summaryrefslogtreecommitdiff
path: root/pkg/rootless/rootless_linux.go
blob: c753228f1402bce36546018cc5c4602532201ecc (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
// +build linux

package rootless

import (
	"fmt"
	"io/ioutil"
	"os"
	"os/exec"
	gosignal "os/signal"
	"os/user"
	"runtime"
	"strconv"
	"sync"
	"syscall"
	"unsafe"

	"github.com/containers/storage/pkg/idtools"
	"github.com/docker/docker/pkg/signal"
	"github.com/pkg/errors"
	"github.com/sirupsen/logrus"
)

/*
extern int reexec_in_user_namespace(int ready);
extern int reexec_in_user_namespace_wait(int pid);
extern int reexec_userns_join(int userns, int mountns);
*/
import "C"

func runInUser() error {
	os.Setenv("_CONTAINERS_USERNS_CONFIGURED", "done")
	return nil
}

var (
	isRootlessOnce sync.Once
	isRootless     bool
)

// IsRootless tells us if we are running in rootless mode
func IsRootless() bool {
	isRootlessOnce.Do(func() {
		isRootless = os.Geteuid() != 0 || os.Getenv("_CONTAINERS_USERNS_CONFIGURED") != ""
	})
	return isRootless
}

// Argument returns the argument that was set for the rootless session.
func Argument() string {
	return os.Getenv("_CONTAINERS_ROOTLESS_ARG")
}

// GetRootlessUID returns the UID of the user in the parent userNS
func GetRootlessUID() int {
	uidEnv := os.Getenv("_CONTAINERS_ROOTLESS_UID")
	if uidEnv != "" {
		u, _ := strconv.Atoi(uidEnv)
		return u
	}
	return os.Geteuid()
}

func tryMappingTool(tool string, pid int, hostID int, mappings []idtools.IDMap) error {
	path, err := exec.LookPath(tool)
	if err != nil {
		return errors.Wrapf(err, "cannot find %s", tool)
	}

	appendTriplet := func(l []string, a, b, c int) []string {
		return append(l, fmt.Sprintf("%d", a), fmt.Sprintf("%d", b), fmt.Sprintf("%d", c))
	}

	args := []string{path, fmt.Sprintf("%d", pid)}
	args = appendTriplet(args, 0, hostID, 1)
	if mappings != nil {
		for _, i := range mappings {
			args = appendTriplet(args, i.ContainerID+1, i.HostID, i.Size)
		}
	}
	cmd := exec.Cmd{
		Path: path,
		Args: args,
	}

	if output, err := cmd.CombinedOutput(); err != nil {
		logrus.Debugf("error from %s: %s", tool, output)
		return errors.Wrapf(err, "cannot setup namespace using %s", tool)
	}
	return nil
}

// JoinNS re-exec podman in a new userNS and join the user namespace of the specified
// PID.
func JoinNS(pid uint, preserveFDs int) (bool, int, error) {
	if os.Geteuid() == 0 || os.Getenv("_CONTAINERS_USERNS_CONFIGURED") != "" {
		return false, -1, nil
	}

	userNS, err := getUserNSForPid(pid)
	if err != nil {
		return false, -1, err
	}
	defer userNS.Close()

	pidC := C.reexec_userns_join(C.int(userNS.Fd()), -1)
	if int(pidC) < 0 {
		return false, -1, errors.Errorf("cannot re-exec process")
	}
	if preserveFDs > 0 {
		for fd := 3; fd < 3+preserveFDs; fd++ {
			// These fds were passed down to the runtime.  Close them
			// and not interfere
			os.NewFile(uintptr(fd), fmt.Sprintf("fd-%d", fd)).Close()
		}
	}

	ret := C.reexec_in_user_namespace_wait(pidC)
	if ret < 0 {
		return false, -1, errors.New("error waiting for the re-exec process")
	}

	return true, int(ret), nil
}

// JoinDirectUserAndMountNS re-exec podman in a new userNS and join the user and mount
// namespace of the specified PID without looking up its parent.  Useful to join directly
// the conmon process.  It is a convenience function for JoinDirectUserAndMountNSWithOpts
// with a default configuration.
func JoinDirectUserAndMountNS(pid uint) (bool, int, error) {
	return JoinDirectUserAndMountNSWithOpts(pid, nil)
}

