ISAR/libmdbx
3
0
Fork 0
mirror of https://github.com/isar/libmdbx.git synced 2025-01-31 10:58:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

lmdb: rework workaround for potential ext3/ext4 corruption issue.

Browse Source 
Reworked from branch 'mdb.master' origin OpenLDAP:
  8b6c425 2015-01-12 More cleanup for fdatasync hack
  ea89e3d 2015-01-11 Tweak conditionals for fdatasync hack
  462dc09 2015-01-08 fdatasync hack, again
  e86072a 2015-01-08 Revert "Note MDB_SAFE_FDATASYNC"
  293d6bb 2015-01-08 Note MDB_SAFE_FDATASYNC
  9585c01 2015-01-08 Simpler fdatasync hack
  0ef1e0b 2015-01-08 Revert "Fix prev commit for env_sync0"

Imported early while forking ReOpenLDAP:
  985bbbb 2014-12-21 Fix prev commit for env_sync0
  0018eeb 2014-12-18 Hack for potential ext3/ext4 corruption issue

Change-Id: I187fd320620b9ced2e3773cac96f281ff65f97d4
This commit is contained in:
Leo Yuriev 2015-01-07 19:09:14 +03:00
parent f00d2cdef6
commit 3fd4f9cce0
2 changed files with 95 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Makefile
View File

@ -11,6 +11,7 @@
# - MDB_USE_POSIX_SEM # - MDB_USE_POSIX_SEM
# - MDB_DSYNC # - MDB_DSYNC
# - MDB_FDATASYNC # - MDB_FDATASYNC
# - MDB_FDATASYNC_WORKS
# - MDB_USE_PWRITEV # - MDB_USE_PWRITEV
# #
# There may be other macros in mdb.c of interest. You should # There may be other macros in mdb.c of interest. You should

