diff options
Diffstat (limited to 'libpod')
-rw-r--r-- | libpod/container.go | 10 | ||||
-rw-r--r-- | libpod/container_api.go | 4 | ||||
-rw-r--r-- | libpod/container_exec.go | 67 | ||||
-rw-r--r-- | libpod/container_internal.go | 65 | ||||
-rw-r--r-- | libpod/container_internal_linux.go | 101 | ||||
-rw-r--r-- | libpod/container_internal_unsupported.go | 5 | ||||
-rw-r--r-- | libpod/container_log.go | 54 | ||||
-rw-r--r-- | libpod/events.go | 39 | ||||
-rw-r--r-- | libpod/events/config.go | 2 | ||||
-rw-r--r-- | libpod/events/events.go | 2 | ||||
-rw-r--r-- | libpod/networking_linux.go | 4 | ||||
-rw-r--r-- | libpod/oci_conmon_linux.go | 6 | ||||
-rw-r--r-- | libpod/options.go | 13 | ||||
-rw-r--r-- | libpod/volume_internal.go | 19 |
14 files changed, 265 insertions, 126 deletions
diff --git a/libpod/container.go b/libpod/container.go index c6f0cd618..4b9bea5fc 100644 --- a/libpod/container.go +++ b/libpod/container.go @@ -957,6 +957,12 @@ func (c *Container) cGroupPath() (string, error) { // is the libpod-specific one we're looking for. // // See #8397 on the need for the longest-path look up. + // + // And another workaround for containers running systemd as the payload. + // containers running systemd moves themselves into a child subgroup of + // the named systemd cgroup hierarchy. Ignore any named cgroups during + // the lookup. + // See #10602 for more details. procPath := fmt.Sprintf("/proc/%d/cgroup", c.state.PID) lines, err := ioutil.ReadFile(procPath) if err != nil { @@ -972,6 +978,10 @@ func (c *Container) cGroupPath() (string, error) { logrus.Debugf("Error parsing cgroup: expected 3 fields but got %d: %s", len(fields), procPath) continue } + // Ignore named cgroups like name=systemd. + if bytes.Contains(fields[1], []byte("=")) { + continue + } path := string(fields[2]) if len(path) > len(cgroupPath) { cgroupPath = path diff --git a/libpod/container_api.go b/libpod/container_api.go index 4ccb240e7..b75d0b41d 100644 --- a/libpod/container_api.go +++ b/libpod/container_api.go @@ -12,6 +12,7 @@ import ( "github.com/containers/podman/v3/libpod/define" "github.com/containers/podman/v3/libpod/events" "github.com/containers/podman/v3/pkg/signal" + "github.com/containers/storage/pkg/archive" "github.com/pkg/errors" "github.com/sirupsen/logrus" ) @@ -776,6 +777,9 @@ type ContainerCheckpointOptions struct { // ImportPrevious tells the API to restore container with two // images. One is TargetFile, the other is ImportPrevious. ImportPrevious string + // Compression tells the API which compression to use for + // the exported checkpoint archive. + Compression archive.Compression } // Checkpoint checkpoints a container diff --git a/libpod/container_exec.go b/libpod/container_exec.go index c359f1e5d..737bf74ad 100644 --- a/libpod/container_exec.go +++ b/libpod/container_exec.go @@ -1,6 +1,7 @@ package libpod import ( + "context" "io/ioutil" "net/http" "os" @@ -539,18 +540,7 @@ func (c *Container) ExecStop(sessionID string, timeout *uint) error { var cleanupErr error // Retrieve exit code and update status - exitCode, err := c.readExecExitCode(session.ID()) - if err != nil { - cleanupErr = err - } - session.ExitCode = exitCode - session.PID = 0 - session.State = define.ExecStateStopped - - if err := c.save(); err != nil { - if cleanupErr != nil { - logrus.Errorf("Error stopping container %s exec session %s: %v", c.ID(), session.ID(), cleanupErr) - } + if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil { cleanupErr = err } @@ -592,15 +582,7 @@ func (c *Container) ExecCleanup(sessionID string) error { return errors.Wrapf(define.ErrExecSessionStateInvalid, "cannot clean up container %s exec session %s as it is running", c.ID(), session.ID()) } - exitCode, err := c.readExecExitCode(session.ID()) - if