]> 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
core: introduce ProtectProc= and ProcSubset= to expose hidepid= and subset= procfs...
[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:
661 * The assumption here is that all used device nodes carry standard
662 * properties. Specifically, the devices nodes we bind-mount should
663 * either be owned by root:root or root:tty (e.g. /dev/tty, /dev/ptmx)
664 * and should not carry ACLs. */
665 if (mount(d, dn, NULL, MS_BIND, NULL) < 0)
666 return log_debug_errno(errno, "Bind mounting failed for '%s': %m", d);
667
668 add_symlink:
669 bn = path_startswith(d, "/dev/");
670 if (!bn)
671 return 0;
672
673 /* Create symlinks like /dev/char/1:9 → ../urandom */
674 if (asprintf(&sl, "%s/dev/%s/%u:%u",
675 temporary_mount,
676 S_ISCHR(st.st_mode) ? "char" : "block",
677 major(st.st_rdev), minor(st.st_rdev)) < 0)
678 return log_oom();
679
680 (void) mkdir_parents(sl, 0755);
681
682 t = strjoina("../", bn);
683 if (symlink(t, sl) < 0)
684 log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl);
685
686 return 0;
687 }
688
689 static int mount_private_dev(MountEntry *m) {
690 static const char devnodes[] =
691 "/dev/null\0"
692 "/dev/zero\0"
693 "/dev/full\0"
694 "/dev/random\0"
695 "/dev/urandom\0"
696 "/dev/tty\0";
697
698 char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
699 const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
700 bool can_mknod = true;
701 _cleanup_umask_ mode_t u;
702 int r;
703
704 assert(m);
705
706 u = umask(0000);
707
708 if (!mkdtemp(temporary_mount))
709 return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount);
710
711 dev = strjoina(temporary_mount, "/dev");
712 (void) mkdir(dev, 0755);
713 if (mount("tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=755" TMPFS_LIMITS_DEV) < 0) {
714 r = log_debug_errno(errno, "Failed to mount tmpfs on '%s': %m", dev);
715 goto fail;
716 }
717 r = label_fix_container(dev, "/dev", 0);
718 if (r < 0) {
719 log_debug_errno(errno, "Failed to fix label of '%s' as /dev: %m", dev);
720 goto fail;
721 }
722
723 devpts = strjoina(temporary_mount, "/dev/pts");
724 (void) mkdir(devpts, 0755);
725 if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
726 r = log_debug_errno(errno, "Failed to bind mount /dev/pts on '%s': %m", devpts);
727 goto fail;
728 }
729
730 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
731 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
732 * Thus, in that case make a clone.
733 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
734 r = is_symlink("/dev/ptmx");
735 if (r < 0) {
736 log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
737 goto fail;
738 } else if (r > 0) {
739 devptmx = strjoina(temporary_mount, "/dev/ptmx");
740 if (symlink("pts/ptmx", devptmx) < 0) {
741 r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx);
742 goto fail;
743 }
744 } else {
745 r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod);
746 if (r < 0)
747 goto fail;
748 }
749
750 devshm = strjoina(temporary_mount, "/dev/shm");
751 (void) mkdir(devshm, 0755);
752 r = mount("/dev/shm", devshm, NULL, MS_BIND, NULL);
753 if (r < 0) {
754 r = log_debug_errno(errno, "Failed to bind mount /dev/shm on '%s': %m", devshm);
755 goto fail;
756 }
757
758 devmqueue = strjoina(temporary_mount, "/dev/mqueue");
759 (void) mkdir(devmqueue, 0755);
760 if (mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL) < 0)
761 log_debug_errno(errno, "Failed to bind mount /dev/mqueue on '%s', ignoring: %m", devmqueue);
762
763 devhugepages = strjoina(temporary_mount, "/dev/hugepages");
764 (void) mkdir(devhugepages, 0755);
765 if (mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL) < 0)
766 log_debug_errno(errno, "Failed to bind mount /dev/hugepages on '%s', ignoring: %m", devhugepages);
767
768 devlog = strjoina(temporary_mount, "/dev/log");
769 if (symlink("/run/systemd/journal/dev-log", devlog) < 0)
770 log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog);
771
772 NULSTR_FOREACH(d, devnodes) {
773 r = clone_device_node(d, temporary_mount, &can_mknod);
774 /* ENXIO means the *source* is not a device file, skip creation in that case */
775 if (r < 0 && r != -ENXIO)
776 goto fail;
777 }
778
779 r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID);
780 if (r < 0)
781 log_debug_errno(r, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount);
782
783 /* Create the /dev directory if missing. It is more likely to be
784 * missing when the service is started with RootDirectory. This is
785 * consistent with mount units creating the mount points when missing.
