Adding upstream version 2:4.17.12+dfsg.upstream/2%4.17.12+dfsgupstream

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

16 files changed, 9959 insertions, 0 deletions

diff --git a/lib/tdb/common/check.c b/lib/tdb/common/check.c
new file mode 100644
index 0000000..d7741f6
--- /dev/null
+++ b/lib/tdb/common/check.c
@@ -0,0 +1,489 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Rusty Russell		   2009
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "tdb_private.h"
+
+/* Since we opened it, these shouldn't fail unless it's recent corruption. */
+static bool tdb_check_header(struct tdb_context *tdb, tdb_off_t *recovery)
+{
+	struct tdb_header hdr;
+	uint32_t h1, h2;
+
+	if (tdb->methods->tdb_read(tdb, 0, &hdr, sizeof(hdr), 0) == -1)
+		return false;
+	if (strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0)
+		goto corrupt;
+
+	CONVERT(hdr);
+	if (hdr.version != TDB_VERSION)
+		goto corrupt;
+
+	if (hdr.rwlocks != 0 &&
+	    hdr.rwlocks != TDB_FEATURE_FLAG_MAGIC &&
+	    hdr.rwlocks != TDB_HASH_RWLOCK_MAGIC)
+		goto corrupt;
+
+	tdb_header_hash(tdb, &h1, &h2);
+	if (hdr.magic1_hash && hdr.magic2_hash &&
+	    (hdr.magic1_hash != h1 || hdr.magic2_hash != h2))
+		goto corrupt;
+
+	if (hdr.hash_size == 0)
+		goto corrupt;
+
+	if (hdr.hash_size != tdb->hash_size)
+		goto corrupt;
+
+	if (hdr.recovery_start != 0 &&
+	    hdr.recovery_start < TDB_DATA_START(tdb->hash_size))
+		goto corrupt;
+
+	*recovery = hdr.recovery_start;
+	return true;
+
+corrupt:
+	tdb->ecode = TDB_ERR_CORRUPT;
+	TDB_LOG((tdb, TDB_DEBUG_ERROR, "Header is corrupt\n"));
+	return false;
+}
+
+/* Generic record header check. */
+static bool tdb_check_record(struct tdb_context *tdb,
+			     tdb_off_t off,
+			     const struct tdb_record *rec)
+{
+	tdb_off_t tailer;
+
+	/* Check rec->next: 0 or points to record offset, aligned. */
+	if (rec->next > 0 && rec->next < TDB_DATA_START(tdb->hash_size)){
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u too small next %u\n",
+			 off, rec->next));
+		goto corrupt;
+	}
+	if (rec->next + sizeof(*rec) < rec->next) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u too large next %u\n",
+			 off, rec->next));
+		goto corrupt;
+	}
+	if ((rec->next % TDB_ALIGNMENT) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u misaligned next %u\n",
+			 off, rec->next));
+		goto corrupt;
+	}
+	if (tdb_oob(tdb, rec->next, sizeof(*rec), 0))
+		goto corrupt;
+
+	/* Check rec_len: similar to rec->next, implies next record. */
+	if ((rec->rec_len % TDB_ALIGNMENT) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u misaligned length %u\n",
+			 off, rec->rec_len));
+		goto corrupt;
+	}
+	/* Must fit tailer. */
+	if (rec->rec_len < sizeof(tailer)) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u too short length %u\n",
+			 off, rec->rec_len));
+		goto corrupt;
+	}
+	/* OOB allows "right at the end" access, so this works for last rec. */
+	if (tdb_oob(tdb, off, sizeof(*rec)+rec->rec_len, 0))
+		goto corrupt;
+
+	/* Check tailer. */
+	if (tdb_ofs_read(tdb, off+sizeof(*rec)+rec->rec_len-sizeof(tailer),
+			 &tailer) == -1)
+		goto corrupt;
+	if (tailer != sizeof(*rec) + rec->rec_len) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u invalid tailer\n", off));
+		goto corrupt;
+	}
+
+	return true;
+
+corrupt:
+	tdb->ecode = TDB_ERR_CORRUPT;
+	return false;
+}
+
+/* Grab some bytes: may copy if can't use mmap.
+   Caller has already done bounds check. */
+static TDB_DATA get_bytes(struct tdb_context *tdb,
+			  tdb_off_t off, tdb_len_t len)
+{
+	TDB_DATA d;
+
+	d.dsize = len;
+
+	if (tdb->transaction == NULL && tdb->map_ptr != NULL)
+		d.dptr = (unsigned char *)tdb->map_ptr + off;
+	else
+		d.dptr = tdb_alloc_read(tdb, off, d.dsize);
+	return d;
+}
+
+/* Frees data if we're not able to simply use mmap. */
+static void put_bytes(struct tdb_context *tdb, TDB_DATA d)
+{
+	if (tdb->transaction == NULL && tdb->map_ptr != NULL)
+		return;
+	free(d.dptr);
+}
+
+/* We use the excellent Jenkins lookup3 hash; this is based on hash_word2.
+ * See: http://burtleburtle.net/bob/c/lookup3.c
+ */
+#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
+static void hash(uint32_t key, uint32_t *pc, uint32_t *pb)
+{
+	uint32_t a,b,c;
+
+	/* Set up the internal state */
+	a = b = c = 0xdeadbeef + *pc;
+	c += *pb;
+	a += key;
+	c ^= b; c -= rot(b,14);
+	a ^= c; a -= rot(c,11);
+	b ^= a; b -= rot(a,25);
+	c ^= b; c -= rot(b,16);
+	a ^= c; a -= rot(c,4);
+	b ^= a; b -= rot(a,14);
+	c ^= b; c -= rot(b,24);
+	*pc=c; *pb=b;
+}
+
+/*
+  We want to check that all free records are in the free list
+  (only once), and all free list entries are free records.  Similarly
+  for each hash chain of used records.
+
+  Doing that naively (without walking hash chains, since we want to be
+  linear) means keeping a list of records which have been seen in each
+  hash chain, and another of records pointed to (ie. next pointers
+  from records and the initial hash chain heads).  These two lists
+  should be equal.  This will take 8 bytes per record, and require
+  sorting at the end.
+
+  So instead, we record each offset in a bitmap such a way that
+  recording it twice will cancel out.  Since each offset should appear
+  exactly twice, the bitmap should be zero at the end.
+
+  The approach was inspired by Bloom Filters (see Wikipedia).  For
+  each value, we flip K bits in a bitmap of size N.  The number of
+  distinct arrangements is:
+
+	N! / (K! * (N-K)!)
+
+  Of course, not all arrangements are actually distinct, but testing
+  shows this formula to be close enough.
+
+  So, if K == 8 and N == 256, the probability of two things flipping the same
+  bits is 1 in 409,663,695,276,000.
+
+  Given that ldb uses a hash size of 10000, using 32 bytes per hash chain
+  (320k) seems reasonable.
+*/
+#define NUM_HASHES 8
+#define BITMAP_BITS 256
+
+static void bit_flip(unsigned char bits[], unsigned int idx)
+{
+	bits[idx / CHAR_BIT] ^= (1 << (idx % CHAR_BIT));
+}
+
+/* We record offsets in a bitmap for the particular chain it should be in.  */
+static void record_offset(unsigned char bits[], tdb_off_t off)
+{
+	uint32_t h1 = off, h2 = 0;
+	unsigned int i;
+
+	/* We get two good hash values out of jhash2, so we use both.  Then
+	 * we keep going to produce further hash values. */
+	for (i = 0; i < NUM_HASHES / 2; i++) {
+		hash(off, &h1, &h2);
+		bit_flip(bits, h1 % BITMAP_BITS);
+		bit_flip(bits, h2 % BITMAP_BITS);
+		h2++;
+	}
+}
+
+/* Check that an in-use record is valid. */
+static bool tdb_check_used_record(struct tdb_context *tdb,
+				  tdb_off_t off,
+				  const struct tdb_record *rec,
+				  unsigned char **hashes,
+				  int (*check)(TDB_DATA, TDB_DATA, void *),
+				  void *private_data)
+{
+	TDB_DATA key, data;
+	tdb_len_t len;
+
+	if (!tdb_check_record(tdb, off, rec))
+		return false;
+
+	/* key + data + tailer must fit in record */
+	len = rec->key_len;
+	len += rec->data_len;
+	if (len < rec->data_len) {
+		/* overflow */
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "Record lengths overflow\n"));
+		return false;
+	}
+	len += sizeof(tdb_off_t);
+	if (len < sizeof(tdb_off_t)) {
+		/* overflow */
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "Record lengths overflow\n"));
+		return false;
+	}
+
+	if (len > rec->rec_len) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u too short for contents\n", off));
+		return false;
+	}
+
+	key = get_bytes(tdb, off + sizeof(*rec), rec->key_len);
+	if (!key.dptr)
+		return false;
+
+	if (tdb->hash_fn(&key) != rec->full_hash) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Record offset %u has incorrect hash\n", off));
+		goto fail_put_key;
+	}
+
+	/* Mark this offset as a known value for this hash bucket. */
+	record_offset(hashes[BUCKET(rec->full_hash)+1], off);
+	/* And similarly if the next pointer is valid. */
+	if (rec->next)
+		record_offset(hashes[BUCKET(rec->full_hash)+1], rec->next);
+
+	/* If they supply a check function and this record isn't dead,
+	   get data and feed it. */
+	if (check && rec->magic != TDB_DEAD_MAGIC) {
+		data = get_bytes(tdb, off + sizeof(*rec) + rec->key_len,
+				 rec->data_len);
+		if (!data.dptr)
+			goto fail_put_key;
+
+		if (check(key, data, private_data) == -1)
+			goto fail_put_data;
+		put_bytes(tdb, data);
+	}
+
+	put_bytes(tdb, key);
+	return true;
+
+fail_put_data:
+	put_bytes(tdb, data);
+fail_put_key:
+	put_bytes(tdb, key);
+	return false;
+}
+
+/* Check that an unused record is valid. */
+static bool tdb_check_free_record(struct tdb_context *tdb,
+				  tdb_off_t off,
+				  const struct tdb_record *rec,
+				  unsigned char **hashes)
+{
+	if (!tdb_check_record(tdb, off, rec))
+		return false;
+
+	/* Mark this offset as a known value for the free list. */
+	record_offset(hashes[0], off);
+	/* And similarly if the next pointer is valid. */
+	if (rec->next)
+		record_offset(hashes[0], rec->next);
+	return true;
+}
+
+/* Slow, but should be very rare. */
+size_t tdb_dead_space(struct tdb_context *tdb, tdb_off_t off)
+{
+	size_t len;
+
+	for (len = 0; off + len < tdb->map_size; len++) {
+		char c;
+		if (tdb->methods->tdb_read(tdb, off, &c, 1, 0))
+			return 0;
+		if (c != 0 && c != 0x42)
+			break;
+	}
+	return len;
+}
+
+_PUBLIC_ int tdb_check(struct tdb_context *tdb,
+	      int (*check)(TDB_DATA key, TDB_DATA data, void *private_data),
+	      void *private_data)
+{
+	unsigned int h;
+	unsigned char **hashes;
+	tdb_off_t off, recovery_start;
+	struct tdb_record rec;
+	bool found_recovery = false;
+	tdb_len_t dead;
+	bool locked;
+
+	/* Read-only databases use no locking at all: it's best-effort.
+	 * We may have a write lock already, so skip that case too. */
+	if (tdb->read_only || tdb->allrecord_lock.count != 0) {
+		locked = false;
+	} else {
+		if (tdb_lockall_read(tdb) == -1)
+			return -1;
+		locked = true;
+	}
+
+	/* Make sure we know true size of the underlying file. */
+	tdb_oob(tdb, tdb->map_size, 1, 1);
+
+	/* Header must be OK: also gets us the recovery ptr, if any. */
+	if (!tdb_check_header(tdb, &recovery_start))
+		goto unlock;
+
+	/* We should have the whole header, too. */
+	if (tdb->map_size < TDB_DATA_START(tdb->hash_size)) {
+		tdb->ecode = TDB_ERR_CORRUPT;
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "File too short for hashes\n"));
+		goto unlock;
+	}
+
+	/* One big malloc: pointers then bit arrays. */
+	hashes = (unsigned char **)calloc(
+			1, sizeof(hashes[0]) * (1+tdb->hash_size)
+			+ BITMAP_BITS / CHAR_BIT * (1+tdb->hash_size));
+	if (!hashes) {
+		tdb->ecode = TDB_ERR_OOM;
+		goto unlock;
+	}
+
+	/* Initialize pointers */
+	hashes[0] = (unsigned char *)(&hashes[1+tdb->hash_size]);
+	for (h = 1; h < 1+tdb->hash_size; h++)
+		hashes[h] = hashes[h-1] + BITMAP_BITS / CHAR_BIT;
+
+	/* Freelist and hash headers are all in a row: read them. */
+	for (h = 0; h < 1+tdb->hash_size; h++) {
+		if (tdb_ofs_read(tdb, FREELIST_TOP + h*sizeof(tdb_off_t),
+				 &off) == -1)
+			goto free;
+		if (off)
+			record_offset(hashes[h], off);
+	}
+
+	/* For each record, read it in and check it's ok. */
+	for (off = TDB_DATA_START(tdb->hash_size);
+	     off < tdb->map_size;
+	     off += sizeof(rec) + rec.rec_len) {
+		if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
+					   DOCONV()) == -1)
+			goto free;
+		switch (rec.magic) {
+		case TDB_MAGIC:
+		case TDB_DEAD_MAGIC:
+			if (!tdb_check_used_record(tdb, off, &rec, hashes,
+						   check, private_data))
+				goto free;
+			break;
+		case TDB_FREE_MAGIC:
+			if (!tdb_check_free_record(tdb, off, &rec, hashes))
+				goto free;
+			break;
+		/* If we crash after ftruncate, we can get zeroes or fill. */
+		case TDB_RECOVERY_INVALID_MAGIC:
+		case 0x42424242:
+			if (recovery_start == off) {
+				found_recovery = true;
+				break;
+			}
+			dead = tdb_dead_space(tdb, off);
+			if (dead < sizeof(rec))
+				goto corrupt;
+
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "Dead space at %u-%u (of %u)\n",
+				 off, off + dead, tdb->map_size));
+			rec.rec_len = dead - sizeof(rec);
+			break;
+		case TDB_RECOVERY_MAGIC:
+			if (recovery_start != off) {
+				TDB_LOG((tdb, TDB_DEBUG_ERROR,
+					 "Unexpected recovery record at offset %u\n",
+					 off));
+				goto free;
+			}
+			found_recovery = true;
+			break;
+		default: ;
+		corrupt:
+			tdb->ecode = TDB_ERR_CORRUPT;
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "Bad magic 0x%x at offset %u\n",
+				 rec.magic, off));
+			goto free;
+		}
+	}
+
+	/* Now, hashes should all be empty: each record exists and is referred
+	 * to by one other. */
+	for (h = 0; h < 1+tdb->hash_size; h++) {
+		unsigned int i;
+		for (i = 0; i < BITMAP_BITS / CHAR_BIT; i++) {
+			if (hashes[h][i] != 0) {
+				tdb->ecode = TDB_ERR_CORRUPT;
+				TDB_LOG((tdb, TDB_DEBUG_ERROR,
+					 "Hashes do not match records\n"));
+				goto free;
+			}
+		}
+	}
+
+	/* We must have found recovery area if there was one. */
+	if (recovery_start != 0 && !found_recovery) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "Expected a recovery area at %u\n",
+			 recovery_start));
+		goto free;
+	}
+
+	free(hashes);
+	if (locked) {
+		tdb_unlockall_read(tdb);
+	}
+	return 0;
+
+free:
+	free(hashes);
+unlock:
+	if (locked) {
+		tdb_unlockall_read(tdb);
+	}
+	return -1;
+}
diff --git a/lib/tdb/common/dump.c b/lib/tdb/common/dump.c
new file mode 100644
index 0000000..adcf591
--- /dev/null
+++ b/lib/tdb/common/dump.c
@@ -0,0 +1,149 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+static tdb_off_t tdb_dump_record(struct tdb_context *tdb, int hash,
+				 tdb_off_t offset)
+{
+	struct tdb_record rec;
+	tdb_off_t tailer_ofs, tailer;
+
+	if (tdb->methods->tdb_read(tdb, offset, (char *)&rec,
+				   sizeof(rec), DOCONV()) == -1) {
+		printf("ERROR: failed to read record at %u\n", offset);
+		return 0;
+	}
+
+	printf(" rec: hash=%d offset=0x%08x next=0x%08x rec_len=%u "
+	       "key_len=%u data_len=%u full_hash=0x%08x magic=0x%08x\n",
+	       hash, offset, rec.next, rec.rec_len, rec.key_len, rec.data_len,
+	       rec.full_hash, rec.magic);
+
+	tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off_t);
+
+	if (tdb_ofs_read(tdb, tailer_ofs, &tailer) == -1) {
+		printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
+		return rec.next;
+	}
+
+	if (tailer != rec.rec_len + sizeof(rec)) {
+		printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
+				(unsigned int)tailer, (unsigned int)(rec.rec_len + sizeof(rec)));
+	}
+	return rec.next;
+}
+
+static int tdb_dump_chain(struct tdb_context *tdb, int i)
+{
+	struct tdb_chainwalk_ctx chainwalk;
+	tdb_off_t rec_ptr, top;
+
+	if (i == -1) {
+		top = FREELIST_TOP;
+	} else {
+		top = TDB_HASH_TOP(i);
+	}
+
+	if (tdb_lock(tdb, i, F_WRLCK) != 0)
+		return -1;
+
+	if (tdb_ofs_read(tdb, top, &rec_ptr) == -1)
+		return tdb_unlock(tdb, i, F_WRLCK);
+
+	tdb_chainwalk_init(&chainwalk, rec_ptr);
+
+	if (rec_ptr)
+		printf("hash=%d\n", i);
+
+	while (rec_ptr) {
+		bool ok;
+		rec_ptr = tdb_dump_record(tdb, i, rec_ptr);
+		ok = tdb_chainwalk_check(tdb, &chainwalk, rec_ptr);
+		if (!ok) {
+			printf("circular hash chain %d\n", i);
+			break;
+		}
+	}
+
+	return tdb_unlock(tdb, i, F_WRLCK);
+}
+
+_PUBLIC_ void tdb_dump_all(struct tdb_context *tdb)
+{
+	uint32_t i;
+	for (i=0;i<tdb->hash_size;i++) {
+		tdb_dump_chain(tdb, i);
+	}
+	printf("freelist:\n");
+	tdb_dump_chain(tdb, -1);
+}
+
+_PUBLIC_ int tdb_printfreelist(struct tdb_context *tdb)
+{
+	int ret;
+	long total_free = 0;
+	tdb_off_t offset, rec_ptr;
+	struct tdb_record rec;
+
+	if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
+		return ret;
+
+	offset = FREELIST_TOP;
+
+	/* read in the freelist top */
+	if (tdb_ofs_read(tdb, offset, &rec_ptr) == -1) {
+		tdb_unlock(tdb, -1, F_WRLCK);
+		return 0;
+	}
+
+	printf("freelist top=[0x%08x]\n", rec_ptr );
+	while (rec_ptr) {
+		if (tdb->methods->tdb_read(tdb, rec_ptr, (char *)&rec,
+					   sizeof(rec), DOCONV()) == -1) {
+			tdb_unlock(tdb, -1, F_WRLCK);
+			return -1;
+		}
+
+		if (rec.magic != TDB_FREE_MAGIC) {
+			printf("bad magic 0x%08x in free list\n", rec.magic);
+			tdb_unlock(tdb, -1, F_WRLCK);
+			return -1;
+		}
+
+		printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%u)] (end = 0x%08x)\n",
+		       rec_ptr, rec.rec_len, rec.rec_len, rec_ptr + rec.rec_len);
+		total_free += rec.rec_len;
+
+		/* move to the next record */
+		rec_ptr = rec.next;
+	}
+	printf("total rec_len = [0x%08lx (%lu)]\n", total_free, total_free);
+
+	return tdb_unlock(tdb, -1, F_WRLCK);
+}
+
diff --git a/lib/tdb/common/error.c b/lib/tdb/common/error.c
new file mode 100644
index 0000000..478eb88
--- /dev/null
+++ b/lib/tdb/common/error.c
@@ -0,0 +1,57 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+_PUBLIC_ enum TDB_ERROR tdb_error(struct tdb_context *tdb)
+{
+	return tdb->ecode;
+}
+
+static struct tdb_errname {
+	enum TDB_ERROR ecode; const char *estring;
+} emap[] = { {TDB_SUCCESS, "Success"},
+	     {TDB_ERR_CORRUPT, "Corrupt database"},
+	     {TDB_ERR_IO, "IO Error"},
+	     {TDB_ERR_LOCK, "Locking error"},
+	     {TDB_ERR_OOM, "Out of memory"},
+	     {TDB_ERR_EXISTS, "Record exists"},
+	     {TDB_ERR_NOLOCK, "Lock exists on other keys"},
+	     {TDB_ERR_EINVAL, "Invalid parameter"},
+	     {TDB_ERR_NOEXIST, "Record does not exist"},
+	     {TDB_ERR_RDONLY, "write not permitted"} };
+
+/* Error string for the last tdb error */
+_PUBLIC_ const char *tdb_errorstr(struct tdb_context *tdb)
+{
+	uint32_t i;
+	for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
+		if (tdb->ecode == emap[i].ecode)
+			return emap[i].estring;
+	return "Invalid error code";
+}
+
diff --git a/lib/tdb/common/freelist.c b/lib/tdb/common/freelist.c
new file mode 100644
index 0000000..046c747
--- /dev/null
+++ b/lib/tdb/common/freelist.c
@@ -0,0 +1,747 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/* read a freelist record and check for simple errors */
+int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off, struct tdb_record *rec)
+{
+	if (tdb->methods->tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
+		return -1;
+
+	if (rec->magic == TDB_MAGIC) {
+		/* this happens when a app is showdown while deleting a record - we should
+		   not completely fail when this happens */
+		TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_rec_free_read non-free magic 0x%x at offset=%u - fixing\n",
+			 rec->magic, off));
+		rec->magic = TDB_FREE_MAGIC;
+		if (tdb_rec_write(tdb, off, rec) == -1)
+			return -1;
+	}
+
+	if (rec->magic != TDB_FREE_MAGIC) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_CORRUPT;
+		TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_rec_free_read bad magic 0x%x at offset=%u\n",
+			   rec->magic, off));
+		return -1;
+	}
+	if (tdb_oob(tdb, rec->next, sizeof(*rec), 0) != 0)
+		return -1;
+	return 0;
+}
+
+/* update a record tailer (must hold allocation lock) */
+static int update_tailer(struct tdb_context *tdb, tdb_off_t offset,
+			 const struct tdb_record *rec)
+{
+	tdb_off_t totalsize;
+
+	/* Offset of tailer from record header */
+	totalsize = sizeof(*rec) + rec->rec_len;
+	return tdb_ofs_write(tdb, offset + totalsize - sizeof(tdb_off_t),
+			 &totalsize);
+}
+
+/**
+ * Read the record directly on the left.
+ * Fail if there is no record on the left.
+ */
+static int read_record_on_left(struct tdb_context *tdb, tdb_off_t rec_ptr,
+			       tdb_off_t *left_p,
+			       struct tdb_record *left_r)
+{
+	tdb_off_t left_ptr;
+	tdb_off_t left_size;
+	struct tdb_record left_rec;
+	int ret;
+
+	left_ptr = rec_ptr - sizeof(tdb_off_t);
+
+	if (left_ptr <= TDB_DATA_START(tdb->hash_size)) {
+		/* no record on the left */
+		return -1;
+	}
+
+	/* Read in tailer and jump back to header */
+	ret = tdb_ofs_read(tdb, left_ptr, &left_size);
+	if (ret == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,
+			"tdb_free: left offset read failed at %u\n", left_ptr));
+		return -1;
+	}
+
+	/* it could be uninitialised data */
+	if (left_size == 0 || left_size == TDB_PAD_U32) {
+		return -1;
+	}
+
+	if (left_size > rec_ptr) {
+		return -1;
+	}
+
+	left_ptr = rec_ptr - left_size;
+
+	if (left_ptr < TDB_DATA_START(tdb->hash_size)) {
+		return -1;
+	}
+
+	/* Now read in the left record */
+	ret = tdb->methods->tdb_read(tdb, left_ptr, &left_rec,
+				     sizeof(left_rec), DOCONV());
+	if (ret == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,
+			 "tdb_free: left read failed at %u (%u)\n",
+			 left_ptr, left_size));
+		return -1;
+	}
+
+	*left_p = left_ptr;
+	*left_r = left_rec;
+
+	return 0;
+}
+
+/**
+ * Merge new freelist record with the direct left neighbour.
+ * This assumes that left_rec represents the record
+ * directly to the left of right_rec and that this is
+ * a freelist record.
+ */
+static int merge_with_left_record(struct tdb_context *tdb,
+				  tdb_off_t left_ptr,
+				  struct tdb_record *left_rec,
+				  struct tdb_record *right_rec)
+{
+	int ret;
+
+	left_rec->rec_len += sizeof(*right_rec) + right_rec->rec_len;
+
+	ret = tdb_rec_write(tdb, left_ptr, left_rec);
+	if (ret == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,
+			 "merge_with_left_record: update_left failed at %u\n",
+			 left_ptr));
+		return -1;
+	}
+
+	ret = update_tailer(tdb, left_ptr, left_rec);
+	if (ret == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,
+			 "merge_with_left_record: update_tailer failed at %u\n",
+			 left_ptr));
+		return -1;
+	}
+
+	return 0;
+}
+
+/**
+ * Check whether the record left of a given freelist record is
+ * also a freelist record, and if so, merge the two records.
+ *
+ * Return code:
+ *  -1 upon error
+ *   0 if left was not a free record
+ *   1 if left was free and successfully merged.
+ *
+ * The current record is handed in with pointer and fully read record.
+ *
+ * The left record pointer and struct can be retrieved as result
+ * in lp and lr;
+ */
+static int check_merge_with_left_record(struct tdb_context *tdb,
+					tdb_off_t rec_ptr,
+					struct tdb_record *rec,
+					tdb_off_t *lp,
+					struct tdb_record *lr)
+{
+	tdb_off_t left_ptr;
+	struct tdb_record left_rec;
+	int ret;
+
+	ret = read_record_on_left(tdb, rec_ptr, &left_ptr, &left_rec);
+	if (ret != 0) {
+		return 0;
+	}
+
+	if (left_rec.magic != TDB_FREE_MAGIC) {
+		return 0;
+	}
+
+	/* It's free - expand to include it. */
+	ret = merge_with_left_record(tdb, left_ptr, &left_rec, rec);
+	if (ret != 0) {
+		return -1;
+	}
+
+	if (lp != NULL) {
+		*lp = left_ptr;
+	}
+
+	if (lr != NULL) {
+		*lr = left_rec;
+	}
+
+	return 1;
+}
+
+/**
+ * Check whether the record left of a given freelist record is
+ * also a freelist record, and if so, merge the two records.
+ *
+ * Return code:
+ *  -1 upon error
+ *   0 if left was not a free record
+ *   1 if left was free and successfully merged.
+ *
+ * In this variant, the input record is specified just as the pointer
+ * and is read from the database if needed.
+ *
+ * next_ptr will contain the original record's next pointer after
+ * successful merging (which will be lost after merging), so that
+ * the caller can update the last pointer.
+ */
+static int check_merge_ptr_with_left_record(struct tdb_context *tdb,
+					    tdb_off_t rec_ptr,
+					    tdb_off_t *next_ptr)
+{
+	tdb_off_t left_ptr;
+	struct tdb_record rec, left_rec;
+	int ret;
+
+	ret = read_record_on_left(tdb, rec_ptr, &left_ptr, &left_rec);
+	if (ret != 0) {
+		return 0;
+	}
+
+	if (left_rec.magic != TDB_FREE_MAGIC) {
+		return 0;
+	}
+
+	/* It's free - expand to include it. */
+
+	ret = tdb->methods->tdb_read(tdb, rec_ptr, &rec,
+				     sizeof(rec), DOCONV());
+	if (ret != 0) {
+		return -1;
+	}
+
+	ret = merge_with_left_record(tdb, left_ptr, &left_rec, &rec);
+	if (ret != 0) {
+		return -1;
+	}
+
+	if (next_ptr != NULL) {
+		*next_ptr = rec.next;
+	}
+
+	return 1;
+}
+
+/**
+ * Add an element into the freelist.
+ *
+ * We merge the new record into the left record if it is also a
+ * free record, but not with the right one. This makes the
+ * operation O(1) instead of O(n): merging with the right record
+ * requires a traverse of the freelist to find the previous
+ * record in the free list.
+ *
+ * This prevents db traverses from being O(n^2) after a lot of deletes.
+ */
+int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+{
+	int ret;
+
+	/* Allocation and tailer lock */
+	if (tdb_lock(tdb, -1, F_WRLCK) != 0)
+		return -1;
+
+	/* set an initial tailer, so if we fail we don't leave a bogus record */
+	if (update_tailer(tdb, offset, rec) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed!\n"));
+		goto fail;
+	}
+
+	ret = check_merge_with_left_record(tdb, offset, rec, NULL, NULL);
+	if (ret == -1) {
+		goto fail;
+	}
+	if (ret == 1) {
+		/* merged */
+		goto done;
+	}
+
+	/* Nothing to merge, prepend to free list */
+
+	rec->magic = TDB_FREE_MAGIC;
+
+	if (tdb_ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
+	    tdb_rec_write(tdb, offset, rec) == -1 ||
+	    tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free record write failed at offset=%u\n", offset));
+		goto fail;
+	}
+
+done:
+	/* And we're done. */
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return 0;
+
+ fail:
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return -1;
+}
+
+
+
+/*
+   the core of tdb_allocate - called when we have decided which
+   free list entry to use
+
+   Note that we try to allocate by grabbing data from the end of an existing record,
+   not the beginning. This is so the left merge in a free is more likely to be
+   able to free up the record without fragmentation
+ */
+static tdb_off_t tdb_allocate_ofs(struct tdb_context *tdb,
+				  tdb_len_t length, tdb_off_t rec_ptr,
+				  struct tdb_record *rec, tdb_off_t last_ptr)
+{
+#define MIN_REC_SIZE (sizeof(struct tdb_record) + sizeof(tdb_off_t) + 8)
+
+	if (rec->rec_len < length + MIN_REC_SIZE) {
+		/* we have to grab the whole record */
+
+		/* unlink it from the previous record */
+		if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1) {
+			return 0;
+		}
+
+		/* mark it not free */
+		rec->magic = TDB_MAGIC;
+		if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
+			return 0;
+		}
+		return rec_ptr;
+	}
+
+	/* we're going to just shorten the existing record */
+	rec->rec_len -= (length + sizeof(*rec));
+	if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
+		return 0;
+	}
+	if (update_tailer(tdb, rec_ptr, rec) == -1) {
+		return 0;
+	}
+
+	/* and setup the new record */
+	rec_ptr += sizeof(*rec) + rec->rec_len;
+
+	memset(rec, '\0', sizeof(*rec));
+	rec->rec_len = length;
+	rec->magic = TDB_MAGIC;
+
+	if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
+		return 0;
+	}
+
+	if (update_tailer(tdb, rec_ptr, rec) == -1) {
+		return 0;
+	}
+
+	return rec_ptr;
+}
+
+/* allocate some space from the free list. The offset returned points
+   to a unconnected tdb_record within the database with room for at
+   least length bytes of total data
+
+   0 is returned if the space could not be allocated
+ */
+static tdb_off_t tdb_allocate_from_freelist(
+	struct tdb_context *tdb, tdb_len_t length, struct tdb_record *rec)
+{
+	tdb_off_t rec_ptr, last_ptr, newrec_ptr;
+	struct tdb_chainwalk_ctx chainwalk;
+	bool modified;
+	struct {
+		tdb_off_t rec_ptr, last_ptr;
+		tdb_len_t rec_len;
+	} bestfit;
+	float multiplier = 1.0;
+	bool merge_created_candidate;
+
+	/* over-allocate to reduce fragmentation */
+	length *= 1.25;
+
+	/* Extra bytes required for tailer */
+	length += sizeof(tdb_off_t);
+	length = TDB_ALIGN(length, TDB_ALIGNMENT);
+
+ again:
+	merge_created_candidate = false;
+	last_ptr = FREELIST_TOP;
+
+	/* read in the freelist top */
+	if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
+		return 0;
+
+	modified = false;
+	tdb_chainwalk_init(&chainwalk, rec_ptr);
+
+	bestfit.rec_ptr = 0;
+	bestfit.last_ptr = 0;
+	bestfit.rec_len = 0;
+
+	/*
+	   this is a best fit allocation strategy. Originally we used
+	   a first fit strategy, but it suffered from massive fragmentation
+	   issues when faced with a slowly increasing record size.
+	 */
+	while (rec_ptr) {
+		int ret;
+		tdb_off_t left_ptr;
+		struct tdb_record left_rec;
+
+		if (tdb_rec_free_read(tdb, rec_ptr, rec) == -1) {
+			return 0;
+		}
+
+		ret = check_merge_with_left_record(tdb, rec_ptr, rec,
+						   &left_ptr, &left_rec);
+		if (ret == -1) {
+			return 0;
+		}
+		if (ret == 1) {
+			/* merged */
+			rec_ptr = rec->next;
+			ret = tdb_ofs_write(tdb, last_ptr, &rec->next);
+			if (ret == -1) {
+				return 0;
+			}
+
+			/*
+			 * We have merged the current record into the left
+			 * neighbour. So our traverse of the freelist will
+			 * skip it and consider the next record in the chain.
+			 *
+			 * But the enlarged left neighbour may be a candidate.
+			 * If it is, we can not directly use it, though.
+			 * The only thing we can do and have to do here is to
+			 * update the current best fit size in the chain if the
+			 * current best fit is the left record. (By that we may
+			 * worsen the best fit we already had, bit this is not a
+			 * problem.)
+			 *
+			 * If the current best fit is not the left record,
+			 * all we can do is remember the fact that a merge
+			 * created a new candidate so that we can trigger
+			 * a second walk of the freelist if at the end of
+			 * the first walk we have not found any fit.
+			 * This way we can avoid expanding the database.
+			 */
+
+			if (bestfit.rec_ptr == left_ptr) {
+				bestfit.rec_len = left_rec.rec_len;
+			}
+
+			if (left_rec.rec_len > length) {
+				merge_created_candidate = true;
+			}
+
+			modified = true;
+
+			continue;
+		}
+
+		if (rec->rec_len >= length) {
+			if (bestfit.rec_ptr == 0 ||
+			    rec->rec_len < bestfit.rec_len) {
+				bestfit.rec_len = rec->rec_len;
+				bestfit.rec_ptr = rec_ptr;
+				bestfit.last_ptr = last_ptr;
+			}
+		}
+
+		/* move to the next record */
+		last_ptr = rec_ptr;
+		rec_ptr = rec->next;
+
+		if (!modified) {
+			bool ok;
+			ok = tdb_chainwalk_check(tdb, &chainwalk, rec_ptr);
+			if (!ok) {
+				return 0;
+			}
+		}
+
+		/* if we've found a record that is big enough, then
+		   stop searching if its also not too big. The
+		   definition of 'too big' changes as we scan
+		   through */
+		if (bestfit.rec_len > 0 &&
+		    bestfit.rec_len < length * multiplier) {
+			break;
+		}
+
+		/* this multiplier means we only extremely rarely
+		   search more than 50 or so records. At 50 records we
+		   accept records up to 11 times larger than what we
+		   want */
+		multiplier *= 1.05;
+	}
+
+	if (bestfit.rec_ptr != 0) {
+		if (tdb_rec_free_read(tdb, bestfit.rec_ptr, rec) == -1) {
+			return 0;
+		}
+
+		newrec_ptr = tdb_allocate_ofs(tdb, length, bestfit.rec_ptr,
+					      rec, bestfit.last_ptr);
+		return newrec_ptr;
+	}
+
+	if (merge_created_candidate) {
+		goto again;
+	}
+
+	/* we didn't find enough space. See if we can expand the
+	   database and if we can then try again */
+	if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
+		goto again;
+
+	return 0;
+}
+
+static bool tdb_alloc_dead(
+	struct tdb_context *tdb, int hash, tdb_len_t length,
+	tdb_off_t *rec_ptr, struct tdb_record *rec)
+{
+	tdb_off_t last_ptr;
+
+	*rec_ptr = tdb_find_dead(tdb, hash, rec, length, &last_ptr);
+	if (*rec_ptr == 0) {
+		return false;
+	}
+	/*
+	 * Unlink the record from the hash chain, it's about to be moved into
+	 * another one.
