summaryrefslogtreecommitdiff
path: root/libpod/oci.go
blob: f25c6bdb9db551accfd1277c514fa5f2cd49df70 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
package libpod

import (
	"bytes"
	"encoding/json"
	"fmt"
	"io/ioutil"
	"os"
	"os/exec"
	"path/filepath"
	"strconv"
	"syscall"
	"time"

	"github.com/containerd/cgroups"
	"github.com/coreos/go-systemd/activation"
	spec "github.com/opencontainers/runtime-spec/specs-go"
	"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
	// Trying to keep libpod from depending on CRI-O code
	"github.com/projectatomic/libpod/utils"
)

// OCI code is undergoing heavy rewrite

const (
	// CgroupfsCgroupsManager represents cgroupfs native cgroup manager
	CgroupfsCgroupsManager = "cgroupfs"
	// SystemdCgroupsManager represents systemd native cgroup manager
	SystemdCgroupsManager = "systemd"

	// ContainerCreateTimeout represents the value of container creating timeout
	ContainerCreateTimeout = 240 * time.Second

	// Timeout before declaring that runc has failed to kill a given container
	killContainerTimeout = 5 * time.Second
	// DefaultShmSize is the default shm size
	DefaultShmSize = 64 * 1024 * 1024
	// NsRunDir is the default directory in which running network namespaces
	// are stored
	NsRunDir = "/var/run/netns"
)

// OCIRuntime represents an OCI-compatible runtime that libpod can call into
// to perform container operations
type OCIRuntime struct {
	name          string
	path          string
	conmonPath    string
	conmonEnv     []string
	cgroupManager string
	tmpDir        string
	exitsDir      string
	socketsDir    string
	logSizeMax    int64
	noPivot       bool
}

// syncInfo is used to return data from monitor process to daemon
type syncInfo struct {
	Pid     int    `json:"pid"`
	Message string `json:"message,omitempty"`
}

// Make a new OCI runtime with provided options
func newOCIRuntime(name string, path string, conmonPath string, conmonEnv []string, cgroupManager string, tmpDir string, logSizeMax int64, noPivotRoot bool) (*OCIRuntime, error) {
	runtime := new(OCIRuntime)
	runtime.name = name
	runtime.path = path
	runtime.conmonPath = conmonPath
	runtime.conmonEnv = conmonEnv
	runtime.cgroupManager = cgroupManager
	runtime.tmpDir = tmpDir
	runtime.logSizeMax = logSizeMax
	runtime.noPivot = noPivotRoot

	runtime.exitsDir = filepath.Join(runtime.tmpDir, "exits")
	runtime.socketsDir = filepath.Join(runtime.tmpDir, "socket")

	if cgroupManager != CgroupfsCgroupsManager && cgroupManager != SystemdCgroupsManager {
		return nil, errors.Wrapf(ErrInvalidArg, "invalid cgroup manager specified: %s", cgroupManager)
	}

	// Create the exit files and attach sockets directories
	if err := os.MkdirAll(runtime.exitsDir, 0750); err != nil {
		// The directory is allowed to exist
		if !os.IsExist(err) {
			return nil, errors.Wrapf(err, "error creating OCI runtime exit files directory %s",
				runtime.exitsDir)
		}
	}
	if err := os.MkdirAll(runtime.socketsDir, 0750); err != nil {
		// The directory is allowed to exist
		if !os.IsExist(err) {
			return nil, errors.Wrapf(err, "error creating OCI runtime attach sockets directory %s",
				runtime.socketsDir)
		}
	}

	return runtime, nil
}

// newPipe creates a unix socket pair for communication
func newPipe() (parent *os.File, child *os.File, err error) {
	fds, err := unix.Socketpair(unix.AF_LOCAL, unix.SOCK_STREAM|unix.SOCK_CLOEXEC, 0)
	if err != nil {
		return nil, nil, err
	}
	return os.NewFile(uintptr(fds[1]), "parent"), os.NewFile(uintptr(fds[0]), "child"), nil
}