786 */
787 (void) mkdir_p_label(mount_entry_path(m), 0755);
788
789 /* Unmount everything in old /dev */
790 r = umount_recursive(mount_entry_path(m), 0);
791 if (r < 0)
792 log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m));
793
794 if (mount(dev, mount_entry_path(m), NULL, MS_MOVE, NULL) < 0) {
795 r = log_debug_errno(errno, "Failed to move mount point '%s' to '%s': %m", dev, mount_entry_path(m));
796 goto fail;
797 }
798
799 (void) rmdir(dev);
800 (void) rmdir(temporary_mount);
801
802 return 0;
803
804 fail:
805 if (devpts)
806 (void) umount(devpts);
807
808 if (devshm)
809 (void) umount(devshm);
810
811 if (devhugepages)
812 (void) umount(devhugepages);
813
814 if (devmqueue)
815 (void) umount(devmqueue);
816
817 (void) umount(dev);
818 (void) rmdir(dev);
819 (void) rmdir(temporary_mount);
820
821 return r;
822 }
823
824 static int mount_bind_dev(const MountEntry *m) {
825 int r;
826
827 assert(m);
828
829 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
830 * /dev. This is only used when RootDirectory= is set. */
831
832 (void) mkdir_p_label(mount_entry_path(m), 0755);
833
834 r = path_is_mount_point(mount_entry_path(m), NULL, 0);
835 if (r < 0)
836 return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m");
837 if (r > 0) /* make this a NOP if /dev is already a mount point */
838 return 0;
839
840 if (mount("/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL) < 0)
841 return log_debug_errno(errno, "Failed to bind mount %s: %m", mount_entry_path(m));
842
843 return 1;
844 }
845
846 static int mount_sysfs(const MountEntry *m) {
847 int r;
848
849 assert(m);
850
851 (void) mkdir_p_label(mount_entry_path(m), 0755);
852
853 r = path_is_mount_point(mount_entry_path(m), NULL, 0);
854 if (r < 0)
855 return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m");
856 if (r > 0) /* make this a NOP if /sys is already a mount point */
857 return 0;
858
859 /* Bind mount the host's version so that we get all child mounts of it, too. */
860 if (mount("/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL) < 0)
861 return log_debug_errno(errno, "Failed to mount %s: %m", mount_entry_path(m));
862
863 return 1;
864 }
865
866 static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) {
867 const char *entry_path;
868
869 assert(m);
870 assert(ns_info);
871
872 entry_path = mount_entry_path(m);
873
874 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
875 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
876 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
877 * mounted on /proc/ first. */
878
879 (void) mkdir_p_label(entry_path, 0755);
880 (void) umount_recursive(entry_path, 0);
881
882 if (ns_info->protect_proc != PROTECT_PROC_DEFAULT ||
883 ns_info->proc_subset != PROC_SUBSET_ALL) {
884 _cleanup_free_ char *opts = NULL;
885
886 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
887 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
888 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
889 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
890 * added in the same commit: if it's supported it is thus also per-instance. */
891
892 opts = strjoin("hidepid=",
893 ns_info->protect_proc == PROTECT_PROC_DEFAULT ? "off" :
894 protect_proc_to_string(ns_info->protect_proc),
895 ns_info->proc_subset == PROC_SUBSET_PID ? ",subset=pid" : "");
896 if (!opts)
897 return -ENOMEM;
898
899 if (mount("proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, opts) < 0) {
900 if (errno != EINVAL)
901 return log_debug_errno(errno, "Failed to mount %s (options=%s): %m", mount_entry_path(m), opts);
902
903 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
904 * not supported by the kernel, and thus the per-instance hidepid= neither, which
905 * means we really don't want to use it, since it would affect our host's /proc
906 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
907 } else
908 return 1;
909 }
910
911 if (mount("proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