err != nil { - return err - } - session.ExitCode = exitCode - session.PID = 0 - session.State = define.ExecStateStopped - - if err := c.save(); err != nil { + if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil { return err } } @@ -637,9 +619,9 @@ func (c *Container) ExecRemove(sessionID string, force bool) error { return err } if !running { - session.State = define.ExecStateStopped - // TODO: should we retrieve exit code here? - // TODO: Might be worth saving state here. + if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil { + return err + } } } @@ -653,6 +635,10 @@ func (c *Container) ExecRemove(sessionID string, force bool) error { return err } + if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil { + return err + } + if err := c.cleanupExecBundle(session.ID()); err != nil { return err } @@ -757,10 +743,25 @@ func (c *Container) Exec(config *ExecConfig, streams *define.AttachStreams, resi session, err := c.ExecSession(sessionID) if err != nil { + if errors.Cause(err) == define.ErrNoSuchExecSession { + // TODO: If a proper Context is ever plumbed in here, we + // should use it. + // As things stand, though, it's not worth it - this + // should always terminate quickly since it's not + // streaming. + diedEvent, err := c.runtime.GetExecDiedEvent(context.Background(), c.ID(), sessionID) + if err != nil { + return -1, errors.Wrapf(err, "error retrieving exec session %s exit code", sessionID) + } + return diedEvent.ContainerExitCode, nil + } return -1, err } exitCode := session.ExitCode if err := c.ExecRemove(sessionID, false); err != nil { + if errors.Cause(err) == define.ErrNoSuchExecSession { + return exitCode, nil + } return -1, err } @@ -927,6 +928,8 @@ func (c *Container) getActiveExecSessions() ([]string, error) { session.PID = 0 session.State = define.ExecStateStopped + c.newExecDiedEvent(session.ID(), exitCode) + needSave = true } if err := c.cleanupExecBundle(id); err != nil { @@ -1036,6 +1039,22 @@ func writeExecExitCode(c *Container, sessionID string, exitCode int) error { return errors.Wrapf(err, "error syncing container %s state to remove exec session %s", c.ID(), sessionID) } + return justWriteExecExitCode(c, sessionID, exitCode) +} + +func retrieveAndWriteExecExitCode(c *Container, sessionID string) error { + exitCode, err := c.readExecExitCode(sessionID) + if err != nil { + return err + } + + return justWriteExecExitCode(c, sessionID, exitCode) +} + +func justWriteExecExitCode(c *Container, sessionID string, exitCode int) error { + // Write an event first + c.newExecDiedEvent(sessionID, exitCode) + session, ok := c.state.ExecSessions[sessionID] if !ok { // Exec session already removed. diff --git a/libpod/container_internal.go b/libpod/container_internal.go index f77825efd..545b78976 100644 --- a/libpod/container_internal.go +++ b/libpod/container_internal.go @@ -42,6 +42,7 @@ const ( // name of the directory holding the artifacts artifactsDir = "artifacts" execDirPermission = 0755 + preCheckpointDir = "pre-checkpoint" ) // rootFsSize gets the size of the container's root filesystem @@ -141,7 +142,7 @@ func (c *Container) CheckpointPath() string { // PreCheckpointPath returns the path to the directory containing the pre-checkpoint-images func (c *Container) PreCheckPointPath() string { - return filepath.Join(c.bundlePath(), "pre-checkpoint") + return filepath.Join(c.bundlePath(), preCheckpointDir) } // AttachSocketPath retrieves the path of the container's attach socket @@ -427,7 +428,7 @@ func (c *Container) setupStorage(ctx context.Context) error { }, LabelOpts: c.config.LabelOpts, } - if c.restoreFromCheckpoint { + if c.restoreFromCheckpoint && !c.config.Privileged { // If restoring from a checkpoint, the root file-system // needs to be mounted with