Debian Codegraph @

src/nspawn/nspawn-mount.c

raw
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <linux/magic.h>
4 #include <sched.h>
5 #include <sys/mount.h>
6 #include <unistd.h>
7
8 #include "alloc-util.h"
9 #include "chase.h"
10 #include "errno-util.h"
11 #include "escape.h"
12 #include "extract-word.h"
13 #include "fd-util.h"
14 #include "format-util.h"
15 #include "fs-util.h"
16 #include "log.h"
17 #include "mkdir.h"
18 #include "mount-util.h"
19 #include "mountpoint-util.h"
20 #include "namespace-util.h"
21 #include "nspawn-mount.h"
22 #include "path-util.h"
23 #include "rm-rf.h"
24 #include "sort-util.h"
25 #include "stat-util.h"
26 #include "string-util.h"
27 #include "strv.h"
28 #include "tmpfile-util.h"
29 #include "user-util.h"
30
31 CustomMount* custom_mount_add(CustomMount **l, size_t *n, CustomMountType t) {
32 CustomMount *ret;
33
34 assert(l);
35 assert(n);
36 assert(t >= 0);
37 assert(t < _CUSTOM_MOUNT_TYPE_MAX);
38
39 if (!GREEDY_REALLOC(*l, *n + 1))
40 return NULL;
41
42 ret = *l + *n;
43 (*n)++;
44
45 *ret = (CustomMount) {
46 .type = t,
47 .destination_uid = UID_INVALID,
48 };
49
50 return ret;
51 }
52
53 void custom_mount_free_all(CustomMount *l, size_t n) {
54 FOREACH_ARRAY(m, l, n) {
55 free(m->source);
56 free(m->destination);
57 free(m->options);
58
59 if (m->work_dir) {
60 (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
61 free(m->work_dir);
62 }
63
64 if (m->rm_rf_tmpdir) {
65 (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL);
66 free(m->rm_rf_tmpdir);
67 }
68
69 strv_free(m->lower);
70 free(m->type_argument);
71 }
72
73 free(l);
74 }
75
76 static int custom_mount_compare(const CustomMount *a, const CustomMount *b) {
77 int r;
78
79 r = path_compare(a->destination, b->destination);
80 if (r != 0)
81 return r;
82
83 return CMP(a->type, b->type);
84 }
85
86 static int source_path_parse(const char *p, char **ret) {
87 assert(p);
88 assert(ret);
89
90 if (isempty(p))
91 return -EINVAL;
92
93 if (*p == '+') {
94 if (!path_is_absolute(p + 1))
95 return -EINVAL;
96
97 char *s = strdup(p);
98 if (!s)
99 return -ENOMEM;
100
101 *ret = TAKE_PTR(s);
102 return 0;
103 }
104
105 return path_make_absolute_cwd(p, ret);
106 }
107
108 static int source_path_parse_nullable(const char *p, char **ret) {
109 assert(p);
110 assert(ret);
111
112 if (isempty(p)) {
113 *ret = NULL;
114 return 0;
115 }
116
117 return source_path_parse(p, ret);
118 }
119
120 static char *resolve_source_path(const char *dest, const char *source) {
121 if (!source)
122 return NULL;
123
124 if (source[0] == '+')
125 return path_join(dest, source + 1);
126
127 return strdup(source);
128 }
129
130 static int allocate_temporary_source(CustomMount *m) {
131 int r;
132
133 assert(m);
134 assert(!m->source);
135 assert(!m->rm_rf_tmpdir);
136
137 r = mkdtemp_malloc("/var/tmp/nspawn-temp-XXXXXX", &m->rm_rf_tmpdir);
138 if (r < 0)
139 return log_error_errno(r, "Failed to acquire temporary directory: %m");
140
141 m->source = path_join(m->rm_rf_tmpdir, "src");
142 if (!m->source)
143 return log_oom();
144
145 if (mkdir(m->source, 0755) < 0)
146 return log_error_errno(errno, "Failed to create %s: %m", m->source);
147
148 return 0;
149 }
150
151 int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n) {
152 int r;
153
154 /* Prepare all custom mounts. This will make sure we know all temporary directories. This is called in the
155 * parent process, so that we know the temporary directories to remove on exit before we fork off the
156 * children. */
157
158 assert(l || n == 0);
159
160 /* Order the custom mounts, and make sure we have a working directory */
161 typesafe_qsort(l, n, custom_mount_compare);
162
163 FOREACH_ARRAY(m, l, n) {
164 /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount
165 * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in
166 * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in
167 * the inner child, not the outer one. Determine this here. */
168 m->in_userns = path_startswith(m->destination, "/proc");
169
170 if (m->type == CUSTOM_MOUNT_BIND) {
171 if (m->source) {
172 char *s;
173
174 s = resolve_source_path(dest, m->source);
175 if (!s)
176 return log_oom();
177
178 free_and_replace(m->source, s);
179 } else {
180 /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
181
182 r = allocate_temporary_source(m);
183 if (r < 0)
184 return r;
185 }
186 }
187
188 if (m->type == CUSTOM_MOUNT_OVERLAY) {
189 STRV_FOREACH(j, m->lower) {
190 char *s;
191
192 s = resolve_source_path(dest, *j);
193 if (!s)
194 return log_oom();
195
196 free_and_replace(*j, s);
197 }
198
199 if (m->source) {
200 char *s;
201
202 s = resolve_source_path(dest, m->source);
203 if (!s)
204 return log_oom();
205
206 free_and_replace(m->source, s);
207 } else {
208 r = allocate_temporary_source(m);
209 if (r < 0)
210 return r;
211 }
212
213 if (m->work_dir) {
214 char *s;
215
216 s = resolve_source_path(dest, m->work_dir);
217 if (!s)
218 return log_oom();
219
220 free_and_replace(m->work_dir, s);
221 } else {
222 r = tempfn_random(m->source, NULL, &m->work_dir);
223 if (r < 0)
224 return log_error_errno(r, "Failed to acquire working directory: %m");
225 }
226
227 (void) mkdir_label(m->work_dir, 0700);
228 }
229 }
230
231 return 0;
232 }
233
234 int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
235 _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL, *p = NULL;
236 CustomMount *m;
237 int r;
238
239 assert(l);
240 assert(n);
241