+	 */
+	return (tdb_ofs_write(tdb, last_ptr, &rec->next) == 0);
+}
+
+static void tdb_purge_dead(struct tdb_context *tdb, uint32_t hash)
+{
+	int max_dead_records = tdb->max_dead_records;
+
+	tdb->max_dead_records = 0;
+
+	tdb_trim_dead(tdb, hash);
+
+	tdb->max_dead_records = max_dead_records;
+}
+
+/*
+ * Chain "hash" is assumed to be locked
+ */
+
+tdb_off_t tdb_allocate(struct tdb_context *tdb, int hash, tdb_len_t length,
+		       struct tdb_record *rec)
+{
+	tdb_off_t ret;
+	uint32_t i;
+
+	if (tdb->max_dead_records == 0) {
+		/*
+		 * No dead records to expect anywhere. Do the blocking
+		 * freelist lock without trying to steal from others
+		 */
+		goto blocking_freelist_allocate;
+	}
+
+	/*
+	 * The following loop tries to get the freelist lock nonblocking. If
+	 * it gets the lock, allocate from there. If the freelist is busy,
+	 * instead of waiting we try to steal dead records from other hash
+	 * chains.
+	 *
+	 * Be aware that we do nonblocking locks on the other hash chains as
+	 * well and fail gracefully. This way we avoid deadlocks (we block two
+	 * hash chains, something which is pretty bad normally)
+	 */
+
+	for (i=0; i<tdb->hash_size; i++) {
+
+		int list;
+
+		list = BUCKET(hash+i);
+
+		if (tdb_lock_nonblock(tdb, list, F_WRLCK) == 0) {
+			bool got_dead;
+
+			got_dead = tdb_alloc_dead(tdb, list, length, &ret, rec);
+			tdb_unlock(tdb, list, F_WRLCK);
+
+			if (got_dead) {
+				return ret;
+			}
+		}
+
+		if (tdb_lock_nonblock(tdb, -1, F_WRLCK) == 0) {
+			/*
+			 * Under the freelist lock take the chance to give
+			 * back our dead records.
+			 */
+			tdb_purge_dead(tdb, hash);
+
+			ret = tdb_allocate_from_freelist(tdb, length, rec);
+			tdb_unlock(tdb, -1, F_WRLCK);
+			return ret;
+		}
+	}
+
+blocking_freelist_allocate:
+
+	if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+		return 0;
+	}
+	/*
+	 * Dead records can happen even if max_dead_records==0, they
+	 * are older than the max_dead_records concept: They happen if
+	 * tdb_delete happens concurrently with a traverse.
+	 */
+	tdb_purge_dead(tdb, hash);
+	ret = tdb_allocate_from_freelist(tdb, length, rec);
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return ret;
+}
+
+/**
+ * Merge adjacent records in the freelist.
+ */
+static int tdb_freelist_merge_adjacent(struct tdb_context *tdb,
+				       int *count_records, int *count_merged)
+{
+	tdb_off_t cur, next;
+	int count = 0;
+	int merged = 0;
+	int ret;
+
+	ret = tdb_lock(tdb, -1, F_RDLCK);
+	if (ret == -1) {
+		return -1;
+	}
+
+	cur = FREELIST_TOP;
+	while (tdb_ofs_read(tdb, cur, &next) == 0 && next != 0) {
+		tdb_off_t next2;
+
+		count++;
+
+		ret = check_merge_ptr_with_left_record(tdb, next, &next2);
+		if (ret == -1) {
+			goto done;
+		}
+		if (ret == 1) {
+			/*
+			 * merged:
+			 * now let cur->next point to next2 instead of next
+			 */
+
+			ret = tdb_ofs_write(tdb, cur, &next2);
+			if (ret != 0) {
+				goto done;
+			}
+
+			next = next2;
+			merged++;
+		}
+
+		cur = next;
+	}
+
+	if (count_records != NULL) {
+		*count_records = count;
+	}
+
+	if (count_merged != NULL) {
+		*count_merged = merged;
+	}
+
+	ret = 0;
+
+done:
+	tdb_unlock(tdb, -1, F_RDLCK);
+	return ret;
+}
+
+/**
+ * return the size of the freelist - no merging done
+ */
+static int tdb_freelist_size_no_merge(struct tdb_context *tdb)
+{
+	tdb_off_t ptr;
+	int count=0;
+
+	if (tdb_lock(tdb, -1, F_RDLCK) == -1) {
+		return -1;
+	}
+
+	ptr = FREELIST_TOP;
+	while (tdb_ofs_read(tdb, ptr, &ptr) == 0 && ptr != 0) {
+		count++;
+	}
+
+	tdb_unlock(tdb, -1, F_RDLCK);
+	return count;
+}
+
+/**
+ * return the size of the freelist - used to decide if we should repack
+ *
+ * As a side effect, adjacent records are merged unless the
+ * database is read-only, in order to reduce the fragmentation
+ * without repacking.
+ */
+_PUBLIC_ int tdb_freelist_size(struct tdb_context *tdb)
+{
+
+	int count = 0;
+
+	if (tdb->read_only) {
+		count = tdb_freelist_size_no_merge(tdb);
+	} else {
+		int ret;
+		ret = tdb_freelist_merge_adjacent(tdb, &count, NULL);
+		if (ret != 0) {
+			return -1;
+		}
+	}
+
+	return count;
+}
diff --git a/lib/tdb/common/freelistcheck.c b/lib/tdb/common/freelistcheck.c
new file mode 100644
index 0000000..2f1e6eb
--- /dev/null
+++ b/lib/tdb/common/freelistcheck.c
@@ -0,0 +1,107 @@
+/*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Jeremy Allison                    2006
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/* Check the freelist is good and contains no loops.
+   Very memory intensive - only do this as a consistency
+   checker. Heh heh - uses an in memory tdb as the storage
+   for the "seen" record list. For some reason this strikes
+   me as extremely clever as I don't have to write another tree
+   data structure implementation :-).
+ */
+
+static int seen_insert(struct tdb_context *mem_tdb, tdb_off_t rec_ptr)
+{
+	TDB_DATA key;
+
+	key.dptr = (unsigned char *)&rec_ptr;
+	key.dsize = sizeof(rec_ptr);
+	return tdb_store(mem_tdb, key, tdb_null, TDB_INSERT);
+}
+
+_PUBLIC_ int tdb_validate_freelist(struct tdb_context *tdb, int *pnum_entries)
+{
+	struct tdb_context *mem_tdb = NULL;
+	struct tdb_record rec;
+	tdb_off_t rec_ptr, last_ptr;
+	int ret = -1;
+
+	*pnum_entries = 0;
+
+	mem_tdb = tdb_open("flval", tdb->hash_size,
+				TDB_INTERNAL, O_RDWR, 0600);
+	if (!mem_tdb) {
+		return -1;
+	}
+
+	if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+		tdb_close(mem_tdb);
+		return 0;
+	}
+
+	last_ptr = FREELIST_TOP;
+
+	/* Store the FREELIST_TOP record. */
+	if (seen_insert(mem_tdb, last_ptr) == -1) {
+		tdb->ecode = TDB_ERR_CORRUPT;
+		ret = -1;
+		goto fail;
+	}
+
+	/* read in the freelist top */
+	if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1) {
+		goto fail;
+	}
+
+	while (rec_ptr) {
+
+		/* If we can't store this record (we've seen it
+		   before) then the free list has a loop and must
+		   be corrupt. */
+
+		if (seen_insert(mem_tdb, rec_ptr)) {
+			tdb->ecode = TDB_ERR_CORRUPT;
+			ret = -1;
+			goto fail;
+		}
+
+		if (tdb_rec_free_read(tdb, rec_ptr, &rec) == -1) {
+			goto fail;
+		}
+
+		/* move to the next record */
+		rec_ptr = rec.next;
+		*pnum_entries += 1;
+	}
+
+	ret = 0;
+
+  fail:
+
+	tdb_close(mem_tdb);
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return ret;
+}
diff --git a/lib/tdb/common/hash.c b/lib/tdb/common/hash.c
new file mode 100644
index 0000000..4de7ba9
--- /dev/null
+++ b/lib/tdb/common/hash.c
@@ -0,0 +1,345 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Rusty Russell		   2010
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "tdb_private.h"
+
+/* This is based on the hash algorithm from gdbm */
+unsigned int tdb_old_hash(TDB_DATA *key)
+{
+	uint32_t value;	/* Used to compute the hash value.  */
+	uint32_t   i;	/* Used to cycle through random values. */
+
+	/* Set the initial value from the key size. */
+	for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
+		value = (value + (key->dptr[i] << (i*5 % 24)));
+
+	return (1103515243 * value + 12345);
+}
+
+#ifndef WORDS_BIGENDIAN
+# define HASH_LITTLE_ENDIAN 1
+# define HASH_BIG_ENDIAN 0
+#else
+# define HASH_LITTLE_ENDIAN 0
+# define HASH_BIG_ENDIAN 1
+#endif
+
+/*
+-------------------------------------------------------------------------------
+lookup3.c, by Bob Jenkins, May 2006, Public Domain.
+
+These are functions for producing 32-bit hashes for hash table lookup.
+hash_word(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
+are externally useful functions.  Routines to test the hash are included
+if SELF_TEST is defined.  You can use this free for any purpose.  It's in
+the public domain.  It has no warranty.
+
+You probably want to use hashlittle().  hashlittle() and hashbig()
+hash byte arrays.  hashlittle() is is faster than hashbig() on
+little-endian machines.  Intel and AMD are little-endian machines.
+On second thought, you probably want hashlittle2(), which is identical to
+hashlittle() except it returns two 32-bit hashes for the price of one.
+You could implement hashbig2() if you wanted but I haven't bothered here.
+
+If you want to find a hash of, say, exactly 7 integers, do
+  a = i1;  b = i2;  c = i3;
+  mix(a,b,c);
+  a += i4; b += i5; c += i6;
+  mix(a,b,c);
+  a += i7;
+  final(a,b,c);
+then use c as the hash value.  If you have a variable length array of
+4-byte integers to hash, use hash_word().  If you have a byte array (like
+a character string), use hashlittle().  If you have several byte arrays, or
+a mix of things, see the comments above hashlittle().
+
+Why is this so big?  I read 12 bytes at a time into 3 4-byte integers,
+then mix those integers.  This is fast (you can do a lot more thorough
+mixing with 12*3 instructions on 3 integers than you can with 3 instructions
+on 1 byte), but shoehorning those bytes into integers efficiently is messy.
+*/
+
+#define hashsize(n) ((uint32_t)1<<(n))
+#define hashmask(n) (hashsize(n)-1)
+#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
+
+/*
+-------------------------------------------------------------------------------
+mix -- mix 3 32-bit values reversibly.
+
+This is reversible, so any information in (a,b,c) before mix() is
+still in (a,b,c) after mix().
+
+If four pairs of (a,b,c) inputs are run through mix(), or through
+mix() in reverse, there are at least 32 bits of the output that
+are sometimes the same for one pair and different for another pair.
+This was tested for:
+* pairs that differed by one bit, by two bits, in any combination
+  of top bits of (a,b,c), or in any combination of bottom bits of
+  (a,b,c).
+* "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
+  the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
+  is commonly produced by subtraction) look like a single 1-bit
+  difference.
+* the base values were pseudorandom, all zero but one bit set, or
+  all zero plus a counter that starts at zero.
+
+Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
+satisfy this are
+    4  6  8 16 19  4
+    9 15  3 18 27 15
+   14  9  3  7 17  3
+Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
+for "differ" defined as + with a one-bit base and a two-bit delta.  I
+used http://burtleburtle.net/bob/hash/avalanche.html to choose
+the operations, constants, and arrangements of the variables.
+
+This does not achieve avalanche.  There are input bits of (a,b,c)
+that fail to affect some output bits of (a,b,c), especially of a.  The
+most thoroughly mixed value is c, but it doesn't really even achieve
+avalanche in c.
+
+This allows some parallelism.  Read-after-writes are good at doubling
+the number of bits affected, so the goal of mixing pulls in the opposite
+direction as the goal of parallelism.  I did what I could.  Rotates
+seem to cost as much as shifts on every machine I could lay my hands
+on, and rotates are much kinder to the top and bottom bits, so I used
+rotates.
+-------------------------------------------------------------------------------
+*/
+#define mix(a,b,c) \
+{ \
+  a -= c;  a ^= rot(c, 4);  c += b; \
+  b -= a;  b ^= rot(a, 6);  a += c; \
+  c -= b;  c ^= rot(b, 8);  b += a; \
+  a -= c;  a ^= rot(c,16);  c += b; \
+  b -= a;  b ^= rot(a,19);  a += c; \
+  c -= b;  c ^= rot(b, 4);  b += a; \
+}
+
+/*
+-------------------------------------------------------------------------------
+final -- final mixing of 3 32-bit values (a,b,c) into c
+
+Pairs of (a,b,c) values differing in only a few bits will usually
+produce values of c that look totally different.  This was tested for
+* pairs that differed by one bit, by two bits, in any combination
+  of top bits of (a,b,c), or in any combination of bottom bits of
+  (a,b,c).
+* "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
+  the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
+  is commonly produced by subtraction) look like a single 1-bit
+  difference.
+* the base values were pseudorandom, all zero but one bit set, or
+  all zero plus a counter that starts at zero.
+
+These constants passed:
+ 14 11 25 16 4 14 24
+ 12 14 25 16 4 14 24
+and these came close:
+  4  8 15 26 3 22 24
+ 10  8 15 26 3 22 24
+ 11  8 15 26 3 22 24
+-------------------------------------------------------------------------------
+*/
+#define final(a,b,c) \
+{ \
+  c ^= b; c -= rot(b,14); \
+  a ^= c; a -= rot(c,11); \
+  b ^= a; b -= rot(a,25); \
+  c ^= b; c -= rot(b,16); \
+  a ^= c; a -= rot(c,4);  \
+  b ^= a; b -= rot(a,14); \
+  c ^= b; c -= rot(b,24); \
+}
+
+
+/*
+-------------------------------------------------------------------------------
+hashlittle() -- hash a variable-length key into a 32-bit value
+  k       : the key (the unaligned variable-length array of bytes)
+  length  : the length of the key, counting by bytes
+  val2    : IN: can be any 4-byte value OUT: second 32 bit hash.
+Returns a 32-bit value.  Every bit of the key affects every bit of
+the return value.  Two keys differing by one or two bits will have
+totally different hash values.  Note that the return value is better
+mixed than val2, so use that first.
+
+The best hash table sizes are powers of 2.  There is no need to do
+mod a prime (mod is sooo slow!).  If you need less than 32 bits,
+use a bitmask.  For example, if you need only 10 bits, do
+  h = (h & hashmask(10));
+In which case, the hash table should have hashsize(10) elements.
+
+If you are hashing n strings (uint8_t **)k, do it like this:
+  for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
+
+By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
+code any way you wish, private, educational, or commercial.  It's free.
+
+Use for hash table lookup, or anything where one collision in 2^^32 is
+acceptable.  Do NOT use for cryptographic purposes.
+-------------------------------------------------------------------------------
+*/
+
+static uint32_t hashlittle( const void *key, size_t length )
+{
+  uint32_t a,b,c;                                          /* internal state */
+  union { const void *ptr; size_t i; } u;     /* needed for Mac Powerbook G4 */
+
+  /* Set up the internal state */
+  a = b = c = 0xdeadbeef + ((uint32_t)length);
+
+  u.ptr = key;
+  if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
+    const uint32_t *k = (const uint32_t *)key;         /* read 32-bit chunks */
+    const uint8_t  *k8;
+
+    /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += k[0];
+      b += k[1];
+      c += k[2];
+      mix(a,b,c);
+      length -= 12;
+      k += 3;
+    }
+
+    /*----------------------------- handle the last (probably partial) block */
+    k8 = (const uint8_t *)k;
+    switch(length)
+    {
+    case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
+    case 11: c+=((uint32_t)k8[10])<<16; FALL_THROUGH;
+    case 10: c+=((uint32_t)k8[9])<<8;   FALL_THROUGH;
+    case 9 : c+=k8[8];                  FALL_THROUGH;
+    case 8 : b+=k[1]; a+=k[0]; break;
+    case 7 : b+=((uint32_t)k8[6])<<16;  FALL_THROUGH;
+    case 6 : b+=((uint32_t)k8[5])<<8;   FALL_THROUGH;
+    case 5 : b+=k8[4];                  FALL_THROUGH;
+    case 4 : a+=k[0]; break;
+    case 3 : a+=((uint32_t)k8[2])<<16;  FALL_THROUGH;
+    case 2 : a+=((uint32_t)k8[1])<<8;   FALL_THROUGH;
+    case 1 : a+=k8[0]; break;
+    case 0 : return c;
+    }
+  } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
+    const uint16_t *k = (const uint16_t *)key;         /* read 16-bit chunks */
+    const uint8_t  *k8;
+
+    /*--------------- all but last block: aligned reads and different mixing */
+    while (length > 12)
+    {
+      a += k[0] + (((uint32_t)k[1])<<16);
+      b += k[2] + (((uint32_t)k[3])<<16);
+      c += k[4] + (((uint32_t)k[5])<<16);
+      mix(a,b,c);
+      length -= 12;
+      k += 6;
+    }
+
+    /*----------------------------- handle the last (probably partial) block */
+    k8 = (const uint8_t *)k;
+    switch(length)
+    {
+    case 12: c+=k[4]+(((uint32_t)k[5])<<16);
+             b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 11: c+=((uint32_t)k8[10])<<16;      FALL_THROUGH;
+    case 10: c+=k[4];
+             b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 9 : c+=k8[8];                       FALL_THROUGH;
+    case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 7 : b+=((uint32_t)k8[6])<<16;       FALL_THROUGH;
+    case 6 : b+=k[2];
+             a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 5 : b+=k8[4];                       FALL_THROUGH;
+    case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
+             break;
+    case 3 : a+=((uint32_t)k8[2])<<16;       FALL_THROUGH;
+    case 2 : a+=k[0];
+             break;
+    case 1 : a+=k8[0];
+             break;
+    case 0 : return c;                     /* zero length requires no mixing */
+    }
+
+  } else {                        /* need to read the key one byte at a time */
+    const uint8_t *k = (const uint8_t *)key;
+
+    /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
+    while (length > 12)
+    {
+      a += k[0];
+      a += ((uint32_t)k[1])<<8;
+      a += ((uint32_t)k[2])<<16;
+      a += ((uint32_t)k[3])<<24;
+      b += k[4];
+      b += ((uint32_t)k[5])<<8;
+      b += ((uint32_t)k[6])<<16;
+      b += ((uint32_t)k[7])<<24;
+      c += k[8];
+      c += ((uint32_t)k[9])<<8;
+      c += ((uint32_t)k[10])<<16;
+      c += ((uint32_t)k[11])<<24;
+      mix(a,b,c);
+      length -= 12;
+      k += 12;
+    }
+
+    /*-------------------------------- last block: affect all 32 bits of (c) */
+    switch(length)
+    {
+    case 12: c+=((uint32_t)k[11])<<24; FALL_THROUGH;
+    case 11: c+=((uint32_t)k[10])<<16; FALL_THROUGH;
+    case 10: c+=((uint32_t)k[9])<<8;   FALL_THROUGH;
+    case 9 : c+=k[8];                  FALL_THROUGH;
+    case 8 : b+=((uint32_t)k[7])<<24;  FALL_THROUGH;
+    case 7 : b+=((uint32_t)k[6])<<16;  FALL_THROUGH;
+    case 6 : b+=((uint32_t)k[5])<<8;   FALL_THROUGH;
+    case 5 : b+=k[4];                  FALL_THROUGH;
+    case 4 : a+=((uint32_t)k[3])<<24;  FALL_THROUGH;
+    case 3 : a+=((uint32_t)k[2])<<16;  FALL_THROUGH;
+    case 2 : a+=((uint32_t)k[1])<<8;   FALL_THROUGH;
+    case 1 : a+=k[0];
+             break;
+    case 0 : return c;
+    }
+  }
+
+  final(a,b,c);
+  return c;
+}
+
+_PUBLIC_ unsigned int tdb_jenkins_hash(TDB_DATA *key)
+{
+	return hashlittle(key->dptr, key->dsize);
+}
diff --git a/lib/tdb/common/io.c b/lib/tdb/common/io.c
new file mode 100644
index 0000000..0de0dab
--- /dev/null
+++ b/lib/tdb/common/io.c
@@ -0,0 +1,806 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+
+#include "tdb_private.h"
+
+/*
+ * We prepend the mutex area, so fixup offsets. See mutex.c for details.
+ * tdb->hdr_ofs is 0 or header.mutex_size.
+ *
+ * Note: that we only have the 4GB limit of tdb_off_t for
+ * tdb->map_size. The file size on disk can be 4GB + tdb->hdr_ofs!
+ */
+
+static bool tdb_adjust_offset(struct tdb_context *tdb, off_t *off)
+{
+	off_t tmp = tdb->hdr_ofs + *off;
+
+	if ((tmp < tdb->hdr_ofs) || (tmp < *off)) {
+		errno = EIO;
+		return false;
+	}
+
+	*off = tmp;
+	return true;
+}
+
+static ssize_t tdb_pwrite(struct tdb_context *tdb, const void *buf,
+			  size_t count, off_t offset)
+{
+	ssize_t ret;
+
+	if (!tdb_adjust_offset(tdb, &offset)) {
+		return -1;
+	}
+
+	do {
+		ret = pwrite(tdb->fd, buf, count, offset);
+	} while ((ret == -1) && (errno == EINTR));
+
+	return ret;
+}
+
+static ssize_t tdb_pread(struct tdb_context *tdb, void *buf,
+			 size_t count, off_t offset)
+{
+	ssize_t ret;
+
+	if (!tdb_adjust_offset(tdb, &offset)) {
+		return -1;
+	}
+
+	do {
+		ret = pread(tdb->fd, buf, count, offset);
+	} while ((ret == -1) && (errno == EINTR));
+
+	return ret;
+}
+
+static int tdb_ftruncate(struct tdb_context *tdb, off_t length)
+{
+	ssize_t ret;
+
+	if (!tdb_adjust_offset(tdb, &length)) {
+		return -1;
+	}
+
+	do {
+		ret = ftruncate(tdb->fd, length);
+	} while ((ret == -1) && (errno == EINTR));
+
+	return ret;
+}
+
+#ifdef HAVE_POSIX_FALLOCATE
+static int tdb_posix_fallocate(struct tdb_context *tdb, off_t offset,
+			       off_t len)
+{
+	ssize_t ret;
+
+	if (!tdb_adjust_offset(tdb, &offset)) {
+		return -1;
+	}
+
+	do {
+		ret = posix_fallocate(tdb->fd, offset, len);
+	} while ((ret == -1) && (errno == EINTR));
+
+	return ret;
+}
+#endif
+
+static int tdb_fstat(struct tdb_context *tdb, struct stat *buf)
+{
+	int ret;
+
+	ret = fstat(tdb->fd, buf);
+	if (ret == -1) {
+		return -1;
+	}
+
+	if (buf->st_size < tdb->hdr_ofs) {
+		errno = EIO;
+		return -1;
+	}
+	buf->st_size -= tdb->hdr_ofs;
+
+	return ret;
+}
+
+/* check for an out of bounds access - if it is out of bounds then
+   see if the database has been expanded by someone else and expand
+   if necessary
+*/
+static int tdb_notrans_oob(
+	struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, int probe)
+{
+	struct stat st;
+	if (len + off < len) {
+		if (!probe) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob off %u len %u wrap\n",
+				 off, len));
+		}
+		return -1;
+	}
+
+	/*
+	 * This duplicates functionality from tdb_oob(). Don't remove:
+	 * we still have direct callers of tdb->methods->tdb_oob()
+	 * inside transaction.c.
+	 */
+	if (off + len <= tdb->map_size)
+		return 0;
+	if (tdb->flags & TDB_INTERNAL) {
+		if (!probe) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %u beyond internal malloc size %u\n",
+				 (int)(off + len), (int)tdb->map_size));
+		}
+		return -1;
+	}
+
+	if (tdb_fstat(tdb, &st) == -1) {
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* Beware >4G files! */
+	if ((tdb_off_t)st.st_size != st.st_size) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_IO;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_oob len %llu too large!\n",
+			 (long long)st.st_size));
+		return -1;
+	}
+
+	/* Unmap, update size, remap.  We do this unconditionally, to handle
+	 * the unusual case where the db is truncated.
+	 *
+	 * This can happen to a child using tdb_reopen_all(true) on a
+	 * TDB_CLEAR_IF_FIRST tdb whose parent crashes: the next
+	 * opener will truncate the database. */
+	if (tdb_munmap(tdb) == -1) {
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+	tdb->map_size = st.st_size;
+	if (tdb_mmap(tdb) != 0) {
+		return -1;
+	}
+
+	if (st.st_size < (size_t)off + len) {
+		if (!probe) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %u beyond eof at %u\n",
+				 (int)(off + len), (int)st.st_size));
+		}
+		return -1;
+	}
+	return 0;
+}
+
+/* write a lump of data at a specified offset */
+static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
+		     const void *buf, tdb_len_t len)
+{
+	if (len == 0) {
+		return 0;
+	}
+
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
+	if (tdb_oob(tdb, off, len, 0) != 0)
+		return -1;
+
+	if (tdb->map_ptr) {
+		memcpy(off + (char *)tdb->map_ptr, buf, len);
+	} else {
+#ifdef HAVE_INCOHERENT_MMAP
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+#else
+		ssize_t written;
+
+		written = tdb_pwrite(tdb, buf, len, off);
+
+		if ((written != (ssize_t)len) && (written != -1)) {
+			/* try once more */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: wrote only "
+				 "%zi of %u bytes at %u, trying once more\n",
+				 written, len, off));
+			written = tdb_pwrite(tdb, (const char *)buf+written,
+					     len-written, off+written);
+		}
+		if (written == -1) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_write failed at %u "
+				 "len=%u (%s)\n", off, len, strerror(errno)));
+			return -1;
+		} else if (written != (ssize_t)len) {
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: failed to "
+				 "write %u bytes at %u in two attempts\n",
+				 len, off));
+			return -1;
+		}
+#endif
+	}
+	return 0;
+}
+
+/* Endian conversion: we only ever deal with 4 byte quantities */
+void *tdb_convert(void *buf, uint32_t size)
+{
+	uint32_t i, *p = (uint32_t *)buf;
+	for (i = 0; i < size / 4; i++)
+		p[i] = TDB_BYTEREV(p[i]);
+	return buf;
+}
+
+
+/* read a lump of data at a specified offset, maybe convert */
+static int tdb_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
+		    tdb_len_t len, int cv)
+{
+	if (tdb_oob(tdb, off, len, 0) != 0) {
+		return -1;
+	}
+
+	if (tdb->map_ptr) {
+		memcpy(buf, off + (char *)tdb->map_ptr, len);
+	} else {
+#ifdef HAVE_INCOHERENT_MMAP
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+#else
+		ssize_t ret;
+
+		ret = tdb_pread(tdb, buf, len, off);
+		if (ret != (ssize_t)len) {
+			/* Ensure ecode is set for log fn. */
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_read failed at %u "
+				 "len=%u ret=%zi (%s) map_size=%u\n",
+				 off, len, ret, strerror(errno),
+				 tdb->map_size));
+			return -1;
+		}
+#endif
+	}
+	if (cv) {
+		tdb_convert(buf, len);
+	}
+	return 0;
+}
+
+
+
+/*
+  do an unlocked scan of the hash table heads to find the next non-zero head. The value
+  will then be confirmed with the lock held
+*/
+static void tdb_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
+{
+	uint32_t h = *chain;
+	if (tdb->map_ptr) {
+		for (;h < tdb->hash_size;h++) {
+			if (0 != *(uint32_t *)(TDB_HASH_TOP(h) + (unsigned char *)tdb->map_ptr)) {
+				break;
+			}
+		}
+	} else {
+		uint32_t off=0;
+		for (;h < tdb->hash_size;h++) {
+			if (tdb_ofs_read(tdb, TDB_HASH_TOP(h), &off) != 0 || off != 0) {
+				break;
+			}
+		}
+	}
+	(*chain) = h;
+}
+
+
+int tdb_munmap(struct tdb_context *tdb)
+{
+	if (tdb->flags & TDB_INTERNAL)
+		return 0;
+
+#ifdef HAVE_MMAP
+	if (tdb->map_ptr) {
+		int ret;
+
+		ret = munmap(tdb->map_ptr, tdb->map_size);
+		if (ret != 0)
+			return ret;
+	}
+#endif
+	tdb->map_ptr = NULL;
+	return 0;
+}
+
+/* If mmap isn't coherent, *everyone* must always mmap. */
+static bool should_mmap(const struct tdb_context *tdb)
+{
+#ifdef HAVE_INCOHERENT_MMAP
+	return true;
+#else
+	return !(tdb->flags & TDB_NOMMAP);
+#endif
+}
+
+int tdb_mmap(struct tdb_context *tdb)
+{
+	if (tdb->flags & TDB_INTERNAL)
+		return 0;
+
+#ifdef HAVE_MMAP
+	if (should_mmap(tdb)) {
+		tdb->map_ptr = mmap(NULL, tdb->map_size,
+				    PROT_READ|(tdb->read_only? 0:PROT_WRITE),
+				    MAP_SHARED|MAP_FILE, tdb->fd,
+				    tdb->hdr_ofs);
+
+		/*
+		 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
+		 */
+
+		if (tdb->map_ptr == MAP_FAILED) {
+			tdb->map_ptr = NULL;
+			TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_mmap failed for size %u (%s)\n",
+				 tdb->map_size, strerror(errno)));
+#ifdef HAVE_INCOHERENT_MMAP
+			tdb->ecode = TDB_ERR_IO;
+			return -1;
+#endif
+		}
+	} else {
+		tdb->map_ptr = NULL;
+	}
+#else
+	tdb->map_ptr = NULL;
+#endif
+	return 0;
+}
+
+/* expand a file.  we prefer to use ftruncate, as that is what posix
+  says to use for mmap expansion */
+static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t addition)
+{
+	char buf[8192];
+	tdb_off_t new_size;
+	int ret;
+
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
+	if (!tdb_add_off_t(size, addition, &new_size)) {
+		tdb->ecode = TDB_ERR_OOM;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write "
+			"overflow detected current size[%u] addition[%u]!\n",
+			(unsigned)size, (unsigned)addition));
+		errno = ENOSPC;
+		return -1;
+	}
+
+#ifdef HAVE_POSIX_FALLOCATE
+	ret = tdb_posix_fallocate(tdb, size, addition);
+	if (ret == 0) {
+		return 0;
+	}
+	if (ret == ENOSPC) {
+		/*
+		 * The Linux glibc (at least as of 2.24) fallback if
+		 * the file system does not support fallocate does not
+		 * reset the file size back to where it was. Also, to
+		 * me it is unclear from the posix spec of
+		 * posix_fallocate whether this is allowed or
+		 * not. Better be safe than sorry and "goto fail" but
+		 * "return -1" here, leaving the EOF pointer too
+		 * large.
+		 */
+		goto fail;
+	}
+
+	/*
+	 * Retry the "old" way. Possibly unnecessary, but looking at
+	 * our configure script there seem to be weird failure modes
+	 * for posix_fallocate. See commit 3264a98ff16de, which
+	 * probably refers to
+	 * https://sourceware.org/bugzilla/show_bug.cgi?id=1083.
+	 */
+#endif
+
+	ret = tdb_ftruncate(tdb, new_size);
+	if (ret == -1) {
+		char b = 0;
+		ssize_t written = tdb_pwrite(tdb, &b, 1, new_size - 1);
+		if (written == 0) {
+			/* try once more, potentially revealing errno */
+			written = tdb_pwrite(tdb, &b, 1, new_size - 1);
+		}
+		if (written == 0) {
+			/* again - give up, guessing errno */
+			errno = ENOSPC;
+		}
+		if (written != 1) {
+			tdb->ecode = TDB_ERR_OOM;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file to %u failed (%s)\n",
+				 (unsigned)new_size, strerror(errno)));
+			return -1;
+		}
+	}
+
+	/* now fill the file with something. This ensures that the
+	   file isn't sparse, which would be very bad if we ran out of
+	   disk. This must be done with write, not via mmap */
+	memset(buf, TDB_PAD_BYTE, sizeof(buf));
+	while (addition) {
+		size_t n = addition>sizeof(buf)?sizeof(buf):addition;
+		ssize_t written = tdb_pwrite(tdb, buf, n, size);
+		if (written == 0) {
+			/* prevent infinite loops: try _once_ more */
+			written = tdb_pwrite(tdb, buf, n, size);
+		}
+		if (written == 0) {
+			/* give up, trying to provide a useful errno */
+			tdb->ecode = TDB_ERR_OOM;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write "
+				"returned 0 twice: giving up!\n"));
+			errno = ENOSPC;
+			goto fail;
+		}
+		if (written == -1) {
+			tdb->ecode = TDB_ERR_OOM;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write of "
+				 "%u bytes failed (%s)\n", (int)n,
+				 strerror(errno)));
+			goto fail;
+		}
+		if (written != n) {
+			TDB_LOG((tdb, TDB_DEBUG_WARNING, "expand_file: wrote "
+				 "only %zu of %zi bytes - retrying\n", written,
+				 n));
+		}
+		addition -= written;
+		size += written;
+	}
+	return 0;
+
+fail:
+	{
+		int err = errno;
+
+		/*
+		 * We're holding the freelist lock or are inside a
+		 * transaction. Cutting the file is safe, the space we
+		 * tried to allocate can't have been used anywhere in
+		 * the meantime.
+		 */
+
+		ret = tdb_ftruncate(tdb, size);
+		if (ret == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_WARNING, "expand_file: "
+				 "retruncate to %ju failed\n",
+				 (uintmax_t)size));
+		}
+		errno = err;
+	}
+
+	return -1;
+}
+
+
+/* You need 'size', this tells you how much you should expand by. */
+tdb_off_t tdb_expand_adjust(tdb_off_t map_size, tdb_off_t size, int page_size)
+{
+	tdb_off_t new_size, top_size, increment;
+	tdb_off_t max_size = UINT32_MAX - map_size;
+
+	if (size > max_size) {
+		/*
+		 * We can't round up anymore, just give back
+		 * what we're asked for.
+		 *
+		 * The caller has to take care of the ENOSPC handling.
+		 */
+		return size;
+	}
+
+	/* limit size in order to avoid using up huge amounts of memory for
+	 * in memory tdbs if an oddball huge record creeps in */
+	if (size > 100 * 1024) {
+		increment = size * 2;
+	} else {
+		increment = size * 100;
+	}
+	if (increment < size) {
+		goto overflow;
+	}
+
+	if (!tdb_add_off_t(map_size, increment, &top_size)) {
+		goto overflow;
+	}
+
+	/* always make room for at least top_size more records, and at
+	   least 25% more space. if the DB is smaller than 100MiB,
+	   otherwise grow it by 10% only. */
+	if (map_size > 100 * 1024 * 1024) {
+		new_size = map_size * 1.10;
+	} else {
+		new_size = map_size * 1.25;
+	}
+	if (new_size < map_size) {
+		goto overflow;
+	}
+
+	/* Round the database up to a multiple of the page size */
+	new_size = MAX(top_size, new_size);
+
+	if (new_size + page_size < new_size) {
+		/* There's a "+" in TDB_ALIGN that might overflow... */
+		goto overflow;
+	}
+
+	return TDB_ALIGN(new_size, page_size) - map_size;
+
+overflow:
+	/*
+	 * Somewhere in between we went over 4GB. Make one big jump to
+	 * exactly 4GB database size.
+	 */
+	return max_size;
+}
+
+/* expand the database at least size bytes by expanding the underlying
+   file and doing the mmap again if necessary */
+int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
+{
+	struct tdb_record rec;
+	tdb_off_t offset;
+	tdb_off_t new_size;
+
+	if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
+		return -1;
+	}
+
+	/* must know about any previous expansions by another process */
+	tdb_oob(tdb, tdb->map_size, 1, 1);
+
+	/*
+	 * Note: that we don't care about tdb->hdr_ofs != 0 here
+	 *
+	 * The 4GB limitation is just related to tdb->map_size
+	 * and the offset calculation in the records.