// JoinDirectUserAndMountNSWithOpts re-exec podman in a new userNS and join the user and
// mount namespace of the specified PID without looking up its parent.  Useful to join
// directly the conmon process.
func JoinDirectUserAndMountNSWithOpts(pid uint, opts *Opts) (bool, int, error) {
	if os.Geteuid() == 0 || os.Getenv("_CONTAINERS_USERNS_CONFIGURED") != "" {
		return false, -1, nil
	}

	userNS, err := os.Open(fmt.Sprintf("/proc/%d/ns/user", pid))
	if err != nil {
		return false, -1, err
	}
	defer userNS.Close()

	mountNS, err := os.Open(fmt.Sprintf("/proc/%d/ns/mnt", pid))
	if err != nil {
		return false, -1, err
	}
	defer userNS.Close()

	if opts != nil && opts.Argument != "" {
		if err := os.Setenv("_CONTAINERS_ROOTLESS_ARG", opts.Argument); err != nil {
			return false, -1, err
		}
	}

	pidC := C.reexec_userns_join(C.int(userNS.Fd()), C.int(mountNS.Fd()))
	if int(pidC) < 0 {
		return false, -1, errors.Errorf("cannot re-exec process")
	}

	ret := C.reexec_in_user_namespace_wait(pidC)
	if ret < 0 {
		return false, -1, errors.New("error waiting for the re-exec process")
	}

	return true, int(ret), nil
}

// JoinNSPath re-exec podman in a new userNS and join the owner user namespace of the
// specified path.
func JoinNSPath(path string) (bool, int, error) {
	if os.Geteuid() == 0 || os.Getenv("_CONTAINERS_USERNS_CONFIGURED") != "" {
		return false, -1, nil
	}

	userNS, err := getUserNSForPath(path)
	if err != nil {
		return false, -1, err
	}
	defer userNS.Close()

	pidC := C.reexec_userns_join(C.int(userNS.Fd()), -1)
	if int(pidC) < 0 {
		return false, -1, errors.Errorf("cannot re-exec process")
	}

	ret := C.reexec_in_user_namespace_wait(pidC)
	if ret < 0 {
		return false, -1, errors.New("error waiting for the re-exec process")
	}

	return true, int(ret), nil
}

// BecomeRootInUserNS re-exec podman in a new userNS.  It returns whether podman was re-executed
// into a new user namespace and the return code from the re-executed podman process.
// If podman was re-executed the caller needs to propagate the error code returned by the child
// process.  It is a convenience function for BecomeRootInUserNSWithOpts with a default configuration.
func BecomeRootInUserNS() (bool, int, error) {
	return BecomeRootInUserNSWithOpts(nil)
}

// BecomeRootInUserNSWithOpts re-exec podman in a new userNS.  It returns whether podman was
// re-execute into a new user namespace and the return code from the re-executed podman process.
// If podman was re-executed the caller needs to propagate the error code returned by the child
// process.
func BecomeRootInUserNSWithOpts(opts *Opts) (bool, int, error) {
	if os.Geteuid() == 0 || os.Getenv("_CONTAINERS_USERNS_CONFIGURED") != "" {
		if os.Getenv("_CONTAINERS_USERNS_CONFIGURED") == "init" {
			return false, 0, runInUser()
		}
		return false, 0, nil
	}

	runtime.LockOSThread()
	defer runtime.UnlockOSThread()

	r, w, err := os.Pipe()
	if err != nil {
		return false, -1, err
	}
	defer r.Close()
	defer w.Close()
	defer w.Write([]byte("0"))

	if opts != nil && opts.Argument != "" {
		if err := os.Setenv("_CONTAINERS_ROOTLESS_ARG", opts.Argument); err != nil {
			return false, -1, err
		}
	}

	pidC := C.reexec_in_user_namespace(C.int(r.Fd()))
	pid := int(pidC)
	if pid < 0 {
		return false, -1, errors.Errorf("cannot re-exec process")
	}

	var uids, gids []idtools.IDMap
	username := os.Getenv("USER")
	if username == "" {
		user, err := user.LookupId(fmt.Sprintf("%d", os.Getuid()))
		if err == nil {
			username = user.Username
		}
	}
	mappings, err := idtools.NewIDMappings(username, username)
	if err != nil {
		logrus.Warnf("cannot find mappings for user %s: %v", username, err)
	} else {
		uids = mappings.UIDs()
		gids = mappings.GIDs()
	}

	uidsMapped := false
	if mappings != nil && uids != nil {
		err := tryMappingTool("newuidmap", pid, os.Getuid(), uids)
		uidsMapped = err == nil
	}
	if !uidsMapped {
		logrus.Warnf("using rootless single mapping into the namespace. This might break some images. Check /etc/subuid and /etc/subgid for adding subids")
		setgroups := fmt.Sprintf("/proc/%d/setgroups", pid)
		err = ioutil.WriteFile(setgroups, []byte("deny\n"), 0666)
		if err != nil {
			return false, -1, errors.Wrapf(err, "cannot write setgroups file")
		}

