Merge pull request #5305 from mheon/check_for_common_deadlocks

Add basic deadlock detection for container start/remove

cmd/podman/rm.go

@@ -4,8 +4,10 @@ import (
 	"fmt"
 
 	"github.com/containers/libpod/cmd/podman/cliconfig"
+	"github.com/containers/libpod/libpod/define"
 	"github.com/containers/libpod/pkg/adapter"
 	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
 	"github.com/spf13/cobra"
 )
 
@@ -77,6 +79,9 @@ func rmCmd(c *cliconfig.RmValues) error {
 
 	if len(failures) > 0 {
 		for _, err := range failures {
+			if errors.Cause(err) == define.ErrWillDeadlock {
+				logrus.Errorf("Potential deadlock detected - please run 'podman system renumber' to resolve")
+			}
 			exitCode = setExitCode(err)
 		}
 	}

libpod/container_internal.go

@@ -1401,6 +1401,9 @@ func (c *Container) mountNamedVolume(v *ContainerNamedVolume, mountpoint string)
 		return nil, errors.Wrapf(err, "error retrieving named volume %s for container %s", v.Name, c.ID())
 	}
 
+	if vol.config.LockID == c.config.LockID {
+		return nil, errors.Wrapf(define.ErrWillDeadlock, "container %s and volume %s share lock ID %d", c.ID(), vol.Name(), c.config.LockID)
+	}
 	vol.lock.Lock()
 	defer vol.lock.Unlock()
 	if vol.needsMount() {

libpod/define/errors.go

@@ -61,6 +61,11 @@ var (
 	// the user.
 	ErrDetach = utils.ErrDetach
 
+	// ErrWillDeadlock indicates that the requested operation will cause a
+	// deadlock. This is usually caused by upgrade issues, and is resolved
+	// by renumbering the locks.
+	ErrWillDeadlock = errors.New("deadlock due to lock mismatch")
+
 	// ErrNoCgroups indicates that the container does not have its own
 	// CGroup.
 	ErrNoCgroups = errors.New("this container does not have a cgroup")

libpod/runtime_ctr.go

@@ -412,6 +412,9 @@ func (r *Runtime) removeContainer(ctx context.Context, c *Container, force bool,
 		}
 
 		// Lock the pod while we're removing container
+		if pod.config.LockID == c.config.LockID {
+			return errors.Wrapf(define.ErrWillDeadlock, "container %s and pod %s share lock ID %d", c.ID(), pod.ID(), c.config.LockID)
+		}
 		pod.lock.Lock()
 		defer pod.lock.Unlock()
 		if err := pod.updatePod(); err != nil {

libpod/runtime_volume.go

@@ -36,9 +36,6 @@ func (r *Runtime) RemoveVolume(ctx context.Context, v *Volume, force bool) error
 		}
 	}
 
-	v.lock.Lock()
-	defer v.lock.Unlock()
-
 	return r.removeVolume(ctx, v, force)
 }
 

libpod/runtime_volume_linux.go

@@ -124,6 +124,9 @@ func (r *Runtime) removeVolume(ctx context.Context, v *Volume, force bool) error
 		return define.ErrVolumeRemoved
 	}
 
+	v.lock.Lock()
+	defer v.lock.Unlock()
+
 	// Update volume status to pick up a potential removal from state
 	if err := v.update(); err != nil {
 		return err

pkg/adapter/containers.go

@@ -469,6 +469,10 @@ func (r *LocalRuntime) Run(ctx context.Context, c *cliconfig.RunValues, exitCode
 				logrus.Debugf("unable to remove container %s after failing to start and attach to it", ctr.ID())
 			}
 		}
+		if errors.Cause(err) == define.ErrWillDeadlock {
+			logrus.Debugf("Deadlock error: %v", err)
+			return define.ExitCode(err), errors.Errorf("attempting to start container %s would cause a deadlock; please run 'podman system renumber' to resolve", ctr.ID())
+		}
 		return define.ExitCode(err), err
 	}
 
@@ -702,6 +706,11 @@ func (r *LocalRuntime) Start(ctx context.Context, c *cliconfig.StartValues, sigP
 				return exitCode, nil
 			}
 
+			if errors.Cause(err) == define.ErrWillDeadlock {
+				logrus.Debugf("Deadlock error: %v", err)
+				return define.ExitCode(err), errors.Errorf("attempting to start container %s would cause a deadlock; please run 'podman system renumber' to resolve", ctr.ID())
+			}
+
 			if ctrRunning {
 				return 0, err
 			}
@@ -735,6 +744,10 @@ func (r *LocalRuntime) Start(ctx context.Context, c *cliconfig.StartValues, sigP
 				if lastError != nil {
 					fmt.Fprintln(os.Stderr, lastError)
 				}
+				if errors.Cause(err) == define.ErrWillDeadlock {
+					lastError = errors.Wrapf(err, "please run 'podman system renumber' to resolve deadlocks")
+					continue
+				}
 				lastError = errors.Wrapf(err, "unable to start container %q", container)
 				continue
 			}