]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/namespace.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #17144 from poettering/mount-nofollow
[thirdparty/systemd.git] / src / core / namespace.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <errno.h>
4 #include <linux/loop.h>
5 #include <sched.h>
6 #include <stdio.h>
7 #include <sys/mount.h>
8 #include <unistd.h>
9 #include <linux/fs.h>
10
11 #include "alloc-util.h"
12 #include "base-filesystem.h"
13 #include "dev-setup.h"
14 #include "fd-util.h"
15 #include "format-util.h"
16 #include "fs-util.h"
17 #include "label.h"
18 #include "list.h"
19 #include "loop-util.h"
20 #include "loopback-setup.h"
21 #include "mkdir.h"
22 #include "mount-util.h"
23 #include "mountpoint-util.h"
24 #include "namespace-util.h"
25 #include "namespace.h"
26 #include "nulstr-util.h"
27 #include "path-util.h"
28 #include "selinux-util.h"
29 #include "socket-util.h"
30 #include "sort-util.h"
31 #include "stat-util.h"
32 #include "string-table.h"
33 #include "string-util.h"
34 #include "strv.h"
35 #include "tmpfile-util.h"
36 #include "umask-util.h"
37 #include "user-util.h"
38
39 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
40
41 typedef enum MountMode {
42 /* This is ordered by priority! */
43 INACCESSIBLE,
44 MOUNT_IMAGES,
45 BIND_MOUNT,
46 BIND_MOUNT_RECURSIVE,
47 PRIVATE_TMP,
48 PRIVATE_TMP_READONLY,
49 PRIVATE_DEV,
50 BIND_DEV,
51 EMPTY_DIR,
52 SYSFS,
53 PROCFS,
54 READONLY,
55 READWRITE,
56 TMPFS,
57 READWRITE_IMPLICIT, /* Should have the lowest priority. */
58 _MOUNT_MODE_MAX,
59 } MountMode;
60
61 typedef struct MountEntry {
62 const char *path_const; /* Memory allocated on stack or static */
63 MountMode mode:5;
64 bool ignore:1; /* Ignore if path does not exist? */
65 bool has_prefix:1; /* Already is prefixed by the root dir? */
66 bool read_only:1; /* Shall this mount point be read-only? */
67 bool nosuid:1; /* Shall set MS_NOSUID on the mount itself */
68 bool applied:1; /* Already applied */
69 char *path_malloc; /* Use this instead of 'path_const' if we had to allocate memory */
70 const char *source_const; /* The source path, for bind mounts or images */
71 char *source_malloc;
72 const char *options_const;/* Mount options for tmpfs */
73 char *options_malloc;
74 unsigned long flags; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
75 unsigned n_followed;
76 LIST_HEAD(MountOptions, image_options);
77 } MountEntry;
78
79 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
80 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
81 static const MountEntry apivfs_table[] = {
82 { "/proc", PROCFS, false },
83 { "/dev", BIND_DEV, false },
84 { "/sys", SYSFS, false },
85 };
86
87 /* ProtectKernelTunables= option and the related filesystem APIs */
88 static const MountEntry protect_kernel_tunables_table[] = {
89 { "/proc/acpi", READONLY, true },
90 { "/proc/apm", READONLY, true }, /* Obsolete API, there's no point in permitting access to this, ever */
91 { "/proc/asound", READONLY, true },
92 { "/proc/bus", READONLY, true },
93 { "/proc/fs", READONLY, true },
94 { "/proc/irq", READONLY, true },
95 { "/proc/kallsyms", INACCESSIBLE, true },
96 { "/proc/kcore", INACCESSIBLE, true },
97 { "/proc/latency_stats", READONLY, true },
98 { "/proc/mtrr", READONLY, true },
99 { "/proc/scsi", READONLY, true },
100 { "/proc/sys", READONLY, true },
101 { "/proc/sysrq-trigger", READONLY, true },
102 { "/proc/timer_stats", READONLY, true },
103 { "/sys", READONLY, false },
104 { "/sys/fs/bpf", READONLY, true },
105 { "/sys/fs/cgroup", READWRITE_IMPLICIT, false }, /* READONLY is set by ProtectControlGroups= option */
106 { "/sys/fs/selinux", READWRITE_IMPLICIT, true },
107 { "/sys/kernel/debug", READONLY, true },
108 { "/sys/kernel/tracing", READONLY, true },
109 };
110
111 /* ProtectKernelModules= option */
112 static const MountEntry protect_kernel_modules_table[] = {
113 #if HAVE_SPLIT_USR
114 { "/lib/modules", INACCESSIBLE, true },
115 #endif
116 { "/usr/lib/modules", INACCESSIBLE, true },
117 };
118
119 /* ProtectKernelLogs= option */
120 static const MountEntry protect_kernel_logs_table[] = {
121 { "/proc/kmsg", INACCESSIBLE, true },
122 { "/dev/kmsg", INACCESSIBLE, true },
123 };
124
125 /*
126 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
127 * system should be protected by ProtectSystem=
128 */
129 static const MountEntry protect_home_read_only_table[] = {
130 { "/home", READONLY, true },
131 { "/run/user", READONLY, true },
132 { "/root", READONLY, true },
133 };
134
135 /* ProtectHome=tmpfs table */
136 static const MountEntry protect_home_tmpfs_table[] = {
137 { "/home", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME },
138 { "/run/user", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME },
139 { "/root", TMPFS, true, .read_only = true, .options_const = "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME },
140 };
141
142 /* ProtectHome=yes table */
143 static const MountEntry protect_home_yes_table[] = {
144 { "/home", INACCESSIBLE, true },
145 { "/run/user", INACCESSIBLE, true },
146 { "/root", INACCESSIBLE, true },
147 };
148
149 /* ProtectSystem=yes table */
150 static const MountEntry protect_system_yes_table[] = {
151 { "/usr", READONLY, false },
152 { "/boot", READONLY, true },
153 { "/efi", READONLY, true },
154 #if HAVE_SPLIT_USR
155 { "/lib", READONLY, true },
156 { "/lib64", READONLY, true },
157 { "/bin", READONLY, true },
158 # if HAVE_SPLIT_BIN
159 { "/sbin", READONLY, true },
160 # endif
161 #endif
162 };
163
164 /* ProtectSystem=full includes ProtectSystem=yes */
165 static const MountEntry protect_system_full_table[] = {
166 { "/usr", READONLY, false },
167 { "/boot", READONLY, true },
168 { "/efi", READONLY, true },
169 { "/etc", READONLY, false },
170 #if HAVE_SPLIT_USR
171 { "/lib", READONLY, true },
172 { "/lib64", READONLY, true },
173 { "/bin", READONLY, true },
174 # if HAVE_SPLIT_BIN
175 { "/sbin", READONLY, true },
176 # endif
177 #endif
178 };
179
180 /*
181 * ProtectSystem=strict table. In this strict mode, we mount everything
182 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
183 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
184 * protect those, and these options should be fully orthogonal.
185 * (And of course /home and friends are also left writable, as ProtectHome=
186 * shall manage those, orthogonally).