		uidMap := fmt.Sprintf("/proc/%d/uid_map", pid)
		err = ioutil.WriteFile(uidMap, []byte(fmt.Sprintf("%d %d 1\n", 0, os.Getuid())), 0666)
		if err != nil {
			return false, -1, errors.Wrapf(err, "cannot write uid_map")
		}
	}

	gidsMapped := false
	if mappings != nil && gids != nil {
		err := tryMappingTool("newgidmap", pid, os.Getgid(), gids)
		gidsMapped = err == nil
	}
	if !gidsMapped {
		gidMap := fmt.Sprintf("/proc/%d/gid_map", pid)
		err = ioutil.WriteFile(gidMap, []byte(fmt.Sprintf("%d %d 1\n", 0, os.Getgid())), 0666)
		if err != nil {
			return false, -1, errors.Wrapf(err, "cannot write gid_map")
		}
	}

	_, err = w.Write([]byte("1"))
	if err != nil {
		return false, -1, errors.Wrapf(err, "write to sync pipe")
	}

	c := make(chan os.Signal, 1)

	gosignal.Notify(c)
	defer gosignal.Reset()
	go func() {
		for s := range c {
			if s == signal.SIGCHLD || s == signal.SIGPIPE {
				continue
			}

			syscall.Kill(int(pidC), s.(syscall.Signal))
		}
	}()

	ret := C.reexec_in_user_namespace_wait(pidC)
	if ret < 0 {
		return false, -1, errors.New("error waiting for the re-exec process")
	}

	return true, int(ret), nil
}

func readUserNs(path string) (string, error) {
	b := make([]byte, 256)
	_, err := syscall.Readlink(path, b)
	if err != nil {
		return "", err
	}
	return string(b), nil
}

func readUserNsFd(fd uintptr) (string, error) {
	return readUserNs(fmt.Sprintf("/proc/self/fd/%d", fd))
}

func getOwner(fd uintptr) (uintptr, error) {
	const nsGetUserns = 0xb701
	ret, _, errno := syscall.Syscall(syscall.SYS_IOCTL, fd, uintptr(nsGetUserns), 0)
	if errno != 0 {
		return 0, errno
	}
	return (uintptr)(unsafe.Pointer(ret)), nil
}

func getParentUserNs(fd uintptr) (uintptr, error) {
	const nsGetParent = 0xb702
	ret, _, errno := syscall.Syscall(syscall.SYS_IOCTL, fd, uintptr(nsGetParent), 0)
	if errno != 0 {
		return 0, errno
	}
	return (uintptr)(unsafe.Pointer(ret)), nil
}

func getUserNSForPath(path string) (*os.File, error) {
	u, err := os.Open(path)
	if err != nil {
		return nil, errors.Wrapf(err, "cannot open %s", path)
	}
	defer u.Close()
	fd, err := getOwner(u.Fd())
	if err != nil {
		return nil, err
	}

	return getUserNSFirstChild(fd)
}

func getUserNSForPid(pid uint) (*os.File, error) {
	path := fmt.Sprintf("/proc/%d/ns/user", pid)
	u, err := os.Open(path)
	if err != nil {
		return nil, errors.Wrapf(err, "cannot open %s", path)
	}

	return getUserNSFirstChild(u.Fd())
}

// getUserNSFirstChild returns an open FD for the first direct child user namespace that created the process
// Each container creates a new user namespace where the runtime runs.  The current process in the container
// might have created new user namespaces that are child of the initial namespace we created.
// This function finds the initial namespace created for the container that is a child of the current namespace.
//
//                                     current ns
//                                       /     \
//                           TARGET ->  a   [other containers]
//                                     /
//                                    b
//                                   /
//        NS READ USING THE PID ->  c
func getUserNSFirstChild(fd uintptr) (*os.File, error) {
	currentNS, err := readUserNs("/proc/self/ns/user")
	if err != nil {
		return nil, err
	}

	ns, err := readUserNsFd(fd)
	if err != nil {
		return nil, errors.Wrapf(err, "cannot read user namespace")
	}
	if ns == currentNS {
		return nil, errors.New("process running in the same user namespace")
	}

	for {
		nextFd, err := getParentUserNs(fd)
		if err != nil {
			return nil, errors.Wrapf(err, "cannot get parent user namespace")
		}

		ns, err = readUserNsFd(nextFd)
		if err != nil {
			return nil, errors.Wrapf(err, "cannot read user namespace")
		}

		if ns == currentNS {
			syscall.Close(int(nextFd))

			// Drop O_CLOEXEC for the fd.
			_, _, errno := syscall.Syscall(syscall.SYS_FCNTL, fd, syscall.F_SETFD, 0)
			if errno != 0 {
				syscall.Close(int(fd))
				return nil, errno
			}

			return os.NewFile(fd, "userns child"), nil
		}
		syscall.Close(int(fd))
		fd = nextFd
	}
}