*
+** Write a blob of data into an output segment being populated by a 
+** merge-worker object. If argument bSep is true, write into the separators
+** array. Otherwise, the main array.
+**
+** This function is used to write the blobs of data for keys and values.
+*/
+static int mergeWorkerData(
+  MergeWorker *pMW,               /* Merge worker object */
+  int bSep,                       /* True to write to separators run */
+  int iFPtr,                      /* Footer ptr for new pages */
+  u8 *aWrite,                     /* Write data from this buffer */
+  int nWrite                      /* Size of aWrite[] in bytes */
+){
+  int rc = LSM_OK;                /* Return code */
+  int nRem = nWrite;              /* Number of bytes still to write */
+
+  while( rc==LSM_OK && nRem>0 ){
+    Merge *pMerge = pMW->pLevel->pMerge;
+    int nCopy;                    /* Number of bytes to copy */
+    u8 *aData;                    /* Pointer to buffer of current output page */
+    int nData;                    /* Size of aData[] in bytes */
+    int nRec;                     /* Number of records on current output page */
+    int iOff;                     /* Offset in aData[] to write to */
+
+    assert( lsmFsPageWritable(pMW->pPage) );
+   
+    aData = fsPageData(pMW->pPage, &nData);
+    nRec = pageGetNRec(aData, nData);
+    iOff = pMerge->iOutputOff;
+    nCopy = LSM_MIN(nRem, SEGMENT_EOF(nData, nRec) - iOff);
+
+    memcpy(&aData[iOff], &aWrite[nWrite-nRem], nCopy);
+    nRem -= nCopy;
+
+    if( nRem>0 ){
+      rc = mergeWorkerNextPage(pMW, iFPtr);
+    }else{
+      pMerge->iOutputOff = iOff + nCopy;
+    }
+  }
+
+  return rc;
+}
+
+
+/*
+** The MergeWorker passed as the only argument is working to merge two or
+** more existing segments together (not to flush an in-memory tree). It
+** has not yet written the first key to the first page of the output.
+*/
+static int mergeWorkerFirstPage(MergeWorker *pMW){
+  int rc = LSM_OK;                /* Return code */
+  Page *pPg = 0;                  /* First page of run pSeg */
+  int iFPtr = 0;                  /* Pointer value read from footer of pPg */
+  MultiCursor *pCsr = pMW->pCsr;
+
+  assert( pMW->pPage==0 );
+
+  if( pCsr->pBtCsr ){
+    rc = LSM_OK;
+    iFPtr = (int)pMW->pLevel->pNext->lhs.iFirst;
+  }else if( pCsr->nPtr>0 ){
+    Segment *pSeg;
+    pSeg = pCsr->aPtr[pCsr->nPtr-1].pSeg;
+    rc = lsmFsDbPageGet(pMW->pDb->pFS, pSeg, pSeg->iFirst, &pPg);
+    if( rc==LSM_OK ){
+      u8 *aData;                    /* Buffer for page pPg */
+      int nData;                    /* Size of aData[] in bytes */
+      aData = fsPageData(pPg, &nData);
+      iFPtr = (int)pageGetPtr(aData, nData);
+      lsmFsPageRelease(pPg);
+    }
+  }
+
+  if( rc==LSM_OK ){
+    rc = mergeWorkerNextPage(pMW, iFPtr);
+    if( pCsr->pPrevMergePtr ) *pCsr->pPrevMergePtr = iFPtr;
+    pMW->aSave[0].bStore = 1;
+  }
+
+  return rc;
+}
+
+static int mergeWorkerWrite(
+  MergeWorker *pMW,               /* Merge worker object to write into */
+  int eType,                      /* One of SORTED_SEPARATOR, WRITE or DELETE */
+  void *pKey, int nKey,           /* Key value */
+  void *pVal, int nVal,           /* Value value */
+  int iPtr                        /* Absolute value of page pointer, or 0 */
+){
+  int rc = LSM_OK;                /* Return code */
+  Merge *pMerge;                  /* Persistent part of level merge state */
+  int nHdr;                       /* Space required for this record header */
+  Page *pPg;                      /* Page to write to */
+  u8 *aData;                      /* Data buffer for page pWriter->pPage */
+  int nData = 0;                  /* Size of buffer aData[] in bytes */
+  int nRec = 0;                   /* Number of records on page pPg */
+  int iFPtr = 0;                  /* Value of pointer in footer of pPg */
+  int iRPtr = 0;                  /* Value of pointer written into record */
+  int iOff = 0;                   /* Current write offset within page pPg */
+  Segment *pSeg;                  /* Segment being written */
+  int flags = 0;                  /* If != 0, flags value for page footer */
+  int bFirst = 0;                 /* True for first key of output run */
+
+  pMerge = pMW->pLevel->pMerge;    
+  pSeg = &pMW->pLevel->lhs;
+
+  if( pSeg->iFirst==0 && pMW->pPage==0 ){
+    rc = mergeWorkerFirstPage(pMW);
+    bFirst = 1;
+  }
+  pPg = pMW->pPage;
+  if( pPg ){
+    aData = fsPageData(pPg, &nData);
+    nRec = pageGetNRec(aData, nData);
+    iFPtr = (int)pageGetPtr(aData, nData);
+    iRPtr = iPtr - iFPtr;
+  }
+     
+  /* Figure out how much space is required by the new record. The space
+  ** required is divided into two sections: the header and the body. The
+  ** header consists of the intial varint fields. The body are the blobs 
+  ** of data that correspond to the key and value data. The entire header 
+  ** must be stored on the page. The body may overflow onto the next and
+  ** subsequent pages.
+  **
+  ** The header space is:
+  **
+  **     1) record type - 1 byte.
+  **     2) Page-pointer-offset - 1 varint
+  **     3) Key size - 1 varint
+  **     4) Value size - 1 varint (only if LSM_INSERT flag is set)
+  */
+  if( rc==LSM_OK ){
+    nHdr = 1 + lsmVarintLen32(iRPtr) + lsmVarintLen32(nKey);
+    if( rtIsWrite(eType) ) nHdr += lsmVarintLen32(nVal);
+
+    /* If the entire header will not fit on page pPg, or if page pPg is 
+    ** marked read-only, advance to the next page of the output run. */
+    iOff = pMerge->iOutputOff;
+    if( iOff<0 || pPg==0 || iOff+nHdr > SEGMENT_EOF(nData, nRec+1) ){
+      if( iOff>=0 && pPg ){
+        /* Zero any free space on the page */
+        assert( aData );
+        memset(&aData[iOff], 0, SEGMENT_EOF(nData, nRec)-iOff);
+      }
+      iFPtr = (int)*pMW->pCsr->pPrevMergePtr;
+      iRPtr = iPtr - iFPtr;
+      iOff = 0;
+      nRec = 0;
+      rc = mergeWorkerNextPage(pMW, iFPtr);
+      pPg = pMW->pPage;
+    }
+  }
+
+  /* If this record header will be the first on the page, and the page is 
+  ** not the very first in the entire run, add a copy of the key to the
+  ** b-tree hierarchy.
+  */
+  if( rc==LSM_OK && nRec==0 && bFirst==0 ){
+    assert( pMerge->nSkip>=0 );
+
+    if( pMerge->nSkip==0 ){
+      rc = mergeWorkerPushHierarchy(pMW, rtTopic(eType), pKey, nKey);
+      assert( pMW->aSave[0].bStore==0 );
+      pMW->aSave[0].bStore = 1;
+      pMerge->nSkip = keyszToSkip(pMW->pDb->pFS, nKey);
+    }else{
+      pMerge->nSkip--;
+      flags = PGFTR_SKIP_THIS_FLAG;
+    }
+
+    if( pMerge->nSkip ) flags |= PGFTR_SKIP_NEXT_FLAG;
+  }
+
+  /* Update the output segment */
+  if( rc==LSM_OK ){
+    aData = fsPageData(pPg, &nData);
+
+    /* Update the page footer. */
+    lsmPutU16(&aData[SEGMENT_NRECORD_OFFSET(nData)], (u16)(nRec+1));
+    lsmPutU16(&aData[SEGMENT_CELLPTR_OFFSET(nData, nRec)], (u16)iOff);
+    if( flags ) lsmPutU16(&aData[SEGMENT_FLAGS_OFFSET(nData)], (u16)flags);
+
+    /* Write the entry header into the current page. */
+    aData[iOff++] = (u8)eType;                                           /* 1 */
+    iOff += lsmVarintPut32(&aData[iOff], iRPtr);                         /* 2 */
+    iOff += lsmVarintPut32(&aData[iOff], nKey);                          /* 3 */
+    if( rtIsWrite(eType) ) iOff += lsmVarintPut32(&aData[iOff], nVal);   /* 4 */
+    pMerge->iOutputOff = iOff;
+
+    /* Write the key and data into the segment. */
+    assert( iFPtr==pageGetPtr(aData, nData) );
+    rc = mergeWorkerData(pMW, 0, iFPtr+iRPtr, pKey, nKey);
+    if( rc==LSM_OK && rtIsWrite(eType) ){
+      if( rc==LSM_OK ){
+        rc = mergeWorkerData(pMW, 0, iFPtr+iRPtr, pVal, nVal);
+      }
+    }
+  }
+
+  return rc;
+}
+
+
+/*
+** Free all resources allocated by mergeWorkerInit().
+*/
+static void mergeWorkerShutdown(MergeWorker *pMW, int *pRc){
+  int i;                          /* Iterator variable */
+  int rc = *pRc;
+  MultiCursor *pCsr = pMW->pCsr;
+
+  /* Unless the merge has finished, save the cursor position in the
+  ** Merge.aInput[] array. See function mergeWorkerInit() for the 
+  ** code to restore a cursor position based on aInput[].  */
+  if( rc==LSM_OK && pCsr ){
+    Merge *pMerge = pMW->pLevel->pMerge;
+    if( lsmMCursorValid(pCsr) ){
+      int bBtree = (pCsr->pBtCsr!=0);
+      int iPtr;
+
+      /* pMerge->nInput==0 indicates that this is a FlushTree() operation. */
+      assert( pMerge->nInput==0 || pMW->pLevel->nRight>0 );
+      assert( pMerge->nInput==0 || pMerge->nInput==(pCsr->nPtr+bBtree) );
+
+      for(i=0; i<(pMerge->nInput-bBtree); i++){
+        SegmentPtr *pPtr = &pCsr->aPtr[i];
+        if( pPtr->pPg ){
+          pMerge->aInput[i].iPg = lsmFsPageNumber(pPtr->pPg);
+          pMerge->aInput[i].iCell = pPtr->iCell;
+        }else{
+          pMerge->aInput[i].iPg = 0;
+          pMerge->aInput[i].iCell = 0;
+        }
+      }
+      if( bBtree && pMerge->nInput ){
+        assert( i==pCsr->nPtr );
+        btreeCursorPosition(pCsr->pBtCsr, &pMerge->aInput[i]);
+      }
+
+      /* Store the location of the split-key */
+      iPtr = pCsr->aTree[1] - CURSOR_DATA_SEGMENT;
+      if( iPtr<pCsr->nPtr ){
+        pMerge->splitkey = pMerge->aInput[iPtr];
+      }else{
+        btreeCursorSplitkey(pCsr->pBtCsr, &pMerge->splitkey);
+      }
+    }
+
+    /* Zero any free space left on the final page. This helps with
+    ** compression if using a compression hook. And prevents valgrind
+    ** from complaining about uninitialized byte passed to write(). */
+    if( pMW->pPage ){
+      int nData;
+      u8 *aData = fsPageData(pMW->pPage, &nData);
+      int iOff = pMerge->iOutputOff;
+      int iEof = SEGMENT_EOF(nData, pageGetNRec(aData, nData));
+      memset(&aData[iOff], 0, iEof - iOff);
+    }
+    
+    pMerge->iOutputOff = -1;
+  }
+
+  lsmMCursorClose(pCsr, 0);
+
+  /* Persist and release the output page. */
+  if( rc==LSM_OK ) rc = mergeWorkerPersistAndRelease(pMW);
+  if( rc==LSM_OK ) rc = mergeWorkerBtreeIndirect(pMW);
+  if( rc==LSM_OK ) rc = mergeWorkerFinishHierarchy(pMW);
+  if( rc==LSM_OK ) rc = mergeWorkerAddPadding(pMW);
+  lsmFsFlushWaiting(pMW->pDb->pFS, &rc);
+  mergeWorkerReleaseAll(pMW);
+
+  lsmFree(pMW->pDb->pEnv, pMW->aGobble);
+  pMW->aGobble = 0;
+  pMW->pCsr = 0;
+
+  *pRc = rc;
+}
+
+/*
+** The cursor passed as the first argument is being used as the input for
+** a merge operation. When this function is called, *piFlags contains the
+** database entry flags for the current entry. The entry about to be written
+** to the output.
+**
+** Note that this function only has to work for cursors configured to 
+** iterate forwards (not backwards).
+*/
+static void mergeRangeDeletes(MultiCursor *pCsr, int *piVal, int *piFlags){
+  int f = *piFlags;
+  int iKey = pCsr->aTree[1];
+  int i;
+
+  assert( pCsr->flags & CURSOR_NEXT_OK );
+  if( pCsr->flags & CURSOR_IGNORE_DELETE ){
+    /* The ignore-delete flag is set when the output of the merge will form
+    ** the oldest level in the database. In this case there is no point in
+    ** retaining any range-delete flags.  */
+    assert( (f & LSM_POINT_DELETE)==0 );
+    f &= ~(LSM_START_DELETE|LSM_END_DELETE);
+  }else{
+    for(i=0; i<(CURSOR_DATA_SEGMENT + pCsr->nPtr); i++){
+      if( i!=iKey ){
+        int eType;
+        void *pKey;
+        int nKey;
+        int res;
+        multiCursorGetKey(pCsr, i, &eType, &pKey, &nKey);
+
+        if( pKey ){
+          res = sortedKeyCompare(pCsr->pDb->xCmp, 
+              rtTopic(pCsr->eType), pCsr->key.pData, pCsr->key.nData,
+              rtTopic(eType), pKey, nKey
+          );
+          assert( res<=0 );
+          if( res==0 ){
+            if( (f & (LSM_INSERT|LSM_POINT_DELETE))==0 ){
+              if( eType & LSM_INSERT ){
+                f |= LSM_INSERT;
+                *piVal = i;
+              }
+              else if( eType & LSM_POINT_DELETE ){
+                f |= LSM_POINT_DELETE;
+              }
+            }
+            f |= (eType & (LSM_END_DELETE|LSM_START_DELETE));
+          }
+
+          if( i>iKey && (eType & LSM_END_DELETE) && res<0 ){
+            if( f & (LSM_INSERT|LSM_POINT_DELETE) ){
+              f |= (LSM_END_DELETE|LSM_START_DELETE);
+            }else{
+              f = 0;
+            }
+            break;
+          }
+        }
+      }
+    }
+
+    assert( (f & LSM_INSERT)==0 || (f & LSM_POINT_DELETE)==0 );
+    if( (f & LSM_START_DELETE) 
+     && (f & LSM_END_DELETE) 
+     && (f & LSM_POINT_DELETE )
+    ){
+      f = 0;
+    }
+  }
+
+  *piFlags = f;
+}
+
+static int mergeWorkerStep(MergeWorker *pMW){
+  lsm_db *pDb = pMW->pDb;       /* Database handle */
+  MultiCursor *pCsr;            /* Cursor to read input data from */
+  int rc = LSM_OK;              /* Return code */
+  int eType;                    /* SORTED_SEPARATOR, WRITE or DELETE */
+  void *pKey; int nKey;         /* Key */
+  LsmPgno iPtr;
+  int iVal;
+
+  pCsr = pMW->pCsr;
+
+  /* Pull the next record out of the source cursor. */
+  lsmMCursorKey(pCsr, &pKey, &nKey);
+  eType = pCsr->eType;
+
+  /* Figure out if the output record may have a different pointer value
+  ** than the previous. This is the case if the current key is identical to
+  ** a key that appears in the lowest level run being merged. If so, set 
+  ** iPtr to the absolute pointer value. If not, leave iPtr set to zero, 
+  ** indicating that the output pointer value should be a copy of the pointer 
+  ** value written with the previous key.  */
+  iPtr = (pCsr->pPrevMergePtr ? *pCsr->pPrevMergePtr : 0);
+  if( pCsr->pBtCsr ){
+    BtreeCursor *pBtCsr = pCsr->pBtCsr;
+    if( pBtCsr->pKey ){
+      int res = rtTopic(pBtCsr->eType) - rtTopic(eType);
+      if( res==0 ) res = pDb->xCmp(pBtCsr->pKey, pBtCsr->nKey, pKey, nKey);
+      if( 0==res ) iPtr = pBtCsr->iPtr;
+      assert( res>=0 );
+    }
+  }else if( pCsr->nPtr ){
+    SegmentPtr *pPtr = &pCsr->aPtr[pCsr->nPtr-1];
+    if( pPtr->pPg
+     && 0==pDb->xCmp(pPtr->pKey, pPtr->nKey, pKey, nKey)
+    ){
+      iPtr = pPtr->iPtr+pPtr->iPgPtr;
+    }
+  }
+
+  iVal = pCsr->aTree[1];
+  mergeRangeDeletes(pCsr, &iVal, &eType);
+
+  if( eType!=0 ){
+    if( pMW->aGobble ){
+      int iGobble = pCsr->aTree[1] - CURSOR_DATA_SEGMENT;
+      if( iGobble<pCsr->nPtr && iGobble>=0 ){
+        SegmentPtr *pGobble = &pCsr->aPtr[iGobble];
+        if( (pGobble->flags & PGFTR_SKIP_THIS_FLAG)==0 ){
+          pMW->aGobble[iGobble] = lsmFsPageNumber(pGobble->pPg);
+        }
+      }
+    }
+
+    /* If this is a separator key and we know that the output pointer has not
+    ** changed, there is no point in writing an output record. Otherwise,
+    ** proceed. */
+    if( rc==LSM_OK && (rtIsSeparator(eType)==0 || iPtr!=0) ){
+      /* Write the record into the main run. */
+      void *pVal; int nVal;
+      rc = multiCursorGetVal(pCsr, iVal, &pVal, &nVal);
+      if( pVal && rc==LSM_OK ){
+        assert( nVal>=0 );
+        rc = sortedBlobSet(pDb->pEnv, &pCsr->val, pVal, nVal);
+        pVal = pCsr->val.pData;
+      }
+      if( rc==LSM_OK ){
+        rc = mergeWorkerWrite(pMW, eType, pKey, nKey, pVal, nVal, (int)iPtr);
+      }
+    }
+  }
+
+  /* Advance the cursor to the next input record (assuming one exists). */
+  assert( lsmMCursorValid(pMW->pCsr) );
+  if( rc==LSM_OK ) rc = lsmMCursorNext(pMW->pCsr);
+
+  return rc;
+}
+
+static int mergeWorkerDone(MergeWorker *pMW){
+  return pMW->pCsr==0 || !lsmMCursorValid(pMW->pCsr);
+}
+
+static void sortedFreeLevel(lsm_env *pEnv, Level *p){
+  if( p ){
+    lsmFree(pEnv, p->pSplitKey);
+    lsmFree(pEnv, p->pMerge);
+    lsmFree(pEnv, p->aRhs);
+    lsmFree(pEnv, p);
+  }
+}
+
+static void sortedInvokeWorkHook(lsm_db *pDb){
+  if( pDb->xWork ){
+    pDb->xWork(pDb, pDb->pWorkCtx);
+  }
+}
+
+static int sortedNewToplevel(
+  lsm_db *pDb,                    /* Connection handle */
+  int eTree,                      /* One of the TREE_XXX constants */
+  int *pnWrite                    /* OUT: Number of database pages written */
+){
+  int rc = LSM_OK;                /* Return Code */
+  MultiCursor *pCsr = 0;
+  Level *pNext = 0;               /* The current top level */
+  Level *pNew;                    /* The new level itself */
+  Segment *pLinked = 0;           /* Delete separators from this segment */
+  Level *pDel = 0;                /* Delete this entire level */
+  int nWrite = 0;                 /* Number of database pages written */
+  Freelist freelist;
+
+  if( eTree!=TREE_NONE ){
+    rc = lsmShmCacheChunks(pDb, pDb->treehdr.nChunk);
+  }
+
+  assert( pDb->bUseFreelist==0 );
+  pDb->pFreelist = &freelist;
+  pDb->bUseFreelist = 1;
+  memset(&freelist, 0, sizeof(freelist));
+
+  /* Allocate the new level structure to write to. */
+  pNext = lsmDbSnapshotLevel(pDb->pWorker);
+  pNew = (Level *)lsmMallocZeroRc(pDb->pEnv, sizeof(Level), &rc);
+  if( pNew ){
+    pNew->pNext = pNext;
+    lsmDbSnapshotSetLevel(pDb->pWorker, pNew);
+  }
+
+  /* Create a cursor to gather the data required by the new segment. The new
+  ** segment contains everything in the tree and pointers to the next segment
+  ** in the database (if any).  */
+  pCsr = multiCursorNew(pDb, &rc);
+  if( pCsr ){
+    pCsr->pDb = pDb;
+    rc = multiCursorVisitFreelist(pCsr);
+    if( rc==LSM_OK ){
+      rc = multiCursorAddTree(pCsr, pDb->pWorker, eTree);
+    }
+    if( rc==LSM_OK && pNext && pNext->pMerge==0 ){
+      if( (pNext->flags & LEVEL_FREELIST_ONLY) ){
+        pDel = pNext;
+        pCsr->aPtr = lsmMallocZeroRc(pDb->pEnv, sizeof(SegmentPtr), &rc);
+        multiCursorAddOne(pCsr, pNext, &rc);
+      }else if( eTree!=TREE_NONE && pNext->lhs.iRoot ){
+        pLinked = &pNext->lhs;
+        rc = btreeCursorNew(pDb, pLinked, &pCsr->pBtCsr);
+      }
+    }
+
+    /* If this will be the only segment in the database, discard any delete
+    ** markers present in the in-memory tree.  */
+    if( pNext==0 ){
+      multiCursorIgnoreDelete(pCsr);
+    }
+  }
+
+  if( rc!=LSM_OK ){
+    lsmMCursorClose(pCsr, 0);
+  }else{
+    LsmPgno iLeftPtr = 0;
+    Merge merge;                  /* Merge object used to create new level */
+    MergeWorker mergeworker;      /* MergeWorker object for the same purpose */
+
+    memset(&merge, 0, sizeof(Merge));
+    memset(&mergeworker, 0, sizeof(MergeWorker));
+
+    pNew->pMerge = &merge;
+    pNew->flags |= LEVEL_INCOMPLETE;
+    mergeworker.pDb = pDb;
+    mergeworker.pLevel = pNew;
+    mergeworker.pCsr = pCsr;
+    pCsr->pPrevMergePtr = &iLeftPtr;
+
+    /* Mark the separators array for the new level as a "phantom". */
+    mergeworker.bFlush = 1;
+
+    /* Do the work to create the new merged segment on disk */
+    if( rc==LSM_OK ) rc = lsmMCursorFirst(pCsr);
+    while( rc==LSM_OK && mergeWorkerDone(&mergeworker)==0 ){
+      rc = mergeWorkerStep(&mergeworker);
+    }
+    mergeWorkerShutdown(&mergeworker, &rc);
+    assert( rc!=LSM_OK || mergeworker.nWork==0 || pNew->lhs.iFirst );
+    if( rc==LSM_OK && pNew->lhs.iFirst ){
+      rc = lsmFsSortedFinish(pDb->pFS, &pNew->lhs);
+    }
+    nWrite = mergeworker.nWork;
+    pNew->flags &= ~LEVEL_INCOMPLETE;
+    if( eTree==TREE_NONE ){
+      pNew->flags |= LEVEL_FREELIST_ONLY;
+    }
+    pNew->pMerge = 0;
+  }
+
+  if( rc!=LSM_OK || pNew->lhs.iFirst==0 ){
+    assert( rc!=LSM_OK || pDb->pWorker->freelist.nEntry==0 );
+    lsmDbSnapshotSetLevel(pDb->pWorker, pNext);
+    sortedFreeLevel(pDb->pEnv, pNew);
+  }else{
+    if( pLinked ){
+      pLinked->iRoot = 0;
+    }else if( pDel ){
+      assert( pNew->pNext==pDel );
+      pNew->pNext = pDel->pNext;
+      lsmFsSortedDelete(pDb->pFS, pDb->pWorker, 1, &pDel->lhs);
+      sortedFreeLevel(pDb->pEnv, pDel);
+    }
+
+#if LSM_LOG_STRUCTURE
+    lsmSortedDumpStructure(pDb, pDb->pWorker, LSM_LOG_DATA, 0, "new-toplevel");
+#endif
+
+    if( freelist.nEntry ){
+      Freelist *p = &pDb->pWorker->freelist;
+      lsmFree(pDb->pEnv, p->aEntry);
+      memcpy(p, &freelist, sizeof(freelist));
+      freelist.aEntry = 0;
+    }else{
+      pDb->pWorker->freelist.nEntry = 0;
+    }
+
+    assertBtreeOk(pDb, &pNew->lhs);
+    sortedInvokeWorkHook(pDb);
+  }
+
+  if( pnWrite ) *pnWrite = nWrite;
+  pDb->pWorker->nWrite += nWrite;
+  pDb->pFreelist = 0;
+  pDb->bUseFreelist = 0;
+  lsmFree(pDb->pEnv, freelist.aEntry);
+  return rc;
+}
+
+/*
+** The nMerge levels in the LSM beginning with pLevel consist of a
+** left-hand-side segment only. Replace these levels with a single new
+** level consisting of a new empty segment on the left-hand-side and the
+** nMerge segments from the replaced levels on the right-hand-side.
+**
+** Also, allocate and populate a Merge object and set Level.pMerge to
+** point to it.
+*/
+static int sortedMergeSetup(
+  lsm_db *pDb,                    /* Database handle */
+  Level *pLevel,                  /* First level to merge */
+  int nMerge,                     /* Merge this many levels together */
+  Level **ppNew                   /* New, merged, level */
+){
+  int rc = LSM_OK;                /* Return Code */
+  Level *pNew;                    /* New Level object */
+  int bUseNext = 0;               /* True to link in next separators */
+  Merge *pMerge;                  /* New Merge object */
+  int nByte;                      /* Bytes of space allocated at pMerge */
+
+#ifdef LSM_DEBUG
+  int iLevel;
+  Level *pX = pLevel;
+  for(iLevel=0; iLevel<nMerge; iLevel++){
+    assert( pX->nRight==0 );
+    pX = pX->pNext;
+  }
+#endif
+
+  /* Allocate the new Level object */
+  pNew = (Level *)lsmMallocZeroRc(pDb->pEnv, sizeof(Level), &rc);
+  if( pNew ){
+    pNew->aRhs = (Segment *)lsmMallocZeroRc(pDb->pEnv, 
+                                        nMerge * sizeof(Segment), &rc);
+  }
+
+  /* Populate the new Level object */
+  if( rc==LSM_OK ){
+    Level *pNext = 0;             /* Level following pNew */
+    int i;
+    int bFreeOnly = 1;
+    Level *pTopLevel;
+    Level *p = pLevel;
+    Level **pp;
+    pNew->nRight = nMerge;
+    pNew->iAge = pLevel->iAge+1;
+    for(i=0; i<nMerge; i++){
+      assert( p->nRight==0 );
+      pNext = p->pNext;
+      pNew->aRhs[i] = p->lhs;
+      if( (p->flags & LEVEL_FREELIST_ONLY)==0 ) bFreeOnly = 0;
+      sortedFreeLevel(pDb->pEnv, p);
+      p = pNext;
+    }
+
+    if( bFreeOnly ) pNew->flags |= LEVEL_FREELIST_ONLY;
+
+    /* Replace the old levels with the new. */
+    pTopLevel = lsmDbSnapshotLevel(pDb->pWorker);
+    pNew->pNext = p;
+    for(pp=&pTopLevel; *pp!=pLevel; pp=&((*pp)->pNext));
+    *pp = pNew;
+    lsmDbSnapshotSetLevel(pDb->pWorker, pTopLevel);
+
+    /* Determine whether or not the next separators will be linked in */
+    if( pNext && pNext->pMerge==0 && pNext->lhs.iRoot && pNext 
+     && (bFreeOnly==0 || (pNext->flags & LEVEL_FREELIST_ONLY))
+    ){
+      bUseNext = 1;
+    }
+  }
+
+  /* Allocate the merge object */
+  nByte = sizeof(Merge) + sizeof(MergeInput) * (nMerge + bUseNext);
+  pMerge = (Merge *)lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
+  if( pMerge ){
+    pMerge->aInput = (MergeInput *)&pMerge[1];
+    pMerge->nInput = nMerge + bUseNext;
+    pNew->pMerge = pMerge;
+  }
+
+  *ppNew = pNew;
+  return rc;
+}
+
+static int mergeWorkerInit(
+  lsm_db *pDb,                    /* Db connection to do merge work */
+  Level *pLevel,                  /* Level to work on merging */
+  MergeWorker *pMW                /* Object to initialize */
+){
+  int rc = LSM_OK;                /* Return code */
+  Merge *pMerge = pLevel->pMerge; /* Persistent part of merge state */
+  MultiCursor *pCsr = 0;          /* Cursor opened for pMW */
+  Level *pNext = pLevel->pNext;   /* Next level in LSM */
+
+  assert( pDb->pWorker );
+  assert( pLevel->pMerge );
+  assert( pLevel->nRight>0 );
+
+  memset(pMW, 0, sizeof(MergeWorker));
+  pMW->pDb = pDb;
+  pMW->pLevel = pLevel;
+  pMW->aGobble = lsmMallocZeroRc(pDb->pEnv, sizeof(LsmPgno)*pLevel->nRight,&rc);
+
+  /* Create a multi-cursor to read the data to write to the new
+  ** segment. The new segment contains:
+  **
+  **   1. Records from LHS of each of the nMerge levels being merged.
+  **   2. Separators from either the last level being merged, or the
+  **      separators attached to the LHS of the following level, or neither.
+  **
+  ** If the new level is the lowest (oldest) in the db, discard any
+  ** delete keys. Key annihilation.
+  */
+  pCsr = multiCursorNew(pDb, &rc);
+  if( pCsr ){
+    pCsr->flags |= CURSOR_NEXT_OK;
+    rc = multiCursorAddRhs(pCsr, pLevel);
+  }
+  if( rc==LSM_OK && pMerge->nInput > pLevel->nRight ){
+    rc = btreeCursorNew(pDb, &pNext->lhs, &pCsr->pBtCsr);
+  }else if( pNext ){
+    multiCursorReadSeparators(pCsr);
+  }else{
+    multiCursorIgnoreDelete(pCsr);
+  }
+
+  assert( rc!=LSM_OK || pMerge->nInput==(pCsr->nPtr+(pCsr->pBtCsr!=0)) );
+  pMW->pCsr = pCsr;
+
+  /* Load the b-tree hierarchy into memory. */
+  if( rc==LSM_OK ) rc = mergeWorkerLoadHierarchy(pMW);
+  if( rc==LSM_OK && pMW->hier.nHier==0 ){
+    pMW->aSave[0].iPgno = pLevel->lhs.iFirst;
+  }
+
+  /* Position the cursor. */
+  if( rc==LSM_OK ){
+    pCsr->pPrevMergePtr = &pMerge->iCurrentPtr;
+    if( pLevel->lhs.iFirst==0 ){
+      /* The output array is still empty. So position the cursor at the very 
+      ** start of the input.  */
+      rc = multiCursorEnd(pCsr, 0);
+    }else{
+      /* The output array is non-empty. Position the cursor based on the
+      ** page/cell data saved in the Merge.aInput[] array.  */
+      int i;
+      for(i=0; rc==LSM_OK && i<pCsr->nPtr; i++){
+        MergeInput *pInput = &pMerge->aInput[i];
+        if( pInput->iPg ){
+          SegmentPtr *pPtr;
+          assert( pCsr->aPtr[i].pPg==0 );
+          pPtr = &pCsr->aPtr[i];
+          rc = segmentPtrLoadPage(pDb->pFS, pPtr, (int)pInput->iPg);
+          if( rc==LSM_OK && pPtr->nCell>0 ){
+            rc = segmentPtrLoadCell(pPtr, pInput->iCell);
+          }
+        }
+      }
+
+      if( rc==LSM_OK && pCsr->pBtCsr ){
+        int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
+        assert( i==pCsr->nPtr );
+        rc = btreeCursorRestore(pCsr->pBtCsr, xCmp, &pMerge->aInput[i]);
+      }
+
+      if( rc==LSM_OK ){
+        rc = multiCursorSetupTree(pCsr, 0);
+      }
+    }
+    pCsr->flags |= CURSOR_NEXT_OK;
+  }
+
+  return rc;
+}
+
+static int sortedBtreeGobble(
+  lsm_db *pDb,                    /* Worker connection */
+  MultiCursor *pCsr,              /* Multi-cursor being used for a merge */
+  int iGobble                     /* pCsr->aPtr[] entry to operate on */
+){
+  int rc = LSM_OK;
+  if( rtTopic(pCsr->eType)==0 ){
+    Segment *pSeg = pCsr->aPtr[iGobble].pSeg;
+    LsmPgno *aPg;
+    int nPg;
+
+    /* Seek from the root of the b-tree to the segment leaf that may contain
+    ** a key equal to the one multi-cursor currently points to. Record the
+    ** page number of each b-tree page and the leaf. The segment may be
+    ** gobbled up to (but not including) the first of these page numbers.
+    */
+    assert( pSeg->iRoot>0 );
+    aPg = lsmMallocZeroRc(pDb->pEnv, sizeof(LsmPgno)*32, &rc);
+    if( rc==LSM_OK ){
+      rc = seekInBtree(pCsr, pSeg, 
+          rtTopic(pCsr->eType), pCsr->key.pData, pCsr->key.nData, aPg, 0
+      ); 
+    }
+
+    if( rc==LSM_OK ){
+      for(nPg=0; aPg[nPg]; nPg++);
+      lsmFsGobble(pDb, pSeg, aPg, nPg);
+    }
+
+    lsmFree(pDb->pEnv, aPg);
+  }
+  return rc;
+}
+
+/*
+** Argument p points to a level of age N. Return the number of levels in
+** the linked list starting at p that have age=N (always at least 1).
+*/
+static int sortedCountLevels(Level *p){
+  int iAge = p->iAge;
+  int nRet = 0;
+  do {
+    nRet++;
+    p = p->pNext;
+  }while( p && p->iAge==iAge );
+  return nRet;
+}
+
+static int sortedSelectLevel(lsm_db *pDb, int nMerge, Level **ppOut){
+  Level *pTopLevel = lsmDbSnapshotLevel(pDb->pWorker);
+  int rc = LSM_OK;
+  Level *pLevel = 0;            /* Output value */
+  Level *pBest = 0;             /* Best level to work on found so far */
+  int nBest;                    /* Number of segments merged at pBest */
+  Level *pThis = 0;             /* First in run of levels with age=iAge */
+  int nThis = 0;                /* Number of levels starting at pThis */
+
+  assert( nMerge>=1 );
+  nBest = LSM_MAX(1, nMerge-1);
+
+  /* Find the longest contiguous run of levels not currently undergoing a 
+  ** merge with the same age in the structure. Or the level being merged
+  ** with the largest number of right-hand segments. Work on it. */
+  for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
+    if( pLevel->nRight==0 && pThis && pLevel->iAge==pThis->iAge ){
+      nThis++;
+    }else{
+      if( nThis>nBest ){
+        if( (pLevel->iAge!=pThis->iAge+1)
+         || (pLevel->nRight==0 && sortedCountLevels(pLevel)<=pDb->nMerge)
+        ){
+          pBest = pThis;
+          nBest = nThis;
+        }
+      }
+      if( pLevel->nRight ){
+        if( pLevel->nRight>nBest ){
+          nBest = pLevel->nRight;
+          pBest = pLevel;
+        }
+        nThis = 0;
+        pThis = 0;
+      }else{
+        pThis = pLevel;
+        nThis = 1;
+      }
+    }
+  }
+  if( nThis>nBest ){
+    assert( pThis );
+    pBest = pThis;
+    nBest = nThis;
+  }
+
+  if( pBest==0 && nMerge==1 ){
+    int nFree = 0;
+    int nUsr = 0;
+    for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
+      assert( !pLevel->nRight );
+      if( pLevel->flags & LEVEL_FREELIST_ONLY ){
+        nFree++;
+      }else{
+        nUsr++;
+      }
+    }
+    if( nUsr>1 ){
+      pBest = pTopLevel;
+      nBest = nFree + nUsr;
+    }
+  }
+
+  if( pBest ){
+    if( pBest->nRight==0 ){
+      rc = sortedMergeSetup(pDb, pBest, nBest, ppOut);
+    }else{
+      *ppOut = pBest;
+    }
+  }
+
+  return rc;
+}
+
+static int sortedDbIsFull(lsm_db *pDb){
+  Level *pTop = lsmDbSnapshotLevel(pDb->pWorker);
+
+  if( lsmDatabaseFull(pDb) ) return 1;
+  if( pTop && pTop->iAge==0
+   && (pTop->nRight || sortedCountLevels(pTop)>=pDb->nMerge)
+  ){
+    return 1;
+  }
+  return 0;
+}
+
+typedef struct MoveBlockCtx MoveBlockCtx;
+struct MoveBlockCtx {
+  int iSeen;                      /* Previous free block on list */
+  int iFrom;                      /* Total number of blocks in file */
+};
+
+static int moveBlockCb(void *pCtx, int iBlk, i64 iSnapshot){
+  MoveBlockCtx *p = (MoveBlockCtx *)pCtx;
+  assert( p->iFrom==0 );
+  if( iBlk==(p->iSeen-1) ){
+    p->iSeen = iBlk;
+    return 0;
+  }
+  p->iFrom = p->iSeen-1;
+  return 1;
+}
+
+/*
+** This function is called to further compact a database for which all 
+** of the content has already been merged into a single segment. If 
+** possible, it moves the contents of a single block from the end of the
+** file to a free-block that lies closer to the start of the file (allowing
+** the file to be eventually truncated).
+*/
+static int sortedMoveBlock(lsm_db *pDb, int *pnWrite){
+  Snapshot *p = pDb->pWorker;
+  Level *pLvl = lsmDbSnapshotLevel(p);
+  int iFrom;                      /* Block to move */
+  int iTo;                        /* Destination to move block to */
+  int rc;                         /* Return code */
+
+  MoveBlockCtx sCtx;
+
+  assert( pLvl->pNext==0 && pLvl->nRight==0 );
+  assert( p->redirect.n<=LSM_MAX_BLOCK_REDIRECTS );
+
+  *pnWrite = 0;
+
+  /* Check that the redirect array is not already full. If it is, return
+  ** without moving any database content.  */
+  if( p->redirect.n>=LSM_MAX_BLOCK_REDIRECTS ) return LSM_OK;
+
+  /* Find the last block of content in the database file. Do this by 
+  ** traversing the free-list in reverse (descending block number) order.
+  ** The first block not on the free list is the one that will be moved.
