diff options
Diffstat (limited to 'libpod')
| -rw-r--r-- | libpod/container_api.go | 2 | ||||
| -rw-r--r-- | libpod/container_internal.go | 18 | ||||
| -rw-r--r-- | libpod/healthcheck.go | 59 | ||||
| -rw-r--r-- | libpod/healthcheck_linux.go | 67 | ||||
| -rw-r--r-- | libpod/healthcheck_unsupported.go | 19 | ||||
| -rw-r--r-- | libpod/oci.go | 447 | ||||
| -rw-r--r-- | libpod/oci_linux.go | 447 | ||||
| -rw-r--r-- | libpod/oci_unsupported.go | 12 | ||||
| -rw-r--r-- | libpod/options.go | 21 | ||||
| -rw-r--r-- | libpod/runtime.go | 18 | ||||
| -rw-r--r-- | libpod/runtime_migrate.go | 5 |
11 files changed, 572 insertions, 543 deletions
diff --git a/libpod/container_api.go b/libpod/container_api.go index 465b23831..5bfd869b3 100644 --- a/libpod/container_api.go +++ b/libpod/container_api.go @@ -40,7 +40,7 @@ func (c *Container) Init(ctx context.Context) (err error) { if !(c.state.State == ContainerStateConfigured || c.state.State == ContainerStateStopped || c.state.State == ContainerStateExited) { - return errors.Wrapf(ErrCtrExists, "container %s has already been created in runtime", c.ID()) + return errors.Wrapf(ErrCtrStateInvalid, "container %s has already been created in runtime", c.ID()) } // don't recursively start diff --git a/libpod/container_internal.go b/libpod/container_internal.go index 927b71b2b..a791df491 100644 --- a/libpod/container_internal.go +++ b/libpod/container_internal.go @@ -299,7 +299,7 @@ func (c *Container) setupStorage(ctx context.Context) error { return errors.Wrapf(err, "error creating container storage") } - if !rootless.IsRootless() && (len(c.config.IDMappings.UIDMap) != 0 || len(c.config.IDMappings.GIDMap) != 0) { + if len(c.config.IDMappings.UIDMap) != 0 || len(c.config.IDMappings.GIDMap) != 0 { if err := os.Chown(containerInfo.RunDir, c.RootUID(), c.RootGID()); err != nil { return err } @@ -811,8 +811,9 @@ func (c *Container) cleanupRuntime(ctx context.Context) error { span.SetTag("struct", "container") defer span.Finish() - // If the container is not ContainerStateStopped, do nothing - if c.state.State != ContainerStateStopped { + // If the container is not ContainerStateStopped or + // ContainerStateCreated, do nothing. + if c.state.State != ContainerStateStopped && c.state.State != ContainerStateCreated { return nil } @@ -825,9 +826,14 @@ func (c *Container) cleanupRuntime(ctx context.Context) error { return err } - // Our state is now Exited, as we've removed ourself from - // the runtime. - c.state.State = ContainerStateExited + // If we were Stopped, we are now Exited, as we've removed ourself + // from the runtime. + // If we were Created, we are now Configured. + if c.state.State == ContainerStateStopped { + c.state.State = ContainerStateExited + } else if c.state.State == ContainerStateCreated { + c.state.State = ContainerStateConfigured + } if c.valid { if err := c.save(); err != nil { diff --git a/libpod/healthcheck.go b/libpod/healthcheck.go index 3a6609740..5c48cc8ee 100644 --- a/libpod/healthcheck.go +++ b/libpod/healthcheck.go @@ -3,16 +3,13 @@ package libpod import ( "bufio" "bytes" - "fmt" "io/ioutil" "os" - "os/exec" "path/filepath" "strings" "time" "github.com/containers/libpod/pkg/inspect" - "github.com/coreos/go-systemd/dbus" "github.com/pkg/errors" "github.com/sirupsen/logrus" ) @@ -241,62 +238,6 @@ func (c *Container) GetHealthCheckLog() (inspect.HealthCheckResults, error) { return healthCheck, nil } -// createTimer systemd timers for healthchecks of a container -func (c *Container) createTimer() error { - if c.disableHealthCheckSystemd() { - return nil - } - podman, err := os.Executable() - if err != nil { - return errors.Wrapf(err, "failed to get path for podman for a health check timer") - } - - var cmd = []string{"--unit", fmt.Sprintf("%s", c.ID()), fmt.Sprintf("--on-unit-inactive=%s", c.HealthCheckConfig().Interval.String()), "--timer-property=AccuracySec=1s", podman, "healthcheck", "run", c.ID()} - - conn, err := dbus.NewSystemdConnection() - if err != nil { - return errors.Wrapf(err, "unable to get systemd connection to add healthchecks") - } - conn.Close() - logrus.Debugf("creating systemd-transient files: %s %s", "systemd-run", cmd) - systemdRun := exec.Command("systemd-run", cmd...) - _, err = systemdRun.CombinedOutput() - if err != nil { - return err - } - return nil -} - -// startTimer starts a systemd timer for the healthchecks -func (c *Container) startTimer() error { - if c.disableHealthCheckSystemd() { - return nil - } - conn, err := dbus.NewSystemdConnection() - if err != nil { - return errors.Wrapf(err, "unable to get systemd connection to start healthchecks") - } - defer conn.Close() - _, err = conn.StartUnit(fmt.Sprintf("%s.service", c.ID()), "fail", nil) - return err -} - -// removeTimer removes the systemd timer and unit files -// for the container -func (c *Container) removeTimer() error { - if c.disableHealthCheckSystemd() { - return nil - } - conn, err := dbus.NewSystemdConnection() - if err != nil { - return errors.Wrapf(err, "unable to get systemd connection to remove healthchecks") - } - defer conn.Close() - serviceFile := fmt.Sprintf("%s.timer", c.ID()) - _, err = conn.StopUnit(serviceFile, "fail", nil) - return err -} - // HealthCheckStatus returns the current state of a container with a healthcheck