187 */
188 static const MountEntry protect_system_strict_table[] = {
189 { "/", READONLY, false },
190 { "/proc", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */
191 { "/sys", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */
192 { "/dev", READWRITE_IMPLICIT, false }, /* PrivateDevices= */
193 { "/home", READWRITE_IMPLICIT, true }, /* ProtectHome= */
194 { "/run/user", READWRITE_IMPLICIT, true }, /* ProtectHome= */
195 { "/root", READWRITE_IMPLICIT, true }, /* ProtectHome= */
196 };
197
198 static const char * const mount_mode_table[_MOUNT_MODE_MAX] = {
199 [INACCESSIBLE] = "inaccessible",
200 [BIND_MOUNT] = "bind",
201 [BIND_MOUNT_RECURSIVE] = "rbind",
202 [PRIVATE_TMP] = "private-tmp",
203 [PRIVATE_DEV] = "private-dev",
204 [BIND_DEV] = "bind-dev",
205 [EMPTY_DIR] = "empty",
206 [SYSFS] = "sysfs",
207 [PROCFS] = "procfs",
208 [READONLY] = "read-only",
209 [READWRITE] = "read-write",
210 [TMPFS] = "tmpfs",
211 [MOUNT_IMAGES] = "mount-images",
212 [READWRITE_IMPLICIT] = "rw-implicit",
213 };
214
215 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode, MountMode);
216
217 static const char *mount_entry_path(const MountEntry *p) {
218 assert(p);
219
220 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
221 * otherwise the stack/static ->path field is returned. */
222
223 return p->path_malloc ?: p->path_const;
224 }
225
226 static bool mount_entry_read_only(const MountEntry *p) {
227 assert(p);
228
229 return p->read_only || IN_SET(p->mode, READONLY, INACCESSIBLE, PRIVATE_TMP_READONLY);
230 }
231
232 static const char *mount_entry_source(const MountEntry *p) {
233 assert(p);
234
235 return p->source_malloc ?: p->source_const;
236 }
237
238 static const char *mount_entry_options(const MountEntry *p) {
239 assert(p);
240
241 return p->options_malloc ?: p->options_const;
242 }
243
244 static void mount_entry_done(MountEntry *p) {
245 assert(p);
246
247 p->path_malloc = mfree(p->path_malloc);
248 p->source_malloc = mfree(p->source_malloc);
249 p->options_malloc = mfree(p->options_malloc);
250 p->image_options = mount_options_free_all(p->image_options);
251 }
252
253 static int append_access_mounts(MountEntry **p, char **strv, MountMode mode, bool forcibly_require_prefix) {
254 char **i;
255
256 assert(p);
257
258 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
259
260 STRV_FOREACH(i, strv) {
261 bool ignore = false, needs_prefix = false;
262 const char *e = *i;
263
264 /* Look for any prefixes */
265 if (startswith(e, "-")) {
266 e++;
267 ignore = true;
268 }
269 if (startswith(e, "+")) {
270 e++;
271 needs_prefix = true;
272 }
273
274 if (!path_is_absolute(e))
275 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL),
276 "Path is not absolute: %s", e);
277
278 *((*p)++) = (MountEntry) {
279 .path_const = e,
280 .mode = mode,
281 .ignore = ignore,
282 .has_prefix = !needs_prefix && !forcibly_require_prefix,
283 };
284 }
285
286 return 0;
287 }
288
289 static int append_empty_dir_mounts(MountEntry **p, char **strv) {
290 char **i;
291
292 assert(p);
293
294 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
295 * "/private/" boundary directories for DynamicUser=1. */
296
297 STRV_FOREACH(i, strv) {
298
299 *((*p)++) = (MountEntry) {
300 .path_const = *i,
301 .mode = EMPTY_DIR,
302 .ignore = false,
303 .read_only = true,
304 .options_const = "mode=755" TMPFS_LIMITS_EMPTY_OR_ALMOST,
305 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
306 };
307 }
308
309 return 0;
310 }
311
312 static int append_bind_mounts(MountEntry **p, const BindMount *binds, size_t n) {
313 size_t i;
314
315 assert(p);
316
317 for (i = 0; i < n; i++) {
318 const BindMount *b = binds + i;
319
320 *((*p)++) = (MountEntry) {
321 .path_const = b->destination,
322 .mode = b->recursive ? BIND_MOUNT_RECURSIVE : BIND_MOUNT,
323 .read_only = b->read_only,
324 .nosuid = b->nosuid,
325 .source_const = b->source,
326 .ignore = b->ignore_enoent,
327 };
328 }
329
330 return 0;
331 }
332
333 static int append_mount_images(MountEntry **p, const MountImage *mount_images, size_t n) {
334 assert(p);
335
336 for (size_t i = 0; i < n; i++) {
337 const MountImage *m = mount_images + i;
338
339 *((*p)++) = (MountEntry) {
340 .path_const = m->destination,
341 .mode = MOUNT_IMAGES,
342 .source_const = m->source,
343 .image_options = m->mount_options,
344 .ignore = m->ignore_enoent,
345 };
346 }
347
348 return 0;
349 }
350
351 static int append_tmpfs_mounts(MountEntry **p, const TemporaryFileSystem *tmpfs, size_t n) {
352 assert(p);
353
354 for (size_t i = 0; i < n; i++) {
355 const TemporaryFileSystem *t = tmpfs + i;
356 _cleanup_free_ char *o = NULL, *str = NULL;
357 unsigned long flags;
358 bool ro = false;
359 int r;
360
361 if (!path_is_absolute(t->path))
362 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL),
363 "Path is not absolute: %s",
364 t->path);
365
366 str = strjoin("mode=0755" NESTED_TMPFS_LIMITS ",", t->options);
367 if (!str)
368 return -ENOMEM;
369
370 r = mount_option_mangle(str, MS_NODEV|MS_STRICTATIME, &flags, &o);
371 if (r < 0)
372 return log_debug_errno(r, "Failed to parse mount option '%s': %m", str);
373
374 ro = flags & MS_RDONLY;
375 if (ro)
376 flags ^= MS_RDONLY;
377
378 *((*p)++) = (MountEntry) {
379 .path_const = t->path,
380 .mode = TMPFS,
381 .read_only = ro,
382 .options_malloc = TAKE_PTR(o),
383 .flags = flags,
384 };
385 }
386
387 return 0;
388 }
389
390 static int append_static_mounts(MountEntry **p, const MountEntry *mounts, size_t n, bool ignore_protect) {
391 size_t i;
392
393 assert(p);
394 assert(mounts);
395
396 /* Adds a list of static pre-defined entries */
397
398 for (i = 0; i < n; i++)
399 *((*p)++) = (MountEntry) {
400 .path_const = mount_entry_path(mounts+i),
401 .mode = mounts[i].mode,
402 .ignore = mounts[i].ignore || ignore_protect,
403 };
404
405 return 0;
406 }
407
408 static int append_protect_home(MountEntry **p, ProtectHome protect_home, bool ignore_protect) {
409 assert(p);
410
411 switch (protect_home) {
412
413 case PROTECT_HOME_NO:
414 return 0;
415
416 case PROTECT_HOME_READ_ONLY:
417 return append_static_mounts(p, protect_home_read_only_table, ELEMENTSOF(protect_home_read_only_table), ignore_protect);
418
419 case PROTECT_HOME_TMPFS:
420 return append_static_mounts(p, protect_home_tmpfs_table, ELEMENTSOF(protect_home_tmpfs_table), ignore_protect);
421
422 case PROTECT_HOME_YES:
423 return append_static_mounts(p, protect_home_yes_table, ELEMENTSOF(protect_home_yes_table), ignore_protect);
424
425 default:
426 assert_not_reached("Unexpected ProtectHome= value");
427 }
428 }
429
430 static int append_protect_system(MountEntry **p, ProtectSystem protect_system, bool ignore_protect) {
431 assert(p);
432
433 switch (protect_system) {
434
435 case PROTECT_SYSTEM_NO:
436 return 0;
437
438 case PROTECT_SYSTEM_STRICT:
439 return append_static_mounts(p, protect_system_strict_table, ELEMENTSOF(protect_system_strict_table), ignore_protect);
440
441 case PROTECT_SYSTEM_YES:
442 return append_static_mounts(p, protect_system_yes_table, ELEMENTSOF(protect_system_yes_table), ignore_protect);
443
444 case PROTECT_SYSTEM_FULL:
445 return append_static_mounts(p, protect_system_full_table, ELEMENTSOF(protect_system_full_table), ignore_protect);
446
447 default:
448 assert_not_reached("Unexpected ProtectSystem= value");
449 }
450 }
451
452 static int mount_path_compare(const MountEntry *a, const MountEntry *b) {
453 int d;
454
455 /* If the paths are not equal, then order prefixes first */
456 d = path_compare(mount_entry_path(a), mount_entry_path(b));
457 if (d != 0)
458 return d;
459
460 /* If the paths are equal, check the mode */
461 return CMP((int) a->mode, (int) b->mode);
462 }
463
464 static int prefix_where_needed(MountEntry *m, size_t n, const char *root_directory) {
465 size_t i;
466
467 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
468
469 for (i = 0; i < n; i++) {
470 char *s;
471
472 if (m[i].has_prefix)
473 continue;
474
475 s = path_join(root_directory, mount_entry_path(m+i));
476 if (!s)
477 return -ENOMEM;
478
479 free_and_replace(m[i].path_malloc, s);
480 m[i].has_prefix = true;
481 }
482
483 return 0;
484 }
485
486 static void drop_duplicates(MountEntry *m, size_t *n) {
487 MountEntry *f, *t, *previous;
488
489 assert(m);
490 assert(n);
491
492 /* Drops duplicate entries. Expects that the array is properly ordered already. */
493
494 for (f = m, t = m, previous = NULL; f < m + *n; f++) {
495
496 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
497 * above. Note that we only drop duplicates that haven't been mounted yet. */
498 if (previous &&
499 path_equal(mount_entry_path(f), mount_entry_path(previous)) &&
500 !f->applied && !previous->applied) {
501 log_debug("%s (%s) is duplicate.", mount_entry_path(f), mount_mode_to_string(f->mode));
502 previous->read_only = previous->read_only || mount_entry_read_only(f); /* Propagate the read-only flag to the remaining entry */
503 mount_entry_done(f);
504 continue;
505 }
506
507 *t = *f;
508 previous = t;
509 t++;
510 }
511
512 *n = t - m;
513 }
514
515 static void drop_inaccessible(MountEntry *m, size_t *n) {
516 MountEntry *f, *t;
517 const char *clear = NULL;
518
519 assert(m);
520 assert(n);
521
522 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
523 * ordered already. */
524
525 for (f = m, t = m; f < m + *n; f++) {
526
527 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
528 * it, as inaccessible paths really should drop the entire subtree. */
529 if (clear && path_startswith(mount_entry_path(f), clear)) {
530 log_debug("%s is masked by %s.", mount_entry_path(f), clear);
531 mount_entry_done(f);
532 continue;
533 }
534
535 clear = f->mode == INACCESSIBLE ? mount_entry_path(f) : NULL;
536
537 *t = *f;
538 t++;
539 }
540
541 *n = t - m;
542 }
543
544 static void drop_nop(MountEntry *m, size_t *n) {
545 MountEntry *f, *t;
546
547 assert(m);
548 assert(n);
549
550 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
551 * list is ordered by prefixes. */
552
553 for (f = m, t = m; f < m + *n; f++) {
554
555 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
556 if (IN_SET(f->mode, READONLY, READWRITE, READWRITE_IMPLICIT)) {
557 MountEntry *p;
558 bool found = false;
559
560 /* Now let's find the first parent of the entry we are looking at. */
561 for (p = t-1; p >= m; p--) {
562 if (path_startswith(mount_entry_path(f), mount_entry_path(p))) {
563 found = true;
564 break;
565 }
566 }
567
568 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
569 if (found && p->mode == f->mode) {
570 log_debug("%s (%s) is made redundant by %s (%s)",
571 mount_entry_path(f), mount_mode_to_string(f->mode),
572 mount_entry_path(p), mount_mode_to_string(p->mode));
573 mount_entry_done(f);
574 continue;
575 }
576 }
577
578 *t = *f;
579 t++;
580 }
581
582 *n = t - m;
583 }
584
585 static void drop_outside_root(const char *root_directory, MountEntry *m, size_t *n) {
586 MountEntry *f, *t;
587
588 assert(m);
589 assert(n);
590
591 /* Nothing to do */
592 if (!root_directory)
593 return;
594
595 /* Drops all mounts that are outside of the root directory. */
596
597 for (f = m, t = m; f < m + *n; f++) {
598
599 if (!path_startswith(mount_entry_path(f), root_directory)) {
600 log_debug("%s is outside of root directory.", mount_entry_path(f));
601 mount_entry_done(f);
602 continue;
603 }
604
605 *t = *f;
606 t++;
607 }
608
609 *n = t - m;
610 }
611
612 static int clone_device_node(
613 const char *d,
614 const char *temporary_mount,
615 bool *make_devnode) {
616
617 _cleanup_free_ char *sl = NULL;
618 const char *dn, *bn, *t;
619 struct stat st;
620 int r;
621
622 if (stat(d, &st) < 0) {
623 if (errno == ENOENT) {
624 log_debug_errno(errno, "Device node '%s' to clone does not exist, ignoring.", d);
625 return -ENXIO;
626 }
627
628 return log_debug_errno(errno, "Failed to stat() device node '%s' to clone, ignoring: %m", d);
629 }
630
631 if (!S_ISBLK(st.st_mode) &&
632 !S_ISCHR(st.st_mode))
633 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL),
634 "Device node '%s' to clone is not a device node, ignoring.",
635 d);
636
637 dn = strjoina(temporary_mount, d);
638
639 /* First, try to create device node properly */
640 if (*make_devnode) {
641 mac_selinux_create_file_prepare(d, st.st_mode);
642 r = mknod(dn, st.st_mode, st.st_rdev);
643 mac_selinux_create_file_clear();
644 if (r >= 0)
645 goto add_symlink;
646 if (errno != EPERM)
647 return log_debug_errno(errno, "mknod failed for %s: %m", d);
648
649 /* This didn't work, let's not try this again for the next iterations. */
650 *make_devnode = false;
651 }
652
653 /* We're about to fall back to bind-mounting the device
654 * node. So create a dummy bind-mount target.