242 r = extract_many_words(&s, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination);
243 if (r < 0)
244 return r;
245 if (r == 0)
246 return -EINVAL;
247 if (r == 1) {
248 destination = strdup(source[0] == '+' ? source+1 : source);
249 if (!destination)
250 return -ENOMEM;
251 }
252 if (r == 2 && !isempty(s)) {
253 opts = strdup(s);
254 if (!opts)
255 return -ENOMEM;
256 }
257
258 r = source_path_parse_nullable(source, &p);
259 if (r < 0)
260 return r;
261
262 if (!path_is_absolute(destination))
263 return -EINVAL;
264
265 m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
266 if (!m)
267 return -ENOMEM;
268
269 m->source = TAKE_PTR(p);
270 m->destination = TAKE_PTR(destination);
271 m->read_only = read_only;
272 m->options = TAKE_PTR(opts);
273
274 return 0;
275 }
276
277 int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s) {
278 _cleanup_free_ char *path = NULL, *opts = NULL;
279 const char *p = ASSERT_PTR(s);
280 CustomMount *m;
281 int r;
282
283 assert(l);
284 assert(n);
285
286 r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
287 if (r < 0)
288 return r;
289 if (r == 0)
290 return -EINVAL;
291
292 if (isempty(p))
293 opts = strdup("mode=0755");
294 else
295 opts = strdup(p);
296 if (!opts)
297 return -ENOMEM;
298
299 if (!path_is_absolute(path))
300 return -EINVAL;
301
302 m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
303 if (!m)
304 return -ENOMEM;
305
306 m->destination = TAKE_PTR(path);
307 m->options = TAKE_PTR(opts);
308
309 return 0;
310 }
311
312 int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
313 _cleanup_free_ char *upper = NULL, *destination = NULL;
314 _cleanup_strv_free_ char **lower = NULL;
315 CustomMount *m;
316 int r, k;
317
318 k = strv_split_full(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
319 if (k < 0)
320 return k;
321 if (k < 2)
322 return -EADDRNOTAVAIL;
323 if (k == 2) {
324 _cleanup_free_ char *p = NULL;
325
326 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
327 * we'll also define the destination mount point the same as the upper. */
328
329 r = source_path_parse(lower[0], &p);
330 if (r < 0)
331 return r;
332
333 free_and_replace(lower[0], p);
334
335 r = source_path_parse(lower[1], &p);
336 if (r < 0)
337 return r;
338
339 free_and_replace(lower[1], p);
340
341 upper = TAKE_PTR(lower[1]);
342
343 destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
344 if (!destination)
345 return -ENOMEM;
346 } else {
347 _cleanup_free_ char *p = NULL;
348
349 /* If more than two parameters are specified, the last one is the destination, the second to last one
350 * the "upper", and all before that the "lower" directories. */
351
352 destination = lower[k - 1];
353 upper = TAKE_PTR(lower[k - 2]);
354
355 STRV_FOREACH(i, lower) {
356 r = source_path_parse(*i, &p);
357 if (r < 0)
358 return r;
359
360 free_and_replace(*i, p);
361 }
362
363 /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
364 * in /var/tmp */
365 r = source_path_parse_nullable(upper, &p);
366 if (r < 0)
367 return r;
368
369 free_and_replace(upper, p);
370
371 if (!path_is_absolute(destination))
372 return -EINVAL;
373 }
374
375 m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
376 if (!m)
377 return -ENOMEM;
378
379 m->destination = TAKE_PTR(destination);
380 m->source = TAKE_PTR(upper);
381 m->lower = TAKE_PTR(lower);
382 m->read_only = read_only;
383
384 return 0;
385 }
386
387 int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s) {
388 _cleanup_free_ char *path = NULL;
389 CustomMount *m;
390
391 assert(l);
392 assert(n);
393 assert(s);
394
395 if (!path_is_absolute(s))
396 return -EINVAL;
397
398 path = strdup(s);
399 if (!path)
400 return -ENOMEM;
401
402 m = custom_mount_add(l, n, CUSTOM_MOUNT_INACCESSIBLE);
403 if (!m)
404 return -ENOMEM;
405
406 m->destination = TAKE_PTR(path);
407 return 0;
408 }
409
410 int tmpfs_patch_options(
411 const char *options,
412 uid_t uid_shift,
413 const char *selinux_apifs_context,
414 char **ret) {
415
416 _cleanup_free_ char *buf = NULL;
417
418 assert(ret);
419
420 if (options) {
421 buf = strdup(options);
422 if (!buf)
423 return -ENOMEM;
424 }
425
426 if (uid_shift != UID_INVALID)
427 if (strextendf_with_separator(&buf, ",", "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift) < 0)
428 return -ENOMEM;
429
430 #if HAVE_SELINUX
431 if (selinux_apifs_context)
432 if (strextendf_with_separator(&buf, ",", "context=\"%s\"", selinux_apifs_context) < 0)
433 return -ENOMEM;
434 #endif
435
436 *ret = TAKE_PTR(buf);
437 return !!*ret;
438 }
439
440 int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
441 _cleanup_free_ char *top = NULL, *full = NULL;;
442 unsigned long extra_flags = 0;
443 int r;
444
445 top = path_join(dest, "/sys");
446 if (!top)
447 return log_oom();
448
449 r = path_is_mount_point(top);
450 if (r < 0)
451 return log_error_errno(r, "Failed to determine if '%s' is a mountpoint: %m", top);
452 if (r == 0) {
453 /* If this is not a mount point yet, then mount a tmpfs there */
454 r = mount_nofollow_verbose(LOG_ERR, "tmpfs", top, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0555" TMPFS_LIMITS_SYS);
455 if (r < 0)
456 return r;
457 } else {
458 r = path_is_fs_type(top, SYSFS_MAGIC);
459 if (r < 0)
460 return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
461
462 /* /sys/ might already be mounted as sysfs by the outer child in the !netns case. In this case, it's
463 * all good. Don't touch it because we don't have the right to do so, see
464 * https://github.com/systemd/systemd/issues/1555.