the same SELinux labels as // it was mounted previously. @@ -1061,7 +1062,7 @@ func (c *Container) init(ctx context.Context, retainRetries bool) error { } for _, v := range c.config.NamedVolumes { - if err := c.chownVolume(v.Name); err != nil { + if err := c.fixVolumePermissions(v); err != nil { return err } } @@ -1680,64 +1681,6 @@ func (c *Container) mountNamedVolume(v *ContainerNamedVolume, mountpoint string) return vol, nil } -// Chown the specified volume if necessary. -func (c *Container) chownVolume(volumeName string) error { - vol, err := c.runtime.state.Volume(volumeName) - if err != nil { - return errors.Wrapf(err, "error retrieving named volume %s for container %s", volumeName, c.ID()) - } - - vol.lock.Lock() - defer vol.lock.Unlock() - - // The volume may need a copy-up. Check the state. - if err := vol.update(); err != nil { - return err - } - - // TODO: For now, I've disabled chowning volumes owned by non-Podman - // drivers. This may be safe, but it's really going to be a case-by-case - // thing, I think - safest to leave disabled now and re-enable later if - // there is a demand. - if vol.state.NeedsChown && !vol.UsesVolumeDriver() { - vol.state.NeedsChown = false - - uid := int(c.config.Spec.Process.User.UID) - gid := int(c.config.Spec.Process.User.GID) - - if c.config.IDMappings.UIDMap != nil { - p := idtools.IDPair{ - UID: uid, - GID: gid, - } - mappings := idtools.NewIDMappingsFromMaps(c.config.IDMappings.UIDMap, c.config.IDMappings.GIDMap) - newPair, err := mappings.ToHost(p) - if err != nil { - return errors.Wrapf(err, "error mapping user %d:%d", uid, gid) - } - uid = newPair.UID - gid = newPair.GID - } - - vol.state.UIDChowned = uid - vol.state.GIDChowned = gid - - if err := vol.save(); err != nil { - return err - } - - mountPoint, err := vol.MountPoint() - if err != nil { - return err - } - - if err := os.Lchown(mountPoint, uid, gid); err != nil { - return err - } - } - return nil -} - // cleanupStorage unmounts and cleans up the container's root filesystem func (c *Container) cleanupStorage() error { if !c.state.Mounted { diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go index 74a3fec32..ea52d7ba0 100644 --- a/libpod/container_internal_linux.go +++ b/libpod/container_internal_linux.go @@ -909,14 +909,15 @@ func (c *Container) exportCheckpoint(options ContainerCheckpointOptions) error { includeFiles := []string{ "artifacts", "ctr.log", - metadata.CheckpointDirectory, metadata.ConfigDumpFile, metadata.SpecDumpFile, metadata.NetworkStatusFile, } if options.PreCheckPoint { - includeFiles[0] = "pre-checkpoint" + includeFiles = append(includeFiles, preCheckpointDir) + } else { + includeFiles = append(includeFiles, metadata.CheckpointDirectory) } // Get root file-system changes included in the checkpoint archive var addToTarFiles []string @@ -985,7 +986,7 @@ func (c *Container) exportCheckpoint(options ContainerCheckpointOptions) error { } input, err := archive.TarWithOptions(c.bundlePath(), &archive.TarOptions{ - Compression: archive.Gzip, + Compression: options.Compression, IncludeSourceDir: true, IncludeFiles: includeFiles, }) @@ -1650,22 +1651,20 @@ func (c *Container) generateResolvConf() (string, error) { } } - // Determine the endpoint for resolv.conf in case it is a symlink - resolvPath, err := filepath.EvalSymlinks(resolvConf) + contents, err := ioutil.ReadFile(resolvConf) // resolv.conf doesn't have to exists if err != nil && !os.IsNotExist(err) { return "", err } - // Determine if symlink points to any of the systemd-resolved files - if strings.HasPrefix(resolvPath, "/run/systemd/resolve/") { - resolvPath = "/run/systemd/resolve/resolv.conf" - } - - contents, err := ioutil.ReadFile(resolvPath) - // resolv.conf doesn't have to exists - if err != nil && !os.IsNotExist(err) { - return "", err + ns := resolvconf.GetNameservers(contents) + // check if systemd-resolved is used, assume it is used when 127.0.0.53 is the only nameserver + if len(ns) == 1 && ns[0] == "127.0.0.53" { + // read the actual resolv.conf file for systemd-resolved + contents, err = ioutil.ReadFile("/run/systemd/resolve/resolv.conf") + if err != nil { + return "", errors.Wrapf(err, "detected that systemd-resolved is in use, but could not locate real resolv.conf") + } } ipv6 := false @@ -2427,3 +2426,77 @@ func (c *Container) createSecretMountDir() error { return err } + +// Fix ownership and permissions of the specified volume if necessary. +func (c *Container) fixVolumePermissions(v *ContainerNamedVolume) error { + vol, err := c.runtime.state.Volume(v.Name) + if err != nil { + return errors.Wrapf(err, "error retrieving named volume %s for container %s", v.Name, c.ID()) + } + + vol.lock.Lock() + defer vol.lock.Unlock() + + // The volume may need a copy-up. Check the state. + if err := vol.update(); err != nil { + return err + } + + // TODO: For now, I've disabled chowning volumes owned by non-Podman + // drivers. This may be safe, but it's really going to be a case-by-case + // thing, I think - safest to leave disabled now and re-enable later if + // there is a demand. + if vol.state.NeedsChown && !vol.UsesVolumeDriver() { + vol.state.NeedsChown = false + + uid := int(c.config.Spec.Process.User.UID) + gid := int(c.config.Spec.Process.User.GID) + + if c.config.IDMappings.UIDMap != nil { + p := idtools.IDPair{ + UID: uid, + GID: gid, + } + mappings := idtools.NewIDMappingsFromMaps(c.config.IDMappings.UIDMap, c.config.IDMappings.GIDMap) + newPair, err := mappings.ToHost(p) + if err != nil { + return errors.Wrapf(err, "error mapping user %d:%d", uid, gid) + } + uid = newPair.UID + gid = newPair.GID + } + + vol.state.UIDChowned = uid + vol.state.GIDChowned = gid + + if err := vol.save(); err != nil { + return err + } + + mountPoint, err := vol.MountPoint() + if err != nil { + return err + } + + if err := os.Lchown(mountPoint, uid, gid); err != nil { + return err + } + + // Make sure the new volume matches the permissions of the target directory. + // https://github.com/containers/podman/issues/10188 + st, err := os.Lstat(filepath.Join(c.state.Mountpoint, v.Dest)) + if err == nil { + if err := os.Chmod(mountPoint, st.Mode()|0111); err != nil { + return err + } + stat := st.Sys().(*syscall.Stat_t) + atime := time.Unix(int64(stat.Atim.Sec), int64(stat.Atim.Nsec)) + if err := os.Chtimes(mountPoint, atime, st.ModTime()); err != nil { + return err + } + } else if !os.IsNotExist(err) { + return err + } + } + return nil +} diff --git a/libpod/container_internal_unsupported.go b/libpod/container_internal_unsupported.go index f979bcbde..125329ce5 100644 --- a/libpod/container_internal_unsupported.go +++ b/libpod/container_internal_unsupported.go @@ -57,3 +57,8 @@ func (c *Container) reloadNetwork() error { func (c *Container) getUserOverrides() *lookup.Overrides { return nil } + +// Fix ownership and permissions of the specified volume if necessary. +func (c *Container) fixVolumePermissions(v *ContainerNamedVolume) error { + return define.ErrNotImplemented +} diff --git a/libpod/container_log.go b/libpod/container_log.go index c207df819..a30e4f5cc 100644 --- a/libpod/container_log.go +++ b/libpod/container_log.go @@ -4,11 +4,10 @@ import ( "context" "fmt" "os" - "time" "github.com/containers/podman/v3/libpod/define" + "github.com/containers/podman/v3/libpod/events" "github.com/containers/podman/v3/libpod/logs" - "github.com/hpcloud/tail/watch" "github.com/pkg/errors" "github.com/sirupsen/logrus" ) @@ -94,27 +93,40 @@ func (c *Container) readFromLogFile(ctx context.Context, options *logs.LogOption }() // Check if container is still running or paused if options.Follow { + state, err := c.State() + if err != nil || state != define.ContainerStateRunning { + // If the container isn't running or if we encountered + // an error getting its state, instruct the logger to + // read the file until EOF. + tailError := t.StopAtEOF() + if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" { + logrus.Error(tailError) + } + if errors.Cause(err) != define.ErrNoSuchCtr { + logrus.Error(err) + } + return nil + } + + // The container is running, so we need to wait until the container exited go func() { - for { - state, err := c.State() - time.Sleep(watch.POLL_DURATION) - if err != nil { - tailError := t.StopAtEOF() - if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" { - logrus.Error(tailError) - } - if errors.Cause(err) != define.ErrNoSuchCtr { - logrus.Error(err) - } - break - } - if state != define.ContainerStateRunning && state != define.ContainerStatePaused { - tailError := t.StopAtEOF() - if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" { - logrus.Error(tailError) - } - break + eventChannel := make(chan *events.Event) + eventOptions := events.ReadOptions{ + EventChannel: eventChannel, + Filters: []string{"event=died", "container=" + c.ID()}, + Stream: true, + } + go func() { + if err := c.runtime.Events(ctx, eventOptions); err != nil { + logrus.Errorf("Error waiting for container to exit: %v", err) } + }() + // Now wait for the died event and signal to finish + // reading the log until EOF. + <-eventChannel + tailError := t.StopAtEOF() + if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" { + logrus.Error(tailError) } }() } diff --git a/libpod/events.go b/libpod/events.go index 839229674..22c51aeec 100644 --- a/libpod/events.go +++ b/libpod/events.go @@ -46,7 +46,22 @@ func (c *Container) newContainerExitedEvent(exitCode int32) { e.Type = events.Container e.ContainerExitCode = int(exitCode) if err := c.runtime.eventer.Write(e); err != nil { - logrus.Errorf("unable to write pod event: %q", err) + logrus.Errorf("unable to write container exited event: %q", err) + } +} + +// newExecDiedEvent creates a new event for an exec session's death +func (c *Container) newExecDiedEvent(sessionID string, exitCode int) { + e := events.NewEvent(events.ExecDied) + e.ID = c.ID() + e.Name = c.Name() + e.Image = c.config.RootfsImageName + e.Type = events.Container + e.ContainerExitCode = exitCode + e.Attributes = make(map[string]string) + e.Attributes["execID"] = sessionID + if err := c.runtime.eventer.Write(e); err != nil { + logrus.Errorf("unable to write exec died event: %q", err) } } @@ -154,3 +169,25 @@ func (r *Runtime) GetLastContainerEvent(ctx context.Context, nameOrID string, co // return the last element in the slice return containerEvents[len(containerEvents)-1], nil } + +// GetExecDiedEvent takes a container name or ID, exec session ID, and returns +// that exec session's Died event (if it has already occurred). +func (r *Runtime) GetExecDiedEvent(ctx context.Context, nameOrID, execSessionID string) (*events.Event, error) { + filters := []string{ + fmt.Sprintf("container=%s", nameOrID), + "event=exec_died", + "type=container", + fmt.Sprintf("label=execID=%s", execSessionID), + } + + containerEvents, err := r.GetEvents(ctx, filters) + if err != nil { + return nil, err + } + // There *should* only be one event maximum. + // But... just in case... let's not blow up if there's more than one. + if len(containerEvents) < 1 { + return nil, errors.Wrapf(events.ErrEventNotFound, "exec died event for session %s (container %s) not found", execSessionID, nameOrID) + } + return containerEvents[len(containerEvents)-1], nil +} diff --git a/libpod/events/config.go b/libpod/events/config.go index 