655 * Do not prepare device-node SELinux label (see issue 13762) */
656 r = mknod(dn, S_IFREG, 0);
657 if (r < 0 && errno != EEXIST)
658 return log_debug_errno(errno, "mknod() fallback failed for '%s': %m", d);
659
660 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
661 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
662 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
663 r = mount_nofollow_verbose(LOG_DEBUG, d, dn, NULL, MS_BIND, NULL);
664 if (r < 0)
665 return r;
666
667 add_symlink:
668 bn = path_startswith(d, "/dev/");
669 if (!bn)
670 return 0;
671
672 /* Create symlinks like /dev/char/1:9 → ../urandom */
673 if (asprintf(&sl, "%s/dev/%s/%u:%u",
674 temporary_mount,
675 S_ISCHR(st.st_mode) ? "char" : "block",
676 major(st.st_rdev), minor(st.st_rdev)) < 0)
677 return log_oom();
678
679 (void) mkdir_parents(sl, 0755);
680
681 t = strjoina("../", bn);
682 if (symlink(t, sl) < 0)
683 log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl);
684
685 return 0;
686 }
687
688 static int mount_private_dev(MountEntry *m) {
689 static const char devnodes[] =
690 "/dev/null\0"
691 "/dev/zero\0"
692 "/dev/full\0"
693 "/dev/random\0"
694 "/dev/urandom\0"
695 "/dev/tty\0";
696
697 char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
698 const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
699 bool can_mknod = true;
700 _cleanup_umask_ mode_t u;
701 int r;
702
703 assert(m);
704
705 u = umask(0000);
706
707 if (!mkdtemp(temporary_mount))
708 return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount);
709
710 dev = strjoina(temporary_mount, "/dev");
711 (void) mkdir(dev, 0755);
712 r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=755" TMPFS_LIMITS_DEV);
713 if (r < 0)
714 goto fail;
715
716 r = label_fix_container(dev, "/dev", 0);
717 if (r < 0) {
718 log_debug_errno(errno, "Failed to fix label of '%s' as /dev: %m", dev);
719 goto fail;
720 }
721
722 devpts = strjoina(temporary_mount, "/dev/pts");
723 (void) mkdir(devpts, 0755);
724 r = mount_nofollow_verbose(LOG_DEBUG, "/dev/pts", devpts, NULL, MS_BIND, NULL);
725 if (r < 0)
726 goto fail;
727
728 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
729 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
730 * Thus, in that case make a clone.
731 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
732 r = is_symlink("/dev/ptmx");
733 if (r < 0) {
734 log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
735 goto fail;
736 } else if (r > 0) {
737 devptmx = strjoina(temporary_mount, "/dev/ptmx");
738 if (symlink("pts/ptmx", devptmx) < 0) {
739 r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx);
740 goto fail;
741 }
742 } else {
743 r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod);
744 if (r < 0)
745 goto fail;
746 }
747
748 devshm = strjoina(temporary_mount, "/dev/shm");
749 (void) mkdir(devshm, 0755);
750 r = mount_nofollow_verbose(LOG_DEBUG, "/dev/shm", devshm, NULL, MS_BIND, NULL);
751 if (r < 0)
752 goto fail;
753
754 devmqueue = strjoina(temporary_mount, "/dev/mqueue");
755 (void) mkdir(devmqueue, 0755);
756 (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/mqueue", devmqueue, NULL, MS_BIND, NULL);
757
758 devhugepages = strjoina(temporary_mount, "/dev/hugepages");
759 (void) mkdir(devhugepages, 0755);
760 (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/hugepages", devhugepages, NULL, MS_BIND, NULL);
761
762 devlog = strjoina(temporary_mount, "/dev/log");
763 if (symlink("/run/systemd/journal/dev-log", devlog) < 0)
764 log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog);
765
766 NULSTR_FOREACH(d, devnodes) {
767 r = clone_device_node(d, temporary_mount, &can_mknod);
768 /* ENXIO means the *source* is not a device file, skip creation in that case */
769 if (r < 0 && r != -ENXIO)
770 goto fail;
771 }
772
773 r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID);
774 if (r < 0)
775 log_debug_errno(r, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount);
776
777 /* Create the /dev directory if missing. It is more likely to be
778 * missing when the service is started with RootDirectory. This is
779 * consistent with mount units creating the mount points when missing.
