Online resizing of filesystem mounted with backup superblock causes corruption

# mkfs.ext4 /dev/loop0 100M # adjust if necessary, should be less than
partition size

Superblock backups stored on blocks:

        8193, 24577, 40961, 57345, 73729

(pick one)

# mount -t ext4 -o sb=40961 /dev/loop0 /media/
# resize2fs /dev/loop0
resize2fs 1.42.12 (29-Aug-2014)
Filesystem at /dev/loop0 is mounted on /media; on-line resizing required
old_desc_blocks = 1, new_desc_blocks = 4

Now, depending on your luck, various things may happen. resize2fs may hang. You might see BUG() in your dmesg. Either way, system will become severely corrupted (as indicated by fsck). You won't be able to mount it, either.

[ 2669.510800] EXT4-fs (loop0): resizing filesystem from 128000 to
256000 blocks
[ 2669.513558] EXT4-fs warning (device loop0): reserve_backup_gdb:969:
reserved block 32770 not at offset 1
[ 2669.513569] EXT4-fs (loop0): resized filesystem to 256000
[ 2669.516328] EXT4-fs warning (device loop0): ext4_group_extend:1734:

can't shrink FS - resize aborted

Maxim Malkov
Software Engineering Department, ISPRAS 

Thanks for the bug report; I've can reproduce this using a 3.18 based kernel.

It may be that simplest bug fix is to simply disallow resizing if we are using a backup superblock.

- Ted

Prevent BUG or corrupted file systems after the following:

mkfs.ext4 /dev/vdc 100M
mount -t ext4 -o sb=40961 /dev/vdc /vdc
resize2fs /dev/vdc

We previously prevented online resizing using the old resize ioctl. Move the code to ext4_resize_begin(), so the check applies for all of the resize ioctl's.

---
 fs/ext4/resize.c | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

index bf76f40..8a8ec62 100644
@@ -24,6 +24,18 @@ int ext4_resize_begin(struct super_block *sb)
 		return -EPERM;
 
 	/*
+	 * If we are not using the primary superblock/GDT copy don't resize,
+         * because the user tools have no way of handling this.  Probably a
+         * bad time to do it anyways.
+         */
+	if (EXT4_SB(sb)->s_sbh->b_blocknr !=
+	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
+		ext4_warning(sb, "won't resize using backup superblock at %llu",
+			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
+		return -EPERM;
+	}
+
+	/*
 	 * We are not allowed to do online-resizing on a filesystem mounted
 	 * with error, because it can destroy the filesystem easily.
 	 */

@@ -758,18 +770,6 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
 		       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 		       gdb_num);
 
-	/*
-	 * If we are not using the primary superblock/GDT copy don't resize,
-         * because the user tools have no way of handling this.  Probably a
-         * bad time to do it anyways.
-         */
-	if (EXT4_SB(sb)->s_sbh->b_blocknr !=
-	    le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
-		ext4_warning(sb, "won't resize using backup superblock at %llu",
-			(unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
-		return -EPERM;
-	}