+  ** Since the db consists of a single segment, there is no ambiguity as
+  ** to which segment the block belongs to.  */
+  sCtx.iSeen = p->nBlock+1;
+  sCtx.iFrom = 0;
+  rc = lsmWalkFreelist(pDb, 1, moveBlockCb, &sCtx);
+  if( rc!=LSM_OK || sCtx.iFrom==0 ) return rc;
+  iFrom = sCtx.iFrom;
+
+  /* Find the first free block in the database, ignoring block 1. Block
+  ** 1 is tricky as it is smaller than the other blocks.  */
+  rc = lsmBlockAllocate(pDb, iFrom, &iTo);
+  if( rc!=LSM_OK || iTo==0 ) return rc;
+  assert( iTo!=1 && iTo<iFrom );
+
+  rc = lsmFsMoveBlock(pDb->pFS, &pLvl->lhs, iTo, iFrom);
+  if( rc==LSM_OK ){
+    if( p->redirect.a==0 ){
+      int nByte = sizeof(struct RedirectEntry) * LSM_MAX_BLOCK_REDIRECTS;
+      p->redirect.a = lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
+    }
+    if( rc==LSM_OK ){
+
+      /* Check if the block just moved was already redirected. */
+      int i;
+      for(i=0; i<p->redirect.n; i++){
+        if( p->redirect.a[i].iTo==iFrom ) break;
+      }
+
+      if( i==p->redirect.n ){
+        /* Block iFrom was not already redirected. Add a new array entry. */
+        memmove(&p->redirect.a[1], &p->redirect.a[0], 
+            sizeof(struct RedirectEntry) * p->redirect.n
+            );
+        p->redirect.a[0].iFrom = iFrom;
+        p->redirect.a[0].iTo = iTo;
+        p->redirect.n++;
+      }else{
+        /* Block iFrom was already redirected. Overwrite existing entry. */
+        p->redirect.a[i].iTo = iTo;
+      }
+
+      rc = lsmBlockFree(pDb, iFrom);
+
+      *pnWrite = lsmFsBlockSize(pDb->pFS) / lsmFsPageSize(pDb->pFS);
+      pLvl->lhs.pRedirect = &p->redirect;
+    }
+  }
+
+#if LSM_LOG_STRUCTURE
+  if( rc==LSM_OK ){
+    char aBuf[64];
+    sprintf(aBuf, "move-block %d/%d", p->redirect.n-1, LSM_MAX_BLOCK_REDIRECTS);
+    lsmSortedDumpStructure(pDb, pDb->pWorker, LSM_LOG_DATA, 0, aBuf);
+  }
+#endif
+  return rc;
+}
+
+/*
+*/
+static int mergeInsertFreelistSegments(
+  lsm_db *pDb, 
+  int nFree,
+  MergeWorker *pMW
+){
+  int rc = LSM_OK;
+  if( nFree>0 ){
+    MultiCursor *pCsr = pMW->pCsr;
+    Level *pLvl = pMW->pLevel;
+    SegmentPtr *aNew1;
+    Segment *aNew2;
+
+    Level *pIter;
+    Level *pNext;
+    int i = 0;
+
+    aNew1 = (SegmentPtr *)lsmMallocZeroRc(
+        pDb->pEnv, sizeof(SegmentPtr) * (pCsr->nPtr+nFree), &rc
+    );
+    if( rc ) return rc;
+    memcpy(&aNew1[nFree], pCsr->aPtr, sizeof(SegmentPtr)*pCsr->nPtr);
+    pCsr->nPtr += nFree;
+    lsmFree(pDb->pEnv, pCsr->aTree);
+    lsmFree(pDb->pEnv, pCsr->aPtr);
+    pCsr->aTree = 0;
+    pCsr->aPtr = aNew1;
+
+    aNew2 = (Segment *)lsmMallocZeroRc(
+        pDb->pEnv, sizeof(Segment) * (pLvl->nRight+nFree), &rc
+    );
+    if( rc ) return rc;
+    memcpy(&aNew2[nFree], pLvl->aRhs, sizeof(Segment)*pLvl->nRight);
+    pLvl->nRight += nFree;
+    lsmFree(pDb->pEnv, pLvl->aRhs);
+    pLvl->aRhs = aNew2;
+
+    for(pIter=pDb->pWorker->pLevel; rc==LSM_OK && pIter!=pLvl; pIter=pNext){
+      Segment *pSeg = &pLvl->aRhs[i];
+      memcpy(pSeg, &pIter->lhs, sizeof(Segment));
+
+      pCsr->aPtr[i].pSeg = pSeg;
+      pCsr->aPtr[i].pLevel = pLvl;
+      rc = segmentPtrEnd(pCsr, &pCsr->aPtr[i], 0);
+
+      pDb->pWorker->pLevel = pNext = pIter->pNext;
+      sortedFreeLevel(pDb->pEnv, pIter);
+      i++;
+    }
+    assert( i==nFree );
+    assert( rc!=LSM_OK || pDb->pWorker->pLevel==pLvl );
+
+    for(i=nFree; i<pCsr->nPtr; i++){
+      pCsr->aPtr[i].pSeg = &pLvl->aRhs[i];
+    }
+
+    lsmFree(pDb->pEnv, pMW->aGobble);
+    pMW->aGobble = 0;
+  }
+  return rc;
+}
+
+static int sortedWork(
+  lsm_db *pDb,                    /* Database handle. Must be worker. */
+  int nWork,                      /* Number of pages of work to do */
+  int nMerge,                     /* Try to merge this many levels at once */
+  int bFlush,                     /* Set if call is to make room for a flush */
+  int *pnWrite                    /* OUT: Actual number of pages written */
+){
+  int rc = LSM_OK;                /* Return Code */
+  int nRemaining = nWork;         /* Units of work to do before returning */
+  Snapshot *pWorker = pDb->pWorker;
+
+  assert( pWorker );
+  if( lsmDbSnapshotLevel(pWorker)==0 ) return LSM_OK;
+
+  while( nRemaining>0 ){
+    Level *pLevel = 0;
+
+    /* Find a level to work on. */
+    rc = sortedSelectLevel(pDb, nMerge, &pLevel);
+    assert( rc==LSM_OK || pLevel==0 );
+
+    if( pLevel==0 ){
+      int nDone = 0;
+      Level *pTopLevel = lsmDbSnapshotLevel(pDb->pWorker);
+      if( bFlush==0 && nMerge==1 && pTopLevel && pTopLevel->pNext==0 ){
+        rc = sortedMoveBlock(pDb, &nDone);
+      }
+      nRemaining -= nDone;
+
+      /* Could not find any work to do. Finished. */
+      if( nDone==0 ) break;
+    }else{
+      int bSave = 0;
+      Freelist freelist = {0, 0, 0};
+      MergeWorker mergeworker;    /* State used to work on the level merge */
+
+      assert( pDb->bIncrMerge==0 );
+      assert( pDb->pFreelist==0 && pDb->bUseFreelist==0 );
+
+      pDb->bIncrMerge = 1;
+      rc = mergeWorkerInit(pDb, pLevel, &mergeworker);
+      assert( mergeworker.nWork==0 );
+      
+      while( rc==LSM_OK 
+          && 0==mergeWorkerDone(&mergeworker) 
+          && (mergeworker.nWork<nRemaining || pDb->bUseFreelist)
+      ){
+        int eType = rtTopic(mergeworker.pCsr->eType);
+        rc = mergeWorkerStep(&mergeworker);
+
+        /* If the cursor now points at the first entry past the end of the
+        ** user data (i.e. either to EOF or to the first free-list entry
+        ** that will be added to the run), then check if it is possible to
+        ** merge in any free-list entries that are either in-memory or in
+        ** free-list-only blocks.  */
+        if( rc==LSM_OK && nMerge==1 && eType==0
+         && (rtTopic(mergeworker.pCsr->eType) || mergeWorkerDone(&mergeworker))
+        ){
+          int nFree = 0;          /* Number of free-list-only levels to merge */
+          Level *pLvl;
+          assert( pDb->pFreelist==0 && pDb->bUseFreelist==0 );
+
+          /* Now check if all levels containing data newer than this one
+          ** are single-segment free-list only levels. If so, they will be
+          ** merged in now.  */
+          for(pLvl=pDb->pWorker->pLevel; 
+              pLvl!=mergeworker.pLevel && (pLvl->flags & LEVEL_FREELIST_ONLY); 
+              pLvl=pLvl->pNext
+          ){
+            assert( pLvl->nRight==0 );
+            nFree++;
+          }
+          if( pLvl==mergeworker.pLevel ){
+
+            rc = mergeInsertFreelistSegments(pDb, nFree, &mergeworker);
+            if( rc==LSM_OK ){
+              rc = multiCursorVisitFreelist(mergeworker.pCsr);
+            }
+            if( rc==LSM_OK ){
+              rc = multiCursorSetupTree(mergeworker.pCsr, 0);
+              pDb->pFreelist = &freelist;
+              pDb->bUseFreelist = 1;
+            }
+          }
+        }
+      }
+      nRemaining -= LSM_MAX(mergeworker.nWork, 1);
+
+      if( rc==LSM_OK ){
+        /* Check if the merge operation is completely finished. If not,
+        ** gobble up (declare eligible for recycling) any pages from rhs
+        ** segments for which the content has been completely merged into 
+        ** the lhs of the level.  */
+        if( mergeWorkerDone(&mergeworker)==0 ){
+          int i;
+          for(i=0; i<pLevel->nRight; i++){
+            SegmentPtr *pGobble = &mergeworker.pCsr->aPtr[i];
+            if( pGobble->pSeg->iRoot ){
+              rc = sortedBtreeGobble(pDb, mergeworker.pCsr, i);
+            }else if( mergeworker.aGobble[i] ){
+              lsmFsGobble(pDb, pGobble->pSeg, &mergeworker.aGobble[i], 1);
+            }
+          }
+        }else{
+          int i;
+          int bEmpty;
+          mergeWorkerShutdown(&mergeworker, &rc);
+          bEmpty = (pLevel->lhs.iFirst==0);
+
+          if( bEmpty==0 && rc==LSM_OK ){
+            rc = lsmFsSortedFinish(pDb->pFS, &pLevel->lhs);
+          }
+
+          if( pDb->bUseFreelist ){
+            Freelist *p = &pDb->pWorker->freelist;
+            lsmFree(pDb->pEnv, p->aEntry);
+            memcpy(p, &freelist, sizeof(freelist));
+            pDb->bUseFreelist = 0;
+            pDb->pFreelist = 0;
+            bSave = 1;
+          }
+
+          for(i=0; i<pLevel->nRight; i++){
+            lsmFsSortedDelete(pDb->pFS, pWorker, 1, &pLevel->aRhs[i]);
+          }
+
+          if( bEmpty ){
+            /* If the new level is completely empty, remove it from the 
+            ** database snapshot. This can only happen if all input keys were
+            ** annihilated. Since keys are only annihilated if the new level
+            ** is the last in the linked list (contains the most ancient of
+            ** database content), this guarantees that pLevel->pNext==0.  */ 
+            Level *pTop;          /* Top level of worker snapshot */
+            Level **pp;           /* Read/write iterator for Level.pNext list */
+
+            assert( pLevel->pNext==0 );
+
+            /* Remove the level from the worker snapshot. */
+            pTop = lsmDbSnapshotLevel(pWorker);
+            for(pp=&pTop; *pp!=pLevel; pp=&((*pp)->pNext));
+            *pp = pLevel->pNext;
+            lsmDbSnapshotSetLevel(pWorker, pTop);
+
+            /* Free the Level structure. */
+            sortedFreeLevel(pDb->pEnv, pLevel);
+          }else{
+
+            /* Free the separators of the next level, if required. */
+            if( pLevel->pMerge->nInput > pLevel->nRight ){
+              assert( pLevel->pNext->lhs.iRoot );
+              pLevel->pNext->lhs.iRoot = 0;
+            }
+
+            /* Zero the right-hand-side of pLevel */
+            lsmFree(pDb->pEnv, pLevel->aRhs);
+            pLevel->nRight = 0;
+            pLevel->aRhs = 0;
+
+            /* Free the Merge object */
+            lsmFree(pDb->pEnv, pLevel->pMerge);
+            pLevel->pMerge = 0;
+          }
+
+          if( bSave && rc==LSM_OK ){
+            pDb->bIncrMerge = 0;
+            rc = lsmSaveWorker(pDb, 0);
+          }
+        }
+      }
+
+      /* Clean up the MergeWorker object initialized above. If no error
+      ** has occurred, invoke the work-hook to inform the application that
+      ** the database structure has changed. */
+      mergeWorkerShutdown(&mergeworker, &rc);
+      pDb->bIncrMerge = 0;
+      if( rc==LSM_OK ) sortedInvokeWorkHook(pDb);
+
+#if LSM_LOG_STRUCTURE
+      lsmSortedDumpStructure(pDb, pDb->pWorker, LSM_LOG_DATA, 0, "work");
+#endif
+      assertBtreeOk(pDb, &pLevel->lhs);
+      assertRunInOrder(pDb, &pLevel->lhs);
+
+      /* If bFlush is true and the database is no longer considered "full",
+      ** break out of the loop even if nRemaining is still greater than
+      ** zero. The caller has an in-memory tree to flush to disk.  */
+      if( bFlush && sortedDbIsFull(pDb)==0 ) break;
+    }
+  }
+
+  if( pnWrite ) *pnWrite = (nWork - nRemaining);
+  pWorker->nWrite += (nWork - nRemaining);
+
+#ifdef LSM_LOG_WORK
+  lsmLogMessage(pDb, rc, "sortedWork(): %d pages", (nWork-nRemaining));
+#endif
+  return rc;
+}
+
+/*
+** The database connection passed as the first argument must be a worker
+** connection. This function checks if there exists an "old" in-memory tree
+** ready to be flushed to disk. If so, true is returned. Otherwise false.
+**
+** If an error occurs, *pRc is set to an LSM error code before returning.
+** It is assumed that *pRc is set to LSM_OK when this function is called.
+*/
+static int sortedTreeHasOld(lsm_db *pDb, int *pRc){
+  int rc = LSM_OK;
+  int bRet = 0;
+
+  assert( pDb->pWorker );
+  if( *pRc==LSM_OK ){
+    if( rc==LSM_OK 
+        && pDb->treehdr.iOldShmid
+        && pDb->treehdr.iOldLog!=pDb->pWorker->iLogOff 
+      ){
+      bRet = 1;
+    }else{
+      bRet = 0;
+    }
+    *pRc = rc;
+  }
+  assert( *pRc==LSM_OK || bRet==0 );
+  return bRet;
+}
+
+/*
+** Create a new free-list only top-level segment. Return LSM_OK if successful
+** or an LSM error code if some error occurs.
+*/
+static int sortedNewFreelistOnly(lsm_db *pDb){
+  return sortedNewToplevel(pDb, TREE_NONE, 0);
+}
+
+int lsmSaveWorker(lsm_db *pDb, int bFlush){
+  Snapshot *p = pDb->pWorker;
+  if( p->freelist.nEntry>pDb->nMaxFreelist ){
+    int rc = sortedNewFreelistOnly(pDb);
+    if( rc!=LSM_OK ) return rc;
+  }
+  return lsmCheckpointSaveWorker(pDb, bFlush);
+}
+
+static int doLsmSingleWork(
+  lsm_db *pDb, 
+  int bShutdown,
+  int nMerge,                     /* Minimum segments to merge together */
+  int nPage,                      /* Number of pages to write to disk */
+  int *pnWrite,                   /* OUT: Pages actually written to disk */
+  int *pbCkpt                     /* OUT: True if an auto-checkpoint is req. */
+){
+  Snapshot *pWorker;              /* Worker snapshot */
+  int rc = LSM_OK;                /* Return code */
+  int bDirty = 0;
+  int nMax = nPage;               /* Maximum pages to write to disk */
+  int nRem = nPage;
+  int bCkpt = 0;
+
+  assert( nPage>0 );
+
+  /* Open the worker 'transaction'. It will be closed before this function
+  ** returns.  */
+  assert( pDb->pWorker==0 );
+  rc = lsmBeginWork(pDb);
+  if( rc!=LSM_OK ) return rc;
+  pWorker = pDb->pWorker;
+
+  /* If this connection is doing auto-checkpoints, set nMax (and nRem) so
+  ** that this call stops writing when the auto-checkpoint is due. The
+  ** caller will do the checkpoint, then possibly call this function again. */
+  if( bShutdown==0 && pDb->nAutockpt ){
+    u32 nSync;
+    u32 nUnsync;
+    int nPgsz;
+
+    lsmCheckpointSynced(pDb, 0, 0, &nSync);
+    nUnsync = lsmCheckpointNWrite(pDb->pShmhdr->aSnap1, 0);
+    nPgsz = lsmCheckpointPgsz(pDb->pShmhdr->aSnap1);
+
+    nMax = (int)LSM_MIN(nMax, (pDb->nAutockpt/nPgsz) - (int)(nUnsync-nSync));
+    if( nMax<nRem ){
+      bCkpt = 1;
+      nRem = LSM_MAX(nMax, 0);
+    }
+  }
+
+  /* If there exists in-memory data ready to be flushed to disk, attempt
+  ** to flush it now.  */
+  if( pDb->nTransOpen==0 ){
+    rc = lsmTreeLoadHeader(pDb, 0);
+  }
+  if( sortedTreeHasOld(pDb, &rc) ){
+    /* sortedDbIsFull() returns non-zero if either (a) there are too many
+    ** levels in total in the db, or (b) there are too many levels with the
+    ** the same age in the db. Either way, call sortedWork() to merge 
+    ** existing segments together until this condition is cleared.  */
+    if( sortedDbIsFull(pDb) ){
+      int nPg = 0;
+      rc = sortedWork(pDb, nRem, nMerge, 1, &nPg);
+      nRem -= nPg;
+      assert( rc!=LSM_OK || nRem<=0 || !sortedDbIsFull(pDb) );
+      bDirty = 1;
+    }
+
+    if( rc==LSM_OK && nRem>0 ){
+      int nPg = 0;
+      rc = sortedNewToplevel(pDb, TREE_OLD, &nPg);
+      nRem -= nPg;
+      if( rc==LSM_OK ){
+        if( pDb->nTransOpen>0 ){
+          lsmTreeDiscardOld(pDb);
+        }
+        rc = lsmSaveWorker(pDb, 1);
+        bDirty = 0;
+      }
+    }
+  }
+
+  /* If nPage is still greater than zero, do some merging. */
+  if( rc==LSM_OK && nRem>0 && bShutdown==0 ){
+    int nPg = 0;
+    rc = sortedWork(pDb, nRem, nMerge, 0, &nPg);
+    nRem -= nPg;
+    if( nPg ) bDirty = 1;
+  }
+
+  /* If the in-memory part of the free-list is too large, write a new 
+  ** top-level containing just the in-memory free-list entries to disk. */
+  if( rc==LSM_OK && pDb->pWorker->freelist.nEntry > pDb->nMaxFreelist ){
+    while( rc==LSM_OK && lsmDatabaseFull(pDb) ){
+      int nPg = 0;
+      rc = sortedWork(pDb, 16, nMerge, 1, &nPg);
+      nRem -= nPg;
+    }
+    if( rc==LSM_OK ){
+      rc = sortedNewFreelistOnly(pDb);
+    }
+    bDirty = 1;
+  }
+
+  if( rc==LSM_OK ){
+    *pnWrite = (nMax - nRem);
+    *pbCkpt = (bCkpt && nRem<=0);
+    if( nMerge==1 && pDb->nAutockpt>0 && *pnWrite>0
+     && pWorker->pLevel 
+     && pWorker->pLevel->nRight==0 
+     && pWorker->pLevel->pNext==0 
+    ){
+      *pbCkpt = 1;
+    }
+  }
+
+  if( rc==LSM_OK && bDirty ){
+    lsmFinishWork(pDb, 0, &rc);
+  }else{
+    int rcdummy = LSM_BUSY;
+    lsmFinishWork(pDb, 0, &rcdummy);
+    *pnWrite = 0;
+  }
+  assert( pDb->pWorker==0 );
+  return rc;
+}
+
+static int doLsmWork(lsm_db *pDb, int nMerge, int nPage, int *pnWrite){
+  int rc = LSM_OK;                /* Return code */
+  int nWrite = 0;                 /* Number of pages written */
+
+  assert( nMerge>=1 );
+
+  if( nPage!=0 ){
+    int bCkpt = 0;
+    do {
+      int nThis = 0;
+      int nReq = (nPage>=0) ? (nPage-nWrite) : ((int)0x7FFFFFFF);
+
+      bCkpt = 0;
+      rc = doLsmSingleWork(pDb, 0, nMerge, nReq, &nThis, &bCkpt);
+      nWrite += nThis;
+      if( rc==LSM_OK && bCkpt ){
+        rc = lsm_checkpoint(pDb, 0);
+      }
+    }while( rc==LSM_OK && bCkpt && (nWrite<nPage || nPage<0) );
+  }
+
+  if( pnWrite ){
+    if( rc==LSM_OK ){
+      *pnWrite = nWrite;
+    }else{
+      *pnWrite = 0;
+    }
+  }
+  return rc;
+}
+
+/*
+** Perform work to merge database segments together.
+*/
+int lsm_work(lsm_db *pDb, int nMerge, int nKB, int *pnWrite){
+  int rc;                         /* Return code */
+  int nPgsz;                      /* Nominal page size in bytes */
+  int nPage;                      /* Equivalent of nKB in pages */
+  int nWrite = 0;                 /* Number of pages written */
+
+  /* This function may not be called if pDb has an open read or write
+  ** transaction. Return LSM_MISUSE if an application attempts this.  */
+  if( pDb->nTransOpen || pDb->pCsr ) return LSM_MISUSE_BKPT;
+  if( nMerge<=0 ) nMerge = pDb->nMerge;
+
+  lsmFsPurgeCache(pDb->pFS);
+
+  /* Convert from KB to pages */
+  nPgsz = lsmFsPageSize(pDb->pFS);
+  if( nKB>=0 ){
+    nPage = ((i64)nKB * 1024 + nPgsz - 1) / nPgsz;
+  }else{
+    nPage = -1;
+  }
+
+  rc = doLsmWork(pDb, nMerge, nPage, &nWrite);
+  
+  if( pnWrite ){
+    /* Convert back from pages to KB */
+    *pnWrite = (int)(((i64)nWrite * 1024 + nPgsz - 1) / nPgsz);
+  }
+  return rc;
+}
+
+int lsm_flush(lsm_db *db){
+  int rc;
+
+  if( db->nTransOpen>0 || db->pCsr ){
+    rc = LSM_MISUSE_BKPT;
+  }else{
+    rc = lsmBeginWriteTrans(db);
+    if( rc==LSM_OK ){
+      lsmFlushTreeToDisk(db);
+      lsmTreeDiscardOld(db);
+      lsmTreeMakeOld(db);
+      lsmTreeDiscardOld(db);
+    }
+
+    if( rc==LSM_OK ){
+      rc = lsmFinishWriteTrans(db, 1);
+    }else{
+      lsmFinishWriteTrans(db, 0);
+    }
+    lsmFinishReadTrans(db);
+  }
+
+  return rc;
+}
+
+/*
+** This function is called in auto-work mode to perform merging work on
+** the data structure. It performs enough merging work to prevent the
+** height of the tree from growing indefinitely assuming that roughly
+** nUnit database pages worth of data have been written to the database
+** (i.e. the in-memory tree) since the last call.
+*/
+int lsmSortedAutoWork(
+  lsm_db *pDb,                    /* Database handle */
+  int nUnit                       /* Pages of data written to in-memory tree */
+){
+  int rc = LSM_OK;                /* Return code */
+  int nDepth = 0;                 /* Current height of tree (longest path) */
+  Level *pLevel;                  /* Used to iterate through levels */
+  int bRestore = 0;
+
+  assert( pDb->pWorker==0 );
+  assert( pDb->nTransOpen>0 );
+
+  /* Determine how many units of work to do before returning. One unit of
+  ** work is achieved by writing one page (~4KB) of merged data.  */
+  for(pLevel=lsmDbSnapshotLevel(pDb->pClient); pLevel; pLevel=pLevel->pNext){
+    /* nDepth += LSM_MAX(1, pLevel->nRight); */
+    nDepth += 1;
+  }
+  if( lsmTreeHasOld(pDb) ){
+    nDepth += 1;
+    bRestore = 1;
+    rc = lsmSaveCursors(pDb);
+    if( rc!=LSM_OK ) return rc;
+  }
+
+  if( nDepth>0 ){
+    int nRemaining;               /* Units of work to do before returning */
+
+    nRemaining = nUnit * nDepth;
+#ifdef LSM_LOG_WORK
+    lsmLogMessage(pDb, rc, "lsmSortedAutoWork(): %d*%d = %d pages", 
+        nUnit, nDepth, nRemaining);
+#endif
+    assert( nRemaining>=0 );
+    rc = doLsmWork(pDb, pDb->nMerge, nRemaining, 0);
+    if( rc==LSM_BUSY ) rc = LSM_OK;
+
+    if( bRestore && pDb->pCsr ){
+      lsmMCursorFreeCache(pDb);
+      lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
+      pDb->pClient = 0;
+      if( rc==LSM_OK ){
+        rc = lsmCheckpointLoad(pDb, 0);
+      }
+      if( rc==LSM_OK ){
+        rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot, &pDb->pClient);
+      }
+      if( rc==LSM_OK ){
+        rc = lsmRestoreCursors(pDb);
+      }
+    }
+  }
+
+  return rc;
+}
+
+/*
+** This function is only called during system shutdown. The contents of
+** any in-memory trees present (old or current) are written out to disk.
+*/
+int lsmFlushTreeToDisk(lsm_db *pDb){
+  int rc;
+
+  rc = lsmBeginWork(pDb);
+  while( rc==LSM_OK && sortedDbIsFull(pDb) ){
+    rc = sortedWork(pDb, 256, pDb->nMerge, 1, 0);
+  }
+
+  if( rc==LSM_OK ){
+    rc = sortedNewToplevel(pDb, TREE_BOTH, 0);
+  }
+
+  lsmFinishWork(pDb, 1, &rc);
+  return rc;
+}
+
+/*
+** Return a string representation of the segment passed as the only argument.
+** Space for the returned string is allocated using lsmMalloc(), and should
+** be freed by the caller using lsmFree().
+*/
+static char *segToString(lsm_env *pEnv, Segment *pSeg, int nMin){
+  int nSize = pSeg->nSize;
+  LsmPgno iRoot = pSeg->iRoot;
+  LsmPgno iFirst = pSeg->iFirst;
+  LsmPgno iLast = pSeg->iLastPg;
+  char *z;
+
+  char *z1;
+  char *z2;
+  int nPad;
+
+  z1 = lsmMallocPrintf(pEnv, "%d.%d", iFirst, iLast);
+  if( iRoot ){
+    z2 = lsmMallocPrintf(pEnv, "root=%d", iRoot);
+  }else{
+    z2 = lsmMallocPrintf(pEnv, "size=%d", nSize);
+  }
+
+  nPad = nMin - 2 - strlen(z1) - 1 - strlen(z2);
+  nPad = LSM_MAX(0, nPad);
+
+  if( iRoot ){
+    z = lsmMallocPrintf(pEnv, "/%s %*s%s\\", z1, nPad, "", z2);
+  }else{
+    z = lsmMallocPrintf(pEnv, "|%s %*s%s|", z1, nPad, "", z2);
+  }
+  lsmFree(pEnv, z1);
+  lsmFree(pEnv, z2);
+
+  return z;
+}
+
+static int fileToString(
+  lsm_db *pDb,                    /* For xMalloc() */
+  char *aBuf, 
+  int nBuf, 
+  int nMin,
+  Segment *pSeg
+){
+  int i = 0;
+  if( pSeg ){
+    char *zSeg;
+
+    zSeg = segToString(pDb->pEnv, pSeg, nMin);
+    snprintf(&aBuf[i], nBuf-i, "%s", zSeg);
+    i += strlen(&aBuf[i]);
+    lsmFree(pDb->pEnv, zSeg);
+
+#ifdef LSM_LOG_FREELIST
+    lsmInfoArrayStructure(pDb, 1, pSeg->iFirst, &zSeg);
+    snprintf(&aBuf[i], nBuf-1, "    (%s)", zSeg);
+    i += strlen(&aBuf[i]);
+    lsmFree(pDb->pEnv, zSeg);
+#endif
+    aBuf[nBuf] = 0;
+  }else{
+    aBuf[0] = '\0';
+  }
+
+  return i;
+}
+
+void sortedDumpPage(lsm_db *pDb, Segment *pRun, Page *pPg, int bVals){
+  LsmBlob blob = {0, 0, 0};       /* LsmBlob used for keys */
+  LsmString s;
+  int i;
+
+  int nRec;
+  int iPtr;
+  int flags;
+  u8 *aData;
+  int nData;
+
+  aData = fsPageData(pPg, &nData);
+
+  nRec = pageGetNRec(aData, nData);
+  iPtr = (int)pageGetPtr(aData, nData);
+  flags = pageGetFlags(aData, nData);
+
+  lsmStringInit(&s, pDb->pEnv);
+  lsmStringAppendf(&s,"nCell=%d iPtr=%d flags=%d {", nRec, iPtr, flags);
+  if( flags&SEGMENT_BTREE_FLAG ) iPtr = 0;
+
+  for(i=0; i<nRec; i++){
+    Page *pRef = 0;               /* Pointer to page iRef */
+    int iChar;
+    u8 *aKey; int nKey = 0;       /* Key */
+    u8 *aVal = 0; int nVal = 0;   /* Value */
+    int iTopic;
+    u8 *aCell;
+    int iPgPtr;
+    int eType;
+
+    aCell = pageGetCell(aData, nData, i);
+    eType = *aCell++;
+    assert( (flags & SEGMENT_BTREE_FLAG) || eType!=0 );
+    aCell += lsmVarintGet32(aCell, &iPgPtr);
+
+    if( eType==0 ){
+      LsmPgno iRef;               /* Page number of referenced page */
+      aCell += lsmVarintGet64(aCell, &iRef);
+      lsmFsDbPageGet(pDb->pFS, pRun, iRef, &pRef);
+      aKey = pageGetKey(pRun, pRef, 0, &iTopic, &nKey, &blob);
+    }else{
+      aCell += lsmVarintGet32(aCell, &nKey);
+      if( rtIsWrite(eType) ) aCell += lsmVarintGet32(aCell, &nVal);
+      sortedReadData(0, pPg, (aCell-aData), nKey+nVal, (void **)&aKey, &blob);
+      aVal = &aKey[nKey];
+      iTopic = eType;
+    }
+
+    lsmStringAppendf(&s, "%s%2X:", (i==0?"":" "), iTopic);
+    for(iChar=0; iChar<nKey; iChar++){
+      lsmStringAppendf(&s, "%c", isalnum(aKey[iChar]) ? aKey[iChar] : '.');
+    }
+    if( nVal>0 && bVals ){
+      lsmStringAppendf(&s, "##");
+      for(iChar=0; iChar<nVal; iChar++){
+        lsmStringAppendf(&s, "%c", isalnum(aVal[iChar]) ? aVal[iChar] : '.');
+      }
+    }
+
+    lsmStringAppendf(&s, " %d", iPgPtr+iPtr);
+    lsmFsPageRelease(pRef);
+  }
+  lsmStringAppend(&s, "}", 1);
+
+  lsmLogMessage(pDb, LSM_OK, "      Page %d: %s", lsmFsPageNumber(pPg), s.z);
+  lsmStringClear(&s);
+
+  sortedBlobFree(&blob);
+}
+
+static void infoCellDump(
+  lsm_db *pDb,                    /* Database handle */
+  Segment *pSeg,                  /* Segment page belongs to */
+  int bIndirect,                  /* True to follow indirect refs */
+  Page *pPg,
+  int iCell,
+  int *peType,
+  int *piPgPtr,
+  u8 **paKey, int *pnKey,
+  u8 **paVal, int *pnVal,
+  LsmBlob *pBlob
+){
+  u8 *aData; int nData;           /* Page data */
+  u8 *aKey; int nKey = 0;         /* Key */
+  u8 *aVal = 0; int nVal = 0;     /* Value */
+  int eType;
+  int iPgPtr;
+  Page *pRef = 0;                 /* Pointer to page iRef */
+  u8 *aCell;
+
+  aData = fsPageData(pPg, &nData);
+
+  aCell = pageGetCell(aData, nData, iCell);
+  eType = *aCell++;
+  aCell += lsmVarintGet32(aCell, &iPgPtr);
+
+  if( eType==0 ){
+    int dummy;
+    LsmPgno iRef;                 /* Page number of referenced page */
+    aCell += lsmVarintGet64(aCell, &iRef);
+    if( bIndirect ){
+      lsmFsDbPageGet(pDb->pFS, pSeg, iRef, &pRef);
+      pageGetKeyCopy(pDb->pEnv, pSeg, pRef, 0, &dummy, pBlob);
+      aKey = (u8 *)pBlob->pData;
+      nKey = pBlob->nData;
+      lsmFsPageRelease(pRef);
+    }else{
+      aKey = (u8 *)"<indirect>";
+      nKey = 11;
+    }
+  }else{
+    aCell += lsmVarintGet32(aCell, &nKey);
+    if( rtIsWrite(eType) ) aCell += lsmVarintGet32(aCell, &nVal);
+    sortedReadData(pSeg, pPg, (aCell-aData), nKey+nVal, (void **)&aKey, pBlob);
+    aVal = &aKey[nKey];
+  }
+
+  if( peType ) *peType = eType;
+  if( piPgPtr ) *piPgPtr = iPgPtr;
+  if( paKey ) *paKey = aKey;
+  if( paVal ) *paVal = aVal;
+  if( pnKey ) *pnKey = nKey;
+  if( pnVal ) *pnVal = nVal;
+}
+
+static int infoAppendBlob(LsmString *pStr, int bHex, u8 *z, int n){
+  int iChar;
+  for(iChar=0; iChar<n; iChar++){
+    if( bHex ){
+      lsmStringAppendf(pStr, "%02X", z[iChar]);
+    }else{
+      lsmStringAppendf(pStr, "%c", isalnum(z[iChar]) ?z[iChar] : '.');
+    }
+  }
+  return LSM_OK;
+}
+
+#define INFO_PAGE_DUMP_DATA     0x01
+#define INFO_PAGE_DUMP_VALUES   0x02
+#define INFO_PAGE_DUMP_HEX      0x04
+#define INFO_PAGE_DUMP_INDIRECT 0x08
+
+static int infoPageDump(
+  lsm_db *pDb,                    /* Database handle */
+  LsmPgno iPg,                    /* Page number of page to dump */
+  int flags,
+  char **pzOut                    /* OUT: lsmMalloc'd string */
+){
+  int rc = LSM_OK;                /* Return code */
+  Page *pPg = 0;                  /* Handle for page iPg */
+  int i, j;                       /* Loop counters */
+  const int perLine = 16;         /* Bytes per line in the raw hex dump */
+  Segment *pSeg = 0;
+  Snapshot *pSnap;
+
+  int bValues = (flags & INFO_PAGE_DUMP_VALUES);
+  int bHex = (flags & INFO_PAGE_DUMP_HEX);
+  int bData = (flags & INFO_PAGE_DUMP_DATA);
+  int bIndirect = (flags & INFO_PAGE_DUMP_INDIRECT);
+
+  *pzOut = 0;
+  if( iPg==0 ) return LSM_ERROR;
+
+  assert( pDb->pClient || pDb->pWorker );
+  pSnap = pDb->pClient;
+  if( pSnap==0 ) pSnap = pDb->pWorker;
+  if( pSnap->redirect.n>0 ){
+    Level *pLvl;
+    int bUse = 0;
+    for(pLvl=pSnap->pLevel; pLvl->pNext; pLvl=pLvl->pNext);
+    pSeg = (pLvl->nRight==0 ? &pLvl->lhs : &pLvl->aRhs[pLvl->nRight-1]);
+    rc = lsmFsSegmentContainsPg(pDb->pFS, pSeg, iPg, &bUse);
+    if( bUse==0 ){
+      pSeg = 0;
+    }
+  }
+
+  /* iPg is a real page number (not subject to redirection). So it is safe 
+  ** to pass a NULL in place of the segment pointer as the second argument
+  ** to lsmFsDbPageGet() here.  */
+  if( rc==LSM_OK ){
+    rc = lsmFsDbPageGet(pDb->pFS, 0, iPg, &pPg);
+  }
+
+  if( rc==LSM_OK ){
+    LsmBlob blob = {0, 0, 0, 0};
+    int nKeyWidth = 0;
+    LsmString str;
+    int nRec;
+    int iPtr;
+    int flags2;
+    int iCell;
+    u8 *aData; int nData;         /* Page data and size thereof */
+
+    aData = fsPageData(pPg, &nData);
+    nRec = pageGetNRec(aData, nData);
+    iPtr = (int)pageGetPtr(aData, nData);
+    flags2 = pageGetFlags(aData, nData);
+
+    lsmStringInit(&str, pDb->pEnv);
+    lsmStringAppendf(&str, "Page : %lld  (%d bytes)\n", iPg, nData);
+    lsmStringAppendf(&str, "nRec : %d\n", nRec);
+    lsmStringAppendf(&str, "iPtr : %d\n", iPtr);
+    lsmStringAppendf(&str, "flags: %04x\n", flags2);
+    lsmStringAppendf(&str, "\n");
+
+    for(iCell=0; iCell<nRec; iCell++){
+      int nKey;
+      infoCellDump(
+          pDb, pSeg, bIndirect, pPg, iCell, 0, 0, 0, &nKey, 0, 0, &blob
+      );
+      if( nKey>nKeyWidth ) nKeyWidth = nKey;
+    }
+    if( bHex ) nKeyWidth = nKeyWidth * 2;
+
+    for(iCell=0; iCell<nRec; iCell++){
+      u8 *aKey; int nKey = 0;       /* Key */
+      u8 *aVal; int nVal = 0;       /* Value */
+      int iPgPtr;
+      int eType;
+      LsmPgno iAbsPtr;
+      char zFlags[8];
+
+      infoCellDump(pDb, pSeg, bIndirect, pPg, iCell, &eType, &iPgPtr,
+          &aKey, &nKey, &aVal, &nVal, &blob
+      );
+      iAbsPtr = iPgPtr + ((flags2 & SEGMENT_BTREE_FLAG) ? 0 : iPtr);
+
+      lsmFlagsToString(eType, zFlags);
+      lsmStringAppendf(&str, "%s %d (%s) ", 
+          zFlags, iAbsPtr, (rtTopic(eType) ? "sys" : "usr")
+      );
+      infoAppendBlob(&str, bHex, aKey, nKey); 
+      if( nVal>0 && bValues ){
+        lsmStringAppendf(&str, "%*s", nKeyWidth - (nKey*(1+bHex)), "");
+        lsmStringAppendf(&str, " ");
+        infoAppendBlob(&str, bHex, aVal, nVal); 
+      }
+      if( rtTopic(eType) ){
+        int iBlk = (int)~lsmGetU32(aKey);
+        lsmStringAppendf(&str, "  (block=%d", iBlk);
+        if( nVal>0 ){
+          i64 iSnap = lsmGetU64(aVal);
+          lsmStringAppendf(&str, " snapshot=%lld", iSnap);
+        }
+        lsmStringAppendf(&str, ")");
+      }
+      lsmStringAppendf(&str, "\n");
+    }
+
+    if( bData ){
+      lsmStringAppendf(&str, "\n-------------------" 
+          "-------------------------------------------------------------\n");
+      lsmStringAppendf(&str, "Page %d\n",
+          iPg, (iPg-1)*nData, iPg*nData - 1);
+      for(i=0; i<nData; i += perLine){
+        lsmStringAppendf(&str, "%04x: ", i);
+        for(j=0; j<perLine; j++){
+          if( i+j>nData ){
+            lsmStringAppendf(&str, "   ");
+          }else{
+            lsmStringAppendf(&str, "%02x ", aData[i+j]);
+          }
+        }
+        lsmStringAppendf(&str, "  ");
+        for(j=0; j<perLine; j++){
+          if( i+j>nData ){
+            lsmStringAppendf(&str, " ");
+          }else{
+            lsmStringAppendf(&str,"%c", isprint(aData[i+j]) ? aData[i+j] : '.');
+          }
+        }
+        lsmStringAppendf(&str,"\n");
+      }
+    }
+
+    *pzOut = str.z;
+    sortedBlobFree(&blob);
+    lsmFsPageRelease(pPg);
+  }
+
+  return rc;
+}
+
+int lsmInfoPageDump(
+  lsm_db *pDb,                    /* Database handle */
+  LsmPgno iPg,                    /* Page number of page to dump */
+  int bHex,                       /* True to output key/value in hex form */
+  char **pzOut                    /* OUT: lsmMalloc'd string */
+){
+  int flags = INFO_PAGE_DUMP_DATA | INFO_PAGE_DUMP_VALUES;
+  if( bHex ) flags |= INFO_PAGE_DUMP_HEX;
+  return infoPageDump(pDb, iPg, flags, pzOut);
+}
+
+void sortedDumpSegment(lsm_db *pDb, Segment *pRun, int bVals){
+  assert( pDb->xLog );
+  if( pRun && pRun->iFirst ){
+    int flags = (bVals ? INFO_PAGE_DUMP_VALUES : 0);
+    char *zSeg;
+    Page *pPg;
+
+    zSeg = segToString(pDb->pEnv, pRun, 0);
+    lsmLogMessage(pDb, LSM_OK, "Segment: %s", zSeg);
+    lsmFree(pDb->pEnv, zSeg);
+
+    lsmFsDbPageGet(pDb->pFS, pRun, pRun->iFirst, &pPg);
+    while( pPg ){
+      Page *pNext;
+      char *z = 0;
+      infoPageDump(pDb, lsmFsPageNumber(pPg), flags, &z);
+      lsmLogMessage(pDb, LSM_OK, "%s", z);
+      lsmFree(pDb->pEnv, z);
+#if 0
+      sortedDumpPage(pDb, pRun, pPg, bVals);
+#endif
+      lsmFsDbPageNext(pRun, pPg, 1, &pNext);
+      lsmFsPageRelease(pPg);
+      pPg = pNext;
+    }
+  }
+}
+
+/*
+** Invoke the log callback zero or more times with messages that describe
+** the current database structure.
