summaryrefslogtreecommitdiff
path: root/libpod/runtime_pod_linux.go
blob: 24e9f3da71704723c2885f40019a9f8a49bda882 (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
//go:build linux
// +build linux

package libpod

import (
	"context"
	"errors"
	"fmt"
	"path"
	"path/filepath"
	"strings"

	"github.com/containers/common/pkg/cgroups"
	"github.com/containers/common/pkg/config"
	"github.com/containers/podman/v4/libpod/define"
	"github.com/containers/podman/v4/libpod/events"
	"github.com/containers/podman/v4/pkg/rootless"
	"github.com/containers/podman/v4/pkg/specgen"
	"github.com/hashicorp/go-multierror"
	"github.com/sirupsen/logrus"
)

// NewPod makes a new, empty pod
func (r *Runtime) NewPod(ctx context.Context, p specgen.PodSpecGenerator, options ...PodCreateOption) (_ *Pod, deferredErr error) {
	if !r.valid {
		return nil, define.ErrRuntimeStopped
	}

	pod := newPod(r)

	// Set default namespace to runtime's namespace
	// Do so before options run so they can override it
	if r.config.Engine.Namespace != "" {
		pod.config.Namespace = r.config.Engine.Namespace
	}

	for _, option := range options {
		if err := option(pod); err != nil {
			return nil, fmt.Errorf("running pod create option: %w", err)
		}
	}

	// Allocate a lock for the pod
	lock, err := r.lockManager.AllocateLock()
	if err != nil {
		return nil, fmt.Errorf("allocating lock for new pod: %w", err)
	}
	pod.lock = lock
	pod.config.LockID = pod.lock.ID()

	defer func() {
		if deferredErr != nil {
			if err := pod.lock.Free(); err != nil {
				logrus.Errorf("Freeing pod lock after failed creation: %v", err)
			}
		}
	}()

	pod.valid = true

	// Check Cgroup parent sanity, and set it if it was not set
	if r.config.Cgroups() != "disabled" {
		switch r.config.Engine.CgroupManager {
		case config.CgroupfsCgroupsManager:
			canUseCgroup := !rootless.IsRootless() || isRootlessCgroupSet(pod.config.CgroupParent)
			if canUseCgroup {
				// need to actually create parent here
				if pod.config.CgroupParent == "" {
					pod.config.CgroupParent = CgroupfsDefaultCgroupParent
				} else if strings.HasSuffix(path.Base(pod.config.CgroupParent), ".slice") {
					return nil, fmt.Errorf("systemd slice received as cgroup parent when using cgroupfs: %w", define.ErrInvalidArg)
				}
				// If we are set to use pod cgroups, set the cgroup parent that
				// all containers in the pod will share
				if pod.config.UsePodCgroup {
					pod.state.CgroupPath = filepath.Join(pod.config.CgroupParent, pod.ID())
					if p.InfraContainerSpec != nil {
						p.InfraContainerSpec.CgroupParent = pod.state.CgroupPath
						// cgroupfs + rootless = permission denied when creating the cgroup.
						if !rootless.IsRootless() {
							res, err := GetLimits(p.ResourceLimits)
							if err != nil {
								return nil, err
							}
							// Need to both create and update the cgroup
							// rather than create a new path in c/common for pod cgroup creation
							// just create as if it is a ctr and then update figures out that we need to
							// populate the resource limits on the pod level
							cgc, err := cgroups.New(pod.state.CgroupPath, &res)
							if err != nil {
								return nil, err
							}
							err = cgc.Update(&res)
							if err != nil {
								return nil, err
							}
						}
					}
				}
			}
		case config.SystemdCgroupsManager:
			if pod.config.CgroupParent == "" {
				if rootless.IsRootless() {
					pod.config.CgroupParent = SystemdDefaultRootlessCgroupParent
				} else {
					pod.config.CgroupParent = SystemdDefaultCgroupParent
				}
			} else if len(pod.config.CgroupParent) < 6 || !strings.HasSuffix(path.Base(pod.config.CgroupParent), ".slice") {
				return nil, fmt.Errorf("did not receive systemd slice as cgroup parent when using systemd to manage cgroups: %w", define.ErrInvalidArg)
			}
			// If we are set to use pod cgroups, set the cgroup parent that
			// all containers in the pod will share
			if pod.config.UsePodCgroup {
				cgroupPath, err := systemdSliceFromPath(pod.config.CgroupParent, fmt.Sprintf("libpod_pod_%s", pod.ID()), p.ResourceLimits)
				if err != nil {
					return nil, fmt.Errorf("unable to create pod cgroup for pod %s: %w", pod.ID(), err)
				}
				pod.state.CgroupPath = cgroupPath
				if p.InfraContainerSpec != nil {
					p.InfraContainerSpec.CgroupParent = pod.state.CgroupPath
				}
			}
		default:
			return nil, fmt.Errorf("unsupported Cgroup manager: %s - cannot validate cgroup parent: %w", r.config.Engine.CgroupManager, define.ErrInvalidArg)
		}
	}

	if pod.config.UsePodCgroup {
		logrus.Debugf("Got pod cgroup as %s", pod.state.CgroupPath)
	}

	if !pod.HasInfraContainer() && pod.SharesNamespaces() {
		return nil, errors.New("Pods must have an infra container to share namespaces")
	}
	if pod.HasInfraContainer() && !pod.SharesNamespaces() {
		logrus.Infof("Pod has an infra container, but shares no namespaces")
	}