465 */
466 if (r > 0)
467 return 0;
468 }
469
470 full = path_join(top, "/full");
471 if (!full)
472 return log_oom();
473
474 if (mkdir(full, 0755) < 0 && errno != EEXIST)
475 return log_error_errno(errno, "Failed to create directory '%s': %m", full);
476
477 if (FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO))
478 extra_flags |= MS_RDONLY;
479
480 r = mount_nofollow_verbose(LOG_ERR, "sysfs", full, "sysfs",
481 MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
482 if (r < 0)
483 return r;
484
485 FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
486 _cleanup_free_ char *from = NULL, *to = NULL;
487
488 from = path_join(full, x);
489 if (!from)
490 return log_oom();
491
492 to = path_join(top, x);
493 if (!to)
494 return log_oom();
495
496 (void) mkdir(to, 0755);
497
498 r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
499 if (r < 0)
500 return r;
501
502 r = mount_nofollow_verbose(LOG_ERR, NULL, to, NULL,
503 MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
504 if (r < 0)
505 return r;
506 }
507
508 r = umount_verbose(LOG_ERR, full, UMOUNT_NOFOLLOW);
509 if (r < 0)
510 return r;
511
512 if (rmdir(full) < 0)
513 return log_error_errno(errno, "Failed to remove %s: %m", full);
514
515 /* Create mountpoints. Otherwise we are not allowed since we remount /sys/ read-only. */
516 FOREACH_STRING(p, "/fs/cgroup", "/fs/bpf") {
517 _cleanup_free_ char *x = path_join(top, p);
518 if (!x)
519 return log_oom();
520
521 (void) mkdir_p(x, 0755);
522 }
523
524 return mount_nofollow_verbose(LOG_ERR, NULL, top, NULL,
525 MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
526 }
527
528 #define PROC_DEFAULT_MOUNT_FLAGS (MS_NOSUID|MS_NOEXEC|MS_NODEV)
529 #define SYS_DEFAULT_MOUNT_FLAGS (MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV)
530
531 int mount_all(const char *dest,
532 MountSettingsMask mount_settings,
533 uid_t uid_shift,
534 const char *selinux_apifs_context) {
535
536 #define PROC_INACCESSIBLE_REG(path) \
537 { "/run/host/inaccessible/reg", (path), NULL, NULL, MS_BIND, \
538 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
539 { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
540 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
541
542 #define PROC_READ_ONLY(path) \
543 { (path), (path), NULL, NULL, MS_BIND, \
544 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
545 { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
546 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
547
548 typedef struct MountPoint {
549 const char *what;
550 const char *where;
551 const char *type;
552 const char *options;
553 unsigned long flags;
554 MountSettingsMask mount_settings;
555 } MountPoint;
556
557 static const MountPoint mount_table[] = {
558 /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing when we are privileged) */
559 { "proc", "/proc", "proc", NULL, PROC_DEFAULT_MOUNT_FLAGS,
560 MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_MKDIR|MOUNT_FOLLOW_SYMLINKS }, /* we follow symlinks here since not following them requires /proc/ already being mounted, which we don't have here. */
561
562 { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND,
563 MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */
564
565 { "/proc/sys/net", "/proc/sys/net", NULL, NULL, MS_BIND,
566 MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
567
568 { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
569 MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* ... then, make it r/o */
570
571 /* Make these files inaccessible to container payloads: they potentially leak information about kernel
572 * internals or the host's execution environment to the container */
573 PROC_INACCESSIBLE_REG("/proc/kallsyms"),
574 PROC_INACCESSIBLE_REG("/proc/kcore"),
575 PROC_INACCESSIBLE_REG("/proc/keys"),
576 PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"),
577 PROC_INACCESSIBLE_REG("/proc/timer_list"),
578
579 /* Make these directories read-only to container payloads: they show hardware information, and in some
580 * cases contain tunables the container really shouldn't have access to. */
581 PROC_READ_ONLY("/proc/acpi"),
582 PROC_READ_ONLY("/proc/apm"),
583 PROC_READ_ONLY("/proc/asound"),
584 PROC_READ_ONLY("/proc/bus"),
585 PROC_READ_ONLY("/proc/fs"),
586 PROC_READ_ONLY("/proc/irq"),
587 PROC_READ_ONLY("/proc/scsi"),
588
589 { "mqueue", "/dev/mqueue", "mqueue", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
590 MOUNT_IN_USERNS|MOUNT_MKDIR },
591
592 /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing when we are privileged) */
593 { "tmpfs", "/tmp", "tmpfs", "mode=01777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
594 MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP|MOUNT_MKDIR|MOUNT_USRQUOTA_GRACEFUL },
595 { "tmpfs", "/sys", "tmpfs", "mode=0555" TMPFS_LIMITS_SYS, MS_NOSUID|MS_NOEXEC|MS_NODEV,
596 MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS|MOUNT_MKDIR|MOUNT_UNMANAGED },
597 { "sysfs", "/sys", "sysfs", NULL, SYS_DEFAULT_MOUNT_FLAGS,
598 MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO|MOUNT_MKDIR|MOUNT_UNMANAGED }, /* skipped if above was mounted */
599 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
600 MOUNT_FATAL|MOUNT_MKDIR|MOUNT_UNMANAGED }, /* skipped if above was mounted */
601 { "tmpfs", "/dev", "tmpfs", "mode=0755" TMPFS_LIMITS_PRIVATE_DEV, MS_NOSUID|MS_STRICTATIME,
602 MOUNT_FATAL|MOUNT_MKDIR },
603 { "tmpfs", "/dev/shm", "tmpfs", "mode=01777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
604 MOUNT_FATAL|MOUNT_MKDIR|MOUNT_USRQUOTA_GRACEFUL },
605 { "tmpfs", "/run", "tmpfs", "mode=0755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
