Fix cleanup of /run/crun state dir when cgroup failed to mount #1456
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
flouthoc
reviewed
Apr 26, 2024
src/libcrun/status.c
Outdated
| @@ -485,7 +486,20 @@ rmdirfd (const char *namedir, int fd, libcrun_error_t *err) | |||
| ret = unlinkat (dirfd (d), de->d_name, 0); | |||
| if (ret < 0) | |||
| { | |||
| retry_unlink: | |||
There was a problem hiding this comment.
Not sure but indentation looks off here, could you do clang-format .
There was a problem hiding this comment.
Yes the CI fails for clang-format as well, could you please run clang-format and repush it. Thanks.
There was a problem hiding this comment.
Sorry, not used/didn't notice this style for goto labels, fixed.
Note make clang-format modifies quite a few other files, might want to do a "fix formatting" commit at some point
martinetd
force-pushed
the
cleanup_cgroup_mount
branch
2 times, most recently
from
749bf3f to
39ab155
Compare
|
I think we should rebase on top of #1457 once it is merged |
When running a container with /sys/fs/cgroup already mounted
(e.g. --volume=/sys:/sys) and another shared volume somewhere
(e.g. --volume=/mnt:/mnt:shared), the move mount fails with EINVAL:
```
mount("/run/crun/6d75ef05e540486f21ef54743f8af97f4dff0d0d7c54d9b7852cd5302e884db8/tmpmount", "/proc/self/fd/8", NULL, MS_MOVE, NULL) = -1 EINVAL (Invalid argument)
```
This leaves tmpmount as a mountpoint behind, which then fails to
cleanup:
```
openat(AT_FDCWD, "/run/crun", O_RDONLY|O_CLOEXEC|O_DIRECTORY) = 3
openat(3, "0d1342f7a7375593813287c743170bb71f30f9a64fa8bc141d3cf9b8e9aa5a89", O_RDONLY|O_DIRECTORY) = 4
unlinkat(4, "tmpmount", 0) = -1 EISDIR (Is a directory)
unlinkat(4, "tmpmount", AT_REMOVEDIR) = -1 EBUSY (Device or resource busy)
close(4) = 0
unlinkat(3, "0d1342f7a7375593813287c743170bb71f30f9a64fa8bc141d3cf9b8e9aa5a89", AT_REMOVEDIR) = -1 ENOTEMPTY (Directory not empty)
```
This in turn ultimately fails libcrun_container_delete_status() and
friends, leading to the non-execution of post hooks in this reproducer:
```
$ mkdir hooks.d
$ cat > hooks.d/hook.json <<'EOF'
{
"version": "1.0.0",
"when": {"always": true},
"hook": {
"path": "/bin/sh",
"args": ["/bin/sh", "-c", "date >> /tmp/hookme"]
},
"stages": ["poststop"]
}
EOF
$ podman --runtime=path/crun run \
--net=none --name test -d --replace --hooks-dir=$PWD/hooks.d \
--volume=/sys:/sys --volume=/mnt:/mnt:shared \
docker.io/alpine true
```
(the hook is eventually run on podman rm, but not by crun immediately on
container stop)
Just cleaning up the mount point makes the problem go away.
Fixes: 523eed3 ("linux: add new fallback when mount fails with EBUSY")
Signed-off-by: Dominique Martinet <[email protected]>
The previous patch left a mount point in /run/crun which could not be
removed, we should try harder to remove directories failing with EBUSY
as we do in create_missing_devs().
This patch also fixes the issue described in the previous commit
("status: rmdirfd: try harder to remove mount points").
Fixes: 523eed3 ("linux: add new fallback when mount fails with EBUSY")
Signed-off-by: Dominique Martinet <[email protected]>
martinetd
force-pushed
the
cleanup_cgroup_mount
branch
from
39ab155 to
4618d50
Compare
Done, tests now pass. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This fixes a regression introduced in crun 1.9.1 with 523eed3
When running a container with /sys/fs/cgroup already mounted (e.g. --volume=/sys:/sys) and another shared volume somewhere (e.g. --volume=/mnt:/mnt:shared), the move mount fails with EINVAL:
This leaves tmpmount as a mountpoint behind, which then fails to cleanup:
This in turn ultimately fails libcrun_container_delete_status() and friends, leading to the non-execution of post hooks in this reproducer:
(the hook is eventually run on podman rm, but not by crun immediately on container stop)
This PR has two commits, either of which fix this issue:
happy to drop either commit, but I think it's worth keeping the safety as such mount points are easy to forget.