// Create systemd unit name for cgroup scopes
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)
		return errors.Errorf("container %s did not die within timeout", ctr.ID())
	}
}

// CreateContainer creates a container in the OCI runtime
// TODO terminal support for container
// Presently just ignoring conmon opts related to it
func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string) (err error) {
	var stderrBuf bytes.Buffer

	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)
	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")
	}
	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))
	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...)
	}

	err = cmd.Start()
	if err != nil {
		childPipe.Close()
		return err
	}

	// We don't need childPipe on the parent side
	childPipe.Close()
	childStartPipe.Close()

	// Move conmon to specified cgroup
	if r.cgroupManager == SystemdCgroupsManager {
		logrus.Infof("Running conmon under slice %s and unitName %s", cgroupParent, createUnitName("libpod-conmon", ctr.ID()))
		if err = utils.RunUnderSystemdScope(cmd.Process.Pid, cgroupParent, createUnitName("libpod-conmon", ctr.ID())); err != nil {
			logrus.Warnf("Failed to add conmon to systemd sandbox cgroup: %v", err)
		}
	} else {
		control, err := cgroups.New(cgroups.V1, cgroups.StaticPath(filepath.Join(cgroupParent, "/libpod-conmon-"+ctr.ID())), &spec.LinuxResources{})
		if err != nil {
			logrus.Warnf("Failed to add conmon to cgroupfs sandbox cgroup: %v", err)
		} else {
			// we need to remove this defer and delete the cgroup once conmon exits
			// maybe need a conmon monitor?
			if err := control.Add(cgroups.Process{Pid: cmd.Process.Pid}); err != nil {
				logrus.Warnf("Failed to add conmon to cgroupfs sandbox cgroup: %v", 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 runc 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
		if err = json.NewDecoder(parentPipe).Decode(&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")
		}
	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.
func (r *OCIRuntime) updateContainerStatus(ctr *Container) error {
	state := new(spec.State)

	// Store old state so we know if we were already stopped
	oldState := ctr.state.State

	out, err := exec.Command(r.path, "state", ctr.ID()).Output()
	if err != nil {
		return errors.Wrapf(err, "error getting container %s state. stderr/out: %s", ctr.ID(), out)
	}

	if err := json.NewDecoder(bytes.NewBuffer(out)).Decode(state); err != nil {
		return errors.Wrapf(err, "error decoding container status for container %s", ctr.ID())
	}

	ctr.state.PID = state.Pid

	switch state.Status {
	case "created":
		ctr.state.State = ContainerStateCreated
	case "paused":
		ctr.state.State = ContainerStatePaused
	case "running":
		ctr.state.State = ContainerStateRunning
	case "stopped":
		ctr.state.State = ContainerStateStopped
	default:
		return errors.Wrapf(ErrInternal, "unrecognized status returned by runc for container %s: %s",
			ctr.ID(), state.Status)
	}

	// Only grab exit status if we were not already stopped
	// If we were, it should already be in the database
	if ctr.state.State == ContainerStateStopped && oldState != ContainerStateStopped {
		exitFile := filepath.Join(r.exitsDir, ctr.ID())
		var fi os.FileInfo
		err = kwait.ExponentialBackoff(
			kwait.Backoff{
				Duration: 500 * time.Millisecond,
				Factor:   1.2,
				Steps:    6,
			},
			func() (bool, error) {
				var err error
				fi, err = os.Stat(exitFile)
				if err != nil {
					// wait longer
					return false, nil
				}
				return true, nil
			})
		if err != nil {
			ctr.state.ExitCode = -1
			ctr.state.FinishedTime = time.Now()
			return errors.Wrapf(err, "no exit file for container %s found", ctr.ID())
		}

		ctr.state.FinishedTime = getFinishedTime(fi)
		statusCodeStr, err := ioutil.ReadFile(exitFile)
		if err != nil {
			return errors.Wrapf(err, "failed to read exit file for container %s", ctr.ID())
		}
		statusCode, err := strconv.Atoi(string(statusCodeStr))
		if err != nil {
			return errors.Wrapf(err, "error convertaing exit status code for container %s to int",
				ctr.ID())
		}
		ctr.state.ExitCode = int32(statusCode)

		oomFilePath := filepath.Join(ctr.bundlePath(), "oom")
		if _, err = os.Stat(oomFilePath); err == nil {
			ctr.state.OOMKilled = true
		}