606 MOUNT_FATAL|MOUNT_MKDIR },
607 { "/run/host", "/run/host", NULL, NULL, MS_BIND,
608 MOUNT_FATAL|MOUNT_MKDIR|MOUNT_PREFIX_ROOT }, /* Prepare this so that we can make it read-only when we are done */
609 { "/etc/os-release", "/run/host/os-release", NULL, NULL, MS_BIND,
610 MOUNT_TOUCH }, /* As per kernel interface requirements, bind mount first (creating mount points) and make read-only later */
611 { "/usr/lib/os-release", "/run/host/os-release", NULL, NULL, MS_BIND,
612 MOUNT_FATAL }, /* If /etc/os-release doesn't exist use the version in /usr/lib as fallback */
613 { NULL, "/run/host/os-release", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
614 MOUNT_FATAL },
615 { NULL, "/run/host/os-release", NULL, NULL, MS_PRIVATE,
616 MOUNT_FATAL }, /* Turn off propagation (we only want that for the mount propagation tunnel dir) */
617 { NULL, "/run/host", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
618 MOUNT_FATAL|MOUNT_IN_USERNS },
619 #if HAVE_SELINUX
620 { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND,
621 MOUNT_MKDIR|MOUNT_PRIVILEGED }, /* Bind mount first (mkdir/chown the mount point in case /sys/ is mounted as minimal skeleton tmpfs) */
622 { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
623 MOUNT_UNMANAGED|MOUNT_PRIVILEGED }, /* Then, make it r/o (don't mkdir/chown the mount point here, the previous entry already did that) */
624 { NULL, "/sys/fs/selinux", NULL, NULL, MS_PRIVATE,
625 MOUNT_UNMANAGED|MOUNT_PRIVILEGED }, /* Turn off propagation (we only want that for the mount propagation tunnel dir) */
626 #endif
627 };
628
629 bool use_userns = FLAGS_SET(mount_settings, MOUNT_USE_USERNS);
630 bool netns = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_NETNS);
631 bool ro = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO);
632 bool in_userns = FLAGS_SET(mount_settings, MOUNT_IN_USERNS);
633 bool tmpfs_tmp = FLAGS_SET(mount_settings, MOUNT_APPLY_TMPFS_TMP);
634 bool unmanaged = FLAGS_SET(mount_settings, MOUNT_UNMANAGED);
635 bool privileged = FLAGS_SET(mount_settings, MOUNT_PRIVILEGED);
636 int r;
637
638 FOREACH_ELEMENT(m, mount_table) {
639 _cleanup_free_ char *where = NULL, *options = NULL, *prefixed = NULL;
640 bool fatal = FLAGS_SET(m->mount_settings, MOUNT_FATAL);
641 const char *o;
642
643 /* If we are in managed user namespace mode but the entry is marked for mount outside of
644 * managed user namespace mode, and to be mounted outside the user namespace, then skip it */
645 if (!unmanaged && FLAGS_SET(m->mount_settings, MOUNT_UNMANAGED) && !FLAGS_SET(m->mount_settings, MOUNT_IN_USERNS))
646 continue;
647
648 if (in_userns != FLAGS_SET(m->mount_settings, MOUNT_IN_USERNS))
649 continue;
650
651 if (!netns && FLAGS_SET(m->mount_settings, MOUNT_APPLY_APIVFS_NETNS))
652 continue;
653
654 if (!ro && FLAGS_SET(m->mount_settings, MOUNT_APPLY_APIVFS_RO))
655 continue;
656
657 if (!tmpfs_tmp && FLAGS_SET(m->mount_settings, MOUNT_APPLY_TMPFS_TMP))
658 continue;
659
660 if (!privileged && FLAGS_SET(m->mount_settings, MOUNT_PRIVILEGED))
661 continue;
662
663 r = chase(m->where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where, NULL);
664 if (r < 0)
665 return log_error_errno(r, "Failed to resolve %s%s: %m", strempty(dest), m->where);
666
667 /* Skip this entry if it is not a remount. */
668 if (m->what) {
669 r = path_is_mount_point(where);
670 if (r < 0 && r != -ENOENT)
671 return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
672 if (r > 0)
673 continue;
674 }
675
676 if ((m->mount_settings & (MOUNT_MKDIR|MOUNT_TOUCH)) != 0) {
677 uid_t u = (use_userns && !in_userns) ? uid_shift : UID_INVALID;
678
679 if (FLAGS_SET(m->mount_settings, MOUNT_TOUCH))
680 r = mkdir_parents_safe(dest, where, 0755, u, u, 0);
681 else
682 r = mkdir_p_safe(dest, where, 0755, u, u, 0);
683 if (r < 0 && r != -EEXIST) {
684 if (fatal && r != -EROFS)
685 return log_error_errno(r, "Failed to create directory %s: %m", where);
686
687 log_debug_errno(r, "Failed to create directory %s: %m", where);
688
689 /* If mkdir() or chown() failed due to the root directory being read only,
690 * attempt to mount this fs anyway and let mount_verbose log any errors */
691 if (r != -EROFS)
692 continue;
693 }
694 }
695
696 if (FLAGS_SET(m->mount_settings, MOUNT_TOUCH)) {
697 r = touch(where);
698 if (r < 0 && r != -EEXIST) {
699 if (fatal && r != -EROFS)
700 return log_error_errno(r, "Failed to create file %s: %m", where);
701
702 log_debug_errno(r, "Failed to create file %s: %m", where);
703 if (r != -EROFS)
704 continue;
705 }
706 }
707
708 o = m->options;
709 if (streq_ptr(m->type, "tmpfs")) {
710 r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options);
711 if (r < 0)
712 return log_oom();
713 if (r > 0)
714 o = options;
715 }
716
717 if (FLAGS_SET(m->mount_settings, MOUNT_USRQUOTA_GRACEFUL)) {
718 r = mount_option_supported(m->type, /* key= */ "usrquota", /* value= */ NULL);
719 if (r < 0)
720 log_warning_errno(r, "Failed to determine if '%s' supports 'usrquota', assuming it doesn't: %m", m->type);
721 else if (r == 0)
722 log_debug("Kernel doesn't support 'usrquota' on '%s', not including in mount options for '%s'.", m->type, m->where);
723 else {
724 _cleanup_free_ char *joined = NULL;
725
726 if (!strextend_with_separator(&joined, ",", o ?: POINTER_MAX, "usrquota"))
727 return log_oom();
728
729 free_and_replace(options, joined);
730 o = options;
731 }
732 }
733
734 if (FLAGS_SET(m->mount_settings, MOUNT_PREFIX_ROOT)) {
735 /* Optionally prefix the mount source with the root dir. This is useful in bind
736 * mounts to be created within the container image before we transition into it. Note