780 */
781 (void) mkdir_p_label(mount_entry_path(m), 0755);
782
783 /* Unmount everything in old /dev */
784 r = umount_recursive(mount_entry_path(m), 0);
785 if (r < 0)
786 log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m));
787
788 r = mount_nofollow_verbose(LOG_DEBUG, dev, mount_entry_path(m), NULL, MS_MOVE, NULL);
789 if (r < 0)
790 goto fail;
791
792 (void) rmdir(dev);
793 (void) rmdir(temporary_mount);
794
795 return 0;
796
797 fail:
798 if (devpts)
799 (void) umount_verbose(LOG_DEBUG, devpts, UMOUNT_NOFOLLOW);
800
801 if (devshm)
802 (void) umount_verbose(LOG_DEBUG, devshm, UMOUNT_NOFOLLOW);
803
804 if (devhugepages)
805 (void) umount_verbose(LOG_DEBUG, devhugepages, UMOUNT_NOFOLLOW);
806
807 if (devmqueue)
808 (void) umount_verbose(LOG_DEBUG, devmqueue, UMOUNT_NOFOLLOW);
809
810 (void) umount_verbose(LOG_DEBUG, dev, UMOUNT_NOFOLLOW);
811 (void) rmdir(dev);
812 (void) rmdir(temporary_mount);
813
814 return r;
815 }
816
817 static int mount_bind_dev(const MountEntry *m) {
818 int r;
819
820 assert(m);
821
822 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
823 * /dev. This is only used when RootDirectory= is set. */
824
825 (void) mkdir_p_label(mount_entry_path(m), 0755);
826
827 r = path_is_mount_point(mount_entry_path(m), NULL, 0);
828 if (r < 0)
829 return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m");
830 if (r > 0) /* make this a NOP if /dev is already a mount point */
831 return 0;
832
833 r = mount_nofollow_verbose(LOG_DEBUG, "/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL);
834 if (r < 0)
835 return r;
836
837 return 1;
838 }
839
840 static int mount_sysfs(const MountEntry *m) {
841 int r;
842
843 assert(m);
844
845 (void) mkdir_p_label(mount_entry_path(m), 0755);
846
847 r = path_is_mount_point(mount_entry_path(m), NULL, 0);
848 if (r < 0)
849 return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m");
850 if (r > 0) /* make this a NOP if /sys is already a mount point */
851 return 0;
852
853 /* Bind mount the host's version so that we get all child mounts of it, too. */
854 r = mount_nofollow_verbose(LOG_DEBUG, "/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL);
855 if (r < 0)
856 return r;
857
858 return 1;
859 }
860
861 static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) {
862 const char *entry_path;
863 int r;
864
865 assert(m);
866 assert(ns_info);
867
868 entry_path = mount_entry_path(m);
869
870 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
871 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
872 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
873 * mounted on /proc/ first. */
874
875 (void) mkdir_p_label(entry_path, 0755);
876 (void) umount_recursive(entry_path, 0);
877
878 if (ns_info->protect_proc != PROTECT_PROC_DEFAULT ||
879 ns_info->proc_subset != PROC_SUBSET_ALL) {
880 _cleanup_free_ char *opts = NULL;
881
882 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
883 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
884 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
885 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
886 * added in the same commit: if it's supported it is thus also per-instance. */
887
888 opts = strjoin("hidepid=",
889 ns_info->protect_proc == PROTECT_PROC_DEFAULT ? "off" :
890 protect_proc_to_string(ns_info->protect_proc),
891 ns_info->proc_subset == PROC_SUBSET_PID ? ",subset=pid" : "");
892 if (!opts)
893 return -ENOMEM;
894
895 r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
896 if (r < 0) {
897 if (r != -EINVAL)
898 return r;
899
900 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
901 * not supported by the kernel, and thus the per-instance hidepid= neither, which
902 * means we really don't want to use it, since it would affect our host's /proc
903 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
904 } else
905 return 1;
906 }
907
908 r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
909 if (r < 0)
910 return r;
911
912 return 1;
913 }
914
915 static int mount_tmpfs(const MountEntry *m) {
916 const char *entry_path, *inner_path;
917 int r;
918
919 assert(m);
920
921 entry_path = mount_entry_path(m);
922 inner_path = m->path_const;
923
924 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
925
926 (void) mkdir_p_label(entry_path, 0755);
927 (void) umount_recursive(entry_path, 0);
928
929 r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", entry_path, "tmpfs", m->flags, mount_entry_options(m));
930 if (r < 0)
931 return r;
932
933 r = label_fix_container(entry_path, inner_path, 0);
934 if (r < 0)
935 return log_debug_errno(r, "Failed to fix label of '%s' as '%s': %m", entry_path, inner_path);
936
937 return 1;
938 }
939
940 static int mount_images(const MountEntry *m) {
941 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
942 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
943 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
944 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
945 DissectImageFlags dissect_image_flags;
946 int r;
947
948 assert(m);
949
950 r = verity_settings_load(&verity, mount_entry_source(m), NULL, NULL);
951 if (r < 0)
952 return log_debug_errno(r, "Failed to load root hash: %m");
953
954 dissect_image_flags =
955 (m->read_only ? DISSECT_IMAGE_READ_ONLY : 0) |
956 (verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0);
957
958 r = loop_device_make_by_path(
959 mount_entry_source(m),
960 m->read_only ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */,
961 verity.data_path ? 0 : LO_FLAGS_PARTSCAN,
962 &loop_device);
963 if (r < 0)
964 return log_debug_errno(r, "Failed to create loop device for image: %m");
965
966 r = dissect_image(
967 loop_device->fd,
968 &verity,
969 m->image_options,
970 dissect_image_flags,
971 &dissected_image);
972 /* No partition table? Might be a single-filesystem image, try again */
973 if (!verity.data_path && r == -ENOPKG)
974 r = dissect_image(
975 loop_device->fd,
976 &verity,
977 m->image_options,
978 dissect_image_flags|DISSECT_IMAGE_NO_PARTITION_TABLE,
979 &dissected_image);
980 if (r < 0)
981 return log_debug_errno(r, "Failed to dissect image: %m");
982
983 r = dissected_image_decrypt(
984 dissected_image,
985 NULL,
986 &verity,
987 dissect_image_flags,
988 &decrypted_image);
989 if (r < 0)
990 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
991
992 r = mkdir_p_label(mount_entry_path(m), 0755);
993 if (r < 0)
994 return log_debug_errno(r, "Failed to create destination directory %s: %m", mount_entry_path(m));
995 r = umount_recursive(mount_entry_path(m), 0);
996 if (r < 0)
997 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", mount_entry_path(m));
998
999 r = dissected_image_mount(dissected_image, mount_entry_path(m), UID_INVALID, dissect_image_flags);
1000 if (r < 0)
1001 return log_debug_errno(r, "Failed to mount image: %m");
1002
1003 if (decrypted_image) {
1004 r = decrypted_image_relinquish(decrypted_image);
1005 if (r < 0)
1006 return log_debug_errno(r, "Failed to relinquish decrypted image: %m");
1007 }
1008
1009 loop_device_relinquish(loop_device);
1010
1011 return 1;
1012 }
1013
1014 static int follow_symlink(
1015 const char *root_directory,
1016 MountEntry *m) {
1017
1018 _cleanup_free_ char *target = NULL;
1019 int r;
1020
1021 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1022 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1023 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1024 * end and already have a fully normalized name. */
1025
1026 r = chase_symlinks(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL);
1027 if (r < 0)
1028 return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m));
1029 if (r > 0) /* Reached the end, nothing more to resolve */
1030 return 1;
1031
1032 if (m->n_followed >= CHASE_SYMLINKS_MAX) /* put a boundary on things */
1033 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP),
1034 "Symlink loop on '%s'.",
1035 mount_entry_path(m));
1036
1037 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m), target);
1038
1039 free_and_replace(m->path_malloc, target);
1040 m->has_prefix = true;
1041
1042 m->n_followed ++;
1043
1044 return 0;
1045 }
1046
1047 static int apply_mount(
1048 const char *root_directory,
1049 MountEntry *m,
1050 const NamespaceInfo *ns_info) {
1051
1052 _cleanup_free_ char *inaccessible = NULL;
1053 bool rbind = true, make = false;
1054 const char *what;
1055 int r;
1056
1057 assert(m);
1058 assert(ns_info);
1059
1060 log_debug("Applying namespace mount on %s", mount_entry_path(m));
1061
1062 switch (m->mode) {
1063
1064 case INACCESSIBLE: {
1065 _cleanup_free_ char *tmp = NULL;
1066 const char *runtime_dir;
1067 struct stat target;
1068
1069 /* First, get rid of everything that is below if there
1070 * is anything... Then, overmount it with an
1071 * inaccessible path. */
1072 (void) umount_recursive(mount_entry_path(m), 0);
1073
1074 if (lstat(mount_entry_path(m), &target) < 0) {
1075 if (errno == ENOENT && m->ignore)
1076 return 0;
1077
1078 return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m",
1079 mount_entry_path(m));
1080 }
1081
1082 if (geteuid() == 0)
1083 runtime_dir = "/run";
1084 else {
1085 if (asprintf(&tmp, "/run/user/" UID_FMT, geteuid()) < 0)
1086 return -ENOMEM;
1087
1088 runtime_dir = tmp;
1089 }
1090
1091 r = mode_to_inaccessible_node(runtime_dir, target.st_mode, &inaccessible);
1092 if (r < 0)
1093 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP),
1094 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1095 what = inaccessible;
1096 break;
1097 }
1098
1099 case READONLY:
1100 case READWRITE:
1101 case READWRITE_IMPLICIT:
1102 r = path_is_mount_point(mount_entry_path(m), root_directory, 0);
1103 if (r == -ENOENT && m->ignore)
1104 return 0;
1105 if (r < 0)
1106 return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m",
1107 mount_entry_path(m));
1108 if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1109 * bit for the mount point if needed. */
1110 return 0;
1111 /* This isn't a mount point yet, let's make it one. */
1112 what = mount_entry_path(m);
1113 break;
1114
1115 case BIND_MOUNT:
1116 rbind = false;
1117
1118 _fallthrough_;
1119 case BIND_MOUNT_RECURSIVE: {
1120 _cleanup_free_ char *chased = NULL;
1121
1122 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1123 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1124 * root directory to chase_symlinks() here. */
1125
1126 r = chase_symlinks(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL);
1127 if (r == -ENOENT && m->ignore) {
1128 log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m));
1129 return 0;
1130 }
1131 if (r < 0)
1132 return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m));
1133
1134 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m), chased);
1135
1136 free_and_replace(m->source_malloc, chased);
1137
1138 what = mount_entry_source(m);
1139 make = true;
1140 break;
1141 }
1142
1143 case EMPTY_DIR:
1144 case TMPFS:
1145 return mount_tmpfs(m);
1146
1147 case PRIVATE_TMP:
1148 case PRIVATE_TMP_READONLY:
1149 what = mount_entry_source(m);
1150 make = true;
1151 break;
1152
1153 case PRIVATE_DEV:
1154 return mount_private_dev(m);
1155
1156 case BIND_DEV:
1157 return mount_bind_dev(m);
1158
1159 case SYSFS:
1160 return mount_sysfs(m);
1161
1162 case PROCFS:
1163 return mount_procfs(m, ns_info);
1164
1165 case MOUNT_IMAGES:
1166 return mount_images(m);
1167
1168 default:
1169 assert_not_reached("Unknown mode");
1170 }
1171
1172 assert(what);
1173
1174 r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL);
1175 if (r < 0) {
1176 bool try_again = false;
1177
1178 if (r == -ENOENT && make) {
1179 struct stat st;
1180
1181 /* Hmm, either the source or the destination are missing. Let's see if we can create
1182 the destination, then try again. */
1183
1184 if (stat(what, &st) < 0)
1185 log_error_errno(errno, "Mount point source '%s' is not accessible: %m", what);
1186 else {
1187 int q;
1188
1189 (void) mkdir_parents(mount_entry_path(m), 0755);
1190
1191 if (S_ISDIR(st.st_mode))
1192 q = mkdir(mount_entry_path(m), 0755) < 0 ? -errno : 0;
1193 else
1194 q = touch(mount_entry_path(m));
1195
1196 if (q < 0)
1197 log_error_errno(q, "Failed to create destination mount point node '%s': %m",
1198 mount_entry_path(m));
1199 else
1200 try_again = true;
1201 }
1202 }
1203
1204 if (try_again)
1205 r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL);
1206 if (r < 0)
1207 return log_error_errno(r, "Failed to mount %s to %s: %m", what, mount_entry_path(m));
1208 }
1209
1210 log_debug("Successfully mounted %s to %s", what, mount_entry_path(m));
1211 return 0;
1212 }
1213
1214 static int make_read_only(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) {
1215 unsigned long new_flags = 0, flags_mask = 0;
1216 bool submounts = false;
1217 int r = 0;
1218
1219 assert(m);
1220 assert(proc_self_mountinfo);
1221
1222 if (mount_entry_read_only(m) || m->mode == PRIVATE_DEV) {
1223 new_flags |= MS_RDONLY;
1224 flags_mask |= MS_RDONLY;
1225 }
1226
1227 if (m->nosuid) {
1228 new_flags |= MS_NOSUID;
1229 flags_mask |= MS_NOSUID;
1230 }
1231
1232 if (flags_mask == 0) /* No Change? */
1233 return 0;
1234
1235 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1236 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1237 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1238 * and running Linux <= 4.17. */
1239 submounts =
1240 mount_entry_read_only(m) &&
1241 !IN_SET(m->mode, EMPTY_DIR, TMPFS);
1242 if (submounts)
1243 r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo);
1244 else
1245 r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo);
1246
1247 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1248 * read-only already stays this way. This improves compatibility with container managers, where we
1249 * won't attempt to undo read-only mounts already applied. */
1250
1251 if (r == -ENOENT && m->ignore)
1252 return 0;
1253 if (r < 0)
1254 return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m),
1255 submounts ? " and its submounts" : "");
1256 return 0;
1257 }
1258
1259 static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) {
1260 assert(ns_info);
1261
1262 /*
1263 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1264 * since to protect the API VFS mounts, they need to be around in the
1265 * first place...