func (c *Container) HealthCheckStatus() (string, error) { if !c.HasHealthCheck() { diff --git a/libpod/healthcheck_linux.go b/libpod/healthcheck_linux.go new file mode 100644 index 000000000..869605ea8 --- /dev/null +++ b/libpod/healthcheck_linux.go @@ -0,0 +1,67 @@ +package libpod + +import ( + "fmt" + "os" + "os/exec" + + "github.com/coreos/go-systemd/dbus" + "github.com/pkg/errors" + "github.com/sirupsen/logrus" +) + +// createTimer systemd timers for healthchecks of a container +func (c *Container) createTimer() error { + if c.disableHealthCheckSystemd() { + return nil + } + podman, err := os.Executable() + if err != nil { + return errors.Wrapf(err, "failed to get path for podman for a health check timer") + } + + var cmd = []string{"--unit", fmt.Sprintf("%s", c.ID()), fmt.Sprintf("--on-unit-inactive=%s", c.HealthCheckConfig().Interval.String()), "--timer-property=AccuracySec=1s", podman, "healthcheck", "run", c.ID()} + + conn, err := dbus.NewSystemdConnection() + if err != nil { + return errors.Wrapf(err, "unable to get systemd connection to add healthchecks") + } + conn.Close() + logrus.Debugf("creating systemd-transient files: %s %s", "systemd-run", cmd) + systemdRun := exec.Command("systemd-run", cmd...) + _, err = systemdRun.CombinedOutput() + if err != nil { + return err + } + return nil +} + +// startTimer starts a systemd timer for the healthchecks +func (c *Container) startTimer() error { + if c.disableHealthCheckSystemd() { + return nil + } + conn, err := dbus.NewSystemdConnection() + if err != nil { + return errors.Wrapf(err, "unable to get systemd connection to start healthchecks") + } + defer conn.Close() + _, err = conn.StartUnit(fmt.Sprintf("%s.service", c.ID()), "fail", nil) + return err +} + +// removeTimer removes the systemd timer and unit files +// for the container +func (c *Container) removeTimer() error { + if c.disableHealthCheckSystemd() { + return nil + } + conn, err := dbus.NewSystemdConnection() + if err != nil { + return errors.Wrapf(err, "unable to get systemd connection to remove healthchecks") + } + defer conn.Close() + serviceFile := fmt.Sprintf("%s.timer", c.ID()) + _, err = conn.StopUnit(serviceFile, "fail", nil) + return err +} diff --git a/libpod/healthcheck_unsupported.go b/libpod/healthcheck_unsupported.go new file mode 100644 index 000000000..d01d1ccd4 --- /dev/null +++ b/libpod/healthcheck_unsupported.go @@ -0,0 +1,19 @@ +// +build !linux + +package libpod + +// createTimer systemd timers for healthchecks of a container +func (c *Container) createTimer() error { + return ErrNotImplemented +} + +// startTimer starts a systemd timer for the healthchecks +func (c *Container) startTimer() error { + return ErrNotImplemented +} + +// removeTimer removes the systemd timer and unit files +// for the container +func (c *Container) removeTimer() error { + return ErrNotImplemented +} diff --git a/libpod/oci.go b/libpod/oci.go index 189359753..3dfde4f24 100644 --- a/libpod/oci.go +++ b/libpod/oci.go @@ -1,7 +1,6 @@ package libpod import ( - "bufio" "bytes" "fmt" "io/ioutil" @@ -9,21 +8,15 @@ import ( "os" "os/exec" "path/filepath" - "runtime" "strings" - "syscall" "time" - "github.com/containers/libpod/pkg/rootless" "github.com/containers/libpod/pkg/util" - "github.com/coreos/go-systemd/activation" "github.com/cri-o/ocicni/pkg/ocicni" spec "github.com/opencontainers/runtime-spec/specs-go" - "github.com/opencontainers/selinux/go-selinux" "github.com/opencontainers/selinux/go-selinux/label" "github.com/pkg/errors" "github.com/sirupsen/logrus" - "golang.org/x/sys/unix" kwait "k8s.io/apimachinery/pkg/util/wait" // TODO import these functions into libpod and remove the import @@ -118,70 +111,6 @@ func createUnitName(prefix string, name string) string { return fmt.Sprintf("%s-%s.scope", prefix, name) } -// Wait for a container which has been sent a signal to stop -func waitContainerStop(ctr *Container, timeout time.Duration) error { - done := make(chan struct{}) - chControl := make(chan struct{}) - go func() { - for { - select { - case <-chControl: - return - default: - // Check if the process is still around - err := unix.Kill(ctr.state.PID, 0) - if err == unix.ESRCH { - close(done) - return - } - time.Sleep(100 * time.Millisecond) - } - } - }() - select { - case <-done: - return nil - case <-time.After(timeout): - close(chControl) - logrus.Debugf("container %s did not die within timeout %d", ctr.ID(), timeout) - return errors.Errorf("container %s did not die within timeout", ctr.ID()) - } -} - -// Wait for a set of given PIDs to stop -func waitPidsStop(pids []int, timeout time.Duration) error { - done := make(chan struct{}) - chControl := make(chan struct{}) - go func() { - for { - select { - case <-chControl: - return - default: - allClosed := true - for _, pid := range pids { - if err := unix.Kill(pid, 0); err != unix.ESRCH { - allClosed = false - break - } - } - if allClosed { - close(done) - return - } - time.Sleep(100 * time.Millisecond) - } - } - }() - select { - case <-done: - return nil - case <-time.After(timeout): - close(chControl) - return errors.Errorf("given PIDs did not die within timeout") - } -} - func bindPorts(ports []ocicni.PortMapping) ([]*os.File, error) { var files []*os.File notifySCTP := false @@ -234,241 +163,6 @@ func