+*/
+void lsmSortedDumpStructure(
+  lsm_db *pDb,                    /* Database handle (used for xLog callback) */
+  Snapshot *pSnap,                /* Snapshot to dump */
+  int bKeys,                      /* Output the keys from each segment */
+  int bVals,                      /* Output the values from each segment */
+  const char *zWhy                /* Caption to print near top of dump */
+){
+  Snapshot *pDump = pSnap;
+  Level *pTopLevel;
+  char *zFree = 0;
+
+  assert( pSnap );
+  pTopLevel = lsmDbSnapshotLevel(pDump);
+  if( pDb->xLog && pTopLevel ){
+    static int nCall = 0;
+    Level *pLevel;
+    int iLevel = 0;
+
+    nCall++;
+    lsmLogMessage(pDb, LSM_OK, "Database structure %d (%s)", nCall, zWhy);
+
+#if 0
+    if( nCall==1031 || nCall==1032 ) bKeys=1;
+#endif
+
+    for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
+      char zLeft[1024];
+      char zRight[1024];
+      int i = 0;
+
+      Segment *aLeft[24];  
+      Segment *aRight[24];
+
+      int nLeft = 0;
+      int nRight = 0;
+
+      Segment *pSeg = &pLevel->lhs;
+      aLeft[nLeft++] = pSeg;
+
+      for(i=0; i<pLevel->nRight; i++){
+        aRight[nRight++] = &pLevel->aRhs[i];
+      }
+
+#ifdef LSM_LOG_FREELIST
+      if( nRight ){
+        memmove(&aRight[1], aRight, sizeof(aRight[0])*nRight);
+        aRight[0] = 0;
+        nRight++;
+      }
+#endif
+
+      for(i=0; i<nLeft || i<nRight; i++){
+        int iPad = 0;
+        char zLevel[32];
+        zLeft[0] = '\0';
+        zRight[0] = '\0';
+
+        if( i<nLeft ){ 
+          fileToString(pDb, zLeft, sizeof(zLeft), 24, aLeft[i]); 
+        }
+        if( i<nRight ){ 
+          fileToString(pDb, zRight, sizeof(zRight), 24, aRight[i]); 
+        }
+
+        if( i==0 ){
+          snprintf(zLevel, sizeof(zLevel), "L%d: (age=%d) (flags=%.4x)",
+              iLevel, (int)pLevel->iAge, (int)pLevel->flags
+          );
+        }else{
+          zLevel[0] = '\0';
+        }
+
+        if( nRight==0 ){
+          iPad = 10;
+        }
+
+        lsmLogMessage(pDb, LSM_OK, "% 25s % *s% -35s %s", 
+            zLevel, iPad, "", zLeft, zRight
+        );
+      }
+
+      iLevel++;
+    }
+
+    if( bKeys ){
+      for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
+        int i;
+        sortedDumpSegment(pDb, &pLevel->lhs, bVals);
+        for(i=0; i<pLevel->nRight; i++){
+          sortedDumpSegment(pDb, &pLevel->aRhs[i], bVals);
+        }
+      }
+    }
+  }
+
+  lsmInfoFreelist(pDb, &zFree);
+  lsmLogMessage(pDb, LSM_OK, "Freelist: %s", zFree);
+  lsmFree(pDb->pEnv, zFree);
+
+  assert( lsmFsIntegrityCheck(pDb) );
+}
+
+void lsmSortedFreeLevel(lsm_env *pEnv, Level *pLevel){
+  Level *pNext;
+  Level *p;
+
+  for(p=pLevel; p; p=pNext){
+    pNext = p->pNext;
+    sortedFreeLevel(pEnv, p);
+  }
+}
+
+void lsmSortedSaveTreeCursors(lsm_db *pDb){
+  MultiCursor *pCsr;
+  for(pCsr=pDb->pCsr; pCsr; pCsr=pCsr->pNext){
+    lsmTreeCursorSave(pCsr->apTreeCsr[0]);
+    lsmTreeCursorSave(pCsr->apTreeCsr[1]);
+  }
+}
+
+void lsmSortedExpandBtreePage(Page *pPg, int nOrig){
+  u8 *aData;
+  int nData;
+  int nEntry;
+  int iHdr;
+
+  aData = lsmFsPageData(pPg, &nData);
+  nEntry = pageGetNRec(aData, nOrig);
+  iHdr = SEGMENT_EOF(nOrig, nEntry);
+  memmove(&aData[iHdr + (nData-nOrig)], &aData[iHdr], nOrig-iHdr);
+}
+
+#ifdef LSM_DEBUG_EXPENSIVE
+static void assertRunInOrder(lsm_db *pDb, Segment *pSeg){
+  Page *pPg = 0;
+  LsmBlob blob1 = {0, 0, 0, 0};
+  LsmBlob blob2 = {0, 0, 0, 0};
+
+  lsmFsDbPageGet(pDb->pFS, pSeg, pSeg->iFirst, &pPg);
+  while( pPg ){
+    u8 *aData; int nData;
+    Page *pNext;
+
+    aData = lsmFsPageData(pPg, &nData);
+    if( 0==(pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG) ){
+      int i;
+      int nRec = pageGetNRec(aData, nData);
+      for(i=0; i<nRec; i++){
+        int iTopic1, iTopic2;
+        pageGetKeyCopy(pDb->pEnv, pSeg, pPg, i, &iTopic1, &blob1);
+
+        if( i==0 && blob2.nData ){
+          assert( sortedKeyCompare(
+                pDb->xCmp, iTopic2, blob2.pData, blob2.nData,
+                iTopic1, blob1.pData, blob1.nData
+          )<0 );
+        }
+
+        if( i<(nRec-1) ){
+          pageGetKeyCopy(pDb->pEnv, pSeg, pPg, i+1, &iTopic2, &blob2);
+          assert( sortedKeyCompare(
+                pDb->xCmp, iTopic1, blob1.pData, blob1.nData,
+                iTopic2, blob2.pData, blob2.nData
+          )<0 );
+        }
+      }
+    }
+
+    lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
+    lsmFsPageRelease(pPg);
+    pPg = pNext;
+  }
+
+  sortedBlobFree(&blob1);
+  sortedBlobFree(&blob2);
+}
+#endif
+
+#ifdef LSM_DEBUG_EXPENSIVE
+/*
+** This function is only included in the build if LSM_DEBUG_EXPENSIVE is 
+** defined. Its only purpose is to evaluate various assert() statements to 
+** verify that the database is well formed in certain respects.
+**
+** More specifically, it checks that the array pOne contains the required 
+** pointers to pTwo. Array pTwo must be a main array. pOne may be either a 
+** separators array or another main array. If pOne does not contain the 
+** correct set of pointers, an assert() statement fails.
+*/
+static int assertPointersOk(
+  lsm_db *pDb,                    /* Database handle */
+  Segment *pOne,                  /* Segment containing pointers */
+  Segment *pTwo,                  /* Segment containing pointer targets */
+  int bRhs                        /* True if pTwo may have been Gobble()d */
+){
+  int rc = LSM_OK;                /* Error code */
+  SegmentPtr ptr1;                /* Iterates through pOne */
+  SegmentPtr ptr2;                /* Iterates through pTwo */
+  LsmPgno iPrev;
+
+  assert( pOne && pTwo );
+
+  memset(&ptr1, 0, sizeof(ptr1));
+  memset(&ptr2, 0, sizeof(ptr1));
+  ptr1.pSeg = pOne;
+  ptr2.pSeg = pTwo;
+  segmentPtrEndPage(pDb->pFS, &ptr1, 0, &rc);
+  segmentPtrEndPage(pDb->pFS, &ptr2, 0, &rc);
+
+  /* Check that the footer pointer of the first page of pOne points to
+  ** the first page of pTwo. */
+  iPrev = pTwo->iFirst;
+  if( ptr1.iPtr!=iPrev && !bRhs ){
+    assert( 0 );
+  }
+
+  if( rc==LSM_OK && ptr1.nCell>0 ){
+    rc = segmentPtrLoadCell(&ptr1, 0);
+  }
+      
+  while( rc==LSM_OK && ptr2.pPg ){
+    LsmPgno iThis;
+
+    /* Advance to the next page of segment pTwo that contains at least
+    ** one cell. Break out of the loop if the iterator reaches EOF.  */
+    do{
+      rc = segmentPtrNextPage(&ptr2, 1);
+      assert( rc==LSM_OK );
+    }while( rc==LSM_OK && ptr2.pPg && ptr2.nCell==0 );
+    if( rc!=LSM_OK || ptr2.pPg==0 ) break;
+    iThis = lsmFsPageNumber(ptr2.pPg);
+
+    if( (ptr2.flags & (PGFTR_SKIP_THIS_FLAG|SEGMENT_BTREE_FLAG))==0 ){
+
+      /* Load the first cell in the array pTwo page. */
+      rc = segmentPtrLoadCell(&ptr2, 0);
+
+      /* Iterate forwards through pOne, searching for a key that matches the
+      ** key ptr2.pKey/nKey. This key should have a pointer to the page that
+      ** ptr2 currently points to. */
+      while( rc==LSM_OK ){
+        int res = rtTopic(ptr1.eType) - rtTopic(ptr2.eType);
+        if( res==0 ){
+          res = pDb->xCmp(ptr1.pKey, ptr1.nKey, ptr2.pKey, ptr2.nKey);
+        }
+
+        if( res<0 ){
+          assert( bRhs || ptr1.iPtr+ptr1.iPgPtr==iPrev );
+        }else if( res>0 ){
+          assert( 0 );
+        }else{
+          assert( ptr1.iPtr+ptr1.iPgPtr==iThis );
+          iPrev = iThis;
+          break;
+        }
+
+        rc = segmentPtrAdvance(0, &ptr1, 0);
+        if( ptr1.pPg==0 ){
+          assert( 0 );
+        }
+      }
+    }
+  }
+
+  segmentPtrReset(&ptr1, 0);
+  segmentPtrReset(&ptr2, 0);
+  return LSM_OK;
+}
+
+/*
+** This function is only included in the build if LSM_DEBUG_EXPENSIVE is 
+** defined. Its only purpose is to evaluate various assert() statements to 
+** verify that the database is well formed in certain respects.
+**
+** More specifically, it checks that the b-tree embedded in array pRun
+** contains the correct keys. If not, an assert() fails.
+*/
+static int assertBtreeOk(
+  lsm_db *pDb,
+  Segment *pSeg
+){
+  int rc = LSM_OK;                /* Return code */
+  if( pSeg->iRoot ){
+    LsmBlob blob = {0, 0, 0};     /* Buffer used to cache overflow keys */
+    FileSystem *pFS = pDb->pFS;   /* File system to read from */
+    Page *pPg = 0;                /* Main run page */
+    BtreeCursor *pCsr = 0;        /* Btree cursor */
+
+    rc = btreeCursorNew(pDb, pSeg, &pCsr);
+    if( rc==LSM_OK ){
+      rc = btreeCursorFirst(pCsr);
+    }
+    if( rc==LSM_OK ){
+      rc = lsmFsDbPageGet(pFS, pSeg, pSeg->iFirst, &pPg);
+    }
+
+    while( rc==LSM_OK ){
+      Page *pNext;
+      u8 *aData;
+      int nData;
+      int flags;
+
+      rc = lsmFsDbPageNext(pSeg, pPg, 1, &pNext);
+      lsmFsPageRelease(pPg);
+      pPg = pNext;
+      if( pPg==0 ) break;
+      aData = fsPageData(pPg, &nData);
+      flags = pageGetFlags(aData, nData);
+      if( rc==LSM_OK 
+       && 0==((SEGMENT_BTREE_FLAG|PGFTR_SKIP_THIS_FLAG) & flags)
+       && 0!=pageGetNRec(aData, nData)
+      ){
+        u8 *pKey;
+        int nKey;
+        int iTopic;
+        pKey = pageGetKey(pSeg, pPg, 0, &iTopic, &nKey, &blob);
+        assert( nKey==pCsr->nKey && 0==memcmp(pKey, pCsr->pKey, nKey) );
+        assert( lsmFsPageNumber(pPg)==pCsr->iPtr );
+        rc = btreeCursorNext(pCsr);
+      }
+    }
+    assert( rc!=LSM_OK || pCsr->pKey==0 );
+
+    if( pPg ) lsmFsPageRelease(pPg);
+
+    btreeCursorFree(pCsr);
+    sortedBlobFree(&blob);
+  }
+
+  return rc;
+}
+#endif /* ifdef LSM_DEBUG_EXPENSIVE */
diff --git a/ext/lsm1/lsm_str.c b/ext/lsm1/lsm_str.c
new file mode 100644
index 0000000..9b1b63c
--- /dev/null
+++ b/ext/lsm1/lsm_str.c
@@ -0,0 +1,148 @@
+/*
+** 2012-04-27
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** Dynamic string functions.
+*/
+#include "lsmInt.h"
+
+/*
+** Turn bulk and uninitialized memory into an LsmString object
+*/
+void lsmStringInit(LsmString *pStr, lsm_env *pEnv){
+  memset(pStr, 0, sizeof(pStr[0]));
+  pStr->pEnv = pEnv;
+}
+
+/*
+** Increase the memory allocated for holding the string.  Realloc as needed.
+**
+** If a memory allocation error occurs, set pStr->n to -1 and free the existing
+** allocation.  If a prior memory allocation has occurred, this routine is a
+** no-op.
+*/
+int lsmStringExtend(LsmString *pStr, int nNew){
+  assert( nNew>0 );
+  if( pStr->n<0 ) return LSM_NOMEM;
+  if( pStr->n + nNew >= pStr->nAlloc ){
+    int nAlloc = pStr->n + nNew + 100;
+    char *zNew = lsmRealloc(pStr->pEnv, pStr->z, nAlloc);
+    if( zNew==0 ){
+      lsmFree(pStr->pEnv, pStr->z);
+      nAlloc = 0;
+      pStr->n = -1;
+    }
+    pStr->nAlloc = nAlloc;
+    pStr->z = zNew;
+  }
+  return (pStr->z ? LSM_OK : LSM_NOMEM_BKPT);
+}
+
+/*
+** Clear an LsmString object, releasing any allocated memory that it holds.
+** This also clears the error indication (if any).
+*/
+void lsmStringClear(LsmString *pStr){
+  lsmFree(pStr->pEnv, pStr->z);
+  lsmStringInit(pStr, pStr->pEnv);
+}
+
+/*
+** Append N bytes of text to the end of an LsmString object.  If
+** N is negative, append the entire string.
+**
+** If the string is in an error state, this routine is a no-op.
+*/
+int lsmStringAppend(LsmString *pStr, const char *z, int N){
+  int rc;
+  if( N<0 ) N = (int)strlen(z);
+  rc = lsmStringExtend(pStr, N+1);
+  if( pStr->nAlloc ){
+    memcpy(pStr->z+pStr->n, z, N+1);
+    pStr->n += N;
+  }
+  return rc;
+}
+
+int lsmStringBinAppend(LsmString *pStr, const u8 *a, int n){
+  int rc;
+  rc = lsmStringExtend(pStr, n);
+  if( pStr->nAlloc ){
+    memcpy(pStr->z+pStr->n, a, n);
+    pStr->n += n;
+  }
+  return rc;
+}
+
+/*
+** Append printf-formatted content to an LsmString.
+*/
+void lsmStringVAppendf(
+  LsmString *pStr, 
+  const char *zFormat, 
+  va_list ap1,
+  va_list ap2
+){
+#if (!defined(__STDC_VERSION__) || (__STDC_VERSION__<199901L)) && \
+    !defined(__APPLE__)
+  extern int vsnprintf(char *str, size_t size, const char *format, va_list ap)
+    /* Compatibility crutch for C89 compilation mode. sqlite3_vsnprintf()
+       does not work identically and causes test failures if used here.
+       For the time being we are assuming that the target has vsnprintf(),
+       but that is not guaranteed to be the case for pure C89 platforms.
+    */;
+#endif
+  int nWrite;
+  int nAvail;
+
+  nAvail = pStr->nAlloc - pStr->n;
+  nWrite = vsnprintf(pStr->z + pStr->n, nAvail, zFormat, ap1);
+
+  if( nWrite>=nAvail ){
+    lsmStringExtend(pStr, nWrite+1);
+    if( pStr->nAlloc==0 ) return;
+    nWrite = vsnprintf(pStr->z + pStr->n, nWrite+1, zFormat, ap2);
+  }
+
+  pStr->n += nWrite;
+  pStr->z[pStr->n] = 0;
+}
+
+void lsmStringAppendf(LsmString *pStr, const char *zFormat, ...){
+  va_list ap, ap2;
+  va_start(ap, zFormat);
+  va_start(ap2, zFormat);
+  lsmStringVAppendf(pStr, zFormat, ap, ap2);
+  va_end(ap);
+  va_end(ap2);
+}
+
+int lsmStrlen(const char *zName){
+  int nRet = 0;
+  while( zName[nRet] ) nRet++;
+  return nRet;
+}
+
+/*
+** Write into memory obtained from lsm_malloc().
+*/
+char *lsmMallocPrintf(lsm_env *pEnv, const char *zFormat, ...){
+  LsmString s;
+  va_list ap, ap2;
+  lsmStringInit(&s, pEnv);
+  va_start(ap, zFormat);
+  va_start(ap2, zFormat);
+  lsmStringVAppendf(&s, zFormat, ap, ap2);
+  va_end(ap);
+  va_end(ap2);
+  if( s.n<0 ) return 0;
+  return (char *)lsmReallocOrFree(pEnv, s.z, s.n+1);
+}
diff --git a/ext/lsm1/lsm_tree.c b/ext/lsm1/lsm_tree.c
new file mode 100644
index 0000000..1a199fc
--- /dev/null
+++ b/ext/lsm1/lsm_tree.c
@@ -0,0 +1,2465 @@
+/*
+** 2011-08-18
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** This file contains the implementation of an in-memory tree structure.
+**
+** Technically the tree is a B-tree of order 4 (in the Knuth sense - each 
+** node may have up to 4 children). Keys are stored within B-tree nodes by
+** reference. This may be slightly slower than a conventional red-black
+** tree, but it is simpler. It is also an easier structure to modify to 
+** create a version that supports nested transaction rollback.
+**
+** This tree does not currently support a delete operation. One is not 
+** required. When LSM deletes a key from a database, it inserts a DELETE
+** marker into the data structure. As a result, although the value associated
+** with a key stored in the in-memory tree structure may be modified, no
+** keys are ever removed. 
+*/
+
+/*
+** MVCC NOTES
+**
+**   The in-memory tree structure supports SQLite-style MVCC. This means
+**   that while one client is writing to the tree structure, other clients
+**   may still be querying an older snapshot of the tree.
+**
+**   One way to implement this is to use an append-only b-tree. In this 
+**   case instead of modifying nodes in-place, a copy of the node is made
+**   and the required modifications made to the copy. The parent of the
+**   node is then modified (to update the pointer so that it points to
+**   the new copy), which causes a copy of the parent to be made, and so on.
+**   This means that each time the tree is written to a new root node is
+**   created. A snapshot is identified by the root node that it uses.
+**
+**   The problem with the above is that each time the tree is written to,
+**   a copy of the node structure modified and all of its ancestor nodes
+**   is made. This may prove excessive with large tree structures.
+**
+**   To reduce this overhead, the data structure used for a tree node is
+**   designed so that it may be edited in place exactly once without 
+**   affecting existing users. In other words, the node structure is capable
+**   of storing two separate versions of the node at the same time.
+**   When a node is to be edited, if the node structure already contains 
+**   two versions, a copy is made as in the append-only approach. Or, if
+**   it only contains a single version, it is edited in place.
+**
+**   This reduces the overhead so that, roughly, one new node structure
+**   must be allocated for each write (on top of those allocations that 
+**   would have been required by a non-MVCC tree). Logic: Assume that at 
+**   any time, 50% of nodes in the tree already contain 2 versions. When
+**   a new entry is written to a node, there is a 50% chance that a copy
+**   of the node will be required. And a 25% chance that a copy of its 
+**   parent is required. And so on.
+**
+** ROLLBACK
+**
+**   The in-memory tree also supports transaction and sub-transaction 
+**   rollback. In order to rollback to point in time X, the following is
+**   necessary:
+**
+**     1. All memory allocated since X must be freed, and 
+**     2. All "v2" data adding to nodes that existed at X should be zeroed.
+**     3. The root node must be restored to its X value.
+**
+**   The Mempool object used to allocate memory for the tree supports 
+**   operation (1) - see the lsmPoolMark() and lsmPoolRevert() functions.
+**
+**   To support (2), all nodes that have v2 data are part of a singly linked 
+**   list, sorted by the age of the v2 data (nodes that have had data added 
+**   most recently are at the end of the list). So to zero all v2 data added
+**   since X, the linked list is traversed from the first node added following
+**   X onwards.
+**
+*/
+
+#ifndef _LSM_INT_H
+# include "lsmInt.h"
+#endif
+
+#include <string.h>
+
+#define MAX_DEPTH 32
+
+typedef struct TreeKey TreeKey;
+typedef struct TreeNode TreeNode;
+typedef struct TreeLeaf TreeLeaf;
+typedef struct NodeVersion NodeVersion;
+
+struct TreeOld {
+  u32 iShmid;                     /* Last shared-memory chunk in use by old */
+  u32 iRoot;                      /* Offset of root node in shm file */
+  u32 nHeight;                    /* Height of tree structure */
+};
+
+#if 0
+/*
+** assert() that a TreeKey.flags value is sane. Usage:
+**
+**   assert( lsmAssertFlagsOk(pTreeKey->flags) );
+*/
+static int lsmAssertFlagsOk(u8 keyflags){
+  /* At least one flag must be set. Otherwise, what is this key doing? */
+  assert( keyflags!=0 );
+
+  /* The POINT_DELETE and INSERT flags cannot both be set. */
+  assert( (keyflags & LSM_POINT_DELETE)==0 || (keyflags & LSM_INSERT)==0 );
+
+  /* If both the START_DELETE and END_DELETE flags are set, then the INSERT
+  ** flag must also be set. In other words - the three DELETE flags cannot
+  ** all be set */
+  assert( (keyflags & LSM_END_DELETE)==0 
+       || (keyflags & LSM_START_DELETE)==0 
+       || (keyflags & LSM_POINT_DELETE)==0 
+  );
+
+  return 1;
+}
+#endif
+static int assert_delete_ranges_match(lsm_db *);
+static int treeCountEntries(lsm_db *db);
+
+/*
+** Container for a key-value pair. Within the *-shm file, each key/value
+** pair is stored in a single allocation (which may not actually be 
+** contiguous in memory). Layout is the TreeKey structure, followed by
+** the nKey bytes of key blob, followed by the nValue bytes of value blob
+** (if nValue is non-negative).
+*/
+struct TreeKey {
+  int nKey;                       /* Size of pKey in bytes */
+  int nValue;                     /* Size of pValue. Or negative. */
+  u8 flags;                       /* Various LSM_XXX flags */
+};
+
+#define TKV_KEY(p) ((void *)&(p)[1])
+#define TKV_VAL(p) ((void *)(((u8 *)&(p)[1]) + (p)->nKey))
+
+/*
+** A single tree node. A node structure may contain up to 3 key/value
+** pairs. Internal (non-leaf) nodes have up to 4 children.
+**
+** TODO: Update the format of this to be more compact. Get it working
+** first though...
+*/
+struct TreeNode {
+  u32 aiKeyPtr[3];                /* Array of pointers to TreeKey objects */
+
+  /* The following fields are present for interior nodes only, not leaves. */
+  u32 aiChildPtr[4];              /* Array of pointers to child nodes */
+
+  /* The extra child pointer slot. */
+  u32 iV2;                        /* Transaction number of v2 */
+  u8 iV2Child;                    /* apChild[] entry replaced by pV2Ptr */
+  u32 iV2Ptr;                     /* Substitute pointer */
+};
+
+struct TreeLeaf {
+  u32 aiKeyPtr[3];                /* Array of pointers to TreeKey objects */
+};
+
+typedef struct TreeBlob TreeBlob;
+struct TreeBlob {
+  int n;
+  u8 *a;
+};
+
+/*
+** Cursor for searching a tree structure.
+**
+** If a cursor does not point to any element (a.k.a. EOF), then the
+** TreeCursor.iNode variable is set to a negative value. Otherwise, the
+** cursor currently points to key aiCell[iNode] on node apTreeNode[iNode].
+**
+** Entries in the apTreeNode[] and aiCell[] arrays contain the node and
+** index of the TreeNode.apChild[] pointer followed to descend to the 
+** current element. Hence apTreeNode[0] always contains the root node of
+** the tree.
+*/
+struct TreeCursor {
+  lsm_db *pDb;                    /* Database handle for this cursor */
+  TreeRoot *pRoot;                /* Root node and height of tree to access */
+  int iNode;                      /* Cursor points at apTreeNode[iNode] */
+  TreeNode *apTreeNode[MAX_DEPTH];/* Current position in tree */
+  u8 aiCell[MAX_DEPTH];           /* Current position in tree */
+  TreeKey *pSave;                 /* Saved key */
+  TreeBlob blob;                  /* Dynamic storage for a key */
+};
+
+/*
+** A value guaranteed to be larger than the largest possible transaction
+** id (TreeHeader.iTransId).
+*/
+#define WORKING_VERSION (1<<30)
+
+static int tblobGrow(lsm_db *pDb, TreeBlob *p, int n, int *pRc){
+  if( n>p->n ){
+    lsmFree(pDb->pEnv, p->a);
+    p->a = lsmMallocRc(pDb->pEnv, n, pRc);
+    p->n = n;
+  }
+  return (p->a==0);
+}
+static void tblobFree(lsm_db *pDb, TreeBlob *p){
+  lsmFree(pDb->pEnv, p->a);
+}
+
+
+/***********************************************************************
+** Start of IntArray methods.  */
+/*
+** Append value iVal to the contents of IntArray *p. Return LSM_OK if 
+** successful, or LSM_NOMEM if an OOM condition is encountered.
+*/
+static int intArrayAppend(lsm_env *pEnv, IntArray *p, u32 iVal){
+  assert( p->nArray<=p->nAlloc );
+  if( p->nArray>=p->nAlloc ){
+    u32 *aNew;
+    int nNew = p->nArray ? p->nArray*2 : 128;
+    aNew = lsmRealloc(pEnv, p->aArray, nNew*sizeof(u32));
+    if( !aNew ) return LSM_NOMEM_BKPT;
+    p->aArray = aNew;
+    p->nAlloc = nNew;
+  }
+
+  p->aArray[p->nArray++] = iVal;
+  return LSM_OK;
+}
+
+/*
+** Zero the IntArray object.
+*/
+static void intArrayFree(lsm_env *pEnv, IntArray *p){
+  p->nArray = 0;
+}
+
+/*
+** Return the number of entries currently in the int-array object.
+*/
+static int intArraySize(IntArray *p){
+  return p->nArray;
+}
+
+/*
+** Return a copy of the iIdx'th entry in the int-array.
+*/
+static u32 intArrayEntry(IntArray *p, int iIdx){
+  return p->aArray[iIdx];
+}
+
+/*
+** Truncate the int-array so that all but the first nVal values are 
+** discarded.
+*/
+static void intArrayTruncate(IntArray *p, int nVal){
+  p->nArray = nVal;
+}
+/* End of IntArray methods.
+***********************************************************************/
+
+static int treeKeycmp(void *p1, int n1, void *p2, int n2){
+  int res;
+  res = memcmp(p1, p2, LSM_MIN(n1, n2));
+  if( res==0 ) res = (n1-n2);
+  return res;
+}
+
+/*
+** The pointer passed as the first argument points to an interior node,
+** not a leaf. This function returns the offset of the iCell'th child
+** sub-tree of the node.
+*/
+static u32 getChildPtr(TreeNode *p, int iVersion, int iCell){
+  assert( iVersion>=0 );
+  assert( iCell>=0 && iCell<=array_size(p->aiChildPtr) );
+  if( p->iV2 && p->iV2<=(u32)iVersion && iCell==p->iV2Child ) return p->iV2Ptr;
+  return p->aiChildPtr[iCell];
+}
+
+/*
+** Given an offset within the *-shm file, return the associated chunk number.
+*/
+static int treeOffsetToChunk(u32 iOff){
+  assert( LSM_SHM_CHUNK_SIZE==(1<<15) );
+  return (int)(iOff>>15);
+}
+
+#define treeShmptrUnsafe(pDb, iPtr) \
+(&((u8*)((pDb)->apShm[(iPtr)>>15]))[(iPtr) & (LSM_SHM_CHUNK_SIZE-1)])
+
+/*
+** Return a pointer to the mapped memory location associated with *-shm 
+** file offset iPtr.
+*/
+static void *treeShmptr(lsm_db *pDb, u32 iPtr){
+
+  assert( (iPtr>>15)<(u32)pDb->nShm );
+  assert( pDb->apShm[iPtr>>15] );
+
+  return iPtr ? treeShmptrUnsafe(pDb, iPtr) : 0;
+}
+
+static ShmChunk * treeShmChunk(lsm_db *pDb, int iChunk){
+  return (ShmChunk *)(pDb->apShm[iChunk]);
+}
+
+static ShmChunk * treeShmChunkRc(lsm_db *pDb, int iChunk, int *pRc){
+  assert( *pRc==LSM_OK );
+  if( iChunk<pDb->nShm || LSM_OK==(*pRc = lsmShmCacheChunks(pDb, iChunk+1)) ){
+    return (ShmChunk *)(pDb->apShm[iChunk]);
+  }
+  return 0;
+}
+
+
+#ifndef NDEBUG
+static void assertIsWorkingChild(
+  lsm_db *db, 
+  TreeNode *pNode, 
+  TreeNode *pParent, 
+  int iCell
+){
+  TreeNode *p;
+  u32 iPtr = getChildPtr(pParent, WORKING_VERSION, iCell);
+  p = treeShmptr(db, iPtr);
+  assert( p==pNode );
+}
+#else
+# define assertIsWorkingChild(w,x,y,z)
+#endif
+
+/* Values for the third argument to treeShmkey(). */
+#define TKV_LOADKEY  1
+#define TKV_LOADVAL  2
+
+static TreeKey *treeShmkey(
+  lsm_db *pDb,                    /* Database handle */
+  u32 iPtr,                       /* Shmptr to TreeKey struct */
+  int eLoad,                      /* Either zero or a TREEKEY_LOADXXX value */
+  TreeBlob *pBlob,                /* Used if dynamic memory is required */
+  int *pRc                        /* IN/OUT: Error code */
+){
+  TreeKey *pRet;
+
+  assert( eLoad==TKV_LOADKEY || eLoad==TKV_LOADVAL );
+  pRet = (TreeKey *)treeShmptr(pDb, iPtr);
+  if( pRet ){
+    int nReq;                     /* Bytes of space required at pRet */
+    int nAvail;                   /* Bytes of space available at pRet */
+
+    nReq = sizeof(TreeKey) + pRet->nKey;
+    if( eLoad==TKV_LOADVAL && pRet->nValue>0 ){
+      nReq += pRet->nValue;
+    }
+    assert( LSM_SHM_CHUNK_SIZE==(1<<15) );
+    nAvail = LSM_SHM_CHUNK_SIZE - (iPtr & (LSM_SHM_CHUNK_SIZE-1));
+
+    if( nAvail<nReq ){
+      if( tblobGrow(pDb, pBlob, nReq, pRc)==0 ){
+        int nLoad = 0;
+        while( *pRc==LSM_OK ){
+          ShmChunk *pChunk;
+          void *p = treeShmptr(pDb, iPtr);
+          int n = LSM_MIN(nAvail, nReq-nLoad);
+
+          memcpy(&pBlob->a[nLoad], p, n);
+          nLoad += n;
+          if( nLoad==nReq ) break;
+
+          pChunk = treeShmChunk(pDb, treeOffsetToChunk(iPtr));
+          assert( pChunk );
+          iPtr = (pChunk->iNext * LSM_SHM_CHUNK_SIZE) + LSM_SHM_CHUNK_HDR;
+          nAvail = LSM_SHM_CHUNK_SIZE - LSM_SHM_CHUNK_HDR;
+        }
+      }
+      pRet = (TreeKey *)(pBlob->a);
+    }
+  }
+
+  return pRet;
+}
+
+#if defined(LSM_DEBUG) && defined(LSM_EXPENSIVE_ASSERT)
+void assert_leaf_looks_ok(TreeNode *pNode){
+  assert( pNode->apKey[1] );
+}
+
+void assert_node_looks_ok(TreeNode *pNode, int nHeight){
+  if( pNode ){
+    assert( pNode->apKey[1] );
+    if( nHeight>1 ){
+      int i;
+      assert( getChildPtr(pNode, WORKING_VERSION, 1) );
+      assert( getChildPtr(pNode, WORKING_VERSION, 2) );
+      for(i=0; i<4; i++){
+        assert_node_looks_ok(getChildPtr(pNode, WORKING_VERSION, i), nHeight-1);
+      }
+    }
+  }
+}
+
+/*
+** Run various assert() statements to check that the working-version of the
+** tree is correct in the following respects:
+**
+**   * todo...
+*/
+void assert_tree_looks_ok(int rc, Tree *pTree){
+}
+#else
+# define assert_tree_looks_ok(x,y)
+#endif
+
+void lsmFlagsToString(int flags, char *zFlags){
+
+  zFlags[0] = (flags & LSM_END_DELETE)   ? ']' : '.';
+
+  /* Only one of LSM_POINT_DELETE, LSM_INSERT and LSM_SEPARATOR should ever
+  ** be set. If this is not true, write a '?' to the output.  */
+  switch( flags & (LSM_POINT_DELETE|LSM_INSERT|LSM_SEPARATOR) ){
+    case 0:                zFlags[1] = '.'; break;
+    case LSM_POINT_DELETE: zFlags[1] = '-'; break;
+    case LSM_INSERT:       zFlags[1] = '+'; break;
+    case LSM_SEPARATOR:    zFlags[1] = '^'; break;
+    default:               zFlags[1] = '?'; break;
+  }
+
+  zFlags[2] = (flags & LSM_SYSTEMKEY)    ? '*' : '.';
+  zFlags[3] = (flags & LSM_START_DELETE) ? '[' : '.';
+  zFlags[4] = '\0';
+}
+
+#ifdef LSM_DEBUG
+
+/*
+** Pointer pBlob points to a buffer containing a blob of binary data
+** nBlob bytes long. Append the contents of this blob to *pStr, with
+** each octet represented by a 2-digit hexadecimal number. For example,
+** if the input blob is three bytes in size and contains {0x01, 0x44, 0xFF},
+** then "0144ff" is appended to *pStr.
+*/
+static void lsmAppendStrBlob(LsmString *pStr, void *pBlob, int nBlob){
+  int i;
+  lsmStringExtend(pStr, nBlob*2);
+  if( pStr->nAlloc==0 ) return;
+  for(i=0; i<nBlob; i++){
+    u8 c = ((u8*)pBlob)[i];
+    if( c>='a' && c<='z' ){
+      pStr->z[pStr->n++] = c;
+    }else if( c!=0 || nBlob==1 || i!=(nBlob-1) ){
+      pStr->z[pStr->n++] = "0123456789abcdef"[(c>>4)&0xf];
+      pStr->z[pStr->n++] = "0123456789abcdef"[c&0xf];
+    }
+  }
+  pStr->z[pStr->n] = 0;
+}
+
+#if 0  /* NOT USED */
+/*
+** Append nIndent space (0x20) characters to string *pStr.
+*/
+static void lsmAppendIndent(LsmString *pStr, int nIndent){
+  int i;
+  lsmStringExtend(pStr, nIndent);
+  for(i=0; i<nIndent; i++) lsmStringAppend(pStr, " ", 1);
+}
+#endif
+
+static void strAppendFlags(LsmString *pStr, u8 flags){
+  char zFlags[8];
+
+  lsmFlagsToString(flags, zFlags);
+  zFlags[4] = ':';
+
+  lsmStringAppend(pStr, zFlags, 5);
+}
+
+void dump_node_contents(
+  lsm_db *pDb,
+  u32 iNode,                      /* Print out the contents of this node */
+  char *zPath,                    /* Path from root to this node */
+  int nPath,                      /* Number of bytes in zPath */
+  int nHeight                     /* Height: (0==leaf) (1==parent-of-leaf) */
+){
+  const char *zSpace = "                                           ";
+  int i;
+  int rc = LSM_OK;
+  LsmString s;
+  TreeNode *pNode;
+  TreeBlob b = {0, 0};
+
+  pNode = (TreeNode *)treeShmptr(pDb, iNode);
+
+  if( nHeight==0 ){
+    /* Append the nIndent bytes of space to string s. */
+    lsmStringInit(&s, pDb->pEnv);
+
+    /* Append each key to string s. */
+    for(i=0; i<3; i++){
+      u32 iPtr = pNode->aiKeyPtr[i];
+      if( iPtr ){
+        TreeKey *pKey = treeShmkey(pDb, pNode->aiKeyPtr[i],TKV_LOADKEY, &b,&rc);
+        strAppendFlags(&s, pKey->flags);
+        lsmAppendStrBlob(&s, TKV_KEY(pKey), pKey->nKey);
+        lsmStringAppend(&s, "     ", -1);
+      }
+    }
+
+    printf("% 6d %.*sleaf%.*s: %s\n", 
+        iNode, nPath, zPath, 20-nPath-4, zSpace, s.z
+    );
+    lsmStringClear(&s);
+  }else{
+    for(i=0; i<4 && nHeight>0; i++){
+      u32 iPtr = getChildPtr(pNode, pDb->treehdr.root.iTransId, i);
+      zPath[nPath] = (char)(i+'0');
+      zPath[nPath+1] = '/';
+
+      if( iPtr ){
+        dump_node_contents(pDb, iPtr, zPath, nPath+2, nHeight-1);
+      }
+      if( i!=3 && pNode->aiKeyPtr[i] ){
+        TreeKey *pKey = treeShmkey(pDb, pNode->aiKeyPtr[i], TKV_LOADKEY,&b,&rc);
+        lsmStringInit(&s, pDb->pEnv);
+        strAppendFlags(&s, pKey->flags);
+        lsmAppendStrBlob(&s, TKV_KEY(pKey), pKey->nKey);
+        printf("% 6d %.*s%.*s: %s\n", 
+            iNode, nPath+1, zPath, 20-nPath-1, zSpace, s.z);
+        lsmStringClear(&s);
+      }
+    }
+  }
+
+  tblobFree(pDb, &b);
+}
+
+void dump_tree_contents(lsm_db *pDb, const char *zCaption){
+  char zPath[64];
+  TreeRoot *p = &pDb->treehdr.root;
+  printf("\n%s\n", zCaption);
+  zPath[0] = '/';
+  if( p->iRoot ){
+    dump_node_contents(pDb, p->iRoot, zPath, 1, p->nHeight-1);
+  }
+  fflush(stdout);
+}
+
+#endif
+
+/*
+** Initialize a cursor object, the space for which has already been
+** allocated.
+*/
+static void treeCursorInit(lsm_db *pDb, int bOld, TreeCursor *pCsr){
+  memset(pCsr, 0, sizeof(TreeCursor));
+  pCsr->pDb = pDb;
+  if( bOld ){
+    pCsr->pRoot = &pDb->treehdr.oldroot;
+  }else{
+    pCsr->pRoot = &pDb->treehdr.root;
+  }
+  pCsr->iNode = -1;
+}
+
+/*
+** Return a pointer to the mapping of the TreeKey object that the cursor
+** is pointing to. 
+*/
+static TreeKey *csrGetKey(TreeCursor *pCsr, TreeBlob *pBlob, int *pRc){
+  TreeKey *pRet;
+  lsm_db *pDb = pCsr->pDb;
+  u32 iPtr = pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[pCsr->aiCell[pCsr->iNode]];
+
+  assert( iPtr );
+  pRet = (TreeKey*)treeShmptrUnsafe(pDb, iPtr);
+  if( !(pRet->flags & LSM_CONTIGUOUS) ){
+    pRet = treeShmkey(pDb, iPtr, TKV_LOADVAL, pBlob, pRc);
+  }
+
+  return pRet;
+}
+
+/*
+** Save the current position of tree cursor pCsr.
+*/
+int lsmTreeCursorSave(TreeCursor *pCsr){
+  int rc = LSM_OK;
+  if( pCsr && pCsr->pSave==0 ){
+    int iNode = pCsr->iNode;
+    if( iNode>=0 ){
+      pCsr->pSave = csrGetKey(pCsr, &pCsr->blob, &rc);
+    }
+    pCsr->iNode = -1;
+  }
+  return rc;
+}
+
+/*
+** Restore the position of a saved tree cursor.
