Hack for potential ext3/ext4 corruption issue

Use regular fsync() if we think this commit grew the DB file.

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index cc6eaa20bb9791354b4f9c2430770b08fea81562..d9ab5bd3830919a8e7ada7c2f969c6e4719e81dd 100644 (file)
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -368,6 +368,7 @@ static int mdb_mutex_failed(MDB_env *env, mdb_mutex_t *mutex, int rc);
  */
 #ifndef MDB_FDATASYNC
 # define MDB_FDATASYNC fdatasync
+# define HAVE_FDATASYNC        1
 #endif
 
 #ifndef MDB_MSYNC
@@ -1154,7 +1155,7 @@ struct MDB_env {
        MDB_txn         *me_txn;                /**< current write transaction */
        MDB_txn         *me_txn0;               /**< prealloc'd write transaction */
        size_t          me_mapsize;             /**< size of the data memory map */
-       off_t           me_size;                /**< current file size */
+       size_t          me_size;                /**< current file size */
        pgno_t          me_maxpg;               /**< me_mapsize / me_psize */
        MDB_dbx         *me_dbxs;               /**< array of static DB info */
        uint16_t        *me_dbflags;    /**< array of flags from MDB_db.md_flags */
@@ -2341,10 +2342,19 @@ fail:
        return rc;
 }
 
-int
-mdb_env_sync(MDB_env *env, int force)
+/* 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
+mdb_env_sync0(MDB_env *env, int flag)
 {
-       int rc = 0;
+       int rc = 0, force = flag & FORCE;
        if (force || !F_ISSET(env->me_flags, MDB_NOSYNC)) {
                if (env->me_flags & MDB_WRITEMAP) {
                        int flags = ((env->me_flags & MDB_MAPASYNC) && !force)
@@ -2356,6 +2366,12 @@ mdb_env_sync(MDB_env *env, int force)
                                rc = ErrCode();
 #endif
                } else {
+#ifdef HAVE_FDATASYNC
+                       if (flag & FGREW) {
+                               if (fsync(env->me_fd))  /* Avoid ext-fs bugs, do full sync */
+                                       rc = ErrCode();
+                       } else
+#endif
                        if (MDB_FDATASYNC(env->me_fd))
                                rc = ErrCode();
                }
@@ -2363,6 +2379,12 @@ mdb_env_sync(MDB_env *env, int force)
        return rc;
 }
 
+int
+mdb_env_sync(MDB_env *env, int force)
+{
+       return mdb_env_sync0(env, force != 0);
+}
+
 /** Back up parent txn's cursors, then grab the originals for tracking */
 static int
 mdb_cursor_shadow(MDB_txn *src, MDB_txn *dst)
@@ -3437,8 +3459,15 @@ mdb_txn_commit(MDB_txn *txn)
        mdb_audit(txn);
 #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)) ||
-               (rc = mdb_env_sync(env, 0)) ||
+               (rc = mdb_env_sync(env, i)) ||
                (rc = mdb_env_write_meta(txn)))
                goto fail;
 
@@ -3979,6 +4008,10 @@ mdb_env_open2(MDB_env *env)
                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);
        if (rc)
                return rc;