Remove infra ID from DB before removing containers

If we interrupt pod removal between removing containers and
removing the whole pod, the infra ID was still in the DB, and
most pod operations would try to retrieve the infra container
(and would this fail). Clear the infra ID from the DB just before
we remove all containers to prevent this.

Fixes #12034

[NO NEW TESTS NEEDED] This is a very narrow race and I have no
idea how to repro it.

Signed-off-by: Matthew Heon <mheon@redhat.com>

1 files changed, 9 insertions, 0 deletions

diff --git a/libpod/runtime_pod_linux.go b/libpod/runtime_pod_linux.go
index 7571fdfff..a90ea6125 100644
--- a/libpod/runtime_pod_linux.go
+++ b/libpod/runtime_pod_linux.go
@@ -264,6 +264,15 @@ func (r *Runtime) removePod(ctx context.Context, p *Pod, removeCtrs, force bool)
 		}
 	}
 
+	// Clear infra container ID before we remove the infra container.
+	// There is a potential issue if we don't do that, and removal is
+	// interrupted between RemoveAllContainers() below and the pod's removal
+	// later - we end up with a reference to a nonexistent infra container.
+	p.state.InfraContainerID = ""
+	if err := p.save(); err != nil {
+		return err
+	}
+
 	// Remove all containers in the pod from the state.
 	if err := r.state.RemovePodContainers(p); err != nil {
 		// If this fails, there isn't much more we can do.