+*/
+static int treeCursorRestore(TreeCursor *pCsr, int *pRes){
+  int rc = LSM_OK;
+  if( pCsr->pSave ){
+    TreeKey *pKey = pCsr->pSave;
+    pCsr->pSave = 0;
+    if( pRes ){
+      rc = lsmTreeCursorSeek(pCsr, TKV_KEY(pKey), pKey->nKey, pRes);
+    }
+  }
+  return rc;
+}
+
+/*
+** Allocate nByte bytes of space within the *-shm file. If successful, 
+** return LSM_OK and set *piPtr to the offset within the file at which
+** the allocated space is located.
+*/
+static u32 treeShmalloc(lsm_db *pDb, int bAlign, int nByte, int *pRc){
+  u32 iRet = 0;
+  if( *pRc==LSM_OK ){
+    const static int CHUNK_SIZE = LSM_SHM_CHUNK_SIZE;
+    const static int CHUNK_HDR = LSM_SHM_CHUNK_HDR;
+    u32 iWrite;                   /* Current write offset */
+    u32 iEof;                     /* End of current chunk */
+    int iChunk;                   /* Current chunk */
+
+    assert( nByte <= (CHUNK_SIZE-CHUNK_HDR) );
+
+    /* Check if there is enough space on the current chunk to fit the
+    ** new allocation. If not, link in a new chunk and put the new
+    ** allocation at the start of it.  */
+    iWrite = pDb->treehdr.iWrite;
+    if( bAlign ){
+      iWrite = (iWrite + 3) & ~0x0003;
+      assert( (iWrite % 4)==0 );
+    }
+
+    assert( iWrite );
+    iChunk = treeOffsetToChunk(iWrite-1);
+    iEof = (iChunk+1) * CHUNK_SIZE;
+    assert( iEof>=iWrite && (iEof-iWrite)<(u32)CHUNK_SIZE );
+    if( (iWrite+nByte)>iEof ){
+      ShmChunk *pHdr;           /* Header of chunk just finished (iChunk) */
+      ShmChunk *pFirst;         /* Header of chunk treehdr.iFirst */
+      ShmChunk *pNext;          /* Header of new chunk */
+      int iNext = 0;            /* Next chunk */
+      int rc = LSM_OK;
+
+      pFirst = treeShmChunk(pDb, pDb->treehdr.iFirst);
+
+      assert( shm_sequence_ge(pDb->treehdr.iUsedShmid, pFirst->iShmid) );
+      assert( (pDb->treehdr.iNextShmid+1-pDb->treehdr.nChunk)==pFirst->iShmid );
+
+      /* Check if the chunk at the start of the linked list is still in
+      ** use. If not, reuse it. If so, allocate a new chunk by appending
+      ** to the *-shm file.  */
+      if( pDb->treehdr.iUsedShmid!=pFirst->iShmid ){
+        int bInUse;
+        rc = lsmTreeInUse(pDb, pFirst->iShmid, &bInUse);
+        if( rc!=LSM_OK ){
+          *pRc = rc;
+          return 0;
+        }
+        if( bInUse==0 ){
+          iNext = pDb->treehdr.iFirst;
+          pDb->treehdr.iFirst = pFirst->iNext;
+          assert( pDb->treehdr.iFirst );
+        }
+      }
+      if( iNext==0 ) iNext = pDb->treehdr.nChunk++;
+
+      /* Set the header values for the new chunk */
+      pNext = treeShmChunkRc(pDb, iNext, &rc);
+      if( pNext ){
+        pNext->iNext = 0;
+        pNext->iShmid = (pDb->treehdr.iNextShmid++);
+      }else{
+        *pRc = rc;
+        return 0;
+      }
+
+      /* Set the header values for the chunk just finished */
+      pHdr = (ShmChunk *)treeShmptr(pDb, iChunk*CHUNK_SIZE);
+      pHdr->iNext = iNext;
+
+      /* Advance to the next chunk */
+      iWrite = iNext * CHUNK_SIZE + CHUNK_HDR;
+    }
+
+    /* Allocate space at iWrite. */
+    iRet = iWrite;
+    pDb->treehdr.iWrite = iWrite + nByte;
+    pDb->treehdr.root.nByte += nByte;
+  }
+  return iRet;
+}
+
+/*
+** Allocate and zero nByte bytes of space within the *-shm file.
+*/
+static void *treeShmallocZero(lsm_db *pDb, int nByte, u32 *piPtr, int *pRc){
+  u32 iPtr;
+  void *p;
+  iPtr = treeShmalloc(pDb, 1, nByte, pRc);
+  p = treeShmptr(pDb, iPtr);
+  if( p ){
+    assert( *pRc==LSM_OK );
+    memset(p, 0, nByte);
+    *piPtr = iPtr;
+  }
+  return p;
+}
+
+static TreeNode *newTreeNode(lsm_db *pDb, u32 *piPtr, int *pRc){
+  return treeShmallocZero(pDb, sizeof(TreeNode), piPtr, pRc);
+}
+
+static TreeLeaf *newTreeLeaf(lsm_db *pDb, u32 *piPtr, int *pRc){
+  return treeShmallocZero(pDb, sizeof(TreeLeaf), piPtr, pRc);
+}
+
+static TreeKey *newTreeKey(
+  lsm_db *pDb, 
+  u32 *piPtr, 
+  void *pKey, int nKey,           /* Key data */
+  void *pVal, int nVal,           /* Value data (or nVal<0 for delete) */
+  int *pRc
+){
+  TreeKey *p;
+  u32 iPtr;
+  u32 iEnd;
+  int nRem;
+  u8 *a;
+  int n;
+
+  /* Allocate space for the TreeKey structure itself */
+  *piPtr = iPtr = treeShmalloc(pDb, 1, sizeof(TreeKey), pRc);
+  p = treeShmptr(pDb, iPtr);
+  if( *pRc ) return 0;
+  p->nKey = nKey;
+  p->nValue = nVal;
+
+  /* Allocate and populate the space required for the key and value. */
+  n = nRem = nKey;
+  a = (u8 *)pKey;
+  while( a ){
+    while( nRem>0 ){
+      u8 *aAlloc;
+      int nAlloc;
+      u32 iWrite;
+
+      iWrite = (pDb->treehdr.iWrite & (LSM_SHM_CHUNK_SIZE-1));
+      iWrite = LSM_MAX(iWrite, LSM_SHM_CHUNK_HDR);
+      nAlloc = LSM_MIN((LSM_SHM_CHUNK_SIZE-iWrite), (u32)nRem);
+
+      aAlloc = treeShmptr(pDb, treeShmalloc(pDb, 0, nAlloc, pRc));
+      if( aAlloc==0 ) break;
+      memcpy(aAlloc, &a[n-nRem], nAlloc);
+      nRem -= nAlloc;
+    }
+    a = pVal;
+    n = nRem = nVal;
+    pVal = 0;
+  }
+
+  iEnd = iPtr + sizeof(TreeKey) + nKey + LSM_MAX(0, nVal);
+  if( (iPtr & ~(LSM_SHM_CHUNK_SIZE-1))!=(iEnd & ~(LSM_SHM_CHUNK_SIZE-1)) ){
+    p->flags = 0;
+  }else{
+    p->flags = LSM_CONTIGUOUS;
+  }
+
+  if( *pRc ) return 0;
+#if 0
+  printf("store: %d %s\n", (int)iPtr, (char *)pKey);
+#endif
+  return p;
+}
+
+static TreeNode *copyTreeNode(
+  lsm_db *pDb, 
+  TreeNode *pOld, 
+  u32 *piNew, 
+  int *pRc
+){
+  TreeNode *pNew;
+
+  pNew = newTreeNode(pDb, piNew, pRc);
+  if( pNew ){
+    memcpy(pNew->aiKeyPtr, pOld->aiKeyPtr, sizeof(pNew->aiKeyPtr));
+    memcpy(pNew->aiChildPtr, pOld->aiChildPtr, sizeof(pNew->aiChildPtr));
+    if( pOld->iV2 ) pNew->aiChildPtr[pOld->iV2Child] = pOld->iV2Ptr;
+  }
+  return pNew;
+}
+
+static TreeNode *copyTreeLeaf(
+  lsm_db *pDb, 
+  TreeLeaf *pOld, 
+  u32 *piNew, 
+  int *pRc
+){
+  TreeLeaf *pNew;
+  pNew = newTreeLeaf(pDb, piNew, pRc);
+  if( pNew ){
+    memcpy(pNew, pOld, sizeof(TreeLeaf));
+  }
+  return (TreeNode *)pNew;
+}
+
+/*
+** The tree cursor passed as the second argument currently points to an 
+** internal node (not a leaf). Specifically, to a sub-tree pointer. This
+** function replaces the sub-tree that the cursor currently points to
+** with sub-tree pNew.
+**
+** The sub-tree may be replaced either by writing the "v2 data" on the
+** internal node, or by allocating a new TreeNode structure and then 
+** calling this function on the parent of the internal node.
+*/
+static int treeUpdatePtr(lsm_db *pDb, TreeCursor *pCsr, u32 iNew){
+  int rc = LSM_OK;
+  if( pCsr->iNode<0 ){
+    /* iNew is the new root node */
+    pDb->treehdr.root.iRoot = iNew;
+  }else{
+    /* If this node already has version 2 content, allocate a copy and
+    ** update the copy with the new pointer value. Otherwise, store the
+    ** new pointer as v2 data within the current node structure.  */
+
+    TreeNode *p;                  /* The node to be modified */
+    int iChildPtr;                /* apChild[] entry to modify */
+
+    p = pCsr->apTreeNode[pCsr->iNode];
+    iChildPtr = pCsr->aiCell[pCsr->iNode];
+
+    if( p->iV2 ){
+      /* The "allocate new TreeNode" option */
+      u32 iCopy;
+      TreeNode *pCopy;
+      pCopy = copyTreeNode(pDb, p, &iCopy, &rc);
+      if( pCopy ){
+        assert( rc==LSM_OK );
+        pCopy->aiChildPtr[iChildPtr] = iNew;
+        pCsr->iNode--;
+        rc = treeUpdatePtr(pDb, pCsr, iCopy);
+      }
+    }else{
+      /* The "v2 data" option */
+      u32 iPtr;
+      assert( pDb->treehdr.root.iTransId>0 );
+
+      if( pCsr->iNode ){
+        iPtr = getChildPtr(
+            pCsr->apTreeNode[pCsr->iNode-1], 
+            pDb->treehdr.root.iTransId, pCsr->aiCell[pCsr->iNode-1]
+        );
+      }else{
+        iPtr = pDb->treehdr.root.iRoot;
+      }
+      rc = intArrayAppend(pDb->pEnv, &pDb->rollback, iPtr);
+
+      if( rc==LSM_OK ){
+        p->iV2 = pDb->treehdr.root.iTransId;
+        p->iV2Child = (u8)iChildPtr;
+        p->iV2Ptr = iNew;
+      }
+    }
+  }
+
+  return rc;
+}
+
+/*
+** Cursor pCsr points at a node that is part of pTree. This function
+** inserts a new key and optionally child node pointer into that node.
+**
+** The position into which the new key and pointer are inserted is
+** determined by the iSlot parameter. The new key will be inserted to
+** the left of the key currently stored in apKey[iSlot]. Or, if iSlot is
+** greater than the index of the rightmost key in the node.
+**
+** Pointer pLeftPtr points to a child tree that contains keys that are
+** smaller than pTreeKey.
+*/
+static int treeInsert(
+  lsm_db *pDb,                    /* Database handle */
+  TreeCursor *pCsr,               /* Cursor indicating path to insert at */
+  u32 iLeftPtr,                   /* Left child pointer */
+  u32 iTreeKey,                   /* Location of key to insert */
+  u32 iRightPtr,                  /* Right child pointer */
+  int iSlot                       /* Position to insert key into */
+){
+  int rc = LSM_OK;
+  TreeNode *pNode = pCsr->apTreeNode[pCsr->iNode];
+
+  /* Check if the node is currently full. If so, split pNode in two and
+  ** call this function recursively to add a key to the parent. Otherwise, 
+  ** insert the new key directly into pNode.  */
+  assert( pNode->aiKeyPtr[1] );
+  if( pNode->aiKeyPtr[0] && pNode->aiKeyPtr[2] ){
+    u32 iLeft; TreeNode *pLeft;   /* New left-hand sibling node */
+    u32 iRight; TreeNode *pRight; /* New right-hand sibling node */
+
+    pLeft = newTreeNode(pDb, &iLeft, &rc);
+    pRight = newTreeNode(pDb, &iRight, &rc);
+    if( rc ) return rc;
+
+    pLeft->aiChildPtr[1] = getChildPtr(pNode, WORKING_VERSION, 0);
+    pLeft->aiKeyPtr[1] = pNode->aiKeyPtr[0];
+    pLeft->aiChildPtr[2] = getChildPtr(pNode, WORKING_VERSION, 1);
+
+    pRight->aiChildPtr[1] = getChildPtr(pNode, WORKING_VERSION, 2);
+    pRight->aiKeyPtr[1] = pNode->aiKeyPtr[2];
+    pRight->aiChildPtr[2] = getChildPtr(pNode, WORKING_VERSION, 3);
+
+    if( pCsr->iNode==0 ){
+      /* pNode is the root of the tree. Grow the tree by one level. */
+      u32 iRoot; TreeNode *pRoot; /* New root node */
+
+      pRoot = newTreeNode(pDb, &iRoot, &rc);
+      pRoot->aiKeyPtr[1] = pNode->aiKeyPtr[1];
+      pRoot->aiChildPtr[1] = iLeft;
+      pRoot->aiChildPtr[2] = iRight;
+
+      pDb->treehdr.root.iRoot = iRoot;
+      pDb->treehdr.root.nHeight++;
+    }else{
+
+      pCsr->iNode--;
+      rc = treeInsert(pDb, pCsr, 
+          iLeft, pNode->aiKeyPtr[1], iRight, pCsr->aiCell[pCsr->iNode]
+      );
+    }
+
+    assert( pLeft->iV2==0 );
+    assert( pRight->iV2==0 );
+    switch( iSlot ){
+      case 0:
+        pLeft->aiKeyPtr[0] = iTreeKey;
+        pLeft->aiChildPtr[0] = iLeftPtr;
+        if( iRightPtr ) pLeft->aiChildPtr[1] = iRightPtr;
+        break;
+      case 1:
+        pLeft->aiChildPtr[3] = (iRightPtr ? iRightPtr : pLeft->aiChildPtr[2]);
+        pLeft->aiKeyPtr[2] = iTreeKey;
+        pLeft->aiChildPtr[2] = iLeftPtr;
+        break;
+      case 2:
+        pRight->aiKeyPtr[0] = iTreeKey;
+        pRight->aiChildPtr[0] = iLeftPtr;
+        if( iRightPtr ) pRight->aiChildPtr[1] = iRightPtr;
+        break;
+      case 3:
+        pRight->aiChildPtr[3] = (iRightPtr ? iRightPtr : pRight->aiChildPtr[2]);
+        pRight->aiKeyPtr[2] = iTreeKey;
+        pRight->aiChildPtr[2] = iLeftPtr;
+        break;
+    }
+
+  }else{
+    TreeNode *pNew;
+    u32 *piKey;
+    u32 *piChild;
+    u32 iStore = 0;
+    u32 iNew = 0;
+    int i;
+
+    /* Allocate a new version of node pNode. */
+    pNew = newTreeNode(pDb, &iNew, &rc);
+    if( rc ) return rc;
+
+    piKey = pNew->aiKeyPtr;
+    piChild = pNew->aiChildPtr;
+
+    for(i=0; i<iSlot; i++){
+      if( pNode->aiKeyPtr[i] ){
+        *(piKey++) = pNode->aiKeyPtr[i];
+        *(piChild++) = getChildPtr(pNode, WORKING_VERSION, i);
+      }
+    }
+
+    *piKey++ = iTreeKey;
+    *piChild++ = iLeftPtr;
+
+    iStore = iRightPtr;
+    for(i=iSlot; i<3; i++){
+      if( pNode->aiKeyPtr[i] ){
+        *(piKey++) = pNode->aiKeyPtr[i];
+        *(piChild++) = iStore ? iStore : getChildPtr(pNode, WORKING_VERSION, i);
+        iStore = 0;
+      }
+    }
+
+    if( iStore ){
+      *piChild = iStore;
+    }else{
+      *piChild = getChildPtr(pNode, WORKING_VERSION, 
+          (pNode->aiKeyPtr[2] ? 3 : 2)
+      );
+    }
+    pCsr->iNode--;
+    rc = treeUpdatePtr(pDb, pCsr, iNew);
+  }
+
+  return rc;
+}
+
+static int treeInsertLeaf(
+  lsm_db *pDb,                    /* Database handle */
+  TreeCursor *pCsr,               /* Cursor structure */
+  u32 iTreeKey,                   /* Key pointer to insert */
+  int iSlot                       /* Insert key to the left of this */
+){
+  int rc = LSM_OK;                /* Return code */
+  TreeNode *pLeaf = pCsr->apTreeNode[pCsr->iNode];
+  TreeLeaf *pNew;
+  u32 iNew;
+
+  assert( iSlot>=0 && iSlot<=4 );
+  assert( pCsr->iNode>0 );
+  assert( pLeaf->aiKeyPtr[1] );
+
+  pCsr->iNode--;
+
+  pNew = newTreeLeaf(pDb, &iNew, &rc);
+  if( pNew ){
+    if( pLeaf->aiKeyPtr[0] && pLeaf->aiKeyPtr[2] ){
+      /* The leaf is full. Split it in two. */
+      TreeLeaf *pRight;
+      u32 iRight;
+      pRight = newTreeLeaf(pDb, &iRight, &rc);
+      if( pRight ){
+        assert( rc==LSM_OK );
+        pNew->aiKeyPtr[1] = pLeaf->aiKeyPtr[0];
+        pRight->aiKeyPtr[1] = pLeaf->aiKeyPtr[2];
+        switch( iSlot ){
+          case 0: pNew->aiKeyPtr[0] = iTreeKey; break;
+          case 1: pNew->aiKeyPtr[2] = iTreeKey; break;
+          case 2: pRight->aiKeyPtr[0] = iTreeKey; break;
+          case 3: pRight->aiKeyPtr[2] = iTreeKey; break;
+        }
+
+        rc = treeInsert(pDb, pCsr, iNew, pLeaf->aiKeyPtr[1], iRight, 
+            pCsr->aiCell[pCsr->iNode]
+        );
+      }
+    }else{
+      int iOut = 0;
+      int i;
+      for(i=0; i<4; i++){
+        if( i==iSlot ) pNew->aiKeyPtr[iOut++] = iTreeKey;
+        if( i<3 && pLeaf->aiKeyPtr[i] ){
+          pNew->aiKeyPtr[iOut++] = pLeaf->aiKeyPtr[i];
+        }
+      }
+      rc = treeUpdatePtr(pDb, pCsr, iNew);
+    }
+  }
+
+  return rc;
+}
+
+void lsmTreeMakeOld(lsm_db *pDb){
+
+  /* A write transaction must be open. Otherwise the code below that
+  ** assumes (pDb->pClient->iLogOff) is current may malfunction. 
+  **
+  ** Update: currently this assert fails due to lsm_flush(), which does
+  ** not set nTransOpen.
+  */
+  assert( /* pDb->nTransOpen>0 && */ pDb->iReader>=0 );
+
+  if( pDb->treehdr.iOldShmid==0 ){
+    pDb->treehdr.iOldLog = (pDb->treehdr.log.aRegion[2].iEnd << 1);
+    pDb->treehdr.iOldLog |= (~(pDb->pClient->iLogOff) & (i64)0x0001);
+
+    pDb->treehdr.oldcksum0 = pDb->treehdr.log.cksum0;
+    pDb->treehdr.oldcksum1 = pDb->treehdr.log.cksum1;
+    pDb->treehdr.iOldShmid = pDb->treehdr.iNextShmid-1;
+    memcpy(&pDb->treehdr.oldroot, &pDb->treehdr.root, sizeof(TreeRoot));
+
+    pDb->treehdr.root.iTransId = 1;
+    pDb->treehdr.root.iRoot = 0;
+    pDb->treehdr.root.nHeight = 0;
+    pDb->treehdr.root.nByte = 0;
+  }
+}
+
+void lsmTreeDiscardOld(lsm_db *pDb){
+  assert( lsmShmAssertLock(pDb, LSM_LOCK_WRITER, LSM_LOCK_EXCL) 
+       || lsmShmAssertLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_EXCL) 
+  );
+  pDb->treehdr.iUsedShmid = pDb->treehdr.iOldShmid;
+  pDb->treehdr.iOldShmid = 0;
+}
+
+int lsmTreeHasOld(lsm_db *pDb){
+  return pDb->treehdr.iOldShmid!=0;
+}
+
+/*
+** This function is called during recovery to initialize the 
+** tree header. Only the database connections private copy of the tree-header
+** is initialized here - it will be copied into shared memory if log file
+** recovery is successful.
+*/
+int lsmTreeInit(lsm_db *pDb){
+  ShmChunk *pOne;
+  int rc = LSM_OK;
+
+  memset(&pDb->treehdr, 0, sizeof(TreeHeader));
+  pDb->treehdr.root.iTransId = 1;
+  pDb->treehdr.iFirst = 1;
+  pDb->treehdr.nChunk = 2;
+  pDb->treehdr.iWrite = LSM_SHM_CHUNK_SIZE + LSM_SHM_CHUNK_HDR;
+  pDb->treehdr.iNextShmid = 2;
+  pDb->treehdr.iUsedShmid = 1;
+
+  pOne = treeShmChunkRc(pDb, 1, &rc);
+  if( pOne ){
+    pOne->iNext = 0;
+    pOne->iShmid = 1;
+  }
+  return rc;
+}
+
+static void treeHeaderChecksum(
+  TreeHeader *pHdr, 
+  u32 *aCksum
+){
+  u32 cksum1 = 0x12345678;
+  u32 cksum2 = 0x9ABCDEF0;
+  u32 *a = (u32 *)pHdr;
+  int i;
+
+  assert( (offsetof(TreeHeader, aCksum) + sizeof(u32)*2)==sizeof(TreeHeader) );
+  assert( (sizeof(TreeHeader) % (sizeof(u32)*2))==0 );
+
+  for(i=0; i<(offsetof(TreeHeader, aCksum) / sizeof(u32)); i+=2){
+    cksum1 += a[i];
+    cksum2 += (cksum1 + a[i+1]);
+  }
+  aCksum[0] = cksum1;
+  aCksum[1] = cksum2;
+}
+
+/*
+** Return true if the checksum stored in TreeHeader object *pHdr is 
+** consistent with the contents of its other fields.
+*/
+static int treeHeaderChecksumOk(TreeHeader *pHdr){
+  u32 aCksum[2];
+  treeHeaderChecksum(pHdr, aCksum);
+  return (0==memcmp(aCksum, pHdr->aCksum, sizeof(aCksum)));
+}
+
+/*
+** This type is used by functions lsmTreeRepair() and treeSortByShmid() to
+** make relinking the linked list of shared-memory chunks easier.
+*/
+typedef struct ShmChunkLoc ShmChunkLoc;
+struct ShmChunkLoc {
+  ShmChunk *pShm;
+  u32 iLoc;
+};
+
+/*
+** This function checks that the linked list of shared memory chunks 
+** that starts at chunk db->treehdr.iFirst:
+**
+**   1) Includes all chunks in the shared-memory region, and
+**   2) Links them together in order of ascending shm-id.
+**
+** If no error occurs and the conditions above are met, LSM_OK is returned.
+**
+** If either of the conditions are untrue, LSM_CORRUPT is returned. Or, if
+** an error is encountered before the checks are completed, another LSM error
+** code (i.e. LSM_IOERR or LSM_NOMEM) may be returned.
+*/
+static int treeCheckLinkedList(lsm_db *db){
+  int rc = LSM_OK;
+  int nVisit = 0;
+  ShmChunk *p;
+
+  p = treeShmChunkRc(db, db->treehdr.iFirst, &rc);
+  while( rc==LSM_OK && p ){
+    if( p->iNext ){
+      if( p->iNext>=db->treehdr.nChunk ){
+        rc = LSM_CORRUPT_BKPT;
+      }else{
+        ShmChunk *pNext = treeShmChunkRc(db, p->iNext, &rc);
+        if( rc==LSM_OK ){
+          if( pNext->iShmid!=p->iShmid+1 ){
+            rc = LSM_CORRUPT_BKPT;
+          }
+          p = pNext;
+        }
+      }
+    }else{
+      p = 0;
+    }
+    nVisit++;
+  }
+
+  if( rc==LSM_OK && (u32)nVisit!=db->treehdr.nChunk-1 ){
+    rc = LSM_CORRUPT_BKPT;
+  }
+  return rc;
+}
+
+/*
+** Iterate through the current in-memory tree. If there are any v2-pointers
+** with transaction ids larger than db->treehdr.iTransId, zero them.
+*/
+static int treeRepairPtrs(lsm_db *db){
+  int rc = LSM_OK;
+
+  if( db->treehdr.root.nHeight>1 ){
+    TreeCursor csr;               /* Cursor used to iterate through tree */
+    u32 iTransId = db->treehdr.root.iTransId;
+
+    /* Initialize the cursor structure. Also decrement the nHeight variable
+    ** in the tree-header. This will prevent the cursor from visiting any
+    ** leaf nodes.  */
+    db->treehdr.root.nHeight--;
+    treeCursorInit(db, 0, &csr);
+
+    rc = lsmTreeCursorEnd(&csr, 0);
+    while( rc==LSM_OK && lsmTreeCursorValid(&csr) ){
+      TreeNode *pNode = csr.apTreeNode[csr.iNode];
+      if( pNode->iV2>iTransId ){
+        pNode->iV2Child = 0;
+        pNode->iV2Ptr = 0;
+        pNode->iV2 = 0;
+      }
+      rc = lsmTreeCursorNext(&csr);
+    }
+    tblobFree(csr.pDb, &csr.blob);
+
+    db->treehdr.root.nHeight++;
+  }
+
+  return rc;
+}
+
+static int treeRepairList(lsm_db *db){
+  int rc = LSM_OK;
+  int i;
+  ShmChunk *p;
+  ShmChunk *pMin = 0;
+  u32 iMin = 0;
+
+  /* Iterate through all shm chunks. Find the smallest shm-id present in
+  ** the shared-memory region. */
+  for(i=1; rc==LSM_OK && (u32)i<db->treehdr.nChunk; i++){
+    p = treeShmChunkRc(db, i, &rc);
+    if( p && (pMin==0 || shm_sequence_ge(pMin->iShmid, p->iShmid)) ){
+      pMin = p;
+      iMin = i;
+    }
+  }
+
+  /* Fix the shm-id values on any chunks with a shm-id greater than or 
+  ** equal to treehdr.iNextShmid. Then do a merge-sort of all chunks to 
+  ** fix the ShmChunk.iNext pointers.
+  */
+  if( rc==LSM_OK ){
+    int nSort;
+    int nByte;
+    u32 iPrevShmid;
+    ShmChunkLoc *aSort;
+
+    /* Allocate space for a merge sort. */
+    nSort = 1;
+    while( (u32)nSort < (db->treehdr.nChunk-1) ) nSort = nSort * 2;
+    nByte = sizeof(ShmChunkLoc) * nSort * 2;
+    aSort = lsmMallocZeroRc(db->pEnv, nByte, &rc);
+    iPrevShmid = pMin->iShmid;
+
+    /* Fix all shm-ids, if required. */
+    if( rc==LSM_OK ){
+      iPrevShmid = pMin->iShmid-1;
+      for(i=1; (u32)i<db->treehdr.nChunk; i++){
+        p = treeShmChunk(db, i);
+        aSort[i-1].pShm = p;
+        aSort[i-1].iLoc = i;
+        if( (u32)i!=db->treehdr.iFirst ){
+          if( shm_sequence_ge(p->iShmid, db->treehdr.iNextShmid) ){
+            p->iShmid = iPrevShmid--;
+          }
+        }
+      }
+      if( iMin!=db->treehdr.iFirst ){
+        p = treeShmChunk(db, db->treehdr.iFirst);
+        p->iShmid = iPrevShmid;
+      }
+    }
+
+    if( rc==LSM_OK ){
+      ShmChunkLoc *aSpace = &aSort[nSort];
+      for(i=0; i<nSort; i++){
+        if( aSort[i].pShm ){
+          assert( shm_sequence_ge(aSort[i].pShm->iShmid, iPrevShmid) );
+          assert( aSpace[aSort[i].pShm->iShmid - iPrevShmid].pShm==0 );
+          aSpace[aSort[i].pShm->iShmid - iPrevShmid] = aSort[i];
+        }
+      }
+
+      if( aSpace[nSort-1].pShm ) aSpace[nSort-1].pShm->iNext = 0;
+      for(i=0; i<nSort-1; i++){
+        if( aSpace[i].pShm ){
+          aSpace[i].pShm->iNext = aSpace[i+1].iLoc;
+        }
+      }
+
+      rc = treeCheckLinkedList(db);
+      lsmFree(db->pEnv, aSort);
+    }
+  }
+
+  return rc;
+}
+
+/*
+** This function is called as part of opening a write-transaction if the
+** writer-flag is already set - indicating that the previous writer 
+** failed before ending its transaction.
+*/
+int lsmTreeRepair(lsm_db *db){
+  int rc = LSM_OK;
+  TreeHeader hdr;
+  ShmHeader *pHdr = db->pShmhdr;
+
+  /* Ensure that the two tree-headers are consistent. Copy one over the other
+  ** if necessary. Prefer the data from a tree-header for which the checksum
+  ** computes. Or, if they both compute, prefer tree-header-1.  */
+  if( memcmp(&pHdr->hdr1, &pHdr->hdr2, sizeof(TreeHeader)) ){
+    if( treeHeaderChecksumOk(&pHdr->hdr1) ){
+      memcpy(&pHdr->hdr2, &pHdr->hdr1, sizeof(TreeHeader));
+    }else{
+      memcpy(&pHdr->hdr1, &pHdr->hdr2, sizeof(TreeHeader));
+    }
+  }
+
+  /* Save the connections current copy of the tree-header. It will be 
+  ** restored before returning.  */
+  memcpy(&hdr, &db->treehdr, sizeof(TreeHeader));
+
+  /* Walk the tree. Zero any v2 pointers with a transaction-id greater than
+  ** the transaction-id currently in the tree-headers.  */
+  rc = treeRepairPtrs(db);
+
+  /* Repair the linked list of shared-memory chunks. */
+  if( rc==LSM_OK ){
+    rc = treeRepairList(db);
+  }
+
+  memcpy(&db->treehdr, &hdr, sizeof(TreeHeader));
+  return rc;
+}
+
+static void treeOverwriteKey(lsm_db *db, TreeCursor *pCsr, u32 iKey, int *pRc){
+  if( *pRc==LSM_OK ){
+    TreeRoot *p = &db->treehdr.root;
+    TreeNode *pNew;
+    u32 iNew;
+    TreeNode *pNode = pCsr->apTreeNode[pCsr->iNode];
+    int iCell = pCsr->aiCell[pCsr->iNode];
+
+    /* Create a copy of this node */
+    if( (pCsr->iNode>0 && (u32)pCsr->iNode==(p->nHeight-1)) ){
+      pNew = copyTreeLeaf(db, (TreeLeaf *)pNode, &iNew, pRc);
+    }else{
+      pNew = copyTreeNode(db, pNode, &iNew, pRc);
+    }
+
+    if( pNew ){
+      /* Modify the value in the new version */
+      pNew->aiKeyPtr[iCell] = iKey;
+
+      /* Change the pointer in the parent (if any) to point at the new 
+       ** TreeNode */
+      pCsr->iNode--;
+      treeUpdatePtr(db, pCsr, iNew);
+    }
+  }
+}
+
+static int treeNextIsEndDelete(lsm_db *db, TreeCursor *pCsr){
+  int iNode = pCsr->iNode;
+  int iCell = pCsr->aiCell[iNode]+1;
+
+  /* Cursor currently points to a leaf node. */
+  assert( (u32)pCsr->iNode==(db->treehdr.root.nHeight-1) );
+
+  while( iNode>=0 ){
+    TreeNode *pNode = pCsr->apTreeNode[iNode];
+    if( iCell<3 && pNode->aiKeyPtr[iCell] ){
+      int rc = LSM_OK;
+      TreeKey *pKey = treeShmptr(db, pNode->aiKeyPtr[iCell]);
+      assert( rc==LSM_OK );
+      return ((pKey->flags & LSM_END_DELETE) ? 1 : 0);
+    }
+    iNode--;
+    iCell = pCsr->aiCell[iNode];
+  }
+
+  return 0;
+}
+
+static int treePrevIsStartDelete(lsm_db *db, TreeCursor *pCsr){
+  int iNode = pCsr->iNode;
+
+  /* Cursor currently points to a leaf node. */
+  assert( (u32)pCsr->iNode==(db->treehdr.root.nHeight-1) );
+
+  while( iNode>=0 ){
+    TreeNode *pNode = pCsr->apTreeNode[iNode];
+    int iCell = pCsr->aiCell[iNode]-1;
+    if( iCell>=0 && pNode->aiKeyPtr[iCell] ){
+      int rc = LSM_OK;
+      TreeKey *pKey = treeShmptr(db, pNode->aiKeyPtr[iCell]);
+      assert( rc==LSM_OK );
+      return ((pKey->flags & LSM_START_DELETE) ? 1 : 0);
+    }
+    iNode--;
+  }
+
+  return 0;
+}
+
+
+static int treeInsertEntry(
+  lsm_db *pDb,                    /* Database handle */
+  int flags,                      /* Flags associated with entry */
+  void *pKey,                     /* Pointer to key data */
+  int nKey,                       /* Size of key data in bytes */
+  void *pVal,                     /* Pointer to value data (or NULL) */
+  int nVal                        /* Bytes in value data (or -ve for delete) */
+){
+  int rc = LSM_OK;                /* Return Code */
+  TreeKey *pTreeKey;              /* New key-value being inserted */
+  u32 iTreeKey;
+  TreeRoot *p = &pDb->treehdr.root;
+  TreeCursor csr;                 /* Cursor to seek to pKey/nKey */
+  int res = 0;                    /* Result of seek operation on csr */
+
+  assert( nVal>=0 || pVal==0 );
+  assert_tree_looks_ok(LSM_OK, pTree);
+  assert( flags==LSM_INSERT       || flags==LSM_POINT_DELETE 
+       || flags==LSM_START_DELETE || flags==LSM_END_DELETE 
+  );
+  assert( (flags & LSM_CONTIGUOUS)==0 );
+#if 0
+  dump_tree_contents(pDb, "before");
+#endif
+
+  if( p->iRoot ){
+    TreeKey *pRes;                /* Key at end of seek operation */
+    treeCursorInit(pDb, 0, &csr);
+
+    /* Seek to the leaf (or internal node) that the new key belongs on */
+    rc = lsmTreeCursorSeek(&csr, pKey, nKey, &res);
+    pRes = csrGetKey(&csr, &csr.blob, &rc);
+    if( rc!=LSM_OK ) return rc;
+    assert( pRes );
+
+    if( flags==LSM_START_DELETE ){
+      /* When inserting a start-delete-range entry, if the key that
+      ** occurs immediately before the new entry is already a START_DELETE,
+      ** then the new entry is not required.  */
+      if( (res<=0 && (pRes->flags & LSM_START_DELETE))
+       || (res>0  && treePrevIsStartDelete(pDb, &csr))
+      ){ 
+        goto insert_entry_out;
+      }
+    }else if( flags==LSM_END_DELETE ){
+      /* When inserting an start-delete-range entry, if the key that
+      ** occurs immediately after the new entry is already an END_DELETE,
+      ** then the new entry is not required.  */
+      if( (res<0  && treeNextIsEndDelete(pDb, &csr))
+       || (res>=0 && (pRes->flags & LSM_END_DELETE))
+      ){
+        goto insert_entry_out;
+      }
+    }
+
+    if( res==0 && (flags & (LSM_END_DELETE|LSM_START_DELETE)) ){
+      if( pRes->flags & LSM_INSERT ){
+        nVal = pRes->nValue;
+        pVal = TKV_VAL(pRes);
+      }
+      flags = flags | pRes->flags;
+    }
+
+    if( flags & (LSM_INSERT|LSM_POINT_DELETE) ){
+      if( (res<0 && (pRes->flags & LSM_START_DELETE))
+       || (res>0 && (pRes->flags & LSM_END_DELETE)) 
+      ){
+        flags = flags | (LSM_END_DELETE|LSM_START_DELETE);
+      }else if( res==0 ){
+        flags = flags | (pRes->flags & (LSM_END_DELETE|LSM_START_DELETE));
+      }
+    }
+  }else{
+    memset(&csr, 0, sizeof(TreeCursor));
+  }
+
+  /* Allocate and populate a new key-value pair structure */
+  pTreeKey = newTreeKey(pDb, &iTreeKey, pKey, nKey, pVal, nVal, &rc);
+  if( rc!=LSM_OK ) return rc;
+  assert( pTreeKey->flags==0 || pTreeKey->flags==LSM_CONTIGUOUS );
+  pTreeKey->flags |= flags;
+
+  if( p->iRoot==0 ){
+    /* The tree is completely empty. Add a new root node and install
+    ** (pKey/nKey) as the middle entry. Even though it is a leaf at the
+    ** moment, use newTreeNode() to allocate the node (i.e. allocate enough
+    ** space for the fields used by interior nodes). This is because the
+    ** treeInsert() routine may convert this node to an interior node. */
+    TreeNode *pRoot = newTreeNode(pDb, &p->iRoot, &rc);
+    if( rc==LSM_OK ){
+      assert( p->nHeight==0 );
+      pRoot->aiKeyPtr[1] = iTreeKey;
+      p->nHeight = 1;
+    }
+  }else{
+    if( res==0 ){
+      /* The search found a match within the tree. */
+      treeOverwriteKey(pDb, &csr, iTreeKey, &rc);
+    }else{
+      /* The cursor now points to the leaf node into which the new entry should
+      ** be inserted. There may or may not be a free slot within the leaf for
+      ** the new key-value pair. 
+      **
+      ** iSlot is set to the index of the key within pLeaf that the new key
+      ** should be inserted to the left of (or to a value 1 greater than the
+      ** index of the rightmost key if the new key is larger than all keys
+      ** currently stored in the node).
+      */
+      int iSlot = csr.aiCell[csr.iNode] + (res<0);
+      if( csr.iNode==0 ){
+        rc = treeInsert(pDb, &csr, 0, iTreeKey, 0, iSlot);
+      }else{
+        rc = treeInsertLeaf(pDb, &csr, iTreeKey, iSlot);
+      }
+    }
+  }
+
+#if 0
+  dump_tree_contents(pDb, "after");
+#endif
+ insert_entry_out:
+  tblobFree(pDb, &csr.blob);
+  assert_tree_looks_ok(rc, pTree);
+  return rc;
+}
+
+/*
+** Insert a new entry into the in-memory tree.
+**
+** If the value of the 5th parameter, nVal, is negative, then a delete-marker
+** is inserted into the tree. In this case the value pointer, pVal, must be
+** NULL.