912 return log_debug_errno(errno, "Failed to mount %s (no options): %m", mount_entry_path(m));
913
914 return 1;
915 }
916
917 static int mount_tmpfs(const MountEntry *m) {
918 const char *entry_path, *inner_path;
919 int r;
920
921 assert(m);
922
923 entry_path = mount_entry_path(m);
924 inner_path = m->path_const;
925
926 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
927
928 (void) mkdir_p_label(entry_path, 0755);
929 (void) umount_recursive(entry_path, 0);
930
931 if (mount("tmpfs", entry_path, "tmpfs", m->flags, mount_entry_options(m)) < 0)
932 return log_debug_errno(errno, "Failed to mount %s: %m", entry_path);
933
934 r = label_fix_container(entry_path, inner_path, 0);
935 if (r < 0)
936 return log_debug_errno(r, "Failed to fix label of '%s' as '%s': %m", entry_path, inner_path);
937
938 return 1;
939 }
940
941 static int mount_images(const MountEntry *m) {
942 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
943 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
944 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
945 _cleanup_free_ void *root_hash_decoded = NULL;
946 _cleanup_free_ char *verity_data = NULL, *hash_sig = NULL;
947 DissectImageFlags dissect_image_flags = m->read_only ? DISSECT_IMAGE_READ_ONLY : 0;
948 size_t root_hash_size = 0;
949 int r;
950
951 r = verity_metadata_load(mount_entry_source(m), NULL, &root_hash_decoded, &root_hash_size, &verity_data, &hash_sig);
952 if (r < 0)
953 return log_debug_errno(r, "Failed to load root hash: %m");
954 dissect_image_flags |= verity_data ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0;
955
956 r = loop_device_make_by_path(mount_entry_source(m),
957 m->read_only ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */,
958 verity_data ? 0 : LO_FLAGS_PARTSCAN,
959 &loop_device);
960 if (r < 0)
961 return log_debug_errno(r, "Failed to create loop device for image: %m");
962
963 r = dissect_image(loop_device->fd, root_hash_decoded, root_hash_size, verity_data, m->image_options, dissect_image_flags, &dissected_image);
964 /* No partition table? Might be a single-filesystem image, try again */
965 if (!verity_data && r < 0 && r == -ENOPKG)
966 r = dissect_image(loop_device->fd, root_hash_decoded, root_hash_size, verity_data, m->image_options, dissect_image_flags|DISSECT_IMAGE_NO_PARTITION_TABLE, &dissected_image);
967 if (r < 0)
968 return log_debug_errno(r, "Failed to dissect image: %m");
969
970 r = dissected_image_decrypt(dissected_image, NULL, root_hash_decoded, root_hash_size, verity_data, hash_sig, NULL, 0, dissect_image_flags, &decrypted_image);
971 if (r < 0)
972 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
973
974 r = mkdir_p_label(mount_entry_path(m), 0755);
975 if (r < 0)
976 return log_debug_errno(r, "Failed to create destination directory %s: %m", mount_entry_path(m));
977 r = umount_recursive(mount_entry_path(m), 0);
978 if (r < 0)
979 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", mount_entry_path(m));
980
981 r = dissected_image_mount(dissected_image, mount_entry_path(m), UID_INVALID, dissect_image_flags);
982 if (r < 0)
983 return log_debug_errno(r, "Failed to mount image: %m");
984
985 if (decrypted_image) {
986 r = decrypted_image_relinquish(decrypted_image);
987 if (r < 0)
988 return log_debug_errno(r, "Failed to relinquish decrypted image: %m");
989 }
990
991 loop_device_relinquish(loop_device);
992
993 return 1;
994 }
995
996 static int follow_symlink(
997 const char *root_directory,
998 MountEntry *m) {
999
1000 _cleanup_free_ char *target = NULL;
1001 int r;
1002
1003 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1004 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1005 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1006 * end and already have a fully normalized name. */
1007
1008 r = chase_symlinks(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL);
1009 if (r < 0)
1010 return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m));
1011 if (r > 0) /* Reached the end, nothing more to resolve */