	}

	return nil
}

// startContainer starts the given container
// Sets time the container was started, but does not save it.
func (r *OCIRuntime) startContainer(ctr *Container) error {
	// TODO: streams should probably *not* be our STDIN/OUT/ERR - redirect to buffers?
	if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, r.path, "start", ctr.ID()); err != nil {
		return err
	}

	ctr.state.StartedTime = time.Now()

	return nil
}

// killContainer sends the given signal to the given container
func (r *OCIRuntime) killContainer(ctr *Container, signal uint) error {
	logrus.Debugf("Sending signal %d to container %s", signal, ctr.ID())
	if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, r.path, "kill", ctr.ID(), fmt.Sprintf("%d", signal)); err != nil {
		return errors.Wrapf(err, "error sending signal to container %s", ctr.ID())
	}

	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 {
	// 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 {
			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
		}
	}

	if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, r.path, "kill", "--all", ctr.ID(), "KILL"); err != nil {
		return errors.Wrapf(err, "error sending SIGKILL to container %s", ctr.ID())
	}

	// Give runc a few seconds to make it happen
	if err := waitContainerStop(ctr, killContainerTimeout); err != nil {
		return err
	}

	return nil
}

// deleteContainer deletes a container from runc
func (r *OCIRuntime) deleteContainer(ctr *Container) error {
	_, err := utils.ExecCmd(r.path, "delete", "--force", ctr.ID())
	return err
}

// pauseContainer pauses the given container
func (r *OCIRuntime) pauseContainer(ctr *Container) error {
	return utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, r.path, "pause", ctr.ID())
}

// unpauseContainer unpauses the given container
func (r *OCIRuntime) unpauseContainer(ctr *Container) error {
	return utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, r.path, "resume", ctr.ID())
}

// execContainer executes a command in a running container
// TODO: Add --detach support
// TODO: Convert to use conmon
// TODO: add --pid-file and use that to generate exec session tracking
func (r *OCIRuntime) execContainer(c *Container, cmd, capAdd, env []string, tty bool, user, sessionID string) (*exec.Cmd, error) {
	if len(cmd) == 0 {
		return nil, errors.Wrapf(ErrInvalidArg, "must provide a command to execute")
	}

	if sessionID == "" {
		return nil, errors.Wrapf(ErrEmptyID, "must provide a session ID for exec")
	}

	args := []string{}

	// TODO - should we maintain separate logpaths for exec sessions?
	args = append(args, "--log", c.LogPath())

	args = append(args, "exec")

	args = append(args, "--cwd", c.config.Spec.Process.Cwd)

	args = append(args, "--pid-file", c.execPidPath(sessionID))

	if tty {
		args = append(args, "--tty")
	}

	if user != "" {
		args = append(args, "--user", user)
	}

	if c.config.Spec.Process.NoNewPrivileges {
		args = append(args, "--no-new-privs")
	}

	for _, cap := range capAdd {
		args = append(args, "--cap", cap)
	}

	for _, envVar := range env {
		args = append(args, "--env", envVar)
	}

	// Append container ID and command
	args = append(args, c.ID())
	args = append(args, cmd...)

	logrus.Debugf("Starting runtime %s with following arguments: %v", r.path, args)

	execCmd := exec.Command(r.path, args...)
	execCmd.Stdout = os.Stdout
	execCmd.Stderr = os.Stderr
	execCmd.Stdin = os.Stdin

	return execCmd, nil
}