On Thu, Jan 24, 2019 at 05:26:06PM +0100, Jan Kara wrote:
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
Thanks, applied.
- Ted