1012 return 1;
1013
1014 if (m->n_followed >= CHASE_SYMLINKS_MAX) /* put a boundary on things */
1015 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP),
1016 "Symlink loop on '%s'.",
1017 mount_entry_path(m));
1018
1019 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m), target);
1020
1021 free_and_replace(m->path_malloc, target);
1022 m->has_prefix = true;
1023
1024 m->n_followed ++;
1025
1026 return 0;
1027 }
1028
1029 static int apply_mount(
1030 const char *root_directory,
1031 MountEntry *m,
1032 const NamespaceInfo *ns_info) {
1033
1034 _cleanup_free_ char *inaccessible = NULL;
1035 bool rbind = true, make = false;
1036 const char *what;
1037 int r;
1038
1039 assert(m);
1040 assert(ns_info);
1041
1042 log_debug("Applying namespace mount on %s", mount_entry_path(m));
1043
1044 switch (m->mode) {
1045
1046 case INACCESSIBLE: {
1047 _cleanup_free_ char *tmp = NULL;
1048 const char *runtime_dir;
1049 struct stat target;
1050
1051 /* First, get rid of everything that is below if there
1052 * is anything... Then, overmount it with an
1053 * inaccessible path. */
1054 (void) umount_recursive(mount_entry_path(m), 0);
1055
1056 if (lstat(mount_entry_path(m), &target) < 0) {
1057 if (errno == ENOENT && m->ignore)
1058 return 0;
1059
1060 return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m",
1061 mount_entry_path(m));
1062 }
1063
1064 if (geteuid() == 0)
1065 runtime_dir = "/run";
1066 else {
1067 if (asprintf(&tmp, "/run/user/" UID_FMT, geteuid()) < 0)
1068 return -ENOMEM;
1069
1070 runtime_dir = tmp;
1071 }
1072
1073 r = mode_to_inaccessible_node(runtime_dir, target.st_mode, &inaccessible);
1074 if (r < 0)
1075 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP),
1076 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1077 what = inaccessible;
1078 break;
1079 }
1080
1081 case READONLY:
1082 case READWRITE:
1083 case READWRITE_IMPLICIT:
1084 r = path_is_mount_point(mount_entry_path(m), root_directory, 0);
1085 if (r == -ENOENT && m->ignore)
1086 return 0;
1087 if (r < 0)
1088 return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m",
1089 mount_entry_path(m));
1090 if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1091 * bit for the mount point if needed. */
1092 return 0;
1093 /* This isn't a mount point yet, let's make it one. */
1094 what = mount_entry_path(m);
1095 break;
1096
1097 case BIND_MOUNT:
1098 rbind = false;
1099
1100 _fallthrough_;
1101 case BIND_MOUNT_RECURSIVE: {
1102 _cleanup_free_ char *chased = NULL;
1103
1104 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1105 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1106 * root directory to chase_symlinks() here. */
1107
1108 r = chase_symlinks(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL);
1109 if (r == -ENOENT && m->ignore) {
1110 log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m));
1111 return 0;
1112 }
1113 if (r < 0)
1114 return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m));
1115
1116 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m), chased);
1117
1118 free_and_replace(m->source_malloc, chased);
1119
1120 what = mount_entry_source(m);
1121 make = true;
1122 break;
1123 }
1124
1125 case EMPTY_DIR:
1126 case TMPFS:
1127 return mount_tmpfs(m);
1128
1129 case PRIVATE_TMP:
1130 case PRIVATE_TMP_READONLY:
1131 what = mount_entry_source(m);
1132 make = true;
1133 break;
1134
1135 case PRIVATE_DEV:
1136 return mount_private_dev(m);
1137
1138 case BIND_DEV:
1139 return mount_bind_dev(m);
1140
1141 case SYSFS:
1142 return mount_sysfs(m);
1143
1144 case PROCFS:
1145 return mount_procfs(m, ns_info);
1146
1147 case MOUNT_IMAGES:
1148 return mount_images(m);
1149
1150 default:
1151 assert_not_reached("Unknown mode");
1152 }
1153
1154 assert(what);
1155
1156 if (mount(what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL) < 0) {
1157 bool try_again = false;
1158 r = -errno;
1159
1160 if (r == -ENOENT && make) {
1161 struct stat st;
1162
1163 /* Hmm, either the source or the destination are missing. Let's see if we can create