737 * that MOUNT_IN_USERNS is run after we transitioned hence prefixing is not necessary
738 * for those. */
739 r = chase(m->what, dest, CHASE_PREFIX_ROOT, &prefixed, NULL);
740 if (r < 0)
741 return log_error_errno(r, "Failed to resolve %s%s: %m", strempty(dest), m->what);
742 }
743
744 r = mount_verbose_full(
745 fatal ? LOG_ERR : LOG_DEBUG,
746 prefixed ?: m->what,
747 where,
748 m->type,
749 m->flags,
750 o,
751 FLAGS_SET(m->mount_settings, MOUNT_FOLLOW_SYMLINKS));
752 if (r < 0 && fatal)
753 return r;
754 }
755
756 return 0;
757 }
758
759 static int parse_mount_bind_options(const char *options, unsigned long *open_tree_flags, char **mount_opts, RemountIdmapping *idmapping) {
760 unsigned long flags = *ASSERT_PTR(open_tree_flags);
761 char *opts = NULL;
762 RemountIdmapping new_idmapping = *ASSERT_PTR(idmapping);
763 int r;
764
765 assert(options);
766 assert(mount_opts);
767
768 for (;;) {
769 _cleanup_free_ char *word = NULL;
770
771 r = extract_first_word(&options, &word, ",", 0);
772 if (r < 0)
773 return log_error_errno(r, "Failed to extract mount option: %m");
774 if (r == 0)
775 break;
776
777 if (streq(word, "rbind"))
778 flags |= AT_RECURSIVE;
779 else if (streq(word, "norbind"))
780 flags &= ~AT_RECURSIVE;
781 else if (streq(word, "idmap"))
782 new_idmapping = REMOUNT_IDMAPPING_HOST_ROOT;
783 else if (streq(word, "noidmap"))
784 new_idmapping = REMOUNT_IDMAPPING_NONE;
785 else if (streq(word, "rootidmap"))
786 new_idmapping = REMOUNT_IDMAPPING_HOST_OWNER;
787 else if (streq(word, "owneridmap"))
788 new_idmapping = REMOUNT_IDMAPPING_HOST_OWNER_TO_TARGET_OWNER;
789 else
790 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
791 "Invalid bind mount option: %s", word);
792 }
793
794 *open_tree_flags = flags;
795 *idmapping = new_idmapping;
796 /* in the future mount_opts will hold string options for mount(2) */
797 *mount_opts = opts;
798
799 return 0;
800 }
801
802 static int mount_bind(const char *dest, CustomMount *m, uid_t uid_shift, uid_t uid_range) {
803 _cleanup_free_ char *mount_opts = NULL, *where = NULL;
804 unsigned long open_tree_flags = OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC | AT_RECURSIVE;
805 struct stat source_st, dest_st;
806 uid_t dest_uid = UID_INVALID;
807 int r;
808 RemountIdmapping idmapping = REMOUNT_IDMAPPING_NONE;
809
810 assert(dest);
811 assert(m);
812
813 if (m->options) {
814 r = parse_mount_bind_options(m->options, &open_tree_flags, &mount_opts, &idmapping);
815 if (r < 0)
816 return r;
817 }
818
819 /* ID remapping cannot be done if user namespaces are not in use (uid_shift is UID_INVALID).
820 * Fail if idmapping was explicitly requested in this state. Otherwise, treat UID_INVALID
821 * as 0 for ownership operations. */
822 if (idmapping != REMOUNT_IDMAPPING_NONE && !uid_is_valid(uid_shift))
823 return log_error_errno(
824 SYNTHETIC_ERRNO(EINVAL),
825 "ID remapping requested for %s, but user namespacing is not enabled.",
826 m->source);
827
828 uid_t chown_uid = uid_is_valid(uid_shift) ? uid_shift : 0;
829
830 /* If this is a bind mount from a temporary sources change ownership of the source to the container's
831 * root UID. Otherwise it would always show up as "nobody" if user namespacing is used. */
832 if (m->rm_rf_tmpdir && chown(m->source, chown_uid, chown_uid) < 0)
833 return log_error_errno(errno, "Failed to chown %s: %m", m->source);
834
835 /* UID/GIDs of idmapped mounts are always resolved in the caller's user namespace. In other
836 * words, they're not nested. If we're doing an idmapped mount from a bind mount that's
837 * already idmapped itself, the old idmap is replaced with the new one. This means that the
838 * source uid which we put in the idmap userns has to be the uid of mount source in the
839 * caller's userns *without* any mount idmapping in place. To get that uid, we clone the
840 * mount source tree and clear any existing idmapping and temporarily mount that tree over
841 * the mount source before we stat the mount source to figure out the source uid. */
842 _cleanup_close_ int fd_clone =
843 idmapping == REMOUNT_IDMAPPING_NONE ?
844 RET_NERRNO(open_tree(
845 AT_FDCWD,
846 m->source,
847 open_tree_flags)) :
848 open_tree_try_drop_idmap(
849 AT_FDCWD,
850 m->source,
851 open_tree_flags);
852 if (fd_clone < 0)
853 return log_error_errno(errno, "Failed to clone %s: %m", m->source);
854
855 if (fstat(fd_clone, &source_st) < 0)
856 return log_error_errno(errno, "Failed to stat %s: %m", m->source);
857
858 r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
859 if (r < 0)
860 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
861 if (r > 0) { /* Path exists already? */
862
863 if (stat(where, &dest_st) < 0)
864 return log_error_errno(errno, "Failed to stat %s: %m", where);
865
866 dest_uid = uid_is_valid(m->destination_uid) ? chown_uid + m->destination_uid : dest_st.st_uid;
867
868 if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode))
869 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
870 "Cannot bind mount directory %s on file %s.",
871 m->source, where);
872
873 if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode))
874 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
875 "Cannot bind mount file %s on directory %s.",
876 m->source, where);
877
878 } else { /* Path doesn't exist yet? */
879 r = mkdir_parents_safe_label(dest, where, 0755, chown_uid, chown_uid, MKDIR_IGNORE_EXISTING);
880 if (r < 0)
881 return log_error_errno(r, "Failed to make parents of %s: %m", where);
882
883 /* Create the mount point. Any non-directory file can be
884 * mounted on any non-directory file (regular, fifo, socket,
885 * char, block).