146
mdb.c
View File

@ -79,6 +79,14 @@ extern int cacheflush(char *addr, int nbytes, int cache);
#define CACHEFLUSH(addr, bytes, cache) #define CACHEFLUSH(addr, bytes, cache)
#endif #endif
#if defined(__linux) && !defined(MDB_FDATASYNC_WORKS)
/** fdatasync is broken on ext3/ext4fs on older kernels, see
* description in #mdb_env_open2 comments. You can safely
* define MDB_FDATASYNC_WORKS if this code will only be run
* on kernels 3.6 and newer.
*/
# define FDATASYNC_MAYBE_BROKEN
#endif
#include <errno.h> #include <errno.h>
#include <limits.h> #include <limits.h>
@ -368,7 +376,6 @@ static int mdb_mutex_failed(MDB_env *env, mdb_mutex_t *mutex, int rc);
*/ */
#ifndef MDB_FDATASYNC #ifndef MDB_FDATASYNC
# define MDB_FDATASYNC fdatasync # define MDB_FDATASYNC fdatasync
# define HAVE_FDATASYNC 1
#endif #endif
#ifndef MDB_MSYNC #ifndef MDB_MSYNC
@ -1142,6 +1149,10 @@ struct MDB_env {
#define MDB_ENV_ACTIVE 0x20000000U #define MDB_ENV_ACTIVE 0x20000000U
/** me_txkey is set */ /** me_txkey is set */
#define MDB_ENV_TXKEY 0x10000000U #define MDB_ENV_TXKEY 0x10000000U
#ifdef FDATASYNC_MAYBE_BROKEN
/** fdatasync may be unreliable */
# define MDB_BROKEN_DATASYNC 0x08000000U
#endif /* FDATASYNC_MAYBE_BROKEN */
uint32_t me_flags; /**< @ref mdb_env */ uint32_t me_flags; /**< @ref mdb_env */
unsigned int me_psize; /**< DB page size, inited from me_os_psize */ unsigned int me_psize; /**< DB page size, inited from me_os_psize */
unsigned int me_os_psize; /**< OS page size, from #GET_PAGESIZE */ unsigned int me_os_psize; /**< OS page size, from #GET_PAGESIZE */
@ -1158,7 +1169,6 @@ struct MDB_env {
MDB_txn *me_txn; /**< current write transaction */ MDB_txn *me_txn; /**< current write transaction */
MDB_txn *me_txn0; /**< prealloc'd write transaction */ MDB_txn *me_txn0; /**< prealloc'd write transaction */
size_t me_mapsize; /**< size of the data memory map */ size_t me_mapsize; /**< size of the data memory map */
size_t me_size; /**< current file size */
pgno_t me_maxpg; /**< me_mapsize / me_psize */ pgno_t me_maxpg; /**< me_mapsize / me_psize */
MDB_dbx *me_dbxs; /**< array of static DB info */ MDB_dbx *me_dbxs; /**< array of static DB info */
uint16_t *me_dbflags; /**< array of flags from MDB_db.md_flags */ uint16_t *me_dbflags; /**< array of flags from MDB_db.md_flags */
@ -1193,7 +1203,9 @@ struct MDB_env {
MDB_assert_func *me_assert_func; /**< Callback for assertion failures */ MDB_assert_func *me_assert_func; /**< Callback for assertion failures */
uint64_t me_sync_pending; /**< Total dirty/commited bytes since the last mdb_env_sync() */ uint64_t me_sync_pending; /**< Total dirty/commited bytes since the last mdb_env_sync() */
uint64_t me_sync_threshold; /**< Treshold of above to force synchronous flush */ uint64_t me_sync_threshold; /**< Treshold of above to force synchronous flush */
size_t me_sync_size; /**< Tracking me_size for FGREW/fsync() */ #ifdef FDATASYNC_MAYBE_BROKEN
size_t me_sync_size; /**< Tracking file size at last sync to decide when fsync() is needed */
#endif /* FDATASYNC_MAYBE_BROKEN */
MDB_oom_func *me_oom_func; /**< Callback for kicking laggard readers */ MDB_oom_func *me_oom_func; /**< Callback for kicking laggard readers */
}; };
@ -2514,34 +2526,18 @@ fail:
return rc; return rc;
} }
/* internal env_sync flags: */
#define FORCE 1 /* as before, force a flush */
#define FGREW 0x8000 /* file has grown, do a full fsync instead of just
fdatasync. We shouldn't have to do this, according to the POSIX spec.