+*/
+int lsmTreeInsert(
+  lsm_db *pDb,                    /* Database handle */
+  void *pKey,                     /* Pointer to key data */
+  int nKey,                       /* Size of key data in bytes */
+  void *pVal,                     /* Pointer to value data (or NULL) */
+  int nVal                        /* Bytes in value data (or -ve for delete) */
+){
+  int flags;
+  if( nVal<0 ){
+    flags = LSM_POINT_DELETE;
+  }else{
+    flags = LSM_INSERT;
+  }
+
+  return treeInsertEntry(pDb, flags, pKey, nKey, pVal, nVal);
+}
+
+static int treeDeleteEntry(lsm_db *db, TreeCursor *pCsr, u32 iNewptr){
+  TreeRoot *p = &db->treehdr.root;
+  TreeNode *pNode = pCsr->apTreeNode[pCsr->iNode];
+  int iSlot = pCsr->aiCell[pCsr->iNode];
+  int bLeaf;
+  int rc = LSM_OK;
+
+  assert( pNode->aiKeyPtr[1] );
+  assert( pNode->aiKeyPtr[iSlot] );
+  assert( iSlot==0 || iSlot==1 || iSlot==2 );
+  assert( ((u32)pCsr->iNode==(db->treehdr.root.nHeight-1))==(iNewptr==0) );
+
+  bLeaf = ((u32)pCsr->iNode==(p->nHeight-1) && p->nHeight>1);
+  
+  if( pNode->aiKeyPtr[0] || pNode->aiKeyPtr[2] ){
+    /* There are currently at least 2 keys on this node. So just create
+    ** a new copy of the node with one of the keys removed. If the node
+    ** happens to be the root node of the tree, allocate an entire 
+    ** TreeNode structure instead of just a TreeLeaf.  */
+    TreeNode *pNew;
+    u32 iNew;
+
+    if( bLeaf ){
+      pNew = (TreeNode *)newTreeLeaf(db, &iNew, &rc);
+    }else{
+      pNew = newTreeNode(db, &iNew, &rc);
+    }
+    if( pNew ){
+      int i;
+      int iOut = 1;
+      for(i=0; i<4; i++){
+        if( i==iSlot ){
+          i++;
+          if( bLeaf==0 ) pNew->aiChildPtr[iOut] = iNewptr;
+          if( i<3 ) pNew->aiKeyPtr[iOut] = pNode->aiKeyPtr[i];
+          iOut++;
+        }else if( bLeaf || p->nHeight==1 ){
+          if( i<3 && pNode->aiKeyPtr[i] ){
+            pNew->aiKeyPtr[iOut++] = pNode->aiKeyPtr[i];
+          }
+        }else{
+          if( getChildPtr(pNode, WORKING_VERSION, i) ){
+            pNew->aiChildPtr[iOut] = getChildPtr(pNode, WORKING_VERSION, i);
+            if( i<3 ) pNew->aiKeyPtr[iOut] = pNode->aiKeyPtr[i];
+            iOut++;
+          }
+        }
+      }
+      assert( iOut<=4 );
+      assert( bLeaf || pNew->aiChildPtr[0]==0 );
+      pCsr->iNode--;
+      rc = treeUpdatePtr(db, pCsr, iNew);
+    }
+
+  }else if( pCsr->iNode==0 ){
+    /* Removing the only key in the root node. iNewptr is the new root. */
+    assert( iSlot==1 );
+    db->treehdr.root.iRoot = iNewptr;
+    db->treehdr.root.nHeight--;
+
+  }else{
+    /* There is only one key on this node and the node is not the root
+    ** node. Find a peer for this node. Then redistribute the contents of
+    ** the peer and the parent cell between the parent and either one or
+    ** two new nodes.  */
+    TreeNode *pParent;            /* Parent tree node */
+    int iPSlot;
+    u32 iPeer;                    /* Pointer to peer leaf node */
+    int iDir;
+    TreeNode *pPeer;              /* The peer leaf node */
+    TreeNode *pNew1; u32 iNew1;   /* First new leaf node */
+
+    assert( iSlot==1 );
+
+    pParent = pCsr->apTreeNode[pCsr->iNode-1];
+    iPSlot = pCsr->aiCell[pCsr->iNode-1];
+
+    if( iPSlot>0 && getChildPtr(pParent, WORKING_VERSION, iPSlot-1) ){
+      iDir = -1;
+    }else{
+      iDir = +1;
+    }
+    iPeer = getChildPtr(pParent, WORKING_VERSION, iPSlot+iDir);
+    pPeer = (TreeNode *)treeShmptr(db, iPeer);
+    assertIsWorkingChild(db, pNode, pParent, iPSlot);
+
+    /* Allocate the first new leaf node. This is always required. */
+    if( bLeaf ){
+      pNew1 = (TreeNode *)newTreeLeaf(db, &iNew1, &rc);
+    }else{
+      pNew1 = (TreeNode *)newTreeNode(db, &iNew1, &rc);
+    }
+
+    if( pPeer->aiKeyPtr[0] && pPeer->aiKeyPtr[2] ){
+      /* Peer node is completely full. This means that two new leaf nodes
+      ** and a new parent node are required. */
+
+      TreeNode *pNew2; u32 iNew2; /* Second new leaf node */
+      TreeNode *pNewP; u32 iNewP; /* New parent node */
+
+      if( bLeaf ){
+        pNew2 = (TreeNode *)newTreeLeaf(db, &iNew2, &rc);
+      }else{
+        pNew2 = (TreeNode *)newTreeNode(db, &iNew2, &rc);
+      }
+      pNewP = copyTreeNode(db, pParent, &iNewP, &rc);
+
+      if( iDir==-1 ){
+        pNew1->aiKeyPtr[1] = pPeer->aiKeyPtr[0];
+        if( bLeaf==0 ){
+          pNew1->aiChildPtr[1] = getChildPtr(pPeer, WORKING_VERSION, 0);
+          pNew1->aiChildPtr[2] = getChildPtr(pPeer, WORKING_VERSION, 1);
+        }
+
+        pNewP->aiChildPtr[iPSlot-1] = iNew1;
+        pNewP->aiKeyPtr[iPSlot-1] = pPeer->aiKeyPtr[1];
+        pNewP->aiChildPtr[iPSlot] = iNew2;
+
+        pNew2->aiKeyPtr[0] = pPeer->aiKeyPtr[2];
+        pNew2->aiKeyPtr[1] = pParent->aiKeyPtr[iPSlot-1];
+        if( bLeaf==0 ){
+          pNew2->aiChildPtr[0] = getChildPtr(pPeer, WORKING_VERSION, 2);
+          pNew2->aiChildPtr[1] = getChildPtr(pPeer, WORKING_VERSION, 3);
+          pNew2->aiChildPtr[2] = iNewptr;
+        }
+      }else{
+        pNew1->aiKeyPtr[1] = pParent->aiKeyPtr[iPSlot];
+        if( bLeaf==0 ){
+          pNew1->aiChildPtr[1] = iNewptr;
+          pNew1->aiChildPtr[2] = getChildPtr(pPeer, WORKING_VERSION, 0);
+        }
+
+        pNewP->aiChildPtr[iPSlot] = iNew1;
+        pNewP->aiKeyPtr[iPSlot] = pPeer->aiKeyPtr[0];
+        pNewP->aiChildPtr[iPSlot+1] = iNew2;
+
+        pNew2->aiKeyPtr[0] = pPeer->aiKeyPtr[1];
+        pNew2->aiKeyPtr[1] = pPeer->aiKeyPtr[2];
+        if( bLeaf==0 ){
+          pNew2->aiChildPtr[0] = getChildPtr(pPeer, WORKING_VERSION, 1);
+          pNew2->aiChildPtr[1] = getChildPtr(pPeer, WORKING_VERSION, 2);
+          pNew2->aiChildPtr[2] = getChildPtr(pPeer, WORKING_VERSION, 3);
+        }
+      }
+      assert( pCsr->iNode>=1 );
+      pCsr->iNode -= 2;
+      if( rc==LSM_OK ){
+        assert( pNew1->aiKeyPtr[1] && pNew2->aiKeyPtr[1] );
+        rc = treeUpdatePtr(db, pCsr, iNewP);
+      }
+    }else{
+      int iKOut = 0;
+      int iPOut = 0;
+      int i;
+
+      pCsr->iNode--;
+
+      if( iDir==1 ){
+        pNew1->aiKeyPtr[iKOut++] = pParent->aiKeyPtr[iPSlot];
+        if( bLeaf==0 ) pNew1->aiChildPtr[iPOut++] = iNewptr;
+      }
+      for(i=0; i<3; i++){
+        if( pPeer->aiKeyPtr[i] ){
+          pNew1->aiKeyPtr[iKOut++] = pPeer->aiKeyPtr[i];
+        }
+      }
+      if( bLeaf==0 ){
+        for(i=0; i<4; i++){
+          if( getChildPtr(pPeer, WORKING_VERSION, i) ){
+            pNew1->aiChildPtr[iPOut++] = getChildPtr(pPeer, WORKING_VERSION, i);
+          }
+        }
+      }
+      if( iDir==-1 ){
+        iPSlot--;
+        pNew1->aiKeyPtr[iKOut++] = pParent->aiKeyPtr[iPSlot];
+        if( bLeaf==0 ) pNew1->aiChildPtr[iPOut++] = iNewptr;
+        pCsr->aiCell[pCsr->iNode] = (u8)iPSlot;
+      }
+
+      rc = treeDeleteEntry(db, pCsr, iNew1);
+    }
+  }
+
+  return rc;
+}
+
+/*
+** Delete a range of keys from the tree structure (i.e. the lsm_delete_range()
+** function, not lsm_delete()).
+**
+** This is a two step process: 
+**
+**     1) Remove all entries currently stored in the tree that have keys
+**        that fall into the deleted range.
+**
+**        TODO: There are surely good ways to optimize this step - removing 
+**        a range of keys from a b-tree. But for now, this function removes
+**        them one at a time using the usual approach.
+**
+**     2) Unless the largest key smaller than or equal to (pKey1/nKey1) is
+**        already marked as START_DELETE, insert a START_DELETE key. 
+**        Similarly, unless the smallest key greater than or equal to
+**        (pKey2/nKey2) is already START_END, insert a START_END key.
+*/
+int lsmTreeDelete(
+  lsm_db *db,
+  void *pKey1, int nKey1,         /* Start of range */
+  void *pKey2, int nKey2          /* End of range */
+){
+  int rc = LSM_OK;
+  int bDone = 0;
+  TreeRoot *p = &db->treehdr.root;
+  TreeBlob blob = {0, 0};
+
+  /* The range must be sensible - that (key1 < key2). */
+  assert( treeKeycmp(pKey1, nKey1, pKey2, nKey2)<0 );
+  assert( assert_delete_ranges_match(db) );
+
+#if 0
+  static int nCall = 0;
+  printf("\n");
+  nCall++;
+  printf("%d delete %s .. %s\n", nCall, (char *)pKey1, (char *)pKey2);
+  dump_tree_contents(db, "before delete");
+#endif
+
+  /* Step 1. This loop runs until the tree contains no keys within the
+  ** range being deleted. Or until an error occurs. */
+  while( bDone==0 && rc==LSM_OK ){
+    int res;
+    TreeCursor csr;               /* Cursor to seek to first key in range */
+    void *pDel; int nDel;         /* Key to (possibly) delete this iteration */
+#ifndef NDEBUG
+    int nEntry = treeCountEntries(db);
+#endif
+
+    /* Seek the cursor to the first entry in the tree greater than pKey1. */
+    treeCursorInit(db, 0, &csr);
+    lsmTreeCursorSeek(&csr, pKey1, nKey1, &res);
+    if( res<=0 && lsmTreeCursorValid(&csr) ) lsmTreeCursorNext(&csr);
+
+    /* If there is no such entry, or if it is greater than pKey2, then the
+    ** tree now contains no keys in the range being deleted. In this case
+    ** break out of the loop.  */
+    bDone = 1;
+    if( lsmTreeCursorValid(&csr) ){
+      lsmTreeCursorKey(&csr, 0, &pDel, &nDel);
+      if( treeKeycmp(pDel, nDel, pKey2, nKey2)<0 ) bDone = 0;
+    }
+
+    if( bDone==0 ){
+      if( (u32)csr.iNode==(p->nHeight-1) ){
+        /* The element to delete already lies on a leaf node */
+        rc = treeDeleteEntry(db, &csr, 0);
+      }else{
+        /* 1. Overwrite the current key with a copy of the next key in the 
+        **    tree (key N).
+        **
+        ** 2. Seek to key N (cursor will stop at the internal node copy of
+        **    N). Move to the next key (original copy of N). Delete
+        **    this entry. 
+        */
+        u32 iKey;
+        TreeKey *pKey;
+        int iNode = csr.iNode;
+        lsmTreeCursorNext(&csr);
+        assert( (u32)csr.iNode==(p->nHeight-1) );
+
+        iKey = csr.apTreeNode[csr.iNode]->aiKeyPtr[csr.aiCell[csr.iNode]];
+        lsmTreeCursorPrev(&csr);
+
+        treeOverwriteKey(db, &csr, iKey, &rc);
+        pKey = treeShmkey(db, iKey, TKV_LOADKEY, &blob, &rc);
+        if( pKey ){
+          rc = lsmTreeCursorSeek(&csr, TKV_KEY(pKey), pKey->nKey, &res);
+        }
+        if( rc==LSM_OK ){
+          assert( res==0 && csr.iNode==iNode );
+          rc = lsmTreeCursorNext(&csr);
+          if( rc==LSM_OK ){
+            rc = treeDeleteEntry(db, &csr, 0);
+          }
+        }
+      }
+    }
+
+    /* Clean up any memory allocated by the cursor. */
+    tblobFree(db, &csr.blob);
+#if 0
+    dump_tree_contents(db, "ddd delete");
+#endif
+    assert( bDone || treeCountEntries(db)==(nEntry-1) );
+  }
+
+#if 0
+  dump_tree_contents(db, "during delete");
+#endif
+
+  /* Now insert the START_DELETE and END_DELETE keys. */
+  if( rc==LSM_OK ){
+    rc = treeInsertEntry(db, LSM_START_DELETE, pKey1, nKey1, 0, -1);
+  }
+#if 0
+  dump_tree_contents(db, "during delete 2");
+#endif
+  if( rc==LSM_OK ){
+    rc = treeInsertEntry(db, LSM_END_DELETE, pKey2, nKey2, 0, -1);
+  }
+
+#if 0
+  dump_tree_contents(db, "after delete");
+#endif
+
+  tblobFree(db, &blob);
+  assert( assert_delete_ranges_match(db) );
+  return rc;
+}
+
+/*
+** Return, in bytes, the amount of memory currently used by the tree 
+** structure.
+*/
+int lsmTreeSize(lsm_db *pDb){
+  return pDb->treehdr.root.nByte;
+}
+
+/*
+** Open a cursor on the in-memory tree pTree.
+*/
+int lsmTreeCursorNew(lsm_db *pDb, int bOld, TreeCursor **ppCsr){
+  TreeCursor *pCsr;
+  *ppCsr = pCsr = lsmMalloc(pDb->pEnv, sizeof(TreeCursor));
+  if( pCsr ){
+    treeCursorInit(pDb, bOld, pCsr);
+    return LSM_OK;
+  }
+  return LSM_NOMEM_BKPT;
+}
+
+/*
+** Close an in-memory tree cursor.
+*/
+void lsmTreeCursorDestroy(TreeCursor *pCsr){
+  if( pCsr ){
+    tblobFree(pCsr->pDb, &pCsr->blob);
+    lsmFree(pCsr->pDb->pEnv, pCsr);
+  }
+}
+
+void lsmTreeCursorReset(TreeCursor *pCsr){
+  if( pCsr ){
+    pCsr->iNode = -1;
+    pCsr->pSave = 0;
+  }
+}
+
+#ifndef NDEBUG
+static int treeCsrCompare(TreeCursor *pCsr, void *pKey, int nKey, int *pRc){
+  TreeKey *p;
+  int cmp = 0;
+  assert( pCsr->iNode>=0 );
+  p = csrGetKey(pCsr, &pCsr->blob, pRc);
+  if( p ){
+    cmp = treeKeycmp(TKV_KEY(p), p->nKey, pKey, nKey);
+  }
+  return cmp;
+}
+#endif
+
+
+/*
+** Attempt to seek the cursor passed as the first argument to key (pKey/nKey)
+** in the tree structure. If an exact match for the key is found, leave the
+** cursor pointing to it and set *pRes to zero before returning. If an
+** exact match cannot be found, do one of the following:
+**
+**   * Leave the cursor pointing to the smallest element in the tree that 
+**     is larger than the key and set *pRes to +1, or
+**
+**   * Leave the cursor pointing to the largest element in the tree that 
+**     is smaller than the key and set *pRes to -1, or
+**
+**   * If the tree is empty, leave the cursor at EOF and set *pRes to -1.
+*/
+int lsmTreeCursorSeek(TreeCursor *pCsr, void *pKey, int nKey, int *pRes){
+  int rc = LSM_OK;                /* Return code */
+  lsm_db *pDb = pCsr->pDb;
+  TreeRoot *pRoot = pCsr->pRoot;
+  u32 iNodePtr;                   /* Location of current node in search */
+
+  /* Discard any saved position data */
+  treeCursorRestore(pCsr, 0);
+
+  iNodePtr = pRoot->iRoot;
+  if( iNodePtr==0 ){
+    /* Either an error occurred or the tree is completely empty. */
+    assert( rc!=LSM_OK || pRoot->iRoot==0 );
+    *pRes = -1;
+    pCsr->iNode = -1;
+  }else{
+    TreeBlob b = {0, 0};
+    int res = 0;                  /* Result of comparison function */
+    int iNode = -1;
+    while( iNodePtr ){
+      TreeNode *pNode;            /* Node at location iNodePtr */
+      int iTest;                  /* Index of second key to test (0 or 2) */
+      u32 iTreeKey;
+      TreeKey *pTreeKey;          /* Key to compare against */
+
+      pNode = (TreeNode *)treeShmptrUnsafe(pDb, iNodePtr);
+      iNode++;
+      pCsr->apTreeNode[iNode] = pNode;
+
+      /* Compare (pKey/nKey) with the key in the middle slot of B-tree node
+      ** pNode. The middle slot is never empty. If the comparison is a match,
+      ** then the search is finished. Break out of the loop. */
+      pTreeKey = (TreeKey*)treeShmptrUnsafe(pDb, pNode->aiKeyPtr[1]);
+      if( !(pTreeKey->flags & LSM_CONTIGUOUS) ){
+        pTreeKey = treeShmkey(pDb, pNode->aiKeyPtr[1], TKV_LOADKEY, &b, &rc);
+        if( rc!=LSM_OK ) break;
+      }
+      res = treeKeycmp((void *)&pTreeKey[1], pTreeKey->nKey, pKey, nKey);
+      if( res==0 ){
+        pCsr->aiCell[iNode] = 1;
+        break;
+      }
+
+      /* Based on the results of the previous comparison, compare (pKey/nKey)
+      ** to either the left or right key of the B-tree node, if such a key
+      ** exists. */
+      iTest = (res>0 ? 0 : 2);
+      iTreeKey = pNode->aiKeyPtr[iTest];
+      if( iTreeKey ){
+        pTreeKey = (TreeKey*)treeShmptrUnsafe(pDb, iTreeKey);
+        if( !(pTreeKey->flags & LSM_CONTIGUOUS) ){
+          pTreeKey = treeShmkey(pDb, iTreeKey, TKV_LOADKEY, &b, &rc);
+          if( rc ) break;
+        }
+        res = treeKeycmp((void *)&pTreeKey[1], pTreeKey->nKey, pKey, nKey);
+        if( res==0 ){
+          pCsr->aiCell[iNode] = (u8)iTest;
+          break;
+        }
+      }else{
+        iTest = 1;
+      }
+
+      if( (u32)iNode<(pRoot->nHeight-1) ){
+        iNodePtr = getChildPtr(pNode, pRoot->iTransId, iTest + (res<0));
+      }else{
+        iNodePtr = 0;
+      }
+      pCsr->aiCell[iNode] = (u8)(iTest + (iNodePtr && (res<0)));
+    }
+
+    *pRes = res;
+    pCsr->iNode = iNode;
+    tblobFree(pDb, &b);
+  }
+
+  /* assert() that *pRes has been set properly */
+#ifndef NDEBUG
+  if( rc==LSM_OK && lsmTreeCursorValid(pCsr) ){
+    int cmp = treeCsrCompare(pCsr, pKey, nKey, &rc);
+    assert( rc!=LSM_OK || *pRes==cmp || (*pRes ^ cmp)>0 );
+  }
+#endif
+
+  return rc;
+}
+
+int lsmTreeCursorNext(TreeCursor *pCsr){
+#ifndef NDEBUG
+  TreeKey *pK1;
+  TreeBlob key1 = {0, 0};
+#endif
+  lsm_db *pDb = pCsr->pDb;
+  TreeRoot *pRoot = pCsr->pRoot;
+  const int iLeaf = pRoot->nHeight-1;
+  int iCell; 
+  int rc = LSM_OK; 
+  TreeNode *pNode; 
+
+  /* Restore the cursor position, if required */
+  int iRestore = 0;
+  treeCursorRestore(pCsr, &iRestore);
+  if( iRestore>0 ) return LSM_OK;
+
+  /* Save a pointer to the current key. This is used in an assert() at the
+  ** end of this function - to check that the 'next' key really is larger
+  ** than the current key. */
+#ifndef NDEBUG
+  pK1 = csrGetKey(pCsr, &key1, &rc);
+  if( rc!=LSM_OK ) return rc;
+#endif
+
+  assert( lsmTreeCursorValid(pCsr) );
+  assert( pCsr->aiCell[pCsr->iNode]<3 );
+
+  pNode = pCsr->apTreeNode[pCsr->iNode];
+  iCell = ++pCsr->aiCell[pCsr->iNode];
+
+  /* If the current node is not a leaf, and the current cell has sub-tree
+  ** associated with it, descend to the left-most key on the left-most
+  ** leaf of the sub-tree.  */
+  if( pCsr->iNode<iLeaf && getChildPtr(pNode, pRoot->iTransId, iCell) ){
+    do {
+      u32 iNodePtr;
+      pCsr->iNode++;
+      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
+      pNode = (TreeNode *)treeShmptr(pDb, iNodePtr);
+      pCsr->apTreeNode[pCsr->iNode] = pNode;
+      iCell = pCsr->aiCell[pCsr->iNode] = (pNode->aiKeyPtr[0]==0);
+    }while( pCsr->iNode < iLeaf );
+  }
+
+  /* Otherwise, the next key is found by following pointer up the tree 
+  ** until there is a key immediately to the right of the pointer followed 
+  ** to reach the sub-tree containing the current key. */
+  else if( iCell>=3 || pNode->aiKeyPtr[iCell]==0 ){
+    while( (--pCsr->iNode)>=0 ){
+      iCell = pCsr->aiCell[pCsr->iNode];
+      if( iCell<3 && pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[iCell] ) break;
+    }
+  }
+
+#ifndef NDEBUG
+  if( pCsr->iNode>=0 ){
+    TreeKey *pK2 = csrGetKey(pCsr, &pCsr->blob, &rc);
+    assert( rc||treeKeycmp(TKV_KEY(pK2),pK2->nKey,TKV_KEY(pK1),pK1->nKey)>=0 );
+  }
+  tblobFree(pDb, &key1);
+#endif
+
+  return rc;
+}
+
+int lsmTreeCursorPrev(TreeCursor *pCsr){
+#ifndef NDEBUG
+  TreeKey *pK1;
+  TreeBlob key1 = {0, 0};
+#endif
+  lsm_db *pDb = pCsr->pDb;
+  TreeRoot *pRoot = pCsr->pRoot;
+  const int iLeaf = pRoot->nHeight-1;
+  int iCell; 
+  int rc = LSM_OK; 
+  TreeNode *pNode; 
+
+  /* Restore the cursor position, if required */
+  int iRestore = 0;
+  treeCursorRestore(pCsr, &iRestore);
+  if( iRestore<0 ) return LSM_OK;
+
+  /* Save a pointer to the current key. This is used in an assert() at the
+  ** end of this function - to check that the 'next' key really is smaller
+  ** than the current key. */
+#ifndef NDEBUG
+  pK1 = csrGetKey(pCsr, &key1, &rc);
+  if( rc!=LSM_OK ) return rc;
+#endif
+
+  assert( lsmTreeCursorValid(pCsr) );
+  pNode = pCsr->apTreeNode[pCsr->iNode];
+  iCell = pCsr->aiCell[pCsr->iNode];
+  assert( iCell>=0 && iCell<3 );
+
+  /* If the current node is not a leaf, and the current cell has sub-tree
+  ** associated with it, descend to the right-most key on the right-most
+  ** leaf of the sub-tree.  */
+  if( pCsr->iNode<iLeaf && getChildPtr(pNode, pRoot->iTransId, iCell) ){
+    do {
+      u32 iNodePtr;
+      pCsr->iNode++;
+      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
+      pNode = (TreeNode *)treeShmptr(pDb, iNodePtr);
+      if( rc!=LSM_OK ) break;
+      pCsr->apTreeNode[pCsr->iNode] = pNode;
+      iCell = 1 + (pNode->aiKeyPtr[2]!=0) + (pCsr->iNode < iLeaf);
+      pCsr->aiCell[pCsr->iNode] = (u8)iCell;
+    }while( pCsr->iNode < iLeaf );
+  }
+
+  /* Otherwise, the next key is found by following pointer up the tree until
+  ** there is a key immediately to the left of the pointer followed to reach
+  ** the sub-tree containing the current key. */
+  else{
+    do {
+      iCell = pCsr->aiCell[pCsr->iNode]-1;
+      if( iCell>=0 && pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[iCell] ) break;
+    }while( (--pCsr->iNode)>=0 );
+    pCsr->aiCell[pCsr->iNode] = (u8)iCell;
+  }
+
+#ifndef NDEBUG
+  if( pCsr->iNode>=0 ){
+    TreeKey *pK2 = csrGetKey(pCsr, &pCsr->blob, &rc);
+    assert( rc || treeKeycmp(TKV_KEY(pK2),pK2->nKey,TKV_KEY(pK1),pK1->nKey)<0 );
+  }
+  tblobFree(pDb, &key1);
+#endif
+
+  return rc;
+}
+
+/*
+** Move the cursor to the first (bLast==0) or last (bLast!=0) entry in the
+** in-memory tree.
+*/
+int lsmTreeCursorEnd(TreeCursor *pCsr, int bLast){
+  lsm_db *pDb = pCsr->pDb;
+  TreeRoot *pRoot = pCsr->pRoot;
+  int rc = LSM_OK;
+
+  u32 iNodePtr;
+  pCsr->iNode = -1;
+
+  /* Discard any saved position data */
+  treeCursorRestore(pCsr, 0);
+
+  iNodePtr = pRoot->iRoot;
+  while( iNodePtr ){
+    int iCell;
+    TreeNode *pNode;
+
+    pNode = (TreeNode *)treeShmptr(pDb, iNodePtr);
+    if( rc ) break;
+
+    if( bLast ){
+      iCell = ((pNode->aiKeyPtr[2]==0) ? 2 : 3);
+    }else{
+      iCell = ((pNode->aiKeyPtr[0]==0) ? 1 : 0);
+    }
+    pCsr->iNode++;
+    pCsr->apTreeNode[pCsr->iNode] = pNode;
+
+    if( (u32)pCsr->iNode<pRoot->nHeight-1 ){
+      iNodePtr = getChildPtr(pNode, pRoot->iTransId, iCell);
+    }else{
+      iNodePtr = 0;
+    }
+    pCsr->aiCell[pCsr->iNode] = (u8)(iCell - (iNodePtr==0 && bLast));
+  }
+
+  return rc;
+}
+
+int lsmTreeCursorFlags(TreeCursor *pCsr){
+  int flags = 0;
+  if( pCsr && pCsr->iNode>=0 ){
+    int rc = LSM_OK;
+    TreeKey *pKey = (TreeKey *)treeShmptrUnsafe(pCsr->pDb,
+        pCsr->apTreeNode[pCsr->iNode]->aiKeyPtr[pCsr->aiCell[pCsr->iNode]]
+    );
+    assert( rc==LSM_OK );
+    flags = (pKey->flags & ~LSM_CONTIGUOUS);
+  }
+  return flags;
+}
+
+int lsmTreeCursorKey(TreeCursor *pCsr, int *pFlags, void **ppKey, int *pnKey){
+  TreeKey *pTreeKey;
+  int rc = LSM_OK;
+
+  assert( lsmTreeCursorValid(pCsr) );
+
+  pTreeKey = pCsr->pSave;
+  if( !pTreeKey ){
+    pTreeKey = csrGetKey(pCsr, &pCsr->blob, &rc);
+  }
+  if( rc==LSM_OK ){
+    *pnKey = pTreeKey->nKey;
+    if( pFlags ) *pFlags = pTreeKey->flags;
+    *ppKey = (void *)&pTreeKey[1];
+  }
+
+  return rc;
+}
+
+int lsmTreeCursorValue(TreeCursor *pCsr, void **ppVal, int *pnVal){
+  int res = 0;
+  int rc;
+
+  rc = treeCursorRestore(pCsr, &res);
+  if( res==0 ){
+    TreeKey *pTreeKey = csrGetKey(pCsr, &pCsr->blob, &rc);
+    if( rc==LSM_OK ){
+      if( pTreeKey->flags & LSM_INSERT ){
+        *pnVal = pTreeKey->nValue;
+        *ppVal = TKV_VAL(pTreeKey);
+      }else{
+        *ppVal = 0;
+        *pnVal = -1;
+      }
+    }
+  }else{
+    *ppVal = 0;
+    *pnVal = 0;
+  }
+
+  return rc;
+}
+
+/*
+** Return true if the cursor currently points to a valid entry. 
+*/
+int lsmTreeCursorValid(TreeCursor *pCsr){
+  return (pCsr && (pCsr->pSave || pCsr->iNode>=0));
+}
+
+/*
+** Store a mark in *pMark. Later on, a call to lsmTreeRollback() with a
+** pointer to the same TreeMark structure may be used to roll the tree
+** contents back to their current state.
+*/
+void lsmTreeMark(lsm_db *pDb, TreeMark *pMark){
+  pMark->iRoot = pDb->treehdr.root.iRoot;
+  pMark->nHeight = pDb->treehdr.root.nHeight;
+  pMark->iWrite = pDb->treehdr.iWrite;
+  pMark->nChunk = pDb->treehdr.nChunk;
+  pMark->iNextShmid = pDb->treehdr.iNextShmid;
+  pMark->iRollback = intArraySize(&pDb->rollback);
+}
+
+/*
+** Roll back to mark pMark. Structure *pMark should have been previously
+** populated by a call to lsmTreeMark().
+*/
+void lsmTreeRollback(lsm_db *pDb, TreeMark *pMark){
+  int iIdx;
+  int nIdx;
+  u32 iNext;
+  ShmChunk *pChunk;
+  u32 iChunk;
+  u32 iShmid;
+
+  /* Revert all required v2 pointers. */
+  nIdx = intArraySize(&pDb->rollback);
+  for(iIdx = pMark->iRollback; iIdx<nIdx; iIdx++){
+    TreeNode *pNode;
+    pNode = treeShmptr(pDb, intArrayEntry(&pDb->rollback, iIdx));
+    assert( pNode );
+    pNode->iV2 = 0;
+    pNode->iV2Child = 0;
+    pNode->iV2Ptr = 0;
+  }
+  intArrayTruncate(&pDb->rollback, pMark->iRollback);
+
+  /* Restore the free-chunk list. */
+  assert( pMark->iWrite!=0 );
+  iChunk = treeOffsetToChunk(pMark->iWrite-1);
+  pChunk = treeShmChunk(pDb, iChunk);
+  iNext = pChunk->iNext;
+  pChunk->iNext = 0;
+
+  pChunk = treeShmChunk(pDb, pDb->treehdr.iFirst);
+  iShmid = pChunk->iShmid-1;
+
+  while( iNext ){
+    u32 iFree = iNext;            /* Current chunk being rollback-freed */
+    ShmChunk *pFree;              /* Pointer to chunk iFree */
+
+    pFree = treeShmChunk(pDb, iFree);
+    iNext = pFree->iNext;
+
+    if( iFree<pMark->nChunk ){
+      pFree->iNext = pDb->treehdr.iFirst;
+      pFree->iShmid = iShmid--;
+      pDb->treehdr.iFirst = iFree;
+    }
+  }
+
+  /* Restore the tree-header fields */
+  pDb->treehdr.root.iRoot = pMark->iRoot;
+  pDb->treehdr.root.nHeight = pMark->nHeight;
+  pDb->treehdr.iWrite = pMark->iWrite;
+  pDb->treehdr.nChunk = pMark->nChunk;
+  pDb->treehdr.iNextShmid = pMark->iNextShmid;
+}
+
+/*
+** Load the in-memory tree header from shared-memory into pDb->treehdr.
+** If the header cannot be loaded, return LSM_PROTOCOL.
+**
+** If the header is successfully loaded and parameter piRead is not NULL,
+** is is set to 1 if the header was loaded from ShmHeader.hdr1, or 2 if
+** the header was loaded from ShmHeader.hdr2.
+*/
+int lsmTreeLoadHeader(lsm_db *pDb, int *piRead){
+  int nRem = LSM_ATTEMPTS_BEFORE_PROTOCOL;
+  while( (nRem--)>0 ){
+    ShmHeader *pShm = pDb->pShmhdr;
+
+    memcpy(&pDb->treehdr, &pShm->hdr1, sizeof(TreeHeader));
+    if( treeHeaderChecksumOk(&pDb->treehdr) ){
+      if( piRead ) *piRead = 1;
+      return LSM_OK;
+    }
+    memcpy(&pDb->treehdr, &pShm->hdr2, sizeof(TreeHeader));
+    if( treeHeaderChecksumOk(&pDb->treehdr) ){
+      if( piRead ) *piRead = 2;
+      return LSM_OK;
+    }
+
+    lsmShmBarrier(pDb);
+  }
+  return LSM_PROTOCOL_BKPT;
+}
+
+int lsmTreeLoadHeaderOk(lsm_db *pDb, int iRead){
+  TreeHeader *p = (iRead==1) ? &pDb->pShmhdr->hdr1 : &pDb->pShmhdr->hdr2;
+  assert( iRead==1 || iRead==2 );
+  return (0==memcmp(pDb->treehdr.aCksum, p->aCksum, sizeof(u32)*2));
+}
+
+/*
+** This function is called to conclude a transaction. If argument bCommit
+** is true, the transaction is committed. Otherwise it is rolled back.
+*/
+int lsmTreeEndTransaction(lsm_db *pDb, int bCommit){
+  ShmHeader *pShm = pDb->pShmhdr;
+
+  treeHeaderChecksum(&pDb->treehdr, pDb->treehdr.aCksum);
+  memcpy(&pShm->hdr2, &pDb->treehdr, sizeof(TreeHeader));
+  lsmShmBarrier(pDb);
+  memcpy(&pShm->hdr1, &pDb->treehdr, sizeof(TreeHeader));
+  pShm->bWriter = 0;
+  intArrayFree(pDb->pEnv, &pDb->rollback);
+
+  return LSM_OK;
+}
+
+#ifndef NDEBUG
+static int assert_delete_ranges_match(lsm_db *db){
+  int prev = 0;
+  TreeBlob blob = {0, 0};
+  TreeCursor csr;               /* Cursor used to iterate through tree */
+  int rc;
+
+  treeCursorInit(db, 0, &csr);
+  for( rc = lsmTreeCursorEnd(&csr, 0);
+       rc==LSM_OK && lsmTreeCursorValid(&csr);
+       rc = lsmTreeCursorNext(&csr)
+  ){
+    TreeKey *pKey = csrGetKey(&csr, &blob, &rc);
+    if( rc!=LSM_OK ) break;
+    assert( ((prev&LSM_START_DELETE)==0)==((pKey->flags&LSM_END_DELETE)==0) );
+    prev = pKey->flags;
+  }
+
+  tblobFree(csr.pDb, &csr.blob);
+  tblobFree(csr.pDb, &blob);
+
+  return 1;
+}
+
+static int treeCountEntries(lsm_db *db){
+  TreeCursor csr;               /* Cursor used to iterate through tree */
+  int rc;
+  int nEntry = 0;
+
+  treeCursorInit(db, 0, &csr);
+  for( rc = lsmTreeCursorEnd(&csr, 0);
+       rc==LSM_OK && lsmTreeCursorValid(&csr);
+       rc = lsmTreeCursorNext(&csr)
+  ){
+    nEntry++;
+  }
+
+  tblobFree(csr.pDb, &csr.blob);
+
+  return nEntry;
+}
+#endif
diff --git a/ext/lsm1/lsm_unix.c b/ext/lsm1/lsm_unix.c
new file mode 100644
index 0000000..88952d1
--- /dev/null
+++ b/ext/lsm1/lsm_unix.c
@@ -0,0 +1,753 @@
+/*
+** 2011-12-03
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** Unix-specific run-time environment implementation for LSM.
+*/
+
+#ifndef _WIN32
+
+#if defined(__GNUC__) || defined(__TINYC__)
+/* workaround for ftruncate() visibility on gcc. */
+# ifndef _XOPEN_SOURCE
+#  define _XOPEN_SOURCE 500
+# endif
+#endif
+
+#include <unistd.h>
+#include <sys/types.h>
+
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <assert.h>
+#include <string.h>
+
+#include <stdlib.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <ctype.h>
+
+#include <unistd.h>
+#include <errno.h>
+
+#include <sys/mman.h>
+#include "lsmInt.h"
+
+/* There is no fdatasync() call on Android */
+#ifdef __ANDROID__
+# define fdatasync(x) fsync(x)
+#endif
+
+/*
+** An open file is an instance of the following object
+*/
+typedef struct PosixFile PosixFile;
+struct PosixFile {
+  lsm_env *pEnv;                  /* The run-time environment */
+  const char *zName;              /* Full path to file */
+  int fd;                         /* The open file descriptor */
+  int shmfd;                      /* Shared memory file-descriptor */
+  void *pMap;                     /* Pointer to mapping of file fd */
+  off_t nMap;                     /* Size of mapping at pMap in bytes */
+  int nShm;                       /* Number of entries in array apShm[] */
+  void **apShm;                   /* Array of 32K shared memory segments */
+};
+
+static char *posixShmFile(PosixFile *p){
+  char *zShm;
+  int nName = strlen(p->zName);
+  zShm = (char *)lsmMalloc(p->pEnv, nName+4+1);
+  if( zShm ){
+    memcpy(zShm, p->zName, nName);
+    memcpy(&zShm[nName], "-shm", 5);
+  }
+  return zShm;
+}
+
+static int lsmPosixOsOpen(
+  lsm_env *pEnv,
+  const char *zFile,
+  int flags,
+  lsm_file **ppFile
+){
+  int rc = LSM_OK;
+  PosixFile *p;
+
+  p = lsm_malloc(pEnv, sizeof(PosixFile));
+  if( p==0 ){
+    rc = LSM_NOMEM;
+  }else{
+    int bReadonly = (flags & LSM_OPEN_READONLY);
+    int oflags = (bReadonly ? O_RDONLY : (O_RDWR|O_CREAT));
+    memset(p, 0, sizeof(PosixFile));
+    p->zName = zFile;
+    p->pEnv = pEnv;
+    p->fd = open(zFile, oflags, 0644);
+    if( p->fd<0 ){
+      lsm_free(pEnv, p);
+      p = 0;
+      if( errno==ENOENT ){
+        rc = lsmErrorBkpt(LSM_IOERR_NOENT);
+      }else{
+        rc = LSM_IOERR_BKPT;
+      }
+    }
+  }
+
+  *ppFile = (lsm_file *)p;
+  return rc;
+}
+
+static int lsmPosixOsWrite(
+  lsm_file *pFile,                /* File to write to */
+  lsm_i64 iOff,                   /* Offset to write to */
+  void *pData,                    /* Write data from this buffer */
+  int nData                       /* Bytes of data to write */
+){
+  int rc = LSM_OK;
+  PosixFile *p = (PosixFile *)pFile;
+  off_t offset;
+
+  offset = lseek(p->fd, (off_t)iOff, SEEK_SET);
+  if( offset!=iOff ){
+    rc = LSM_IOERR_BKPT;
+  }else{
+    ssize_t prc = write(p->fd, pData, (size_t)nData);
+    if( prc<0 ) rc = LSM_IOERR_BKPT;
+  }
+
+  return rc;
+}
+
+static int lsmPosixOsTruncate(
+  lsm_file *pFile,                /* File to write to */
+  lsm_i64 nSize                   /* Size to truncate file to */
+){
+  PosixFile *p = (PosixFile *)pFile;
+  int rc = LSM_OK;                /* Return code */
+  int prc;                        /* Posix Return Code */
+  struct stat sStat;              /* Result of fstat() invocation */
+  
+  prc = fstat(p->fd, &sStat);
+  if( prc==0 && sStat.st_size>nSize ){
+    prc = ftruncate(p->fd, (off_t)nSize);
+  }
+  if( prc<0 ) rc = LSM_IOERR_BKPT;
+
+  return rc;
+}
+
+static int lsmPosixOsRead(
+  lsm_file *pFile,                /* File to read from */
+  lsm_i64 iOff,                   /* Offset to read from */
+  void *pData,                    /* Read data into this buffer */
+  int nData                       /* Bytes of data to read */
+){
+  int rc = LSM_OK;
+  PosixFile *p = (PosixFile *)pFile;
+  off_t offset;
+
+  offset = lseek(p->fd, (off_t)iOff, SEEK_SET);
+  if( offset!=iOff ){
+    rc = LSM_IOERR_BKPT;
+  }else{
+    ssize_t prc = read(p->fd, pData, (size_t)nData);
+    if( prc<0 ){ 
+      rc = LSM_IOERR_BKPT;
+    }else if( prc<nData ){
+      memset(&((u8 *)pData)[prc], 0, nData - prc);
+    }
+
+  }
+
+  return rc;
+}
+
+static int lsmPosixOsSync(lsm_file *pFile){
+  int rc = LSM_OK;
+
+#ifndef LSM_NO_SYNC
+  PosixFile *p = (PosixFile *)pFile;
+  int prc = 0;
+
+  if( p->pMap ){
+    prc = msync(p->pMap, p->nMap, MS_SYNC);
+  }
+  if( prc==0 ) prc = fdatasync(p->fd);
+  if( prc<0 ) rc = LSM_IOERR_BKPT;
+#else
+  (void)pFile;
+#endif
+
+  return rc;
+}
+
+static int lsmPosixOsSectorSize(lsm_file *pFile){
+  return 512;
+}
+
+static int lsmPosixOsRemap(
+  lsm_file *pFile, 
+  lsm_i64 iMin, 
+  void **ppOut,
+  lsm_i64 *pnOut
+){
+  off_t iSz;
+  int prc;
+  PosixFile *p = (PosixFile *)pFile;
+  struct stat buf;
+
+  /* If the file is between 0 and 2MB in size, extend it in chunks of 256K.