886 */
887 if (S_ISDIR(source_st.st_mode))
888 r = mkdir_label(where, 0755);
889 else
890 r = touch(where);
891 if (r < 0)
892 return log_error_errno(r, "Failed to create mount point %s: %m", where);
893
894 if (chown(where, chown_uid, chown_uid) < 0)
895 return log_error_errno(errno, "Failed to chown %s: %m", where);
896
897 dest_uid = chown_uid + (uid_is_valid(m->destination_uid) ? m->destination_uid : 0);
898 }
899
900 if (move_mount(fd_clone, "", AT_FDCWD, where, MOVE_MOUNT_F_EMPTY_PATH) < 0)
901 return log_error_errno(errno, "Failed to mount %s to %s: %m", m->source, where);
902
903 fd_clone = safe_close(fd_clone);
904
905 if (m->read_only) {
906 r = bind_remount_recursive(where, MS_RDONLY, MS_RDONLY, NULL);
907 if (r < 0)
908 return log_error_errno(r, "Read-only bind mount failed: %m");
909 }
910
911 if (idmapping != REMOUNT_IDMAPPING_NONE) {
912 r = remount_idmap(STRV_MAKE(where), uid_shift, uid_range, source_st.st_uid, dest_uid, idmapping);
913 if (r < 0)
914 return log_error_errno(r, "Failed to map ids for bind mount %s: %m", where);
915 }
916
917 return 0;
918 }
919
920 static int mount_tmpfs(const char *dest, CustomMount *m, uid_t uid_shift, const char *selinux_apifs_context) {
921 const char *options;
922 _cleanup_free_ char *buf = NULL, *where = NULL;
923 int r;
924
925 assert(dest);
926 assert(m);
927
928 r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
929 if (r < 0)
930 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
931 if (r == 0) { /* Doesn't exist yet? */
932 r = mkdir_p_label(where, 0755);
933 if (r < 0)
934 return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
935 }
936
937 r = tmpfs_patch_options(m->options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
938 if (r < 0)
939 return log_oom();
940 options = r > 0 ? buf : m->options;
941
942 return mount_nofollow_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
943 }
944
945 static char *joined_and_escaped_lower_dirs(char **lower) {
946 _cleanup_strv_free_ char **sv = NULL;
947
948 sv = strv_copy(lower);
949 if (!sv)
950 return NULL;
951
952 strv_reverse(sv);
953
954 if (!strv_shell_escape(sv, ",:"))
955 return NULL;
956
957 return strv_join(sv, ":");
958 }
959
960 static int mount_overlay(const char *dest, CustomMount *m) {
961 _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
962 const char *options;
963 int r;
964
965 assert(dest);
966 assert(m);
967
968 r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
969 if (r < 0)
970 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
971 if (r == 0) { /* Doesn't exist yet? */
972 r = mkdir_label(where, 0755);
973 if (r < 0)
974 return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
975 }
976
977 (void) mkdir_p_label(m->source, 0755);
978
979 lower = joined_and_escaped_lower_dirs(m->lower);
980 if (!lower)
981 return log_oom();
982
983 escaped_source = shell_escape(m->source, ",:");
984 if (!escaped_source)
985 return log_oom();
986
987 if (m->read_only)
988 options = strjoina("lowerdir=", escaped_source, ":", lower);
989 else {
990 _cleanup_free_ char *escaped_work_dir = NULL;
991
992 escaped_work_dir = shell_escape(m->work_dir, ",:");
993 if (!escaped_work_dir)
994 return log_oom();
995
996 options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
997 }
998
999 return mount_nofollow_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
1000 }
1001
1002 static int mount_inaccessible(const char *dest, CustomMount *m) {
1003 _cleanup_free_ char *where = NULL, *source = NULL;
1004 struct stat st;
1005 int r;
1006
1007 assert(dest);
1008 assert(m);
1009
1010 r = chase_and_stat(m->destination, dest, CHASE_PREFIX_ROOT, &where, &st);
1011 if (r < 0) {
1012 log_full_errno(m->graceful ? LOG_DEBUG : LOG_ERR, r, "Failed to resolve %s/%s: %m", dest, m->destination);
1013 return m->graceful ? 0 : r;
1014 }
1015
1016 r = mode_to_inaccessible_node(NULL, st.st_mode, &source);
1017 if (r < 0)
1018 return m->graceful ? 0 : r;
1019
1020 r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, source, where, NULL, MS_BIND, NULL);
1021 if (r < 0)
1022 return m->graceful ? 0 : r;
1023
1024 r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, NULL, where, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, NULL);
1025 if (r < 0) {
1026 (void) umount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, where, UMOUNT_NOFOLLOW);
1027 return m->graceful ? 0 : r;
1028 }
1029
1030 return 0;
1031 }
1032
1033 static int mount_arbitrary(const char *dest, CustomMount *m) {
1034 _cleanup_free_ char *where = NULL;
1035 int r;
1036
1037 assert(dest);
1038 assert(m);
1039
1040 r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
1041 if (r < 0)
1042 return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
1043 if (r == 0) { /* Doesn't exist yet? */
1044 r = mkdir_p_label(where, 0755);
1045 if (r < 0)
1046 return log_error_errno(r, "Creating mount point for mount %s failed: %m", where);
1047 }
1048
1049 return mount_nofollow_verbose(LOG_ERR, m->source, where, m->type_argument, 0, m->options);
1050 }
1051
1052 int mount_custom(
1053 const char *dest,
1054 CustomMount *mounts, size_t n,
1055 uid_t uid_shift,
1056 uid_t uid_range,
1057 const char *selinux_apifs_context,
1058 MountSettingsMask mount_settings) {
1059 int r;
1060
1061 assert(dest);
1062
1063 FOREACH_ARRAY(m, mounts, n) {
1064 if (FLAGS_SET(mount_settings, MOUNT_IN_USERNS) != m->in_userns)
1065 continue;
1066
1067 if (FLAGS_SET(mount_settings, MOUNT_ROOT_ONLY) && !path_equal(m->destination, "/"))
1068 continue;
1069
1070 if (FLAGS_SET(mount_settings, MOUNT_NON_ROOT_ONLY) && path_equal(m->destination, "/"))
1071 continue;
1072
1073 switch (m->type) {
1074
1075 case CUSTOM_MOUNT_BIND:
1076 r = mount_bind(dest, m, uid_shift, uid_range);
1077 break;
1078
1079 case CUSTOM_MOUNT_TMPFS:
1080 r = mount_tmpfs(dest, m, uid_shift, selinux_apifs_context);
1081 break;
1082
1083 case CUSTOM_MOUNT_OVERLAY:
1084 r = mount_overlay(dest, m);
1085 break;
1086
1087 case CUSTOM_MOUNT_INACCESSIBLE:
1088 r = mount_inaccessible(dest, m);
1089 break;
1090
1091 case CUSTOM_MOUNT_ARBITRARY:
1092 r = mount_arbitrary(dest, m);
1093 break;
1094
1095 default:
1096 assert_not_reached();
1097 }
1098
1099 if (r < 0)
1100 return r;
1101 }
1102
1103 return 0;
1104 }
1105
1106 bool has_custom_root_mount(const CustomMount *mounts, size_t n) {
1107 FOREACH_ARRAY(m, mounts, n)
1108 if (path_equal(m->destination, "/"))
1109 return true;
1110
1111 return false;
1112 }
1113