085fa9d52..d88d7b6e3 100644 --- a/libpod/events/config.go +++ b/libpod/events/config.go @@ -127,6 +127,8 @@ const ( Create Status = "create" // Exec ... Exec Status = "exec" + // ExecDied indicates that an exec session in a container died. + ExecDied Status = "exec_died" // Exited indicates that a container's process died Exited Status = "died" // Export ... diff --git a/libpod/events/events.go b/libpod/events/events.go index 01ea6a386..e03215eff 100644 --- a/libpod/events/events.go +++ b/libpod/events/events.go @@ -149,6 +149,8 @@ func StringToStatus(name string) (Status, error) { return Create, nil case Exec.String(): return Exec, nil + case ExecDied.String(): + return ExecDied, nil case Exited.String(): return Exited, nil case Export.String(): diff --git a/libpod/networking_linux.go b/libpod/networking_linux.go index c928e02a6..5446841f6 100644 --- a/libpod/networking_linux.go +++ b/libpod/networking_linux.go @@ -1090,7 +1090,7 @@ func (c *Container) NetworkDisconnect(nameOrID, netName string, force bool) erro } c.newNetworkEvent(events.NetworkDisconnect, netName) - if c.state.State != define.ContainerStateRunning { + if !c.ensureState(define.ContainerStateRunning, define.ContainerStateCreated) { return nil } @@ -1145,7 +1145,7 @@ func (c *Container) NetworkConnect(nameOrID, netName string, aliases []string) e return err } c.newNetworkEvent(events.NetworkConnect, netName) - if c.state.State != define.ContainerStateRunning { + if !c.ensureState(define.ContainerStateRunning, define.ContainerStateCreated) { return nil } if c.state.NetNS == nil { diff --git a/libpod/oci_conmon_linux.go b/libpod/oci_conmon_linux.go index 3da49b85f..2914bd1a1 100644 --- a/libpod/oci_conmon_linux.go +++ b/libpod/oci_conmon_linux.go @@ -787,7 +787,11 @@ func (r *ConmonOCIRuntime) CheckpointContainer(ctr *Container, options Container args = append(args, "--pre-dump") } if !options.PreCheckPoint && options.WithPrevious { - args = append(args, "--parent-path", ctr.PreCheckPointPath()) + args = append( + args, + "--parent-path", + filepath.Join("..", preCheckpointDir), + ) } runtimeDir, err := util.GetRuntimeDir() if err != nil { diff --git a/libpod/options.go b/libpod/options.go index f942d264b..d3be46ad8 100644 --- a/libpod/options.go +++ b/libpod/options.go @@ -1641,6 +1641,19 @@ func WithVolumeGID(gid int) VolumeCreateOption { } } +// WithVolumeNoChown prevents the volume from being chowned to the process uid at first use. +func WithVolumeNoChown() VolumeCreateOption { + return func(volume *Volume) error { + if volume.valid { + return define.ErrVolumeFinalized + } + + volume.state.NeedsChown = false + + return nil + } +} + // withSetAnon sets a bool notifying libpod that this volume is anonymous and // should be removed when containers using it are removed and volumes are // specified for removal. diff --git a/libpod/volume_internal.go b/libpod/volume_internal.go index 694cdd149..19008a253 100644 --- a/libpod/volume_internal.go +++ b/libpod/volume_internal.go @@ -39,8 +39,23 @@ func (v *Volume) needsMount() bool { return true } - // Local driver with options needs mount - return len(v.config.Options) > 0 + // Commit 28138dafcc added the UID and GID options to this map + // However we should only mount when options other than uid and gid are set. + // see https://github.com/containers/podman/issues/10620 + index := 0 + if _, ok := v.config.Options["UID"]; ok { + index++ + } + if _, ok := v.config.Options["GID"]; ok { + index++ + } + // when uid or gid is set there is also the "o" option + // set so we have to ignore this one as well + if index > 0 { + index++ + } + // Local driver with options other than uid,gid needs mount + return len(v.config.Options) > index } // update() updates the volume state from the DB. |