+  ** Thereafter, in chunks of 1MB at a time.  */
+  const int aIncrSz[] = {256*1024, 1024*1024};
+  int nIncrSz = aIncrSz[iMin>(2*1024*1024)];
+
+  if( p->pMap ){
+    munmap(p->pMap, p->nMap);
+    *ppOut = p->pMap = 0;
+    *pnOut = p->nMap = 0;
+  }
+
+  if( iMin>=0 ){
+    memset(&buf, 0, sizeof(buf));
+    prc = fstat(p->fd, &buf);
+    if( prc!=0 ) return LSM_IOERR_BKPT;
+    iSz = buf.st_size;
+    if( iSz<iMin ){
+      iSz = ((iMin + nIncrSz-1) / nIncrSz) * nIncrSz;
+      prc = ftruncate(p->fd, iSz);
+      if( prc!=0 ) return LSM_IOERR_BKPT;
+    }
+
+    p->pMap = mmap(0, iSz, PROT_READ|PROT_WRITE, MAP_SHARED, p->fd, 0);
+    if( p->pMap==MAP_FAILED ){
+      p->pMap = 0;
+      return LSM_IOERR_BKPT;
+    }
+    p->nMap = iSz;
+  }
+
+  *ppOut = p->pMap;
+  *pnOut = p->nMap;
+  return LSM_OK;
+}
+
+static int lsmPosixOsFullpath(
+  lsm_env *pEnv,
+  const char *zName,
+  char *zOut,
+  int *pnOut
+){
+  int nBuf = *pnOut;
+  int nReq;
+
+  if( zName[0]!='/' ){
+    char *z;
+    char *zTmp;
+    int nTmp = 512;
+    zTmp = lsmMalloc(pEnv, nTmp);
+    while( zTmp ){
+      z = getcwd(zTmp, nTmp);
+      if( z || errno!=ERANGE ) break;
+      nTmp = nTmp*2;
+      zTmp = lsmReallocOrFree(pEnv, zTmp, nTmp);
+    }
+    if( zTmp==0 ) return LSM_NOMEM_BKPT;
+    if( z==0 ) return LSM_IOERR_BKPT;
+    assert( z==zTmp );
+
+    nTmp = strlen(zTmp);
+    nReq = nTmp + 1 + strlen(zName) + 1;
+    if( nReq<=nBuf ){
+      memcpy(zOut, zTmp, nTmp);
+      zOut[nTmp] = '/';
+      memcpy(&zOut[nTmp+1], zName, strlen(zName)+1);
+    }
+    lsmFree(pEnv, zTmp);
+  }else{
+    nReq = strlen(zName)+1;
+    if( nReq<=nBuf ){
+      memcpy(zOut, zName, strlen(zName)+1);
+    }
+  }
+
+  *pnOut = nReq;
+  return LSM_OK;
+}
+
+static int lsmPosixOsFileid(
+  lsm_file *pFile, 
+  void *pBuf,
+  int *pnBuf
+){
+  int prc;
+  int nBuf;
+  int nReq;
+  PosixFile *p = (PosixFile *)pFile;
+  struct stat buf;
+
+  nBuf = *pnBuf;
+  nReq = (sizeof(buf.st_dev) + sizeof(buf.st_ino));
+  *pnBuf = nReq;
+  if( nReq>nBuf ) return LSM_OK;
+
+  memset(&buf, 0, sizeof(buf));
+  prc = fstat(p->fd, &buf);
+  if( prc!=0 ) return LSM_IOERR_BKPT;
+
+  memcpy(pBuf, &buf.st_dev, sizeof(buf.st_dev));
+  memcpy(&(((u8 *)pBuf)[sizeof(buf.st_dev)]), &buf.st_ino, sizeof(buf.st_ino));
+  return LSM_OK;
+}
+
+static int lsmPosixOsUnlink(lsm_env *pEnv, const char *zFile){
+  int prc = unlink(zFile);
+  return prc ? LSM_IOERR_BKPT : LSM_OK;
+}
+
+static int lsmPosixOsLock(lsm_file *pFile, int iLock, int eType){
+  int rc = LSM_OK;
+  PosixFile *p = (PosixFile *)pFile;
+  static const short aType[3] = { F_UNLCK, F_RDLCK, F_WRLCK };
+  struct flock lock;
+
+  assert( aType[LSM_LOCK_UNLOCK]==F_UNLCK );
+  assert( aType[LSM_LOCK_SHARED]==F_RDLCK );
+  assert( aType[LSM_LOCK_EXCL]==F_WRLCK );
+  assert( eType>=0 && eType<array_size(aType) );
+  assert( iLock>0 && iLock<=32 );
+
+  memset(&lock, 0, sizeof(lock));
+  lock.l_whence = SEEK_SET;
+  lock.l_len = 1;
+  lock.l_type = aType[eType];
+  lock.l_start = (4096-iLock);
+
+  if( fcntl(p->fd, F_SETLK, &lock) ){
+    int e = errno;
+    if( e==EACCES || e==EAGAIN ){
+      rc = LSM_BUSY;
+    }else{
+      rc = LSM_IOERR_BKPT;
+    }
+  }
+
+  return rc;
+}
+
+static int lsmPosixOsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
+  int rc = LSM_OK;
+  PosixFile *p = (PosixFile *)pFile;
+  static const short aType[3] = { 0, F_RDLCK, F_WRLCK };
+  struct flock lock;
+
+  assert( eType==LSM_LOCK_SHARED || eType==LSM_LOCK_EXCL );
+  assert( aType[LSM_LOCK_SHARED]==F_RDLCK );
+  assert( aType[LSM_LOCK_EXCL]==F_WRLCK );
+  assert( eType>=0 && eType<array_size(aType) );
+  assert( iLock>0 && iLock<=32 );
+
+  memset(&lock, 0, sizeof(lock));
+  lock.l_whence = SEEK_SET;
+  lock.l_len = nLock;
+  lock.l_type = aType[eType];
+  lock.l_start = (4096-iLock-nLock+1);
+
+  if( fcntl(p->fd, F_GETLK, &lock) ){
+    rc = LSM_IOERR_BKPT;
+  }else if( lock.l_type!=F_UNLCK ){
+    rc = LSM_BUSY;
+  }
+
+  return rc;
+}
+
+static int lsmPosixOsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
+  PosixFile *p = (PosixFile *)pFile;
+
+  *ppShm = 0;
+  assert( sz==LSM_SHM_CHUNK_SIZE );
+  if( iChunk>=p->nShm ){
+    int i;
+    void **apNew;
+    int nNew = iChunk+1;
+    off_t nReq = nNew * LSM_SHM_CHUNK_SIZE;
+    struct stat sStat;
+
+    /* If the shared-memory file has not been opened, open it now. */
+    if( p->shmfd<=0 ){
+      char *zShm = posixShmFile(p);
+      if( !zShm ) return LSM_NOMEM_BKPT;
+      p->shmfd = open(zShm, O_RDWR|O_CREAT, 0644);
+      lsmFree(p->pEnv, zShm);
+      if( p->shmfd<0 ){ 
+        return LSM_IOERR_BKPT;
+      }
+    }
+
+    /* If the shared-memory file is not large enough to contain the 
+    ** requested chunk, cause it to grow.  */
+    if( fstat(p->shmfd, &sStat) ){
+      return LSM_IOERR_BKPT;
+    }
+    if( sStat.st_size<nReq ){
+      if( ftruncate(p->shmfd, nReq) ){
+        return LSM_IOERR_BKPT;
+      }
+    }
+
+    apNew = (void **)lsmRealloc(p->pEnv, p->apShm, sizeof(void *) * nNew);
+    if( !apNew ) return LSM_NOMEM_BKPT;
+    for(i=p->nShm; i<nNew; i++){
+      apNew[i] = 0;
+    }
+    p->apShm = apNew;
+    p->nShm = nNew;
+  }
+
+  if( p->apShm[iChunk]==0 ){
+    p->apShm[iChunk] = mmap(0, LSM_SHM_CHUNK_SIZE, 
+        PROT_READ|PROT_WRITE, MAP_SHARED, p->shmfd, iChunk*LSM_SHM_CHUNK_SIZE
+    );
+    if( p->apShm[iChunk]==MAP_FAILED ){
+      p->apShm[iChunk] = 0;
+      return LSM_IOERR_BKPT;
+    }
+  }
+
+  *ppShm = p->apShm[iChunk];
+  return LSM_OK;
+}
+
+static void lsmPosixOsShmBarrier(void){
+}
+
+static int lsmPosixOsShmUnmap(lsm_file *pFile, int bDelete){
+  PosixFile *p = (PosixFile *)pFile;
+  if( p->shmfd>0 ){
+    int i;
+    for(i=0; i<p->nShm; i++){
+      if( p->apShm[i] ){
+        munmap(p->apShm[i], LSM_SHM_CHUNK_SIZE);
+        p->apShm[i] = 0;
+      }
+    }
+    close(p->shmfd);
+    p->shmfd = 0;
+    if( bDelete ){
+      char *zShm = posixShmFile(p);
+      if( zShm ) unlink(zShm);
+      lsmFree(p->pEnv, zShm);
+    }
+  }
+  return LSM_OK;
+}
+
+
+static int lsmPosixOsClose(lsm_file *pFile){
+   PosixFile *p = (PosixFile *)pFile;
+   lsmPosixOsShmUnmap(pFile, 0);
+   if( p->pMap ) munmap(p->pMap, p->nMap);
+   close(p->fd);
+   lsm_free(p->pEnv, p->apShm);
+   lsm_free(p->pEnv, p);
+   return LSM_OK;
+}
+
+static int lsmPosixOsSleep(lsm_env *pEnv, int us){
+#if 0
+  /* Apparently on Android usleep() returns void */
+  if( usleep(us) ) return LSM_IOERR;
+#endif
+  usleep(us);
+  return LSM_OK;
+}
+
+/****************************************************************************
+** Memory allocation routines.
+*/
+#define BLOCK_HDR_SIZE ROUND8( sizeof(size_t) )
+
+static void *lsmPosixOsMalloc(lsm_env *pEnv, size_t N){
+  unsigned char * m;
+  N += BLOCK_HDR_SIZE;
+  m = (unsigned char *)malloc(N);
+  *((size_t*)m) = N;
+  return m + BLOCK_HDR_SIZE;
+}
+
+static void lsmPosixOsFree(lsm_env *pEnv, void *p){
+  if(p){
+    free( ((unsigned char *)p) - BLOCK_HDR_SIZE );
+  }
+}
+
+static void *lsmPosixOsRealloc(lsm_env *pEnv, void *p, size_t N){
+  unsigned char * m = (unsigned char *)p;
+  if(1>N){
+    lsmPosixOsFree( pEnv, p );
+    return NULL;
+  }else if(NULL==p){
+    return lsmPosixOsMalloc(pEnv, N);
+  }else{
+    void * re = NULL;
+    m -= BLOCK_HDR_SIZE;
+#if 0 /* arguable: don't shrink */
+    size_t * sz = (size_t*)m;
+    if(*sz >= (size_t)N){
+      return p;
+    }
+#endif
+    re = realloc( m, N + BLOCK_HDR_SIZE );
+    if(re){
+      m = (unsigned char *)re;
+      *((size_t*)m) = N;
+      return m + BLOCK_HDR_SIZE;
+    }else{
+      return NULL;
+    }
+  }
+}
+
+static size_t lsmPosixOsMSize(lsm_env *pEnv, void *p){
+  unsigned char * m = (unsigned char *)p;
+  return *((size_t*)(m-BLOCK_HDR_SIZE));
+}
+#undef BLOCK_HDR_SIZE
+
+
+#ifdef LSM_MUTEX_PTHREADS 
+/*************************************************************************
+** Mutex methods for pthreads based systems.  If LSM_MUTEX_PTHREADS is
+** missing then a no-op implementation of mutexes found in lsm_mutex.c
+** will be used instead.
+*/
+#include <pthread.h>
+
+typedef struct PthreadMutex PthreadMutex;
+struct PthreadMutex {
+  lsm_env *pEnv;
+  pthread_mutex_t mutex;
+#ifdef LSM_DEBUG
+  pthread_t owner;
+#endif
+};
+
+#ifdef LSM_DEBUG
+# define LSM_PTHREAD_STATIC_MUTEX { 0, PTHREAD_MUTEX_INITIALIZER, 0 }
+#else
+# define LSM_PTHREAD_STATIC_MUTEX { 0, PTHREAD_MUTEX_INITIALIZER }
+#endif
+
+static int lsmPosixOsMutexStatic(
+  lsm_env *pEnv,
+  int iMutex,
+  lsm_mutex **ppStatic
+){
+  static PthreadMutex sMutex[2] = {
+    LSM_PTHREAD_STATIC_MUTEX,
+    LSM_PTHREAD_STATIC_MUTEX
+  };
+
+  assert( iMutex==LSM_MUTEX_GLOBAL || iMutex==LSM_MUTEX_HEAP );
+  assert( LSM_MUTEX_GLOBAL==1 && LSM_MUTEX_HEAP==2 );
+
+  *ppStatic = (lsm_mutex *)&sMutex[iMutex-1];
+  return LSM_OK;
+}
+
+static int lsmPosixOsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
+  PthreadMutex *pMutex;           /* Pointer to new mutex */
+  pthread_mutexattr_t attr;       /* Attributes object */
+
+  pMutex = (PthreadMutex *)lsmMallocZero(pEnv, sizeof(PthreadMutex));
+  if( !pMutex ) return LSM_NOMEM_BKPT;
+
+  pMutex->pEnv = pEnv;
+  pthread_mutexattr_init(&attr);
+  pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
+  pthread_mutex_init(&pMutex->mutex, &attr);
+  pthread_mutexattr_destroy(&attr);
+
+  *ppNew = (lsm_mutex *)pMutex;
+  return LSM_OK;
+}
+
+static void lsmPosixOsMutexDel(lsm_mutex *p){
+  PthreadMutex *pMutex = (PthreadMutex *)p;
+  pthread_mutex_destroy(&pMutex->mutex);
+  lsmFree(pMutex->pEnv, pMutex);
+}
+
+static void lsmPosixOsMutexEnter(lsm_mutex *p){
+  PthreadMutex *pMutex = (PthreadMutex *)p;
+  pthread_mutex_lock(&pMutex->mutex);
+
+#ifdef LSM_DEBUG
+  assert( !pthread_equal(pMutex->owner, pthread_self()) );
+  pMutex->owner = pthread_self();
+  assert( pthread_equal(pMutex->owner, pthread_self()) );
+#endif
+}
+
+static int lsmPosixOsMutexTry(lsm_mutex *p){
+  int ret;
+  PthreadMutex *pMutex = (PthreadMutex *)p;
+  ret = pthread_mutex_trylock(&pMutex->mutex);
+#ifdef LSM_DEBUG
+  if( ret==0 ){
+    assert( !pthread_equal(pMutex->owner, pthread_self()) );
+    pMutex->owner = pthread_self();
+    assert( pthread_equal(pMutex->owner, pthread_self()) );
+  }
+#endif
+  return ret;
+}
+
+static void lsmPosixOsMutexLeave(lsm_mutex *p){
+  PthreadMutex *pMutex = (PthreadMutex *)p;
+#ifdef LSM_DEBUG
+  assert( pthread_equal(pMutex->owner, pthread_self()) );
+  pMutex->owner = 0;
+  assert( !pthread_equal(pMutex->owner, pthread_self()) );
+#endif
+  pthread_mutex_unlock(&pMutex->mutex);
+}
+
+#ifdef LSM_DEBUG
+static int lsmPosixOsMutexHeld(lsm_mutex *p){
+  PthreadMutex *pMutex = (PthreadMutex *)p;
+  return pMutex ? pthread_equal(pMutex->owner, pthread_self()) : 1;
+}
+static int lsmPosixOsMutexNotHeld(lsm_mutex *p){
+  PthreadMutex *pMutex = (PthreadMutex *)p;
+  return pMutex ? !pthread_equal(pMutex->owner, pthread_self()) : 1;
+}
+#endif
+/*
+** End of pthreads mutex implementation.
+*************************************************************************/
+#else
+/*************************************************************************
+** Noop mutex implementation
+*/
+typedef struct NoopMutex NoopMutex;
+struct NoopMutex {
+  lsm_env *pEnv;                  /* Environment handle (for xFree()) */
+  int bHeld;                      /* True if mutex is held */
+  int bStatic;                    /* True for a static mutex */
+};
+static NoopMutex aStaticNoopMutex[2] = {
+  {0, 0, 1},
+  {0, 0, 1},
+};
+
+static int lsmPosixOsMutexStatic(
+  lsm_env *pEnv,
+  int iMutex,
+  lsm_mutex **ppStatic
+){
+  assert( iMutex>=1 && iMutex<=(int)array_size(aStaticNoopMutex) );
+  *ppStatic = (lsm_mutex *)&aStaticNoopMutex[iMutex-1];
+  return LSM_OK;
+}
+static int lsmPosixOsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
+  NoopMutex *p;
+  p = (NoopMutex *)lsmMallocZero(pEnv, sizeof(NoopMutex));
+  if( p ) p->pEnv = pEnv;
+  *ppNew = (lsm_mutex *)p;
+  return (p ? LSM_OK : LSM_NOMEM_BKPT);
+}
+static void lsmPosixOsMutexDel(lsm_mutex *pMutex)  { 
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bStatic==0 && p->pEnv );
+  lsmFree(p->pEnv, p);
+}
+static void lsmPosixOsMutexEnter(lsm_mutex *pMutex){ 
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bHeld==0 );
+  p->bHeld = 1;
+}
+static int lsmPosixOsMutexTry(lsm_mutex *pMutex){
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bHeld==0 );
+  p->bHeld = 1;
+  return 0;
+}
+static void lsmPosixOsMutexLeave(lsm_mutex *pMutex){ 
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bHeld==1 );
+  p->bHeld = 0;
+}
+#ifdef LSM_DEBUG
+static int lsmPosixOsMutexHeld(lsm_mutex *pMutex){ 
+  NoopMutex *p = (NoopMutex *)pMutex;
+  return p ? p->bHeld : 1;
+}
+static int lsmPosixOsMutexNotHeld(lsm_mutex *pMutex){ 
+  NoopMutex *p = (NoopMutex *)pMutex;
+  return p ? !p->bHeld : 1;
+}
+#endif
+/***************************************************************************/
+#endif /* else LSM_MUTEX_NONE */
+
+/* Without LSM_DEBUG, the MutexHeld tests are never called */
+#ifndef LSM_DEBUG
+# define lsmPosixOsMutexHeld    0
+# define lsmPosixOsMutexNotHeld 0
+#endif
+
+lsm_env *lsm_default_env(void){
+  static lsm_env posix_env = {
+    sizeof(lsm_env),         /* nByte */
+    1,                       /* iVersion */
+    /***** file i/o ******************/
+    0,                       /* pVfsCtx */
+    lsmPosixOsFullpath,      /* xFullpath */
+    lsmPosixOsOpen,          /* xOpen */
+    lsmPosixOsRead,          /* xRead */
+    lsmPosixOsWrite,         /* xWrite */
+    lsmPosixOsTruncate,      /* xTruncate */
+    lsmPosixOsSync,          /* xSync */
+    lsmPosixOsSectorSize,    /* xSectorSize */
+    lsmPosixOsRemap,         /* xRemap */
+    lsmPosixOsFileid,        /* xFileid */
+    lsmPosixOsClose,         /* xClose */
+    lsmPosixOsUnlink,        /* xUnlink */
+    lsmPosixOsLock,          /* xLock */
+    lsmPosixOsTestLock,      /* xTestLock */
+    lsmPosixOsShmMap,        /* xShmMap */
+    lsmPosixOsShmBarrier,    /* xShmBarrier */
+    lsmPosixOsShmUnmap,      /* xShmUnmap */
+    /***** memory allocation *********/
+    0,                       /* pMemCtx */
+    lsmPosixOsMalloc,        /* xMalloc */
+    lsmPosixOsRealloc,       /* xRealloc */
+    lsmPosixOsFree,          /* xFree */
+    lsmPosixOsMSize,         /* xSize */
+    /***** mutexes *********************/
+    0,                       /* pMutexCtx */
+    lsmPosixOsMutexStatic,   /* xMutexStatic */
+    lsmPosixOsMutexNew,      /* xMutexNew */
+    lsmPosixOsMutexDel,      /* xMutexDel */
+    lsmPosixOsMutexEnter,    /* xMutexEnter */
+    lsmPosixOsMutexTry,      /* xMutexTry */
+    lsmPosixOsMutexLeave,    /* xMutexLeave */
+    lsmPosixOsMutexHeld,     /* xMutexHeld */
+    lsmPosixOsMutexNotHeld,  /* xMutexNotHeld */
+    /***** other *********************/
+    lsmPosixOsSleep,         /* xSleep */
+  };
+  return &posix_env;
+}
+
+#endif
diff --git a/ext/lsm1/lsm_varint.c b/ext/lsm1/lsm_varint.c
new file mode 100644
index 0000000..c550a64
--- /dev/null
+++ b/ext/lsm1/lsm_varint.c
@@ -0,0 +1,196 @@
+
+/*
+** 2012-02-08
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** SQLite4-compatible varint implementation.
+*/
+#include "lsmInt.h"
+
+/*************************************************************************
+** The following is a copy of the varint.c module from SQLite 4.
+*/
+
+/*
+** Decode the varint in z[].  Write the integer value into *pResult and
+** return the number of bytes in the varint.
+*/
+static int lsmSqlite4GetVarint64(const unsigned char *z, u64 *pResult){
+  unsigned int x;
+  if( z[0]<=240 ){
+    *pResult = z[0];
+    return 1;
+  }
+  if( z[0]<=248 ){
+    *pResult = (z[0]-241)*256 + z[1] + 240;
+    return 2;
+  }
+  if( z[0]==249 ){
+    *pResult = 2288 + 256*z[1] + z[2];
+    return 3;
+  }
+  if( z[0]==250 ){
+    *pResult = (z[1]<<16) + (z[2]<<8) + z[3];
+    return 4;
+  }
+  x = (z[1]<<24) + (z[2]<<16) + (z[3]<<8) + z[4];
+  if( z[0]==251 ){
+    *pResult = x;
+    return 5;
+  }
+  if( z[0]==252 ){
+    *pResult = (((u64)x)<<8) + z[5];
+    return 6;
+  }
+  if( z[0]==253 ){
+    *pResult = (((u64)x)<<16) + (z[5]<<8) + z[6];
+    return 7;
+  }
+  if( z[0]==254 ){
+    *pResult = (((u64)x)<<24) + (z[5]<<16) + (z[6]<<8) + z[7];
+    return 8;
+  }
+  *pResult = (((u64)x)<<32) +
+               (0xffffffff & ((z[5]<<24) + (z[6]<<16) + (z[7]<<8) + z[8]));
+  return 9;
+}
+
+/*
+** Write a 32-bit unsigned integer as 4 big-endian bytes.
+*/
+static void lsmVarintWrite32(unsigned char *z, unsigned int y){
+  z[0] = (unsigned char)(y>>24);
+  z[1] = (unsigned char)(y>>16);
+  z[2] = (unsigned char)(y>>8);
+  z[3] = (unsigned char)(y);
+}
+
+/*
+** Write a varint into z[].  The buffer z[] must be at least 9 characters
+** long to accommodate the largest possible varint.  Return the number of
+** bytes of z[] used.
+*/
+static int lsmSqlite4PutVarint64(unsigned char *z, u64 x){
+  unsigned int w, y;
+  if( x<=240 ){
+    z[0] = (unsigned char)x;
+    return 1;
+  }
+  if( x<=2287 ){
+    y = (unsigned int)(x - 240);
+    z[0] = (unsigned char)(y/256 + 241);
+    z[1] = (unsigned char)(y%256);
+    return 2;
+  }
+  if( x<=67823 ){
+    y = (unsigned int)(x - 2288);
+    z[0] = 249;
+    z[1] = (unsigned char)(y/256);
+    z[2] = (unsigned char)(y%256);
+    return 3;
+  }
+  y = (unsigned int)x;
+  w = (unsigned int)(x>>32);
+  if( w==0 ){
+    if( y<=16777215 ){
+      z[0] = 250;
+      z[1] = (unsigned char)(y>>16);
+      z[2] = (unsigned char)(y>>8);
+      z[3] = (unsigned char)(y);
+      return 4;
+    }
+    z[0] = 251;
+    lsmVarintWrite32(z+1, y);
+    return 5;
+  }
+  if( w<=255 ){
+    z[0] = 252;
+    z[1] = (unsigned char)w;
+    lsmVarintWrite32(z+2, y);
+    return 6;
+  }
+  if( w<=32767 ){
+    z[0] = 253;
+    z[1] = (unsigned char)(w>>8);
+    z[2] = (unsigned char)w;
+    lsmVarintWrite32(z+3, y);
+    return 7;
+  }
+  if( w<=16777215 ){
+    z[0] = 254;
+    z[1] = (unsigned char)(w>>16);
+    z[2] = (unsigned char)(w>>8);
+    z[3] = (unsigned char)w;
+    lsmVarintWrite32(z+4, y);
+    return 8;
+  }
+  z[0] = 255;
+  lsmVarintWrite32(z+1, w);
+  lsmVarintWrite32(z+5, y);
+  return 9;
+}
+
+/*
+** End of SQLite 4 code.
+*************************************************************************/
+
+int lsmVarintPut64(u8 *aData, i64 iVal){
+  return lsmSqlite4PutVarint64(aData, (u64)iVal);
+}
+
+int lsmVarintGet64(const u8 *aData, i64 *piVal){
+  return lsmSqlite4GetVarint64(aData, (u64 *)piVal);
+}
+
+int lsmVarintPut32(u8 *aData, int iVal){
+  return lsmSqlite4PutVarint64(aData, (u64)iVal);
+}
+
+int lsmVarintGet32(u8 *z, int *piVal){
+  u64 i;
+  int ret;
+
+  if( z[0]<=240 ){
+    *piVal = z[0];
+    return 1;
+  }
+  if( z[0]<=248 ){
+    *piVal = (z[0]-241)*256 + z[1] + 240;
+    return 2;
+  }
+  if( z[0]==249 ){
+    *piVal = 2288 + 256*z[1] + z[2];
+    return 3;
+  }
+  if( z[0]==250 ){
+    *piVal = (z[1]<<16) + (z[2]<<8) + z[3];
+    return 4;
+  }
+
+  ret = lsmSqlite4GetVarint64(z, &i);
+  *piVal = (int)i;
+  return ret;
+}
+
+int lsmVarintLen32(int n){
+  u8 aData[9];
+  return lsmVarintPut32(aData, n);
+}
+
+/*
+** The argument is the first byte of a varint. This function returns the
+** total number of bytes in the entire varint (including the first byte).
+*/
+int lsmVarintSize(u8 c){
+  if( c<241 ) return 1;
+  if( c<249 ) return 2;
+  return (int)(c - 246);
+}
diff --git a/ext/lsm1/lsm_vtab.c b/ext/lsm1/lsm_vtab.c
new file mode 100644
index 0000000..bb14602
--- /dev/null
+++ b/ext/lsm1/lsm_vtab.c
@@ -0,0 +1,1079 @@
+/*
+** 2015-11-16
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** This file implements a virtual table for SQLite3 around the LSM
+** storage engine from SQLite4.
+**
+** USAGE
+**
+**   CREATE VIRTUAL TABLE demo USING lsm1(filename,key,keytype,value1,...);
+**
+** The filename parameter is the name of the LSM database file, which is
+** separate and distinct from the SQLite3 database file.
+**
+** The keytype must be one of: UINT, TEXT, BLOB.  All keys must be of that
+** one type.  "UINT" means unsigned integer.  The values may be of any
+** SQLite datatype: BLOB, TEXT, INTEGER, FLOAT, or NULL.
+**
+** The virtual table contains read-only hidden columns:
+**
+**     lsm1_key       A BLOB which is the raw LSM key.  If the "keytype"
+**                    is BLOB or TEXT then this column is exactly the
+**                    same as the key.  For the UINT keytype, this column
+**                    will be a variable-length integer encoding of the key.
+**
+**     lsm1_value     A BLOB which is the raw LSM value.  All of the value
+**                    columns are packed into this BLOB using the encoding
+**                    described below.
+**
+** Attempts to write values into the lsm1_key and lsm1_value columns are
+** silently ignored.
+**
+** EXAMPLE
+**
+** The virtual table declared this way:
+**
+**    CREATE VIRTUAL TABLE demo2 USING lsm1('x.lsm',id,UINT,a,b,c,d);
+**
+** Results in a new virtual table named "demo2" that acts as if it has
+** the following schema:
+**
+**    CREATE TABLE demo2(
+**      id UINT PRIMARY KEY ON CONFLICT REPLACE,
+**      a ANY,
+**      b ANY,
+**      c ANY,
+**      d ANY,
+**      lsm1_key BLOB HIDDEN,
+**      lsm1_value BLOB HIDDEN
+**    ) WITHOUT ROWID;
+**
+** 
+**
+** INTERNALS
+**
+** The key encoding for BLOB and TEXT is just a copy of the blob or text.
+** UTF-8 is used for text.  The key encoding for UINT is the variable-length
+** integer format at https://sqlite.org/src4/doc/trunk/www/varint.wiki.
+**
+** The values are encoded as a single blob (since that is what lsm stores as
+** its content).  There is a "type integer" followed by "content" for each
+** value, alternating back and forth.  The content might be empty.
+**
+**    TYPE1  CONTENT1  TYPE2  CONTENT2  TYPE3  CONTENT3 ....
+**
+** Each "type integer" is encoded as a variable-length integer in the
+** format of the link above.  Let the type integer be T.  The actual
+** datatype is an integer 0-5 equal to T%6.  Values 1 through 5 correspond
+** to SQLITE_INTEGER through SQLITE_NULL.  The size of the content in bytes
+** is T/6.  Type value 0 means that the value is an integer whose actual
+** values is T/6 and there is no content.  The type-value-0 integer format
+** only works for integers in the range of 0 through 40.
+**
+** There is no content for NULL or type-0 integers.  For BLOB and TEXT
+** values, the content is the blob data or the UTF-8 text data.  For
+** non-negative integers X, the content is a variable-length integer X*2.
+** For negative integers Y, the content is varaible-length integer (1-Y)*2+1.
+** For FLOAT values, the content is the IEEE754 floating point value in
+** native byte-order.  This means that FLOAT values will be corrupted when
+** database file is moved between big-endian and little-endian machines.
+*/
+#include "sqlite3ext.h"
+SQLITE_EXTENSION_INIT1
+#include "lsm.h"
+#include <assert.h>
+#include <string.h>
+
+/* Forward declaration of subclasses of virtual table objects */
+typedef struct lsm1_vtab lsm1_vtab;
+typedef struct lsm1_cursor lsm1_cursor;
+typedef struct lsm1_vblob lsm1_vblob;
+
+/* Primitive types */
+typedef unsigned char u8;
+typedef unsigned int u32;
+typedef sqlite3_uint64 u64;
+
+/* An open connection to an LSM table */
+struct lsm1_vtab {
+  sqlite3_vtab base;          /* Base class - must be first */
+  lsm_db *pDb;                /* Open connection to the LSM table */
+  u8 keyType;                 /* SQLITE_BLOB, _TEXT, or _INTEGER */
+  u32 nVal;                   /* Number of value columns */
+};
+
+
+/* lsm1_cursor is a subclass of sqlite3_vtab_cursor which will
+** serve as the underlying representation of a cursor that scans
+** over rows of the result
+*/
+struct lsm1_cursor {
+  sqlite3_vtab_cursor base;  /* Base class - must be first */
+  lsm_cursor *pLsmCur;       /* The LSM cursor */
+  u8 isDesc;                 /* 0: scan forward.  1: scan reverse */
+  u8 atEof;                  /* True if the scan is complete */
+  u8 bUnique;                /* True if no more than one row of output */
+  u8 *zData;                 /* Content of the current row */
+  u32 nData;                 /* Number of bytes in the current row */
+  u8 *aeType;                /* Types for all column values */
+  u32 *aiOfst;               /* Offsets to the various fields */
+  u32 *aiLen;                /* Length of each field */
+  u8 *pKey2;                 /* Loop termination key, or NULL */
+  u32 nKey2;                 /* Length of the loop termination key */
+};
+
+/* An extensible buffer object.
+**
+** Content can be appended.  Space to hold new content is automatically
+** allocated.
+*/
+struct lsm1_vblob {
+  u8 *a;             /* Space to hold content, from sqlite3_malloc64() */
+  u64 n;             /* Bytes of space used */
+  u64 nAlloc;        /* Bytes of space allocated */
+  u8 errNoMem;       /* True if a memory allocation error has been seen */
+};
+
+#if defined(__GNUC__)
+#  define LSM1_NOINLINE  __attribute__((noinline))
+#elif defined(_MSC_VER) && _MSC_VER>=1310
+#  define LSM1_NOINLINE  __declspec(noinline)
+#else
+#  define LSM1_NOINLINE
+#endif
+
+
+/* Increase the available space in the vblob object so that it can hold
+** at least N more bytes.  Return the number of errors.
+*/
+static int lsm1VblobEnlarge(lsm1_vblob *p, u32 N){
+  if( p->n+N>p->nAlloc ){
+    if( p->errNoMem ) return 1;
+    p->nAlloc += N + (p->nAlloc ? p->nAlloc : N);
+    p->a = sqlite3_realloc64(p->a, p->nAlloc);
+    if( p->a==0 ){
+      p->n = 0;
+      p->nAlloc = 0;
+      p->errNoMem = 1;
+      return 1;
+    }
+    p->nAlloc = sqlite3_msize(p->a);
+  }
+  return 0;
+}
+
+/* Append N bytes to a vblob after first enlarging it */
+static LSM1_NOINLINE void lsm1VblobEnlargeAndAppend(
+  lsm1_vblob *p,
+  const u8 *pData,
+  u32 N
+){
+  if( p->n+N>p->nAlloc && lsm1VblobEnlarge(p, N) ) return;
+  memcpy(p->a+p->n, pData, N);
+  p->n += N;
+}
+
+/* Append N bytes to a vblob */
+static void lsm1VblobAppend(lsm1_vblob *p, const u8 *pData, u32 N){
+  sqlite3_int64 n = p->n;
+  if( n+N>p->nAlloc ){
+    lsm1VblobEnlargeAndAppend(p, pData, N);
+  }else{
+    p->n += N;
+    memcpy(p->a+n, pData, N);
+  }
+}
+
+/* append text to a vblob */
+static void lsm1VblobAppendText(lsm1_vblob *p, const char *z){
+  lsm1VblobAppend(p, (u8*)z, (u32)strlen(z));
+}
+
+/* Dequote the string */
+static void lsm1Dequote(char *z){
+  int j;
+  char cQuote = z[0];
+  size_t i, n;
+
+  if( cQuote!='\'' && cQuote!='"' ) return;
+  n = strlen(z);
+  if( n<2 || z[n-1]!=z[0] ) return;
+  for(i=1, j=0; i<n-1; i++){
+    if( z[i]==cQuote && z[i+1]==cQuote ) i++;
+    z[j++] = z[i];
+  }
+  z[j] = 0;
+}
+
+
+/*
+** The lsm1Connect() method is invoked to create a new
+** lsm1_vtab that describes the virtual table.
+*/
+static int lsm1Connect(
+  sqlite3 *db,
+  void *pAux,
+  int argc, const char *const*argv,
+  sqlite3_vtab **ppVtab,
+  char **pzErr
+){
+  lsm1_vtab *pNew;
+  int rc;
+  char *zFilename;
+  u8 keyType = 0;
+  int i;
+  lsm1_vblob sql;
+  static const char *azTypes[] = { "UINT",         "TEXT",     "BLOB" };
+  static const u8 aeTypes[] =    { SQLITE_INTEGER, SQLITE_TEXT, SQLITE_BLOB };
+  static const char *azArgName[] = {"filename", "key", "key type", "value1" };
+
+  for(i=0; i<sizeof(azArgName)/sizeof(azArgName[0]); i++){
+    if( argc<i+4 || argv[i+3]==0 || argv[i+3][0]==0 ){
+      *pzErr = sqlite3_mprintf("%s (%r) argument missing",
+                               azArgName[i], i+1);
+      return SQLITE_ERROR;
+    }
+  }
+  for(i=0; i<sizeof(azTypes)/sizeof(azTypes[0]); i++){
+    if( sqlite3_stricmp(azTypes[i],argv[5])==0 ){
+      keyType = aeTypes[i];
+      break;
+    }
+  }
+  if( keyType==0 ){
+    *pzErr = sqlite3_mprintf("key type should be INT, TEXT, or BLOB");
+    return SQLITE_ERROR;
+  }
+  *ppVtab = sqlite3_malloc( sizeof(*pNew) );
+  pNew = (lsm1_vtab*)*ppVtab;
+  if( pNew==0 ){
+    return SQLITE_NOMEM;
+  }
+  memset(pNew, 0, sizeof(*pNew));
+  pNew->keyType = keyType;
+  rc = lsm_new(0, &pNew->pDb);
+  if( rc ){
+    *pzErr = sqlite3_mprintf("lsm_new failed with error code %d",  rc);
+    rc = SQLITE_ERROR;
+    goto connect_failed;
+  }
+  zFilename = sqlite3_mprintf("%s", argv[3]);
+  lsm1Dequote(zFilename);
+  rc = lsm_open(pNew->pDb, zFilename);
+  sqlite3_free(zFilename);
+  if( rc ){
+    *pzErr = sqlite3_mprintf("lsm_open failed with %d", rc);
+    rc = SQLITE_ERROR;
+    goto connect_failed;
+  }
+
+  memset(&sql, 0, sizeof(sql));
+  lsm1VblobAppendText(&sql, "CREATE TABLE x(");
+  lsm1VblobAppendText(&sql, argv[4]);
+  lsm1VblobAppendText(&sql, " ");
+  lsm1VblobAppendText(&sql, argv[5]);
+  lsm1VblobAppendText(&sql, " PRIMARY KEY");
+  for(i=6; i<argc; i++){
+    lsm1VblobAppendText(&sql, ", ");
+    lsm1VblobAppendText(&sql, argv[i]);
+    pNew->nVal++;
+  }
+  lsm1VblobAppendText(&sql, 
+      ", lsm1_command HIDDEN"
+      ", lsm1_key HIDDEN"
+      ", lsm1_value HIDDEN) WITHOUT ROWID");
+  lsm1VblobAppend(&sql, (u8*)"", 1);
+  if( sql.errNoMem ){
+    rc = SQLITE_NOMEM;
+    goto connect_failed;
+  }
+  rc = sqlite3_declare_vtab(db, (const char*)sql.a);
+  sqlite3_free(sql.a);
+
+connect_failed:
+  if( rc!=SQLITE_OK ){
+    if( pNew ){
+      if( pNew->pDb ) lsm_close(pNew->pDb);
+      sqlite3_free(pNew);
+    }
+    *ppVtab = 0;
+  }
+  return rc;
+}
+
+/*
+** This method is the destructor for lsm1_cursor objects.
+*/
+static int lsm1Disconnect(sqlite3_vtab *pVtab){
+  lsm1_vtab *p = (lsm1_vtab*)pVtab;
+  lsm_close(p->pDb);
+  sqlite3_free(p);
+  return SQLITE_OK;
+}
+
+/*
+** Constructor for a new lsm1_cursor object.