1114 static int setup_volatile_state(const char *directory) {
1115 int r;
1116
1117 assert(directory);
1118
1119 /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
1120
1121 /* First, remount the root directory. */
1122 r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
1123 if (r < 0)
1124 return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
1125
1126 return 0;
1127 }
1128
1129 static int setup_volatile_state_after_remount_idmap(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1130 _cleanup_free_ char *buf = NULL;
1131 int r;
1132
1133 assert(directory);
1134
1135 /* Then, after remount_idmap(), overmount /var/ with a tmpfs. */
1136
1137 _cleanup_free_ char *p = path_join(directory, "/var");
1138 if (!p)
1139 return log_oom();
1140
1141 r = mkdir(p, 0755);
1142 if (r < 0 && errno != EEXIST)
1143 return log_error_errno(errno, "Failed to create %s: %m", directory);
1144
1145 const char *options = "mode=0755" TMPFS_LIMITS_VOLATILE_STATE;
1146 r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1147 if (r < 0)
1148 return log_oom();
1149 if (r > 0)
1150 options = buf;
1151
1152 return mount_nofollow_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
1153 }
1154
1155 static int setup_volatile_yes(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1156 bool tmpfs_mounted = false, bind_mounted = false;
1157 _cleanup_(rmdir_and_freep) char *template = NULL;
1158 _cleanup_free_ char *buf = NULL, *bindir = NULL, *f = NULL, *t = NULL;
1159 struct stat st;
1160 int r;
1161
1162 assert(directory);
1163
1164 /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and
1165 * that read-only. Before we start setting this up let's validate if the image has the /usr merge
1166 * implemented, and let's output a friendly log message if it hasn't. */
1167
1168 bindir = path_join(directory, "/bin");
1169 if (!bindir)
1170 return log_oom();
1171 if (lstat(bindir, &st) < 0) {
1172 if (errno != ENOENT)
1173 return log_error_errno(errno, "Failed to stat /bin directory below image: %m");
1174
1175 /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the
1176 * rest. */
1177 } else if (S_ISDIR(st.st_mode))
1178 return log_error_errno(SYNTHETIC_ERRNO(EISDIR),
1179 "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. "
1180 "Please work with your distribution and help them adopt the merged /usr scheme.");
1181 else if (!S_ISLNK(st.st_mode))
1182 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
1183 "Error starting image: if --volatile=yes is used /bin must be a symlink (for merged /usr support) or non-existent (in which case a symlink is created automatically).");
1184
1185 r = mkdtemp_malloc("/tmp/nspawn-volatile-XXXXXX", &template);
1186 if (r < 0)
1187 return log_error_errno(r, "Failed to create temporary directory: %m");
1188
1189 const char *options = "mode=0755" TMPFS_LIMITS_ROOTFS;
1190 r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1191 if (r < 0)
1192 goto fail;
1193 if (r > 0)
1194 options = buf;
1195
1196 r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1197 if (r < 0)
1198 goto fail;
1199
1200 tmpfs_mounted = true;
1201
1202 f = path_join(directory, "/usr");
1203 if (!f) {
1204 r = log_oom();
1205 goto fail;
1206 }
1207
1208 t = path_join(template, "/usr");
1209 if (!t) {
1210 r = log_oom();
1211 goto fail;
1212 }
1213
1214 r = mkdir(t, 0755);
1215 if (r < 0 && errno != EEXIST) {
1216 r = log_error_errno(errno, "Failed to create %s: %m", t);
1217 goto fail;
1218 }
1219
1220 r = mount_nofollow_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
1221 if (r < 0)
1222 goto fail;
1223
1224 bind_mounted = true;
1225
1226 r = bind_remount_recursive(t, MS_RDONLY, MS_RDONLY, NULL);
1227 if (r < 0) {
1228 log_error_errno(r, "Failed to remount %s read-only: %m", t);
1229 goto fail;
1230 }
1231
1232 r = mount_nofollow_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
1233 if (r < 0)
1234 goto fail;
1235
1236 (void) rmdir(template);
1237
1238 return 0;
1239
1240 fail:
1241 if (bind_mounted)
1242 (void) umount_verbose(LOG_ERR, t, UMOUNT_NOFOLLOW);
1243
1244 if (tmpfs_mounted)
1245 (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW);
1246
1247 return r;
1248 }
1249
1250 static int setup_volatile_overlay(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
1251 _cleanup_free_ char *buf = NULL, *escaped_directory = NULL, *escaped_upper = NULL, *escaped_work = NULL;
1252 _cleanup_(rmdir_and_freep) char *template = NULL;
1253 const char *upper, *work, *options;
1254 bool tmpfs_mounted = false;
1255 int r;
1256
1257 assert(directory);
1258
1259 /* --volatile=overlay means we mount an overlayfs to the root dir. */
1260
1261 r = mkdtemp_malloc("/tmp/nspawn-volatile-XXXXXX", &template);
1262 if (r < 0)
1263 return log_error_errno(r, "Failed to create temporary directory: %m");
1264
1265 options = "mode=0755" TMPFS_LIMITS_ROOTFS;
1266 r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
1267 if (r < 0)
1268 goto finish;
1269 if (r > 0)
1270 options = buf;
1271
1272 r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
1273 if (r < 0)
1274 goto finish;
1275
1276 tmpfs_mounted = true;
1277
1278 upper = strjoina(template, "/upper");
1279 work = strjoina(template, "/work");
1280
1281 if (mkdir(upper, 0755) < 0) {
1282 r = log_error_errno(errno, "Failed to create %s: %m", upper);
1283 goto finish;
1284 }
1285 if (mkdir(work, 0755) < 0) {
1286 r = log_error_errno(errno, "Failed to create %s: %m", work);
1287 goto finish;
1288 }
1289
1290 /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice
1291 * that the kernel allows us to do that without going through some mount point rearrangements. */
1292
1293 escaped_directory = shell_escape(directory, ",:");
1294 escaped_upper = shell_escape(upper, ",:");
1295 escaped_work = shell_escape(work, ",:");
1296 if (!escaped_directory || !escaped_upper || !escaped_work) {
1297 r = -ENOMEM;
1298 goto finish;
1299 }
1300
1301 options = strjoina("lowerdir=", escaped_directory, ",upperdir=", escaped_upper, ",workdir=", escaped_work);
1302 r = mount_nofollow_verbose(LOG_ERR, "overlay", directory, "overlay", 0, options);
1303
1304 finish:
1305 if (tmpfs_mounted)
1306 (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW);
1307
1308 return r;
1309 }
1310
1311 int setup_volatile_mode(
1312 const char *directory,
1313 VolatileMode mode,
1314 uid_t uid_shift,
1315 const char *selinux_apifs_context) {
1316
1317 switch (mode) {
1318
1319 case VOLATILE_YES:
1320 return setup_volatile_yes(directory, uid_shift, selinux_apifs_context);