bindPorts(ports []ocicni.PortMapping) ([]*os.File, error) { return files, nil } -func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, restoreOptions *ContainerCheckpointOptions) (err error) { - var stderrBuf bytes.Buffer - - runtimeDir, err := util.GetRootlessRuntimeDir() - if err != nil { - return err - } - - parentPipe, childPipe, err := newPipe() - if err != nil { - return errors.Wrapf(err, "error creating socket pair") - } - - childStartPipe, parentStartPipe, err := newPipe() - if err != nil { - return errors.Wrapf(err, "error creating socket pair for start pipe") - } - - defer parentPipe.Close() - defer parentStartPipe.Close() - - args := []string{} - if r.cgroupManager == SystemdCgroupsManager { - args = append(args, "-s") - } - args = append(args, "-c", ctr.ID()) - args = append(args, "-u", ctr.ID()) - args = append(args, "-r", r.path) - args = append(args, "-b", ctr.bundlePath()) - args = append(args, "-p", filepath.Join(ctr.state.RunDir, "pidfile")) - args = append(args, "-l", ctr.LogPath()) - args = append(args, "--exit-dir", r.exitsDir) - if ctr.config.ConmonPidFile != "" { - args = append(args, "--conmon-pidfile", ctr.config.ConmonPidFile) - } - if len(ctr.config.ExitCommand) > 0 { - args = append(args, "--exit-command", ctr.config.ExitCommand[0]) - for _, arg := range ctr.config.ExitCommand[1:] { - args = append(args, []string{"--exit-command-arg", arg}...) - } - } - args = append(args, "--socket-dir-path", r.socketsDir) - if ctr.config.Spec.Process.Terminal { - args = append(args, "-t") - } else if ctr.config.Stdin { - args = append(args, "-i") - } - if r.logSizeMax >= 0 { - args = append(args, "--log-size-max", fmt.Sprintf("%v", r.logSizeMax)) - } - if r.noPivot { - args = append(args, "--no-pivot") - } - - logLevel := logrus.GetLevel() - args = append(args, "--log-level", logLevel.String()) - - if logLevel == logrus.DebugLevel { - logrus.Debugf("%s messages will be logged to syslog", r.conmonPath) - args = append(args, "--syslog") - } - - if restoreOptions != nil { - args = append(args, "--restore", ctr.CheckpointPath()) - if restoreOptions.TCPEstablished { - args = append(args, "--restore-arg", "--tcp-established") - } - } - - logrus.WithFields(logrus.Fields{ - "args": args, - }).Debugf("running conmon: %s", r.conmonPath) - - cmd := exec.Command(r.conmonPath, args...) - cmd.Dir = ctr.bundlePath() - cmd.SysProcAttr = &syscall.SysProcAttr{ - Setpgid: true, - } - // TODO this is probably a really bad idea for some uses - // Make this configurable - cmd.Stdin = os.Stdin - cmd.Stdout = os.Stdout - cmd.Stderr = os.Stderr - if ctr.config.Spec.Process.Terminal { - cmd.Stderr = &stderrBuf - } - - cmd.ExtraFiles = append(cmd.ExtraFiles, childPipe, childStartPipe) - // 0, 1 and 2 are stdin, stdout and stderr - cmd.Env = append(r.conmonEnv, fmt.Sprintf("_OCI_SYNCPIPE=%d", 3)) - cmd.Env = append(cmd.Env, fmt.Sprintf("_OCI_STARTPIPE=%d", 4)) - cmd.Env = append(cmd.Env, fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)) - cmd.Env = append(cmd.Env, fmt.Sprintf("_CONTAINERS_USERNS_CONFIGURED=%s", os.Getenv("_CONTAINERS_USERNS_CONFIGURED"))) - cmd.Env = append(cmd.Env, fmt.Sprintf("_CONTAINERS_ROOTLESS_UID=%s", os.Getenv("_CONTAINERS_ROOTLESS_UID"))) - cmd.Env = append(cmd.Env, fmt.Sprintf("HOME=%s", os.Getenv("HOME"))) - - if r.reservePorts && !ctr.config.NetMode.IsSlirp4netns() { - ports, err := bindPorts(ctr.config.PortMappings) - if err != nil { - return err - } - - // Leak the port we bound in the conmon process. These fd's won't be used - // by the container and conmon will keep the ports busy so that another - // process cannot use them. - cmd.ExtraFiles = append(cmd.ExtraFiles, ports...) - } - - if ctr.config.NetMode.IsSlirp4netns() { - ctr.rootlessSlirpSyncR, ctr.rootlessSlirpSyncW, err = os.Pipe() - if err != nil { - return errors.Wrapf(err, "failed to create rootless network sync pipe") - } - // Leak one end in conmon, the other one will be leaked into slirp4netns - cmd.ExtraFiles = append(cmd.ExtraFiles, ctr.rootlessSlirpSyncW) - } - - if notify, ok := os.LookupEnv("NOTIFY_SOCKET"); ok { - cmd.Env = append(cmd.Env, fmt.Sprintf("NOTIFY_SOCKET=%s", notify)) - } - if listenfds, ok := os.LookupEnv("LISTEN_FDS"); ok { - cmd.Env = append(cmd.Env, fmt.Sprintf("LISTEN_FDS=%s", listenfds), "LISTEN_PID=1") - fds := activation.Files(false) - cmd.ExtraFiles = append(cmd.ExtraFiles, fds...) - } - if selinux.GetEnabled() { - // Set the label of the conmon process to be level :s0 - // This will allow the container processes to talk to fifo-files - // passed into the container by conmon - var ( - plabel string - con selinux.Context - ) - plabel, err = selinux.CurrentLabel() - if err != nil { - childPipe.Close() - return errors.Wrapf(err, "Failed to get current SELinux label") - } - - con, err = selinux.NewContext(plabel) - if err != nil { - return errors.Wrapf(err, "Failed to get new context from SELinux label") - } - - runtime.LockOSThread() - if con["level"] != "s0" && con["level"] != "" { - con["level"] = "s0" - if err = label.SetProcessLabel(con.Get()); err != nil { - runtime.UnlockOSThread() - return err - } - } - err = cmd.Start() - // Ignore error returned from SetProcessLabel("") call, - // can't recover. - label.SetProcessLabel("") - runtime.UnlockOSThread() - } else { - err = cmd.Start() - } - if err != nil { - childPipe.Close() - return err - } - defer cmd.Wait() - - // We don't need childPipe on the parent side - childPipe.Close() - childStartPipe.Close() - - // Move conmon to specified cgroup - if err := r.moveConmonToCgroup(ctr, cgroupParent, cmd); err != nil { - return err - } - - /* We set the cgroup, now the child can start creating children */ - someData := []byte{0} - _, err = parentStartPipe.Write(someData) - if err != nil { - return err - } - - /* Wait for initial setup and fork, and reap child */ - err = cmd.Wait() - if err != nil { - return err - } - - defer func() { - if err != nil { - if err2 := r.deleteContainer(ctr); err2 != nil { - logrus.Errorf("Error removing container %s from runtime after creation failed", ctr.ID()) - } - } - }() - - // Wait to get container pid from conmon - type syncStruct struct { - si *syncInfo - err error - } - ch := make(chan syncStruct) - go func() { - var si *syncInfo - rdr := bufio.NewReader(parentPipe) - b, err := rdr.ReadBytes('\n') - if err != nil { - ch <- syncStruct{err: err} - } - if err := json.Unmarshal(b, &si); err != nil { - ch <- syncStruct{err: err} - return - } - ch <- syncStruct{si: si} - }() - - select { - case ss := <-ch: - if ss.err != nil { - return errors.Wrapf(ss.err, "error reading container (probably exited) json message") - } - logrus.Debugf("Received container pid: %d", ss.si.Pid) - if ss.si.Pid == -1 { - if ss.si.Message != "" { - return errors.Wrapf(ErrInternal, "container create failed: %s", ss.si.Message) - } - return errors.Wrapf(ErrInternal, "container create failed") - } - ctr.state.PID = ss.si.Pid - case <-time.After(ContainerCreateTimeout): - return errors.Wrapf(ErrInternal, "container creation timeout") - } - return nil -} - // updateContainerStatus retrieves the current status of the container from the // runtime. It updates the container's state but does not save it. // If useRunc is false, we will not directly hit runc to see the container's @@ -631,82 +325,6 @@ func (r *OCIRuntime) killContainer(ctr *Container, signal uint) error { return nil } -// stopContainer stops a container, first using its given stop signal (or -// SIGTERM if no signal was specified), then using SIGKILL -// Timeout is given in seconds. If timeout is 0, the container will be -// immediately kill with SIGKILL -// Does not set finished time for container, assumes you will run updateStatus -// after to pull the exit code -func (r *OCIRuntime) stopContainer(ctr *Container, timeout uint) error { - logrus.Debugf("Stopping container %s (PID %d)", ctr.ID(), ctr.state.PID) - - // Ping the container to see if it's alive - // If it's not, it's already stopped, return - err := unix.Kill(ctr.state.PID, 0) - if err == unix.ESRCH { - return nil - } - - stopSignal := ctr.config.StopSignal - if stopSignal == 0 { - stopSignal = uint(syscall.SIGTERM) - } - - if timeout > 0 { - if err := r.killContainer(ctr, stopSignal); err != nil { - // Is the container gone? - // If so, it probably died between the first check and - // our sending the signal - // The container is stopped, so exit cleanly - err := unix.Kill(ctr.state.PID, 0) - if err == unix.ESRCH { - return nil - } - - return err - } - - if err := waitContainerStop(ctr, time.Duration(timeout)*time.Second); err != nil { - logrus.Warnf("Timed out stopping container %s, resorting to SIGKILL", ctr.ID()) - } else { - // No error, the container is dead - return nil - } - } - - var args []string - if rootless.IsRootless() { - // we don't use --all for rootless containers as the OCI runtime might use - // the cgroups to determine the PIDs, but for rootless containers there is - // not any. - args = []string{"kill", ctr.ID(), "KILL"} - } else { - args = []string{"kill", "--all", ctr.ID(), "KILL"} - } - - runtimeDir, err := util.GetRootlessRuntimeDir() - if err != nil { - return err - } - env := []string{fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)} - if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, args...); err != nil { - // Again, check if the container is gone. If it is, exit cleanly. - err := unix.Kill(ctr.state.PID, 0) - if err == unix.ESRCH { - return nil - } - - return errors.Wrapf(err, "error sending SIGKILL to container %s", ctr.ID()) - } - - // Give runtime a few seconds to make it happen - if err := waitContainerStop(ctr, killContainerTimeout); err != nil { - return err - } - - return nil -} - // deleteContainer deletes a container from the OCI runtime func (r *OCIRuntime) deleteContainer(ctr *Container) error { runtimeDir, err := util.GetRootlessRuntimeDir() @@ -834,71 +452,6 @@ func (r *OCIRuntime) execContainer(c *Container, cmd, capAdd, env []string, tty return execCmd, nil } -// execStopContainer stops all active exec sessions in a container -// It will also stop all other processes in the container. It is only intended -// to be used to assist in cleanup when removing a container. -// SIGTERM is used by default to stop processes. If SIGTERM fails, SIGKILL will be used. -func (r *OCIRuntime) execStopContainer(ctr *Container, timeout uint) error { - // Do we have active exec sessions? - if len(ctr.state.ExecSessions) == 0 { - return nil - } - - // Get a list of active exec sessions - execSessions := []int{} - for _, session := range ctr.state.ExecSessions { - pid := session.PID - // Ping the PID with signal 0 to see if it still exists - if err := unix.Kill(pid, 0); err == unix.ESRCH { - continue - } - - execSessions = append(execSessions, pid) - } - - // All the sessions may be dead - // If they are, just return - if len(execSessions) == 0 { - return nil - } - runtimeDir, err := util.GetRootlessRuntimeDir() - if err != nil { - return err - } - env := []string{fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)} - - // If timeout is 0, just use SIGKILL - if timeout > 0 { - // Stop using SIGTERM by default - // Use SIGSTOP after a timeout - logrus.Debugf("Killing all processes in container %s with SIGTERM", ctr.ID()) - if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, "kill", "--all", ctr.ID(), "TERM"); err != nil { - return errors.Wrapf(err, "error sending SIGTERM to container %s processes", ctr.ID()) - } - - // Wait for all processes to stop - if err := waitPidsStop(execSessions, time.Duration(timeout)*time.Second); err != nil { - logrus.Warnf("Timed out stopping container %s exec sessions", ctr.ID()) - } else { - // No error, all exec sessions are dead - return nil - } - } - - // Send SIGKILL - logrus.Debugf("Killing all processes in container %s with SIGKILL", ctr.ID()) - if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, "kill", "--all", ctr.ID(), "KILL"); err != nil { - return errors.Wrapf(err, "error sending SIGKILL to container %s processes", ctr.ID()) - } - - // Give the processes a few seconds to go down - if err := waitPidsStop(execSessions, killContainerTimeout); err != nil { - return errors.Wrapf(err, "failed to kill container %s exec sessions", ctr.ID()) - } - - return nil -} - // checkpointContainer checkpoints the given container func (r *OCIRuntime) checkpointContainer(ctr *Container, options ContainerCheckpointOptions) error { label.SetSocketLabel(ctr.ProcessLabel()) diff --git a/libpod/oci_linux.go b/libpod/oci_linux.go index 1f5411c1f..1c1e4a203 100644 --- a/libpod/oci_linux.go +++ b/libpod/oci_linux.go @@ -3,6 +3,8 @@ package libpod import ( + "bufio" + "bytes" "fmt" "os" "os/exec" @@ -10,12 +12,17 @@ import ( "runtime" "strings" "syscall" + "time" "github.com/containerd/cgroups" "github.com/containers/libpod/pkg/rootless" + "github.com/containers/libpod/pkg/util" "github.com/containers/libpod/utils" pmount "github.com/containers/storage/pkg/mount" + "github.com/coreos/go-systemd/activation" spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/opencontainers/selinux/go-selinux" + "github.com/opencontainers/selinux/go-selinux/label" "github.com/pkg/errors" "github.com/sirupsen/logrus" "golang.org/x/sys/unix" @@ -179,3 +186,443 @@ func (r *OCIRuntime) conmonPackage() string { } return dpkgVersion(r.conmonPath) } + +func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, restoreOptions *ContainerCheckpointOptions) (err error) { + var stderrBuf bytes.Buffer + + runtimeDir, err := util.GetRootlessRuntimeDir() + if err != nil { + return err + } + + parentPipe, childPipe, err := newPipe() + if err != nil { + return errors.Wrapf(err, "error creating socket pair") + } + + childStartPipe, parentStartPipe, err := newPipe() + if err != nil { + return errors.Wrapf(err, "error creating socket pair for start pipe") + } + + defer parentPipe.Close() + defer parentStartPipe.Close() + + args := []string{} + if r.cgroupManager == SystemdCgroupsManager { + args = append(args, "-s") + } + args = append(args, "-c", ctr.ID()) + args = append(args, "-u", ctr.ID()) + args = append(args, "-r", r.path) + args = append(args, "-b", ctr.bundlePath()) + args = append(args, "-p", filepath.Join(ctr.state.RunDir, "pidfile")) + args = append(args, "-l", ctr.LogPath()) + args = append(args, "--exit-dir", r.exitsDir) + if ctr.config.ConmonPidFile != "" { + args = append(args, "--conmon-pidfile", ctr.config.ConmonPidFile) + } + if len(ctr.config.ExitCommand) > 0 { + args = append(args, "--exit-command", ctr.config.ExitCommand[0]) + for _, arg := range ctr.config.ExitCommand[1:] { + args = append(args, []string{"--exit-command-arg", arg}...) + } + } + args = append(args, "--socket-dir-path", r.socketsDir) + if ctr.config.Spec.Process.Terminal { + args = append(args, "-t") + } else if ctr.config.Stdin { + args = append(args, "-i") + } + if r.logSizeMax >= 0 { + args = append(args, "--log-size-max", fmt.Sprintf("%v", r.logSizeMax)) + } + if r.noPivot { + args = append(args, "--no-pivot") + } + + logLevel := logrus.GetLevel() + args = append(args, "--log-level", logLevel.String()) + + if logLevel == logrus.DebugLevel { + logrus.Debugf("%s messages will be logged to syslog", r.conmonPath) + args = append(args, "--syslog") + } + + if restoreOptions != nil { + args = append(args, "--restore", ctr.CheckpointPath()) + if restoreOptions.TCPEstablished { + args = append(args, "--restore-arg", "--tcp-established") + } + } + + logrus.WithFields(logrus.Fields{ + "args": args, + }).Debugf("running conmon: %s", r.conmonPath) + + cmd := exec.Command(r.conmonPath, args...) + cmd.Dir = ctr.bundlePath() + cmd.SysProcAttr = &syscall.SysProcAttr{ + Setpgid: true, + } + // TODO this is probably a really bad idea for some uses + // Make this configurable + cmd.Stdin = os.Stdin + cmd.Stdout = os.Stdout + cmd.Stderr = os.Stderr + if ctr.config.Spec.Process.Terminal { + cmd.Stderr = &stderrBuf + } + + cmd.ExtraFiles = append(cmd.ExtraFiles, childPipe, childStartPipe) + // 0, 1 and 2 are stdin, stdout and stderr + cmd.Env = append(r.conmonEnv, fmt.Sprintf("_OCI_SYNCPIPE=%d", 3)) + cmd.Env = append(cmd.Env, fmt.Sprintf("_OCI_STARTPIPE=%d", 4)) + cmd.Env = append(cmd.Env, fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)) + cmd.Env = append(cmd.Env, fmt.Sprintf("_CONTAINERS_USERNS_CONFIGURED=%s", os.Getenv("_CONTAINERS_USERNS_CONFIGURED"))) + cmd.Env = append(cmd.Env, fmt.Sprintf("_CONTAINERS_ROOTLESS_UID=%s", os.Getenv("_CONTAINERS_ROOTLESS_UID"))) + cmd.Env = append(cmd.Env, fmt.Sprintf("HOME=%s", os.Getenv("HOME"))) + + if r.reservePorts && !ctr.config.NetMode.IsSlirp4netns() { + ports, err := bindPorts(ctr.config.PortMappings) + if err != nil { + return err + } + + // Leak the port we bound in the conmon process. These fd's won't be used + // by the container and conmon will keep the ports busy so that another + // process cannot use them. + cmd.ExtraFiles = append(cmd.ExtraFiles, ports...) + } + + if ctr.config.NetMode.IsSlirp4netns() { + ctr.rootlessSlirpSyncR, ctr.rootlessSlirpSyncW, err = os.Pipe() + if err != nil { + return errors.Wrapf(err, "failed to create rootless network sync pipe") + } + // Leak one end in conmon, the other one will be leaked into slirp4netns + cmd.ExtraFiles = append(cmd.ExtraFiles, ctr.rootlessSlirpSyncW) + } + + if notify, ok := os.LookupEnv("NOTIFY_SOCKET"); ok { + cmd.Env = append(cmd.Env, fmt.Sprintf("NOTIFY_SOCKET=%s", notify)) + } + if listenfds, ok := os.LookupEnv("LISTEN_FDS"); ok { + cmd.Env = append(cmd.Env, fmt.Sprintf("LISTEN_FDS=%s", listenfds), "LISTEN_PID=1") + fds := activation.Files(false) + cmd.ExtraFiles = append(cmd.ExtraFiles, fds...) + } + if selinux.GetEnabled() { + // Set the label of the conmon process to be level :s0 + // This will allow the container processes to talk to fifo-files + // passed into the container by conmon + var ( + plabel string + con selinux.Context + ) + plabel, err = selinux.CurrentLabel() + if err != nil { + childPipe.Close() + return errors.Wrapf(err, "Failed to get current SELinux label") + } + + con, err = selinux.NewContext(plabel) + if err != nil { + return errors.Wrapf(err, "Failed to get new context from SELinux label") + } + + runtime.LockOSThread() + if con["level"] != "s0" && con["level"] != "" { + con["level"] = "s0" + if err = label.SetProcessLabel(con.Get()); err != nil { + runtime.UnlockOSThread() + return err + } + } + err = cmd.Start() + // Ignore error returned from SetProcessLabel("") call, + // can't recover. + label.SetProcessLabel("") + runtime.UnlockOSThread() + } else { + err = cmd.Start() + } + if err != nil { + childPipe.Close() + return err + } + defer cmd.Wait() + + // We don't need childPipe on the parent side + childPipe.Close() + childStartPipe.Close() + + // Move conmon to specified cgroup + if err := r.moveConmonToCgroup(ctr, cgroupParent, cmd); err != nil { + return err + } + + /* We set the cgroup, now the child can start creating children */ + someData := []byte{0} + _, err = parentStartPipe.Write(someData) + if err != nil { + return err + } + + /* Wait for initial setup and fork, and reap child */ + err = cmd.Wait() + if err != nil { + return err + } + + defer func() { + if err != nil { + if err2 := r.deleteContainer(ctr); err2 != nil { + logrus.Errorf("Error removing container %s from runtime after creation failed", ctr.ID()) + } + } + }() + + // Wait to get container pid from conmon + type syncStruct struct { + si *syncInfo + err error + } + ch := make(chan syncStruct) + go func() { + var si *syncInfo + rdr := bufio.NewReader(parentPipe) + b, err := rdr.ReadBytes('\n') + if err != nil { + ch <- syncStruct{err: err} + } + if err := json.Unmarshal(b, &si); err != nil { + ch <- syncStruct{err: err} + return + } + ch <- syncStruct{si: si} + }() + + select { + case ss := <-ch: + if ss.err != nil { + return errors.Wrapf(ss.err, "error reading container (probably exited) json message") + } + logrus.Debugf("Received container pid: %d", ss.si.Pid) + if ss.si.Pid == -1 { + if ss.si.Message != "" { + return errors.Wrapf(ErrInternal, "container create failed: %s", ss.si.Message) + } + return errors.Wrapf(ErrInternal, "container create failed") + } + ctr.state.PID = ss.si.Pid + case <-time.After(ContainerCreateTimeout): + return errors.Wrapf(ErrInternal, "container creation timeout") + } + return nil +} + +// Wait for a container which has been sent a signal to stop +func waitContainerStop(ctr *Container, timeout time.Duration) error { + done := make(chan struct{}) + chControl := make(chan struct{}) + go func() { + for { + select { + case <-chControl: + return + default: + // Check if the process is still around + err := unix.Kill(ctr.state.PID, 0) + if err == unix.ESRCH { + close(done) + return + } + time.Sleep(100 * time.Millisecond) + } + } + }() + select { + case <-done: + return nil + case <-time.After(timeout): + close(chControl) + logrus.Debugf("container %s did not die within timeout %d", ctr.ID(), timeout) + return errors.Errorf("container %s did not die within timeout", ctr.ID()) + } +} + +// Wait for a set of given PIDs to stop +func waitPidsStop(pids []int, timeout