+	 *
+	 * The file on disk can be up to 4GB + tdb->hdr_ofs
+	 */
+	size = tdb_expand_adjust(tdb->map_size, size, tdb->page_size);
+
+	if (!tdb_add_off_t(tdb->map_size, size, &new_size)) {
+		tdb->ecode = TDB_ERR_OOM;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_expand "
+			"overflow detected current map_size[%u] size[%u]!\n",
+			(unsigned)tdb->map_size, (unsigned)size));
+		goto fail;
+	}
+
+	/* form a new freelist record */
+	offset = tdb->map_size;
+	memset(&rec,'\0',sizeof(rec));
+	rec.rec_len = size - sizeof(rec);
+
+	if (tdb->flags & TDB_INTERNAL) {
+		char *new_map_ptr;
+
+		new_map_ptr = (char *)realloc(tdb->map_ptr, new_size);
+		if (!new_map_ptr) {
+			tdb->ecode = TDB_ERR_OOM;
+			goto fail;
+		}
+		tdb->map_ptr = new_map_ptr;
+		tdb->map_size = new_size;
+	} else {
+		int ret;
+
+		/*
+		 * expand the file itself
+		 */
+		ret = tdb->methods->tdb_expand_file(tdb, tdb->map_size, size);
+		if (ret != 0) {
+			goto fail;
+		}
+
+		/* Explicitly remap: if we're in a transaction, this won't
+		 * happen automatically! */
+		tdb_munmap(tdb);
+		tdb->map_size = new_size;
+		if (tdb_mmap(tdb) != 0) {
+			goto fail;
+		}
+	}
+
+	/* link it into the free list */
+	if (tdb_free(tdb, offset, &rec) == -1)
+		goto fail;
+
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return 0;
+ fail:
+	tdb_unlock(tdb, -1, F_WRLCK);
+	return -1;
+}
+
+int _tdb_oob(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, int probe)
+{
+	int ret = tdb->methods->tdb_oob(tdb, off, len, probe);
+	return ret;
+}
+
+/* read/write a tdb_off_t */
+int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+{
+	return tdb->methods->tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
+}
+
+int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
+{
+	tdb_off_t off = *d;
+	return tdb->methods->tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
+}
+
+
+/* read a lump of data, allocating the space for it */
+unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
+{
+	unsigned char *buf;
+
+	/* some systems don't like zero length malloc */
+
+	if (!(buf = (unsigned char *)malloc(len ? len : 1))) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_OOM;
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_alloc_read malloc failed len=%u (%s)\n",
+			   len, strerror(errno)));
+		return NULL;
+	}
+	if (tdb->methods->tdb_read(tdb, offset, buf, len, 0) == -1) {
+		SAFE_FREE(buf);
+		return NULL;
+	}
+	return buf;
+}
+
+/* Give a piece of tdb data to a parser */
+
+int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
+		   tdb_off_t offset, tdb_len_t len,
+		   int (*parser)(TDB_DATA key, TDB_DATA data,
+				 void *private_data),
+		   void *private_data)
+{
+	TDB_DATA data;
+	int result;
+
+	data.dsize = len;
+
+	if ((tdb->transaction == NULL) && (tdb->map_ptr != NULL)) {
+		/*
+		 * Optimize by avoiding the malloc/memcpy/free, point the
+		 * parser directly at the mmap area.
+		 */
+		if (tdb_oob(tdb, offset, len, 0) != 0) {
+			return -1;
+		}
+		data.dptr = offset + (unsigned char *)tdb->map_ptr;
+		return parser(key, data, private_data);
+	}
+
+	if (!(data.dptr = tdb_alloc_read(tdb, offset, len))) {
+		return -1;
+	}
+
+	result = parser(key, data, private_data);
+	free(data.dptr);
+	return result;
+}
+
+/* read/write a record */
+int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+{
+	int ret;
+	tdb_len_t overall_len;
+
+	if (tdb->methods->tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
+		return -1;
+	if (TDB_BAD_MAGIC(rec)) {
+		/* Ensure ecode is set for log fn. */
+		tdb->ecode = TDB_ERR_CORRUPT;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_rec_read bad magic 0x%x at offset=%u\n", rec->magic, offset));
+		return -1;
+	}
+
+	overall_len = rec->key_len + rec->data_len;
+	if (overall_len < rec->data_len) {
+		/* overflow */
+		return -1;
+	}
+
+	if (overall_len > rec->rec_len) {
+		/* invalid record */
+		return -1;
+	}
+
+	ret = tdb_oob(tdb, offset, rec->key_len, 1);
+	if (ret == -1) {
+		return -1;
+	}
+	ret = tdb_oob(tdb, offset, rec->data_len, 1);
+	if (ret == -1) {
+		return -1;
+	}
+	ret = tdb_oob(tdb, offset, rec->rec_len, 1);
+	if (ret == -1) {
+		return -1;
+	}
+
+	return tdb_oob(tdb, rec->next, sizeof(*rec), 0);
+}
+
+int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
+{
+	struct tdb_record r = *rec;
+	return tdb->methods->tdb_write(tdb, offset, CONVERT(r), sizeof(r));
+}
+
+static const struct tdb_methods io_methods = {
+	tdb_read,
+	tdb_write,
+	tdb_next_hash_chain,
+	tdb_notrans_oob,
+	tdb_expand_file,
+};
+
+/*
+  initialise the default methods table
+*/
+void tdb_io_init(struct tdb_context *tdb)
+{
+	tdb->methods = &io_methods;
+}
diff --git a/lib/tdb/common/lock.c b/lib/tdb/common/lock.c
new file mode 100644
index 0000000..c4e276f
--- /dev/null
+++ b/lib/tdb/common/lock.c
@@ -0,0 +1,1033 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+_PUBLIC_ void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
+{
+	tdb->interrupt_sig_ptr = ptr;
+}
+
+static int fcntl_lock(struct tdb_context *tdb,
+		      int rw, off_t off, off_t len, bool waitflag)
+{
+	struct flock fl;
+	int cmd;
+
+#ifdef USE_TDB_MUTEX_LOCKING
+	{
+		int ret;
+		if (tdb_mutex_lock(tdb, rw, off, len, waitflag, &ret)) {
+			return ret;
+		}
+	}
+#endif
+
+	fl.l_type = rw;
+	fl.l_whence = SEEK_SET;
+	fl.l_start = off;
+	fl.l_len = len;
+	fl.l_pid = 0;
+
+	cmd = waitflag ? F_SETLKW : F_SETLK;
+
+	return fcntl(tdb->fd, cmd, &fl);
+}
+
+static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
+{
+	struct flock fl;
+#if 0 /* Check they matched up locks and unlocks correctly. */
+	char line[80];
+	FILE *locks;
+	bool found = false;
+
+	locks = fopen("/proc/locks", "r");
+
+	while (fgets(line, 80, locks)) {
+		char *p;
+		int type, start, l;
+
+		/* eg. 1: FLOCK  ADVISORY  WRITE 2440 08:01:2180826 0 EOF */
+		p = strchr(line, ':') + 1;
+		if (strncmp(p, " POSIX  ADVISORY  ", strlen(" POSIX  ADVISORY  ")))
+			continue;
+		p += strlen(" FLOCK  ADVISORY  ");
+		if (strncmp(p, "READ  ", strlen("READ  ")) == 0)
+			type = F_RDLCK;
+		else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
+			type = F_WRLCK;
+		else
+			abort();
+		p += 6;
+		if (atoi(p) != getpid())
+			continue;
+		p = strchr(strchr(p, ' ') + 1, ' ') + 1;
+		start = atoi(p);
+		p = strchr(p, ' ') + 1;
+		if (strncmp(p, "EOF", 3) == 0)
+			l = 0;
+		else
+			l = atoi(p) - start + 1;
+
+		if (off == start) {
+			if (len != l) {
+				fprintf(stderr, "Len %u should be %u: %s",
+					(int)len, l, line);
+				abort();
+			}
+			if (type != rw) {
+				fprintf(stderr, "Type %s wrong: %s",
+					rw == F_RDLCK ? "READ" : "WRITE", line);
+				abort();
+			}
+			found = true;
+			break;
+		}
+	}
+
+	if (!found) {
+		fprintf(stderr, "Unlock on %u@%u not found!\n",
+			(int)off, (int)len);
+		abort();
+	}
+
+	fclose(locks);
+#endif
+
+#ifdef USE_TDB_MUTEX_LOCKING
+	{
+		int ret;
+		if (tdb_mutex_unlock(tdb, rw, off, len, &ret)) {
+			return ret;
+		}
+	}
+#endif
+
+	fl.l_type = F_UNLCK;
+	fl.l_whence = SEEK_SET;
+	fl.l_start = off;
+	fl.l_len = len;
+	fl.l_pid = 0;
+
+	return fcntl(tdb->fd, F_SETLKW, &fl);
+}
+
+/*
+ * Calculate the lock offset for a list
+ *
+ * list -1 is the freelist, otherwise a hash chain.
+ *
+ * Note that we consistently (but without real reason) lock hash chains at an
+ * offset that is 4 bytes below the real offset of the corresponding list head
+ * in the db.
+ *
+ * This is the memory layout of the hashchain array:
+ *
+ * FREELIST_TOP + 0 = freelist
+ * FREELIST_TOP + 4 = hashtable list 0
+ * FREELIST_TOP + 8 = hashtable list 1
+ * ...
+ *
+ * Otoh lock_offset computes:
+ *
+ * freelist = FREELIST_TOP - 4
+ * list 0   = FREELIST_TOP + 0
+ * list 1   = FREELIST_TOP + 4
+ * ...
+ *
+ * Unfortunately we can't change this calculation in order to align the locking
+ * offset with the memory layout, as that would make the locking incompatible
+ * between different tdb versions.
+ */
+static tdb_off_t lock_offset(int list)
+{
+	return FREELIST_TOP + 4*list;
+}
+
+/* a byte range locking function - return 0 on success
+   this functions locks/unlocks "len" byte at the specified offset.
+
+   On error, errno is also set so that errors are passed back properly
+   through tdb_open().
+
+   note that a len of zero means lock to end of file
+*/
+int tdb_brlock(struct tdb_context *tdb,
+	       int rw_type, tdb_off_t offset, size_t len,
+	       enum tdb_lock_flags flags)
+{
+	int ret;
+
+	if (tdb->flags & TDB_NOLOCK) {
+		return 0;
+	}
+
+	if (flags & TDB_LOCK_MARK_ONLY) {
+		return 0;
+	}
+
+	if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
+	do {
+		ret = fcntl_lock(tdb, rw_type, offset, len,
+				 flags & TDB_LOCK_WAIT);
+		/* Check for a sigalarm break. */
+		if (ret == -1 && errno == EINTR &&
+				tdb->interrupt_sig_ptr &&
+				*tdb->interrupt_sig_ptr) {
+			break;
+		}
+	} while (ret == -1 && errno == EINTR);
+
+	if (ret == -1) {
+		tdb->ecode = TDB_ERR_LOCK;
+		/* Generic lock error. errno set by fcntl.
+		 * EAGAIN is an expected return from non-blocking
+		 * locks. */
+		if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
+			TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %u rw_type=%d flags=%d len=%zu\n",
+				 tdb->fd, offset, rw_type, flags, len));
+		}
+		return -1;
+	}
+	return 0;
+}
+
+int tdb_brunlock(struct tdb_context *tdb,
+		 int rw_type, tdb_off_t offset, size_t len)
+{
+	int ret;
+
+	if (tdb->flags & TDB_NOLOCK) {
+		return 0;
+	}
+
+	do {
+		ret = fcntl_unlock(tdb, rw_type, offset, len);
+	} while (ret == -1 && errno == EINTR);
+
+	if (ret == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brunlock failed (fd=%d) at offset %u rw_type=%u len=%zu\n",
+			 tdb->fd, offset, rw_type, len));
+	}
+	return ret;
+}
+
+/*
+ * Do a tdb_brlock in a loop. Some OSes (such as solaris) have too
+ * conservative deadlock detection and claim a deadlock when progress can be
+ * made. For those OSes we may loop for a while.
+ */
+
+static int tdb_brlock_retry(struct tdb_context *tdb,
+			    int rw_type, tdb_off_t offset, size_t len,
+			    enum tdb_lock_flags flags)
+{
+	int count = 1000;
+
+	while (count--) {
+		struct timeval tv;
+		int ret;
+
+		ret = tdb_brlock(tdb, rw_type, offset, len, flags);
+		if (ret == 0) {
+			return 0;
+		}
+		if (errno != EDEADLK) {
+			break;
+		}
+		/* sleep for as short a time as we can - more portable than usleep() */
+		tv.tv_sec = 0;
+		tv.tv_usec = 1;
+		select(0, NULL, NULL, NULL, &tv);
+	}
+	return -1;
+}
+
+/*
+  upgrade a read lock to a write lock.
+*/
+int tdb_allrecord_upgrade(struct tdb_context *tdb)
+{
+	int ret;
+
+	if (tdb->allrecord_lock.count != 1) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "tdb_allrecord_upgrade failed: count %u too high\n",
+			 tdb->allrecord_lock.count));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb->allrecord_lock.off != 1) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "tdb_allrecord_upgrade failed: already upgraded?\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb_have_mutexes(tdb)) {
+		ret = tdb_mutex_allrecord_upgrade(tdb);
+		if (ret == -1) {
+			goto fail;
+		}
+		ret = tdb_brlock_retry(tdb, F_WRLCK, lock_offset(tdb->hash_size),
+				       0, TDB_LOCK_WAIT|TDB_LOCK_PROBE);
+		if (ret == -1) {
+			tdb_mutex_allrecord_downgrade(tdb);
+		}
+	} else {
+		ret = tdb_brlock_retry(tdb, F_WRLCK, FREELIST_TOP, 0,
+				       TDB_LOCK_WAIT|TDB_LOCK_PROBE);
+	}
+
+	if (ret == 0) {
+		tdb->allrecord_lock.ltype = F_WRLCK;
+		tdb->allrecord_lock.off = 0;
+		return 0;
+	}
+fail:
+	TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_allrecord_upgrade failed\n"));
+	return -1;
+}
+
+static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
+					   tdb_off_t offset)
+{
+	int i;
+
+	for (i=0; i<tdb->num_lockrecs; i++) {
+		if (tdb->lockrecs[i].off == offset) {
+			return &tdb->lockrecs[i];
+		}
+	}
+	return NULL;
+}
+
+/* lock an offset in the database. */
+int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
+		  enum tdb_lock_flags flags)
+{
+	struct tdb_lock_type *new_lck;
+
+	if (offset >= lock_offset(tdb->hash_size)) {
+		tdb->ecode = TDB_ERR_LOCK;
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid offset %u for ltype=%d\n",
+			 offset, ltype));
+		return -1;
+	}
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
+
+	new_lck = find_nestlock(tdb, offset);
+	if (new_lck) {
+		if ((new_lck->ltype == F_RDLCK) && (ltype == F_WRLCK)) {
+			if (!tdb_have_mutexes(tdb)) {
+				int ret;
+				/*
+				 * Upgrade the underlying fcntl
+				 * lock. Mutexes don't do readlocks,
+				 * so this only applies to fcntl
+				 * locking.
+				 */
+				ret = tdb_brlock(tdb, ltype, offset, 1, flags);
+				if (ret != 0) {
+					return ret;
+				}
+			}
+			new_lck->ltype = F_WRLCK;
+		}
+		/*
+		 * Just increment the in-memory struct, posix locks
+		 * don't stack.
+		 */
+		new_lck->count++;
+		return 0;
+	}
+
+	if (tdb->num_lockrecs == tdb->lockrecs_array_length) {
+		new_lck = (struct tdb_lock_type *)realloc(
+			tdb->lockrecs,
+			sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
+		if (new_lck == NULL) {
+			errno = ENOMEM;
+			return -1;
+		}
+		tdb->lockrecs_array_length = tdb->num_lockrecs+1;
+		tdb->lockrecs = new_lck;
+	}
+
+	/* Since fcntl locks don't nest, we do a lock for the first one,
+	   and simply bump the count for future ones */
+	if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
+		return -1;
+	}
+
+	new_lck = &tdb->lockrecs[tdb->num_lockrecs];
+
+	new_lck->off = offset;
+	new_lck->count = 1;
+	new_lck->ltype = ltype;
+	tdb->num_lockrecs++;
+
+	return 0;
+}
+
+static int tdb_lock_and_recover(struct tdb_context *tdb)
+{
+	int ret;
+
+	/* We need to match locking order in transaction commit. */
+	if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT)) {
+		return -1;
+	}
+
+	if (tdb_brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT)) {
+		tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
+		return -1;
+	}
+
+	ret = tdb_transaction_recover(tdb);
+
+	tdb_brunlock(tdb, F_WRLCK, OPEN_LOCK, 1);
+	tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
+
+	return ret;
+}
+
+static bool have_data_locks(const struct tdb_context *tdb)
+{
+	int i;
+
+	for (i = 0; i < tdb->num_lockrecs; i++) {
+		if (tdb->lockrecs[i].off >= lock_offset(-1))
+			return true;
+	}
+	return false;
+}
+
+/*
+ * A allrecord lock allows us to avoid per chain locks. Check if the allrecord
+ * lock is strong enough.
+ */
+static int tdb_lock_covered_by_allrecord_lock(struct tdb_context *tdb,
+					      int ltype)
+{
+	if (ltype == F_RDLCK) {
+		/*
+		 * The allrecord_lock is equal (F_RDLCK) or stronger
+		 * (F_WRLCK). Pass.
+		 */
+		return 0;
+	}
+
+	if (tdb->allrecord_lock.ltype == F_RDLCK) {
+		/*
+		 * We ask for ltype==F_WRLCK, but the allrecord_lock
+		 * is too weak. We can't upgrade here, so fail.
+		 */
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	/*
+	 * Asking for F_WRLCK, allrecord is F_WRLCK as well. Pass.
+	 */
+	return 0;
+}
+
+static int tdb_lock_list(struct tdb_context *tdb, int list, int ltype,
+			 enum tdb_lock_flags waitflag)
+{
+	int ret;
+	bool check = false;
+
+	if (tdb->allrecord_lock.count) {
+		return tdb_lock_covered_by_allrecord_lock(tdb, ltype);
+	}
+
+	/*
+	 * Check for recoveries: Someone might have kill -9'ed a process
+	 * during a commit.
+	 */
+	check = !have_data_locks(tdb);
+	ret = tdb_nest_lock(tdb, lock_offset(list), ltype, waitflag);
+
+	if (ret == 0 && check && tdb_needs_recovery(tdb)) {
+		tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
+
+		if (tdb_lock_and_recover(tdb) == -1) {
+			return -1;
+		}
+		return tdb_lock_list(tdb, list, ltype, waitflag);
+	}
+	return ret;
+}
+
+/* lock a list in the database. list -1 is the alloc list */
+int tdb_lock(struct tdb_context *tdb, int list, int ltype)
+{
+	int ret;
+
+	ret = tdb_lock_list(tdb, list, ltype, TDB_LOCK_WAIT);
+	if (ret) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
+			 "ltype=%d (%s)\n",  list, ltype, strerror(errno)));
+	}
+	return ret;
+}
+
+/* lock a list in the database. list -1 is the alloc list. non-blocking lock */
+_PUBLIC_ int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
+_PUBLIC_ int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
+{
+	return tdb_lock_list(tdb, list, ltype, TDB_LOCK_NOWAIT);
+}
+
+
+int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
+		    bool mark_lock)
+{
+	int ret = -1;
+	struct tdb_lock_type *lck;
+
+	if (tdb->flags & TDB_NOLOCK)
+		return 0;
+
+	/* Sanity checks */
+	if (offset >= lock_offset(tdb->hash_size)) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: offset %u invalid (%d)\n", offset, tdb->hash_size));
+		return ret;
+	}
+
+	lck = find_nestlock(tdb, offset);
+	if ((lck == NULL) || (lck->count == 0)) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
+		return -1;
+	}
+
+	if (lck->count > 1) {
+		lck->count--;
+		return 0;
+	}
+
+	/*
+	 * This lock has count==1 left, so we need to unlock it in the
+	 * kernel. We don't bother with decrementing the in-memory array
+	 * element, we're about to overwrite it with the last array element
+	 * anyway.
+	 */
+
+	if (mark_lock) {
+		ret = 0;
+	} else {
+		ret = tdb_brunlock(tdb, ltype, offset, 1);
+	}
+
+	/*
+	 * Shrink the array by overwriting the element just unlocked with the
+	 * last array element.
+	 */
+	*lck = tdb->lockrecs[--tdb->num_lockrecs];
+
+	/*
+	 * We don't bother with realloc when the array shrinks, but if we have
+	 * a completely idle tdb we should get rid of the locked array.
+	 */
+
+	if (ret)
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
+	return ret;
+}
+
+_PUBLIC_ int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
+_PUBLIC_ int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
+{
+	/* a global lock allows us to avoid per chain locks */
+	if (tdb->allrecord_lock.count) {
+		return tdb_lock_covered_by_allrecord_lock(tdb, ltype);
+	}
+
+	return tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
+}
+
+/*
+  get the transaction lock
+ */
+int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
+			 enum tdb_lock_flags lockflags)
+{
+	return tdb_nest_lock(tdb, TRANSACTION_LOCK, ltype, lockflags);
+}
+
+/*
+  release the transaction lock
+ */
+int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
+{
+	return tdb_nest_unlock(tdb, TRANSACTION_LOCK, ltype, false);
+}
+
+/* Returns 0 if all done, -1 if error, 1 if ok. */
+static int tdb_allrecord_check(struct tdb_context *tdb, int ltype,
+			       enum tdb_lock_flags flags, bool upgradable)
+{
+	/* There are no locks on read-only dbs */
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb->allrecord_lock.count &&
+	    tdb->allrecord_lock.ltype == (uint32_t)ltype) {
+		tdb->allrecord_lock.count++;
+		return 0;
+	}
+
+	if (tdb->allrecord_lock.count) {
+		/* a global lock of a different type exists */
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb_have_extra_locks(tdb)) {
+		/* can't combine global and chain locks */
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (upgradable && ltype != F_RDLCK) {
+		/* tdb error: you can't upgrade a write lock! */
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+	return 1;
+}
+
+/* We only need to lock individual bytes, but Linux merges consecutive locks
+ * so we lock in contiguous ranges. */
+static int tdb_chainlock_gradual(struct tdb_context *tdb,
+				 int ltype, enum tdb_lock_flags flags,
+				 size_t off, size_t len)
+{
+	int ret;
+	enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
+
+	if (len <= 4) {
+		/* Single record.  Just do blocking lock. */
+		return tdb_brlock(tdb, ltype, off, len, flags);
+	}
+
+	/* First we try non-blocking. */
+	ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
+	if (ret == 0) {
+		return 0;
+	}
+
+	/* Try locking first half, then second. */
+	ret = tdb_chainlock_gradual(tdb, ltype, flags, off, len / 2);
+	if (ret == -1)
+		return -1;
+
+	ret = tdb_chainlock_gradual(tdb, ltype, flags,
+				    off + len / 2, len - len / 2);
+	if (ret == -1) {
+		tdb_brunlock(tdb, ltype, off, len / 2);
+		return -1;
+	}
+	return 0;
+}
+
+/* lock/unlock entire database.  It can only be upgradable if you have some
+ * other way of guaranteeing exclusivity (ie. transaction write lock).
+ * We do the locking gradually to avoid being starved by smaller locks. */
+int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
+		       enum tdb_lock_flags flags, bool upgradable)
+{
+	int ret;
+
+	switch (tdb_allrecord_check(tdb, ltype, flags, upgradable)) {
+	case -1:
+		return -1;
+	case 0:
+		return 0;
+	}
+
+	/* We cover two kinds of locks:
+	 * 1) Normal chain locks.  Taken for almost all operations.
+	 * 2) Individual records locks.  Taken after normal or free
+	 *    chain locks.
+	 *
+	 * It is (1) which cause the starvation problem, so we're only
+	 * gradual for that. */
+
+	if (tdb_have_mutexes(tdb)) {
+		ret = tdb_mutex_allrecord_lock(tdb, ltype, flags);
+	} else {
+		ret = tdb_chainlock_gradual(tdb, ltype, flags, FREELIST_TOP,
+					    tdb->hash_size * 4);
+	}
+
+	if (ret == -1) {
+		return -1;
+	}
+
+	/* Grab individual record locks. */
+	if (tdb_brlock(tdb, ltype, lock_offset(tdb->hash_size), 0,
+		       flags) == -1) {
+		if (tdb_have_mutexes(tdb)) {
+			tdb_mutex_allrecord_unlock(tdb);
+		} else {
+			tdb_brunlock(tdb, ltype, FREELIST_TOP,
+				     tdb->hash_size * 4);
+		}
+		return -1;
+	}
+
+	tdb->allrecord_lock.count = 1;
+	/* If it's upgradable, it's actually exclusive so we can treat
+	 * it as a write lock. */
+	tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
+	tdb->allrecord_lock.off = upgradable;
+
+	if (tdb_needs_recovery(tdb)) {
+		bool mark = flags & TDB_LOCK_MARK_ONLY;
+		tdb_allrecord_unlock(tdb, ltype, mark);
+		if (mark) {
+			tdb->ecode = TDB_ERR_LOCK;
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "tdb_lockall_mark cannot do recovery\n"));
+			return -1;
+		}
+		if (tdb_lock_and_recover(tdb) == -1) {
+			return -1;
+		}
+		return tdb_allrecord_lock(tdb, ltype, flags, upgradable);
+	}
+
+	return 0;
+}
+
+
+
+/* unlock entire db */
+int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock)
+{
+	/* There are no locks on read-only dbs */
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb->allrecord_lock.count == 0) {
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	/* Upgradable locks are marked as write locks. */
+	if (tdb->allrecord_lock.ltype != (uint32_t)ltype
+	    && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb->allrecord_lock.count > 1) {
+		tdb->allrecord_lock.count--;
+		return 0;
+	}
+
+	if (!mark_lock) {
+		int ret;
+
+		if (tdb_have_mutexes(tdb)) {
+			ret = tdb_mutex_allrecord_unlock(tdb);
+			if (ret == 0) {
+				ret = tdb_brunlock(tdb, ltype,
+						   lock_offset(tdb->hash_size),
+						   0);
+			}
+		} else {
+			ret = tdb_brunlock(tdb, ltype, FREELIST_TOP, 0);
+		}
+
+		if (ret != 0) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed "
+				 "(%s)\n", strerror(errno)));
+			return -1;
+		}
+	}
+
+	tdb->allrecord_lock.count = 0;
+	tdb->allrecord_lock.ltype = 0;
+
+	return 0;
+}
+
+/* lock entire database with write lock */
+_PUBLIC_ int tdb_lockall(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_lockall");
+	return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
+}
+
+/* lock entire database with write lock - mark only */
+_PUBLIC_ int tdb_lockall_mark(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_lockall_mark");
+	return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY, false);
+}
+
+/* unlock entire database with write lock - unmark only */
+_PUBLIC_ int tdb_lockall_unmark(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_lockall_unmark");
+	return tdb_allrecord_unlock(tdb, F_WRLCK, true);
+}
+
+/* lock entire database with write lock - nonblocking varient */
+_PUBLIC_ int tdb_lockall_nonblock(struct tdb_context *tdb)
+{
+	int ret = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_NOWAIT, false);
+	tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
+	return ret;
+}
+
+/* unlock entire database with write lock */
+_PUBLIC_ int tdb_unlockall(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_unlockall");
+	return tdb_allrecord_unlock(tdb, F_WRLCK, false);
+}
+
+/* lock entire database with read lock */
+_PUBLIC_ int tdb_lockall_read(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_lockall_read");
+	return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
+}
+
+/* lock entire database with read lock - nonblock varient */
+_PUBLIC_ int tdb_lockall_read_nonblock(struct tdb_context *tdb)
+{
+	int ret = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_NOWAIT, false);
+	tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
+	return ret;
+}
+
+/* unlock entire database with read lock */
+_PUBLIC_ int tdb_unlockall_read(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_unlockall_read");
+	return tdb_allrecord_unlock(tdb, F_RDLCK, false);
+}
+
+/* lock/unlock one hash chain. This is meant to be used to reduce
+   contention - it cannot guarantee how many records will be locked */
+_PUBLIC_ int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
+{
+	int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+	tdb_trace_1rec(tdb, "tdb_chainlock", key);
+	return ret;
+}
+
+/* lock/unlock one hash chain, non-blocking. This is meant to be used
+   to reduce contention - it cannot guarantee how many records will be
+   locked */
+_PUBLIC_ int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
+{
+	int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+	tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
+	return ret;
+}
+
+/* mark a chain as locked without actually locking it. Warning! use with great caution! */
+_PUBLIC_ int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
+{
+	int ret = tdb_nest_lock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
+				F_WRLCK, TDB_LOCK_MARK_ONLY);
+	tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
+	return ret;
+}
+
+/* unmark a chain as locked without actually locking it. Warning! use with great caution! */
+_PUBLIC_ int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
+{
+	tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
+	return tdb_nest_unlock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
+			       F_WRLCK, true);
+}
+
+_PUBLIC_ int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
+{
+	tdb_trace_1rec(tdb, "tdb_chainunlock", key);
+	return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+}
+
+_PUBLIC_ int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+	int ret;
+	ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+	tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
+	return ret;
+}
+
+_PUBLIC_ int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+	tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
+	return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+}
+
+_PUBLIC_ int tdb_chainlock_read_nonblock(struct tdb_context *tdb, TDB_DATA key)
+{
+	int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+	tdb_trace_1rec_ret(tdb, "tdb_chainlock_read_nonblock", key, ret);
+	return ret;
+}
+
+/* record lock stops delete underneath */
+int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	if (tdb->allrecord_lock.count) {
+		return 0;
+	}
+	return off ? tdb_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
+}
+
+/*
+  Write locks override our own fcntl readlocks, so check it here.
+  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
+  an error to fail to get the lock here.
+*/
+int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	struct tdb_traverse_lock *i;
+	if (tdb == NULL) {
+		return -1;
+	}
+	for (i = &tdb->travlocks; i; i = i->next)
+		if (i->off == off)
+			return -1;
+	if (tdb->allrecord_lock.count) {
+		if (tdb->allrecord_lock.ltype == F_WRLCK) {
+			return 0;
+		}
+		return -1;
+	}
+	return tdb_brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
+}
+
+int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	if (tdb->allrecord_lock.count) {
+		return 0;
+	}
+	return tdb_brunlock(tdb, F_WRLCK, off, 1);
+}
+
+/* fcntl locks don't stack: avoid unlocking someone else's */
+int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
+{
+	struct tdb_traverse_lock *i;
+	uint32_t count = 0;
+
+	if (tdb->allrecord_lock.count) {
+		return 0;
+	}
+
+	if (off == 0)
+		return 0;
+	for (i = &tdb->travlocks; i; i = i->next)
+		if (i->off == off)
+			count++;
+	return (count == 1 ? tdb_brunlock(tdb, F_RDLCK, off, 1) : 0);
+}
+
+bool tdb_have_extra_locks(struct tdb_context *tdb)
+{
+	unsigned int extra = tdb->num_lockrecs;
+
+	/* A transaction holds the lock for all records. */
+	if (!tdb->transaction && tdb->allrecord_lock.count) {
+		return true;
+	}
+
+	/* We always hold the active lock if CLEAR_IF_FIRST. */
+	if (find_nestlock(tdb, ACTIVE_LOCK)) {
+		extra--;
+	}
+
+	/* In a transaction, we expect to hold the transaction lock */
+	if (tdb->transaction && find_nestlock(tdb, TRANSACTION_LOCK)) {
+		extra--;
+	}
+
+	return extra;
+}
+
+/* The transaction code uses this to remove all locks. */
+void tdb_release_transaction_locks(struct tdb_context *tdb)
+{
+	int i;
+	unsigned int active = 0;
+
+	if (tdb->allrecord_lock.count != 0) {
+		tdb_allrecord_unlock(tdb, tdb->allrecord_lock.ltype, false);
+		tdb->allrecord_lock.count = 0;
+	}
+
+	for (i=0;i<tdb->num_lockrecs;i++) {
+		struct tdb_lock_type *lck = &tdb->lockrecs[i];
+
+		/* Don't release the active lock!  Copy it to first entry. */
+		if (lck->off == ACTIVE_LOCK) {
+			tdb->lockrecs[active++] = *lck;
+		} else {
+			tdb_brunlock(tdb, lck->ltype, lck->off, 1);
+		}
+	}
+	tdb->num_lockrecs = active;
+}
+
+/* Following functions are added specifically to support CTDB. */
+
+/* Don't do actual fcntl locking, just mark tdb locked */
+_PUBLIC_ int tdb_transaction_write_lock_mark(struct tdb_context *tdb);
+_PUBLIC_ int tdb_transaction_write_lock_mark(struct tdb_context *tdb)
+{
+	return tdb_transaction_lock(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY);
+}
+
+/* Don't do actual fcntl unlocking, just mark tdb unlocked */
+_PUBLIC_ int tdb_transaction_write_lock_unmark(struct tdb_context *tdb);
+_PUBLIC_ int tdb_transaction_write_lock_unmark(struct tdb_context *tdb)
+{
+	return tdb_nest_unlock(tdb, TRANSACTION_LOCK, F_WRLCK, true);
+}
diff --git a/lib/tdb/common/mutex.c b/lib/tdb/common/mutex.c
new file mode 100644
index 0000000..78926db
--- /dev/null
+++ b/lib/tdb/common/mutex.c
@@ -0,0 +1,1078 @@
+/*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Volker Lendecke 2012,2013
+   Copyright (C) Stefan Metzmacher 2013,2014
+   Copyright (C) Michael Adam 2014
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "tdb_private.h"
+#include "system/threads.h"
+
+#ifdef USE_TDB_MUTEX_LOCKING
+
+/*
+ * If we run with mutexes, we store the "struct tdb_mutexes" at the
+ * beginning of the file. We store an additional tdb_header right
+ * beyond the mutex area, page aligned. All the offsets within the tdb
+ * are relative to the area behind the mutex area. tdb->map_ptr points
+ * behind the mmap area as well, so the read and write path in the
+ * mutex case can remain unchanged.
+ *
+ * Early in the mutex development the mutexes were placed between the hash
+ * chain pointers and the real tdb data. This had two drawbacks: First, it
+ * made pointer calculations more complex. Second, we had to mmap the mutex
+ * area twice. One was the normal map_ptr in the tdb. This frequently changed
+ * from within tdb_oob. At least the Linux glibc robust mutex code assumes
+ * constant pointers in memory, so a constantly changing mmap area destroys
+ * the mutex list. So we had to mmap the first bytes of the file with a second
+ * mmap call. With that scheme, very weird errors happened that could be
+ * easily fixed by doing the mutex mmap in a second file. It seemed that
+ * mapping the same memory area twice does not end up in accessing the same
+ * physical page, looking at the mutexes in gdb it seemed that old data showed
+ * up after some re-mapping. To avoid a separate mutex file, the code now puts
+ * the real content of the tdb file after the mutex area. This way we do not
+ * have overlapping mmap areas, the mutex area is mmapped once and not
+ * changed, the tdb data area's mmap is constantly changed but does not
+ * overlap.
+ */
+
+struct tdb_mutexes {
+	struct tdb_header hdr;
+
+	/* protect allrecord_lock */
+	pthread_mutex_t allrecord_mutex;
+
+	/*
+	 * F_UNLCK: free,
+	 * F_RDLCK: shared,
+	 * F_WRLCK: exclusive
+	 */
+	short int allrecord_lock;
+
+	/*
+	 * Index 0 is the freelist mutex, followed by
+	 * one mutex per hashchain.
+	 */
+	pthread_mutex_t hashchains[1];
+};
+
+bool tdb_have_mutexes(struct tdb_context *tdb)
+{
+	return ((tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) != 0);
+}
+
+size_t tdb_mutex_size(struct tdb_context *tdb)
+{
+	size_t mutex_size;
+
+	if (!tdb_have_mutexes(tdb)) {
+		return 0;
+	}
+
+	mutex_size = sizeof(struct tdb_mutexes);
+	mutex_size += tdb->hash_size * sizeof(pthread_mutex_t);
+
+	return TDB_ALIGN(mutex_size, tdb->page_size);
+}
+
+/*
+ * Get the index for a chain mutex
+ */
+static bool tdb_mutex_index(struct tdb_context *tdb, off_t off, off_t len,
+			    unsigned *idx)
+{
+	/*
+	 * Weird but true: We fcntl lock 1 byte at an offset 4 bytes before
+	 * the 4 bytes of the freelist start and the hash chain that is about
+	 * to be locked. See lock_offset() where the freelist is -1 vs the
+	 * "+1" in TDB_HASH_TOP(). Because the mutex array is represented in
+	 * the tdb file itself as data, we need to adjust the offset here.
