ITS#8704 Fix PREVMETA, rename to PREVSNAPSHOT

and enforce exclusive access to environment. Also fix txn_begin/pick_meta
to use correct meta page, and reset the flag after successful commit.

libraries/liblmdb/lmdb.h

@@ -354,8 +354,8 @@ typedef void (MDB_enc_func)(const MDB_val *src, MDB_val *dst, const MDB_val *key
 #define MDB_NORDAHEAD	0x800000
 	/** don't initialize malloc'd memory before writing to datafile */
 #define MDB_NOMEMINIT	0x1000000
-	/** use the previous meta page rather than the latest one */
-#define MDB_PREVMETA	0x2000000
+	/** use the previous snapshot rather than the latest one */
+#define MDB_PREVSNAPSHOT	0x2000000
 	/** don't use a single mmap, remap individual chunks (needs MDB_RPAGE_CACHE) */
 #define MDB_REMAP_CHUNKS	0x4000000
 /** @} */
@@ -672,10 +672,12 @@ 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_PREVMETA
-	 *		Open the environment with the previous meta page rather than the latest
+	 *	<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.
+	 *		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.

libraries/liblmdb/mdb.c

@@ -4047,6 +4047,8 @@ done:
 	return MDB_SUCCESS;
 }
 
+static int ESECT mdb_env_share_locks(MDB_env *env, int *excl);
+
 int
 mdb_txn_commit(MDB_txn *txn)
 {
@@ -4274,6 +4276,15 @@ 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);
@@ -4564,7 +4575,8 @@ 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
@@ -5139,6 +5151,9 @@ mdb_env_open2(MDB_env *env, int prev)
 #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);
@@ -5238,9 +5253,6 @@ 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
 	{
@@ -5684,7 +5696,7 @@ mdb_env_envflags(MDB_env *env)
 		/*f*/ MDB_FIXEDMAP, /*h*/ MDB_NORDAHEAD,    /*i*/ MDB_NOMEMINIT,
 		/*l*/ MDB_NOLOCK,   /*m*/ MDB_NOMETASYNC,   /*n*/ MDB_NOSUBDIR,
 		/*r*/ MDB_RDONLY,   /*s*/ MDB_NOSYNC,       /*t*/ MDB_NOTLS,
-		/*v*/ MDB_PREVMETA, /*w*/ MDB_WRITEMAP,
+		/*v*/ MDB_PREVSNAPSHOT, /*w*/ MDB_WRITEMAP,
 	};
 	unsigned flags = 0;
 	const char *s, *opts = getenv("LMDB_FLAGS");
@@ -5724,7 +5736,7 @@ mdb_env_envflags(MDB_env *env)
 	 */
 #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_PREVMETA|MDB_REMAP_CHUNKS)
+	MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK|MDB_NORDAHEAD|MDB_PREVSNAPSHOT|MDB_REMAP_CHUNKS)
 #define EXPOSED		(CHANGEABLE|CHANGELESS | MDB_ENCRYPT)
 
 #if VALID_FLAGS & PERSISTENT_FLAGS & EXPOSED
@@ -5823,6 +5835,10 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 		rc = mdb_env_setup_locks(env, &fname, mode, &excl);
 		if (rc)
 			goto leave;
+		if ((flags & MDB_PREVSNAPSHOT) && !excl) {
+			rc = EAGAIN;
+			goto leave;
+		}
 	}
 
 	rc = mdb_fopen(env, &fname,
@@ -5837,7 +5853,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 			goto leave;
 	}
 
-	if ((rc = mdb_env_open2(env, flags & MDB_PREVMETA)) == MDB_SUCCESS) {
+	if ((rc = mdb_env_open2(env, flags & MDB_PREVSNAPSHOT)) == MDB_SUCCESS) {
 		if (!(flags & (MDB_RDONLY|MDB_WRITEMAP))) {
 			/* Synchronous fd for meta writes. Needed even with
 			 * MDB_NOSYNC/MDB_NOMETASYNC, in case these get reset.
@@ -5847,7 +5863,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 				goto leave;
 		}
 		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;

libraries/liblmdb/mdb_copy.c

@@ -39,7 +39,7 @@ int main(int argc,char * 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_PREVMETA;
+			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') {

libraries/liblmdb/mdb_dump.c

@@ -175,7 +175,7 @@ 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 metapage
+	 * -v: use previous snapshot
 	 * -V: print version and exit
 	 * (default) dump only the main DB
 	 */
@@ -204,7 +204,7 @@ int main(int argc, char *argv[])
 			envflags |= MDB_NOSUBDIR;
 			break;
 		case 'v':
-			envflags |= MDB_PREVMETA;
+			envflags |= MDB_PREVSNAPSHOT;
 			break;
 		case 'p':
 			mode |= PRINT;

libraries/liblmdb/mdb_stat.c

@@ -61,7 +61,7 @@ int main(int argc, char *argv[])
 	 * -f: print freelist info
 	 * -r: print reader info
 	 * -n: use NOSUBDIR flag on env_open
-	 * -v: use previous metapage
+	 * -v: use previous snapshot
 	 * -V: print version and exit
 	 * (default) print stat of only the main DB
 	 */
@@ -86,7 +86,7 @@ int main(int argc, char *argv[])
 			envflags |= MDB_NOSUBDIR;
 			break;
 		case 'v':
-			envflags |= MDB_PREVMETA;
+			envflags |= MDB_PREVSNAPSHOT;
 			break;
 		case 'r':
 			rdrinfo++;