diff options
Diffstat (limited to 'test/pendingrace.test')
| -rw-r--r-- | test/pendingrace.test | 123 |
1 files changed, 123 insertions, 0 deletions
diff --git a/test/pendingrace.test b/test/pendingrace.test new file mode 100644 index 0000000..ef42578 --- /dev/null +++ b/test/pendingrace.test @@ -0,0 +1,123 @@ +# 2023 January 31 +# +# The author disclaims copyright to this source code. In place of +# a legal notice, here is a blessing: +# +# May you do good and not evil. +# May you find forgiveness for yourself and forgive others. +# May you share freely, never taking more than you give. +# +#*********************************************************************** +# + +set testdir [file dirname $argv0] +source $testdir/tester.tcl +set testprefix pendingrace + +# This test file tests that a race condition surrounding hot-journal +# rollback that once existed has been resolved. The problem was that +# if, when attempting to upgrade from a SHARED to EXCLUSIVE lock in +# order to roll back a hot journal, a connection failed to take the +# lock, the file-descriptor was left holding a PENDING lock for +# a very short amount of time. In a multi-threaded deployment, this +# could allow a second connection to read the database without rolling +# back the hot journal. +# + +testvfs tvfs +db close +sqlite3 db test.db -vfs tvfs + +# Create a 20 page database using connection [db]. Connection [db] uses +# Tcl VFS wrapper "tvfs", but it is configured to do straight pass-through +# for now. +# +do_execsql_test 1.0 { + PRAGMA cache_size = 5; + CREATE TABLE t1(a, b); + CREATE INDEX i1 ON t1(a, b); + WITH s(i) AS ( + SELECT 1 UNION ALL SELECT i+1 FROM s WHERE i<10 + ) + INSERT INTO t1 SELECT hex(randomblob(100)), hex(randomblob(100)) FROM s; +} {} +do_test 1.1a { + set nPg [db one { PRAGMA page_count }] + expr ($nPg==20 || $nPg==21) +} 1 + +# Simulate a crash in another process. This leaves the db with a hot-journal. +# Without the journal the db is corrupt. +# +sqlite3 db2 test.db +do_execsql_test -db db2 1.1 { + PRAGMA cache_size = 5; + BEGIN; + UPDATE t1 SET b=hex(randomblob(100)); +} +db_save +db2 close +proc my_db_restore {} { + forcecopy sv_test.db-journal test.db-journal + + set fd1 [open sv_test.db r] + fconfigure $fd1 -encoding binary -translation binary + set data [read $fd1] + close $fd1 + + set fd1 [open test.db w] + fconfigure $fd1 -encoding binary -translation binary + puts -nonewline $fd1 $data + close $fd1 +} +my_db_restore +do_test 1.2 { + file exists test.db-journal +} {1} + +# Set up connection [db2] to use Tcl VFS wrapper [tvfs2]. Which is configured +# so that the first call to xUnlock() fails. And then all VFS calls thereafter +# fail as well. +# +testvfs tvfs2 +tvfs2 filter xUnlock +tvfs2 script xUnlock +set ::seen_unlock 0 +proc xUnlock {args} { + if {$::seen_unlock==0} { + set ::seen_unlock 1 + tvfs2 ioerr 1 1 + tvfs2 filter {xLock xUnlock} + } + return "" +} +sqlite3 db2 test.db -vfs tvfs2 + +# Configure [tvfs] (used by [db]) so that within the first call to xAccess, +# [db2] attempts to read the db. This causes [db2] to fail to upgrade to +# EXCLUSIVE, leaving it with a PENDING lock. Which it holds on to, +# as the xUnlock() and all subsequent VFS calls fail. +# +tvfs filter xAccess +tvfs script xAccess +set ::seen_access 0 +proc xAccess {args} { + if {$::seen_access==0} { + set ::seen_access 1 + catch { db2 eval { SELECT count(*)+0 FROM t1 } } + breakpoint + } + return "" +} + +# Run an integrity check using [db]. +do_catchsql_test 1.3 { + PRAGMA integrity_check +} {1 {database is locked}} + +db close +db2 close +tvfs delete +tvfs2 delete + +finish_test |