Comment 25 for bug 1387214

So this is not a normal fs corruption issue. I was able to run the aa-corruption test quite a few times with the extra e2fsck logic we added, and was easily able to get the corrupted files on krillin, but which is even more interesting is that the files are not actually corrupted for e2fsck (not pointing out a single fs error, even though the files are incorrect).

There's clearly a pattern with click paths under /opt/click.ubuntu.com inside the corrupted files. /opt/click.ubuntu.com is also bind-mounted, and it seems to be where it's mostly heavily used by the system.

So in the end it looks like with a quite bizarre error under rw filesystems (ext4) that are bind-mounted. After migrating /opt/click.ubuntu.com, /var/lib/apparmor and /var/cache/apparmor under '/' (not bind-mounted), I was able to run the same aa-corruption test case for a few without any issues (by default I can always get the issue after the first test iteration).

Will let it running for a few more iterations and report back, but one bad thing is that running e2fsck is really not going to help much here.