	// Unless the user has specified a name, use a randomly generated one.
	// Note that name conflicts may occur (see #11735), so we need to loop.
	generateName := pod.config.Name == ""
	var addPodErr error
	for {
		if generateName {
			name, err := r.generateName()
			if err != nil {
				return nil, err
			}
			pod.config.Name = name
		}

		if p.InfraContainerSpec != nil && p.InfraContainerSpec.Hostname == "" {
			p.InfraContainerSpec.Hostname = pod.config.Name
		}
		if addPodErr = r.state.AddPod(pod); addPodErr == nil {
			return pod, nil
		}
		if !generateName || (!errors.Is(addPodErr, define.ErrPodExists) && !errors.Is(addPodErr, define.ErrCtrExists)) {
			break
		}
	}
	if addPodErr != nil {
		return nil, fmt.Errorf("adding pod to state: %w", addPodErr)
	}

	return pod, nil
}

// AddInfra adds the created infra container to the pod state
func (r *Runtime) AddInfra(ctx context.Context, pod *Pod, infraCtr *Container) (*Pod, error) {
	if !r.valid {
		return nil, define.ErrRuntimeStopped
	}
	pod.state.InfraContainerID = infraCtr.ID()
	if err := pod.save(); err != nil {
		return nil, err
	}
	pod.newPodEvent(events.Create)
	return pod, nil
}

// SavePod is a helper function to save the pod state from outside of libpod
func (r *Runtime) SavePod(pod *Pod) error {
	if !r.valid {
		return define.ErrRuntimeStopped
	}
	if err := pod.save(); err != nil {
		return err
	}
	pod.newPodEvent(events.Create)
	return nil
}

// DO NOT USE THIS FUNCTION DIRECTLY. Use removePod(), below. It will call
// removeMalformedPod() if necessary.
func (r *Runtime) removeMalformedPod(ctx context.Context, p *Pod, ctrs []*Container, force bool, timeout *uint, ctrNamedVolumes map[string]*ContainerNamedVolume) error {
	var removalErr error
	for _, ctr := range ctrs {
		err := func() error {
			ctrLock := ctr.lock
			ctrLock.Lock()
			defer func() {
				ctrLock.Unlock()
			}()

			if err := ctr.syncContainer(); err != nil {
				return err
			}

			for _, vol := range ctr.config.NamedVolumes {
				ctrNamedVolumes[vol.Name] = vol
			}

			return r.removeContainer(ctx, ctr, force, false, true, true, timeout)
		}()

		if removalErr == nil {
			removalErr = err
		} else {
			logrus.Errorf("Removing container %s from pod %s: %v", ctr.ID(), p.ID(), err)
		}
	}
	if removalErr != nil {
		return removalErr
	}

	// 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.
		// The containers in the pod are unusable, but they still exist,
		// so pod removal will fail.
		return err
	}

	return nil
}

func (r *Runtime) removePod(ctx context.Context, p *Pod, removeCtrs, force bool, timeout *uint) error {
	if err := p.updatePod(); err != nil {
		return err
	}

	ctrs, err := r.state.PodContainers(p)
	if err != nil {
		return err
	}
	numCtrs := len(ctrs)

	// If the only running container in the pod is the pause container, remove the pod and container unconditionally.
	pauseCtrID := p.state.InfraContainerID
	if numCtrs == 1 && ctrs[0].ID() == pauseCtrID {
		removeCtrs = true
		force = true
	}
	if !removeCtrs && numCtrs > 0 {
		return fmt.Errorf("pod %s contains containers and cannot be removed: %w", p.ID(), define.ErrCtrExists)
	}

	var removalErr error
	ctrNamedVolumes := make(map[string]*ContainerNamedVolume)

	// Build a graph of all containers in the pod.
	graph, err := BuildContainerGraph(ctrs)
	if err != nil {
		// We have to allow the pod to be removed.
		// But let's only do it if force is set.
		if !force {
			return fmt.Errorf("cannot create container graph for pod %s: %w", p.ID(), err)
		}

		removalErr = fmt.Errorf("creating container graph for pod %s failed, fell back to loop removal: %w", p.ID(), err)

		if err := r.removeMalformedPod(ctx, p, ctrs, force, timeout, ctrNamedVolumes); err != nil {
			logrus.Errorf("Error creating container graph for pod %s: %v. Falling back to loop removal.", p.ID(), err)
			return err
		}
	} else {
		ctrErrors := make(map[string]error)
		ctrsVisited := make(map[string]bool)

		for _, node := range graph.notDependedOnNodes {
			removeNode(ctx, node, p, force, timeout, false, ctrErrors, ctrsVisited, ctrNamedVolumes)
		}

