If the lockfile directory gets removed but the container runtime is
still running with an open storage instance, then we can trigger a
runtime panic by:
1. Starting CRI-O
2. Removing the lockfile dir, e.g.:
`sudo rm -rf /var/lib/containers/storage/overlay-images`
3. Running `crictl images`
```
panic: error opening "/var/lib/containers/storage/overlay-images/images.lock": no such file or directory
goroutine 93 [running]:
github.com/containers/storage/pkg/lockfile.(*lockfile).lock(0xc0006b83c0, 0x2000000)
/go/src/github.com/cri-o/cri-o/vendor/github.com/containers/storage/pkg/lockfile/lockfile_unix.go:107 +0x388
github.com/containers/storage/pkg/lockfile.(*lockfile).RLock(0xc0006b83c0)
/go/src/github.com/cri-o/cri-o/vendor/github.com/containers/storage/pkg/lockfile/lockfile_unix.go:146 +0x37
github.com/containers/storage.(*imageStore).RLock(0xc00006e420)
/go/src/github.com/cri-o/cri-o/vendor/github.com/containers/storage/images.go:778 +0x33
github.com/containers/storage.(*store).Images(0xc0003f3b80, 0x0, 0x0, 0x0, 0x0, 0x0)
/go/src/github.com/cri-o/cri-o/vendor/github.com/containers/storage/store.go:3089 +0x1dc
```
We now ensure the parent lockfile directory, but only if it does not
exist to not degrade the performance of c/stroage.
Unit tests around that case have been added as well.
Signed-off-by: Sascha Grunert <sgrunert@redhat.com>
drop a call to Seek and use the Pread and Pwrite syscalls instead that
already accept an offset.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Fix a race where the state lock in a Unixy lockfile structure wasn't
being used to protect access to all of the structure's state fields.
Fix usage of sync/atomic in the corresponding unit tests to not mix
atomic adds and reads with non-atomic reads, something which the race
detector complained about.
Signed-off-by: Nalin Dahyabhai <nalin@redhat.com>
do not flush the modified metadata to disk each time the store is
modified.
It is not needed for other processes stat'ing the lock file to see the
updated mtime.
introduced with commit 9b0633295b
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
We're seeing some panics with Podman on Ubuntu 19 in this code,
and debugging without knowing what's going wrong is very
difficult.
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
Daniel J Walsh