1266 */
1267
1268 return ns_info->mount_apivfs ||
1269 ns_info->protect_control_groups ||
1270 ns_info->protect_kernel_tunables ||
1271 ns_info->protect_proc != PROTECT_PROC_DEFAULT ||
1272 ns_info->proc_subset != PROC_SUBSET_ALL;
1273 }
1274
1275 static size_t namespace_calculate_mounts(
1276 const NamespaceInfo *ns_info,
1277 char** read_write_paths,
1278 char** read_only_paths,
1279 char** inaccessible_paths,
1280 char** empty_directories,
1281 size_t n_bind_mounts,
1282 size_t n_temporary_filesystems,
1283 size_t n_mount_images,
1284 const char* tmp_dir,
1285 const char* var_tmp_dir,
1286 const char *creds_path,
1287 const char* log_namespace) {
1288
1289 size_t protect_home_cnt;
1290 size_t protect_system_cnt =
1291 (ns_info->protect_system == PROTECT_SYSTEM_STRICT ?
1292 ELEMENTSOF(protect_system_strict_table) :
1293 ((ns_info->protect_system == PROTECT_SYSTEM_FULL) ?
1294 ELEMENTSOF(protect_system_full_table) :
1295 ((ns_info->protect_system == PROTECT_SYSTEM_YES) ?
1296 ELEMENTSOF(protect_system_yes_table) : 0)));
1297
1298 protect_home_cnt =
1299 (ns_info->protect_home == PROTECT_HOME_YES ?
1300 ELEMENTSOF(protect_home_yes_table) :
1301 ((ns_info->protect_home == PROTECT_HOME_READ_ONLY) ?
1302 ELEMENTSOF(protect_home_read_only_table) :
1303 ((ns_info->protect_home == PROTECT_HOME_TMPFS) ?
1304 ELEMENTSOF(protect_home_tmpfs_table) : 0)));
1305
1306 return !!tmp_dir + !!var_tmp_dir +
1307 strv_length(read_write_paths) +
1308 strv_length(read_only_paths) +
1309 strv_length(inaccessible_paths) +
1310 strv_length(empty_directories) +
1311 n_bind_mounts +
1312 n_mount_images +
1313 n_temporary_filesystems +
1314 ns_info->private_dev +
1315 (ns_info->protect_kernel_tunables ? ELEMENTSOF(protect_kernel_tunables_table) : 0) +
1316 (ns_info->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) +
1317 (ns_info->protect_kernel_logs ? ELEMENTSOF(protect_kernel_logs_table) : 0) +
1318 (ns_info->protect_control_groups ? 1 : 0) +
1319 protect_home_cnt + protect_system_cnt +
1320 (ns_info->protect_hostname ? 2 : 0) +
1321 (namespace_info_mount_apivfs(ns_info) ? ELEMENTSOF(apivfs_table) : 0) +
1322 (creds_path ? 2 : 1) +
1323 !!log_namespace;
1324 }
1325
1326 static void normalize_mounts(const char *root_directory, MountEntry *mounts, size_t *n_mounts) {
1327 assert(root_directory);
1328 assert(n_mounts);
1329 assert(mounts || *n_mounts == 0);
1330
1331 typesafe_qsort(mounts, *n_mounts, mount_path_compare);
1332
1333 drop_duplicates(mounts, n_mounts);
1334 drop_outside_root(root_directory, mounts, n_mounts);
1335 drop_inaccessible(mounts, n_mounts);
1336 drop_nop(mounts, n_mounts);
1337 }
1338
1339 static bool root_read_only(
1340 char **read_only_paths,
1341 ProtectSystem protect_system) {
1342
1343 /* Determine whether the root directory is going to be read-only given the configured settings. */
1344
1345 if (protect_system == PROTECT_SYSTEM_STRICT)
1346 return true;
1347
1348 if (prefixed_path_strv_contains(read_only_paths, "/"))
1349 return true;
1350
1351 return false;
1352 }
1353
1354 static bool home_read_only(
1355 char** read_only_paths,
1356 char** inaccessible_paths,
1357 char** empty_directories,
1358 const BindMount *bind_mounts,
1359 size_t n_bind_mounts,
1360 const TemporaryFileSystem *temporary_filesystems,
1361 size_t n_temporary_filesystems,
1362 ProtectHome protect_home) {
1363
1364 size_t i;
1365
1366 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1367 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1368 * settings. */
1369
1370 if (protect_home != PROTECT_HOME_NO)
1371 return true;
1372
1373 if (prefixed_path_strv_contains(read_only_paths, "/home") ||
1374 prefixed_path_strv_contains(inaccessible_paths, "/home") ||
1375 prefixed_path_strv_contains(empty_directories, "/home"))
1376 return true;
1377
1378 for (i = 0; i < n_temporary_filesystems; i++)
1379 if (path_equal(temporary_filesystems[i].path, "/home"))
1380 return true;
1381
1382 /* If /home is overmounted with some dir from the host it's not writable. */
1383 for (i = 0; i < n_bind_mounts; i++)
1384 if (path_equal(bind_mounts[i].destination, "/home"))
1385 return true;
1386
1387 return false;
1388 }
1389
1390 static int verity_settings_prepare(
1391 VeritySettings *verity,
1392 const char *root_image,
1393 const void *root_hash,
1394 size_t root_hash_size,
1395 const char *root_hash_path,
1396 const void *root_hash_sig,
1397 size_t root_hash_sig_size,
1398 const char *root_hash_sig_path,
1399 const char *verity_data_path) {
1400
1401 int r;
1402
1403 assert(verity);
1404
1405 if (root_hash) {
1406 void *d;
1407
1408 d = memdup(root_hash, root_hash_size);
1409 if (!d)
1410 return -ENOMEM;
1411
1412 free_and_replace(verity->root_hash, d);
1413 verity->root_hash_size = root_hash_size;
1414 verity->designator = PARTITION_ROOT;
1415 }
1416
1417 if (root_hash_sig) {
1418 void *d;
1419
1420 d = memdup(root_hash_sig, root_hash_sig_size);
1421 if (!d)
1422 return -ENOMEM;
1423
1424 free_and_replace(verity->root_hash_sig, d);
1425 verity->root_hash_sig_size = root_hash_sig_size;
1426 verity->designator = PARTITION_ROOT;
1427 }
1428
1429 if (verity_data_path) {
1430 r = free_and_strdup(&verity->data_path, verity_data_path);
1431 if (r < 0)
1432 return r;
1433 }
1434
1435 r = verity_settings_load(
1436 verity,
1437 root_image,
1438 root_hash_path,
1439 root_hash_sig_path);
1440 if (r < 0)
1441 return log_debug_errno(r, "Failed to load root hash: %m");
1442
1443 return 0;
1444 }
1445
1446 int setup_namespace(
1447 const char* root_directory,
1448 const char* root_image,
1449 const MountOptions *root_image_options,
1450 const NamespaceInfo *ns_info,
1451 char** read_write_paths,
1452 char** read_only_paths,
1453 char** inaccessible_paths,
1454 char** empty_directories,
1455 const BindMount *bind_mounts,
1456 size_t n_bind_mounts,
1457 const TemporaryFileSystem *temporary_filesystems,
1458 size_t n_temporary_filesystems,
1459 const MountImage *mount_images,
1460 size_t n_mount_images,
1461 const char* tmp_dir,
1462 const char* var_tmp_dir,
1463 const char *creds_path,
1464 const char *log_namespace,
1465 unsigned long mount_flags,
1466 const void *root_hash,
1467 size_t root_hash_size,
1468 const char *root_hash_path,
1469 const void *root_hash_sig,
1470 size_t root_hash_sig_size,
1471 const char *root_hash_sig_path,
1472 const char *verity_data_path,
1473 DissectImageFlags dissect_image_flags,
1474 char **error_path) {
1475
1476 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
1477 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
1478 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
1479 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
1480 MountEntry *m = NULL, *mounts = NULL;
1481 bool require_prefix = false;
1482 const char *root;
1483 size_t n_mounts;
1484 int r;
1485
1486 assert(ns_info);
1487
1488 if (mount_flags == 0)
1489 mount_flags = MS_SHARED;
1490
1491 if (root_image) {
1492 dissect_image_flags |= DISSECT_IMAGE_REQUIRE_ROOT;
1493
1494 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
1495 if (root_read_only(read_only_paths,
1496 ns_info->protect_system) &&
1497 home_read_only(read_only_paths, inaccessible_paths, empty_directories,
1498 bind_mounts, n_bind_mounts, temporary_filesystems, n_temporary_filesystems,
1499 ns_info->protect_home) &&
1500 strv_isempty(read_write_paths))
1501 dissect_image_flags |= DISSECT_IMAGE_READ_ONLY;
1502
1503 r = verity_settings_prepare(
1504 &verity,
1505 root_image,
1506 root_hash, root_hash_size, root_hash_path,
1507 root_hash_sig, root_hash_sig_size, root_hash_sig_path,
1508 verity_data_path);
1509 if (r < 0)
1510 return r;
1511
1512 SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, verity.data_path);
1513
1514 r = loop_device_make_by_path(
1515 root_image,
1516 FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */,
1517 FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
1518 &loop_device);
1519 if (r < 0)
1520 return log_debug_errno(r, "Failed to create loop device for root image: %m");
1521
1522 r = dissect_image(
1523 loop_device->fd,
1524 &verity,
1525 root_image_options,
1526 dissect_image_flags,
1527 &dissected_image);
1528 if (r < 0)
1529 return log_debug_errno(r, "Failed to dissect image: %m");
1530
1531 r = dissected_image_decrypt(
1532 dissected_image,
1533 NULL,
1534 &verity,
1535 dissect_image_flags,
1536 &decrypted_image);
1537 if (r < 0)
1538 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
1539 }
1540
1541 if (root_directory)
1542 root = root_directory;
1543 else {
1544 /* Always create the mount namespace in a temporary directory, instead of operating
1545 * directly in the root. The temporary directory prevents any mounts from being
1546 * potentially obscured my other mounts we already applied.