		// This is gross, but I don't want to change the signature on
		// removePod - especially since any change here eventually has
		// to map down to one error unless we want to make a breaking
		// API change.
		if len(ctrErrors) > 0 {
			var allErrs error
			for id, err := range ctrErrors {
				allErrs = multierror.Append(allErrs, fmt.Errorf("removing container %s from pod %s: %w", id, p.ID(), err))
			}
			return allErrs
		}
	}

	for volName := range ctrNamedVolumes {
		volume, err := r.state.Volume(volName)
		if err != nil && !errors.Is(err, define.ErrNoSuchVolume) {
			logrus.Errorf("Retrieving volume %s: %v", volName, err)
			continue
		}
		if !volume.Anonymous() {
			continue
		}
		if err := r.removeVolume(ctx, volume, false, timeout, false); err != nil {
			if errors.Is(err, define.ErrNoSuchVolume) || errors.Is(err, define.ErrVolumeRemoved) {
				continue
			}
			logrus.Errorf("Removing volume %s: %v", volName, err)
		}
	}

	// Remove pod cgroup, if present
	if p.state.CgroupPath != "" {
		logrus.Debugf("Removing pod cgroup %s", p.state.CgroupPath)

		switch p.runtime.config.Engine.CgroupManager {
		case config.SystemdCgroupsManager:
			if err := deleteSystemdCgroup(p.state.CgroupPath, p.ResourceLim()); err != nil {
				if removalErr == nil {
					removalErr = fmt.Errorf("removing pod %s cgroup: %w", p.ID(), err)
				} else {
					logrus.Errorf("Deleting pod %s cgroup %s: %v", p.ID(), p.state.CgroupPath, err)
				}
			}
		case config.CgroupfsCgroupsManager:
			// Delete the cgroupfs cgroup
			// Make sure the conmon cgroup is deleted first
			// Since the pod is almost gone, don't bother failing
			// hard - instead, just log errors.
			conmonCgroupPath := filepath.Join(p.state.CgroupPath, "conmon")
			conmonCgroup, err := cgroups.Load(conmonCgroupPath)
			if err != nil && err != cgroups.ErrCgroupDeleted && err != cgroups.ErrCgroupV1Rootless {
				if removalErr == nil {
					removalErr = fmt.Errorf("retrieving pod %s conmon cgroup: %w", p.ID(), err)
				} else {
					logrus.Debugf("Error retrieving pod %s conmon cgroup %s: %v", p.ID(), conmonCgroupPath, err)
				}
			}
			if err == nil {
				if err = conmonCgroup.Delete(); err != nil {
					if removalErr == nil {
						removalErr = fmt.Errorf("removing pod %s conmon cgroup: %w", p.ID(), err)
					} else {
						logrus.Errorf("Deleting pod %s conmon cgroup %s: %v", p.ID(), conmonCgroupPath, err)
					}
				}
			}
			cgroup, err := cgroups.Load(p.state.CgroupPath)
			if err != nil && err != cgroups.ErrCgroupDeleted && err != cgroups.ErrCgroupV1Rootless {
				if removalErr == nil {
					removalErr = fmt.Errorf("retrieving pod %s cgroup: %w", p.ID(), err)
				} else {
					logrus.Errorf("Retrieving pod %s cgroup %s: %v", p.ID(), p.state.CgroupPath, err)
				}
			}
			if err == nil {
				if err := cgroup.Delete(); err != nil {
					if removalErr == nil {
						removalErr = fmt.Errorf("removing pod %s cgroup: %w", p.ID(), err)
					} else {
						logrus.Errorf("Deleting pod %s cgroup %s: %v", p.ID(), p.state.CgroupPath, err)
					}
				}
			}
		default:
			// This should be caught much earlier, but let's still
			// keep going so we make sure to evict the pod before
			// ending up with an inconsistent state.
			if removalErr == nil {
				removalErr = fmt.Errorf("unrecognized cgroup manager %s when removing pod %s cgroups: %w", p.runtime.config.Engine.CgroupManager, p.ID(), define.ErrInternal)
			} else {
				logrus.Errorf("Unknown cgroups manager %s specified - cannot remove pod %s cgroup", p.runtime.config.Engine.CgroupManager, p.ID())
			}
		}
	}

	if err := p.maybeRemoveServiceContainer(); err != nil {
		return err
	}

	// Remove pod from state
	if err := r.state.RemovePod(p); err != nil {
		if removalErr != nil {
			logrus.Errorf("%v", removalErr)
		}
		return err
	}

	// Mark pod invalid
	p.valid = false
	p.newPodEvent(events.Remove)

	// Deallocate the pod lock
	if err := p.lock.Free(); err != nil {
		if removalErr == nil {
			removalErr = fmt.Errorf("freeing pod %s lock: %w", p.ID(), err)
		} else {
			logrus.Errorf("Freeing pod %s lock: %v", p.ID(), err)
		}
	}

	return removalErr
}