+	 */
+	const off_t freelist_lock_ofs = FREELIST_TOP - sizeof(tdb_off_t);
+
+	if (!tdb_have_mutexes(tdb)) {
+		return false;
+	}
+	if (len != 1) {
+		/* Possibly the allrecord lock */
+		return false;
+	}
+	if (off < freelist_lock_ofs) {
+		/* One of the special locks */
+		return false;
+	}
+	if (tdb->hash_size == 0) {
+		/* tdb not initialized yet, called from tdb_open_ex() */
+		return false;
+	}
+	if (off >= TDB_DATA_START(tdb->hash_size)) {
+		/* Single record lock from traverses */
+		return false;
+	}
+
+	/*
+	 * Now we know it's a freelist or hash chain lock. Those are always 4
+	 * byte aligned. Paranoia check.
+	 */
+	if ((off % sizeof(tdb_off_t)) != 0) {
+		abort();
+	}
+
+	/*
+	 * Re-index the fcntl offset into an offset into the mutex array
+	 */
+	off -= freelist_lock_ofs; /* rebase to index 0 */
+	off /= sizeof(tdb_off_t); /* 0 for freelist 1-n for hashchain */
+
+	*idx = off;
+	return true;
+}
+
+static bool tdb_have_mutex_chainlocks(struct tdb_context *tdb)
+{
+	int i;
+
+	for (i=0; i < tdb->num_lockrecs; i++) {
+		bool ret;
+		unsigned idx;
+
+		ret = tdb_mutex_index(tdb,
+				      tdb->lockrecs[i].off,
+				      tdb->lockrecs[i].count,
+				      &idx);
+		if (!ret) {
+			continue;
+		}
+
+		if (idx == 0) {
+			/* this is the freelist mutex */
+			continue;
+		}
+
+		return true;
+	}
+
+	return false;
+}
+
+static int chain_mutex_lock(pthread_mutex_t *m, bool waitflag)
+{
+	int ret;
+
+	if (waitflag) {
+		ret = pthread_mutex_lock(m);
+	} else {
+		ret = pthread_mutex_trylock(m);
+	}
+	if (ret != EOWNERDEAD) {
+		return ret;
+	}
+
+	/*
+	 * For chainlocks, we don't do any cleanup (yet?)
+	 */
+	return pthread_mutex_consistent(m);
+}
+
+static int allrecord_mutex_lock(struct tdb_mutexes *m, bool waitflag)
+{
+	int ret;
+
+	if (waitflag) {
+		ret = pthread_mutex_lock(&m->allrecord_mutex);
+	} else {
+		ret = pthread_mutex_trylock(&m->allrecord_mutex);
+	}
+	if (ret != EOWNERDEAD) {
+		return ret;
+	}
+
+	/*
+	 * The allrecord lock holder died. We need to reset the allrecord_lock
+	 * to F_UNLCK. This should also be the indication for
+	 * tdb_needs_recovery.
+	 */
+	m->allrecord_lock = F_UNLCK;
+
+	return pthread_mutex_consistent(&m->allrecord_mutex);
+}
+
+bool tdb_mutex_lock(struct tdb_context *tdb, int rw, off_t off, off_t len,
+		    bool waitflag, int *pret)
+{
+	struct tdb_mutexes *m = tdb->mutexes;
+	pthread_mutex_t *chain;
+	int ret;
+	unsigned idx;
+	bool allrecord_ok;
+
+	if (!tdb_mutex_index(tdb, off, len, &idx)) {
+		return false;
+	}
+	chain = &m->hashchains[idx];
+
+again:
+	ret = chain_mutex_lock(chain, waitflag);
+	if (ret == EBUSY) {
+		ret = EAGAIN;
+	}
+	if (ret != 0) {
+		errno = ret;
+		goto fail;
+	}
+
+	if (idx == 0) {
+		/*
+		 * This is a freelist lock, which is independent to
+		 * the allrecord lock. So we're done once we got the
+		 * freelist mutex.
+		 */
+		*pret = 0;
+		return true;
+	}
+
+	if (tdb_have_mutex_chainlocks(tdb)) {
+		/*
+		 * We can only check the allrecord lock once. If we do it with
+		 * one chain mutex locked, we will deadlock with the allrecord
+		 * locker process in the following way: We lock the first hash
+		 * chain, we check for the allrecord lock. We keep the hash
+		 * chain locked. Then the allrecord locker locks the
+		 * allrecord_mutex. It walks the list of chain mutexes,
+		 * locking them all in sequence. Meanwhile, we have the chain
+		 * mutex locked, so the allrecord locker blocks trying to lock
+		 * our chain mutex. Then we come in and try to lock the second
+		 * chain lock, which in most cases will be the freelist. We
+		 * see that the allrecord lock is locked and put ourselves on
+		 * the allrecord_mutex. This will never be signalled though
+		 * because the allrecord locker waits for us to give up the
+		 * chain lock.
+		 */
+
+		*pret = 0;
+		return true;
+	}
+
+	/*
+	 * Check if someone is has the allrecord lock: queue if so.
+	 */
+
+	allrecord_ok = false;
+
+	if (m->allrecord_lock == F_UNLCK) {
+		/*
+		 * allrecord lock not taken
+		 */
+		allrecord_ok = true;
+	}
+
+	if ((m->allrecord_lock == F_RDLCK) && (rw == F_RDLCK)) {
+		/*
+		 * allrecord shared lock taken, but we only want to read
+		 */
+		allrecord_ok = true;
+	}
+
+	if (allrecord_ok) {
+		*pret = 0;
+		return true;
+	}
+
+	ret = pthread_mutex_unlock(chain);
+	if (ret != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_unlock"
+			 "(chain_mutex) failed: %s\n", strerror(ret)));
+		errno = ret;
+		goto fail;
+	}
+	ret = allrecord_mutex_lock(m, waitflag);
+	if (ret == EBUSY) {
+		ret = EAGAIN;
+	}
+	if (ret != 0) {
+		if (waitflag || (ret != EAGAIN)) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_%slock"
+				 "(allrecord_mutex) failed: %s\n",
+				 waitflag ? "" : "try_",  strerror(ret)));
+		}
+		errno = ret;
+		goto fail;
+	}
+	ret = pthread_mutex_unlock(&m->allrecord_mutex);
+	if (ret != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_unlock"
+			 "(allrecord_mutex) failed: %s\n", strerror(ret)));
+		errno = ret;
+		goto fail;
+	}
+	goto again;
+
+fail:
+	*pret = -1;
+	return true;
+}
+
+bool tdb_mutex_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len,
+		      int *pret)
+{
+	struct tdb_mutexes *m = tdb->mutexes;
+	pthread_mutex_t *chain;
+	int ret;
+	unsigned idx;
+
+	if (!tdb_mutex_index(tdb, off, len, &idx)) {
+		return false;
+	}
+	chain = &m->hashchains[idx];
+
+	ret = pthread_mutex_unlock(chain);
+	if (ret == 0) {
+		*pret = 0;
+		return true;
+	}
+	errno = ret;
+	*pret = -1;
+	return true;
+}
+
+int tdb_mutex_allrecord_lock(struct tdb_context *tdb, int ltype,
+			     enum tdb_lock_flags flags)
+{
+	struct tdb_mutexes *m = tdb->mutexes;
+	int ret;
+	uint32_t i;
+	bool waitflag = (flags & TDB_LOCK_WAIT);
+	int saved_errno;
+
+	if (tdb->flags & TDB_NOLOCK) {
+		return 0;
+	}
+
+	if (flags & TDB_LOCK_MARK_ONLY) {
+		return 0;
+	}
+
+	ret = allrecord_mutex_lock(m, waitflag);
+	if (!waitflag && (ret == EBUSY)) {
+		errno = EAGAIN;
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+	if (ret != 0) {
+		if (!(flags & TDB_LOCK_PROBE)) {
+			TDB_LOG((tdb, TDB_DEBUG_TRACE,
+				 "allrecord_mutex_lock() failed: %s\n",
+				 strerror(ret)));
+		}
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (m->allrecord_lock != F_UNLCK) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "allrecord_lock == %d\n",
+			 (int)m->allrecord_lock));
+		goto fail_unlock_allrecord_mutex;
+	}
+	m->allrecord_lock = (ltype == F_RDLCK) ? F_RDLCK : F_WRLCK;
+
+	for (i=0; i<tdb->hash_size; i++) {
+
+		/* ignore hashchains[0], the freelist */
+		pthread_mutex_t *chain = &m->hashchains[i+1];
+
+		ret = chain_mutex_lock(chain, waitflag);
+		if (!waitflag && (ret == EBUSY)) {
+			errno = EAGAIN;
+			goto fail_unroll_allrecord_lock;
+		}
+		if (ret != 0) {
+			if (!(flags & TDB_LOCK_PROBE)) {
+				TDB_LOG((tdb, TDB_DEBUG_TRACE,
+					 "chain_mutex_lock() failed: %s\n",
+					 strerror(ret)));
+			}
+			errno = ret;
+			goto fail_unroll_allrecord_lock;
+		}
+
+		ret = pthread_mutex_unlock(chain);
+		if (ret != 0) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_unlock"
+				 "(chainlock) failed: %s\n", strerror(ret)));
+			errno = ret;
+			goto fail_unroll_allrecord_lock;
+		}
+	}
+	/*
+	 * We leave this routine with m->allrecord_mutex locked
+	 */
+	return 0;
+
+fail_unroll_allrecord_lock:
+	m->allrecord_lock = F_UNLCK;
+
+fail_unlock_allrecord_mutex:
+	saved_errno = errno;
+	ret = pthread_mutex_unlock(&m->allrecord_mutex);
+	if (ret != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_unlock"
+			 "(allrecord_mutex) failed: %s\n", strerror(ret)));
+	}
+	errno = saved_errno;
+	tdb->ecode = TDB_ERR_LOCK;
+	return -1;
+}
+
+int tdb_mutex_allrecord_upgrade(struct tdb_context *tdb)
+{
+	struct tdb_mutexes *m = tdb->mutexes;
+	int ret;
+	uint32_t i;
+
+	if (tdb->flags & TDB_NOLOCK) {
+		return 0;
+	}
+
+	/*
+	 * Our only caller tdb_allrecord_upgrade()
+	 * garantees that we already own the allrecord lock.
+	 *
+	 * Which means m->allrecord_mutex is still locked by us.
+	 */
+
+	if (m->allrecord_lock != F_RDLCK) {
+		tdb->ecode = TDB_ERR_LOCK;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "allrecord_lock == %d\n",
+			 (int)m->allrecord_lock));
+		return -1;
+	}
+
+	m->allrecord_lock = F_WRLCK;
+
+	for (i=0; i<tdb->hash_size; i++) {
+
+		/* ignore hashchains[0], the freelist */
+		pthread_mutex_t *chain = &m->hashchains[i+1];
+
+		ret = chain_mutex_lock(chain, true);
+		if (ret != 0) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_lock"
+				 "(chainlock) failed: %s\n", strerror(ret)));
+			goto fail_unroll_allrecord_lock;
+		}
+
+		ret = pthread_mutex_unlock(chain);
+		if (ret != 0) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_unlock"
+				 "(chainlock) failed: %s\n", strerror(ret)));
+			goto fail_unroll_allrecord_lock;
+		}
+	}
+
+	return 0;
+
+fail_unroll_allrecord_lock:
+	m->allrecord_lock = F_RDLCK;
+	tdb->ecode = TDB_ERR_LOCK;
+	return -1;
+}
+
+void tdb_mutex_allrecord_downgrade(struct tdb_context *tdb)
+{
+	struct tdb_mutexes *m = tdb->mutexes;
+
+	/*
+	 * Our only caller tdb_allrecord_upgrade() (in the error case)
+	 * garantees that we already own the allrecord lock.
+	 *
+	 * Which means m->allrecord_mutex is still locked by us.
+	 */
+
+	if (m->allrecord_lock != F_WRLCK) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "allrecord_lock == %d\n",
+			 (int)m->allrecord_lock));
+		return;
+	}
+
+	m->allrecord_lock = F_RDLCK;
+	return;
+}
+
+
+int tdb_mutex_allrecord_unlock(struct tdb_context *tdb)
+{
+	struct tdb_mutexes *m = tdb->mutexes;
+	short old;
+	int ret;
+
+	if (tdb->flags & TDB_NOLOCK) {
+		return 0;
+	}
+
+	/*
+	 * Our only callers tdb_allrecord_unlock() and
+	 * tdb_allrecord_lock() (in the error path)
+	 * garantee that we already own the allrecord lock.
+	 *
+	 * Which means m->allrecord_mutex is still locked by us.
+	 */
+
+	if ((m->allrecord_lock != F_RDLCK) && (m->allrecord_lock != F_WRLCK)) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "allrecord_lock == %d\n",
+			 (int)m->allrecord_lock));
+		return -1;
+	}
+
+	old = m->allrecord_lock;
+	m->allrecord_lock = F_UNLCK;
+
+	ret = pthread_mutex_unlock(&m->allrecord_mutex);
+	if (ret != 0) {
+		m->allrecord_lock = old;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "pthread_mutex_unlock"
+			 "(allrecord_mutex) failed: %s\n", strerror(ret)));
+		return -1;
+	}
+	return 0;
+}
+
+int tdb_mutex_init(struct tdb_context *tdb)
+{
+	struct tdb_mutexes *m;
+	pthread_mutexattr_t ma;
+	uint32_t i;
+	int ret;
+
+	ret = tdb_mutex_mmap(tdb);
+	if (ret == -1) {
+		return -1;
+	}
+	m = tdb->mutexes;
+
+	ret = pthread_mutexattr_init(&ma);
+	if (ret != 0) {
+		goto fail_munmap;
+	}
+	ret = pthread_mutexattr_settype(&ma, PTHREAD_MUTEX_ERRORCHECK);
+	if (ret != 0) {
+		goto fail;
+	}
+	ret = pthread_mutexattr_setpshared(&ma, PTHREAD_PROCESS_SHARED);
+	if (ret != 0) {
+		goto fail;
+	}
+	ret = pthread_mutexattr_setrobust(&ma, PTHREAD_MUTEX_ROBUST);
+	if (ret != 0) {
+		goto fail;
+	}
+
+	for (i=0; i<tdb->hash_size+1; i++) {
+		pthread_mutex_t *chain = &m->hashchains[i];
+
+		ret = pthread_mutex_init(chain, &ma);
+		if (ret != 0) {
+			goto fail;
+		}
+	}
+
+	m->allrecord_lock = F_UNLCK;
+
+	ret = pthread_mutex_init(&m->allrecord_mutex, &ma);
+	if (ret != 0) {
+		goto fail;
+	}
+	ret = 0;
+fail:
+	pthread_mutexattr_destroy(&ma);
+fail_munmap:
+
+	if (ret == 0) {
+		return 0;
+	}
+
+	tdb_mutex_munmap(tdb);
+
+	errno = ret;
+	return -1;
+}
+
+int tdb_mutex_mmap(struct tdb_context *tdb)
+{
+	size_t len;
+	void *ptr;
+
+	len = tdb_mutex_size(tdb);
+	if (len == 0) {
+		return 0;
+	}
+
+	if (tdb->mutexes != NULL) {
+		return 0;
+	}
+
+	ptr = mmap(NULL, len, PROT_READ|PROT_WRITE, MAP_SHARED|MAP_FILE,
+		   tdb->fd, 0);
+	if (ptr == MAP_FAILED) {
+		return -1;
+	}
+	tdb->mutexes = (struct tdb_mutexes *)ptr;
+
+	return 0;
+}
+
+int tdb_mutex_munmap(struct tdb_context *tdb)
+{
+	size_t len;
+	int ret;
+
+	len = tdb_mutex_size(tdb);
+	if (len == 0) {
+		return 0;
+	}
+
+	ret = munmap(tdb->mutexes, len);
+	if (ret == -1) {
+		return -1;
+	}
+	tdb->mutexes = NULL;
+
+	return 0;
+}
+
+static bool tdb_mutex_locking_cached;
+
+static bool tdb_mutex_locking_supported(void)
+{
+	pthread_mutexattr_t ma;
+	pthread_mutex_t m;
+	int ret;
+	static bool initialized;
+
+	if (initialized) {
+		return tdb_mutex_locking_cached;
+	}
+
+	initialized = true;
+
+	ret = pthread_mutexattr_init(&ma);
+	if (ret != 0) {
+		return false;
+	}
+	ret = pthread_mutexattr_settype(&ma, PTHREAD_MUTEX_ERRORCHECK);
+	if (ret != 0) {
+		goto cleanup_ma;
+	}
+	ret = pthread_mutexattr_setpshared(&ma, PTHREAD_PROCESS_SHARED);
+	if (ret != 0) {
+		goto cleanup_ma;
+	}
+	ret = pthread_mutexattr_setrobust(&ma, PTHREAD_MUTEX_ROBUST);
+	if (ret != 0) {
+		goto cleanup_ma;
+	}
+	ret = pthread_mutex_init(&m, &ma);
+	if (ret != 0) {
+		goto cleanup_ma;
+	}
+	ret = pthread_mutex_lock(&m);
+	if (ret != 0) {
+		goto cleanup_m;
+	}
+	/*
+	 * This makes sure we have real mutexes
+	 * from a threading library instead of just
+	 * stubs from libc.
+	 */
+	ret = pthread_mutex_lock(&m);
+	if (ret != EDEADLK) {
+		goto cleanup_lock;
+	}
+	ret = pthread_mutex_unlock(&m);
+	if (ret != 0) {
+		goto cleanup_m;
+	}
+
+	tdb_mutex_locking_cached = true;
+	goto cleanup_m;
+
+cleanup_lock:
+	pthread_mutex_unlock(&m);
+cleanup_m:
+	pthread_mutex_destroy(&m);
+cleanup_ma:
+	pthread_mutexattr_destroy(&ma);
+	return tdb_mutex_locking_cached;
+}
+
+static void (*tdb_robust_mutext_old_handler)(int) = SIG_ERR;
+static pid_t tdb_robust_mutex_pid = -1;
+
+static bool tdb_robust_mutex_setup_sigchild(void (*handler)(int),
+			void (**p_old_handler)(int))
+{
+#ifdef HAVE_SIGACTION
+	struct sigaction act;
+	struct sigaction oldact;
+
+	memset(&act, '\0', sizeof(act));
+
+	act.sa_handler = handler;
+#ifdef SA_RESTART
+	act.sa_flags = SA_RESTART;
+#endif
+	sigemptyset(&act.sa_mask);
+	sigaddset(&act.sa_mask, SIGCHLD);
+	sigaction(SIGCHLD, &act, &oldact);
+	if (p_old_handler) {
+		*p_old_handler = oldact.sa_handler;
+	}
+	return true;
+#else /* !HAVE_SIGACTION */
+	return false;
+#endif
+}
+
+static void tdb_robust_mutex_handler(int sig)
+{
+	pid_t child_pid = tdb_robust_mutex_pid;
+
+	if (child_pid != -1) {
+		pid_t pid;
+
+		pid = waitpid(child_pid, NULL, WNOHANG);
+		if (pid == -1) {
+			switch (errno) {
+			case ECHILD:
+				tdb_robust_mutex_pid = -1;
+				return;
+
+			default:
+				return;
+			}
+		}
+		if (pid == child_pid) {
+			tdb_robust_mutex_pid = -1;
+			return;
+		}
+	}
+
+	if (tdb_robust_mutext_old_handler == SIG_DFL) {
+		return;
+	}
+	if (tdb_robust_mutext_old_handler == SIG_IGN) {
+		return;
+	}
+	if (tdb_robust_mutext_old_handler == SIG_ERR) {
+		return;
+	}
+
+	tdb_robust_mutext_old_handler(sig);
+}
+
+static void tdb_robust_mutex_wait_for_child(pid_t *child_pid)
+{
+	int options = WNOHANG;
+
+	if (*child_pid == -1) {
+		return;
+	}
+
+	while (tdb_robust_mutex_pid > 0) {
+		pid_t pid;
+
+		/*
+		 * First we try with WNOHANG, as the process might not exist
+		 * anymore. Once we've sent SIGKILL we block waiting for the
+		 * exit.
+		 */
+		pid = waitpid(*child_pid, NULL, options);
+		if (pid == -1) {
+			if (errno == EINTR) {
+				continue;
+			} else if (errno == ECHILD) {
+				break;
+			} else {
+				abort();
+			}
+		}
+		if (pid == *child_pid) {
+			break;
+		}
+
+		kill(*child_pid, SIGKILL);
+		options = 0;
+	}
+
+	tdb_robust_mutex_pid = -1;
+	*child_pid = -1;
+}
+
+_PUBLIC_ bool tdb_runtime_check_for_robust_mutexes(void)
+{
+	void *ptr = NULL;
+	pthread_mutex_t *m = NULL;
+	pthread_mutexattr_t ma;
+	int ret = 1;
+	int pipe_down[2] = { -1, -1 };
+	int pipe_up[2] = { -1, -1 };
+	ssize_t nread;
+	char c = 0;
+	bool ok;
+	static bool initialized;
+	pid_t saved_child_pid = -1;
+	bool cleanup_ma = false;
+
+	if (initialized) {
+		return tdb_mutex_locking_cached;
+	}
+
+	initialized = true;
+
+	ok = tdb_mutex_locking_supported();
+	if (!ok) {
+		return false;
+	}
+
+	tdb_mutex_locking_cached = false;
+
+	ptr = mmap(NULL, sizeof(pthread_mutex_t), PROT_READ|PROT_WRITE,
+		   MAP_SHARED|MAP_ANON, -1 /* fd */, 0);
+	if (ptr == MAP_FAILED) {
+		return false;
+	}
+
+	ret = pipe(pipe_down);
+	if (ret != 0) {
+		goto cleanup;
+	}
+	ret = pipe(pipe_up);
+	if (ret != 0) {
+		goto cleanup;
+	}
+
+	ret = pthread_mutexattr_init(&ma);
+	if (ret != 0) {
+		goto cleanup;
+	}
+	cleanup_ma = true;
+	ret = pthread_mutexattr_settype(&ma, PTHREAD_MUTEX_ERRORCHECK);
+	if (ret != 0) {
+		goto cleanup;
+	}
+	ret = pthread_mutexattr_setpshared(&ma, PTHREAD_PROCESS_SHARED);
+	if (ret != 0) {
+		goto cleanup;
+	}
+	ret = pthread_mutexattr_setrobust(&ma, PTHREAD_MUTEX_ROBUST);
+	if (ret != 0) {
+		goto cleanup;
+	}
+	ret = pthread_mutex_init(ptr, &ma);
+	if (ret != 0) {
+		goto cleanup;
+	}
+	m = (pthread_mutex_t *)ptr;
+
+	if (tdb_robust_mutex_setup_sigchild(tdb_robust_mutex_handler,
+			&tdb_robust_mutext_old_handler) == false) {
+		goto cleanup;
+	}
+
+	tdb_robust_mutex_pid = fork();
+	saved_child_pid = tdb_robust_mutex_pid;
+	if (tdb_robust_mutex_pid == 0) {
+		size_t nwritten;
+		close(pipe_down[1]);
+		close(pipe_up[0]);
+		ret = pthread_mutex_lock(m);
+		nwritten = write(pipe_up[1], &ret, sizeof(ret));
+		if (nwritten != sizeof(ret)) {
+			_exit(1);
+		}
+		if (ret != 0) {
+			_exit(1);
+		}
+		nread = read(pipe_down[0], &c, 1);
+		if (nread != 1) {
+			_exit(1);
+		}
+		/* leave locked */
+		_exit(0);
+	}
+	if (tdb_robust_mutex_pid == -1) {
+		goto cleanup;
+	}
+	close(pipe_down[0]);
+	pipe_down[0] = -1;
+	close(pipe_up[1]);
+	pipe_up[1] = -1;
+
+	nread = read(pipe_up[0], &ret, sizeof(ret));
+	if (nread != sizeof(ret)) {
+		goto cleanup;
+	}
+
+	ret = pthread_mutex_trylock(m);
+	if (ret != EBUSY) {
+		if (ret == 0) {
+			pthread_mutex_unlock(m);
+		}
+		goto cleanup;
+	}
+
+	if (write(pipe_down[1], &c, 1) != 1) {
+		goto cleanup;
+	}
+
+	nread = read(pipe_up[0], &c, 1);
+	if (nread != 0) {
+		goto cleanup;
+	}
+
+	tdb_robust_mutex_wait_for_child(&saved_child_pid);
+
+	ret = pthread_mutex_trylock(m);
+	if (ret != EOWNERDEAD) {
+		if (ret == 0) {
+			pthread_mutex_unlock(m);
+		}
+		goto cleanup;
+	}
+
+	ret = pthread_mutex_consistent(m);
+	if (ret != 0) {
+		goto cleanup;
+	}
+
+	ret = pthread_mutex_trylock(m);
+	if (ret != EDEADLK && ret != EBUSY) {
+		pthread_mutex_unlock(m);
+		goto cleanup;
+	}
+
+	ret = pthread_mutex_unlock(m);
+	if (ret != 0) {
+		goto cleanup;
+	}
+
+	tdb_mutex_locking_cached = true;
+
+cleanup:
+	/*
+	 * Note that we don't reset the signal handler we just reset
+	 * tdb_robust_mutex_pid to -1. This is ok as this code path is only
+	 * called once per process.
+	 *
+	 * Leaving our signal handler avoids races with other threads potentialy
+	 * setting up their SIGCHLD handlers.
+	 *
+	 * The worst thing that can happen is that the other newer signal
+	 * handler will get the SIGCHLD signal for our child and/or reap the
+	 * child with a wait() function. tdb_robust_mutex_wait_for_child()
+	 * handles the case where waitpid returns ECHILD.
+	 */
+	tdb_robust_mutex_wait_for_child(&saved_child_pid);
+
+	if (m != NULL) {
+		pthread_mutex_destroy(m);
+	}
+	if (cleanup_ma) {
+		pthread_mutexattr_destroy(&ma);
+	}
+	if (pipe_down[0] != -1) {
+		close(pipe_down[0]);
+	}
+	if (pipe_down[1] != -1) {
+		close(pipe_down[1]);
+	}
+	if (pipe_up[0] != -1) {
+		close(pipe_up[0]);
+	}
+	if (pipe_up[1] != -1) {
+		close(pipe_up[1]);
+	}
+	if (ptr != NULL) {
+		munmap(ptr, sizeof(pthread_mutex_t));
+	}
+
+	return tdb_mutex_locking_cached;
+}
+
+#else
+
+size_t tdb_mutex_size(struct tdb_context *tdb)
+{
+	return 0;
+}
+
+bool tdb_have_mutexes(struct tdb_context *tdb)
+{
+	return false;
+}
+
+int tdb_mutex_allrecord_lock(struct tdb_context *tdb, int ltype,
+			     enum tdb_lock_flags flags)
+{
+	tdb->ecode = TDB_ERR_LOCK;
+	return -1;
+}
+
+int tdb_mutex_allrecord_unlock(struct tdb_context *tdb)
+{
+	return -1;
+}
+
+int tdb_mutex_allrecord_upgrade(struct tdb_context *tdb)
+{
+	tdb->ecode = TDB_ERR_LOCK;
+	return -1;
+}
+
+void tdb_mutex_allrecord_downgrade(struct tdb_context *tdb)
+{
+	return;
+}
+
+int tdb_mutex_mmap(struct tdb_context *tdb)
+{
+	errno = ENOSYS;
+	return -1;
+}
+
+int tdb_mutex_munmap(struct tdb_context *tdb)
+{
+	errno = ENOSYS;
+	return -1;
+}
+
+int tdb_mutex_init(struct tdb_context *tdb)
+{
+	errno = ENOSYS;
+	return -1;
+}
+
+_PUBLIC_ bool tdb_runtime_check_for_robust_mutexes(void)
+{
+	return false;
+}
+
+#endif
diff --git a/lib/tdb/common/open.c b/lib/tdb/common/open.c
new file mode 100644
index 0000000..f7f65b0
--- /dev/null
+++ b/lib/tdb/common/open.c
@@ -0,0 +1,962 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
+static struct tdb_context *tdbs = NULL;
+
+/* We use two hashes to double-check they're using the right hash function. */
+void tdb_header_hash(struct tdb_context *tdb,
+		     uint32_t *magic1_hash, uint32_t *magic2_hash)
+{
+	TDB_DATA hash_key;
+	uint32_t tdb_magic = TDB_MAGIC;
+
+	hash_key.dptr = discard_const_p(unsigned char, TDB_MAGIC_FOOD);
+	hash_key.dsize = sizeof(TDB_MAGIC_FOOD);
+	*magic1_hash = tdb->hash_fn(&hash_key);
+
+	hash_key.dptr = (unsigned char *)CONVERT(tdb_magic);
+	hash_key.dsize = sizeof(tdb_magic);
+	*magic2_hash = tdb->hash_fn(&hash_key);
+
+	/* Make sure at least one hash is non-zero! */
+	if (*magic1_hash == 0 && *magic2_hash == 0)
+		*magic1_hash = 1;
+}
+
+/* initialise a new database with a specified hash size */
+static int tdb_new_database(struct tdb_context *tdb, struct tdb_header *header,
+			    int hash_size)
+{
+	struct tdb_header *newdb;
+	size_t size;
+	int ret = -1;
+
+	/* We make it up in memory, then write it out if not internal */
+	size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
+	if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
+		tdb->ecode = TDB_ERR_OOM;
+		return -1;
+	}
+
+	/* Fill in the header */
+	newdb->version = TDB_VERSION;
+	newdb->hash_size = hash_size;
+
+	tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
+
+	/* Make sure older tdbs (which don't check the magic hash fields)
+	 * will refuse to open this TDB. */
+	if (tdb->flags & TDB_INCOMPATIBLE_HASH)
+		newdb->rwlocks = TDB_HASH_RWLOCK_MAGIC;
+
+	/*
+	 * We create a tdb with TDB_FEATURE_FLAG_MUTEX support,
+	 * the flag combination and runtime feature checks
+	 * are done by the caller already.
+	 */
+	if (tdb->flags & TDB_MUTEX_LOCKING) {
+		newdb->feature_flags |= TDB_FEATURE_FLAG_MUTEX;
+	}
+
+	/*
+	 * If we have any features we add the FEATURE_FLAG_MAGIC, overwriting the
+	 * TDB_HASH_RWLOCK_MAGIC above.
+	 */
+	if (newdb->feature_flags != 0) {
+		newdb->rwlocks = TDB_FEATURE_FLAG_MAGIC;
+	}
+
+	/*
+	 * It's required for some following code pathes
+	 * to have the fields on 'tdb' up-to-date.
+	 *
+	 * E.g. tdb_mutex_size() requires it
+	 */
+	tdb->feature_flags = newdb->feature_flags;
+	tdb->hash_size = newdb->hash_size;
+
+	if (tdb->flags & TDB_INTERNAL) {
+		tdb->map_size = size;
+		tdb->map_ptr = (char *)newdb;
+		memcpy(header, newdb, sizeof(*header));
+		/* Convert the `ondisk' version if asked. */
+		CONVERT(*newdb);
+		return 0;
+	}
+	if (lseek(tdb->fd, 0, SEEK_SET) == -1)
+		goto fail;
+
+	if (ftruncate(tdb->fd, 0) == -1)
+		goto fail;
+
+	if (newdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+		newdb->mutex_size = tdb_mutex_size(tdb);
+		tdb->hdr_ofs = newdb->mutex_size;
+	}
+
+	/* This creates an endian-converted header, as if read from disk */
+	CONVERT(*newdb);
+	memcpy(header, newdb, sizeof(*header));
+	/* Don't endian-convert the magic food! */
+	memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
+
+	if (!tdb_write_all(tdb->fd, newdb, size))
+		goto fail;
+
+	if (newdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+
+		/*
+		 * Now we init the mutex area
+		 * followed by a second header.
+		 */
+
+		ret = ftruncate(
+			tdb->fd,
+			newdb->mutex_size + sizeof(struct tdb_header));
+		if (ret == -1) {
+			goto fail;
+		}
+		ret = tdb_mutex_init(tdb);
+		if (ret == -1) {
+			goto fail;
+		}
+
+		/*
+		 * Write a second header behind the mutexes. That's the area
+		 * that will be mmapp'ed.
+		 */
+		ret = lseek(tdb->fd, newdb->mutex_size, SEEK_SET);
+		if (ret == -1) {
+			goto fail;
+		}
+		if (!tdb_write_all(tdb->fd, newdb, size)) {
+			goto fail;
+		}
+	}
+
+	ret = 0;
+  fail:
+	SAFE_FREE(newdb);
+	return ret;
+}
+
+
+
+static int tdb_already_open(dev_t device,
+			    ino_t ino)
+{
+	struct tdb_context *i;
+
+	for (i = tdbs; i; i = i->next) {
+		if (i->device == device && i->inode == ino) {
+			return 1;
+		}
+	}
+
+	return 0;
+}
+
+/* open the database, creating it if necessary
+
+   The open_flags and mode are passed straight to the open call on the
+   database file. A flags value of O_WRONLY is invalid. The hash size
+   is advisory, use zero for a default value.
+
+   Return is NULL on error, in which case errno is also set.  Don't
+   try to call tdb_error or tdb_errname, just do strerror(errno).
+
+   @param name may be NULL for internal databases. */
+_PUBLIC_ struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
+		      int open_flags, mode_t mode)
+{
+	return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
+}
+
+/* a default logging function */
+static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
+static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
+{
+}
+
+static bool check_header_hash(struct tdb_context *tdb,
+			      struct tdb_header *header,
+			      bool default_hash, uint32_t *m1, uint32_t *m2)
+{
+	tdb_header_hash(tdb, m1, m2);
+	if (header->magic1_hash == *m1 &&
+	    header->magic2_hash == *m2) {
+		return true;
+	}
+
+	/* If they explicitly set a hash, always respect it. */
+	if (!default_hash)
+		return false;
+
+	/* Otherwise, try the other inbuilt hash. */
+	if (tdb->hash_fn == tdb_old_hash)
+		tdb->hash_fn = tdb_jenkins_hash;
+	else
+		tdb->hash_fn = tdb_old_hash;
+	return check_header_hash(tdb, header, false, m1, m2);
+}
+
+static bool tdb_mutex_open_ok(struct tdb_context *tdb,
+			      const struct tdb_header *header)
+{
+	if (tdb->flags & TDB_NOLOCK) {
+		/*
+		 * We don't look at locks, so it does not matter to have a
+		 * compatible mutex implementation. Allow the open.
+		 */
+		return true;
+	}
+
+	if (!(tdb->flags & TDB_MUTEX_LOCKING)) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_mutex_open_ok[%s]: "
+			 "Can use mutexes only with "
+			 "MUTEX_LOCKING or NOLOCK\n",
+			 tdb->name));
+		return false;
+	}
+
+	if (tdb_mutex_size(tdb) != header->mutex_size) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_mutex_open_ok[%s]: "
+			 "Mutex size changed from %"PRIu32" to %zu\n.",
+			 tdb->name,
+			 header->mutex_size,
+			 tdb_mutex_size(tdb)));
+		return false;
+	}
+
+	return true;
+}
+
+_PUBLIC_ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+				int open_flags, mode_t mode,
+				const struct tdb_logging_context *log_ctx,
+				tdb_hash_func hash_fn)
+{
+	int orig_errno = errno;
+	struct tdb_header header = {
+		.version = 0,
+	};
+	struct tdb_context *tdb;
+	struct stat st;
+	int rev = 0;
+	bool locked = false;
+	unsigned char *vp;
+	uint32_t vertest;
+	unsigned v;
+	const char *hash_alg;
+	uint32_t magic1, magic2;
+	int ret;
+
+	if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
+		/* Can't log this */
+		errno = ENOMEM;
+		goto fail;
+	}
+	tdb_io_init(tdb);
+
+	if (tdb_flags & TDB_INTERNAL) {
+		tdb_flags |= TDB_INCOMPATIBLE_HASH;
+	}
+	if (tdb_flags & TDB_MUTEX_LOCKING) {
+		tdb_flags |= TDB_INCOMPATIBLE_HASH;
+	}
+
+	tdb->fd = -1;
+#ifdef TDB_TRACE
+	tdb->tracefd = -1;
+#endif
+	tdb->name = NULL;
+	tdb->map_ptr = NULL;
+	tdb->flags = tdb_flags;
+	tdb->open_flags = open_flags;
+	if (log_ctx) {
+		tdb->log = *log_ctx;
+	} else {
+		tdb->log.log_fn = null_log_fn;
+		tdb->log.log_private = NULL;
+	}
+
+	if (name == NULL && (tdb_flags & TDB_INTERNAL)) {
+		name = "__TDB_INTERNAL__";
+	}
+
+	if (name == NULL) {
+		tdb->name = discard_const_p(char, "__NULL__");
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: called with name == NULL\n"));
+		tdb->name = NULL;
+		errno = EINVAL;
+		goto fail;
+	}
+
+	/* now make a copy of the name, as the caller memory might go away */
+	if (!(tdb->name = (char *)strdup(name))) {
+		/*
+		 * set the name as the given string, so that tdb_name() will
+		 * work in case of an error.