But common Linux FSs violate the spec and won't sync required metadata
correctly when the file grows. This only makes a difference if the
platform actually distinguishes fdatasync from fsync.
http://www.openldap.org/lists/openldap-devel/201411/msg00000.html */
static int static int
mdb_env_sync0(MDB_env *env, unsigned int *flags) mdb_env_sync0(MDB_env *env, int *force)
{ {
int rc = 0, force; int rc = 0;
if (env->me_sync_threshold && env->me_sync_pending >= env->me_sync_threshold) if (env->me_sync_threshold && env->me_sync_pending >= env->me_sync_threshold)
*flags |= FORCE; *force = 1;
force = *flags & FORCE; if (*force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
if (env->me_sync_size != env->me_size)
*flags |= FGREW;
if (env->me_flags & MDB_WRITEMAP) { if (env->me_flags & MDB_WRITEMAP) {
int mode = ((env->me_flags & MDB_MAPASYNC) && !force) int mode = ((env->me_flags & MDB_MAPASYNC) && *force == 0) ? MS_ASYNC : MS_SYNC;
? MS_ASYNC : MS_SYNC;
/* LY: skip meta-pages */
size_t data_offset = env->me_os_psize;
while (data_offset < env->me_psize + env->me_psize)
data_offset += env->me_os_psize;
/* LY: skip meta-pages, sync ones explicit later */
size_t data_offset = (env->me_psize * 2 + env->me_os_psize - 1) & ~(env->me_os_psize - 1);
if (MDB_MSYNC(env->me_map + data_offset, env->me_mapsize - data_offset, mode)) if (MDB_MSYNC(env->me_map + data_offset, env->me_mapsize - data_offset, mode))
rc = ErrCode(); rc = ErrCode();
#ifdef _WIN32 #ifdef _WIN32
@ -2549,20 +2545,22 @@ mdb_env_sync0(MDB_env *env, unsigned int *flags)
rc = ErrCode(); rc = ErrCode();
#endif #endif
} else { } else {
#ifdef HAVE_FDATASYNC /* (LY) TODO: sync_file_range() for data and later fdatasync() for meta,
if (*flags & FGREW) { ALSO sync_file_range() needed before calling fsync().
if (fsync(env->me_fd)) /* Avoid ext-fs bugs, do full sync */ */
#ifdef FDATASYNC_MAYBE_BROKEN
if (env->me_sync_size != env->me_mapsize && (env->me_flags & MDB_BROKEN_DATASYNC)) {
if (fsync(env->me_fd))
rc = ErrCode(); rc = ErrCode();
else
env->me_sync_size = env->me_mapsize;
} else } else
#endif #endif /* FDATASYNC_MAYBE_BROKEN */
if (MDB_FDATASYNC(env->me_fd)) if (MDB_FDATASYNC(env->me_fd))
rc = ErrCode(); rc = ErrCode();
} }
if (! rc) { if (! rc)
env->me_sync_pending = 0; env->me_sync_pending = 0;
if (*flags & FGREW)
env->me_sync_size = env->me_size;
}
} }
return rc; return rc;
} }
@ -2570,7 +2568,6 @@ mdb_env_sync0(MDB_env *env, unsigned int *flags)
int int
mdb_env_sync(MDB_env *env, int force) mdb_env_sync(MDB_env *env, int force)
{ {
unsigned int flags = force ? FORCE | FGREW : 0;
MDB_meta *meta; MDB_meta *meta;
txnid_t checkpoint; txnid_t checkpoint;
int rc, lockfree_countdown = 3; int rc, lockfree_countdown = 3;
@ -2590,7 +2587,7 @@ mdb_env_sync(MDB_env *env, int force)
checkpoint = meta->mm_txnid; checkpoint = meta->mm_txnid;
/* first sync data. */ /* first sync data. */
rc = mdb_env_sync0(env, &flags); rc = mdb_env_sync0(env, &force);
/* then sync meta-pages. */ /* then sync meta-pages. */
if (rc == 0 && (env->me_flags & MDB_WRITEMAP)) { if (rc == 0 && (env->me_flags & MDB_WRITEMAP)) {
@ -2598,7 +2595,7 @@ mdb_env_sync(MDB_env *env, int force)
if (MDB_MSYNC(env->me_map, env->me_psize * 2, mode)) if (MDB_MSYNC(env->me_map, env->me_psize * 2, mode))
rc = ErrCode(); rc = ErrCode();
#ifdef _WIN32 #ifdef _WIN32
if (rc == 0 && mode == MS_SYNC && MDB_FDATASYNC(env->me_fd)) else if (mode == MS_SYNC && MDB_FDATASYNC(env->me_fd))
rc = ErrCode(); rc = ErrCode();
#endif #endif
} }
