ITS#8704 add MDB_PREVSNAPSHOT flag to mdb_env_open

used to open the previous snapshot, in case the latest one
is corrupted

external/db_drivers/liblmdb/lmdb.h

@@ -311,6 +311,8 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
 #define MDB_NORDAHEAD	0x800000
 	/** don't initialize malloc'd memory before writing to datafile */
 #define MDB_NOMEMINIT	0x1000000
+	/** use the previous snapshot rather than the latest one */
+#define MDB_PREVSNAPSHOT	0x2000000
 /** @} */
 
 /**	@defgroup	mdb_dbi_open	Database Flags
@@ -622,6 +624,12 @@ int  mdb_env_create(MDB_env **env);
 	 *		caller is expected to overwrite all of the memory that was
 	 *		reserved in that case.
 	 *		This flag may be changed at any time using #mdb_env_set_flags().
+	 *	<li>#MDB_PREVSNAPSHOT
+	 *		Open the environment with the previous snapshot rather than the latest
+	 *		one. This loses the latest transaction, but may help work around some
+	 *		types of corruption. If opened with write access, this must be the
+	 *		only process using the environment. This flag is automatically reset
+	 *		after a write transaction is successfully committed.
 	 * </ul>
 	 * @param[in] mode The UNIX permissions to set on created files and semaphores.
 	 * This parameter is ignored on Windows.

external/db_drivers/liblmdb/mdb.c

@@ -1468,7 +1468,7 @@ static int	mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst);
 static int	mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata,
 				pgno_t newpgno, unsigned int nflags);
 
-static int  mdb_env_read_header(MDB_env *env, MDB_meta *meta);
+static int  mdb_env_read_header(MDB_env *env, int prev, MDB_meta *meta);
 static MDB_meta *mdb_env_pick_meta(const MDB_env *env);
 static int  mdb_env_write_meta(MDB_txn *txn);
 #ifdef MDB_USE_POSIX_MUTEX /* Drop unused excl arg */
@@ -3632,6 +3632,8 @@ done:
 	return MDB_SUCCESS;
 }
 
+static int ESECT mdb_env_share_locks(MDB_env *env, int *excl);
+
 int
 mdb_txn_commit(MDB_txn *txn)
 {
@@ -3854,6 +3856,15 @@ mdb_txn_commit(MDB_txn *txn)
 	if ((rc = mdb_env_write_meta(txn)))
 		goto fail;
 	end_mode = MDB_END_COMMITTED|MDB_END_UPDATE;
+	if (env->me_flags & MDB_PREVSNAPSHOT) {
+		if (!(env->me_flags & MDB_NOLOCK)) {
+			int excl;
+			rc = mdb_env_share_locks(env, &excl);
+			if (rc)
+				goto fail;
+		}
+		env->me_flags ^= MDB_PREVSNAPSHOT;
+	}
 
 done:
 	mdb_txn_end(txn, end_mode);
@@ -3867,11 +3878,12 @@ fail:
 /** Read the environment parameters of a DB environment before
  * mapping it into memory.
  * @param[in] env the environment handle
+ * @param[in] prev whether to read the backup meta page
  * @param[out] meta address of where to store the meta information
  * @return 0 on success, non-zero on failure.
  */
 static int ESECT
-mdb_env_read_header(MDB_env *env, MDB_meta *meta)
+mdb_env_read_header(MDB_env *env, int prev, MDB_meta *meta)
 {
 	MDB_metabuf	pbuf;
 	MDB_page	*p;
@@ -3922,7 +3934,7 @@ mdb_env_read_header(MDB_env *env, MDB_meta *meta)
 			return MDB_VERSION_MISMATCH;
 		}
 
-		if (off == 0 || m->mm_txnid > meta->mm_txnid)
+		if (off == 0 || (prev ? m->mm_txnid < meta->mm_txnid : m->mm_txnid > meta->mm_txnid))
 			*meta = *m;
 	}
 	return 0;
@@ -4131,7 +4143,8 @@ static MDB_meta *
 mdb_env_pick_meta(const MDB_env *env)
 {
 	MDB_meta *const *metas = env->me_metas;
-	return metas[ metas[0]->mm_txnid < metas[1]->mm_txnid ];
+	return metas[ (metas[0]->mm_txnid < metas[1]->mm_txnid) ^
+		((env->me_flags & MDB_PREVSNAPSHOT) != 0) ];
 }
 
 int ESECT
@@ -4366,7 +4379,7 @@ mdb_fsize(HANDLE fd, mdb_size_t *size)
 /** Further setup required for opening an LMDB environment
  */
 static int ESECT
-mdb_env_open2(MDB_env *env)
+mdb_env_open2(MDB_env *env, int prev)
 {
 	unsigned int flags = env->me_flags;
 	int i, newenv = 0, rc;
@@ -4429,7 +4442,7 @@ mdb_env_open2(MDB_env *env)
 	}
 #endif
 