1321
1322 case VOLATILE_STATE:
1323 return setup_volatile_state(directory);
1324
1325 case VOLATILE_OVERLAY:
1326 return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context);
1327
1328 default:
1329 return 0;
1330 }
1331 }
1332
1333 int setup_volatile_mode_after_remount_idmap(
1334 const char *directory,
1335 VolatileMode mode,
1336 uid_t uid_shift,
1337 const char *selinux_apifs_context) {
1338
1339 switch (mode) {
1340
1341 case VOLATILE_STATE:
1342 return setup_volatile_state_after_remount_idmap(directory, uid_shift, selinux_apifs_context);
1343
1344 default:
1345 return 0;
1346 }
1347 }
1348
1349 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1350 int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
1351 _cleanup_free_ char *root_new = NULL, *root_old = NULL;
1352 const char *p = s;
1353 int r;
1354
1355 assert(pivot_root_new);
1356 assert(pivot_root_old);
1357
1358 r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
1359 if (r < 0)
1360 return r;
1361 if (r == 0)
1362 return -EINVAL;
1363
1364 if (isempty(p))
1365 root_old = NULL;
1366 else {
1367 root_old = strdup(p);
1368 if (!root_old)
1369 return -ENOMEM;
1370 }
1371
1372 if (!path_is_absolute(root_new))
1373 return -EINVAL;
1374 if (!path_is_normalized(root_new))
1375 return -EINVAL;
1376 if (root_old && (!path_is_absolute(root_old) || !path_is_normalized(root_old)))
1377 return -EINVAL;
1378
1379 free_and_replace(*pivot_root_new, root_new);
1380 free_and_replace(*pivot_root_old, root_old);
1381
1382 return 0;
1383 }
1384
1385 int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
1386 _cleanup_free_ char *directory_pivot_root_new = NULL;
1387 _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
1388 _cleanup_(rmdir_and_freep) char *pivot_tmp = NULL;
1389 int r;
1390
1391 assert(directory);
1392
1393 if (!pivot_root_new)
1394 return 0;
1395
1396 /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1397 * If pivot_root_old is NULL, the existing / disappears.
1398 * This requires a temporary directory, pivot_tmp, which is
1399 * not a child of either.
1400 *
1401 * This is typically used for OSTree-style containers, where the root partition contains several
1402 * sysroots which could be run. Normally, one would be chosen by the bootloader and pivoted to / by
1403 * initrd.
1404 *
1405 * For example, for an OSTree deployment, pivot_root_new
1406 * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1407 * code doesn’t do the /var mount which OSTree expects: use
1408 * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1409 *
1410 * So in the OSTree case, we’ll end up with something like:
1411 * - directory = /tmp/nspawn-root-123456
1412 * - pivot_root_new = /ostree/deploy/os/deploy/123abc
1413 * - pivot_root_old = /sysroot
1414 * - directory_pivot_root_new =
1415 * /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1416 * - pivot_tmp = /tmp/nspawn-pivot-123456
1417 * - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1418 *
1419 * Requires all file systems at directory and below to be mounted
1420 * MS_PRIVATE or MS_SLAVE so they can be moved.
1421 */
1422 directory_pivot_root_new = path_join(directory, pivot_root_new);
1423 if (!directory_pivot_root_new)
1424 return log_oom();
1425
1426 /* Remount directory_pivot_root_new to make it movable. */
1427 r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
1428 if (r < 0)
1429 return r;
1430
1431 if (pivot_root_old) {
1432 r = mkdtemp_malloc("/tmp/nspawn-pivot-XXXXXX", &pivot_tmp);
1433 if (r < 0)
1434 return log_error_errno(r, "Failed to create temporary directory: %m");
1435
1436 pivot_tmp_pivot_root_old = path_join(pivot_tmp, pivot_root_old);
1437 if (!pivot_tmp_pivot_root_old)
1438 return log_oom();
1439
1440 r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
1441 if (r < 0)
1442 return r;
1443
1444 r = mount_nofollow_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
1445 if (r < 0)
1446 return r;
1447
1448 r = mount_nofollow_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
1449 } else
1450 r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
1451
1452 if (r < 0)
1453 return r;
1454
1455 return 0;
1456 }
1457
1458 #define NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS "/run/host/proc"
1459 #define NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS "/run/host/sys"
1460
1461 int pin_fully_visible_api_fs(void) {
1462 int r;
1463
1464 log_debug("Pinning fully visible API FS");
1465
1466 (void) mkdir_p(NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS, 0755);
1467 (void) mkdir_p(NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS, 0755);
1468
1469 r = mount_follow_verbose(LOG_ERR, "proc", NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS, "proc", PROC_DEFAULT_MOUNT_FLAGS, NULL);
1470 if (r < 0)
1471 return r;
1472
1473 r = mount_follow_verbose(LOG_ERR, "sysfs", NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS, "sysfs", SYS_DEFAULT_MOUNT_FLAGS, NULL);
1474 if (r < 0)
1475 return r;
1476
1477 return 0;
1478 }
1479
1480 static int do_wipe_fully_visible_api_fs(void) {
1481 if (umount2(NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS, MNT_DETACH) < 0)
1482 return log_error_errno(errno, "Failed to unmount temporary proc: %m");
1483
1484 if (rmdir(NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS) < 0)
1485 return log_error_errno(errno, "Failed to remove temporary proc mountpoint: %m");
1486
1487 if (umount2(NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS, MNT_DETACH) < 0)
1488 return log_error_errno(errno, "Failed to unmount temporary sys: %m");
1489
1490 if (rmdir(NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS) < 0)
1491 return log_error_errno(errno, "Failed to remove temporary sys mountpoint: %m");
1492
1493 return 0;
1494 }
1495
1496 int wipe_fully_visible_api_fs(int mntns_fd) {
1497 _cleanup_close_ int orig_mntns_fd = -EBADF;
1498 int r;
1499
1500 log_debug("Wiping fully visible API FS");
1501
1502 orig_mntns_fd = namespace_open_by_type(NAMESPACE_MOUNT);
1503 if (orig_mntns_fd < 0)
1504 return log_error_errno(orig_mntns_fd, "Failed to pin originating mount namespace: %m");
1505
1506 if (setns(mntns_fd, CLONE_NEWNS) < 0)
1507 return log_error_errno(errno, "Failed to enter mount namespace: %m");
1508
1509 r = do_wipe_fully_visible_api_fs();
1510
1511 if (setns(orig_mntns_fd, CLONE_NEWNS) < 0)
1512 return log_error_errno(errno, "Failed to enter original mount namespace: %m");
1513
1514 return r;
1515 }