@ -3657,8 +3654,8 @@ done:
int int
mdb_txn_commit(MDB_txn *txn) mdb_txn_commit(MDB_txn *txn)
{ {
int rc; int rc, force = 0;
unsigned int i; unsigned i;
MDB_env *env; MDB_env *env;
if (txn == NULL || txn->mt_env == NULL) if (txn == NULL || txn->mt_env == NULL)
@ -3862,16 +3859,9 @@ mdb_txn_commit(MDB_txn *txn)
mdb_audit(txn); mdb_audit(txn);
#endif #endif
i = 0;
#ifdef HAVE_FDATASYNC
if (txn->mt_next_pgno * env->me_psize > env->me_size) {
i |= FGREW;
env->me_size = txn->mt_next_pgno * env->me_psize;
}
#endif
if ((rc = mdb_page_flush(txn, 0)) || if ((rc = mdb_page_flush(txn, 0)) ||
(rc = mdb_env_sync0(env, &i)) || (rc = mdb_env_sync0(env, &force)) ||
(rc = mdb_env_write_meta(txn, i != 0))) (rc = mdb_env_write_meta(txn, force)))
goto fail; goto fail;
/* Free P_LOOSE pages left behind in dirty_list */ /* Free P_LOOSE pages left behind in dirty_list */
@ -4372,6 +4362,11 @@ mdb_fsize(HANDLE fd, size_t *size)
return MDB_SUCCESS; return MDB_SUCCESS;
} }
#ifdef FDATASYNC_MAYBE_BROKEN
# include <sys/utsname.h>
# include <sys/vfs.h>
#endif /* FDATASYNC_MAYBE_BROKEN */
/** Further setup required for opening an LMDB environment /** Further setup required for opening an LMDB environment
*/ */
static int ESECT static int ESECT
@ -4390,6 +4385,54 @@ mdb_env_open2(MDB_env *env)
env->me_pidquery = PROCESS_QUERY_INFORMATION; env->me_pidquery = PROCESS_QUERY_INFORMATION;
#endif /* _WIN32 */ #endif /* _WIN32 */
#ifdef FDATASYNC_MAYBE_BROKEN
/* ext3/ext4 fdatasync is broken on some older Linux kernels.
* https://lkml.org/lkml/2012/9/3/83
* Kernels after 3.6-rc6 are known good.
* https://lkml.org/lkml/2012/9/10/556
* See if the DB is on ext3/ext4, then check for new enough kernel
* Kernels 2.6.32.60, 2.6.34.15, 3.2.30, and 3.5.4 are also known
* to be patched.
*/
{
struct statfs st;
fstatfs(env->me_fd, &st);
while (st.f_type == 0xEF53) {
struct utsname uts;
int i;
uname(&uts);
if (uts.release[0] < '3') {
if (!strncmp(uts.release, "2.6.32.", 7)) {
i = atoi(uts.release+7);
if (i >= 60)
break; /* 2.6.32.60 and newer is OK */
} else if (!strncmp(uts.release, "2.6.34.", 7)) {
i = atoi(uts.release+7);
if (i >= 15)
break; /* 2.6.34.15 and newer is OK */
}
} else if (uts.release[0] == '3') {
i = atoi(uts.release+2);
if (i > 5)
break; /* 3.6 and newer is OK */
if (i == 5) {
i = atoi(uts.release+4);
if (i >= 4)
break; /* 3.5.4 and newer is OK */
} else if (i == 2) {
i = atoi(uts.release+4);
if (i >= 30)
break; /* 3.2.30 and newer is OK */
}
} else { /* 4.x and newer is OK */
break;
}
env->me_flags |= MDB_BROKEN_DATASYNC;
break;
}
}
#endif /* FDATASYNC_MAYBE_BROKEN */
if ((i = mdb_env_read_header(env, &meta)) != 0) { if ((i = mdb_env_read_header(env, &meta)) != 0) {
if (i != ENOENT) if (i != ENOENT)
return i; return i;
@ -4433,10 +4476,6 @@ mdb_env_open2(MDB_env *env)
newenv = 0; newenv = 0;
} }
rc = mdb_fsize(env->me_fd, &env->me_size);
if (rc)
return rc;
rc = mdb_env_map(env, (flags & MDB_FIXEDMAP) ? meta.mm_address : NULL); rc = mdb_env_map(env, (flags & MDB_FIXEDMAP) ? meta.mm_address : NULL);
if (rc) if (rc)
return rc; return rc;
@ -4457,6 +4496,9 @@ mdb_env_open2(MDB_env *env)
env->me_maxkey = env->me_nodemax - (NODESIZE + sizeof(MDB_db)); env->me_maxkey = env->me_nodemax - (NODESIZE + sizeof(MDB_db));
#endif #endif
env->me_maxpg = env->me_mapsize / env->me_psize; env->me_maxpg = env->me_mapsize / env->me_psize;
#ifdef FDATASYNC_MAYBE_BROKEN
env->me_sync_size = env->me_mapsize;
#endif /* FDATASYNC_MAYBE_BROKEN */
#if MDB_DEBUG #if MDB_DEBUG
{ {