-	if ((i = mdb_env_read_header(env, &meta)) != 0) {
+	if ((i = mdb_env_read_header(env, prev, &meta)) != 0) {
 		if (i != ENOENT)
 			return i;
 		DPUTS("new mdbenv");
@@ -4505,6 +4518,9 @@ mdb_env_open2(MDB_env *env)
 #endif
 	env->me_maxpg = env->me_mapsize / env->me_psize;
 
+	if (env->me_txns)
+		env->me_txns->mti_txnid = meta.mm_txnid;
+
 #if MDB_DEBUG
 	{
 		MDB_meta *meta = mdb_env_pick_meta(env);
@@ -4600,9 +4616,6 @@ static int ESECT
 mdb_env_share_locks(MDB_env *env, int *excl)
 {
 	int rc = 0;
-	MDB_meta *meta = mdb_env_pick_meta(env);
-
-	env->me_txns->mti_txnid = meta->mm_txnid;
 
 #ifdef _WIN32
 	{
@@ -5056,7 +5069,7 @@ fail:
 	 */
 #define	CHANGEABLE	(MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC|MDB_NOMEMINIT)
 #define	CHANGELESS	(MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY| \
-	MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD)
+	MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD|MDB_PREVSNAPSHOT)
 
 #if VALID_FLAGS & PERSISTENT_FLAGS & (CHANGEABLE|CHANGELESS)
 # error "Persistent DB flags & env flags overlap, but both go in mm_flags"
@@ -5178,9 +5191,13 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 		rc = mdb_env_setup_locks(env, lpath, mode, &excl);
 		if (rc)
 			goto leave;
+		if ((flags & MDB_PREVSNAPSHOT) && !excl) {
+			rc = EAGAIN;
+			goto leave;
+		}
 	}
 
-	if ((rc = mdb_env_open2(env)) == MDB_SUCCESS) {
+	if ((rc = mdb_env_open2(env, flags & MDB_PREVSNAPSHOT)) == MDB_SUCCESS) {
 		if (flags & (MDB_RDONLY|MDB_WRITEMAP)) {
 			env->me_mfd = env->me_fd;
 		} else {
@@ -5206,7 +5223,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 			}
 		}
 		DPRINTF(("opened dbenv %p", (void *) env));
-		if (excl > 0) {
+		if (excl > 0 && !(flags & MDB_PREVSNAPSHOT)) {
 			rc = mdb_env_share_locks(env, &excl);
 			if (rc)
 				goto leave;

external/db_drivers/liblmdb/mdb_copy.1

@@ -11,6 +11,8 @@ mdb_copy \- LMDB environment copy tool
 .BR \-c ]
 [\c
 .BR \-n ]
+[\c
+.BR \-v ]
 .B srcpath
 [\c
 .BR dstpath ]
@@ -39,6 +41,10 @@ slow down the backup process as it is more CPU-intensive.
 .TP
 .BR \-n
 Open LDMB environment(s) which do not use subdirectories.
+.TP
+.BR \-v
+Use the previous environment state instead of the latest state.
+This may be useful if the latest state has been corrupted.
 
 .SH DIAGNOSTICS
 Exit status is zero if no errors occur.

external/db_drivers/liblmdb/mdb_copy.c

@@ -38,6 +38,8 @@ int main(int argc,char * argv[])
 	for (; argc > 1 && argv[1][0] == '-'; argc--, argv++) {
 		if (argv[1][1] == 'n' && argv[1][2] == '\0')
 			flags |= MDB_NOSUBDIR;
+		else if (argv[1][1] == 'v' && argv[1][2] == '\0')
+			flags |= MDB_PREVSNAPSHOT;
 		else if (argv[1][1] == 'c' && argv[1][2] == '\0')
 			cpflags |= MDB_CP_COMPACT;
 		else if (argv[1][1] == 'V' && argv[1][2] == '\0') {
@@ -48,7 +50,7 @@ int main(int argc,char * argv[])
 	}
 