1547 * We use the same mount point for all images, which is safe, since they all live
1548 * in their own namespaces after all, and hence won't see each other. */
1549
1550 root = "/run/systemd/unit-root";
1551 (void) mkdir_label(root, 0700);
1552 require_prefix = true;
1553 }
1554
1555 n_mounts = namespace_calculate_mounts(
1556 ns_info,
1557 read_write_paths,
1558 read_only_paths,
1559 inaccessible_paths,
1560 empty_directories,
1561 n_bind_mounts,
1562 n_temporary_filesystems,
1563 n_mount_images,
1564 tmp_dir, var_tmp_dir,
1565 creds_path,
1566 log_namespace);
1567
1568 if (n_mounts > 0) {
1569 m = mounts = new0(MountEntry, n_mounts);
1570 if (!mounts)
1571 return -ENOMEM;
1572
1573 r = append_access_mounts(&m, read_write_paths, READWRITE, require_prefix);
1574 if (r < 0)
1575 goto finish;
1576
1577 r = append_access_mounts(&m, read_only_paths, READONLY, require_prefix);
1578 if (r < 0)
1579 goto finish;
1580
1581 r = append_access_mounts(&m, inaccessible_paths, INACCESSIBLE, require_prefix);
1582 if (r < 0)
1583 goto finish;
1584
1585 r = append_empty_dir_mounts(&m, empty_directories);
1586 if (r < 0)
1587 goto finish;
1588
1589 r = append_bind_mounts(&m, bind_mounts, n_bind_mounts);
1590 if (r < 0)
1591 goto finish;
1592
1593 r = append_tmpfs_mounts(&m, temporary_filesystems, n_temporary_filesystems);
1594 if (r < 0)
1595 goto finish;
1596
1597 if (tmp_dir) {
1598 bool ro = streq(tmp_dir, RUN_SYSTEMD_EMPTY);
1599
1600 *(m++) = (MountEntry) {
1601 .path_const = "/tmp",
1602 .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP,
1603 .source_const = tmp_dir,
1604 };
1605 }
1606
1607 if (var_tmp_dir) {
1608 bool ro = streq(var_tmp_dir, RUN_SYSTEMD_EMPTY);
1609
1610 *(m++) = (MountEntry) {
1611 .path_const = "/var/tmp",
1612 .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP,
1613 .source_const = var_tmp_dir,
1614 };
1615 }
1616
1617 r = append_mount_images(&m, mount_images, n_mount_images);
1618 if (r < 0)
1619 goto finish;
1620
1621 if (ns_info->private_dev) {
1622 *(m++) = (MountEntry) {
1623 .path_const = "/dev",
1624 .mode = PRIVATE_DEV,
1625 .flags = DEV_MOUNT_OPTIONS,
1626 };
1627 }
1628
1629 if (ns_info->protect_kernel_tunables) {
1630 r = append_static_mounts(&m,
1631 protect_kernel_tunables_table,
1632 ELEMENTSOF(protect_kernel_tunables_table),
1633 ns_info->ignore_protect_paths);
1634 if (r < 0)
1635 goto finish;
1636 }
1637
1638 if (ns_info->protect_kernel_modules) {
1639 r = append_static_mounts(&m,
1640 protect_kernel_modules_table,
1641 ELEMENTSOF(protect_kernel_modules_table),
1642 ns_info->ignore_protect_paths);
1643 if (r < 0)
1644 goto finish;
1645 }
1646
1647 if (ns_info->protect_kernel_logs) {
1648 r = append_static_mounts(&m,
1649 protect_kernel_logs_table,
1650 ELEMENTSOF(protect_kernel_logs_table),
1651 ns_info->ignore_protect_paths);
1652 if (r < 0)
1653 goto finish;
1654 }
1655
1656 if (ns_info->protect_control_groups) {
1657 *(m++) = (MountEntry) {
1658 .path_const = "/sys/fs/cgroup",
1659 .mode = READONLY,
1660 };
1661 }
1662
1663 r = append_protect_home(&m, ns_info->protect_home, ns_info->ignore_protect_paths);
1664 if (r < 0)
1665 goto finish;
1666
1667 r = append_protect_system(&m, ns_info->protect_system, false);
1668 if (r < 0)
1669 goto finish;
1670
1671 if (namespace_info_mount_apivfs(ns_info)) {
1672 r = append_static_mounts(&m,
1673 apivfs_table,
1674 ELEMENTSOF(apivfs_table),
1675 ns_info->ignore_protect_paths);
1676 if (r < 0)
1677 goto finish;
1678 }
1679
1680 if (ns_info->protect_hostname) {
1681 *(m++) = (MountEntry) {
1682 .path_const = "/proc/sys/kernel/hostname",
1683 .mode = READONLY,
1684 };
1685 *(m++) = (MountEntry) {
1686 .path_const = "/proc/sys/kernel/domainname",
1687 .mode = READONLY,
1688 };
1689 }
1690
1691 if (creds_path) {
1692 /* If our service has a credentials store configured, then bind that one in, but hide
1693 * everything else. */
1694
1695 *(m++) = (MountEntry) {
1696 .path_const = "/run/credentials",
1697 .mode = TMPFS,
1698 .read_only = true,
1699 .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST,
1700 .flags = MS_NODEV|MS_STRICTATIME|MS_NOSUID|MS_NOEXEC,
1701 };
1702
1703 *(m++) = (MountEntry) {
1704 .path_const = creds_path,
1705 .mode = BIND_MOUNT,
1706 .read_only = true,
1707 .source_const = creds_path,
1708 };
1709 } else {
1710 /* If our service has no credentials store configured, then make the whole
1711 * credentials tree inaccessible wholesale. */
1712
1713 *(m++) = (MountEntry) {
1714 .path_const = "/run/credentials",
1715 .mode = INACCESSIBLE,
1716 .ignore = true,
1717 };
1718 }
1719
1720 if (log_namespace) {
1721 _cleanup_free_ char *q;
1722
1723 q = strjoin("/run/systemd/journal.", log_namespace);
1724 if (!q) {
1725 r = -ENOMEM;
1726 goto finish;
1727 }
1728
1729 *(m++) = (MountEntry) {
1730 .path_const = "/run/systemd/journal",
1731 .mode = BIND_MOUNT_RECURSIVE,
1732 .read_only = true,
1733 .source_malloc = TAKE_PTR(q),
1734 };
1735 }
1736
1737 assert(mounts + n_mounts == m);
1738
1739 /* Prepend the root directory where that's necessary */
1740 r = prefix_where_needed(mounts, n_mounts, root);
1741 if (r < 0)
1742 goto finish;
1743
1744 normalize_mounts(root, mounts, &n_mounts);
1745 }
1746
1747 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
1748
1749 if (unshare(CLONE_NEWNS) < 0) {
1750 r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m");
1751 if (IN_SET(r, -EACCES, -EPERM, -EOPNOTSUPP, -ENOSYS))
1752 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
1753 * in place that doesn't allow us to create namespaces (or a missing cap), then
1754 * propagate a recognizable error back, which the caller can use to detect this case
1755 * (and only this) and optionally continue without namespacing applied. */
1756 r = -ENOANO;
1757
1758 goto finish;
1759 }
1760
1761 /* Remount / as SLAVE so that nothing now mounted in the namespace
1762 * shows up in the parent */
1763 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
1764 r = log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m");
1765 goto finish;
1766 }
1767
1768 if (root_image) {
1769 /* A root image is specified, mount it to the right place */
1770 r = dissected_image_mount(dissected_image, root, UID_INVALID, dissect_image_flags);
1771 if (r < 0) {
1772 log_debug_errno(r, "Failed to mount root image: %m");
1773 goto finish;
1774 }
1775
1776 if (decrypted_image) {
1777 r = decrypted_image_relinquish(decrypted_image);
1778 if (r < 0) {
1779 log_debug_errno(r, "Failed to relinquish decrypted image: %m");
1780 goto finish;
1781 }
1782 }
1783
1784 loop_device_relinquish(loop_device);
1785
1786 } else if (root_directory) {
1787
1788 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1789 r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW);
1790 if (r < 0) {
1791 log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root);
1792 goto finish;
1793 }
1794 if (r == 0) {
1795 r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL);
1796 if (r < 0)
1797 goto finish;
1798 }
1799
1800 } else {
1801 /* Let's mount the main root directory to the root directory to use */
1802 r = mount_nofollow_verbose(LOG_DEBUG, "/", root, NULL, MS_BIND|MS_REC, NULL);
1803 if (r < 0)
1804 goto finish;
1805 }
1806
