cptofs: remount as read-only prior to exiting #534
Conversation
| @@ -675,6 +675,26 @@ int main(int argc, char **argv) | |||
| break; | |||
| } | |||
|
|
|||
| /* FIXME: Remouting as readonly seems to be required to make sure the filessytem is cleanly umounted. Otherwise the journal will be still dirty... */ | |||
There was a problem hiding this comment.
Looks good as a workaround, but I would like to investigate the root cause. Do you have a way to reproduce this?
There was a problem hiding this comment.
$ truncate -s 10G /tmp/foo.img
$ mkfs.ext4 /tmp/foo.img
$ ./tools/lkl/cptofs -p -o 1000 -g 1000 -t ext4 -i /tmp/foo.img ./tools/lkl/tests /
Now if you try to mount the image read-only, you can see hat it fails because of the unclean journal
$ mount -o ro /tmp/foo.img /mnt
The error will show up in dmesg.
There was a problem hiding this comment.
@tavip did you managed to reproduce the issue? This would block us updating lkl in NixOS, where we use it for building disk images without the need for KVM.
There was a problem hiding this comment.
I think I found the root cause but it is turning out more difficult to fix than I expected. Lets merge this for now. Thanks for the ping.
also umount returns and /proc/self/mounts no longer lists the filesystem, ext4 seems to only flush the filesystem properly after a certain delay. To work around this, we now remount the filesystem as read-only to ensure that it is actually finished by the time we exited Signed-off-by: Jörg Thalheim <[email protected]>
|
I can fix the coding style issues in case you cannot find a better solution to fix the issue |
|
Ok. It's probably good to drop some insights on what the actual root cause is. |
|
@tavip If you don't mind, care to give an update on the root cause of this? I suspect #466 might be linked to this workaround and issue in general based on the trace provided in NixOS/nixpkgs#59867, but that was as far as my limited C skills got me. From what I can tell, the lkl umount call follows an identical call path (sans wrapper) as the mainline kernel so I really am rather curious what is causing it to fail a journal flush. |
depends on #532