+		 */
+		tdb->name = discard_const_p(char, name);
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't strdup(%s)\n",
+			 name));
+		tdb->name = NULL;
+		errno = ENOMEM;
+		goto fail;
+	}
+
+	if (hash_fn) {
+		tdb->hash_fn = hash_fn;
+		hash_alg = "the user defined";
+	} else {
+		/* This controls what we use when creating a tdb. */
+		if (tdb->flags & TDB_INCOMPATIBLE_HASH) {
+			tdb->hash_fn = tdb_jenkins_hash;
+		} else {
+			tdb->hash_fn = tdb_old_hash;
+		}
+		hash_alg = "either default";
+	}
+
+	/* cache the page size */
+	tdb->page_size = getpagesize();
+	if (tdb->page_size <= 0) {
+		tdb->page_size = 0x2000;
+	}
+
+	tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
+
+	if ((open_flags & O_ACCMODE) == O_WRONLY) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
+			 name));
+		errno = EINVAL;
+		goto fail;
+	}
+
+	if (hash_size == 0)
+		hash_size = DEFAULT_HASH_SIZE;
+	if ((open_flags & O_ACCMODE) == O_RDONLY) {
+		tdb->read_only = 1;
+		/* read only databases don't do locking or clear if first */
+		tdb->flags |= TDB_NOLOCK;
+		tdb->flags &= ~(TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING);
+	}
+
+	if ((tdb->flags & TDB_ALLOW_NESTING) &&
+	    (tdb->flags & TDB_DISALLOW_NESTING)) {
+		tdb->ecode = TDB_ERR_NESTING;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+			"allow_nesting and disallow_nesting are not allowed together!"));
+		errno = EINVAL;
+		goto fail;
+	}
+
+	if (tdb->flags & TDB_MUTEX_LOCKING) {
+		/*
+		 * Here we catch bugs in the callers,
+		 * the runtime check for existing tdb's comes later.
+		 */
+
+		if (tdb->flags & TDB_INTERNAL) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+				"invalid flags for %s - TDB_MUTEX_LOCKING and "
+				"TDB_INTERNAL are not allowed together\n", name));
+			errno = EINVAL;
+			goto fail;
+		}
+
+		if (tdb->flags & TDB_NOMMAP) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+				"invalid flags for %s - TDB_MUTEX_LOCKING and "
+				"TDB_NOMMAP are not allowed together\n", name));
+			errno = EINVAL;
+			goto fail;
+		}
+
+		if (tdb->read_only) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+				"invalid flags for %s - TDB_MUTEX_LOCKING "
+				"not allowed read only\n", name));
+			errno = EINVAL;
+			goto fail;
+		}
+
+		/*
+		 * The callers should have called
+		 * tdb_runtime_check_for_robust_mutexes()
+		 * before using TDB_MUTEX_LOCKING!
+		 *
+		 * This makes sure the caller understands
+		 * that the locking may behave a bit differently
+		 * than with pure fcntl locking. E.g. multiple
+		 * read locks are not supported.
+		 */
+		if (!tdb_runtime_check_for_robust_mutexes()) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+				"invalid flags for %s - TDB_MUTEX_LOCKING "
+				"requires support for robust_mutexes\n",
+				name));
+			errno = ENOSYS;
+			goto fail;
+		}
+	}
+
+	if (getenv("TDB_NO_FSYNC")) {
+		tdb->flags |= TDB_NOSYNC;
+	}
+
+	/*
+	 * TDB_ALLOW_NESTING is the default behavior.
+	 * Note: this may change in future versions!
+	 */
+	if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
+		tdb->flags |= TDB_ALLOW_NESTING;
+	}
+
+	/* internal databases don't mmap or lock, and start off cleared */
+	if (tdb->flags & TDB_INTERNAL) {
+		tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
+		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+		if (tdb_new_database(tdb, &header, hash_size) != 0) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
+			goto fail;
+		}
+		tdb->hash_size = hash_size;
+		goto internal;
+	}
+
+	if ((tdb->fd = open(name, open_flags, mode)) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
+			 name, strerror(errno)));
+		goto fail;	/* errno set by open(2) */
+	}
+
+	/* on exec, don't inherit the fd */
+	v = fcntl(tdb->fd, F_GETFD, 0);
+        fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
+
+	/* ensure there is only one process initialising at once */
+	if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
+			 name, strerror(errno)));
+		goto fail;	/* errno set by tdb_brlock */
+	}
+
+	/* we need to zero database if we are the only one with it open */
+	if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
+	    (!tdb->read_only)) {
+		ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK,
+				    TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
+		locked = (ret == 0);
+
+		if (locked) {
+			ret = tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0,
+					 TDB_LOCK_WAIT);
+			if (ret == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+					 "tdb_brlock failed for %s: %s\n",
+					 name, strerror(errno)));
+				goto fail;
+			}
+			ret = tdb_new_database(tdb, &header, hash_size);
+			if (ret == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+					 "tdb_new_database failed for "
+					 "%s: %s\n", name, strerror(errno)));
+				tdb_unlockall(tdb);
+				goto fail;
+			}
+			ret = tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
+			if (ret == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+					 "tdb_unlockall failed for %s: %s\n",
+					 name, strerror(errno)));
+				goto fail;
+			}
+			ret = lseek(tdb->fd, 0, SEEK_SET);
+			if (ret == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+					 "lseek failed for %s: %s\n",
+					 name, strerror(errno)));
+				goto fail;
+			}
+		}
+	}
+
+	errno = 0;
+	if (read(tdb->fd, &header, sizeof(header)) != sizeof(header)
+	    || strcmp(header.magic_food, TDB_MAGIC_FOOD) != 0) {
+		if (!(open_flags & O_CREAT) ||
+		    tdb_new_database(tdb, &header, hash_size) == -1) {
+			if (errno == 0) {
+				errno = EIO; /* ie bad format or something */
+			}
+			goto fail;
+		}
+		rev = (tdb->flags & TDB_CONVERT);
+	} else if (header.version != TDB_VERSION
+		   && !(rev = (header.version==TDB_BYTEREV(TDB_VERSION)))) {
+		/* wrong version */
+		errno = EIO;
+		goto fail;
+	}
+	vp = (unsigned char *)&header.version;
+	vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
+		  (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
+	tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
+	if (!rev)
+		tdb->flags &= ~TDB_CONVERT;
+	else {
+		tdb->flags |= TDB_CONVERT;
+		tdb_convert(&header, sizeof(header));
+	}
+
+	/*
+	 * We only use st.st_dev and st.st_ino from the raw fstat()
+	 * call, everything else needs to use tdb_fstat() in order
+	 * to skip tdb->hdr_ofs!
+	 */
+	if (fstat(tdb->fd, &st) == -1) {
+		goto fail;
+	}
+	tdb->device = st.st_dev;
+	tdb->inode = st.st_ino;
+	ZERO_STRUCT(st);
+
+	if (header.rwlocks != 0 &&
+	    header.rwlocks != TDB_FEATURE_FLAG_MAGIC &&
+	    header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
+		errno = ENOSYS;
+		goto fail;
+	}
+
+	if (header.hash_size == 0) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: invalid database: 0 hash_size\n"));
+		errno = ENOSYS;
+		goto fail;
+	}
+
+	tdb->hash_size = header.hash_size;
+
+	if (header.rwlocks == TDB_FEATURE_FLAG_MAGIC) {
+		tdb->feature_flags = header.feature_flags;
+	}
+
+	if (tdb->feature_flags & ~TDB_SUPPORTED_FEATURE_FLAGS) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: unsupported "
+			 "features in tdb %s: 0x%08x (supported: 0x%08x)\n",
+			 name, (unsigned)tdb->feature_flags,
+			 (unsigned)TDB_SUPPORTED_FEATURE_FLAGS));
+		errno = ENOSYS;
+		goto fail;
+	}
+
+	if (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+		if (!tdb_mutex_open_ok(tdb, &header)) {
+			errno = EINVAL;
+			goto fail;
+		}
+
+		/*
+		 * We need to remember the hdr_ofs
+		 * also for the TDB_NOLOCK case
+		 * if the current library doesn't support
+		 * mutex locking.
+		 */
+		tdb->hdr_ofs = header.mutex_size;
+
+		if ((!(tdb_flags & TDB_CLEAR_IF_FIRST)) && (!tdb->read_only)) {
+			/*
+			 * Open an existing mutexed tdb, but without
+			 * CLEAR_IF_FIRST. We need to initialize the
+			 * mutex array and keep the CLEAR_IF_FIRST
+			 * lock locked.
+			 */
+			ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK,
+					    TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
+			locked = (ret == 0);
+
+			if (locked) {
+				ret = tdb_mutex_init(tdb);
+				if (ret == -1) {
+					TDB_LOG((tdb,
+						 TDB_DEBUG_FATAL,
+						 "tdb_open_ex: tdb_mutex_init "
+						 "failed for ""%s: %s\n",
+						 name, strerror(errno)));
+					goto fail;
+				}
+			}
+		}
+	}
+
+	if ((header.magic1_hash == 0) && (header.magic2_hash == 0)) {
+		/* older TDB without magic hash references */
+		tdb->hash_fn = tdb_old_hash;
+	} else if (!check_header_hash(tdb, &header, !hash_fn,
+				      &magic1, &magic2)) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+			 "%s was not created with %s hash function we are using\n"
+			 "magic1_hash[0x%08X %s 0x%08X] "
+			 "magic2_hash[0x%08X %s 0x%08X]\n",
+			 name, hash_alg,
+			 header.magic1_hash,
+			 (header.magic1_hash == magic1) ? "==" : "!=",
+			 magic1,
+			 header.magic2_hash,
+			 (header.magic2_hash == magic2) ? "==" : "!=",
+			 magic2));
+		errno = EINVAL;
+		goto fail;
+	}
+
+	/* Is it already in the open list?  If so, fail. */
+	if (tdb_already_open(tdb->device, tdb->inode)) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+			 "%s (%d,%d) is already open in this process\n",
+			 name, (int)tdb->device, (int)tdb->inode));
+		errno = EBUSY;
+		goto fail;
+	}
+
+	/*
+	 * We had tdb_mmap(tdb) here before,
+	 * but we need to use tdb_fstat(),
+	 * which is triggered from tdb_oob() before calling tdb_mmap().
+	 * As this skips tdb->hdr_ofs.
+	 */
+	tdb->map_size = 0;
+	ret = tdb_oob(tdb, 0, 1, 0);
+	if (ret == -1) {
+		errno = EIO;
+		goto fail;
+	}
+
+	if (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+		if (!(tdb->flags & TDB_NOLOCK)) {
+			ret = tdb_mutex_mmap(tdb);
+			if (ret != 0) {
+				goto fail;
+			}
+		}
+	}
+
+	if (tdb->hash_size > UINT32_MAX/4) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+			 "hash size %"PRIu32" too large\n", tdb->hash_size));
+		errno = EINVAL;
+		goto fail;
+	}
+
+	ret = tdb_oob(tdb, FREELIST_TOP, 4*tdb->hash_size, 1);
+	if (ret == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+			 "hash size %"PRIu32" does not fit\n", tdb->hash_size));
+		errno = EINVAL;
+		goto fail;
+	}
+
+	if (locked) {
+		if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+				 "failed to release ACTIVE_LOCK on %s: %s\n",
+				 name, strerror(errno)));
+			goto fail;
+		}
+
+
+	}
+
+	if (locked || (tdb_flags & TDB_CLEAR_IF_FIRST)) {
+		/*
+		 * We always need to do this if the CLEAR_IF_FIRST
+		 * flag is set, even if we didn't get the initial
+		 * exclusive lock as we need to let all other users
+		 * know we're using it.
+		 */
+
+		ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT);
+		if (ret == -1) {
+			goto fail;
+		}
+	}
+
+	/* if needed, run recovery */
+	if (tdb_transaction_recover(tdb) == -1) {
+		goto fail;
+	}
+
+#ifdef TDB_TRACE
+	{
+		char tracefile[strlen(name) + 32];
+
+		snprintf(tracefile, sizeof(tracefile),
+			 "%s.trace.%li", name, (long)getpid());
+		tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
+		if (tdb->tracefd >= 0) {
+			tdb_enable_seqnum(tdb);
+			tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
+				       open_flags);
+		} else
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
+	}
+#endif
+
+ internal:
+	/* Internal (memory-only) databases skip all the code above to
+	 * do with disk files, and resume here by releasing their
+	 * open lock and hooking into the active list. */
+	if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
+		goto fail;
+	}
+	tdb->next = tdbs;
+	tdbs = tdb;
+	errno = orig_errno;
+	return tdb;
+
+ fail:
+	{ int save_errno = errno;
+
+	if (!tdb)
+		return NULL;
+
+#ifdef TDB_TRACE
+	close(tdb->tracefd);
+#endif
+	if (tdb->map_ptr) {
+		if (tdb->flags & TDB_INTERNAL)
+			SAFE_FREE(tdb->map_ptr);
+		else
+			tdb_munmap(tdb);
+	}
+	if (tdb->fd != -1)
+		if (close(tdb->fd) != 0)
+			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
+	SAFE_FREE(tdb->lockrecs);
+	SAFE_FREE(tdb->name);
+	SAFE_FREE(tdb);
+	errno = save_errno;
+	return NULL;
+	}
+}
+
+/*
+ * Set the maximum number of dead records per hash chain
+ */
+
+_PUBLIC_ void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
+{
+	tdb->max_dead_records = max_dead;
+}
+
+/**
+ * Close a database.
+ *
+ * @returns -1 for error; 0 for success.
+ **/
+_PUBLIC_ int tdb_close(struct tdb_context *tdb)
+{
+	struct tdb_context **i;
+	int ret = 0;
+
+	if (tdb->transaction) {
+		tdb_transaction_cancel(tdb);
+	}
+	tdb_trace(tdb, "tdb_close");
+
+	if (tdb->map_ptr) {
+		if (tdb->flags & TDB_INTERNAL)
+			SAFE_FREE(tdb->map_ptr);
+		else
+			tdb_munmap(tdb);
+	}
+
+	tdb_mutex_munmap(tdb);
+
+	SAFE_FREE(tdb->name);
+	if (tdb->fd != -1) {
+		ret = close(tdb->fd);
+		tdb->fd = -1;
+	}
+	SAFE_FREE(tdb->lockrecs);
+
+	/* Remove from contexts list */
+	for (i = &tdbs; *i; i = &(*i)->next) {
+		if (*i == tdb) {
+			*i = tdb->next;
+			break;
+		}
+	}
+
+#ifdef TDB_TRACE
+	close(tdb->tracefd);
+#endif
+	memset(tdb, 0, sizeof(*tdb));
+	SAFE_FREE(tdb);
+
+	return ret;
+}
+
+/* register a loging function */
+_PUBLIC_ void tdb_set_logging_function(struct tdb_context *tdb,
+                                       const struct tdb_logging_context *log_ctx)
+{
+        tdb->log = *log_ctx;
+}
+
+_PUBLIC_ void *tdb_get_logging_private(struct tdb_context *tdb)
+{
+	return tdb->log.log_private;
+}
+
+static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
+{
+#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
+	!defined(LIBREPLACE_PWRITE_NOT_REPLACED)
+	struct stat st;
+#endif
+
+	if (tdb->flags & TDB_INTERNAL) {
+		return 0; /* Nothing to do. */
+	}
+
+	if (tdb_have_extra_locks(tdb)) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
+		goto fail;
+	}
+
+	if (tdb->transaction != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
+		goto fail;
+	}
+
+/* If we have real pread & pwrite, we can skip reopen. */
+#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
+	!defined(LIBREPLACE_PWRITE_NOT_REPLACED)
+	if (tdb_munmap(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
+		goto fail;
+	}
+	if (close(tdb->fd) != 0)
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
+	tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
+	if (tdb->fd == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
+		goto fail;
+	}
+	/*
+	 * We only use st.st_dev and st.st_ino from the raw fstat()
+	 * call, everything else needs to use tdb_fstat() in order
+	 * to skip tdb->hdr_ofs!
+	 */
+	if (fstat(tdb->fd, &st) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
+		goto fail;
+	}
+	if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
+		goto fail;
+	}
+	ZERO_STRUCT(st);
+
+	/*
+	 * We had tdb_mmap(tdb) here before,
+	 * but we need to use tdb_fstat(),
+	 * which is triggered from tdb_oob() before calling tdb_mmap().
+	 * As this skips tdb->hdr_ofs.
+	 */
+	tdb->map_size = 0;
+	if (tdb_oob(tdb, 0, 1, 0) != 0) {
+		goto fail;
+	}
+#endif /* fake pread or pwrite */
+
+	/* We may still think we hold the active lock. */
+	tdb->num_lockrecs = 0;
+	SAFE_FREE(tdb->lockrecs);
+	tdb->lockrecs_array_length = 0;
+
+	if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
+		goto fail;
+	}
+
+	return 0;
+
+fail:
+	tdb_close(tdb);
+	return -1;
+}
+
+/* reopen a tdb - this can be used after a fork to ensure that we have an independent
+   seek pointer from our parent and to re-establish locks */
+_PUBLIC_ int tdb_reopen(struct tdb_context *tdb)
+{
+	bool active_lock;
+	active_lock = (tdb->flags & (TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING));
+
+	return tdb_reopen_internal(tdb, active_lock);
+}
+
+/* reopen all tdb's */
+_PUBLIC_ int tdb_reopen_all(int parent_longlived)
+{
+	struct tdb_context *tdb;
+
+	for (tdb=tdbs; tdb; tdb = tdb->next) {
+		bool active_lock;
+
+		active_lock =
+			(tdb->flags & (TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING));
+
+		/*
+		 * If the parent is longlived (ie. a
+		 * parent daemon architecture), we know
+		 * it will keep it's active lock on a
+		 * tdb opened with CLEAR_IF_FIRST. Thus
+		 * for child processes we don't have to
+		 * add an active lock. This is essential
+		 * to improve performance on systems that
+		 * keep POSIX locks as a non-scalable data
+		 * structure in the kernel.
+		 */
+		if (parent_longlived) {
+			/* Ensure no clear-if-first. */
+			active_lock = false;
+		}
+
+		if (tdb_reopen_internal(tdb, active_lock) != 0)
+			return -1;
+	}
+
+	return 0;
+}
diff --git a/lib/tdb/common/rescue.c b/lib/tdb/common/rescue.c
new file mode 100644
index 0000000..7a85ebc
--- /dev/null
+++ b/lib/tdb/common/rescue.c
@@ -0,0 +1,351 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library, rescue attempt code.
+
+   Copyright (C) Rusty Russell		   2012
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "tdb_private.h"
+#include <assert.h>
+
+
+struct found {
+	tdb_off_t head; /* 0 -> invalid. */
+	struct tdb_record rec;
+	TDB_DATA key;
+	bool in_hash;
+	bool in_free;
+};
+
+struct found_table {
+	/* As an ordered array (by head offset). */
+	struct found *arr;
+	unsigned int num, max;
+};
+
+static bool looks_like_valid_record(struct tdb_context *tdb,
+				    tdb_off_t off,
+				    const struct tdb_record *rec,
+				    TDB_DATA *key)
+{
+	unsigned int hval;
+
+	if (rec->magic != TDB_MAGIC)
+		return false;
+
+	if (rec->key_len + rec->data_len > rec->rec_len)
+		return false;
+
+	if (rec->rec_len % TDB_ALIGNMENT)
+		return false;
+
+	/* Next pointer must make some sense. */
+	if (rec->next > 0 && rec->next < TDB_DATA_START(tdb->hash_size))
+		return false;
+
+	if (tdb_oob(tdb, rec->next, sizeof(*rec), 1))
+		return false;
+
+	key->dsize = rec->key_len;
+	key->dptr = tdb_alloc_read(tdb, off + sizeof(*rec), key->dsize);
+	if (!key->dptr)
+		return false;
+
+	hval = tdb->hash_fn(key);
+	if (hval != rec->full_hash) {
+		free(key->dptr);
+		return false;
+	}
+
+	/* Caller frees up key->dptr */
+	return true;
+}
+
+static bool add_to_table(struct found_table *found,
+			 tdb_off_t off,
+			 struct tdb_record *rec,
+			 TDB_DATA key)
+{
+	if (found->num + 1 > found->max) {
+		struct found *new;
+		found->max = (found->max ? found->max * 2 : 128);
+		new = realloc(found->arr, found->max * sizeof(found->arr[0]));
+		if (!new)
+			return false;
+		found->arr = new;
+	}
+
+	found->arr[found->num].head = off;
+	found->arr[found->num].rec = *rec;
+	found->arr[found->num].key = key;
+	found->arr[found->num].in_hash = false;
+	found->arr[found->num].in_free = false;
+
+	found->num++;
+	return true;
+}
+
+static bool walk_record(struct tdb_context *tdb,
+			const struct found *f,
+			void (*walk)(TDB_DATA, TDB_DATA, void *private_data),
+			void *private_data)
+{
+	TDB_DATA data;
+
+	data.dsize = f->rec.data_len;
+	data.dptr = tdb_alloc_read(tdb,
+				   f->head + sizeof(f->rec) + f->rec.key_len,
+				   data.dsize);
+	if (!data.dptr) {
+		if (tdb->ecode == TDB_ERR_OOM)
+			return false;
+		/* I/O errors are expected. */
+		return true;
+	}
+
+	walk(f->key, data, private_data);
+	free(data.dptr);
+	return true;
+}
+
+/* First entry which has offset >= this one. */
+static unsigned int find_entry(struct found_table *found, tdb_off_t off)
+{
+	unsigned int start = 0, end = found->num;
+
+	while (start < end) {
+		/* We can't overflow here. */
+		unsigned int mid = (start + end) / 2;
+
+		if (off < found->arr[mid].head) {
+			end = mid;
+		} else if (off > found->arr[mid].head) {
+			start = mid + 1;
+		} else {
+			return mid;
+		}
+	}
+
+	assert(start == end);
+	return end;
+}
+
+static void found_in_hashchain(struct found_table *found, tdb_off_t head)
+{
+	unsigned int match;
+
+	match = find_entry(found, head);
+	if (match < found->num && found->arr[match].head == head) {
+		found->arr[match].in_hash = true;
+	}
+}
+
+static void mark_free_area(struct found_table *found, tdb_off_t head,
+			   tdb_len_t len)
+{
+	unsigned int match;
+
+	match = find_entry(found, head);
+	/* Mark everything within this free entry. */
+	while (match < found->num) {
+		if (found->arr[match].head >= head + len) {
+			break;
+		}
+		found->arr[match].in_free = true;
+		match++;
+	}
+}
+
+static int cmp_key(const void *a, const void *b)
+{
+	const struct found *fa = a, *fb = b;
+
+	if (fa->key.dsize < fb->key.dsize) {
+		return -1;
+	} else if (fa->key.dsize > fb->key.dsize) {
+		return 1;
+	}
+	return memcmp(fa->key.dptr, fb->key.dptr, fa->key.dsize);
+}
+
+static bool key_eq(TDB_DATA a, TDB_DATA b)
+{
+	return a.dsize == b.dsize
+		&& memcmp(a.dptr, b.dptr, a.dsize) == 0;
+}
+
+static void free_table(struct found_table *found)
+{
+	unsigned int i;
+
+	for (i = 0; i < found->num; i++) {
+		free(found->arr[i].key.dptr);
+	}
+	free(found->arr);
+}
+
+static void logging_suppressed(struct tdb_context *tdb,
+			       enum tdb_debug_level level, const char *fmt, ...)
+{
+}
+
+_PUBLIC_ int tdb_rescue(struct tdb_context *tdb,
+			void (*walk)(TDB_DATA, TDB_DATA, void *private_data),
+			void *private_data)
+{
+	struct found_table found = { NULL, 0, 0 };
+	tdb_off_t h, off, i;
+	tdb_log_func oldlog = tdb->log.log_fn;
+	struct tdb_record rec;
+	TDB_DATA key;
+	bool locked;
+
+	/* Read-only databases use no locking at all: it's best-effort.
+	 * We may have a write lock already, so skip that case too. */
+	if (tdb->read_only || tdb->allrecord_lock.count != 0) {
+		locked = false;
+	} else {
+		if (tdb_lockall_read(tdb) == -1)
+			return -1;
+		locked = true;
+	}
+
+	/* Make sure we know true size of the underlying file. */
+	tdb_oob(tdb, tdb->map_size, 1, 1);
+
+	/* Suppress logging, since we anticipate errors. */
+	tdb->log.log_fn = logging_suppressed;
+
+	/* Now walk entire db looking for records. */
+	for (off = TDB_DATA_START(tdb->hash_size);
+	     off < tdb->map_size;
+	     off += TDB_ALIGNMENT) {
+		if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
+					   DOCONV()) == -1)
+			continue;
+
+		if (looks_like_valid_record(tdb, off, &rec, &key)) {
+			if (!add_to_table(&found, off, &rec, key)) {
+				goto oom;
+			}
+		}
+	}
+
+	/* Walk hash chains to positive vet. */
+	for (h = 0; h < 1+tdb->hash_size; h++) {
+		bool slow_chase = false;
+		tdb_off_t slow_off = FREELIST_TOP + h*sizeof(tdb_off_t);
+
+		if (tdb_ofs_read(tdb, FREELIST_TOP + h*sizeof(tdb_off_t),
+				 &off) == -1)
+			continue;
+
+		while (off && off != slow_off) {
+			if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
+						   DOCONV()) != 0) {
+				break;
+			}
+
+			/* 0 is the free list, rest are hash chains. */
+			if (h == 0) {
+				/* Don't mark garbage as free. */
+				if (rec.magic != TDB_FREE_MAGIC) {
+					break;
+				}
+				mark_free_area(&found, off,
+					       sizeof(rec) + rec.rec_len);
+			} else {
+				found_in_hashchain(&found, off);
+			}
+
+			off = rec.next;
+
+			/* Loop detection using second pointer at half-speed */
+			if (slow_chase) {
+				/* First entry happens to be next ptr */
+				tdb_ofs_read(tdb, slow_off, &slow_off);
+			}
+			slow_chase = !slow_chase;
+		}
+	}
+
+	/* Recovery area: must be marked as free, since it often has old
+	 * records in there! */
+	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &off) == 0 && off != 0) {
+		if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
+					   DOCONV()) == 0) {
+			mark_free_area(&found, off, sizeof(rec) + rec.rec_len);
+		}
+	}
+
+	/* Now sort by key! */
+	if (found.arr != NULL) {
+		qsort(found.arr, found.num, sizeof(found.arr[0]), cmp_key);
+	}
+
+	for (i = 0; (found.arr != NULL) && i < found.num; ) {
+		unsigned int num, num_in_hash = 0;
+
+		/* How many are identical? */
+		for (num = 0; num < found.num - i; num++) {
+			if (!key_eq(found.arr[i].key, found.arr[i+num].key)) {
+				break;
+			}
+			if (found.arr[i+num].in_hash) {
+				if (!walk_record(tdb, &found.arr[i+num],
+						 walk, private_data))
+					goto oom;
+				num_in_hash++;
+			}
+		}
+		assert(num);
+
+		/* If none were in the hash, we print any not in free list. */
+		if (num_in_hash == 0) {
+			unsigned int j;
+
+			for (j = i; j < i + num; j++) {
+				if (!found.arr[j].in_free) {
+					if (!walk_record(tdb, &found.arr[j],
+							 walk, private_data))
+						goto oom;
+				}
+			}
+		}
+
+		i += num;
+	}
+
+	tdb->log.log_fn = oldlog;
+	if (locked) {
+		tdb_unlockall_read(tdb);
+	}
+	return 0;
+
+oom:
+	tdb->log.log_fn = oldlog;
+	tdb->ecode = TDB_ERR_OOM;
+	TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_rescue: failed allocating\n"));
+	free_table(&found);
+	if (locked) {
+		tdb_unlockall_read(tdb);
+	}
+	return -1;
+}
diff --git a/lib/tdb/common/summary.c b/lib/tdb/common/summary.c
new file mode 100644
index 0000000..a93eb93
--- /dev/null
+++ b/lib/tdb/common/summary.c
@@ -0,0 +1,219 @@
+ /*
+   Trivial Database: human-readable summary code
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "tdb_private.h"
+
+#define SUMMARY_FORMAT \
+	"Size of file/data: %llu/%zu\n" \
+	"Header offset/logical size: %zu/%zu\n" \
+	"Number of records: %zu\n" \
+	"Incompatible hash: %s\n" \
+	"Active/supported feature flags: 0x%08x/0x%08x\n" \
+	"Robust mutexes locking: %s\n" \
+	"Smallest/average/largest keys: %zu/%zu/%zu\n" \
+	"Smallest/average/largest data: %zu/%zu/%zu\n" \
+	"Smallest/average/largest padding: %zu/%zu/%zu\n" \
+	"Number of dead records: %zu\n" \
+	"Smallest/average/largest dead records: %zu/%zu/%zu\n" \
+	"Number of free records: %zu\n" \
+	"Smallest/average/largest free records: %zu/%zu/%zu\n" \
+	"Number of hash chains: %zu\n" \
+	"Smallest/average/largest hash chains: %zu/%zu/%zu\n" \
+	"Number of uncoalesced records: %zu\n" \
+	"Smallest/average/largest uncoalesced runs: %zu/%zu/%zu\n" \
+	"Percentage keys/data/padding/free/dead/rechdrs&tailers/hashes: %.0f/%.0f/%.0f/%.0f/%.0f/%.0f/%.0f\n"
+
+/* We don't use tally module, to keep upstream happy. */
+struct tally {
+	size_t min, max, total;
+	size_t num;
+};
+
+static void tally_init(struct tally *tally)
+{
+	tally->total = 0;
+	tally->num = 0;
+	tally->min = tally->max = 0;
+}
+
+static void tally_add(struct tally *tally, size_t len)
+{
+	if (tally->num == 0)
+		tally->max = tally->min = len;
+	else if (len > tally->max)
+		tally->max = len;
+	else if (len < tally->min)
+		tally->min = len;
+	tally->num++;
+	tally->total += len;
+}
+
+static size_t tally_mean(const struct tally *tally)
+{
+	if (!tally->num)
+		return 0;
+	return tally->total / tally->num;
+}
+
+static size_t get_hash_length(struct tdb_context *tdb, unsigned int i)
+{
+	tdb_off_t rec_ptr;
+	struct tdb_chainwalk_ctx chainwalk;
+	size_t count = 0;
+
+	if (tdb_ofs_read(tdb, TDB_HASH_TOP(i), &rec_ptr) == -1)
+		return 0;
+
+	tdb_chainwalk_init(&chainwalk, rec_ptr);
+
+	/* keep looking until we find the right record */
+	while (rec_ptr) {
+		struct tdb_record r;
+		bool ok;
+		++count;
+		if (tdb_rec_read(tdb, rec_ptr, &r) == -1)
+			return 0;
+		rec_ptr = r.next;
+		ok = tdb_chainwalk_check(tdb, &chainwalk, rec_ptr);
+		if (!ok) {
+			return SIZE_MAX;
+		}
+	}
+	return count;
+}
+
+_PUBLIC_ char *tdb_summary(struct tdb_context *tdb)
+{
+	off_t file_size;
+	tdb_off_t off, rec_off;
+	struct tally freet, keys, data, dead, extra, hashval, uncoal;
+	struct tdb_record rec;
+	char *ret = NULL;
+	bool locked;
+	size_t unc = 0;
+	int len;
+	struct tdb_record recovery;
+
+	/* Read-only databases use no locking at all: it's best-effort.
+	 * We may have a write lock already, so skip that case too. */
+	if (tdb->read_only || tdb->allrecord_lock.count != 0) {
+		locked = false;
+	} else {
+		if (tdb_lockall_read(tdb) == -1)
+			return NULL;
+		locked = true;
+	}
+
+	if (tdb_recovery_area(tdb, tdb->methods, &rec_off, &recovery) != 0) {
+		goto unlock;
+	}
+
+	tally_init(&freet);
+	tally_init(&keys);
+	tally_init(&data);
+	tally_init(&dead);
+	tally_init(&extra);
+	tally_init(&hashval);
+	tally_init(&uncoal);
+
+	for (off = TDB_DATA_START(tdb->hash_size);
+	     off < tdb->map_size - 1;
+	     off += sizeof(rec) + rec.rec_len) {
+		if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
+					   DOCONV()) == -1)
+			goto unlock;
+		switch (rec.magic) {
+		case TDB_MAGIC:
+			tally_add(&keys, rec.key_len);
+			tally_add(&data, rec.data_len);
+			tally_add(&extra, rec.rec_len - (rec.key_len
+							 + rec.data_len));
+			if (unc > 1)
+				tally_add(&uncoal, unc - 1);
+			unc = 0;
+			break;
+		case TDB_FREE_MAGIC:
+			tally_add(&freet, rec.rec_len);
+			unc++;
+			break;
+		/* If we crash after ftruncate, we can get zeroes or fill. */
+		case TDB_RECOVERY_INVALID_MAGIC:
+		case 0x42424242:
+			unc++;
+			/* If it's a valid recovery, we can trust rec_len. */
+			if (off != rec_off) {
+				rec.rec_len = tdb_dead_space(tdb, off)
+					- sizeof(rec);
+			}
+
+			FALL_THROUGH;
+		case TDB_DEAD_MAGIC:
+			tally_add(&dead, rec.rec_len);
+			break;
+		default:
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "Unexpected record magic 0x%x at offset %u\n",
+				 rec.magic, off));
+			goto unlock;
+		}
+	}
+	if (unc > 1)
+		tally_add(&uncoal, unc - 1);
+
+	for (off = 0; off < tdb->hash_size; off++)
+		tally_add(&hashval, get_hash_length(tdb, off));
+
+	file_size = tdb->hdr_ofs + tdb->map_size;
+
+	len = asprintf(&ret, SUMMARY_FORMAT,
+		 (unsigned long long)file_size, keys.total+data.total,
+		 (size_t)tdb->hdr_ofs, (size_t)tdb->map_size,
+		 keys.num,
+		 (tdb->hash_fn == tdb_jenkins_hash)?"yes":"no",
+		 (unsigned)tdb->feature_flags, TDB_SUPPORTED_FEATURE_FLAGS,
+		 (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX)?"yes":"no",
+		 keys.min, tally_mean(&keys), keys.max,
+		 data.min, tally_mean(&data), data.max,
+		 extra.min, tally_mean(&extra), extra.max,
+		 dead.num,
+		 dead.min, tally_mean(&dead), dead.max,
+		 freet.num,
+		 freet.min, tally_mean(&freet), freet.max,
+		 hashval.num,
+		 hashval.min, tally_mean(&hashval), hashval.max,
+		 uncoal.total,
+		 uncoal.min, tally_mean(&uncoal), uncoal.max,
+		 keys.total * 100.0 / file_size,
+		 data.total * 100.0 / file_size,
+		 extra.total * 100.0 / file_size,
+		 freet.total * 100.0 / file_size,
+		 dead.total * 100.0 / file_size,
+		 (keys.num + freet.num + dead.num)
+		 * (sizeof(struct tdb_record) + sizeof(uint32_t))
+		 * 100.0 / file_size,
+		 tdb->hash_size * sizeof(tdb_off_t)
+		 * 100.0 / file_size);
+	if (len == -1) {
+		goto unlock;
+	}
+
+unlock:
+	if (locked) {
+		tdb_unlockall_read(tdb);
+	}
+	return ret;
+}
diff --git a/lib/tdb/common/tdb.c b/lib/tdb/common/tdb.c
new file mode 100644
index 0000000..de829bb
--- /dev/null
+++ b/lib/tdb/common/tdb.c
@@ -0,0 +1,1348 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+_PUBLIC_ TDB_DATA tdb_null;
+
+/*
+  non-blocking increment of the tdb sequence number if the tdb has been opened using
+  the TDB_SEQNUM flag
+*/
+_PUBLIC_ void tdb_increment_seqnum_nonblock(struct tdb_context *tdb)
+{
+	tdb_off_t seqnum=0;
+
+	if (!(tdb->flags & TDB_SEQNUM)) {
+		return;
+	}
+
+	/* we ignore errors from this, as we have no sane way of
+	   dealing with them.