+*/
+static int lsm1Open(sqlite3_vtab *pVtab, sqlite3_vtab_cursor **ppCursor){
+  lsm1_vtab *p = (lsm1_vtab*)pVtab;
+  lsm1_cursor *pCur;
+  int rc;
+  pCur = sqlite3_malloc64( sizeof(*pCur)
+                 + p->nVal*(sizeof(pCur->aiOfst)+sizeof(pCur->aiLen)+1) );
+  if( pCur==0 ) return SQLITE_NOMEM;
+  memset(pCur, 0, sizeof(*pCur));
+  pCur->aiOfst = (u32*)&pCur[1];
+  pCur->aiLen = &pCur->aiOfst[p->nVal];
+  pCur->aeType = (u8*)&pCur->aiLen[p->nVal];
+  *ppCursor = &pCur->base;
+  rc = lsm_csr_open(p->pDb, &pCur->pLsmCur);
+  if( rc==LSM_OK ){
+    rc = SQLITE_OK;
+  }else{
+    sqlite3_free(pCur);
+    *ppCursor = 0;
+    rc = SQLITE_ERROR;
+  }
+  return rc;
+}
+
+/*
+** Destructor for a lsm1_cursor.
+*/
+static int lsm1Close(sqlite3_vtab_cursor *cur){
+  lsm1_cursor *pCur = (lsm1_cursor*)cur;
+  sqlite3_free(pCur->pKey2);
+  lsm_csr_close(pCur->pLsmCur);
+  sqlite3_free(pCur);
+  return SQLITE_OK;
+}
+
+
+/*
+** Advance a lsm1_cursor to its next row of output.
+*/
+static int lsm1Next(sqlite3_vtab_cursor *cur){
+  lsm1_cursor *pCur = (lsm1_cursor*)cur;
+  int rc = LSM_OK;
+  if( pCur->bUnique ){
+    pCur->atEof = 1;
+  }else{
+    if( pCur->isDesc ){
+      rc = lsm_csr_prev(pCur->pLsmCur);
+    }else{
+      rc = lsm_csr_next(pCur->pLsmCur);
+    }
+    if( rc==LSM_OK && lsm_csr_valid(pCur->pLsmCur)==0 ){
+      pCur->atEof = 1;
+    }
+    if( pCur->pKey2 && pCur->atEof==0 ){
+      const u8 *pVal;
+      u32 nVal;
+      assert( pCur->isDesc==0 );
+      rc = lsm_csr_key(pCur->pLsmCur, (const void**)&pVal, (int*)&nVal);
+      if( rc==LSM_OK ){
+        u32 len = pCur->nKey2;
+        int c;
+        if( len>nVal ) len = nVal;
+        c = memcmp(pVal, pCur->pKey2, len);
+        if( c==0 ) c = nVal - pCur->nKey2;
+        if( c>0 ) pCur->atEof = 1;
+      }
+    }
+    pCur->zData = 0;
+  }
+  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
+}
+
+/*
+** Return TRUE if the cursor has been moved off of the last
+** row of output.
+*/
+static int lsm1Eof(sqlite3_vtab_cursor *cur){
+  lsm1_cursor *pCur = (lsm1_cursor*)cur;
+  return pCur->atEof;
+}
+
+/*
+** Rowids are not supported by the underlying virtual table.  So always
+** return 0 for the rowid.
+*/
+static int lsm1Rowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){
+  *pRowid = 0;
+  return SQLITE_OK;
+}
+
+/*
+** Type prefixes on LSM keys
+*/
+#define LSM1_TYPE_NEGATIVE   0
+#define LSM1_TYPE_POSITIVE   1
+#define LSM1_TYPE_TEXT       2
+#define LSM1_TYPE_BLOB       3
+
+/*
+** Write a 32-bit unsigned integer as 4 big-endian bytes.
+*/
+static void varintWrite32(unsigned char *z, unsigned int y){
+  z[0] = (unsigned char)(y>>24);
+  z[1] = (unsigned char)(y>>16);
+  z[2] = (unsigned char)(y>>8);
+  z[3] = (unsigned char)(y);
+}
+
+/*
+** Write a varint into z[].  The buffer z[] must be at least 9 characters
+** long to accommodate the largest possible varint.  Return the number of
+** bytes of z[] used.
+*/
+static int lsm1PutVarint64(unsigned char *z, sqlite3_uint64 x){
+  unsigned int w, y;
+  if( x<=240 ){
+    z[0] = (unsigned char)x;
+    return 1;
+  }
+  if( x<=2287 ){
+    y = (unsigned int)(x - 240);
+    z[0] = (unsigned char)(y/256 + 241);
+    z[1] = (unsigned char)(y%256);
+    return 2;
+  }
+  if( x<=67823 ){
+    y = (unsigned int)(x - 2288);
+    z[0] = 249;
+    z[1] = (unsigned char)(y/256);
+    z[2] = (unsigned char)(y%256);
+    return 3;
+  }
+  y = (unsigned int)x;
+  w = (unsigned int)(x>>32);
+  if( w==0 ){
+    if( y<=16777215 ){
+      z[0] = 250;
+      z[1] = (unsigned char)(y>>16);
+      z[2] = (unsigned char)(y>>8);
+      z[3] = (unsigned char)(y);
+      return 4;
+    }
+    z[0] = 251;
+    varintWrite32(z+1, y);
+    return 5;
+  }
+  if( w<=255 ){
+    z[0] = 252;
+    z[1] = (unsigned char)w;
+    varintWrite32(z+2, y);
+    return 6;
+  }
+  if( w<=65535 ){
+    z[0] = 253;
+    z[1] = (unsigned char)(w>>8);
+    z[2] = (unsigned char)w;
+    varintWrite32(z+3, y);
+    return 7;
+  }
+  if( w<=16777215 ){
+    z[0] = 254;
+    z[1] = (unsigned char)(w>>16);
+    z[2] = (unsigned char)(w>>8);
+    z[3] = (unsigned char)w;
+    varintWrite32(z+4, y);
+    return 8;
+  }
+  z[0] = 255;
+  varintWrite32(z+1, w);
+  varintWrite32(z+5, y);
+  return 9;
+}
+
+/* Append non-negative integer x as a variable-length integer.
+*/
+static void lsm1VblobAppendVarint(lsm1_vblob *p, sqlite3_uint64 x){
+  sqlite3_int64 n = p->n;
+  if( n+9>p->nAlloc && lsm1VblobEnlarge(p, 9) ) return;
+  p->n += lsm1PutVarint64(p->a+p->n, x);
+}
+
+/*
+** Decode the varint in the first n bytes z[].  Write the integer value
+** into *pResult and return the number of bytes in the varint.
+**
+** If the decode fails because there are not enough bytes in z[] then
+** return 0;
+*/
+static int lsm1GetVarint64(
+  const unsigned char *z,
+  int n,
+  sqlite3_uint64 *pResult
+){
+  unsigned int x;
+  if( n<1 ) return 0;
+  if( z[0]<=240 ){
+    *pResult = z[0];
+    return 1;
+  }
+  if( z[0]<=248 ){
+    if( n<2 ) return 0;
+    *pResult = (z[0]-241)*256 + z[1] + 240;
+    return 2;
+  }
+  if( n<z[0]-246 ) return 0;
+  if( z[0]==249 ){
+    *pResult = 2288 + 256*z[1] + z[2];
+    return 3;
+  }
+  if( z[0]==250 ){
+    *pResult = (z[1]<<16) + (z[2]<<8) + z[3];
+    return 4;
+  }
+  x = (z[1]<<24) + (z[2]<<16) + (z[3]<<8) + z[4];
+  if( z[0]==251 ){
+    *pResult = x;
+    return 5;
+  }
+  if( z[0]==252 ){
+    *pResult = (((sqlite3_uint64)x)<<8) + z[5];
+    return 6;
+  }
+  if( z[0]==253 ){
+    *pResult = (((sqlite3_uint64)x)<<16) + (z[5]<<8) + z[6];
+    return 7;
+  }
+  if( z[0]==254 ){
+    *pResult = (((sqlite3_uint64)x)<<24) + (z[5]<<16) + (z[6]<<8) + z[7];
+    return 8;
+  }
+  *pResult = (((sqlite3_uint64)x)<<32) +
+               (0xffffffff & ((z[5]<<24) + (z[6]<<16) + (z[7]<<8) + z[8]));
+  return 9;
+}
+
+/* Encoded a signed integer as a varint.  Numbers close to zero uses fewer
+** bytes than numbers far away from zero.  However, the result is not in
+** lexicographical order.
+**
+** Encoding:  Non-negative integer X is encoding as an unsigned
+** varint X*2.  Negative integer Y is encoding as an unsigned
+** varint (1-Y)*2 + 1.
+*/
+static int lsm1PutSignedVarint64(u8 *z, sqlite3_int64 v){
+  sqlite3_uint64 u;
+  if( v>=0 ){
+    u = (sqlite3_uint64)v;
+    return lsm1PutVarint64(z, u*2);
+  }else{
+    u = (sqlite3_uint64)(-1-v);
+    return lsm1PutVarint64(z, u*2+1);
+  }
+}
+
+/* Decoded a signed varint. */
+static int lsm1GetSignedVarint64(
+  const unsigned char *z,
+  int n,
+  sqlite3_int64 *pResult
+){
+  sqlite3_uint64 u = 0;
+  n = lsm1GetVarint64(z, n, &u);
+  if( u&1 ){
+    *pResult = -1 - (sqlite3_int64)(u>>1);
+  }else{
+    *pResult = (sqlite3_int64)(u>>1);
+  }
+  return n;
+}
+
+
+/*
+** Read the value part of the key-value pair and decode it into columns.
+*/
+static int lsm1DecodeValues(lsm1_cursor *pCur){
+  lsm1_vtab *pTab = (lsm1_vtab*)(pCur->base.pVtab);
+  int i, n;
+  int rc;
+  u8 eType;
+  sqlite3_uint64 v;
+
+  if( pCur->zData ) return 1;
+  rc = lsm_csr_value(pCur->pLsmCur, (const void**)&pCur->zData,
+                     (int*)&pCur->nData);
+  if( rc ) return 0;
+  for(i=n=0; i<pTab->nVal; i++){
+    v = 0;
+    n += lsm1GetVarint64(pCur->zData+n, pCur->nData-n, &v);
+    pCur->aeType[i] = eType = (u8)(v%6);
+    if( eType==0 ){
+      pCur->aiOfst[i] = (u32)(v/6);
+      pCur->aiLen[i] = 0;
+    }else{ 
+      pCur->aiOfst[i] = n;
+      n += (pCur->aiLen[i] = (u32)(v/6));
+    }
+    if( n>pCur->nData ) break;
+  }
+  if( i<pTab->nVal ){
+    pCur->zData = 0;
+    return 0;
+  }
+  return 1;
+}
+
+/*
+** Return values of columns for the row at which the lsm1_cursor
+** is currently pointing.
+*/
+static int lsm1Column(
+  sqlite3_vtab_cursor *cur,   /* The cursor */
+  sqlite3_context *ctx,       /* First argument to sqlite3_result_...() */
+  int i                       /* Which column to return */
+){
+  lsm1_cursor *pCur = (lsm1_cursor*)cur;
+  lsm1_vtab *pTab = (lsm1_vtab*)(cur->pVtab);
+  if( i==0 ){
+    /* The key column */
+    const void *pVal;
+    int nVal;
+    if( lsm_csr_key(pCur->pLsmCur, &pVal, &nVal)==LSM_OK ){
+      if( pTab->keyType==SQLITE_BLOB ){
+        sqlite3_result_blob(ctx, pVal, nVal, SQLITE_TRANSIENT);
+      }else if( pTab->keyType==SQLITE_TEXT ){
+        sqlite3_result_text(ctx,(const char*)pVal, nVal, SQLITE_TRANSIENT);
+      }else{
+        const unsigned char *z = (const unsigned char*)pVal;
+        sqlite3_uint64 v1;
+        lsm1GetVarint64(z, nVal, &v1);
+        sqlite3_result_int64(ctx, (sqlite3_int64)v1);
+      }
+    }
+  }else if( i>pTab->nVal ){
+    if( i==pTab->nVal+2 ){  /* lsm1_key */
+      const void *pVal;
+      int nVal;
+      if( lsm_csr_key(pCur->pLsmCur, &pVal, &nVal)==LSM_OK ){
+        sqlite3_result_blob(ctx, pVal, nVal, SQLITE_TRANSIENT);
+      }
+    }else if( i==pTab->nVal+3 ){  /* lsm1_value */
+      const void *pVal;
+      int nVal;
+      if( lsm_csr_value(pCur->pLsmCur, &pVal, &nVal)==LSM_OK ){
+        sqlite3_result_blob(ctx, pVal, nVal, SQLITE_TRANSIENT);
+      }
+    }
+  }else if( lsm1DecodeValues(pCur) ){
+    /* The i-th value column (where leftmost is 1) */
+    const u8 *zData;
+    u32 nData;
+    i--;
+    zData = pCur->zData + pCur->aiOfst[i];
+    nData = pCur->aiLen[i];
+    switch( pCur->aeType[i] ){
+      case 0: {  /* in-line integer */
+        sqlite3_result_int(ctx, pCur->aiOfst[i]);
+        break;
+      }
+      case SQLITE_INTEGER: {
+        sqlite3_int64 v;
+        lsm1GetSignedVarint64(zData, nData, &v);
+        sqlite3_result_int64(ctx, v);
+        break;
+      }
+      case SQLITE_FLOAT: {
+        double v;
+        if( nData==sizeof(v) ){
+          memcpy(&v, zData, sizeof(v));
+          sqlite3_result_double(ctx, v);
+        }
+        break;
+      }
+      case SQLITE_TEXT: {
+        sqlite3_result_text(ctx, (const char*)zData, nData, SQLITE_TRANSIENT);
+        break;
+      }
+      case SQLITE_BLOB: {
+        sqlite3_result_blob(ctx, zData, nData, SQLITE_TRANSIENT);
+        break;
+      }
+      default: {
+         /* A NULL.  Do nothing */
+      }
+    }
+  }
+  return SQLITE_OK;
+}
+
+/* Parameter "pValue" contains an SQL value that is to be used as
+** a key in an LSM table.  The type of the key is determined by
+** "keyType".  Extract the raw bytes used for the key in LSM1.
+*/
+static void lsm1KeyFromValue(
+  int keyType,                 /* The key type */
+  sqlite3_value *pValue,       /* The key value */
+  u8 *pBuf,                    /* Storage space for a generated key */
+  const u8 **ppKey,            /* OUT: the bytes of the key */
+  int *pnKey                   /* OUT: size of the key */
+){
+  if( keyType==SQLITE_BLOB ){
+    *ppKey = (const u8*)sqlite3_value_blob(pValue);
+    *pnKey = sqlite3_value_bytes(pValue);
+  }else if( keyType==SQLITE_TEXT ){
+    *ppKey = (const u8*)sqlite3_value_text(pValue);
+    *pnKey = sqlite3_value_bytes(pValue);
+  }else{
+    sqlite3_int64 v = sqlite3_value_int64(pValue);
+    if( v<0 ) v = 0;
+    *pnKey = lsm1PutVarint64(pBuf, v);
+    *ppKey = pBuf;
+  }
+}
+
+/* Move to the first row to return.
+*/
+static int lsm1Filter(
+  sqlite3_vtab_cursor *pVtabCursor, 
+  int idxNum, const char *idxStr,
+  int argc, sqlite3_value **argv
+){
+  lsm1_cursor *pCur = (lsm1_cursor *)pVtabCursor;
+  lsm1_vtab *pTab = (lsm1_vtab*)(pCur->base.pVtab);
+  int rc = LSM_OK;
+  int seekType = -1;
+  const u8 *pVal = 0;
+  int nVal;
+  u8 keyType = pTab->keyType;
+  u8 aKey1[16];
+
+  pCur->atEof = 1;
+  sqlite3_free(pCur->pKey2);
+  pCur->pKey2 = 0;
+  if( idxNum<99 ){
+    lsm1KeyFromValue(keyType, argv[0], aKey1, &pVal, &nVal);
+  }
+  switch( idxNum ){
+    case 0: {   /* key==argv[0] */
+      assert( argc==1 );
+      seekType = LSM_SEEK_EQ;
+      pCur->isDesc = 0;
+      pCur->bUnique = 1;
+      break;
+    }
+    case 1: {  /* key>=argv[0] AND key<=argv[1] */
+      u8 aKey[12];
+      seekType = LSM_SEEK_GE;
+      pCur->isDesc = 0;
+      pCur->bUnique = 0;
+      if( keyType==SQLITE_INTEGER ){
+        sqlite3_int64 v = sqlite3_value_int64(argv[1]);
+        if( v<0 ) v = 0;
+        pCur->nKey2 = lsm1PutVarint64(aKey, (sqlite3_uint64)v);
+        pCur->pKey2 = sqlite3_malloc( pCur->nKey2 );
+        if( pCur->pKey2==0 ) return SQLITE_NOMEM;
+        memcpy(pCur->pKey2, aKey, pCur->nKey2);
+      }else{
+        pCur->nKey2 = sqlite3_value_bytes(argv[1]);
+        pCur->pKey2 = sqlite3_malloc( pCur->nKey2 );
+        if( pCur->pKey2==0 ) return SQLITE_NOMEM;
+        if( keyType==SQLITE_BLOB ){
+          memcpy(pCur->pKey2, sqlite3_value_blob(argv[1]), pCur->nKey2);
+        }else{
+          memcpy(pCur->pKey2, sqlite3_value_text(argv[1]), pCur->nKey2);
+        }
+      }
+      break;
+    }
+    case 2: {  /* key>=argv[0] */
+      seekType = LSM_SEEK_GE;
+      pCur->isDesc = 0;
+      pCur->bUnique = 0;
+      break;
+    }
+    case 3: {  /* key<=argv[0] */
+      seekType = LSM_SEEK_LE;
+      pCur->isDesc = 1;
+      pCur->bUnique = 0;
+      break;
+    }
+    default: { /* full table scan */
+      pCur->isDesc = 0;
+      pCur->bUnique = 0;
+      break;
+    }
+  }
+  if( pVal ){
+    rc = lsm_csr_seek(pCur->pLsmCur, pVal, nVal, seekType);
+  }else{
+    rc = lsm_csr_first(pCur->pLsmCur);
+  }
+  if( rc==LSM_OK && lsm_csr_valid(pCur->pLsmCur)!=0 ){
+    pCur->atEof = 0;
+  }
+  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
+}
+
+/*
+** Only comparisons against the key are allowed.  The idxNum defines
+** which comparisons are available:
+**
+**     0        key==?1
+**     1        key>=?1 AND key<=?2
+**     2        key>?1 or key>=?1
+**     3        key<?1 or key<=?1
+**    99        Full table scan only
+*/
+static int lsm1BestIndex(
+  sqlite3_vtab *tab,
+  sqlite3_index_info *pIdxInfo
+){
+  int i;                 /* Loop over constraints */
+  int idxNum = 99;       /* The query plan */
+  int nArg = 0;          /* Number of arguments to xFilter */
+  int argIdx = -1;       /* Index of the key== constraint, or -1 if none */
+  int iIdx2 = -1;        /* The index of the second key */
+  int omit1 = 0;
+  int omit2 = 0;
+
+  const struct sqlite3_index_constraint *pConstraint;
+  pConstraint = pIdxInfo->aConstraint;
+  for(i=0; i<pIdxInfo->nConstraint; i++, pConstraint++){
+    if( pConstraint->usable==0 ) continue;
+    if( pConstraint->iColumn!=0 ) continue;
+    switch( pConstraint->op ){
+      case SQLITE_INDEX_CONSTRAINT_EQ: {
+        if( idxNum>0 ){
+          argIdx = i;
+          iIdx2 = -1;
+          idxNum = 0;
+          omit1 = 1;
+        }
+        break;
+      }
+      case SQLITE_INDEX_CONSTRAINT_GE:
+      case SQLITE_INDEX_CONSTRAINT_GT: {
+        if( idxNum==99 ){
+          argIdx = i;
+          idxNum = 2;
+          omit1 = pConstraint->op==SQLITE_INDEX_CONSTRAINT_GE;
+        }else if( idxNum==3 ){
+          iIdx2 = idxNum;
+          omit2 = omit1;
+          argIdx = i;
+          idxNum = 1;
+          omit1 = pConstraint->op==SQLITE_INDEX_CONSTRAINT_GE;
+        }
+        break;
+      }
+      case SQLITE_INDEX_CONSTRAINT_LE:
+      case SQLITE_INDEX_CONSTRAINT_LT: {
+        if( idxNum==99 ){
+          argIdx = i;
+          idxNum = 3;
+          omit1 = pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE;
+        }else if( idxNum==2 ){
+          iIdx2 = i;
+          idxNum = 1;
+          omit1 = pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE;
+        }
+        break;
+      }
+    }
+  }
+  if( argIdx>=0 ){
+    pIdxInfo->aConstraintUsage[argIdx].argvIndex = ++nArg;
+    pIdxInfo->aConstraintUsage[argIdx].omit = omit1;
+  }
+  if( iIdx2>=0 ){
+    pIdxInfo->aConstraintUsage[iIdx2].argvIndex = ++nArg;
+    pIdxInfo->aConstraintUsage[iIdx2].omit = omit2;
+  }
+  if( idxNum==0 ){
+    pIdxInfo->estimatedCost = (double)1;
+    pIdxInfo->estimatedRows = 1;
+    pIdxInfo->orderByConsumed = 1;
+  }else if( idxNum==1 ){
+    pIdxInfo->estimatedCost = (double)100;
+    pIdxInfo->estimatedRows = 100;
+  }else if( idxNum<99 ){
+    pIdxInfo->estimatedCost = (double)5000;
+    pIdxInfo->estimatedRows = 5000;
+  }else{
+    /* Full table scan */
+    pIdxInfo->estimatedCost = (double)2147483647;
+    pIdxInfo->estimatedRows = 2147483647;
+  }
+  pIdxInfo->idxNum = idxNum;
+  return SQLITE_OK;
+}
+
+/*
+** The xUpdate method is normally used for INSERT, REPLACE, UPDATE, and
+** DELETE.  But this virtual table only supports INSERT and REPLACE.
+** DELETE is accomplished by inserting a record with a value of NULL.
+** UPDATE is achieved by using REPLACE.
+*/
+int lsm1Update(
+  sqlite3_vtab *pVTab,
+  int argc,
+  sqlite3_value **argv,
+  sqlite_int64 *pRowid
+){
+  lsm1_vtab *p = (lsm1_vtab*)pVTab;
+  int nKey, nKey2;
+  int i;
+  int rc = LSM_OK;
+  const u8 *pKey, *pKey2;
+  unsigned char aKey[16];
+  unsigned char pSpace[16];
+  lsm1_vblob val;
+
+  if( argc==1 ){
+    /* DELETE the record whose key is argv[0] */
+    lsm1KeyFromValue(p->keyType, argv[0], aKey, &pKey, &nKey);
+    lsm_delete(p->pDb, pKey, nKey);
+    return SQLITE_OK;
+  }
+
+  if( sqlite3_value_type(argv[0])!=SQLITE_NULL ){
+    /* An UPDATE */
+    lsm1KeyFromValue(p->keyType, argv[0], aKey, &pKey, &nKey);
+    lsm1KeyFromValue(p->keyType, argv[1], pSpace, &pKey2, &nKey2);
+    if( nKey!=nKey2 || memcmp(pKey, pKey2, nKey)!=0 ){
+      /* The UPDATE changes the PRIMARY KEY value.  DELETE the old key */
+      lsm_delete(p->pDb, pKey, nKey);
+    }
+    /* Fall through into the INSERT case to complete the UPDATE */
+  }
+
+  /* "INSERT INTO tab(lsm1_command) VALUES('....')" is used to implement
+  ** special commands.
+  */
+  if( sqlite3_value_type(argv[3+p->nVal])!=SQLITE_NULL ){
+    return SQLITE_OK;
+  }
+  lsm1KeyFromValue(p->keyType, argv[2], aKey, &pKey, &nKey);
+  memset(&val, 0, sizeof(val));
+  for(i=0; i<p->nVal; i++){
+    sqlite3_value *pArg = argv[3+i];
+    u8 eType = sqlite3_value_type(pArg);
+    switch( eType ){
+      case SQLITE_NULL: {
+        lsm1VblobAppendVarint(&val, SQLITE_NULL);
+        break;
+      }
+      case SQLITE_INTEGER: {
+        sqlite3_int64 v = sqlite3_value_int64(pArg);
+        if( v>=0 && v<=240/6 ){
+          lsm1VblobAppendVarint(&val, v*6);
+        }else{
+          int n = lsm1PutSignedVarint64(pSpace, v);
+          lsm1VblobAppendVarint(&val, SQLITE_INTEGER + n*6);
+          lsm1VblobAppend(&val, pSpace, n);
+        }
+        break;
+      }
+      case SQLITE_FLOAT: {
+        double r = sqlite3_value_double(pArg);
+        lsm1VblobAppendVarint(&val, SQLITE_FLOAT + 8*6);
+        lsm1VblobAppend(&val, (u8*)&r, sizeof(r));
+        break;
+      }
+      case SQLITE_BLOB: {
+        int n = sqlite3_value_bytes(pArg);
+        lsm1VblobAppendVarint(&val, n*6 + SQLITE_BLOB);
+        lsm1VblobAppend(&val, sqlite3_value_blob(pArg), n);
+        break;
+      }
+      case SQLITE_TEXT: {
+        int n = sqlite3_value_bytes(pArg);
+        lsm1VblobAppendVarint(&val, n*6 + SQLITE_TEXT);
+        lsm1VblobAppend(&val, sqlite3_value_text(pArg), n);
+        break;
+      }
+    }
+  }
+  if( val.errNoMem ){
+    return SQLITE_NOMEM;
+  }
+  rc = lsm_insert(p->pDb, pKey, nKey, val.a, val.n);
+  sqlite3_free(val.a);
+  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
+}      
+
+/* Begin a transaction
+*/
+static int lsm1Begin(sqlite3_vtab *pVtab){
+  lsm1_vtab *p = (lsm1_vtab*)pVtab;
+  int rc = lsm_begin(p->pDb, 1);
+  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
+}
+
+/* Phase 1 of a transaction commit.
+*/
+static int lsm1Sync(sqlite3_vtab *pVtab){
+  return SQLITE_OK;
+}
+
+/* Commit a transaction
+*/
+static int lsm1Commit(sqlite3_vtab *pVtab){
+  lsm1_vtab *p = (lsm1_vtab*)pVtab;
+  int rc = lsm_commit(p->pDb, 0);
+  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
+}
+
+/* Rollback a transaction
+*/
+static int lsm1Rollback(sqlite3_vtab *pVtab){
+  lsm1_vtab *p = (lsm1_vtab*)pVtab;
+  int rc = lsm_rollback(p->pDb, 0);
+  return rc==LSM_OK ? SQLITE_OK : SQLITE_ERROR;
+}
+
+/*
+** This following structure defines all the methods for the 
+** generate_lsm1 virtual table.
+*/
+static sqlite3_module lsm1Module = {
+  0,                       /* iVersion */
+  lsm1Connect,             /* xCreate */
+  lsm1Connect,             /* xConnect */
+  lsm1BestIndex,           /* xBestIndex */
+  lsm1Disconnect,          /* xDisconnect */
+  lsm1Disconnect,          /* xDestroy */
+  lsm1Open,                /* xOpen - open a cursor */
+  lsm1Close,               /* xClose - close a cursor */
+  lsm1Filter,              /* xFilter - configure scan constraints */
+  lsm1Next,                /* xNext - advance a cursor */
+  lsm1Eof,                 /* xEof - check for end of scan */
+  lsm1Column,              /* xColumn - read data */
+  lsm1Rowid,               /* xRowid - read data */
+  lsm1Update,              /* xUpdate */
+  lsm1Begin,               /* xBegin */
+  lsm1Sync,                /* xSync */
+  lsm1Commit,              /* xCommit */
+  lsm1Rollback,            /* xRollback */
+  0,                       /* xFindMethod */
+  0,                       /* xRename */
+};
+
+
+#ifdef _WIN32
+__declspec(dllexport)
+#endif
+int sqlite3_lsm_init(
+  sqlite3 *db, 
+  char **pzErrMsg, 
+  const sqlite3_api_routines *pApi
+){
+  int rc = SQLITE_OK;
+  SQLITE_EXTENSION_INIT2(pApi);
+  rc = sqlite3_create_module(db, "lsm1", &lsm1Module, 0);
+  return rc;
+}
diff --git a/ext/lsm1/lsm_win32.c b/ext/lsm1/lsm_win32.c
new file mode 100644
index 0000000..6c5d06b
--- /dev/null
+++ b/ext/lsm1/lsm_win32.c
@@ -0,0 +1,1063 @@
+/*
+** 2011-12-03
+**
+** The author disclaims copyright to this source code.  In place of
+** a legal notice, here is a blessing:
+**
+**    May you do good and not evil.
+**    May you find forgiveness for yourself and forgive others.
+**    May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** Win32-specific run-time environment implementation for LSM.
+*/
+
+#ifdef _WIN32
+
+#include <assert.h>
+#include <string.h>
+
+#include <stdlib.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <ctype.h>
+
+#include "windows.h"
+
+#include "lsmInt.h"
+
+/*
+** An open file is an instance of the following object
+*/
+typedef struct Win32File Win32File;
+struct Win32File {
+  lsm_env *pEnv;                  /* The run-time environment */
+  const char *zName;              /* Full path to file */
+
+  HANDLE hFile;                   /* Open file handle */
+  HANDLE hShmFile;                /* File handle for *-shm file */
+
+  SYSTEM_INFO sysInfo;            /* Operating system information */
+  HANDLE hMap;                    /* File handle for mapping */
+  LPVOID pMap;                    /* Pointer to mapping of file fd */
+  size_t nMap;                    /* Size of mapping at pMap in bytes */
+  int nShm;                       /* Number of entries in ahShm[]/apShm[] */
+  LPHANDLE ahShm;                 /* Array of handles for shared mappings */
+  LPVOID *apShm;                  /* Array of 32K shared memory segments */
+};
+
+static char *win32ShmFile(Win32File *pWin32File){
+  char *zShm;
+  int nName = strlen(pWin32File->zName);
+  zShm = (char *)lsmMallocZero(pWin32File->pEnv, nName+4+1);
+  if( zShm ){
+    memcpy(zShm, pWin32File->zName, nName);
+    memcpy(&zShm[nName], "-shm", 5);
+  }
+  return zShm;
+}
+
+static int win32Sleep(int us){
+  Sleep((us + 999) / 1000);
+  return LSM_OK;
+}
+
+/*
+** The number of times that an I/O operation will be retried following a
+** locking error - probably caused by antivirus software.  Also the initial
+** delay before the first retry.  The delay increases linearly with each
+** retry.
+*/
+#ifndef LSM_WIN32_IOERR_RETRY
+# define LSM_WIN32_IOERR_RETRY 10
+#endif
+#ifndef LSM_WIN32_IOERR_RETRY_DELAY
+# define LSM_WIN32_IOERR_RETRY_DELAY 25000
+#endif
+static int win32IoerrRetry = LSM_WIN32_IOERR_RETRY;
+static int win32IoerrRetryDelay = LSM_WIN32_IOERR_RETRY_DELAY;
+
+/*
+** The "win32IoerrCanRetry1" macro is used to determine if a particular
+** I/O error code obtained via GetLastError() is eligible to be retried.
+** It must accept the error code DWORD as its only argument and should
+** return non-zero if the error code is transient in nature and the
+** operation responsible for generating the original error might succeed
+** upon being retried.  The argument to this macro should be a variable.
+**
+** Additionally, a macro named "win32IoerrCanRetry2" may be defined.  If
+** it is defined, it will be consulted only when the macro
+** "win32IoerrCanRetry1" returns zero.  The "win32IoerrCanRetry2" macro
+** is completely optional and may be used to include additional error
+** codes in the set that should result in the failing I/O operation being
+** retried by the caller.  If defined, the "win32IoerrCanRetry2" macro
+** must exhibit external semantics identical to those of the
+** "win32IoerrCanRetry1" macro.
+*/
+#if !defined(win32IoerrCanRetry1)
+#define win32IoerrCanRetry1(a) (((a)==ERROR_ACCESS_DENIED)        || \
+                                ((a)==ERROR_SHARING_VIOLATION)    || \
+                                ((a)==ERROR_LOCK_VIOLATION)       || \
+                                ((a)==ERROR_DEV_NOT_EXIST)        || \
+                                ((a)==ERROR_NETNAME_DELETED)      || \
+                                ((a)==ERROR_SEM_TIMEOUT)          || \
+                                ((a)==ERROR_NETWORK_UNREACHABLE))
+#endif
+
+/*
+** If an I/O error occurs, invoke this routine to see if it should be
+** retried.  Return TRUE to retry.  Return FALSE to give up with an
+** error.
+*/
+static int win32RetryIoerr(
+  lsm_env *pEnv,
+  int *pnRetry
+){
+  DWORD lastErrno;
+  if( *pnRetry>=win32IoerrRetry ){
+    return 0;
+  }
+  lastErrno = GetLastError();
+  if( win32IoerrCanRetry1(lastErrno) ){
+    win32Sleep(win32IoerrRetryDelay*(1+*pnRetry));
+    ++*pnRetry;
+    return 1;
+  }
+#if defined(win32IoerrCanRetry2)
+  else if( win32IoerrCanRetry2(lastErrno) ){
+    win32Sleep(win32IoerrRetryDelay*(1+*pnRetry));
+    ++*pnRetry;
+    return 1;
+  }
+#endif
+  return 0;
+}
+
+/*
+** Convert a UTF-8 string to Microsoft Unicode.
+**
+** Space to hold the returned string is obtained from lsmMalloc().
+*/
+static LPWSTR win32Utf8ToUnicode(lsm_env *pEnv, const char *zText){
+  int nChar;
+  LPWSTR zWideText;
+
+  nChar = MultiByteToWideChar(CP_UTF8, 0, zText, -1, NULL, 0);
+  if( nChar==0 ){
+    return 0;
+  }
+  zWideText = lsmMallocZero(pEnv, nChar * sizeof(WCHAR));
+  if( zWideText==0 ){
+    return 0;
+  }
+  nChar = MultiByteToWideChar(CP_UTF8, 0, zText, -1, zWideText, nChar);
+  if( nChar==0 ){
+    lsmFree(pEnv, zWideText);
+    zWideText = 0;
+  }
+  return zWideText;
+}
+
+/*
+** Convert a Microsoft Unicode string to UTF-8.
+**
+** Space to hold the returned string is obtained from lsmMalloc().
+*/
+static char *win32UnicodeToUtf8(lsm_env *pEnv, LPCWSTR zWideText){
+  int nByte;
+  char *zText;
+
+  nByte = WideCharToMultiByte(CP_UTF8, 0, zWideText, -1, 0, 0, 0, 0);
+  if( nByte == 0 ){
+    return 0;
+  }
+  zText = lsmMallocZero(pEnv, nByte);
+  if( zText==0 ){
+    return 0;
+  }
+  nByte = WideCharToMultiByte(CP_UTF8, 0, zWideText, -1, zText, nByte, 0, 0);
+  if( nByte == 0 ){
+    lsmFree(pEnv, zText);
+    zText = 0;
+  }
+  return zText;
+}
+
+#if !defined(win32IsNotFound)
+#define win32IsNotFound(a) (((a)==ERROR_FILE_NOT_FOUND)  || \
+                            ((a)==ERROR_PATH_NOT_FOUND))
+#endif
+
+static int win32Open(
+  lsm_env *pEnv,
+  const char *zFile,
+  int flags,
+  LPHANDLE phFile
+){
+  int rc;
+  LPWSTR zConverted;
+
+  zConverted = win32Utf8ToUnicode(pEnv, zFile);
+  if( zConverted==0 ){
+    rc = LSM_NOMEM_BKPT;
+  }else{
+    int bReadonly = (flags & LSM_OPEN_READONLY);
+    DWORD dwDesiredAccess;
+    DWORD dwShareMode = FILE_SHARE_READ | FILE_SHARE_WRITE;
+    DWORD dwCreationDisposition;
+    DWORD dwFlagsAndAttributes = FILE_ATTRIBUTE_NORMAL;
+    HANDLE hFile;
+    int nRetry = 0;
+    if( bReadonly ){
+      dwDesiredAccess = GENERIC_READ;
+      dwCreationDisposition = OPEN_EXISTING;
+    }else{
+      dwDesiredAccess = GENERIC_READ | GENERIC_WRITE;
+      dwCreationDisposition = OPEN_ALWAYS;
+    }
+    while( (hFile = CreateFileW((LPCWSTR)zConverted,
+                                dwDesiredAccess,
+                                dwShareMode, NULL,
+                                dwCreationDisposition,
+                                dwFlagsAndAttributes,
+                                NULL))==INVALID_HANDLE_VALUE &&
+                                win32RetryIoerr(pEnv, &nRetry) ){
+      /* Noop */
+    }
+    lsmFree(pEnv, zConverted);
+    if( hFile!=INVALID_HANDLE_VALUE ){
+      *phFile = hFile;
+      rc = LSM_OK;
+    }else{
+      if( win32IsNotFound(GetLastError()) ){
+        rc = lsmErrorBkpt(LSM_IOERR_NOENT);
+      }else{
+        rc = LSM_IOERR_BKPT;
+      }
+    }
+  }
+  return rc;
+}
+
+static int lsmWin32OsOpen(
+  lsm_env *pEnv,
+  const char *zFile,
+  int flags,
+  lsm_file **ppFile
+){
+  int rc = LSM_OK;
+  Win32File *pWin32File;
+
+  pWin32File = lsmMallocZero(pEnv, sizeof(Win32File));
+  if( pWin32File==0 ){
+    rc = LSM_NOMEM_BKPT;
+  }else{
+    HANDLE hFile = NULL;
+
+    rc = win32Open(pEnv, zFile, flags, &hFile);
+    if( rc==LSM_OK ){
+      memset(&pWin32File->sysInfo, 0, sizeof(SYSTEM_INFO));
+      GetSystemInfo(&pWin32File->sysInfo);
+      pWin32File->pEnv = pEnv;
+      pWin32File->zName = zFile;
+      pWin32File->hFile = hFile;
+    }else{
+      lsmFree(pEnv, pWin32File);
+      pWin32File = 0;
+    }
+  }
+  *ppFile = (lsm_file *)pWin32File;
+  return rc;
+}
+
+static int lsmWin32OsWrite(
+  lsm_file *pFile, /* File to write to */
+  lsm_i64 iOff,    /* Offset to write to */
+  void *pData,     /* Write data from this buffer */
+  int nData        /* Bytes of data to write */
+){
+  Win32File *pWin32File = (Win32File *)pFile;
+  OVERLAPPED overlapped;  /* The offset for WriteFile. */
+  u8 *aRem = (u8 *)pData; /* Data yet to be written */
+  int nRem = nData;       /* Number of bytes yet to be written */
+  int nRetry = 0;         /* Number of retrys */
+
+  memset(&overlapped, 0, sizeof(OVERLAPPED));
+  overlapped.Offset = (LONG)(iOff & 0XFFFFFFFF);
+  overlapped.OffsetHigh = (LONG)((iOff>>32) & 0x7FFFFFFF);
+  while( nRem>0 ){
+    DWORD nWrite = 0; /* Bytes written using WriteFile */
+    if( !WriteFile(pWin32File->hFile, aRem, nRem, &nWrite, &overlapped) ){
+      if( win32RetryIoerr(pWin32File->pEnv, &nRetry) ) continue;
+      break;
+    }
+    assert( nWrite==0 || nWrite<=(DWORD)nRem );
+    if( nWrite==0 || nWrite>(DWORD)nRem ){
+      break;
+    }
+    iOff += nWrite;
+    overlapped.Offset = (LONG)(iOff & 0xFFFFFFFF);
+    overlapped.OffsetHigh = (LONG)((iOff>>32) & 0x7FFFFFFF);
+    aRem += nWrite;
+    nRem -= nWrite;
+  }
+  if( nRem!=0 ) return LSM_IOERR_BKPT;
+  return LSM_OK;
+}
+
+static int win32Truncate(
+  HANDLE hFile,
+  lsm_i64 nSize
+){
+  LARGE_INTEGER offset;
+  offset.QuadPart = nSize;
+  if( !SetFilePointerEx(hFile, offset, 0, FILE_BEGIN) ){
+    return LSM_IOERR_BKPT;
+  }
+  if (!SetEndOfFile(hFile) ){
+    return LSM_IOERR_BKPT;
+  }
+  return LSM_OK;
+}
+
+static int lsmWin32OsTruncate(
+  lsm_file *pFile, /* File to write to */
+  lsm_i64 nSize    /* Size to truncate file to */
+){
+  Win32File *pWin32File = (Win32File *)pFile;
+  return win32Truncate(pWin32File->hFile, nSize);
+}
+
+static int lsmWin32OsRead(
+  lsm_file *pFile, /* File to read from */
+  lsm_i64 iOff,    /* Offset to read from */
+  void *pData,     /* Read data into this buffer */
+  int nData        /* Bytes of data to read */
+){
+  Win32File *pWin32File = (Win32File *)pFile;
+  OVERLAPPED overlapped; /* The offset for ReadFile */
+  DWORD nRead = 0;       /* Bytes read using ReadFile */
+  int nRetry = 0;        /* Number of retrys */
+
+  memset(&overlapped, 0, sizeof(OVERLAPPED));
+  overlapped.Offset = (LONG)(iOff & 0XFFFFFFFF);
+  overlapped.OffsetHigh = (LONG)((iOff>>32) & 0X7FFFFFFF);
+  while( !ReadFile(pWin32File->hFile, pData, nData, &nRead, &overlapped) &&
+         GetLastError()!=ERROR_HANDLE_EOF ){
+    if( win32RetryIoerr(pWin32File->pEnv, &nRetry) ) continue;
+    return LSM_IOERR_BKPT;
+  }
+  if( nRead<(DWORD)nData ){
+    /* Unread parts of the buffer must be zero-filled */
+    memset(&((char*)pData)[nRead], 0, nData - nRead);
+  }
+  return LSM_OK;
+}
+
+static int lsmWin32OsSync(lsm_file *pFile){
+  int rc = LSM_OK;
+
+#ifndef LSM_NO_SYNC
+  Win32File *pWin32File = (Win32File *)pFile;
+
+  if( pWin32File->pMap!=NULL ){
+    if( !FlushViewOfFile(pWin32File->pMap, 0) ){
+      rc = LSM_IOERR_BKPT;
+    }
+  }
+  if( rc==LSM_OK && !FlushFileBuffers(pWin32File->hFile) ){
+    rc = LSM_IOERR_BKPT;
+  }
+#else
+  unused_parameter(pFile);
+#endif
+
+  return rc;
+}
+
+static int lsmWin32OsSectorSize(lsm_file *pFile){
+  return 512;
+}
+
+static void win32Unmap(Win32File *pWin32File){
+  if( pWin32File->pMap!=NULL ){
+    UnmapViewOfFile(pWin32File->pMap);
+    pWin32File->pMap = NULL;
+    pWin32File->nMap = 0;
+  }
+  if( pWin32File->hMap!=NULL ){
+    CloseHandle(pWin32File->hMap);
+    pWin32File->hMap = NULL;
+  }
+}
+
+static int lsmWin32OsRemap(
+  lsm_file *pFile,
+  lsm_i64 iMin,
+  void **ppOut,
+  lsm_i64 *pnOut
+){
+  Win32File *pWin32File = (Win32File *)pFile;
+
+  /* If the file is between 0 and 2MB in size, extend it in chunks of 256K.