1807 /* Try to set up the new root directory before mounting anything else there. */
1808 if (root_image || root_directory)
1809 (void) base_filesystem_create(root, UID_INVALID, GID_INVALID);
1810
1811 if (n_mounts > 0) {
1812 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
1813 _cleanup_free_ char **deny_list = NULL;
1814 size_t j;
1815
1816 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1817 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1818 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
1819 if (!proc_self_mountinfo) {
1820 r = log_debug_errno(errno, "Failed to open /proc/self/mountinfo: %m");
1821 if (error_path)
1822 *error_path = strdup("/proc/self/mountinfo");
1823 goto finish;
1824 }
1825
1826 /* First round, establish all mounts we need */
1827 for (;;) {
1828 bool again = false;
1829
1830 for (m = mounts; m < mounts + n_mounts; ++m) {
1831
1832 if (m->applied)
1833 continue;
1834
1835 r = follow_symlink(root, m);
1836 if (r < 0) {
1837 if (error_path && mount_entry_path(m))
1838 *error_path = strdup(mount_entry_path(m));
1839 goto finish;
1840 }
1841 if (r == 0) {
1842 /* We hit a symlinked mount point. The entry got rewritten and might
1843 * point to a very different place now. Let's normalize the changed
1844 * list, and start from the beginning. After all to mount the entry
1845 * at the new location we might need some other mounts first */
1846 again = true;
1847 break;
1848 }
1849
1850 r = apply_mount(root, m, ns_info);
1851 if (r < 0) {
1852 if (error_path && mount_entry_path(m))
1853 *error_path = strdup(mount_entry_path(m));
1854 goto finish;
1855 }
1856
1857 m->applied = true;
1858 }
1859
1860 if (!again)
1861 break;
1862
1863 normalize_mounts(root, mounts, &n_mounts);
1864 }
1865
1866 /* Create a deny list we can pass to bind_mount_recursive() */
1867 deny_list = new(char*, n_mounts+1);
1868 if (!deny_list) {
1869 r = -ENOMEM;
1870 goto finish;
1871 }
1872 for (j = 0; j < n_mounts; j++)
1873 deny_list[j] = (char*) mount_entry_path(mounts+j);
1874 deny_list[j] = NULL;
1875
1876 /* Second round, flip the ro bits if necessary. */
1877 for (m = mounts; m < mounts + n_mounts; ++m) {
1878 r = make_read_only(m, deny_list, proc_self_mountinfo);
1879 if (r < 0) {
1880 if (error_path && mount_entry_path(m))
1881 *error_path = strdup(mount_entry_path(m));
1882 goto finish;
1883 }
1884 }
1885 }
1886
1887 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1888 r = mount_move_root(root);
1889 if (r < 0) {
1890 log_debug_errno(r, "Failed to mount root with MS_MOVE: %m");
1891 goto finish;
1892 }
1893
1894 /* Remount / as the desired mode. Note that this will not
1895 * reestablish propagation from our side to the host, since
1896 * what's disconnected is disconnected. */
1897 if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
1898 r = log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m");
1899 goto finish;
1900 }
1901
1902 r = 0;
1903
1904 finish:
1905 if (n_mounts > 0)
1906 for (m = mounts; m < mounts + n_mounts; m++)
1907 mount_entry_done(m);
1908
1909 free(mounts);
1910
1911 return r;
1912 }
1913
1914 void bind_mount_free_many(BindMount *b, size_t n) {
1915 size_t i;
1916
1917 assert(b || n == 0);
1918
1919 for (i = 0; i < n; i++) {
1920 free(b[i].source);
1921 free(b[i].destination);
1922 }
1923
1924 free(b);
1925 }
1926
1927 int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) {
1928 _cleanup_free_ char *s = NULL, *d = NULL;
1929 BindMount *c;
1930
1931 assert(b);
1932 assert(n);
1933 assert(item);
1934
1935 s = strdup(item->source);
1936 if (!s)
1937 return -ENOMEM;
1938
1939 d = strdup(item->destination);
1940 if (!d)
1941 return -ENOMEM;
1942
1943 c = reallocarray(*b, *n + 1, sizeof(BindMount));
1944 if (!c)
1945 return -ENOMEM;
1946
1947 *b = c;
1948
1949 c[(*n) ++] = (BindMount) {
1950 .source = TAKE_PTR(s),
1951 .destination = TAKE_PTR(d),
1952 .read_only = item->read_only,
1953 .nosuid = item->nosuid,
1954 .recursive = item->recursive,
1955 .ignore_enoent = item->ignore_enoent,
1956 };
1957
1958 return 0;
1959 }
1960
1961 MountImage* mount_image_free_many(MountImage *m, size_t *n) {
1962 size_t i;
1963
1964 assert(n);
1965 assert(m || *n == 0);
1966
1967 for (i = 0; i < *n; i++) {
1968 free(m[i].source);
1969 free(m[i].destination);
1970 mount_options_free_all(m[i].mount_options);
1971 }
1972
1973 free(m);
1974 *n = 0;
1975 return NULL;
1976 }
1977
1978 int mount_image_add(MountImage **m, size_t *n, const MountImage *item) {
1979 _cleanup_free_ char *s = NULL, *d = NULL;
1980 _cleanup_(mount_options_free_allp) MountOptions *options = NULL;
1981 MountOptions *i;
1982 MountImage *c;
1983
1984 assert(m);
1985 assert(n);
1986 assert(item);
1987
1988 s = strdup(item->source);
1989 if (!s)
1990 return -ENOMEM;
1991
1992 d = strdup(item->destination);
1993 if (!d)
1994 return -ENOMEM;
1995
1996 LIST_FOREACH(mount_options, i, item->mount_options) {
1997 _cleanup_(mount_options_free_allp) MountOptions *o;
1998
1999 o = new(MountOptions, 1);
2000 if (!o)
2001 return -ENOMEM;
2002
2003 *o = (MountOptions) {
2004 .partition_designator = i->partition_designator,
2005 .options = strdup(i->options),
2006 };
2007 if (!o->options)
2008 return -ENOMEM;
2009
2010 LIST_APPEND(mount_options, options, TAKE_PTR(o));
2011 }
2012
2013 c = reallocarray(*m, *n + 1, sizeof(MountImage));
2014 if (!c)
2015 return -ENOMEM;
2016
2017 *m = c;
2018
2019 c[(*n) ++] = (MountImage) {
2020 .source = TAKE_PTR(s),
2021 .destination = TAKE_PTR(d),
2022 .mount_options = TAKE_PTR(options),
2023 .ignore_enoent = item->ignore_enoent,
2024 };
2025
2026 return 0;
2027 }
2028
2029 void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) {
2030 size_t i;
2031
2032 assert(t || n == 0);
2033
2034 for (i = 0; i < n; i++) {
2035 free(t[i].path);
2036 free(t[i].options);
2037 }
2038
2039 free(t);
2040 }
2041
2042 int temporary_filesystem_add(
2043 TemporaryFileSystem **t,
2044 size_t *n,
2045 const char *path,
2046 const char *options) {
2047
2048 _cleanup_free_ char *p = NULL, *o = NULL;
2049 TemporaryFileSystem *c;
2050
2051 assert(t);
2052 assert(n);
2053 assert(path);
2054
2055 p = strdup(path);
2056 if (!p)
2057 return -ENOMEM;
2058
2059 if (!isempty(options)) {
2060 o = strdup(options);
2061 if (!o)
2062 return -ENOMEM;
2063 }
2064
2065 c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem));
2066 if (!c)
2067 return -ENOMEM;
2068
2069 *t = c;
2070
2071 c[(*n) ++] = (TemporaryFileSystem) {
2072 .path = TAKE_PTR(p),
2073 .options = TAKE_PTR(o),
2074 };
2075
2076 return 0;
2077 }
2078
2079 static int make_tmp_prefix(const char *prefix) {
2080 _cleanup_free_ char *t = NULL;
2081 int r;
2082
2083 /* Don't do anything unless we know the dir is actually missing */
2084 r = access(prefix, F_OK);
2085 if (r >= 0)
2086 return 0;
2087 if (errno != ENOENT)
2088 return -errno;
2089
2090 r = mkdir_parents(prefix, 0755);
2091 if (r < 0)
2092 return r;
2093
2094 r = tempfn_random(prefix, NULL, &t);
2095 if (r < 0)
2096 return r;
2097
2098 if (mkdir(t, 0777) < 0)
2099 return -errno;
2100
2101 if (chmod(t, 01777) < 0) {
2102 r = -errno;
2103 (void) rmdir(t);
2104 return r;
2105 }
2106
2107 if (rename(t, prefix) < 0) {
2108 r = -errno;
2109 (void) rmdir(t);
2110 return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */
2111 }
2112
2113 return 0;
2114
2115 }
2116
2117 static int make_tmp_subdir(const char *parent, char **ret) {
2118 _cleanup_free_ char *y = NULL;
2119
2120 y = path_join(parent, "/tmp");
2121 if (!y)
2122 return -ENOMEM;
2123
2124 RUN_WITH_UMASK(0000) {
2125 if (mkdir(y, 0777 | S_ISVTX) < 0)
2126 return -errno;
2127 }
2128
2129 if (ret)
2130 *ret = TAKE_PTR(y);
2131 return 0;
2132 }
2133
2134 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path, char **tmp_path) {
2135 _cleanup_free_ char *x = NULL;
2136 char bid[SD_ID128_STRING_MAX];
2137 sd_id128_t boot_id;
2138 bool rw = true;
2139 int r;
2140
2141 assert(id);
2142 assert(prefix);
2143 assert(path);
2144
2145 /* We include the boot id in the directory so that after a
2146 * reboot we can easily identify obsolete directories. */
2147
2148 r = sd_id128_get_boot(&boot_id);
2149 if (r < 0)
2150 return r;
2151
2152 x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX");
2153 if (!x)
2154 return -ENOMEM;
2155
2156 r = make_tmp_prefix(prefix);
2157 if (r < 0)
2158 return r;
2159
2160 RUN_WITH_UMASK(0077)
2161 if (!mkdtemp(x)) {
2162 if (errno == EROFS || ERRNO_IS_DISK_SPACE(errno))
2163 rw = false;
2164 else
2165 return -errno;
2166 }
2167
2168 if (rw) {
2169 r = make_tmp_subdir(x, tmp_path);
2170 if (r < 0)
2171 return r;
2172 } else {
2173 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2174 * read-only. This way the service will get the EROFS result as if it was writing to the real
2175 * file system. */
2176 r = mkdir_p(RUN_SYSTEMD_EMPTY, 0500);
2177 if (r < 0)
2178 return r;
2179
2180 r = free_and_strdup(&x, RUN_SYSTEMD_EMPTY);
2181 if (r < 0)
2182 return r;
2183 }
2184
2185 *path = TAKE_PTR(x);
2186 return 0;
2187 }
2188
2189 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
2190 _cleanup_(namespace_cleanup_tmpdirp) char *a = NULL;
2191 _cleanup_(rmdir_and_freep) char *a_tmp = NULL;
2192 char *b;
2193 int r;
2194
2195 assert(id);
2196 assert(tmp_dir);
2197 assert(var_tmp_dir);
2198
2199 r = setup_one_tmp_dir(id, "/tmp", &a, &a_tmp);
2200 if (r < 0)
2201 return r;
2202
2203 r = setup_one_tmp_dir(id, "/var/tmp", &b, NULL);
2204 if (r < 0)
2205 return r;
2206
2207 a_tmp = mfree(a_tmp); /* avoid rmdir */
2208 *tmp_dir = TAKE_PTR(a);
2209 *var_tmp_dir = TAKE_PTR(b);
2210
2211 return 0;
2212 }
2213
2214 int setup_netns(const int netns_storage_socket[static 2]) {
2215 _cleanup_close_ int netns = -1;
2216 int r, q;
2217
2218 assert(netns_storage_socket);
2219 assert(netns_storage_socket[0] >= 0);
2220 assert(netns_storage_socket[1] >= 0);
2221
2222 /* We use the passed socketpair as a storage buffer for our
2223 * namespace reference fd. Whatever process runs this first
2224 * shall create a new namespace, all others should just join
2225 * it. To serialize that we use a file lock on the socket
2226 * pair.
2227 *
2228 * It's a bit crazy, but hey, works great! */
2229
2230 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
2231 return -errno;
2232
2233 netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
2234 if (netns == -EAGAIN) {
2235 /* Nothing stored yet, so let's create a new namespace. */
2236
2237 if (unshare(CLONE_NEWNET) < 0) {
2238 r = -errno;
2239 goto fail;
2240 }
2241
2242 (void) loopback_setup();
2243
2244 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
2245 if (netns < 0) {
2246 r = -errno;
2247 goto fail;
2248 }
2249
2250 r = 1;
2251
2252 } else if (netns < 0) {
2253 r = netns;
2254 goto fail;
2255
2256 } else {
2257 /* Yay, found something, so let's join the namespace */
2258 if (setns(netns, CLONE_NEWNET) < 0) {
2259 r = -errno;
2260 goto fail;
2261 }
2262
2263 r = 0;
2264 }
2265
2266 q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
2267 if (q < 0) {
2268 r = q;
2269 goto fail;
2270 }
2271
2272 fail:
2273 (void) lockf(netns_storage_socket[0], F_ULOCK, 0);
2274 return r;
2275 }
2276
2277 int open_netns_path(const int netns_storage_socket[static 2], const char *path) {
2278 _cleanup_close_ int netns = -1;
2279 int q, r;
2280
2281 assert(netns_storage_socket);
2282 assert(netns_storage_socket[0] >= 0);
2283 assert(netns_storage_socket[1] >= 0);
2284 assert(path);
2285
2286 /* If the storage socket doesn't contain a netns fd yet, open one via the file system and store it in
2287 * it. This is supposed to be called ahead of time, i.e. before setup_netns() which will allocate a
2288 * new anonymous netns if needed. */
2289
2290 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
2291 return -errno;
2292
2293 netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
2294 if (netns == -EAGAIN) {
2295 /* Nothing stored yet. Open the file from the file system. */
2296
2297 netns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
2298 if (netns < 0) {
2299 r = -errno;
2300 goto fail;
2301 }
2302
2303 r = fd_is_network_ns(netns);
2304 if (r == 0) { /* Not a netns? Refuse early. */
2305 r = -EINVAL;
2306 goto fail;
2307 }
2308 if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */
2309 goto fail;
2310
2311 r = 1;
2312
2313 } else if (netns < 0) {
2314 r = netns;
2315 goto fail;
2316 } else
2317 r = 0; /* Already allocated */
2318
2319 q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
2320 if (q < 0) {
2321 r = q;
2322 goto fail;
2323 }
2324
2325 fail:
2326 (void) lockf(netns_storage_socket[0], F_ULOCK, 0);
2327 return r;
2328 }
2329
2330 bool ns_type_supported(NamespaceType type) {
2331 const char *t, *ns_proc;
2332
2333 t = namespace_type_to_string(type);
2334 if (!t) /* Don't know how to translate this? Then it's not supported */
2335 return false;
2336
2337 ns_proc = strjoina("/proc/self/ns/", t);
2338 return access(ns_proc, F_OK) == 0;
2339 }
2340
2341 static const char *const protect_home_table[_PROTECT_HOME_MAX] = {
2342 [PROTECT_HOME_NO] = "no",
2343 [PROTECT_HOME_YES] = "yes",
2344 [PROTECT_HOME_READ_ONLY] = "read-only",
2345 [PROTECT_HOME_TMPFS] = "tmpfs",
2346 };
2347
2348 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES);
2349
2350 static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
2351 [PROTECT_SYSTEM_NO] = "no",
2352 [PROTECT_SYSTEM_YES] = "yes",
2353 [PROTECT_SYSTEM_FULL] = "full",
2354 [PROTECT_SYSTEM_STRICT] = "strict",
2355 };
2356
2357 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES);
2358
2359 static const char* const namespace_type_table[] = {
2360 [NAMESPACE_MOUNT] = "mnt",
2361 [NAMESPACE_CGROUP] = "cgroup",
2362 [NAMESPACE_UTS] = "uts",
2363 [NAMESPACE_IPC] = "ipc",
2364 [NAMESPACE_USER] = "user",
2365 [NAMESPACE_PID] = "pid",
2366 [NAMESPACE_NET] = "net",
2367 };
2368
2369 DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType);
2370
2371 static const char* const protect_proc_table[_PROTECT_PROC_MAX] = {
2372 [PROTECT_PROC_DEFAULT] = "default",
2373 [PROTECT_PROC_NOACCESS] = "noaccess",
2374 [PROTECT_PROC_INVISIBLE] = "invisible",
2375 [PROTECT_PROC_PTRACEABLE] = "ptraceable",
2376 };
2377
2378 DEFINE_STRING_TABLE_LOOKUP(protect_proc, ProtectProc);
2379
2380 static const char* const proc_subset_table[_PROC_SUBSET_MAX] = {
2381 [PROC_SUBSET_ALL] = "all",
2382 [PROC_SUBSET_PID] = "pid",
2383 };
2384
2385 DEFINE_STRING_TABLE_LOOKUP(proc_subset, ProcSubset);
Unnamed repository; edit this file 'description' to name the repository.