 	if (argc<2 || argc>3) {
-		fprintf(stderr, "usage: %s [-V] [-c] [-n] srcpath [dstpath]\n", progname);
+		fprintf(stderr, "usage: %s [-V] [-c] [-n] [-v] srcpath [dstpath]\n", progname);
 		exit(EXIT_FAILURE);
 	}
 

external/db_drivers/liblmdb/mdb_dump.1

@@ -14,6 +14,8 @@ mdb_dump \- LMDB environment export tool
 [\c
 .BR \-n ]
 [\c
+.BR \-v ]
+[\c
 .BR \-p ]
 [\c
 .BR \-a \ |
@@ -42,6 +44,10 @@ names will be listed, no data will be output.
 .BR \-n
 Dump an LMDB database which does not use subdirectories.
 .TP
+.BR \-v
+Use the previous environment state instead of the latest state.
+This may be useful if the latest state has been corrupted.
+.TP
 .BR \-p
 If characters in either the key or data items are printing characters (as
 defined by isprint(3)), output them directly. This option permits users to

external/db_drivers/liblmdb/mdb_dump.c

@@ -164,7 +164,7 @@ static int dumpit(MDB_txn *txn, MDB_dbi dbi, char *name)
 
 static void usage(char *prog)
 {
-	fprintf(stderr, "usage: %s [-V] [-f output] [-l] [-n] [-p] [-a|-s subdb] dbpath\n", prog);
+	fprintf(stderr, "usage: %s [-V] [-f output] [-l] [-n] [-p] [-v] [-a|-s subdb] dbpath\n", prog);
 	exit(EXIT_FAILURE);
 }
 
@@ -188,6 +188,7 @@ int main(int argc, char *argv[])
 	 * -n: use NOSUBDIR flag on env_open
 	 * -p: use printable characters
 	 * -f: write to file instead of stdout
+	 * -v: use previous snapshot
 	 * -V: print version and exit
 	 * (default) dump only the main DB
 	 */
@@ -215,6 +216,9 @@ int main(int argc, char *argv[])
 		case 'n':
 			envflags |= MDB_NOSUBDIR;
 			break;
+		case 'v':
+			envflags |= MDB_PREVSNAPSHOT;
+			break;
 		case 'p':
 			mode |= PRINT;
 			break;

external/db_drivers/liblmdb/mdb_stat.1

@@ -14,6 +14,8 @@ mdb_stat \- LMDB environment status tool
 [\c
 .BR \-n ]
 [\c
+.BR \-v ]
+[\c
 .BR \-r [ r ]]
 [\c
 .BR \-a \ |
@@ -39,6 +41,10 @@ If \fB\-fff\fP is given, display the full list of page IDs in the freelist.
 .BR \-n
 Display the status of an LMDB database which does not use subdirectories.
 .TP
+.BR \-v
+Use the previous environment state instead of the latest state.
+This may be useful if the latest state has been corrupted.
+.TP
 .BR \-r
 Display information about the environment reader table.
 Shows the process ID, thread ID, and transaction ID for each active

external/db_drivers/liblmdb/mdb_stat.c

@@ -46,7 +46,7 @@ static void prstat(MDB_stat *ms)
 
 static void usage(char *prog)
 {
-	fprintf(stderr, "usage: %s [-V] [-n] [-e] [-r[r]] [-f[f[f]]] [-a|-s subdb] dbpath\n", prog);
+	fprintf(stderr, "usage: %s [-V] [-n] [-e] [-r[r]] [-f[f[f]]] [-v] [-a|-s subdb] dbpath\n", prog);
 	exit(EXIT_FAILURE);
 }
 
@@ -73,6 +73,7 @@ int main(int argc, char *argv[])
 	 * -f: print freelist info
 	 * -r: print reader info
 	 * -n: use NOSUBDIR flag on env_open
+	 * -v: use previous snapshot
 	 * -V: print version and exit
 	 * (default) print stat of only the main DB
 	 */
@@ -96,6 +97,9 @@ int main(int argc, char *argv[])
 		case 'n':
 			envflags |= MDB_NOSUBDIR;
 			break;
+		case 'v':
+			envflags |= MDB_PREVSNAPSHOT;
+			break;
 		case 'r':
 			rdrinfo++;
 			break;