+  ** Thereafter, in chunks of 1MB at a time.  */
+  const int aIncrSz[] = {256*1024, 1024*1024};
+  int nIncrSz = aIncrSz[iMin>(2*1024*1024)];
+
+  *ppOut = NULL;
+  *pnOut = 0;
+
+  win32Unmap(pWin32File);
+  if( iMin>=0 ){
+    LARGE_INTEGER fileSize;
+    DWORD dwSizeHigh;
+    DWORD dwSizeLow;
+    HANDLE hMap;
+    LPVOID pMap;
+    memset(&fileSize, 0, sizeof(LARGE_INTEGER));
+    if( !GetFileSizeEx(pWin32File->hFile, &fileSize) ){
+      return LSM_IOERR_BKPT;
+    }
+    assert( fileSize.QuadPart>=0 );
+    if( fileSize.QuadPart<iMin ){
+      int rc;
+      fileSize.QuadPart = ((iMin + nIncrSz-1) / nIncrSz) * nIncrSz;
+      rc = lsmWin32OsTruncate(pFile, fileSize.QuadPart);
+      if( rc!=LSM_OK ){
+        return rc;
+      }
+    }
+    dwSizeLow = (DWORD)(fileSize.QuadPart & 0xFFFFFFFF);
+    dwSizeHigh = (DWORD)((fileSize.QuadPart & 0x7FFFFFFFFFFFFFFF) >> 32);
+    hMap = CreateFileMappingW(pWin32File->hFile, NULL, PAGE_READWRITE,
+                              dwSizeHigh, dwSizeLow, NULL);
+    if( hMap==NULL ){
+      return LSM_IOERR_BKPT;
+    }
+    pWin32File->hMap = hMap;
+    assert( fileSize.QuadPart<=0xFFFFFFFF );
+    pMap = MapViewOfFile(hMap, FILE_MAP_WRITE | FILE_MAP_READ, 0, 0,
+                         (SIZE_T)fileSize.QuadPart);
+    if( pMap==NULL ){
+      return LSM_IOERR_BKPT;
+    }
+    pWin32File->pMap = pMap;
+    pWin32File->nMap = (SIZE_T)fileSize.QuadPart;
+  }
+  *ppOut = pWin32File->pMap;
+  *pnOut = pWin32File->nMap;
+  return LSM_OK;
+}
+
+static BOOL win32IsDriveLetterAndColon(
+  const char *zPathname
+){
+  return ( isalpha(zPathname[0]) && zPathname[1]==':' );
+}
+
+static int lsmWin32OsFullpath(
+  lsm_env *pEnv,
+  const char *zName,
+  char *zOut,
+  int *pnOut
+){
+  DWORD nByte;
+  void *zConverted;
+  LPWSTR zTempWide;
+  char *zTempUtf8;
+
+  if( zName[0]=='/' && win32IsDriveLetterAndColon(zName+1) ){
+    zName++;
+  }
+  zConverted = win32Utf8ToUnicode(pEnv, zName);
+  if( zConverted==0 ){
+    return LSM_NOMEM_BKPT;
+  }
+  nByte = GetFullPathNameW((LPCWSTR)zConverted, 0, 0, 0);
+  if( nByte==0 ){
+    lsmFree(pEnv, zConverted);
+    return LSM_IOERR_BKPT;
+  }
+  nByte += 3;
+  zTempWide = lsmMallocZero(pEnv, nByte * sizeof(zTempWide[0]));
+  if( zTempWide==0 ){
+    lsmFree(pEnv, zConverted);
+    return LSM_NOMEM_BKPT;
+  }
+  nByte = GetFullPathNameW((LPCWSTR)zConverted, nByte, zTempWide, 0);
+  if( nByte==0 ){
+    lsmFree(pEnv, zConverted);
+    lsmFree(pEnv, zTempWide);
+    return LSM_IOERR_BKPT;
+  }
+  lsmFree(pEnv, zConverted);
+  zTempUtf8 = win32UnicodeToUtf8(pEnv, zTempWide);
+  lsmFree(pEnv, zTempWide);
+  if( zTempUtf8 ){
+    int nOut = *pnOut;
+    int nLen = strlen(zTempUtf8) + 1;
+    if( nLen<=nOut ){
+      snprintf(zOut, nOut, "%s", zTempUtf8);
+    }
+    lsmFree(pEnv, zTempUtf8);
+    *pnOut = nLen;
+    return LSM_OK;
+  }else{
+    return LSM_NOMEM_BKPT;
+  }
+}
+
+static int lsmWin32OsFileid(
+  lsm_file *pFile,
+  void *pBuf,
+  int *pnBuf
+){
+  int nBuf;
+  int nReq;
+  u8 *pBuf2 = (u8 *)pBuf;
+  Win32File *pWin32File = (Win32File *)pFile;
+  BY_HANDLE_FILE_INFORMATION fileInfo;
+
+  nBuf = *pnBuf;
+  nReq = (sizeof(fileInfo.dwVolumeSerialNumber) +
+          sizeof(fileInfo.nFileIndexHigh) +
+          sizeof(fileInfo.nFileIndexLow));
+  *pnBuf = nReq;
+  if( nReq>nBuf ) return LSM_OK;
+  memset(&fileInfo, 0, sizeof(BY_HANDLE_FILE_INFORMATION));
+  if( !GetFileInformationByHandle(pWin32File->hFile, &fileInfo) ){
+    return LSM_IOERR_BKPT;
+  }
+  nReq = sizeof(fileInfo.dwVolumeSerialNumber);
+  memcpy(pBuf2, &fileInfo.dwVolumeSerialNumber, nReq);
+  pBuf2 += nReq;
+  nReq = sizeof(fileInfo.nFileIndexHigh);
+  memcpy(pBuf, &fileInfo.nFileIndexHigh, nReq);
+  pBuf2 += nReq;
+  nReq = sizeof(fileInfo.nFileIndexLow);
+  memcpy(pBuf2, &fileInfo.nFileIndexLow, nReq);
+  return LSM_OK;
+}
+
+static int win32Delete(
+  lsm_env *pEnv,
+  const char *zFile
+){
+  int rc;
+  LPWSTR zConverted;
+
+  zConverted = win32Utf8ToUnicode(pEnv, zFile);
+  if( zConverted==0 ){
+    rc = LSM_NOMEM_BKPT;
+  }else{
+    int nRetry = 0;
+    DWORD attr;
+
+    do {
+      attr = GetFileAttributesW(zConverted);
+      if ( attr==INVALID_FILE_ATTRIBUTES ){
+        rc = LSM_IOERR_BKPT;
+        break;
+      }
+      if ( attr&FILE_ATTRIBUTE_DIRECTORY ){
+        rc = LSM_IOERR_BKPT; /* Files only. */
+        break;
+      }
+      if ( DeleteFileW(zConverted) ){
+        rc = LSM_OK; /* Deleted OK. */
+        break;
+      }
+      if ( !win32RetryIoerr(pEnv, &nRetry) ){
+        rc = LSM_IOERR_BKPT; /* No more retries. */
+        break;
+      }
+    }while( 1 );
+  }
+  lsmFree(pEnv, zConverted);
+  return rc;
+}
+
+static int lsmWin32OsUnlink(lsm_env *pEnv, const char *zFile){
+  return win32Delete(pEnv, zFile);
+}
+
+#if !defined(win32IsLockBusy)
+#define win32IsLockBusy(a) (((a)==ERROR_LOCK_VIOLATION) || \
+                            ((a)==ERROR_IO_PENDING))
+#endif
+
+static int win32LockFile(
+  Win32File *pWin32File,
+  int iLock,
+  int nLock,
+  int eType
+){
+  OVERLAPPED ovlp;
+
+  assert( LSM_LOCK_UNLOCK==0 );
+  assert( LSM_LOCK_SHARED==1 );
+  assert( LSM_LOCK_EXCL==2 );
+  assert( eType>=LSM_LOCK_UNLOCK && eType<=LSM_LOCK_EXCL );
+  assert( nLock>=0 );
+  assert( iLock>0 && iLock<=32 );
+
+  memset(&ovlp, 0, sizeof(OVERLAPPED));
+  ovlp.Offset = (4096-iLock-nLock+1);
+  if( eType>LSM_LOCK_UNLOCK ){
+    DWORD flags = LOCKFILE_FAIL_IMMEDIATELY;
+    if( eType>=LSM_LOCK_EXCL ) flags |= LOCKFILE_EXCLUSIVE_LOCK;
+    if( !LockFileEx(pWin32File->hFile, flags, 0, (DWORD)nLock, 0, &ovlp) ){
+      if( win32IsLockBusy(GetLastError()) ){
+        return LSM_BUSY;
+      }else{
+        return LSM_IOERR_BKPT;
+      }
+    }
+  }else{
+    if( !UnlockFileEx(pWin32File->hFile, 0, (DWORD)nLock, 0, &ovlp) ){
+      return LSM_IOERR_BKPT;
+    }
+  }
+  return LSM_OK;
+}
+
+static int lsmWin32OsLock(lsm_file *pFile, int iLock, int eType){
+  Win32File *pWin32File = (Win32File *)pFile;
+  return win32LockFile(pWin32File, iLock, 1, eType);
+}
+
+static int lsmWin32OsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
+  int rc;
+  Win32File *pWin32File = (Win32File *)pFile;
+  rc = win32LockFile(pWin32File, iLock, nLock, eType);
+  if( rc!=LSM_OK ) return rc;
+  win32LockFile(pWin32File, iLock, nLock, LSM_LOCK_UNLOCK);
+  return LSM_OK;
+}
+
+static int lsmWin32OsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
+  int rc;
+  Win32File *pWin32File = (Win32File *)pFile;
+  int iOffset = iChunk * sz;
+  int iOffsetShift = iOffset % pWin32File->sysInfo.dwAllocationGranularity;
+  int nNew = iChunk + 1;
+  lsm_i64 nReq = nNew * sz;
+
+  *ppShm = NULL;
+  assert( sz>=0 );
+  assert( sz==LSM_SHM_CHUNK_SIZE );
+  if( iChunk>=pWin32File->nShm ){
+    LPHANDLE ahNew;
+    LPVOID *apNew;
+    LARGE_INTEGER fileSize;
+
+    /* If the shared-memory file has not been opened, open it now. */
+    if( pWin32File->hShmFile==NULL ){
+      char *zShm = win32ShmFile(pWin32File);
+      if( !zShm ) return LSM_NOMEM_BKPT;
+      rc = win32Open(pWin32File->pEnv, zShm, 0, &pWin32File->hShmFile);
+      lsmFree(pWin32File->pEnv, zShm);
+      if( rc!=LSM_OK ){
+        return rc;
+      }
+    }
+
+    /* If the shared-memory file is not large enough to contain the
+    ** requested chunk, cause it to grow.  */
+    memset(&fileSize, 0, sizeof(LARGE_INTEGER));
+    if( !GetFileSizeEx(pWin32File->hShmFile, &fileSize) ){
+      return LSM_IOERR_BKPT;
+    }
+    assert( fileSize.QuadPart>=0 );
+    if( fileSize.QuadPart<nReq ){
+      rc = win32Truncate(pWin32File->hShmFile, nReq);
+      if( rc!=LSM_OK ){
+        return rc;
+      }
+    }
+
+    ahNew = (LPHANDLE)lsmMallocZero(pWin32File->pEnv, sizeof(HANDLE) * nNew);
+    if( !ahNew ) return LSM_NOMEM_BKPT;
+    apNew = (LPVOID *)lsmMallocZero(pWin32File->pEnv, sizeof(LPVOID) * nNew);
+    if( !apNew ){
+      lsmFree(pWin32File->pEnv, ahNew);
+      return LSM_NOMEM_BKPT;
+    }
+    memcpy(ahNew, pWin32File->ahShm, sizeof(HANDLE) * pWin32File->nShm);
+    memcpy(apNew, pWin32File->apShm, sizeof(LPVOID) * pWin32File->nShm);
+    lsmFree(pWin32File->pEnv, pWin32File->ahShm);
+    pWin32File->ahShm = ahNew;
+    lsmFree(pWin32File->pEnv, pWin32File->apShm);
+    pWin32File->apShm = apNew;
+    pWin32File->nShm = nNew;
+  }
+
+  if( pWin32File->ahShm[iChunk]==NULL ){
+    HANDLE hMap;
+    assert( nReq<=0xFFFFFFFF );
+    hMap = CreateFileMappingW(pWin32File->hShmFile, NULL, PAGE_READWRITE, 0,
+                              (DWORD)nReq, NULL);
+    if( hMap==NULL ){
+      return LSM_IOERR_BKPT;
+    }
+    pWin32File->ahShm[iChunk] = hMap;
+  }
+  if( pWin32File->apShm[iChunk]==NULL ){
+    LPVOID pMap;
+    pMap = MapViewOfFile(pWin32File->ahShm[iChunk],
+                         FILE_MAP_WRITE | FILE_MAP_READ, 0,
+                         iOffset - iOffsetShift, sz + iOffsetShift);
+    if( pMap==NULL ){
+      return LSM_IOERR_BKPT;
+    }
+    pWin32File->apShm[iChunk] = pMap;
+  }
+  if( iOffsetShift!=0 ){
+    char *p = (char *)pWin32File->apShm[iChunk];
+    *ppShm = (void *)&p[iOffsetShift];
+  }else{
+    *ppShm = pWin32File->apShm[iChunk];
+  }
+  return LSM_OK;
+}
+
+static void lsmWin32OsShmBarrier(void){
+  MemoryBarrier();
+}
+
+static int lsmWin32OsShmUnmap(lsm_file *pFile, int bDelete){
+  Win32File *pWin32File = (Win32File *)pFile;
+
+  if( pWin32File->hShmFile!=NULL ){
+    int i;
+    for(i=0; i<pWin32File->nShm; i++){
+      if( pWin32File->apShm[i]!=NULL ){
+        UnmapViewOfFile(pWin32File->apShm[i]);
+        pWin32File->apShm[i] = NULL;
+      }
+      if( pWin32File->ahShm[i]!=NULL ){
+        CloseHandle(pWin32File->ahShm[i]);
+        pWin32File->ahShm[i] = NULL;
+      }
+    }
+    CloseHandle(pWin32File->hShmFile);
+    pWin32File->hShmFile = NULL;
+    if( bDelete ){
+      char *zShm = win32ShmFile(pWin32File);
+      if( zShm ){ win32Delete(pWin32File->pEnv, zShm); }
+      lsmFree(pWin32File->pEnv, zShm);
+    }
+  }
+  return LSM_OK;
+}
+
+#define MX_CLOSE_ATTEMPT 3
+static int lsmWin32OsClose(lsm_file *pFile){
+  int rc;
+  int nRetry = 0;
+  Win32File *pWin32File = (Win32File *)pFile;
+  lsmWin32OsShmUnmap(pFile, 0);
+  win32Unmap(pWin32File);
+  do{
+    if( pWin32File->hFile==NULL ){
+      rc = LSM_IOERR_BKPT;
+      break;
+    }
+    rc = CloseHandle(pWin32File->hFile);
+    if( rc ){
+      pWin32File->hFile = NULL;
+      rc = LSM_OK;
+      break;
+    }
+    if( ++nRetry>=MX_CLOSE_ATTEMPT ){
+      rc = LSM_IOERR_BKPT;
+      break;
+    }
+  }while( 1 );
+  lsmFree(pWin32File->pEnv, pWin32File->ahShm);
+  lsmFree(pWin32File->pEnv, pWin32File->apShm);
+  lsmFree(pWin32File->pEnv, pWin32File);
+  return rc;
+}
+
+static int lsmWin32OsSleep(lsm_env *pEnv, int us){
+  unused_parameter(pEnv);
+  return win32Sleep(us);
+}
+
+/****************************************************************************
+** Memory allocation routines.
+*/
+
+static void *lsmWin32OsMalloc(lsm_env *pEnv, size_t N){
+  assert( HeapValidate(GetProcessHeap(), 0, NULL) );
+  return HeapAlloc(GetProcessHeap(), 0, (SIZE_T)N);
+}
+
+static void lsmWin32OsFree(lsm_env *pEnv, void *p){
+  assert( HeapValidate(GetProcessHeap(), 0, NULL) );
+  if( p ){
+    HeapFree(GetProcessHeap(), 0, p);
+  }
+}
+
+static void *lsmWin32OsRealloc(lsm_env *pEnv, void *p, size_t N){
+  unsigned char *m = (unsigned char *)p;
+  assert( HeapValidate(GetProcessHeap(), 0, NULL) );
+  if( 1>N ){
+    lsmWin32OsFree(pEnv, p);
+    return NULL;
+  }else if( NULL==p ){
+    return lsmWin32OsMalloc(pEnv, N);
+  }else{
+#if 0 /* arguable: don't shrink */
+    SIZE_T sz = HeapSize(GetProcessHeap(), 0, m);
+    if( sz>=(SIZE_T)N ){
+      return p;
+    }
+#endif
+    return HeapReAlloc(GetProcessHeap(), 0, m, N);
+  }
+}
+
+static size_t lsmWin32OsMSize(lsm_env *pEnv, void *p){
+  assert( HeapValidate(GetProcessHeap(), 0, NULL) );
+  return (size_t)HeapSize(GetProcessHeap(), 0, p);
+}
+
+
+#ifdef LSM_MUTEX_WIN32
+/*************************************************************************
+** Mutex methods for Win32 based systems.  If LSM_MUTEX_WIN32 is
+** missing then a no-op implementation of mutexes found below will be
+** used instead.
+*/
+#include "windows.h"
+
+typedef struct Win32Mutex Win32Mutex;
+struct Win32Mutex {
+  lsm_env *pEnv;
+  CRITICAL_SECTION mutex;
+#ifdef LSM_DEBUG
+  DWORD owner;
+#endif
+};
+
+#ifndef WIN32_MUTEX_INITIALIZER
+# define WIN32_MUTEX_INITIALIZER { 0 }
+#endif
+
+#ifdef LSM_DEBUG
+# define LSM_WIN32_STATIC_MUTEX { 0, WIN32_MUTEX_INITIALIZER, 0 }
+#else
+# define LSM_WIN32_STATIC_MUTEX { 0, WIN32_MUTEX_INITIALIZER }
+#endif
+
+static int lsmWin32OsMutexStatic(
+  lsm_env *pEnv,
+  int iMutex,
+  lsm_mutex **ppStatic
+){
+  static volatile LONG initialized = 0;
+  static Win32Mutex sMutex[2] = {
+    LSM_WIN32_STATIC_MUTEX,
+    LSM_WIN32_STATIC_MUTEX
+  };
+
+  assert( iMutex==LSM_MUTEX_GLOBAL || iMutex==LSM_MUTEX_HEAP );
+  assert( LSM_MUTEX_GLOBAL==1 && LSM_MUTEX_HEAP==2 );
+
+  if( InterlockedCompareExchange(&initialized, 1, 0)==0 ){
+    int i;
+    for(i=0; i<array_size(sMutex); i++){
+      InitializeCriticalSection(&sMutex[i].mutex);
+    }
+  }
+  *ppStatic = (lsm_mutex *)&sMutex[iMutex-1];
+  return LSM_OK;
+}
+
+static int lsmWin32OsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
+  Win32Mutex *pMutex;           /* Pointer to new mutex */
+
+  pMutex = (Win32Mutex *)lsmMallocZero(pEnv, sizeof(Win32Mutex));
+  if( !pMutex ) return LSM_NOMEM_BKPT;
+
+  pMutex->pEnv = pEnv;
+  InitializeCriticalSection(&pMutex->mutex);
+
+  *ppNew = (lsm_mutex *)pMutex;
+  return LSM_OK;
+}
+
+static void lsmWin32OsMutexDel(lsm_mutex *p){
+  Win32Mutex *pMutex = (Win32Mutex *)p;
+  DeleteCriticalSection(&pMutex->mutex);
+  lsmFree(pMutex->pEnv, pMutex);
+}
+
+static void lsmWin32OsMutexEnter(lsm_mutex *p){
+  Win32Mutex *pMutex = (Win32Mutex *)p;
+  EnterCriticalSection(&pMutex->mutex);
+
+#ifdef LSM_DEBUG
+  assert( pMutex->owner!=GetCurrentThreadId() );
+  pMutex->owner = GetCurrentThreadId();
+  assert( pMutex->owner==GetCurrentThreadId() );
+#endif
+}
+
+static int lsmWin32OsMutexTry(lsm_mutex *p){
+  BOOL bRet;
+  Win32Mutex *pMutex = (Win32Mutex *)p;
+  bRet = TryEnterCriticalSection(&pMutex->mutex);
+#ifdef LSM_DEBUG
+  if( bRet ){
+    assert( pMutex->owner!=GetCurrentThreadId() );
+    pMutex->owner = GetCurrentThreadId();
+    assert( pMutex->owner==GetCurrentThreadId() );
+  }
+#endif
+  return !bRet;
+}
+
+static void lsmWin32OsMutexLeave(lsm_mutex *p){
+  Win32Mutex *pMutex = (Win32Mutex *)p;
+#ifdef LSM_DEBUG
+  assert( pMutex->owner==GetCurrentThreadId() );
+  pMutex->owner = 0;
+  assert( pMutex->owner!=GetCurrentThreadId() );
+#endif
+  LeaveCriticalSection(&pMutex->mutex);
+}
+
+#ifdef LSM_DEBUG
+static int lsmWin32OsMutexHeld(lsm_mutex *p){
+  Win32Mutex *pMutex = (Win32Mutex *)p;
+  return pMutex ? pMutex->owner==GetCurrentThreadId() : 1;
+}
+static int lsmWin32OsMutexNotHeld(lsm_mutex *p){
+  Win32Mutex *pMutex = (Win32Mutex *)p;
+  return pMutex ? pMutex->owner!=GetCurrentThreadId() : 1;
+}
+#endif
+/*
+** End of Win32 mutex implementation.
+*************************************************************************/
+#else
+/*************************************************************************
+** Noop mutex implementation
+*/
+typedef struct NoopMutex NoopMutex;
+struct NoopMutex {
+  lsm_env *pEnv;                  /* Environment handle (for xFree()) */
+  int bHeld;                      /* True if mutex is held */
+  int bStatic;                    /* True for a static mutex */
+};
+static NoopMutex aStaticNoopMutex[2] = {
+  {0, 0, 1},
+  {0, 0, 1},
+};
+
+static int lsmWin32OsMutexStatic(
+  lsm_env *pEnv,
+  int iMutex,
+  lsm_mutex **ppStatic
+){
+  assert( iMutex>=1 && iMutex<=(int)array_size(aStaticNoopMutex) );
+  *ppStatic = (lsm_mutex *)&aStaticNoopMutex[iMutex-1];
+  return LSM_OK;
+}
+static int lsmWin32OsMutexNew(lsm_env *pEnv, lsm_mutex **ppNew){
+  NoopMutex *p;
+  p = (NoopMutex *)lsmMallocZero(pEnv, sizeof(NoopMutex));
+  if( p ) p->pEnv = pEnv;
+  *ppNew = (lsm_mutex *)p;
+  return (p ? LSM_OK : LSM_NOMEM_BKPT);
+}
+static void lsmWin32OsMutexDel(lsm_mutex *pMutex)  {
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bStatic==0 && p->pEnv );
+  lsmFree(p->pEnv, p);
+}
+static void lsmWin32OsMutexEnter(lsm_mutex *pMutex){
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bHeld==0 );
+  p->bHeld = 1;
+}
+static int lsmWin32OsMutexTry(lsm_mutex *pMutex){
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bHeld==0 );
+  p->bHeld = 1;
+  return 0;
+}
+static void lsmWin32OsMutexLeave(lsm_mutex *pMutex){
+  NoopMutex *p = (NoopMutex *)pMutex;
+  assert( p->bHeld==1 );
+  p->bHeld = 0;
+}
+#ifdef LSM_DEBUG
+static int lsmWin32OsMutexHeld(lsm_mutex *pMutex){
+  NoopMutex *p = (NoopMutex *)pMutex;
+  return p ? p->bHeld : 1;
+}
+static int lsmWin32OsMutexNotHeld(lsm_mutex *pMutex){
+  NoopMutex *p = (NoopMutex *)pMutex;
+  return p ? !p->bHeld : 1;
+}
+#endif
+/***************************************************************************/
+#endif /* else LSM_MUTEX_NONE */
+
+/* Without LSM_DEBUG, the MutexHeld tests are never called */
+#ifndef LSM_DEBUG
+# define lsmWin32OsMutexHeld    0
+# define lsmWin32OsMutexNotHeld 0
+#endif
+
+lsm_env *lsm_default_env(void){
+  static lsm_env win32_env = {
+    sizeof(lsm_env),         /* nByte */
+    1,                       /* iVersion */
+    /***** file i/o ******************/
+    0,                       /* pVfsCtx */
+    lsmWin32OsFullpath,      /* xFullpath */
+    lsmWin32OsOpen,          /* xOpen */
+    lsmWin32OsRead,          /* xRead */
+    lsmWin32OsWrite,         /* xWrite */
+    lsmWin32OsTruncate,      /* xTruncate */
+    lsmWin32OsSync,          /* xSync */
+    lsmWin32OsSectorSize,    /* xSectorSize */
+    lsmWin32OsRemap,         /* xRemap */
+    lsmWin32OsFileid,        /* xFileid */
+    lsmWin32OsClose,         /* xClose */
+    lsmWin32OsUnlink,        /* xUnlink */
+    lsmWin32OsLock,          /* xLock */
+    lsmWin32OsTestLock,      /* xTestLock */
+    lsmWin32OsShmMap,        /* xShmMap */
+    lsmWin32OsShmBarrier,    /* xShmBarrier */
+    lsmWin32OsShmUnmap,      /* xShmUnmap */
+    /***** memory allocation *********/
+    0,                       /* pMemCtx */
+    lsmWin32OsMalloc,        /* xMalloc */
+    lsmWin32OsRealloc,       /* xRealloc */
+    lsmWin32OsFree,          /* xFree */
+    lsmWin32OsMSize,         /* xSize */
+    /***** mutexes *********************/
+    0,                       /* pMutexCtx */
+    lsmWin32OsMutexStatic,   /* xMutexStatic */
+    lsmWin32OsMutexNew,      /* xMutexNew */
+    lsmWin32OsMutexDel,      /* xMutexDel */
+    lsmWin32OsMutexEnter,    /* xMutexEnter */
+    lsmWin32OsMutexTry,      /* xMutexTry */
+    lsmWin32OsMutexLeave,    /* xMutexLeave */
+    lsmWin32OsMutexHeld,     /* xMutexHeld */
+    lsmWin32OsMutexNotHeld,  /* xMutexNotHeld */
+    /***** other *********************/
+    lsmWin32OsSleep,         /* xSleep */
+  };
+  return &win32_env;
+}
+
+#endif
diff --git a/ext/lsm1/test/lsm1_common.tcl b/ext/lsm1/test/lsm1_common.tcl
new file mode 100644
index 0000000..0e6cd84
--- /dev/null
+++ b/ext/lsm1/test/lsm1_common.tcl
@@ -0,0 +1,38 @@
+# 2014 Dec 19
+#
+# The author disclaims copyright to this source code.  In place of
+# a legal notice, here is a blessing:
+#
+#    May you do good and not evil.
+#    May you find forgiveness for yourself and forgive others.
+#    May you share freely, never taking more than you give.
+#
+#***********************************************************************
+#
+
+if {![info exists testdir]} {
+  set testdir [file join [file dirname [info script]] .. .. .. test]
+}
+source $testdir/tester.tcl
+
+# Check if the lsm1 extension has been compiled.
+if {$::tcl_platform(platform) == "windows"} {
+  set lsm1 lsm.dll
+} else {
+  set lsm1 lsm.so
+}
+
+if {[file exists [file join .. $lsm1]]} {
+  proc return_if_no_lsm1 {} {}
+} else {
+  proc return_if_no_lsm1 {} {
+    finish_test
+    return -code return
+  }
+  return
+}
+
+proc load_lsm1_vtab {db} {
+  db enable_load_extension 1
+  db eval {SELECT load_extension('../lsm')}
+}
diff --git a/ext/lsm1/test/lsm1_simple.test b/ext/lsm1/test/lsm1_simple.test
new file mode 100644
index 0000000..2eab50a
--- /dev/null
+++ b/ext/lsm1/test/lsm1_simple.test
@@ -0,0 +1,152 @@
+# 2017 July 14
+#
+# The author disclaims copyright to this source code.  In place of
+# a legal notice, here is a blessing:
+#
+#    May you do good and not evil.
+#    May you find forgiveness for yourself and forgive others.
+#    May you share freely, never taking more than you give.
+#
+#*************************************************************************
+# This file implements regression tests for SQLite library.  The
+# focus of this script is testing the lsm1 virtual table module.
+#
+
+source [file join [file dirname [info script]] lsm1_common.tcl]
+set testprefix lsm1_simple
+return_if_no_lsm1
+load_lsm1_vtab db
+
+forcedelete testlsm.db
+
+do_execsql_test 100 {
+  CREATE VIRTUAL TABLE x1 USING lsm1(testlsm.db,a,UINT,b,c,d);
+  PRAGMA table_info(x1);
+} {
+  0 a UINT 1 {} 1 
+  1 b {} 0 {} 0 
+  2 c {} 0 {} 0 
+  3 d {} 0 {} 0
+}
+
+do_execsql_test 110 {
+  INSERT INTO x1(a,b,c,d) VALUES(15, 11, 22, 33),(8,'banjo',x'333231',NULL),
+      (12,NULL,3.25,-559281390);
+  SELECT a, quote(b), quote(c), quote(d) FROM x1;
+} {8 'banjo' X'333231' NULL 12 NULL 3.25 -559281390 15 11 22 33}
+do_execsql_test 111 {
+  SELECT a, quote(lsm1_key), quote(lsm1_value) FROM x1;
+} {8 X'08' X'2162616E6A6F1633323105' 12 X'0C' X'05320000000000000A401FFB42ABE9DB' 15 X'0F' X'4284C6'}
+
+do_execsql_test 120 {
+  UPDATE x1 SET d = d+1.0 WHERE a=15;
+  SELECT a, quote(b), quote(c), quote(d) FROM x1;
+} {8 'banjo' X'333231' NULL 12 NULL 3.25 -559281390 15 11 22 34.0}
+
+do_execsql_test 130 {
+  UPDATE x1 SET a=123456789 WHERE a=12;
+  SELECT a, quote(b), quote(c), quote(d) FROM x1;
+} {8 'banjo' X'333231' NULL 15 11 22 34.0 123456789 NULL 3.25 -559281390}
+do_execsql_test 131 {
+  SELECT quote(lsm1_key), printf('0x%x',a) FROM x1 WHERE a > 100000000;
+} {X'FB075BCD15' 0x75bcd15}
+
+do_execsql_test 140 {
+  DELETE FROM x1 WHERE a=15;
+  SELECT a, quote(b), quote(c), quote(d) FROM x1;
+} {8 'banjo' X'333231' NULL 123456789 NULL 3.25 -559281390}
+
+do_test 150 {
+  lsort [glob testlsm.db*]
+} {testlsm.db testlsm.db-log testlsm.db-shm}
+
+db close
+do_test 160 {
+  lsort [glob testlsm.db*]
+} {testlsm.db}
+
+forcedelete testlsm.db
+forcedelete test.db
+sqlite3 db test.db
+load_lsm1_vtab db
+
+
+do_execsql_test 200 {
+  CREATE VIRTUAL TABLE x1 USING lsm1(testlsm.db,a,TEXT,b,c,d);
+  PRAGMA table_info(x1);
+} {
+  0 a TEXT 1 {} 1 
+  1 b {} 0 {} 0 
+  2 c {} 0 {} 0 
+  3 d {} 0 {} 0
+}
+do_execsql_test 210 {
+  INSERT INTO x1(a,b,c,d) VALUES(15, 11, 22, 33),(8,'banjo',x'333231',NULL),
+      (12,NULL,3.25,-559281390);
+  SELECT quote(a), quote(b), quote(c), quote(d), '|' FROM x1;
+} {'12' NULL 3.25 -559281390 | '15' 11 22 33 | '8' 'banjo' X'333231' NULL |}
+do_execsql_test 211 {
+  SELECT quote(a), quote(lsm1_key), quote(lsm1_value), '|' FROM x1;
+} {'12' X'3132' X'05320000000000000A401FFB42ABE9DB' | '15' X'3135' X'4284C6' | '8' X'38' X'2162616E6A6F1633323105' |}
+do_execsql_test 212 {
+  SELECT quote(a), quote(lsm1_key), quote(lsm1_value) FROM x1 WHERE a='12';
+} {'12' X'3132' X'05320000000000000A401FFB42ABE9DB'}
+
+#-------------------------------------------------------------------------
+reset_db
+forcedelete testlsm.db
+load_lsm1_vtab db
+do_execsql_test 300 {
+  CREATE VIRTUAL TABLE x1 USING lsm1(testlsm.db,a,TEXT,b,c,d);
+}
+do_eqp_test 310 {
+  SELECT * FROM x1 WHERE a=?
+} {SCAN TABLE x1 VIRTUAL TABLE INDEX 0:}
+
+do_eqp_test 320 {
+  SELECT * FROM x1 WHERE a>?
+} {SCAN TABLE x1 VIRTUAL TABLE INDEX 2:}
+
+do_eqp_test 330 {
+  SELECT * FROM x1 WHERE a<?
+} {SCAN TABLE x1 VIRTUAL TABLE INDEX 3:}
+do_eqp_test 340 {
+  SELECT * FROM x1 WHERE a BETWEEN ? AND ?
+} {SCAN TABLE x1 VIRTUAL TABLE INDEX 1:}
+
+#-------------------------------------------------------------------------
+reset_db
+forcedelete testlsm.db
+load_lsm1_vtab db
+do_execsql_test 400 {
+  CREATE VIRTUAL TABLE x1 USING lsm1(testlsm.db,a,TEXT,b);
+  INSERT INTO x1 VALUES('one', 1);
+  INSERT INTO x1 VALUES('two', 2);
+  INSERT INTO x1 VALUES('three', 3);
+  INSERT INTO x1 VALUES('four', 4);
+  INSERT INTO x1 VALUES('five', 5);
+}
+do_execsql_test 410 {
+  SELECT b FROM x1 WHERE a = 'two'
+} {2}
+do_execsql_test 411 {
+  SELECT b FROM x1 WHERE a = 'one'
+} {1}
+do_execsql_test 412 {
+  SELECT b FROM x1 WHERE a = 'five'
+} {5}
+
+do_execsql_test 420 {
+  SELECT b FROM x1 WHERE a BETWEEN 'one' AND 'three';
+} {1 3}
+do_execsql_test 421 {
+  SELECT b FROM x1 WHERE a BETWEEN 'five' AND 'two';
+} {5 4 1 3 2}
+do_execsql_test 421 {
+  SELECT b FROM x1 WHERE a > 'five';
+} {4 1 3 2}
+do_execsql_test 421 {
+  SELECT b FROM x1 WHERE a <= 'three';
+} {3 1 4 5}
+
+finish_test
diff --git a/ext/lsm1/tool/mklsm1c.tcl b/ext/lsm1/tool/mklsm1c.tcl
new file mode 100644
index 0000000..d4a317b
--- /dev/null
+++ b/ext/lsm1/tool/mklsm1c.tcl
@@ -0,0 +1,88 @@
+#!/bin/sh
+# restart with tclsh \
+exec tclsh "$0" "$@"
+
+set srcdir [file dirname [file dirname [info script]]]
+set G(src) [string map [list %dir% $srcdir] {
+  %dir%/lsm.h
+  %dir%/lsmInt.h
+  %dir%/lsm_vtab.c
+  %dir%/lsm_ckpt.c
+  %dir%/lsm_file.c
+  %dir%/lsm_log.c
+  %dir%/lsm_main.c
+  %dir%/lsm_mem.c
+  %dir%/lsm_mutex.c
+  %dir%/lsm_shared.c
+  %dir%/lsm_sorted.c
+  %dir%/lsm_str.c
+  %dir%/lsm_tree.c
+  %dir%/lsm_unix.c
+  %dir%/lsm_varint.c
+  %dir%/lsm_win32.c
+}]
+
+set G(hdr) {
+
+#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_LSM1) 
+
+#if !defined(NDEBUG) && !defined(SQLITE_DEBUG) 
+# define NDEBUG 1
+#endif
+#if defined(NDEBUG) && defined(SQLITE_DEBUG)
+# undef NDEBUG
+#endif
+
+}
+
+set G(footer) {
+    
+#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_LSM1) */
+}
+
+#-------------------------------------------------------------------------
+# Read and return the entire contents of text file $zFile from disk.
+#
+proc readfile {zFile} {
+  set fd [open $zFile]
+  set data [read $fd]
+  close $fd
+  return $data
+}
+
+proc lsm1c_init {zOut} {
+  global G
+  set G(fd) stdout
+  set G(fd) [open $zOut w]
+
+  puts -nonewline $G(fd) $G(hdr)
+}
+
+proc lsm1c_printfile {zIn} {
+  global G
+  set data [readfile $zIn]
+  set zTail [file tail $zIn]
+  puts $G(fd) "#line 1 \"$zTail\""
+
+  foreach line [split $data "\n"] {
+    if {[regexp {^# *include.*lsm} $line]} {
+      set line "/* $line */"
+    } elseif { [regexp {^(const )?[a-zA-Z][a-zA-Z0-9]* [*]?lsm[^_]} $line] } {
+      set line "static $line"
+    }
+    puts $G(fd) $line
+  }
+}
+
+proc lsm1c_close {} {
+  global G
+  puts -nonewline $G(fd) $G(footer)
+  if {$G(fd)!="stdout"} {
+    close $G(fd)
+  }
+}
+
+
+lsm1c_init lsm1.c
+foreach f $G(src) { lsm1c_printfile $f }
+lsm1c_close