+	*/
+	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+	seqnum++;
+	tdb_ofs_write(tdb, TDB_SEQNUM_OFS, &seqnum);
+}
+
+/*
+  increment the tdb sequence number if the tdb has been opened using
+  the TDB_SEQNUM flag
+*/
+static void tdb_increment_seqnum(struct tdb_context *tdb)
+{
+	if (!(tdb->flags & TDB_SEQNUM)) {
+		return;
+	}
+
+	if (tdb->transaction != NULL) {
+		tdb_increment_seqnum_nonblock(tdb);
+		return;
+	}
+
+#if defined(HAVE___ATOMIC_ADD_FETCH) && defined(HAVE___ATOMIC_ADD_LOAD)
+	if (tdb->map_ptr != NULL) {
+		uint32_t *pseqnum = (uint32_t *)(
+			TDB_SEQNUM_OFS + (char *)tdb->map_ptr);
+		__atomic_add_fetch(pseqnum, 1, __ATOMIC_SEQ_CST);
+		return;
+	}
+#endif
+
+	if (tdb_nest_lock(tdb, TDB_SEQNUM_OFS, F_WRLCK,
+			  TDB_LOCK_WAIT|TDB_LOCK_PROBE) != 0) {
+		return;
+	}
+
+	tdb_increment_seqnum_nonblock(tdb);
+
+	tdb_nest_unlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, false);
+}
+
+static int tdb_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
+{
+	return memcmp(data.dptr, key.dptr, data.dsize);
+}
+
+void tdb_chainwalk_init(struct tdb_chainwalk_ctx *ctx, tdb_off_t ptr)
+{
+	*ctx = (struct tdb_chainwalk_ctx) { .slow_ptr = ptr };
+}
+
+bool tdb_chainwalk_check(struct tdb_context *tdb,
+			 struct tdb_chainwalk_ctx *ctx,
+			 tdb_off_t next_ptr)
+{
+	int ret;
+
+	if (ctx->slow_chase) {
+		ret = tdb_ofs_read(tdb, ctx->slow_ptr, &ctx->slow_ptr);
+		if (ret == -1) {
+			return false;
+		}
+	}
+	ctx->slow_chase = !ctx->slow_chase;
+
+	if (next_ptr == ctx->slow_ptr) {
+		tdb->ecode = TDB_ERR_CORRUPT;
+		TDB_LOG((tdb, TDB_DEBUG_ERROR,
+			 "tdb_chainwalk_check: circular chain\n"));
+		return false;
+	}
+
+	return true;
+}
+
+/* Returns 0 on fail.  On success, return offset of record, and fills
+   in rec */
+static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, uint32_t hash,
+			struct tdb_record *r)
+{
+	tdb_off_t rec_ptr;
+	struct tdb_chainwalk_ctx chainwalk;
+
+	/* read in the hash top */
+	if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
+		return 0;
+
+	tdb_chainwalk_init(&chainwalk, rec_ptr);
+
+	/* keep looking until we find the right record */
+	while (rec_ptr) {
+		bool ok;
+
+		if (tdb_rec_read(tdb, rec_ptr, r) == -1)
+			return 0;
+
+		if (!TDB_DEAD(r) && hash==r->full_hash
+		    && key.dsize==r->key_len
+		    && tdb_parse_data(tdb, key, rec_ptr + sizeof(*r),
+				      r->key_len, tdb_key_compare,
+				      NULL) == 0) {
+			return rec_ptr;
+		}
+		rec_ptr = r->next;
+
+		ok = tdb_chainwalk_check(tdb, &chainwalk, rec_ptr);
+		if (!ok) {
+			return 0;
+		}
+	}
+	tdb->ecode = TDB_ERR_NOEXIST;
+	return 0;
+}
+
+/* As tdb_find, but if you succeed, keep the lock */
+tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
+			   struct tdb_record *rec)
+{
+	uint32_t rec_ptr;
+
+	if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
+		return 0;
+	if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
+		tdb_unlock(tdb, BUCKET(hash), locktype);
+	return rec_ptr;
+}
+
+static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key);
+
+struct tdb_update_hash_state {
+	const TDB_DATA *dbufs;
+	int num_dbufs;
+	tdb_len_t dbufs_len;
+};
+
+static int tdb_update_hash_cmp(TDB_DATA key, TDB_DATA data, void *private_data)
+{
+	struct tdb_update_hash_state *state = private_data;
+	unsigned char *dptr = data.dptr;
+	int i;
+
+	if (state->dbufs_len != data.dsize) {
+		return -1;
+	}
+
+	for (i=0; i<state->num_dbufs; i++) {
+		TDB_DATA dbuf = state->dbufs[i];
+		if( dbuf.dsize > 0) {
+			int ret;
+			ret = memcmp(dptr, dbuf.dptr, dbuf.dsize);
+			if (ret != 0) {
+				return -1;
+			}
+			dptr += dbuf.dsize;
+		}
+	}
+
+	return 0;
+}
+
+/* update an entry in place - this only works if the new data size
+   is <= the old data size and the key exists.
+   on failure return -1.
+*/
+static int tdb_update_hash(struct tdb_context *tdb, TDB_DATA key,
+			   uint32_t hash,
+			   const TDB_DATA *dbufs, int num_dbufs,
+			   tdb_len_t dbufs_len)
+{
+	struct tdb_record rec;
+	tdb_off_t rec_ptr, ofs;
+	int i;
+
+	/* find entry */
+	if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
+		return -1;
+
+	/* it could be an exact duplicate of what is there - this is
+	 * surprisingly common (eg. with a ldb re-index). */
+	if (rec.data_len == dbufs_len) {
+		struct tdb_update_hash_state state = {
+			.dbufs = dbufs, .num_dbufs = num_dbufs,
+			.dbufs_len = dbufs_len
+		};
+		int ret;
+
+		ret = tdb_parse_record(tdb, key, tdb_update_hash_cmp, &state);
+		if (ret == 0) {
+			return 0;
+		}
+	}
+
+	/* must be long enough key, data and tailer */
+	if (rec.rec_len < key.dsize + dbufs_len + sizeof(tdb_off_t)) {
+		tdb->ecode = TDB_SUCCESS; /* Not really an error */
+		return -1;
+	}
+
+	ofs = rec_ptr + sizeof(rec) + rec.key_len;
+
+	for (i=0; i<num_dbufs; i++) {
+		TDB_DATA dbuf = dbufs[i];
+		int ret;
+
+		ret = tdb->methods->tdb_write(tdb, ofs, dbuf.dptr, dbuf.dsize);
+		if (ret == -1) {
+			return -1;
+		}
+		ofs += dbuf.dsize;
+	}
+
+	if (dbufs_len != rec.data_len) {
+		/* update size */
+		rec.data_len = dbufs_len;
+		return tdb_rec_write(tdb, rec_ptr, &rec);
+	}
+
+	return 0;
+}
+
+/* find an entry in the database given a key */
+/* If an entry doesn't exist tdb_err will be set to
+ * TDB_ERR_NOEXIST. If a key has no data attached
+ * then the TDB_DATA will have zero length but
+ * a non-zero pointer
+ */
+static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
+{
+	tdb_off_t rec_ptr;
+	struct tdb_record rec;
+	TDB_DATA ret;
+	uint32_t hash;
+
+	/* find which hash bucket it is in */
+	hash = tdb->hash_fn(&key);
+	if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
+		return tdb_null;
+
+	ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+				  rec.data_len);
+	ret.dsize = rec.data_len;
+	tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
+	return ret;
+}
+
+_PUBLIC_ TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
+{
+	TDB_DATA ret = _tdb_fetch(tdb, key);
+
+	tdb_trace_1rec_retrec(tdb, "tdb_fetch", key, ret);
+	return ret;
+}
+
+/*
+ * Find an entry in the database and hand the record's data to a parsing
+ * function. The parsing function is executed under the chain read lock, so it
+ * should be fast and should not block on other syscalls.
+ *
+ * DON'T CALL OTHER TDB CALLS FROM THE PARSER, THIS MIGHT LEAD TO SEGFAULTS.
+ *
+ * For mmapped tdb's that do not have a transaction open it points the parsing
+ * function directly at the mmap area, it avoids the malloc/memcpy in this
+ * case. If a transaction is open or no mmap is available, it has to do
+ * malloc/read/parse/free.
+ *
+ * This is interesting for all readers of potentially large data structures in
+ * the tdb records, ldb indexes being one example.
+ *
+ * Return -1 if the record was not found.
+ */
+
+_PUBLIC_ int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
+		     int (*parser)(TDB_DATA key, TDB_DATA data,
+				   void *private_data),
+		     void *private_data)
+{
+	tdb_off_t rec_ptr;
+	struct tdb_record rec;
+	int ret;
+	uint32_t hash;
+
+	/* find which hash bucket it is in */
+	hash = tdb->hash_fn(&key);
+
+	if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
+		/* record not found */
+		tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, -1);
+		tdb->ecode = TDB_ERR_NOEXIST;
+		return -1;
+	}
+	tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, 0);
+
+	ret = tdb_parse_data(tdb, key, rec_ptr + sizeof(rec) + rec.key_len,
+			     rec.data_len, parser, private_data);
+
+	tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
+
+	return ret;
+}
+
+/* check if an entry in the database exists
+
+   note that 1 is returned if the key is found and 0 is returned if not found
+   this doesn't match the conventions in the rest of this module, but is
+   compatible with gdbm
+*/
+static int tdb_exists_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
+{
+	struct tdb_record rec;
+
+	if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0)
+		return 0;
+	tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
+	return 1;
+}
+
+_PUBLIC_ int tdb_exists(struct tdb_context *tdb, TDB_DATA key)
+{
+	uint32_t hash = tdb->hash_fn(&key);
+	int ret;
+
+	ret = tdb_exists_hash(tdb, key, hash);
+	tdb_trace_1rec_ret(tdb, "tdb_exists", key, ret);
+	return ret;
+}
+
+/*
+ * Move a dead record to the freelist. The hash chain and freelist
+ * must be locked.
+ */
+static int tdb_del_dead(struct tdb_context *tdb,
+			uint32_t last_ptr,
+			uint32_t rec_ptr,
+			struct tdb_record *rec,
+			bool *deleted)
+{
+	int ret;
+
+	ret = tdb_write_lock_record(tdb, rec_ptr);
+	if (ret == -1) {
+		/* Someone traversing here: Just leave it dead */
+		return 0;
+	}
+	ret = tdb_write_unlock_record(tdb, rec_ptr);
+	if (ret == -1) {
+		return -1;
+	}
+	ret = tdb_ofs_write(tdb, last_ptr, &rec->next);
+	if (ret == -1) {
+		return -1;
+	}
+
+	*deleted = true;
+
+	ret = tdb_free(tdb, rec_ptr, rec);
+	return ret;
+}
+
+/*
+ * Walk the hash chain and leave tdb->max_dead_records around. Move
+ * the rest of dead records to the freelist.
+ */
+int tdb_trim_dead(struct tdb_context *tdb, uint32_t hash)
+{
+	struct tdb_chainwalk_ctx chainwalk;
+	struct tdb_record rec;
+	tdb_off_t last_ptr, rec_ptr;
+	bool locked_freelist = false;
+	int num_dead = 0;
+	int ret;
+
+	last_ptr = TDB_HASH_TOP(hash);
+
+	/*
+	 * Init chainwalk with the pointer to the hash top. It might
+	 * be that the very first record in the chain is a dead one
+	 * that we have to delete.
+	 */
+	tdb_chainwalk_init(&chainwalk, last_ptr);
+
+	ret = tdb_ofs_read(tdb, last_ptr, &rec_ptr);
+	if (ret == -1) {
+		return -1;
+	}
+
+	while (rec_ptr != 0) {
+		bool deleted = false;
+		uint32_t next;
+
+		ret = tdb_rec_read(tdb, rec_ptr, &rec);
+		if (ret == -1) {
+			goto fail;
+		}
+
+		/*
+		 * Make a copy of rec.next: Further down we might
+		 * delete and put the record on the freelist. Make
+		 * sure that modifications in that code path can't
+		 * break the chainwalk here.
+		 */
+		next = rec.next;
+
+		if (rec.magic == TDB_DEAD_MAGIC) {
+			num_dead += 1;
+
+			if (num_dead > tdb->max_dead_records) {
+
+				if (!locked_freelist) {
+					/*
+					 * Lock the freelist only if
+					 * it's really required.
+					 */
+					ret = tdb_lock(tdb, -1, F_WRLCK);
+					if (ret == -1) {
+						goto fail;
+					};
+					locked_freelist = true;
+				}
+
+				ret = tdb_del_dead(
+					tdb,
+					last_ptr,
+					rec_ptr,
+					&rec,
+					&deleted);
+
+				if (ret == -1) {
+					goto fail;
+				}
+			}
+		}
+
+		/*
+		 * Don't do the chainwalk check if "rec_ptr" was
+		 * deleted. We reduced the chain, and the chainwalk
+		 * check might catch up early. Imagine a valid chain
+		 * with just dead records: We never can bump the
+		 * "slow" pointer in chainwalk_check, as there isn't
+		 * anything left to jump to and compare.
+		 */
+		if (!deleted) {
+			bool ok;
+
+			last_ptr = rec_ptr;
+
+			ok = tdb_chainwalk_check(tdb, &chainwalk, next);
+			if (!ok) {
+				ret = -1;
+				goto fail;
+			}
+		}
+		rec_ptr = next;
+	}
+	ret = 0;
+fail:
+	if (locked_freelist) {
+		tdb_unlock(tdb, -1, F_WRLCK);
+	}
+	return ret;
+}
+
+/* delete an entry in the database given a key */
+static int tdb_delete_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
+{
+	tdb_off_t rec_ptr;
+	struct tdb_record rec;
+	int ret;
+
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		return -1;
+	}
+
+	rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK, &rec);
+	if (rec_ptr == 0) {
+		return -1;
+	}
+
+	/*
+	 * Mark the record dead
+	 */
+	rec.magic = TDB_DEAD_MAGIC;
+	ret = tdb_rec_write(tdb, rec_ptr, &rec);
+	if (ret == -1) {
+		goto done;
+	}
+
+	tdb_increment_seqnum(tdb);
+
+	ret = tdb_trim_dead(tdb, hash);
+done:
+	if (tdb_unlock(tdb, BUCKET(hash), F_WRLCK) != 0)
+		TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_delete: WARNING tdb_unlock failed!\n"));
+	return ret;
+}
+
+_PUBLIC_ int tdb_delete(struct tdb_context *tdb, TDB_DATA key)
+{
+	uint32_t hash = tdb->hash_fn(&key);
+	int ret;
+
+	ret = tdb_delete_hash(tdb, key, hash);
+	tdb_trace_1rec_ret(tdb, "tdb_delete", key, ret);
+	return ret;
+}
+
+/*
+ * See if we have a dead record around with enough space
+ */
+tdb_off_t tdb_find_dead(struct tdb_context *tdb, uint32_t hash,
+			struct tdb_record *r, tdb_len_t length,
+			tdb_off_t *p_last_ptr)
+{
+	tdb_off_t rec_ptr, last_ptr;
+	struct tdb_chainwalk_ctx chainwalk;
+	tdb_off_t best_rec_ptr = 0;
+	tdb_off_t best_last_ptr = 0;
+	struct tdb_record best = { .rec_len = UINT32_MAX };
+
+	length += sizeof(tdb_off_t); /* tailer */
+
+	last_ptr = TDB_HASH_TOP(hash);
+
+	/* read in the hash top */
+	if (tdb_ofs_read(tdb, last_ptr, &rec_ptr) == -1)
+		return 0;
+
+	tdb_chainwalk_init(&chainwalk, rec_ptr);
+
+	/* keep looking until we find the right record */
+	while (rec_ptr) {
+		bool ok;
+
+		if (tdb_rec_read(tdb, rec_ptr, r) == -1)
+			return 0;
+
+		if (TDB_DEAD(r) && (r->rec_len >= length) &&
+		    (r->rec_len < best.rec_len)) {
+			best_rec_ptr = rec_ptr;
+			best_last_ptr = last_ptr;
+			best = *r;
+		}
+		last_ptr = rec_ptr;
+		rec_ptr = r->next;
+
+		ok = tdb_chainwalk_check(tdb, &chainwalk, rec_ptr);
+		if (!ok) {
+			return 0;
+		}
+	}
+
+	if (best.rec_len == UINT32_MAX) {
+		return 0;
+	}
+
+	*r = best;
+	*p_last_ptr = best_last_ptr;
+	return best_rec_ptr;
+}
+
+static int _tdb_storev(struct tdb_context *tdb, TDB_DATA key,
+		       const TDB_DATA *dbufs, int num_dbufs,
+		       int flag, uint32_t hash)
+{
+	struct tdb_record rec;
+	tdb_off_t rec_ptr, ofs;
+	tdb_len_t rec_len, dbufs_len;
+	int i;
+	int ret = -1;
+
+	dbufs_len = 0;
+
+	for (i=0; i<num_dbufs; i++) {
+		size_t dsize = dbufs[i].dsize;
+
+		if ((dsize != 0) && (dbufs[i].dptr == NULL)) {
+			tdb->ecode = TDB_ERR_EINVAL;
+			goto fail;
+		}
+
+		dbufs_len += dsize;
+		if (dbufs_len < dsize) {
+			tdb->ecode = TDB_ERR_OOM;
+			goto fail;
+		}
+	}
+
+	rec_len = key.dsize + dbufs_len;
+	if ((rec_len < key.dsize) || (rec_len < dbufs_len)) {
+		tdb->ecode = TDB_ERR_OOM;
+		goto fail;
+	}
+
+	/* check for it existing, on insert. */
+	if (flag == TDB_INSERT) {
+		if (tdb_exists_hash(tdb, key, hash)) {
+			tdb->ecode = TDB_ERR_EXISTS;
+			goto fail;
+		}
+	} else {
+		/* first try in-place update, on modify or replace. */
+		if (tdb_update_hash(tdb, key, hash, dbufs, num_dbufs,
+				    dbufs_len) == 0) {
+			goto done;
+		}
+		if (tdb->ecode == TDB_ERR_NOEXIST &&
+		    flag == TDB_MODIFY) {
+			/* if the record doesn't exist and we are in TDB_MODIFY mode then
+			 we should fail the store */
+			goto fail;
+		}
+	}
+	/* reset the error code potentially set by the tdb_update_hash() */
+	tdb->ecode = TDB_SUCCESS;
+
+	/* delete any existing record - if it doesn't exist we don't
+           care.  Doing this first reduces fragmentation, and avoids
+           coalescing with `allocated' block before it's updated. */
+	if (flag != TDB_INSERT)
+		tdb_delete_hash(tdb, key, hash);
+
+	/* we have to allocate some space */
+	rec_ptr = tdb_allocate(tdb, hash, rec_len, &rec);
+
+	if (rec_ptr == 0) {
+		goto fail;
+	}
+
+	/* Read hash top into next ptr */
+	if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
+		goto fail;
+
+	rec.key_len = key.dsize;
+	rec.data_len = dbufs_len;
+	rec.full_hash = hash;
+	rec.magic = TDB_MAGIC;
+
+	ofs = rec_ptr;
+
+	/* write out and point the top of the hash chain at it */
+	ret = tdb_rec_write(tdb, ofs, &rec);
+	if (ret == -1) {
+		goto fail;
+	}
+	ofs += sizeof(rec);
+
+	ret = tdb->methods->tdb_write(tdb, ofs, key.dptr, key.dsize);
+	if (ret == -1) {
+		goto fail;
+	}
+	ofs += key.dsize;
+
+	for (i=0; i<num_dbufs; i++) {
+		if (dbufs[i].dsize == 0) {
+			continue;
+		}
+
+		ret = tdb->methods->tdb_write(tdb, ofs, dbufs[i].dptr,
+					      dbufs[i].dsize);
+		if (ret == -1) {
+			goto fail;
+		}
+		ofs += dbufs[i].dsize;
+	}
+
+	ret = tdb_ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr);
+	if (ret == -1) {
+		/* Need to tdb_unallocate() here */
+		goto fail;
+	}
+
+ done:
+	ret = 0;
+ fail:
+	if (ret == 0) {
+		tdb_increment_seqnum(tdb);
+	}
+	return ret;
+}
+
+static int _tdb_store(struct tdb_context *tdb, TDB_DATA key,
+		      TDB_DATA dbuf, int flag, uint32_t hash)
+{
+	return _tdb_storev(tdb, key, &dbuf, 1, flag, hash);
+}
+
+/* store an element in the database, replacing any existing element
+   with the same key
+
+   return 0 on success, -1 on failure
+*/
+_PUBLIC_ int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
+{
+	uint32_t hash;
+	int ret;
+
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag, -1);
+		return -1;
+	}
+
+	/* find which hash bucket it is in */
+	hash = tdb->hash_fn(&key);
+	if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
+		return -1;
+
+	ret = _tdb_store(tdb, key, dbuf, flag, hash);
+	tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag, ret);
+	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+	return ret;
+}
+
+_PUBLIC_ int tdb_storev(struct tdb_context *tdb, TDB_DATA key,
+			const TDB_DATA *dbufs, int num_dbufs, int flag)
+{
+	uint32_t hash;
+	int ret;
+
+	if (tdb->read_only || tdb->traverse_read) {
+		tdb->ecode = TDB_ERR_RDONLY;
+		tdb_trace_1plusn_rec_flag_ret(tdb, "tdb_storev", key,
+					      dbufs, num_dbufs, flag, -1);
+		return -1;
+	}
+
+	/* find which hash bucket it is in */
+	hash = tdb->hash_fn(&key);
+	if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
+		return -1;
+
+	ret = _tdb_storev(tdb, key, dbufs, num_dbufs, flag, hash);
+	tdb_trace_1plusn_rec_flag_ret(tdb, "tdb_storev", key,
+				      dbufs, num_dbufs, flag, -1);
+	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+	return ret;
+}
+
+/* Append to an entry. Create if not exist. */
+_PUBLIC_ int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
+{
+	uint32_t hash;
+	TDB_DATA dbufs[2];
+	int ret = -1;
+
+	/* find which hash bucket it is in */
+	hash = tdb->hash_fn(&key);
+	if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
+		return -1;
+
+	dbufs[0] = _tdb_fetch(tdb, key);
+	dbufs[1] = new_dbuf;
+
+	ret = _tdb_storev(tdb, key, dbufs, 2, 0, hash);
+	tdb_trace_2rec_retrec(tdb, "tdb_append", key, dbufs[0], dbufs[1]);
+
+	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+	SAFE_FREE(dbufs[0].dptr);
+	return ret;
+}
+
+
+/*
+  return the name of the current tdb file
+  useful for external logging functions
+*/
+_PUBLIC_ const char *tdb_name(struct tdb_context *tdb)
+{
+	return tdb->name;
+}
+
+/*
+  return the underlying file descriptor being used by tdb, or -1
+  useful for external routines that want to check the device/inode
+  of the fd
+*/
+_PUBLIC_ int tdb_fd(struct tdb_context *tdb)
+{
+	return tdb->fd;
+}
+
+/*
+  return the current logging function
+  useful for external tdb routines that wish to log tdb errors
+*/
+_PUBLIC_ tdb_log_func tdb_log_fn(struct tdb_context *tdb)
+{
+	return tdb->log.log_fn;
+}
+
+
+/*
+  get the tdb sequence number. Only makes sense if the writers opened
+  with TDB_SEQNUM set. Note that this sequence number will wrap quite
+  quickly, so it should only be used for a 'has something changed'
+  test, not for code that relies on the count of the number of changes
+  made. If you want a counter then use a tdb record.
+
+  The aim of this sequence number is to allow for a very lightweight
+  test of a possible tdb change.
+*/
+_PUBLIC_ int tdb_get_seqnum(struct tdb_context *tdb)
+{
+	tdb_off_t seqnum=0;
+
+	if (tdb->transaction != NULL) {
+		tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+		return seqnum;
+	}
+
+#if defined(HAVE___ATOMIC_ADD_FETCH) && defined(HAVE___ATOMIC_ADD_LOAD)
+	if (tdb->map_ptr != NULL) {
+		uint32_t *pseqnum = (uint32_t *)(
+			TDB_SEQNUM_OFS + (char *)tdb->map_ptr);
+		uint32_t ret;
+		__atomic_load(pseqnum, &ret,__ATOMIC_SEQ_CST);
+		return ret;
+	}
+#endif
+
+	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+	return seqnum;
+}
+
+_PUBLIC_ int tdb_hash_size(struct tdb_context *tdb)
+{
+	return tdb->hash_size;
+}
+
+_PUBLIC_ size_t tdb_map_size(struct tdb_context *tdb)
+{
+	return tdb->map_size;
+}
+
+_PUBLIC_ int tdb_get_flags(struct tdb_context *tdb)
+{
+	return tdb->flags;
+}
+
+_PUBLIC_ void tdb_add_flags(struct tdb_context *tdb, unsigned flags)
+{
+	if ((flags & TDB_ALLOW_NESTING) &&
+	    (flags & TDB_DISALLOW_NESTING)) {
+		tdb->ecode = TDB_ERR_NESTING;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_add_flags: "
+			"allow_nesting and disallow_nesting are not allowed together!"));
+		return;
+	}
+
+	if (flags & TDB_ALLOW_NESTING) {
+		tdb->flags &= ~TDB_DISALLOW_NESTING;
+	}
+	if (flags & TDB_DISALLOW_NESTING) {
+		tdb->flags &= ~TDB_ALLOW_NESTING;
+	}
+
+	tdb->flags |= flags;
+}
+
+_PUBLIC_ void tdb_remove_flags(struct tdb_context *tdb, unsigned flags)
+{
+	if ((flags & TDB_ALLOW_NESTING) &&
+	    (flags & TDB_DISALLOW_NESTING)) {
+		tdb->ecode = TDB_ERR_NESTING;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_remove_flags: "
+			"allow_nesting and disallow_nesting are not allowed together!"));
+		return;
+	}
+
+	if ((flags & TDB_NOLOCK) &&
+	    (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) &&
+	    (tdb->mutexes == NULL)) {
+		tdb->ecode = TDB_ERR_LOCK;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_remove_flags: "
+			 "Can not remove NOLOCK flag on mutexed databases"));
+		return;
+	}
+
+	if (flags & TDB_ALLOW_NESTING) {
+		tdb->flags |= TDB_DISALLOW_NESTING;
+	}
+	if (flags & TDB_DISALLOW_NESTING) {
+		tdb->flags |= TDB_ALLOW_NESTING;
+	}
+
+	tdb->flags &= ~flags;
+}
+
+
+/*
+  enable sequence number handling on an open tdb
+*/
+_PUBLIC_ void tdb_enable_seqnum(struct tdb_context *tdb)
+{
+	tdb->flags |= TDB_SEQNUM;
+}
+
+
+/*
+  add a region of the file to the freelist. Length is the size of the region in bytes,
+  which includes the free list header that needs to be added
+ */
+static int tdb_free_region(struct tdb_context *tdb, tdb_off_t offset, ssize_t length)
+{
+	struct tdb_record rec;
+	if (length <= sizeof(rec)) {
+		/* the region is not worth adding */
+		return 0;
+	}
+	if (length + offset > tdb->map_size) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_free_region: adding region beyond end of file\n"));
+		return -1;
+	}
+	memset(&rec,'\0',sizeof(rec));
+	rec.rec_len = length - sizeof(rec);
+	if (tdb_free(tdb, offset, &rec) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_free_region: failed to add free record\n"));
+		return -1;
+	}
+	return 0;
+}
+
+/*
+  wipe the entire database, deleting all records. This can be done
+  very fast by using a allrecord lock. The entire data portion of the
+  file becomes a single entry in the freelist.
+
+  This code carefully steps around the recovery area, leaving it alone
+ */
+_PUBLIC_ int tdb_wipe_all(struct tdb_context *tdb)
+{
+	uint32_t i;
+	tdb_off_t offset = 0;
+	ssize_t data_len;
+	tdb_off_t recovery_head;
+	tdb_len_t recovery_size = 0;
+
+	if (tdb_lockall(tdb) != 0) {
+		return -1;
+	}
+
+	tdb_trace(tdb, "tdb_wipe_all");
+
+	/* see if the tdb has a recovery area, and remember its size
+	   if so. We don't want to lose this as otherwise each
+	   tdb_wipe_all() in a transaction will increase the size of
+	   the tdb by the size of the recovery area */
+	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_wipe_all: failed to read recovery head\n"));
+		goto failed;
+	}
+
+	if (recovery_head != 0) {
+		struct tdb_record rec;
+		if (tdb->methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_wipe_all: failed to read recovery record\n"));
+			return -1;
+		}
+		recovery_size = rec.rec_len + sizeof(rec);
+	}
+
+	/* wipe the hashes */
+	for (i=0;i<tdb->hash_size;i++) {
+		if (tdb_ofs_write(tdb, TDB_HASH_TOP(i), &offset) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write hash %d\n", i));
+			goto failed;
+		}
+	}
+
+	/* wipe the freelist */
+	if (tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write freelist\n"));
+		goto failed;
+	}
+
+	/* add all the rest of the file to the freelist, possibly leaving a gap
+	   for the recovery area */
+	if (recovery_size == 0) {
+		/* the simple case - the whole file can be used as a freelist */
+		data_len = (tdb->map_size - TDB_DATA_START(tdb->hash_size));
+		if (tdb_free_region(tdb, TDB_DATA_START(tdb->hash_size), data_len) != 0) {
+			goto failed;
+		}
+	} else {
+		/* we need to add two freelist entries - one on either
+		   side of the recovery area
+
+		   Note that we cannot shift the recovery area during
+		   this operation. Only the transaction.c code may
+		   move the recovery area or we risk subtle data
+		   corruption
+		*/
+		data_len = (recovery_head - TDB_DATA_START(tdb->hash_size));
+		if (tdb_free_region(tdb, TDB_DATA_START(tdb->hash_size), data_len) != 0) {
+			goto failed;
+		}
+		/* and the 2nd free list entry after the recovery area - if any */
+		data_len = tdb->map_size - (recovery_head+recovery_size);
+		if (tdb_free_region(tdb, recovery_head+recovery_size, data_len) != 0) {
+			goto failed;
+		}
+	}
+
+	tdb_increment_seqnum_nonblock(tdb);
+
+	if (tdb_unlockall(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to unlock\n"));
+		goto failed;
+	}
+
+	return 0;
+
+failed:
+	tdb_unlockall(tdb);
+	return -1;
+}
+
+struct traverse_state {
+	bool error;
+	struct tdb_context *dest_db;
+};
+
+/*
+  traverse function for repacking
+ */
+static int repack_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private_data)
+{
+	struct traverse_state *state = (struct traverse_state *)private_data;
+	if (tdb_store(state->dest_db, key, data, TDB_INSERT) != 0) {
+		state->error = true;
+		return -1;
+	}
+	return 0;
+}
+
+/*
+  repack a tdb
+ */
+_PUBLIC_ int tdb_repack(struct tdb_context *tdb)
+{
+	struct tdb_context *tmp_db;
+	struct traverse_state state;
+
+	tdb_trace(tdb, "tdb_repack");
+
+	if (tdb_transaction_start(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to start transaction\n"));
+		return -1;
+	}
+
+	tmp_db = tdb_open("tmpdb", tdb_hash_size(tdb), TDB_INTERNAL, O_RDWR|O_CREAT, 0);
+	if (tmp_db == NULL) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to create tmp_db\n"));
+		tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	state.error = false;
+	state.dest_db = tmp_db;
+
+	if (tdb_traverse_read(tdb, repack_traverse, &state) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying out\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	if (state.error) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during traversal\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	if (tdb_wipe_all(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to wipe database\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	state.error = false;
+	state.dest_db = tdb;
+
+	if (tdb_traverse_read(tmp_db, repack_traverse, &state) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying back\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	if (state.error) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during second traversal\n"));
+		tdb_transaction_cancel(tdb);
+		tdb_close(tmp_db);
+		return -1;
+	}
+
+	tdb_close(tmp_db);
+
+	if (tdb_transaction_commit(tdb) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to commit\n"));
+		return -1;
+	}
+
+	return 0;
+}
+
+/* Even on files, we can get partial writes due to signals. */
+bool tdb_write_all(int fd, const void *buf, size_t count)
+{
+	while (count) {
+		ssize_t ret;
+		ret = write(fd, buf, count);
+		if (ret < 0)
+			return false;
+		buf = (const char *)buf + ret;
+		count -= ret;
+	}
+	return true;
+}
+
+bool tdb_add_off_t(tdb_off_t a, tdb_off_t b, tdb_off_t *pret)
+{
+	tdb_off_t ret = a + b;
+
+	if ((ret < a) || (ret < b)) {
+		return false;
+	}
+	*pret = ret;
+	return true;
+}
+
+#ifdef TDB_TRACE
+static void tdb_trace_write(struct tdb_context *tdb, const char *str)
+{
+	if (!tdb_write_all(tdb->tracefd, str, strlen(str))) {
+		close(tdb->tracefd);
+		tdb->tracefd = -1;
+	}
+}
+
+static void tdb_trace_start(struct tdb_context *tdb)
+{
+	tdb_off_t seqnum=0;
+	char msg[sizeof(tdb_off_t) * 4 + 1];
+
+	tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
+	snprintf(msg, sizeof(msg), "%u ", seqnum);
+	tdb_trace_write(tdb, msg);
+}
+
+static void tdb_trace_end(struct tdb_context *tdb)
+{
+	tdb_trace_write(tdb, "\n");
+}
+
+static void tdb_trace_end_ret(struct tdb_context *tdb, int ret)
+{
+	char msg[sizeof(ret) * 4 + 4];
+	snprintf(msg, sizeof(msg), " = %i\n", ret);
+	tdb_trace_write(tdb, msg);
+}
+
+static void tdb_trace_record(struct tdb_context *tdb, TDB_DATA rec)
+{
+	char msg[20 + rec.dsize*2], *p;
+	unsigned int i;
+
+	/* We differentiate zero-length records from non-existent ones. */
+	if (rec.dptr == NULL) {
+		tdb_trace_write(tdb, " NULL");
+		return;
+	}
+
+	/* snprintf here is purely cargo-cult programming. */
+	p = msg;
+	p += snprintf(p, sizeof(msg), " %zu:", rec.dsize);
+	for (i = 0; i < rec.dsize; i++)
+		p += snprintf(p, 2, "%02x", rec.dptr[i]);
+
+	tdb_trace_write(tdb, msg);
+}
+
+void tdb_trace(struct tdb_context *tdb, const char *op)
+{
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_end(tdb);
+}
+
+void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op)
+{
+	char msg[sizeof(tdb_off_t) * 4 + 1];
+
+	snprintf(msg, sizeof(msg), "%u ", seqnum);
+	tdb_trace_write(tdb, msg);
+	tdb_trace_write(tdb, op);
+	tdb_trace_end(tdb);
+}
+
+void tdb_trace_open(struct tdb_context *tdb, const char *op,
+		    unsigned hash_size, unsigned tdb_flags, unsigned open_flags)
+{
+	char msg[128];
+
+	snprintf(msg, sizeof(msg),
+		 "%s %u 0x%x 0x%x", op, hash_size, tdb_flags, open_flags);
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, msg);
+	tdb_trace_end(tdb);
+}
+
+void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret)
+{
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_end_ret(tdb, ret);
+}
+
+void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret)
+{
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_write(tdb, " =");
+	tdb_trace_record(tdb, ret);
+	tdb_trace_end(tdb);
+}
+
+void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
+		    TDB_DATA rec)
+{
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_record(tdb, rec);
+	tdb_trace_end(tdb);
+}
+
+void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
+			TDB_DATA rec, int ret)
+{
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_record(tdb, rec);
+	tdb_trace_end_ret(tdb, ret);
+}
+
+void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
+			   TDB_DATA rec, TDB_DATA ret)
+{
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_record(tdb, rec);
+	tdb_trace_write(tdb, " =");
+	tdb_trace_record(tdb, ret);
+	tdb_trace_end(tdb);
+}
+
+void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
+			     TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
+			     int ret)
+{
+	char msg[1 + sizeof(ret) * 4];
+
+	snprintf(msg, sizeof(msg), " %#x", flag);
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_record(tdb, rec1);
+	tdb_trace_record(tdb, rec2);
+	tdb_trace_write(tdb, msg);
+	tdb_trace_end_ret(tdb, ret);
+}
+
+void tdb_trace_1plusn_rec_flag_ret(struct tdb_context *tdb, const char *op,
+				   TDB_DATA rec,
+				   const TDB_DATA *recs, int num_recs,
+				   unsigned flag, int ret)
+{
+	char msg[1 + sizeof(ret) * 4];
+	int i;
+
+	snprintf(msg, sizeof(msg), " %#x", flag);
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_record(tdb, rec);
+	for (i=0; i<num_recs; i++) {
+		tdb_trace_record(tdb, recs[i]);
+	}
+	tdb_trace_write(tdb, msg);
+	tdb_trace_end_ret(tdb, ret);
+}
+
+void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
+			   TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret)
+{
+	tdb_trace_start(tdb);
+	tdb_trace_write(tdb, op);
+	tdb_trace_record(tdb, rec1);
+	tdb_trace_record(tdb, rec2);
+	tdb_trace_write(tdb, " =");
+	tdb_trace_record(tdb, ret);
+	tdb_trace_end(tdb);
+}
+#endif
diff --git a/lib/tdb/common/tdb_private.h b/lib/tdb/common/tdb_private.h
new file mode 100644
index 0000000..2979043
--- /dev/null
+++ b/lib/tdb/common/tdb_private.h
@@ -0,0 +1,370 @@
+#ifndef TDB_PRIVATE_H
+#define TDB_PRIVATE_H
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library - private includes
+
+   Copyright (C) Andrew Tridgell              2005
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "replace.h"
+#include "system/filesys.h"
+#include "system/time.h"
+#include "system/shmem.h"
+#include "system/select.h"
+#include "system/wait.h"
+#include "tdb.h"
+
+/* #define TDB_TRACE 1 */
+#ifndef HAVE_GETPAGESIZE
+#define getpagesize() 0x2000
+#endif
+
+typedef uint32_t tdb_len_t;
+typedef uint32_t tdb_off_t;
+
+#ifndef offsetof
+#define offsetof(t,f) ((unsigned int)&((t *)0)->f)
+#endif
+
+#define TDB_MAGIC_FOOD "TDB file\n"
+#define TDB_VERSION (0x26011967 + 6)
+#define TDB_MAGIC (0x26011999U)
+#define TDB_FREE_MAGIC (~TDB_MAGIC)
+#define TDB_DEAD_MAGIC (0xFEE1DEAD)
+#define TDB_RECOVERY_MAGIC (0xf53bc0e7U)
+#define TDB_RECOVERY_INVALID_MAGIC (0x0)
+#define TDB_HASH_RWLOCK_MAGIC (0xbad1a51U)
+#define TDB_FEATURE_FLAG_MAGIC (0xbad1a52U)
+#define TDB_ALIGNMENT 4
+#define DEFAULT_HASH_SIZE 131
+#define FREELIST_TOP (sizeof(struct tdb_header))
+#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
+#define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
+#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
+#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
+#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off_t))
+#define TDB_HASHTABLE_SIZE(tdb) ((tdb->hash_size+1)*sizeof(tdb_off_t))
+#define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + sizeof(tdb_off_t))
+#define TDB_RECOVERY_HEAD offsetof(struct tdb_header, recovery_start)
+#define TDB_SEQNUM_OFS    offsetof(struct tdb_header, sequence_number)
+#define TDB_PAD_BYTE 0x42
+#define TDB_PAD_U32  0x42424242
+
+#define TDB_FEATURE_FLAG_MUTEX 0x00000001
+
+#define TDB_SUPPORTED_FEATURE_FLAGS ( \
+	TDB_FEATURE_FLAG_MUTEX | \
+	0)
+
+/* NB assumes there is a local variable called "tdb" that is the
+ * current context, also takes doubly-parenthesized print-style
+ * argument. */
+#define TDB_LOG(x) tdb->log.log_fn x
+
+#ifdef TDB_TRACE
+void tdb_trace(struct tdb_context *tdb, const char *op);
+void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
+void tdb_trace_open(struct tdb_context *tdb, const char *op,
+		    unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
+void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
+void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
+void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
+		    TDB_DATA rec);
+void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
+			TDB_DATA rec, int ret);
+void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
+			   TDB_DATA rec, TDB_DATA ret);
+void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
+			     TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
+			     int ret);
+void tdb_trace_1plusn_rec_flag_ret(struct tdb_context *tdb, const char *op,
+				   TDB_DATA rec,
+				   const TDB_DATA *recs, int num_recs,
+				   unsigned flag, int ret);
+void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
+			   TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
+#else
+#define tdb_trace(tdb, op)
+#define tdb_trace_seqnum(tdb, seqnum, op)
+#define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
+#define tdb_trace_ret(tdb, op, ret)
+#define tdb_trace_retrec(tdb, op, ret)
+#define tdb_trace_1rec(tdb, op, rec)
+#define tdb_trace_1rec_ret(tdb, op, rec, ret)
+#define tdb_trace_1rec_retrec(tdb, op, rec, ret)
+#define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
+#define tdb_trace_1plusn_rec_flag_ret(tdb, op, rec, recs, num_recs, flag, ret);
+#define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
+#endif /* !TDB_TRACE */
+
+/* lock offsets */
+#define OPEN_LOCK        0
+#define ACTIVE_LOCK      4
+#define TRANSACTION_LOCK 8
+
+/* free memory if the pointer is valid and zero the pointer */
+#ifndef SAFE_FREE
+#define SAFE_FREE(x) do { if ((x) != NULL) {free(x); (x)=NULL;} } while(0)
+#endif
+
+/*
+ * Note: the BUCKET macro is broken as it returns an unexpected result when
+ * called as BUCKET(-1) for the freelist:
+ *
+ * -1 is sign converted to an unsigned int 4294967295 and then the modulo
+ * tdb->hashtable_size is computed. So with a hashtable_size of 10 the result
+ * is
+ *
+ *   4294967295 % hashtable_size = 5.
