On 2024/9/26 19:32, Aleksandr Mikhalitsyn wrote:
Question to you and Jan. Do you guys think that it makes sense to try to create a minimal reproducer for this problem without Incus/LXD involved? (only e2fsprogs, lvm tools, etc)
I guess this test can be put in the xfstests test suite, right?
Kind regards, Alex
I think it makes sense, and it's good to have more use cases to look around some corners. If you have an idea, let it go.
Minimal reproducer:
mkdir -p /tmp/ext4_crash/mnt EXT4_CRASH_IMG="/tmp/ext4_crash/disk.img" rm -f $EXT4_CRASH_IMG truncate $EXT4_CRASH_IMG --size 25MiB EXT4_CRASH_DEV=$(losetup --find --nooverlap --direct-io=on --show $EXT4_CRASH_IMG) mkfs.ext4 -E nodiscard,lazy_itable_init=0,lazy_journal_init=0 $EXT4_CRASH_DEV mount $EXT4_CRASH_DEV /tmp/ext4_crash/mnt truncate $EXT4_CRASH_IMG --size 3GiB losetup -c $EXT4_CRASH_DEV resize2fs $EXT4_CRASH_DEV
Hi Alex,
This replicator didn't replicate the issue in my VM, so I took a deeper look. The reproduction of the problem requires the following:
o_group = flexbg_size * 2 * n; o_size = (o_group + 1) * group_size; n_group: [o_group + flexbg_size, o_group + flexbg_size * 2)
Take n=1,flexbg_size=16 as an example: last:47 |----------------|----------------|o---------------|--------------n-| old:32 >>> new:62
Thus the replicator can be simplified as:
img=test.img truncate -s 600M $img mkfs.ext4 -F $img -b 1024 -G 16 264M dev=`losetup -f --show $img` mkdir -p /tmp/test mount $dev /tmp/test resize2fs $dev 504M