When computing maximum size of filesystem possible with given number of group descriptor blocks, we forget to include s_first_data_block into the number of blocks. Thus for filesystems with non-zero s_first_data_block it can happen that computed maximum filesystem size is actually lower than current filesystem size which confuses the code and eventually leads to a BUG_ON in ext4_alloc_group_tables() hitting on flex_gd->count == 0. The problem can be reproduced like:
truncate -s 100g /tmp/image mkfs.ext4 -b 1024 -E resize=262144 /tmp/image 32768 mount -t ext4 -o loop /tmp/image /mnt resize2fs /dev/loop0 262145 resize2fs /dev/loop0 300000
Fix the problem by properly including s_first_data_block into the computed number of filesystem blocks.
CC: stable@vger.kernel.org Fixes: 1c6bd7173d66 "ext4: convert file system to meta_bg if needed..." Signed-off-by: Jan Kara jack@suse.cz --- fs/ext4/resize.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c index 48421de803b7..3d9b18505c0c 100644 --- a/fs/ext4/resize.c +++ b/fs/ext4/resize.c @@ -1960,7 +1960,8 @@ int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count) le16_to_cpu(es->s_reserved_gdt_blocks); n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb); n_blocks_count = (ext4_fsblk_t)n_group * - EXT4_BLOCKS_PER_GROUP(sb); + EXT4_BLOCKS_PER_GROUP(sb) + + le32_to_cpu(es->s_first_data_block); n_group--; /* set to last group number */ }