+ *
+ * where it should be -1 (C uses symmetric modulo).
+ *
+ * As all callers will lock the same wrong list consistently locking is still
+ * consistent. We can not change this without an incompatible on-disk format
+ * change, otherwise different tdb versions would use incompatible locking.
+ */
+#define BUCKET(hash) ((hash) % tdb->hash_size)
+
+#define DOCONV() (tdb->flags & TDB_CONVERT)
+#define CONVERT(x) (DOCONV() ? tdb_convert(&x, sizeof(x)) : &x)
+
+
+/* the body of the database is made of one tdb_record for the free space
+   plus a separate data list for each hash value */
+struct tdb_record {
+	tdb_off_t next; /* offset of the next record in the list */
+	tdb_len_t rec_len; /* total byte length of record */
+	tdb_len_t key_len; /* byte length of key */
+	tdb_len_t data_len; /* byte length of data */
+	uint32_t full_hash; /* the full 32 bit hash of the key */
+	uint32_t magic;   /* try to catch errors */
+	/* the following union is implied:
+		union {
+			char record[rec_len];
+			struct {
+				char key[key_len];
+				char data[data_len];
+			}
+			uint32_t totalsize; (tailer)
+		}
+	*/
+};
+
+
+/* this is stored at the front of every database */
+struct tdb_header {
+	char magic_food[32]; /* for /etc/magic */
+	uint32_t version; /* version of the code */
+	uint32_t hash_size; /* number of hash entries */
+	tdb_off_t rwlocks; /* obsolete - kept to detect old formats */
+	tdb_off_t recovery_start; /* offset of transaction recovery region */
+	tdb_off_t sequence_number; /* used when TDB_SEQNUM is set */
+	uint32_t magic1_hash; /* hash of TDB_MAGIC_FOOD. */
+	uint32_t magic2_hash; /* hash of TDB_MAGIC. */
+	uint32_t feature_flags;
+	tdb_len_t mutex_size; /* set if TDB_FEATURE_FLAG_MUTEX is set */
+	tdb_off_t reserved[25];
+};
+
+struct tdb_lock_type {
+	uint32_t off;
+	uint32_t count;
+	uint32_t ltype;
+};
+
+struct tdb_chainwalk_ctx {
+	tdb_off_t slow_ptr;
+	bool slow_chase;
+};
+
+struct tdb_traverse_lock {
+	struct tdb_traverse_lock *next;
+	uint32_t off;
+	uint32_t list;
+	int lock_rw;
+};
+
+void tdb_chainwalk_init(struct tdb_chainwalk_ctx *ctx, tdb_off_t ptr);
+bool tdb_chainwalk_check(struct tdb_context *tdb,
+			 struct tdb_chainwalk_ctx *ctx,
+			 tdb_off_t next_ptr);
+
+enum tdb_lock_flags {
+	/* WAIT == F_SETLKW, NOWAIT == F_SETLK */
+	TDB_LOCK_NOWAIT = 0,
+	TDB_LOCK_WAIT = 1,
+	/* If set, don't log an error on failure. */
+	TDB_LOCK_PROBE = 2,
+	/* If set, don't actually lock at all. */
+	TDB_LOCK_MARK_ONLY = 4,
+};
+
+struct tdb_methods {
+	int (*tdb_read)(struct tdb_context *, tdb_off_t , void *, tdb_len_t , int );
+	int (*tdb_write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
+	void (*next_hash_chain)(struct tdb_context *, uint32_t *);
+	int (*tdb_oob)(struct tdb_context *, tdb_off_t , tdb_len_t, int );
+	int (*tdb_expand_file)(struct tdb_context *, tdb_off_t , tdb_off_t );
+};
+
+struct tdb_mutexes;
+
+struct tdb_context {
+	char *name; /* the name of the database */
+	void *map_ptr; /* where it is currently mapped */
+	int fd; /* open file descriptor for the database */
+	tdb_len_t map_size; /* how much space has been mapped */
+	int read_only; /* opened read-only */
+	int traverse_read; /* read-only traversal */
+	int traverse_write; /* read-write traversal */
+	struct tdb_lock_type allrecord_lock; /* .offset == upgradable */
+	int num_lockrecs;
+	struct tdb_lock_type *lockrecs; /* only real locks, all with count>0 */
+	int lockrecs_array_length;
+
+	tdb_off_t hdr_ofs; /* this is 0 or header.mutex_size */
+	struct tdb_mutexes *mutexes; /* mmap of the mutex area */
+
+	enum TDB_ERROR ecode; /* error code for last tdb error */
+	uint32_t hash_size;
+	uint32_t feature_flags;
+	uint32_t flags; /* the flags passed to tdb_open */
+	struct tdb_traverse_lock travlocks; /* current traversal locks */
+	struct tdb_context *next; /* all tdbs to avoid multiple opens */
+	dev_t device;	/* uniquely identifies this tdb */
+	ino_t inode;	/* uniquely identifies this tdb */
+	struct tdb_logging_context log;
+	unsigned int (*hash_fn)(TDB_DATA *key);
+	int open_flags; /* flags used in the open - needed by reopen */
+	const struct tdb_methods *methods;
+	struct tdb_transaction *transaction;
+	int page_size;
+	int max_dead_records;
+#ifdef TDB_TRACE
+	int tracefd;
+#endif
+	volatile sig_atomic_t *interrupt_sig_ptr;
+};
+
+
+/*
+  internal prototypes
+*/
+int tdb_munmap(struct tdb_context *tdb);
+int tdb_mmap(struct tdb_context *tdb);
+int tdb_lock(struct tdb_context *tdb, int list, int ltype);
+int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
+int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
+		  enum tdb_lock_flags flags);
+int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
+		    bool mark_lock);
+int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
+int tdb_brlock(struct tdb_context *tdb,
+	       int rw_type, tdb_off_t offset, size_t len,
+	       enum tdb_lock_flags flags);
+int tdb_brunlock(struct tdb_context *tdb,
+		 int rw_type, tdb_off_t offset, size_t len);
+bool tdb_have_extra_locks(struct tdb_context *tdb);
+void tdb_release_transaction_locks(struct tdb_context *tdb);
+int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
+			 enum tdb_lock_flags lockflags);
+int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
+int tdb_recovery_area(struct tdb_context *tdb,
+		      const struct tdb_methods *methods,
+		      tdb_off_t *recovery_offset,
+		      struct tdb_record *rec);
+int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
+		       enum tdb_lock_flags flags, bool upgradable);
+int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock);
+int tdb_allrecord_upgrade(struct tdb_context *tdb);
+int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off);
+int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off);
+int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+void *tdb_convert(void *buf, uint32_t size);
+int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
+tdb_off_t tdb_allocate(struct tdb_context *tdb, int hash, tdb_len_t length,
+		       struct tdb_record *rec);
+
+int _tdb_oob(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, int probe);
+
+static inline int tdb_oob(
+	struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, int probe)
+{
+	if (likely((off + len >= off) && (off + len <= tdb->map_size))) {
+		return 0;
+	}
+	return _tdb_oob(tdb, off, len, probe);
+}
+
+
+int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
+int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off);
+int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off);
+bool tdb_needs_recovery(struct tdb_context *tdb);
+int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
+int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
+unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
+int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
+		   tdb_off_t offset, tdb_len_t len,
+		   int (*parser)(TDB_DATA key, TDB_DATA data,
+				 void *private_data),
+		   void *private_data);
+tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
+			   struct tdb_record *rec);
+tdb_off_t tdb_find_dead(struct tdb_context *tdb, uint32_t hash,
+			struct tdb_record *r, tdb_len_t length,
+			tdb_off_t *p_last_ptr);
+int tdb_trim_dead(struct tdb_context *tdb, uint32_t hash);
+void tdb_io_init(struct tdb_context *tdb);
+int tdb_expand(struct tdb_context *tdb, tdb_off_t size);
+tdb_off_t tdb_expand_adjust(tdb_off_t map_size, tdb_off_t size, int page_size);
+int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off,
+		      struct tdb_record *rec);
+bool tdb_write_all(int fd, const void *buf, size_t count);
+int tdb_transaction_recover(struct tdb_context *tdb);
+void tdb_header_hash(struct tdb_context *tdb,
+		     uint32_t *magic1_hash, uint32_t *magic2_hash);
+unsigned int tdb_old_hash(TDB_DATA *key);
+size_t tdb_dead_space(struct tdb_context *tdb, tdb_off_t off);
+bool tdb_add_off_t(tdb_off_t a, tdb_off_t b, tdb_off_t *pret);
+
+/* tdb_off_t and tdb_len_t right now are both uint32_t */
+#define tdb_add_len_t tdb_add_off_t
+
+size_t tdb_mutex_size(struct tdb_context *tdb);
+bool tdb_have_mutexes(struct tdb_context *tdb);
+int tdb_mutex_init(struct tdb_context *tdb);
+int tdb_mutex_mmap(struct tdb_context *tdb);
+int tdb_mutex_munmap(struct tdb_context *tdb);
+bool tdb_mutex_lock(struct tdb_context *tdb, int rw, off_t off, off_t len,
+		    bool waitflag, int *pret);
+bool tdb_mutex_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len,
+		      int *pret);
+int tdb_mutex_allrecord_lock(struct tdb_context *tdb, int ltype,
+			     enum tdb_lock_flags flags);
+int tdb_mutex_allrecord_unlock(struct tdb_context *tdb);
+int tdb_mutex_allrecord_upgrade(struct tdb_context *tdb);
+void tdb_mutex_allrecord_downgrade(struct tdb_context *tdb);
+
+#endif /* TDB_PRIVATE_H */
diff --git a/lib/tdb/common/transaction.c b/lib/tdb/common/transaction.c
new file mode 100644
index 0000000..5ab38d0
--- /dev/null
+++ b/lib/tdb/common/transaction.c
@@ -0,0 +1,1388 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              2005
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+/*
+  transaction design:
+
+  - only allow a single transaction at a time per database. This makes
+    using the transaction API simpler, as otherwise the caller would
+    have to cope with temporary failures in transactions that conflict
+    with other current transactions
+
+  - keep the transaction recovery information in the same file as the
+    database, using a special 'transaction recovery' record pointed at
+    by the header. This removes the need for extra journal files as
+    used by some other databases
+
+  - dynamically allocated the transaction recover record, re-using it
+    for subsequent transactions. If a larger record is needed then
+    tdb_free() the old record to place it on the normal tdb freelist
+    before allocating the new record
+
+  - during transactions, keep a linked list of all writes that have
+    been performed by intercepting all tdb_write() calls. The hooked
+    transaction versions of tdb_read() and tdb_write() check this
+    linked list and try to use the elements of the list in preference
+    to the real database.
+
+  - don't allow any locks to be held when a transaction starts,
+    otherwise we can end up with deadlock (plus lack of lock nesting
+    in posix locks would mean the lock is lost)
+
+  - if the caller gains a lock during the transaction but doesn't
+    release it then fail the commit
+
+  - allow for nested calls to tdb_transaction_start(), re-using the
+    existing transaction record. If the inner transaction is cancelled
+    then a subsequent commit will fail
+
+  - keep a mirrored copy of the tdb hash chain heads to allow for the
+    fast hash heads scan on traverse, updating the mirrored copy in
+    the transaction version of tdb_write
+
+  - allow callers to mix transaction and non-transaction use of tdb,
+    although once a transaction is started then an exclusive lock is
+    gained until the transaction is committed or cancelled
+
+  - the commit stategy involves first saving away all modified data
+    into a linearised buffer in the transaction recovery area, then
+    marking the transaction recovery area with a magic value to
+    indicate a valid recovery record. In total 4 fsync/msync calls are
+    needed per commit to prevent race conditions. It might be possible
+    to reduce this to 3 or even 2 with some more work.
+
+  - check for a valid recovery record on open of the tdb, while the
+    open lock is held. Automatically recover from the transaction
+    recovery area if needed, then continue with the open as
+    usual. This allows for smooth crash recovery with no administrator
+    intervention.
+
+  - if TDB_NOSYNC is passed to flags in tdb_open then transactions are
+    still available, but no fsync/msync calls are made.  This means we
+    are still proof against a process dying during transaction commit,
+    but not against machine reboot.
+
+  - if TDB_ALLOW_NESTING is passed to flags in tdb open, or added using
+    tdb_add_flags() transaction nesting is enabled.
+    It resets the TDB_DISALLOW_NESTING flag, as both cannot be used together.
+    The default is that transaction nesting is allowed.
+    Note: this default may change in future versions of tdb.
+
+    Beware. when transactions are nested a transaction successfully
+    completed with tdb_transaction_commit() can be silently unrolled later.
+
+  - if TDB_DISALLOW_NESTING is passed to flags in tdb open, or added using
+    tdb_add_flags() transaction nesting is disabled.
+    It resets the TDB_ALLOW_NESTING flag, as both cannot be used together.
+    An attempt create a nested transaction will fail with TDB_ERR_NESTING.
+    The default is that transaction nesting is allowed.
+    Note: this default may change in future versions of tdb.
+*/
+
+
+/*
+  hold the context of any current transaction
+*/
+struct tdb_transaction {
+	/* we keep a mirrored copy of the tdb hash heads here so
+	   tdb_next_hash_chain() can operate efficiently */
+	uint32_t *hash_heads;
+
+	/* the original io methods - used to do IOs to the real db */
+	const struct tdb_methods *io_methods;
+
+	/* the list of transaction blocks. When a block is first
+	   written to, it gets created in this list */
+	uint8_t **blocks;
+	uint32_t num_blocks;
+	uint32_t block_size;      /* bytes in each block */
+	uint32_t last_block_size; /* number of valid bytes in the last block */
+
+	/* non-zero when an internal transaction error has
+	   occurred. All write operations will then fail until the
+	   transaction is ended */
+	int transaction_error;
+
+	/* when inside a transaction we need to keep track of any
+	   nested tdb_transaction_start() calls, as these are allowed,
+	   but don't create a new transaction */
+	int nesting;
+
+	/* set when a prepare has already occurred */
+	bool prepared;
+	tdb_off_t magic_offset;
+
+	/* old file size before transaction */
+	tdb_len_t old_map_size;
+
+	/* did we expand in this transaction */
+	bool expanded;
+};
+
+
+/*
+  read while in a transaction. We need to check first if the data is in our list
+  of transaction elements, then if not do a real read
+*/
+static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
+			    tdb_len_t len, int cv)
+{
+	uint32_t blk;
+
+	/* break it down into block sized ops */
+	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+		if (transaction_read(tdb, off, buf, len2, cv) != 0) {
+			return -1;
+		}
+		len -= len2;
+		off += len2;
+		buf = (void *)(len2 + (char *)buf);
+	}
+
+	if (len == 0) {
+		return 0;
+	}
+
+	blk = off / tdb->transaction->block_size;
+
+	/* see if we have it in the block list */
+	if (tdb->transaction->num_blocks <= blk ||
+	    tdb->transaction->blocks[blk] == NULL) {
+		/* nope, do a real read */
+		if (tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv) != 0) {
+			goto fail;
+		}
+		return 0;
+	}
+
+	/* it is in the block list. Now check for the last block */
+	if (blk == tdb->transaction->num_blocks-1) {
+		if (len > tdb->transaction->last_block_size) {
+			goto fail;
+		}
+	}
+
+	/* now copy it out of this block */
+	memcpy(buf, tdb->transaction->blocks[blk] + (off % tdb->transaction->block_size), len);
+	if (cv) {
+		tdb_convert(buf, len);
+	}
+	return 0;
+
+fail:
+	TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_read: failed at off=%u len=%u\n", off, len));
+	tdb->ecode = TDB_ERR_IO;
+	tdb->transaction->transaction_error = 1;
+	return -1;
+}
+
+
+/*
+  write while in a transaction
+*/
+static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
+			     const void *buf, tdb_len_t len)
+{
+	uint32_t blk;
+
+	if (buf == NULL) {
+		return -1;
+	}
+
+	/* Only a commit is allowed on a prepared transaction */
+	if (tdb->transaction->prepared) {
+		tdb->ecode = TDB_ERR_EINVAL;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: transaction already prepared, write not allowed\n"));
+		tdb->transaction->transaction_error = 1;
+		return -1;
+	}
+
+	/* if the write is to a hash head, then update the transaction
+	   hash heads */
+	if (len == sizeof(tdb_off_t) && off >= FREELIST_TOP &&
+	    off < FREELIST_TOP+TDB_HASHTABLE_SIZE(tdb)) {
+		uint32_t chain = (off-FREELIST_TOP) / sizeof(tdb_off_t);
+		memcpy(&tdb->transaction->hash_heads[chain], buf, len);
+	}
+
+	/* break it up into block sized chunks */
+	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+		if (transaction_write(tdb, off, buf, len2) != 0) {
+			return -1;
+		}
+		len -= len2;
+		off += len2;
+		buf = (const void *)(len2 + (const char *)buf);
+	}
+
+	if (len == 0) {
+		return 0;
+	}
+
+	blk = off / tdb->transaction->block_size;
+	off = off % tdb->transaction->block_size;
+
+	if (tdb->transaction->num_blocks <= blk) {
+		uint8_t **new_blocks;
+		/* expand the blocks array */
+		new_blocks = (uint8_t **)realloc(tdb->transaction->blocks,
+						 (blk+1)*sizeof(uint8_t *));
+		if (new_blocks == NULL) {
+			tdb->ecode = TDB_ERR_OOM;
+			goto fail;
+		}
+		memset(&new_blocks[tdb->transaction->num_blocks], 0,
+		       (1+(blk - tdb->transaction->num_blocks))*sizeof(uint8_t *));
+		tdb->transaction->blocks = new_blocks;
+		tdb->transaction->num_blocks = blk+1;
+		tdb->transaction->last_block_size = 0;
+	}
+
+	/* allocate and fill a block? */
+	if (tdb->transaction->blocks[blk] == NULL) {
+		tdb->transaction->blocks[blk] = (uint8_t *)calloc(tdb->transaction->block_size, 1);
+		if (tdb->transaction->blocks[blk] == NULL) {
+			tdb->ecode = TDB_ERR_OOM;
+			tdb->transaction->transaction_error = 1;
+			return -1;
+		}
+		if (tdb->transaction->old_map_size > blk * tdb->transaction->block_size) {
+			tdb_len_t len2 = tdb->transaction->block_size;
+			if (len2 + (blk * tdb->transaction->block_size) > tdb->transaction->old_map_size) {
+				len2 = tdb->transaction->old_map_size - (blk * tdb->transaction->block_size);
+			}
+			if (tdb->transaction->io_methods->tdb_read(tdb, blk * tdb->transaction->block_size,
+								   tdb->transaction->blocks[blk],
+								   len2, 0) != 0) {
+				SAFE_FREE(tdb->transaction->blocks[blk]);
+				tdb->ecode = TDB_ERR_IO;
+				goto fail;
+			}
+			if (blk == tdb->transaction->num_blocks-1) {
+				tdb->transaction->last_block_size = len2;
+			}
+		}
+	}
+
+	/* overwrite part of an existing block */
+	memcpy(tdb->transaction->blocks[blk] + off, buf, len);
+	if (blk == tdb->transaction->num_blocks-1) {
+		if (len + off > tdb->transaction->last_block_size) {
+			tdb->transaction->last_block_size = len + off;
+		}
+	}
+
+	return 0;
+
+fail:
+	TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%u len=%u\n",
+		 (blk*tdb->transaction->block_size) + off, len));
+	tdb->transaction->transaction_error = 1;
+	return -1;
+}
+
+
+/*
+  write while in a transaction - this variant never expands the transaction blocks, it only
+  updates existing blocks. This means it cannot change the recovery size
+*/
+static int transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
+				      const void *buf, tdb_len_t len)
+{
+	uint32_t blk;
+
+	/* break it up into block sized chunks */
+	while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
+		tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
+		if (transaction_write_existing(tdb, off, buf, len2) != 0) {
+			return -1;
+		}
+		len -= len2;
+		off += len2;
+		if (buf != NULL) {
+			buf = (const void *)(len2 + (const char *)buf);
+		}
+	}
+
+	if (len == 0 || buf == NULL) {
+		return 0;
+	}
+
+	blk = off / tdb->transaction->block_size;
+	off = off % tdb->transaction->block_size;
+
+	if (tdb->transaction->num_blocks <= blk ||
+	    tdb->transaction->blocks[blk] == NULL) {
+		return 0;
+	}
+
+	if (blk == tdb->transaction->num_blocks-1 &&
+	    off + len > tdb->transaction->last_block_size) {
+		if (off >= tdb->transaction->last_block_size) {
+			return 0;
+		}
+		len = tdb->transaction->last_block_size - off;
+	}
+
+	/* overwrite part of an existing block */
+	memcpy(tdb->transaction->blocks[blk] + off, buf, len);
+
+	return 0;
+}
+
+
+/*
+  accelerated hash chain head search, using the cached hash heads
+*/
+static void transaction_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
+{
+	uint32_t h = *chain;
+	for (;h < tdb->hash_size;h++) {
+		/* the +1 takes account of the freelist */
+		if (0 != tdb->transaction->hash_heads[h+1]) {
+			break;
+		}
+	}
+	(*chain) = h;
+}
+
+/*
+  out of bounds check during a transaction
+*/
+static int transaction_oob(struct tdb_context *tdb, tdb_off_t off,
+			   tdb_len_t len, int probe)
+{
+	/*
+	 * This duplicates functionality from tdb_oob(). Don't remove:
+	 * we still have direct callers of tdb->methods->tdb_oob()
+	 * inside transaction.c.
+	 */
+	if (off + len >= off && off + len <= tdb->map_size) {
+		return 0;
+	}
+	tdb->ecode = TDB_ERR_IO;
+	return -1;
+}
+
+/*
+  transaction version of tdb_expand().
+*/
+static int transaction_expand_file(struct tdb_context *tdb, tdb_off_t size,
+				   tdb_off_t addition)
+{
+	const char buf_zero[8192] = {0};
+	size_t buf_len = sizeof(buf_zero);
+
+	while (addition > 0) {
+		size_t n = MIN(addition, buf_len);
+		int ret;
+
+		ret = transaction_write(tdb, size, buf_zero, n);
+		if (ret != 0) {
+			return ret;
+		}
+
+		addition -= n;
+		size += n;
+	}
+
+	tdb->transaction->expanded = true;
+
+	return 0;
+}
+
+static const struct tdb_methods transaction_methods = {
+	transaction_read,
+	transaction_write,
+	transaction_next_hash_chain,
+	transaction_oob,
+	transaction_expand_file,
+};
+
+/*
+ * Is a transaction currently active on this context?