time.Duration) error { + done := make(chan struct{}) + chControl := make(chan struct{}) + go func() { + for { + select { + case <-chControl: + return + default: + allClosed := true + for _, pid := range pids { + if err := unix.Kill(pid, 0); err != unix.ESRCH { + allClosed = false + break + } + } + if allClosed { + close(done) + return + } + time.Sleep(100 * time.Millisecond) + } + } + }() + select { + case <-done: + return nil + case <-time.After(timeout): + close(chControl) + return errors.Errorf("given PIDs did not die within timeout") + } +} + +// stopContainer stops a container, first using its given stop signal (or +// SIGTERM if no signal was specified), then using SIGKILL +// Timeout is given in seconds. If timeout is 0, the container will be +// immediately kill with SIGKILL +// Does not set finished time for container, assumes you will run updateStatus +// after to pull the exit code +func (r *OCIRuntime) stopContainer(ctr *Container, timeout uint) error { + logrus.Debugf("Stopping container %s (PID %d)", ctr.ID(), ctr.state.PID) + + // Ping the container to see if it's alive + // If it's not, it's already stopped, return + err := unix.Kill(ctr.state.PID, 0) + if err == unix.ESRCH { + return nil + } + + stopSignal := ctr.config.StopSignal + if stopSignal == 0 { + stopSignal = uint(syscall.SIGTERM) + } + + if timeout > 0 { + if err := r.killContainer(ctr, stopSignal); err != nil { + // Is the container gone? + // If so, it probably died between the first check and + // our sending the signal + // The container is stopped, so exit cleanly + err := unix.Kill(ctr.state.PID, 0) + if err == unix.ESRCH { + return nil + } + + return err + } + + if err := waitContainerStop(ctr, time.Duration(timeout)*time.Second); err != nil { + logrus.Warnf("Timed out stopping container %s, resorting to SIGKILL", ctr.ID()) + } else { + // No error, the container is dead + return nil + } + } + + var args []string + if rootless.IsRootless() { + // we don't use --all for rootless containers as the OCI runtime might use + // the cgroups to determine the PIDs, but for rootless containers there is + // not any. + args = []string{"kill", ctr.ID(), "KILL"} + } else { + args = []string{"kill", "--all", ctr.ID(), "KILL"} + } + + runtimeDir, err := util.GetRootlessRuntimeDir() + if err != nil { + return err + } + env := []string{fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)} + if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, args...); err != nil { + // Again, check if the container is gone. If it is, exit cleanly. + err := unix.Kill(ctr.state.PID, 0) + if err == unix.ESRCH { + return nil + } + + return errors.Wrapf(err, "error sending SIGKILL to container %s", ctr.ID()) + } + + // Give runtime a few seconds to make it happen + if err := waitContainerStop(ctr, killContainerTimeout); err != nil { + return err + } + + return nil +} + +// execStopContainer stops all active exec sessions in a container +// It will also stop all other processes in the container. It is only intended +// to be used to assist in cleanup when removing a container. +// SIGTERM is used by default to stop processes. If SIGTERM fails, SIGKILL will be used. +func (r *OCIRuntime) execStopContainer(ctr *Container, timeout uint) error { + // Do we have active exec sessions? + if len(ctr.state.ExecSessions) == 0 { + return nil + } + + // Get a list of active exec sessions + execSessions := []int{} + for _, session := range ctr.state.ExecSessions { + pid := session.PID + // Ping the PID with signal 0 to see if it still exists + if err := unix.Kill(pid, 0); err == unix.ESRCH { + continue + } + + execSessions = append(execSessions, pid) + } + + // All the sessions may be dead + // If they are, just return + if len(execSessions) == 0 { + return nil + } + runtimeDir, err := util.GetRootlessRuntimeDir() + if err != nil { + return err + } + env := []string{fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)} + + // If timeout is 0, just use SIGKILL + if timeout > 0 { + // Stop using SIGTERM by default + // Use SIGSTOP after a timeout + logrus.Debugf("Killing all processes in container %s with SIGTERM", ctr.ID()) + if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, "kill", "--all", ctr.ID(), "TERM"); err != nil { + return errors.Wrapf(err, "error sending SIGTERM to container %s processes", ctr.ID()) + } + + // Wait for all processes to stop + if err := waitPidsStop(execSessions, time.Duration(timeout)*time.Second); err != nil { + logrus.Warnf("Timed out stopping container %s exec sessions", ctr.ID()) + } else { + // No error, all exec sessions are dead + return nil + } + } + + // Send SIGKILL + logrus.Debugf("Killing all processes in container %s with SIGKILL", ctr.ID()) + if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, "kill", "--all", ctr.ID(), "KILL"); err != nil { + return errors.Wrapf(err, "error sending SIGKILL to container %s processes", ctr.ID()) + } + + // Give the processes a few seconds to go down + if err := waitPidsStop(execSessions, killContainerTimeout); err != nil { + return errors.Wrapf(err, "failed to kill container %s exec sessions", ctr.ID()) + } + + return nil +} diff --git a/libpod/oci_unsupported.go b/libpod/oci_unsupported.go index 8c084d1e2..12183faf3 100644 --- a/libpod/oci_unsupported.go +++ b/libpod/oci_unsupported.go @@ -26,3 +26,15 @@ func (r *OCIRuntime) pathPackage() string { func (r *OCIRuntime) conmonPackage() string { return "" } + +func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, restoreOptions *ContainerCheckpointOptions) (err error) { + return ErrOSNotSupported +} + +func (r *OCIRuntime) execStopContainer(ctr *Container, timeout uint) error { + return ErrOSNotSupported +} + +func (r *OCIRuntime) stopContainer(ctr *Container, timeout uint) error { + return ErrOSNotSupported +} diff --git a/libpod/options.go b/libpod/options.go index dde9bc64c..86c04db09 100644 --- a/libpod/options.go +++ b/libpod/options.go @@ -1,7 +1,6 @@ package libpod import ( - "context" "net" "os" "path/filepath" @@ -438,10 +437,9 @@ func WithRenumber() RuntimeOption { } } -// WithMigrate instructs libpod to perform a lock migrateing while -// initializing. This will handle migrations from early versions of libpod with -// file locks to newer versions with SHM locking, as well as changes in the -// number of configured locks. +// WithMigrate instructs libpod to migrate container configurations to account +// for changes between Libpod versions. All running containers will be stopped +// during a migration, then restarted after the migration is complete. func WithMigrate() RuntimeOption { return func(rt *Runtime) error { if rt.valid { @@ -468,19 +466,6 @@ func WithShmDir(dir string) CtrCreateOption { } } -// WithContext sets the context to use. -func WithContext(ctx context.Context) RuntimeOption { - return func(rt *Runtime) error { - if rt.valid { - return ErrRuntimeFinalized - } - - rt.ctx = ctx - - return nil - } -} - // WithSystemd turns on systemd mode in the container func WithSystemd() CtrCreateOption { return func(ctr *Container) error { diff --git a/libpod/runtime.go b/libpod/runtime.go index e85242028..6b8d97fd9 100644 --- a/libpod/runtime.go +++ b/libpod/runtime.go @@ -112,8 +112,6 @@ type Runtime struct { // mechanism to read and write even logs eventer events.Eventer - - ctx context.Context } // OCIRuntimePath contains information about an OCI runtime. @@ -353,8 +351,8 @@ func SetXdgRuntimeDir(val string) error { // NewRuntime creates a new container runtime // Options can be passed to override the default configuration for the runtime -func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { - return newRuntimeFromConfig("", options...) +func NewRuntime(ctx context.Context, options ...RuntimeOption) (runtime *Runtime, err error) { + return newRuntimeFromConfig(ctx, "", options...) } // NewRuntimeFromConfig creates a new container runtime using the given @@ -362,14 +360,14 @@ func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { // functions can be used to mutate this configuration further. // An error will be returned if the configuration file at the given path does // not exist or cannot be loaded -func NewRuntimeFromConfig(userConfigPath string, options ...RuntimeOption) (runtime *Runtime, err error) { +func NewRuntimeFromConfig(ctx context.Context, userConfigPath string, options ...RuntimeOption) (runtime *Runtime, err error) { if userConfigPath == "" { return nil, errors.New("invalid configuration file specified") } - return newRuntimeFromConfig(userConfigPath, options...) + return newRuntimeFromConfig(ctx, userConfigPath, options...) } -func newRuntimeFromConfig(userConfigPath string, options ...RuntimeOption) (runtime *Runtime, err error) { +func newRuntimeFromConfig(ctx context.Context, userConfigPath string, options ...RuntimeOption) (runtime *Runtime, err error) { runtime = new(Runtime) runtime.config = new(RuntimeConfig) runtime.configuredFrom = new(runtimeConfiguredFrom) @@ -563,7 +561,7 @@ func newRuntimeFromConfig(userConfigPath string, options ...RuntimeOption) (runt } } } - if err := makeRuntime(runtime); err != nil { + if err := makeRuntime(ctx, runtime); err != nil { return nil, err } return runtime, nil @@ -571,7 +569,7 @@ func newRuntimeFromConfig(userConfigPath string, options ...RuntimeOption) (runt // Make a new runtime based on the given configuration // Sets up containers/storage, state store, OCI runtime -func makeRuntime(runtime *Runtime) (err error) { +func makeRuntime(ctx context.Context, runtime *Runtime) (err error) { // Backward compatibility for `runtime_path` if runtime.config.RuntimePath != nil { // Don't print twice in rootless mode. @@ -980,7 +978,7 @@ func makeRuntime(runtime *Runtime) (err error) { os.Exit(ret) } } - if err := runtime.migrate(); err != nil { + if err := runtime.migrate(ctx); err != nil { return err } } diff --git a/libpod/runtime_migrate.go b/libpod/runtime_migrate.go index a084df289..0bb8e952f 100644 --- a/libpod/runtime_migrate.go +++ b/libpod/runtime_migrate.go @@ -1,13 +1,14 @@ package libpod import ( + "context" "path/filepath" "github.com/pkg/errors" "github.com/sirupsen/logrus" ) -func (r *Runtime) migrate() error { +func (r *Runtime) migrate(ctx context.Context) error { runningContainers, err := r.GetRunningContainers() if err != nil { return err @@ -38,7 +39,7 @@ func (r *Runtime) migrate() error { } for _, ctr := range runningContainers { - if err := ctr.Start(r.ctx, true); err != nil { + if err := ctr.Start(ctx, true); err != nil { logrus.Errorf("error restarting container %s", ctr.ID()) } } |