+ *
+ */
+_PUBLIC_ bool tdb_transaction_active(struct tdb_context *tdb)
+{
+	return (tdb->transaction != NULL);
+}
+
+/*
+  start a tdb transaction. No token is returned, as only a single
+  transaction is allowed to be pending per tdb_context
+*/
+static int _tdb_transaction_start(struct tdb_context *tdb,
+				  enum tdb_lock_flags lockflags)
+{
+	/* some sanity checks */
+	if (tdb->read_only || (tdb->flags & TDB_INTERNAL)
+	    || tdb->traverse_read) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction on a read-only or internal db\n"));
+		tdb->ecode = TDB_ERR_EINVAL;
+		return -1;
+	}
+
+	/* cope with nested tdb_transaction_start() calls */
+	if (tdb->transaction != NULL) {
+		if (!(tdb->flags & TDB_ALLOW_NESTING)) {
+			tdb->ecode = TDB_ERR_NESTING;
+			return -1;
+		}
+		tdb->transaction->nesting++;
+		TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_start: nesting %d\n",
+			 tdb->transaction->nesting));
+		return 0;
+	}
+
+	if (tdb_have_extra_locks(tdb)) {
+		/* the caller must not have any locks when starting a
+		   transaction as otherwise we'll be screwed by lack
+		   of nested locks in posix */
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction with locks held\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	if (tdb->travlocks.next != NULL) {
+		/* you cannot use transactions inside a traverse (although you can use
+		   traverse inside a transaction) as otherwise you can end up with
+		   deadlock */
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction within a traverse\n"));
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	tdb->transaction = (struct tdb_transaction *)
+		calloc(sizeof(struct tdb_transaction), 1);
+	if (tdb->transaction == NULL) {
+		tdb->ecode = TDB_ERR_OOM;
+		return -1;
+	}
+
+	/* a page at a time seems like a reasonable compromise between compactness and efficiency */
+	tdb->transaction->block_size = tdb->page_size;
+
+	/* get the transaction write lock. This is a blocking lock. As
+	   discussed with Volker, there are a number of ways we could
+	   make this async, which we will probably do in the future */
+	if (tdb_transaction_lock(tdb, F_WRLCK, lockflags) == -1) {
+		SAFE_FREE(tdb->transaction->blocks);
+		SAFE_FREE(tdb->transaction);
+		if ((lockflags & TDB_LOCK_WAIT) == 0) {
+			tdb->ecode = TDB_ERR_NOLOCK;
+		} else {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "tdb_transaction_start: "
+				 "failed to get transaction lock\n"));
+		}
+		return -1;
+	}
+
+	/* get a read lock from the freelist to the end of file. This
+	   is upgraded to a write lock during the commit */
+	if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get hash locks\n"));
+		goto fail_allrecord_lock;
+	}
+
+	/* setup a copy of the hash table heads so the hash scan in
+	   traverse can be fast */
+	tdb->transaction->hash_heads = (uint32_t *)
+		calloc(tdb->hash_size+1, sizeof(uint32_t));
+	if (tdb->transaction->hash_heads == NULL) {
+		tdb->ecode = TDB_ERR_OOM;
+		goto fail;
+	}
+	if (tdb->methods->tdb_read(tdb, FREELIST_TOP, tdb->transaction->hash_heads,
+				   TDB_HASHTABLE_SIZE(tdb), 0) != 0) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_start: failed to read hash heads\n"));
+		tdb->ecode = TDB_ERR_IO;
+		goto fail;
+	}
+
+	/* make sure we know about any file expansions already done by
+	   anyone else */
+	tdb_oob(tdb, tdb->map_size, 1, 1);
+	tdb->transaction->old_map_size = tdb->map_size;
+
+	/* finally hook the io methods, replacing them with
+	   transaction specific methods */
+	tdb->transaction->io_methods = tdb->methods;
+	tdb->methods = &transaction_methods;
+
+	/* Trace at the end, so we get sequence number correct. */
+	tdb_trace(tdb, "tdb_transaction_start");
+	return 0;
+
+fail:
+	tdb_allrecord_unlock(tdb, F_RDLCK, false);
+fail_allrecord_lock:
+	tdb_transaction_unlock(tdb, F_WRLCK);
+	SAFE_FREE(tdb->transaction->blocks);
+	SAFE_FREE(tdb->transaction->hash_heads);
+	SAFE_FREE(tdb->transaction);
+	return -1;
+}
+
+_PUBLIC_ int tdb_transaction_start(struct tdb_context *tdb)
+{
+	return _tdb_transaction_start(tdb, TDB_LOCK_WAIT);
+}
+
+_PUBLIC_ int tdb_transaction_start_nonblock(struct tdb_context *tdb)
+{
+	return _tdb_transaction_start(tdb, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
+}
+
+/*
+  sync to disk
+*/
+static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t length)
+{
+	if (tdb->flags & TDB_NOSYNC) {
+		return 0;
+	}
+
+#ifdef HAVE_FDATASYNC
+	if (fdatasync(tdb->fd) != 0) {
+#else
+	if (fsync(tdb->fd) != 0) {
+#endif
+		tdb->ecode = TDB_ERR_IO;
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: fsync failed\n"));
+		return -1;
+	}
+#ifdef HAVE_MMAP
+	if (tdb->map_ptr) {
+		tdb_off_t moffset = offset & ~(tdb->page_size-1);
+		if (msync(moffset + (char *)tdb->map_ptr,
+			  length + (offset - moffset), MS_SYNC) != 0) {
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: msync failed - %s\n",
+				 strerror(errno)));
+			return -1;
+		}
+	}
+#endif
+	return 0;
+}
+
+
+static int _tdb_transaction_cancel(struct tdb_context *tdb)
+{
+	uint32_t i;
+	int ret = 0;
+
+	if (tdb->transaction == NULL) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_cancel: no transaction\n"));
+		return -1;
+	}
+
+	if (tdb->transaction->nesting != 0) {
+		tdb->transaction->transaction_error = 1;
+		tdb->transaction->nesting--;
+		return 0;
+	}
+
+	tdb->map_size = tdb->transaction->old_map_size;
+
+	/* free all the transaction blocks */
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		if ((tdb->transaction->blocks != NULL) &&
+		    tdb->transaction->blocks[i] != NULL) {
+			free(tdb->transaction->blocks[i]);
+		}
+	}
+	SAFE_FREE(tdb->transaction->blocks);
+
+	if (tdb->transaction->magic_offset) {
+		const struct tdb_methods *methods = tdb->transaction->io_methods;
+		const uint32_t invalid = TDB_RECOVERY_INVALID_MAGIC;
+
+		/* remove the recovery marker */
+		if (methods->tdb_write(tdb, tdb->transaction->magic_offset, &invalid, 4) == -1 ||
+		transaction_sync(tdb, tdb->transaction->magic_offset, 4) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_cancel: failed to remove recovery magic\n"));
+			ret = -1;
+		}
+	}
+
+	/* This also removes the OPEN_LOCK, if we have it. */
+	tdb_release_transaction_locks(tdb);
+
+	/* restore the normal io methods */
+	tdb->methods = tdb->transaction->io_methods;
+
+	SAFE_FREE(tdb->transaction->hash_heads);
+	SAFE_FREE(tdb->transaction);
+
+	return ret;
+}
+
+/*
+  cancel the current transaction
+*/
+_PUBLIC_ int tdb_transaction_cancel(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_transaction_cancel");
+	return _tdb_transaction_cancel(tdb);
+}
+
+/*
+  work out how much space the linearised recovery data will consume
+*/
+static bool tdb_recovery_size(struct tdb_context *tdb, tdb_len_t *result)
+{
+	tdb_len_t recovery_size = 0;
+	uint32_t i;
+
+	recovery_size = sizeof(uint32_t);
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		tdb_len_t block_size;
+		if (i * tdb->transaction->block_size >= tdb->transaction->old_map_size) {
+			break;
+		}
+		if (tdb->transaction->blocks[i] == NULL) {
+			continue;
+		}
+		if (!tdb_add_len_t(recovery_size, 2*sizeof(tdb_off_t),
+				   &recovery_size)) {
+			return false;
+		}
+		if (i == tdb->transaction->num_blocks-1) {
+			block_size = tdb->transaction->last_block_size;
+		} else {
+			block_size =  tdb->transaction->block_size;
+		}
+		if (!tdb_add_len_t(recovery_size, block_size,
+				   &recovery_size)) {
+			return false;
+		}
+	}
+
+	*result = recovery_size;
+	return true;
+}
+
+int tdb_recovery_area(struct tdb_context *tdb,
+		      const struct tdb_methods *methods,
+		      tdb_off_t *recovery_offset,
+		      struct tdb_record *rec)
+{
+	int ret;
+
+	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, recovery_offset) == -1) {
+		return -1;
+	}
+
+	if (*recovery_offset == 0) {
+		rec->rec_len = 0;
+		return 0;
+	}
+
+	if (methods->tdb_read(tdb, *recovery_offset, rec, sizeof(*rec),
+			      DOCONV()) == -1) {
+		return -1;
+	}
+
+	/* ignore invalid recovery regions: can happen in crash */
+	if (rec->magic != TDB_RECOVERY_MAGIC &&
+	    rec->magic != TDB_RECOVERY_INVALID_MAGIC) {
+		*recovery_offset = 0;
+		rec->rec_len = 0;
+	}
+
+	ret = methods->tdb_oob(tdb, *recovery_offset, rec->rec_len, 1);
+	if (ret == -1) {
+		*recovery_offset = 0;
+		rec->rec_len = 0;
+	}
+
+	return 0;
+}
+
+/*
+  allocate the recovery area, or use an existing recovery area if it is
+  large enough
+*/
+static int tdb_recovery_allocate(struct tdb_context *tdb,
+				 tdb_len_t *recovery_size,
+				 tdb_off_t *recovery_offset,
+				 tdb_len_t *recovery_max_size)
+{
+	struct tdb_record rec;
+	const struct tdb_methods *methods = tdb->transaction->io_methods;
+	tdb_off_t recovery_head, new_end;
+
+	if (tdb_recovery_area(tdb, methods, &recovery_head, &rec) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery head\n"));
+		return -1;
+	}
+
+	if (!tdb_recovery_size(tdb, recovery_size)) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: "
+			 "overflow recovery size\n"));
+		return -1;
+	}
+
+	/* Existing recovery area? */
+	if (recovery_head != 0 && *recovery_size <= rec.rec_len) {
+		/* it fits in the existing area */
+		*recovery_max_size = rec.rec_len;
+		*recovery_offset = recovery_head;
+		return 0;
+	}
+
+	/* If recovery area in middle of file, we need a new one. */
+	if (recovery_head == 0
+	    || recovery_head + sizeof(rec) + rec.rec_len != tdb->map_size) {
+		/* we need to free up the old recovery area, then allocate a
+		   new one at the end of the file. Note that we cannot use
+		   tdb_allocate() to allocate the new one as that might return
+		   us an area that is being currently used (as of the start of
+		   the transaction) */
+		if (recovery_head) {
+			if (tdb_free(tdb, recovery_head, &rec) == -1) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL,
+					 "tdb_recovery_allocate: failed to"
+					 " free previous recovery area\n"));
+				return -1;
+			}
+
+			/* the tdb_free() call might have increased
+			 * the recovery size */
+			if (!tdb_recovery_size(tdb, recovery_size)) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL,
+					 "tdb_recovery_allocate: "
+					 "overflow recovery size\n"));
+				return -1;
+			}
+		}
+
+		/* New head will be at end of file. */
+		recovery_head = tdb->map_size;
+	}
+
+	/* Now we know where it will be. */
+	*recovery_offset = recovery_head;
+
+	/* Expand by more than we need, so we don't do it often. */
+	*recovery_max_size = tdb_expand_adjust(tdb->map_size,
+					       *recovery_size,
+					       tdb->page_size)
+		- sizeof(rec);
+
+	if (!tdb_add_off_t(recovery_head, sizeof(rec), &new_end) ||
+	    !tdb_add_off_t(new_end, *recovery_max_size, &new_end)) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: "
+			 "overflow recovery area\n"));
+		return -1;
+	}
+
+	if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size,
+				     new_end - tdb->transaction->old_map_size)
+	    == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to create recovery area\n"));
+		return -1;
+	}
+
+	/* remap the file (if using mmap) */
+	methods->tdb_oob(tdb, tdb->map_size, 1, 1);
+
+	/* we have to reset the old map size so that we don't try to expand the file
+	   again in the transaction commit, which would destroy the recovery area */
+	tdb->transaction->old_map_size = tdb->map_size;
+
+	/* write the recovery header offset and sync - we can sync without a race here
+	   as the magic ptr in the recovery record has not been set */
+	CONVERT(recovery_head);
+	if (methods->tdb_write(tdb, TDB_RECOVERY_HEAD,
+			       &recovery_head, sizeof(tdb_off_t)) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
+		return -1;
+	}
+	if (transaction_write_existing(tdb, TDB_RECOVERY_HEAD, &recovery_head, sizeof(tdb_off_t)) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
+		return -1;
+	}
+
+	return 0;
+}
+
+
+/*
+  setup the recovery data that will be used on a crash during commit
+*/
+static int transaction_setup_recovery(struct tdb_context *tdb,
+				      tdb_off_t *magic_offset)
+{
+	tdb_len_t recovery_size;
+	unsigned char *data, *p;
+	const struct tdb_methods *methods = tdb->transaction->io_methods;
+	struct tdb_record *rec;
+	tdb_off_t recovery_offset, recovery_max_size;
+	tdb_off_t old_map_size = tdb->transaction->old_map_size;
+	uint32_t magic, tailer;
+	uint32_t i;
+
+	/*
+	  check that the recovery area has enough space
+	*/
+	if (tdb_recovery_allocate(tdb, &recovery_size,
+				  &recovery_offset, &recovery_max_size) == -1) {
+		return -1;
+	}
+
+	rec = malloc(recovery_size + sizeof(*rec));
+	if (rec == NULL) {
+		tdb->ecode = TDB_ERR_OOM;
+		return -1;
+	}
+
+	memset(rec, 0, sizeof(*rec));
+
+	rec->magic    = TDB_RECOVERY_INVALID_MAGIC;
+	rec->data_len = recovery_size;
+	rec->rec_len  = recovery_max_size;
+	rec->key_len  = old_map_size;
+	CONVERT(*rec);
+
+	data = (unsigned char *)rec;
+
+	/* build the recovery data into a single blob to allow us to do a single
+	   large write, which should be more efficient */
+	p = data + sizeof(*rec);
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		tdb_off_t offset;
+		tdb_len_t length;
+
+		if (tdb->transaction->blocks[i] == NULL) {
+			continue;
+		}
+
+		offset = i * tdb->transaction->block_size;
+		length = tdb->transaction->block_size;
+		if (i == tdb->transaction->num_blocks-1) {
+			length = tdb->transaction->last_block_size;
+		}
+
+		if (offset >= old_map_size) {
+			continue;
+		}
+		if (offset + length > tdb->transaction->old_map_size) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: transaction data over new region boundary\n"));
+			free(data);
+			tdb->ecode = TDB_ERR_CORRUPT;
+			return -1;
+		}
+		memcpy(p, &offset, 4);
+		memcpy(p+4, &length, 4);
+		if (DOCONV()) {
+			tdb_convert(p, 8);
+		}
+		/* the recovery area contains the old data, not the
+		   new data, so we have to call the original tdb_read
+		   method to get it */
+		if (methods->tdb_read(tdb, offset, p + 8, length, 0) != 0) {
+			free(data);
+			tdb->ecode = TDB_ERR_IO;
+			return -1;
+		}
+		p += 8 + length;
+	}
+
+	/* and the tailer */
+	tailer = sizeof(*rec) + recovery_max_size;
+	memcpy(p, &tailer, 4);
+	if (DOCONV()) {
+		tdb_convert(p, 4);
+	}
+
+	/* write the recovery data to the recovery area */
+	if (methods->tdb_write(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery data\n"));
+		free(data);
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+	if (transaction_write_existing(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery data\n"));
+		free(data);
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* as we don't have ordered writes, we have to sync the recovery
+	   data before we update the magic to indicate that the recovery
+	   data is present */
+	if (transaction_sync(tdb, recovery_offset, sizeof(*rec) + recovery_size) == -1) {
+		free(data);
+		return -1;
+	}
+
+	free(data);
+
+	magic = TDB_RECOVERY_MAGIC;
+	CONVERT(magic);
+
+	*magic_offset = recovery_offset + offsetof(struct tdb_record, magic);
+
+	if (methods->tdb_write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery magic\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+	if (transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery magic\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* ensure the recovery magic marker is on disk */
+	if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
+		return -1;
+	}
+
+	return 0;
+}
+
+static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
+{
+	const struct tdb_methods *methods;
+
+	if (tdb->transaction == NULL) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: no transaction\n"));
+		return -1;
+	}
+
+	if (tdb->transaction->prepared) {
+		tdb->ecode = TDB_ERR_EINVAL;
+		_tdb_transaction_cancel(tdb);
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction already prepared\n"));
+		return -1;
+	}
+
+	if (tdb->transaction->transaction_error) {
+		tdb->ecode = TDB_ERR_IO;
+		_tdb_transaction_cancel(tdb);
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction error pending\n"));
+		return -1;
+	}
+
+
+	if (tdb->transaction->nesting != 0) {
+		return 0;
+	}
+
+	/* check for a null transaction */
+	if (tdb->transaction->blocks == NULL) {
+		return 0;
+	}
+
+	methods = tdb->transaction->io_methods;
+
+	/* if there are any locks pending then the caller has not
+	   nested their locks properly, so fail the transaction */
+	if (tdb_have_extra_locks(tdb)) {
+		tdb->ecode = TDB_ERR_LOCK;
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: locks pending on commit\n"));
+		_tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	/* upgrade the main transaction lock region to a write lock */
+	if (tdb_allrecord_upgrade(tdb) == -1) {
+		if (tdb->ecode == TDB_ERR_RDONLY && tdb->read_only) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "tdb_transaction_prepare_commit: "
+				 "failed to upgrade hash locks: "
+				 "database is read only\n"));
+		} else if (tdb->ecode == TDB_ERR_RDONLY
+			   && tdb->traverse_read) {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "tdb_transaction_prepare_commit: "
+				 "failed to upgrade hash locks: "
+				 "a database traverse is in progress\n"));
+		} else {
+			TDB_LOG((tdb, TDB_DEBUG_ERROR,
+				 "tdb_transaction_prepare_commit: "
+				 "failed to upgrade hash locks: %s\n",
+				 tdb_errorstr(tdb)));
+		}
+		_tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	/* get the open lock - this prevents new users attaching to the database
+	   during the commit */
+	if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: failed to get open lock\n"));
+		_tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	/* write the recovery data to the end of the file */
+	if (transaction_setup_recovery(tdb, &tdb->transaction->magic_offset) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: failed to setup recovery data\n"));
+		_tdb_transaction_cancel(tdb);
+		return -1;
+	}
+
+	tdb->transaction->prepared = true;
+
+	/* expand the file to the new size if needed */
+	if (tdb->map_size != tdb->transaction->old_map_size) {
+		if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size,
+					     tdb->map_size -
+					     tdb->transaction->old_map_size) == -1) {
+			tdb->ecode = TDB_ERR_IO;
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: expansion failed\n"));
+			_tdb_transaction_cancel(tdb);
+			return -1;
+		}
+		tdb->map_size = tdb->transaction->old_map_size;
+		methods->tdb_oob(tdb, tdb->map_size, 1, 1);
+	}
+
+	/* Keep the open lock until the actual commit */
+
+	return 0;
+}
+
+/*
+   prepare to commit the current transaction
+*/
+_PUBLIC_ int tdb_transaction_prepare_commit(struct tdb_context *tdb)
+{
+	tdb_trace(tdb, "tdb_transaction_prepare_commit");
+	return _tdb_transaction_prepare_commit(tdb);
+}
+
+/* A repack is worthwhile if the largest is less than half total free. */
+static bool repack_worthwhile(struct tdb_context *tdb)
+{
+	tdb_off_t ptr;
+	struct tdb_record rec;
+	tdb_len_t total = 0, largest = 0;
+
+	if (tdb_ofs_read(tdb, FREELIST_TOP, &ptr) == -1) {
+		return false;
+	}
+
+	while (ptr != 0 && tdb_rec_free_read(tdb, ptr, &rec) == 0) {
+		total += rec.rec_len;
+		if (rec.rec_len > largest) {
+			largest = rec.rec_len;
+		}
+		ptr = rec.next;
+	}
+
+	return total > largest * 2;
+}
+
+/*
+  commit the current transaction
+*/
+_PUBLIC_ int tdb_transaction_commit(struct tdb_context *tdb)
+{
+	const struct tdb_methods *methods;
+	uint32_t i;
+	bool need_repack = false;
+
+	if (tdb->transaction == NULL) {
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: no transaction\n"));
+		return -1;
+	}
+
+	tdb_trace(tdb, "tdb_transaction_commit");
+
+	if (tdb->transaction->transaction_error) {
+		tdb->ecode = TDB_ERR_IO;
+		_tdb_transaction_cancel(tdb);
+		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: transaction error pending\n"));
+		return -1;
+	}
+
+
+	if (tdb->transaction->nesting != 0) {
+		tdb->transaction->nesting--;
+		return 0;
+	}
+
+	/* check for a null transaction */
+	if (tdb->transaction->blocks == NULL) {
+		_tdb_transaction_cancel(tdb);
+		return 0;
+	}
+
+	if (!tdb->transaction->prepared) {
+		int ret = _tdb_transaction_prepare_commit(tdb);
+		if (ret)
+			return ret;
+	}
+
+	methods = tdb->transaction->io_methods;
+
+	/* perform all the writes */
+	for (i=0;i<tdb->transaction->num_blocks;i++) {
+		tdb_off_t offset;
+		tdb_len_t length;
+
+		if (tdb->transaction->blocks[i] == NULL) {
+			continue;
+		}
+
+		offset = i * tdb->transaction->block_size;
+		length = tdb->transaction->block_size;
+		if (i == tdb->transaction->num_blocks-1) {
+			length = tdb->transaction->last_block_size;
+		}
+
+		if (methods->tdb_write(tdb, offset, tdb->transaction->blocks[i], length) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed during commit\n"));
+
+			/* we've overwritten part of the data and
+			   possibly expanded the file, so we need to
+			   run the crash recovery code */
+			tdb->methods = methods;
+			tdb_transaction_recover(tdb);
+
+			_tdb_transaction_cancel(tdb);
+
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
+			return -1;
+		}
+		SAFE_FREE(tdb->transaction->blocks[i]);
+	}
+
+	/* Do this before we drop lock or blocks. */
+	if (tdb->transaction->expanded) {
+		need_repack = repack_worthwhile(tdb);
+	}
+
+	SAFE_FREE(tdb->transaction->blocks);
+	tdb->transaction->num_blocks = 0;
+
+	/* ensure the new data is on disk */
+	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
+		return -1;
+	}
+
+	/*
+	  TODO: maybe write to some dummy hdr field, or write to magic
+	  offset without mmap, before the last sync, instead of the
+	  utime() call
+	*/
+
+	/* on some systems (like Linux 2.6.x) changes via mmap/msync
+	   don't change the mtime of the file, this means the file may
+	   not be backed up (as tdb rounding to block sizes means that
+	   file size changes are quite rare too). The following forces
+	   mtime changes when a transaction completes */
+#ifdef HAVE_UTIME
+	utime(tdb->name, NULL);
+#endif
+
+	/* use a transaction cancel to free memory and remove the
+	   transaction locks */
+	_tdb_transaction_cancel(tdb);
+
+	if (need_repack) {
+		int ret = tdb_repack(tdb);
+		if (ret != 0) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL,
+				 __location__ " Failed to repack database (not fatal)\n"));
+		}
+		/*
+		 * Ignore the error.
+		 *
+		 * Why?
+		 *
+		 * We just committed to the DB above, so anything
+		 * written during the transaction is committed, the
+		 * caller needs to know that the long-term state was
+		 * successfully modified.
+		 *
+		 * tdb_repack is an optimization that can fail for
+		 * reasons like lock ordering and we cannot recover
+		 * the transaction lock at this point, having released
+		 * it above.
+		 *
+		 * If we return a failure the caller thinks the
+		 * transaction was rolled back.
+		 */
+	}
+
+	return 0;
+}
+
+
+/*
+  recover from an aborted transaction. Must be called with exclusive
+  database write access already established (including the open
+  lock to prevent new processes attaching)
+*/
+int tdb_transaction_recover(struct tdb_context *tdb)
+{
+	tdb_off_t recovery_head, recovery_eof;
+	unsigned char *data, *p;
+	uint32_t zero = 0;
+	struct tdb_record rec;
+
+	/* find the recovery area */
+	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery head\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	if (recovery_head == 0) {
+		/* we have never allocated a recovery record */
+		return 0;
+	}
+
+	/* read the recovery record */
+	if (tdb->methods->tdb_read(tdb, recovery_head, &rec,
+				   sizeof(rec), DOCONV()) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery record\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	if (rec.magic != TDB_RECOVERY_MAGIC) {
+		/* there is no valid recovery data */
+		return 0;
+	}
+
+	if (tdb->read_only) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: attempt to recover read only database\n"));
+		tdb->ecode = TDB_ERR_CORRUPT;
+		return -1;
+	}
+
+	recovery_eof = rec.key_len;
+
+	data = (unsigned char *)malloc(rec.data_len);
+	if (data == NULL) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to allocate recovery data\n"));
+		tdb->ecode = TDB_ERR_OOM;
+		return -1;
+	}
+
+	/* read the full recovery data */
+	if (tdb->methods->tdb_read(tdb, recovery_head + sizeof(rec), data,
+				   rec.data_len, 0) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery data\n"));
+		tdb->ecode = TDB_ERR_IO;
+		free(data);
+		return -1;
+	}
+
+	/* recover the file data */
+	p = data;
+	while (p+8 < data + rec.data_len) {
+		uint32_t ofs, len;
+		if (DOCONV()) {
+			tdb_convert(p, 8);
+		}
+		memcpy(&ofs, p, 4);
+		memcpy(&len, p+4, 4);
+
+		if (tdb->methods->tdb_write(tdb, ofs, p+8, len) == -1) {
+			free(data);
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to recover %u bytes at offset %u\n", len, ofs));
+			tdb->ecode = TDB_ERR_IO;
+			return -1;
+		}
+		p += 8 + len;
+	}
+
+	free(data);
+
+	if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync recovery\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	/* if the recovery area is after the recovered eof then remove it */
+	if (recovery_eof <= recovery_head) {
+		if (tdb_ofs_write(tdb, TDB_RECOVERY_HEAD, &zero) == -1) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery head\n"));
+			tdb->ecode = TDB_ERR_IO;
+			return -1;
+		}
+	}
+
+	/* remove the recovery magic */
+	if (tdb_ofs_write(tdb, recovery_head + offsetof(struct tdb_record, magic),
+			  &zero) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery magic\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	if (transaction_sync(tdb, 0, recovery_eof) == -1) {
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync2 recovery\n"));
+		tdb->ecode = TDB_ERR_IO;
+		return -1;
+	}
+
+	TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_recover: recovered %u byte database\n",
+		 recovery_eof));
+
+	/* all done */
+	return 0;
+}
+
+/* Any I/O failures we say "needs recovery". */
+bool tdb_needs_recovery(struct tdb_context *tdb)
+{
+	tdb_off_t recovery_head;
+	struct tdb_record rec;
+
+	/* find the recovery area */
+	if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
+		return true;
+	}
+
+	if (recovery_head == 0) {
+		/* we have never allocated a recovery record */
+		return false;
+	}
+
+	/* read the recovery record */
+	if (tdb->methods->tdb_read(tdb, recovery_head, &rec,
+				   sizeof(rec), DOCONV()) == -1) {
+		return true;
+	}
+
+	return (rec.magic == TDB_RECOVERY_MAGIC);
+}
diff --git a/lib/tdb/common/traverse.c b/lib/tdb/common/traverse.c
new file mode 100644
index 0000000..d69e7df
--- /dev/null
+++ b/lib/tdb/common/traverse.c
@@ -0,0 +1,510 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "tdb_private.h"
+
+#define TDB_NEXT_LOCK_ERR ((tdb_off_t)-1)
+
+/* Uses traverse lock: 0 = finish, TDB_NEXT_LOCK_ERR = error,
+   other = record offset */
+static tdb_off_t tdb_next_lock(struct tdb_context *tdb, struct tdb_traverse_lock *tlock,
+			 struct tdb_record *rec)
+{
+	int want_next = (tlock->off != 0);
+
+	/* Lock each chain from the start one. */
+	for (; tlock->list < tdb->hash_size; tlock->list++) {
+		if (!tlock->off && tlock->list != 0) {
+			/* this is an optimisation for the common case where
+			   the hash chain is empty, which is particularly
+			   common for the use of tdb with ldb, where large
+			   hashes are used. In that case we spend most of our
+			   time in tdb_brlock(), locking empty hash chains.
+
+			   To avoid this, we do an unlocked pre-check to see
+			   if the hash chain is empty before starting to look
+			   inside it. If it is empty then we can avoid that
+			   hash chain. If it isn't empty then we can't believe
+			   the value we get back, as we read it without a
+			   lock, so instead we get the lock and re-fetch the
+			   value below.
+
+			   Notice that not doing this optimisation on the
+			   first hash chain is critical. We must guarantee
+			   that we have done at least one fcntl lock at the
+			   start of a search to guarantee that memory is
+			   coherent on SMP systems. If records are added by
+			   others during the search then thats OK, and we
+			   could possibly miss those with this trick, but we
+			   could miss them anyway without this trick, so the
+			   semantics don't change.
+
+			   With a non-indexed ldb search this trick gains us a
+			   factor of around 80 in speed on a linux 2.6.x
+			   system (testing using ldbtest).
+			*/
+			tdb->methods->next_hash_chain(tdb, &tlock->list);
+			if (tlock->list == tdb->hash_size) {
+				continue;
+			}
+		}
+
+		if (tdb_lock(tdb, tlock->list, tlock->lock_rw) == -1)
+			return TDB_NEXT_LOCK_ERR;
+
+		/* No previous record?  Start at top of chain. */
+		if (!tlock->off) {
+			if (tdb_ofs_read(tdb, TDB_HASH_TOP(tlock->list),
+				     &tlock->off) == -1)
+				goto fail;
+		} else {
+			/* Otherwise unlock the previous record. */
+			if (tdb_unlock_record(tdb, tlock->off) != 0)
+				goto fail;
+		}
+
+		if (want_next) {
+			/* We have offset of old record: grab next */
+			if (tdb_rec_read(tdb, tlock->off, rec) == -1)
+				goto fail;
+			tlock->off = rec->next;
+		}
+
+		/* Iterate through chain */
+		while( tlock->off) {
+			if (tdb_rec_read(tdb, tlock->off, rec) == -1)
+				goto fail;
+
+			/* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
+			if (tlock->off == rec->next) {
+				tdb->ecode = TDB_ERR_CORRUPT;
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: loop detected.\n"));
+				goto fail;
+			}
+
+			if (!TDB_DEAD(rec)) {
+				/* Woohoo: we found one! */
+				if (tdb_lock_record(tdb, tlock->off) != 0)
+					goto fail;
+				return tlock->off;
+			}
+
+			tlock->off = rec->next;
+		}
+		tdb_unlock(tdb, tlock->list, tlock->lock_rw);
+		want_next = 0;
+	}
+	/* We finished iteration without finding anything */
+	tdb->ecode = TDB_SUCCESS;
+	return 0;
+
+ fail:
+	tlock->off = 0;
+	if (tdb_unlock(tdb, tlock->list, tlock->lock_rw) != 0)
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: On error unlock failed!\n"));
+	return TDB_NEXT_LOCK_ERR;
+}
+
+/* traverse the entire database - calling fn(tdb, key, data) on each element.
+   return -1 on error or the record count traversed
+   if fn is NULL then it is not called
+   a non-zero return value from fn() indicates that the traversal should stop
+  */
+static int tdb_traverse_internal(struct tdb_context *tdb,
+				 tdb_traverse_func fn, void *private_data,
+				 struct tdb_traverse_lock *tl)
+{
+	TDB_DATA key, dbuf;
+	struct tdb_record rec;
+	int ret = 0, count = 0;
+	tdb_off_t off;
+	size_t recbuf_len;
+
+	recbuf_len = 4096;
+	key.dptr = malloc(recbuf_len);
+	if (key.dptr == NULL) {
+		return -1;
+	}
+
+	/* This was in the initialization, above, but the IRIX compiler
+	 * did not like it.  crh
+	 */
+	tl->next = tdb->travlocks.next;
+
+	/* fcntl locks don't stack: beware traverse inside traverse */
+	tdb->travlocks.next = tl;
+
+	/* tdb_next_lock places locks on the record returned, and its chain */
+	while ((off = tdb_next_lock(tdb, tl, &rec)) != 0) {
+		tdb_len_t full_len;
+		int nread;
+
+		if (off == TDB_NEXT_LOCK_ERR) {
+			ret = -1;
+			goto out;
+		}
+
+		full_len = rec.key_len + rec.data_len;
+
+		if (full_len > recbuf_len) {
+			recbuf_len = full_len;
+
+			/*
+			 * No realloc, we don't need the old data and thus can
+			 * do without the memcpy
+			 */
+			free(key.dptr);
+			key.dptr = malloc(recbuf_len);
+
+			if (key.dptr == NULL) {
+				ret = -1;
+				if (tdb_unlock(tdb, tl->list, tl->lock_rw)
+				    != 0) {
+					goto out;
+				}
+				if (tdb_unlock_record(tdb, tl->off) != 0) {
+					TDB_LOG((tdb, TDB_DEBUG_FATAL,
+						 "tdb_traverse: malloc "
+						 "failed and unlock_record "
+						 "failed!\n"));
+				}
+				goto out;
+			}
+		}
+
+		count++;
+		/* now read the full record */
+		nread = tdb->methods->tdb_read(tdb, tl->off + sizeof(rec),
+					       key.dptr, full_len, 0);
+		if (nread == -1) {
+			ret = -1;
+			if (tdb_unlock(tdb, tl->list, tl->lock_rw) != 0)
+				goto out;
+			if (tdb_unlock_record(tdb, tl->off) != 0)
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
+			goto out;
+		}
+		key.dsize = rec.key_len;
+		dbuf.dptr = key.dptr + rec.key_len;
+		dbuf.dsize = rec.data_len;
+
+		tdb_trace_1rec_retrec(tdb, "traverse", key, dbuf);
+
+		/* Drop chain lock, call out */
+		if (tdb_unlock(tdb, tl->list, tl->lock_rw) != 0) {
+			ret = -1;
+			goto out;
+		}
+		if (fn && fn(tdb, key, dbuf, private_data)) {
+			/* They want us to terminate traversal */
+			tdb_trace_ret(tdb, "tdb_traverse_end", count);
+			if (tdb_unlock_record(tdb, tl->off) != 0) {
+				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: unlock_record failed!\n"));;
+				ret = -1;
+			}
+			goto out;
+		}
+	}
+	tdb_trace(tdb, "tdb_traverse_end");
+out:
+	SAFE_FREE(key.dptr);
+	tdb->travlocks.next = tl->next;
+	if (ret < 0)
+		return -1;
+	else
+		return count;
+}
+
+
+/*
+  a read style traverse - temporarily marks each record read only
+*/
+_PUBLIC_ int tdb_traverse_read(struct tdb_context *tdb,
+		      tdb_traverse_func fn, void *private_data)
+{
+	struct tdb_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
+	int ret;
+
+	tdb->traverse_read++;
+	tdb_trace(tdb, "tdb_traverse_read_start");
+	ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
+	tdb->traverse_read--;
+
+	return ret;
+}
+
+/*
+  a write style traverse - needs to get the transaction lock to
+  prevent deadlocks
+
+  WARNING: The data buffer given to the callback fn does NOT meet the
+  alignment guarantees malloc gives you.
+*/
+_PUBLIC_ int tdb_traverse(struct tdb_context *tdb,
+		 tdb_traverse_func fn, void *private_data)
+{
+	struct tdb_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
+	enum tdb_lock_flags lock_flags;
+	int ret;
+
+	if (tdb->read_only || tdb->traverse_read) {
+		return tdb_traverse_read(tdb, fn, private_data);
+	}
+
+	lock_flags = TDB_LOCK_WAIT;
+
+	if (tdb->allrecord_lock.count != 0) {
+		/*
+		 * This avoids a deadlock between tdb_lockall() and
+		 * tdb_traverse(). See
+		 * https://bugzilla.samba.org/show_bug.cgi?id=11381
+		 */
+		lock_flags = TDB_LOCK_NOWAIT;
+	}
+
+	if (tdb_transaction_lock(tdb, F_WRLCK, lock_flags)) {
+		return -1;
+	}
+
+	tdb->traverse_write++;
+	tdb_trace(tdb, "tdb_traverse_start");
+	ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
+	tdb->traverse_write--;
+
+	tdb_transaction_unlock(tdb, F_WRLCK);
+
+	return ret;
+}
+
+
+/* find the first entry in the database and return its key */
+_PUBLIC_ TDB_DATA tdb_firstkey(struct tdb_context *tdb)
+{
+	TDB_DATA key;
+	struct tdb_record rec;
+	tdb_off_t off;
+
+	/* release any old lock */
+	if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0)
+		return tdb_null;
+	tdb->travlocks.off = tdb->travlocks.list = 0;
+	tdb->travlocks.lock_rw = F_RDLCK;
+
+	/* Grab first record: locks chain and returned record. */
+	off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
+	if (off == 0 || off == TDB_NEXT_LOCK_ERR) {
+		tdb_trace_retrec(tdb, "tdb_firstkey", tdb_null);
+		return tdb_null;
+	}
+	/* now read the key */
+	key.dsize = rec.key_len;
+	key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
+
+	tdb_trace_retrec(tdb, "tdb_firstkey", key);
+
+	/* Unlock the hash chain of the record we just read. */
+	if (tdb_unlock(tdb, tdb->travlocks.list, tdb->travlocks.lock_rw) != 0)
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
+	return key;
+}
+
+/* find the next entry in the database, returning its key */
+_PUBLIC_ TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
+{
+	uint32_t oldlist;
+	TDB_DATA key = tdb_null;
+	struct tdb_record rec;
+	unsigned char *k = NULL;
+	tdb_off_t off;
+
+	/* Is locked key the old key?  If so, traverse will be reliable. */
+	if (tdb->travlocks.off) {
+		if (tdb_lock(tdb,tdb->travlocks.list,tdb->travlocks.lock_rw))
+			return tdb_null;
+		if (tdb_rec_read(tdb, tdb->travlocks.off, &rec) == -1
+		    || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
+					    rec.key_len))
+		    || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
+			/* No, it wasn't: unlock it and start from scratch */
+			if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0) {
+				tdb_trace_1rec_retrec(tdb, "tdb_nextkey",
+						      oldkey, tdb_null);
+				SAFE_FREE(k);
+				return tdb_null;
+			}
+			if (tdb_unlock(tdb, tdb->travlocks.list, tdb->travlocks.lock_rw) != 0) {
+				SAFE_FREE(k);
+				return tdb_null;
+			}
+			tdb->travlocks.off = 0;
+		}
+
+		SAFE_FREE(k);
+	}
+
+	if (!tdb->travlocks.off) {
+		/* No previous element: do normal find, and lock record */
+		tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), tdb->travlocks.lock_rw, &rec);
+		if (!tdb->travlocks.off) {
+			tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, tdb_null);
+			return tdb_null;
+		}
+		tdb->travlocks.list = BUCKET(rec.full_hash);
+		if (tdb_lock_record(tdb, tdb->travlocks.off) != 0) {
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
+			return tdb_null;
+		}
+	}
+	oldlist = tdb->travlocks.list;
+
+	/* Grab next record: locks chain and returned record,
+	   unlocks old record */
+	off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
+	if (off != TDB_NEXT_LOCK_ERR && off != 0) {
+		key.dsize = rec.key_len;
+		key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
+					  key.dsize);
+		/* Unlock the chain of this new record */
+		if (tdb_unlock(tdb, tdb->travlocks.list, tdb->travlocks.lock_rw) != 0)
+			TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+	}
+	/* Unlock the chain of old record */
+	if (tdb_unlock(tdb, oldlist, tdb->travlocks.lock_rw) != 0)
+		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+	tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, key);
+	return key;
+}
+
+_PUBLIC_ int tdb_traverse_chain(struct tdb_context *tdb,
+				unsigned chain,
+				tdb_traverse_func fn,
+				void *private_data)
+{
+	tdb_off_t rec_ptr;
+	struct tdb_chainwalk_ctx chainwalk;
+	int count = 0;
+	int ret;
+
+	if (chain >= tdb->hash_size) {
+		tdb->ecode = TDB_ERR_EINVAL;
+		return -1;
+	}
+
+	if (tdb->traverse_read != 0) {
+		tdb->ecode = TDB_ERR_LOCK;
+		return -1;
+	}
+
+	ret = tdb_lock(tdb, chain, F_RDLCK);
+	if (ret == -1) {
+		return -1;
+	}
+
+	tdb->traverse_read += 1;
+
+	ret = tdb_ofs_read(tdb, TDB_HASH_TOP(chain), &rec_ptr);
+	if (ret == -1) {
+		goto fail;
+	}
+
+	tdb_chainwalk_init(&chainwalk, rec_ptr);
+
+	while (rec_ptr != 0) {
+		struct tdb_record rec;
+		bool ok;
+
+		ret = tdb_rec_read(tdb, rec_ptr, &rec);
+		if (ret == -1) {
+			goto fail;
+		}
+
+		if (!TDB_DEAD(&rec)) {
+			/* no overflow checks, tdb_rec_read checked it */
+			tdb_off_t key_ofs = rec_ptr + sizeof(rec);
+			size_t full_len = rec.key_len + rec.data_len;
+			uint8_t *buf = NULL;
+
+			TDB_DATA key = { .dsize = rec.key_len };
+			TDB_DATA data = { .dsize = rec.data_len };
+
+			if ((tdb->transaction == NULL) &&
+			    (tdb->map_ptr != NULL)) {
+				ret = tdb_oob(tdb, key_ofs, full_len, 0);
+				if (ret == -1) {
+					goto fail;
+				}
+				key.dptr = (uint8_t *)tdb->map_ptr + key_ofs;
+			} else {
+				buf = tdb_alloc_read(tdb, key_ofs, full_len);
+				if (buf == NULL) {
+					goto fail;
+				}
+				key.dptr = buf;
+			}
+			data.dptr = key.dptr + key.dsize;
+
+			ret = fn(tdb, key, data, private_data);
+			free(buf);
+
+			count += 1;
+
+			if (ret != 0) {
+				break;
+			}
+		}
+
+		rec_ptr = rec.next;
+
+		ok = tdb_chainwalk_check(tdb, &chainwalk, rec_ptr);
+		if (!ok) {
+			goto fail;
+		}
+	}
+	tdb->traverse_read -= 1;
+	tdb_unlock(tdb, chain, F_RDLCK);
+	return count;
+
+fail:
+	tdb->traverse_read -= 1;
+	tdb_unlock(tdb, chain, F_RDLCK);
+	return -1;
+}
+
+_PUBLIC_ int tdb_traverse_key_chain(struct tdb_context *tdb,
+				    TDB_DATA key,
+				    tdb_traverse_func fn,
+				    void *private_data)
+{
+	uint32_t hash, chain;
+	int ret;
+
+	hash = tdb->hash_fn(&key);
+	chain = BUCKET(hash);
+	ret = tdb_traverse_chain(tdb, chain, fn, private_data);
+
+	return ret;
+}