1164 the destination, then try again. */
1165
1166 if (stat(what, &st) < 0)
1167 log_error_errno(errno, "Mount point source '%s' is not accessible: %m", what);
1168 else {
1169 int q;
1170
1171 (void) mkdir_parents(mount_entry_path(m), 0755);
1172
1173 if (S_ISDIR(st.st_mode))
1174 q = mkdir(mount_entry_path(m), 0755) < 0 ? -errno : 0;
1175 else
1176 q = touch(mount_entry_path(m));
1177
1178 if (q < 0)
1179 log_error_errno(q, "Failed to create destination mount point node '%s': %m",
1180 mount_entry_path(m));
1181 else
1182 try_again = true;
1183 }
1184 }
1185
1186 if (try_again) {
1187 if (mount(what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL) < 0)
1188 r = -errno;
1189 else
1190 r = 0;
1191 }
1192
1193 if (r < 0)
1194 return log_error_errno(r, "Failed to mount %s to %s: %m", what, mount_entry_path(m));
1195 }
1196
1197 log_debug("Successfully mounted %s to %s", what, mount_entry_path(m));
1198 return 0;
1199 }
1200
1201 static int make_read_only(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) {
1202 unsigned long new_flags = 0, flags_mask = 0;
1203 bool submounts = false;
1204 int r = 0;
1205
1206 assert(m);
1207 assert(proc_self_mountinfo);
1208
1209 if (mount_entry_read_only(m) || m->mode == PRIVATE_DEV) {
1210 new_flags |= MS_RDONLY;
1211 flags_mask |= MS_RDONLY;
1212 }
1213
1214 if (m->nosuid) {
1215 new_flags |= MS_NOSUID;
1216 flags_mask |= MS_NOSUID;
1217 }
1218
1219 if (flags_mask == 0) /* No Change? */
1220 return 0;
1221
1222 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1223 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1224 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1225 * and running Linux <= 4.17. */
1226 submounts =
1227 mount_entry_read_only(m) &&
1228 !IN_SET(m->mode, EMPTY_DIR, TMPFS);
1229 if (submounts)
1230 r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo);
1231 else
1232 r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo);
1233
1234 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1235 * read-only already stays this way. This improves compatibility with container managers, where we
1236 * won't attempt to undo read-only mounts already applied. */
1237
1238 if (r == -ENOENT && m->ignore)
1239 return 0;
1240 if (r < 0)
1241 return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m),
1242 submounts ? " and its submounts" : "");
1243 return 0;
1244 }
1245
1246 static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) {
1247 assert(ns_info);
1248
1249 /*
1250 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1251 * since to protect the API VFS mounts, they need to be around in the
1252 * first place...
1253 */
1254
1255 return ns_info->mount_apivfs ||
1256 ns_info->protect_control_groups ||
1257 ns_info->protect_kernel_tunables ||
1258 ns_info->protect_proc != PROTECT_PROC_DEFAULT ||
1259 ns_info->proc_subset != PROC_SUBSET_ALL;
1260 }
1261
1262 static size_t namespace_calculate_mounts(
1263 const NamespaceInfo *ns_info,
1264 char** read_write_paths,
1265 char** read_only_paths,
1266 char** inaccessible_paths,
1267 char** empty_directories,
1268 size_t n_bind_mounts,
1269 size_t n_temporary_filesystems,
1270 size_t n_mount_images,
1271 const char* tmp_dir,
1272 const char* var_tmp_dir,
1273 const char* log_namespace) {
1274
1275 size_t protect_home_cnt;
1276 size_t protect_system_cnt =
1277 (ns_info->protect_system == PROTECT_SYSTEM_STRICT ?
1278 ELEMENTSOF(protect_system_strict_table) :
1279 ((ns_info->protect_system == PROTECT_SYSTEM_FULL) ?
1280 ELEMENTSOF(protect_system_full_table) :
1281 ((ns_info->protect_system == PROTECT_SYSTEM_YES) ?
1282 ELEMENTSOF(protect_system_yes_table) : 0)));
1283
1284 protect_home_cnt =
1285 (ns_info->protect_home == PROTECT_HOME_YES ?
1286 ELEMENTSOF(protect_home_yes_table) :
1287 ((ns_info->protect_home == PROTECT_HOME_READ_ONLY) ?
1288 ELEMENTSOF(protect_home_read_only_table) :
1289 ((ns_info->protect_home == PROTECT_HOME_TMPFS) ?
1290 ELEMENTSOF(protect_home_tmpfs_table) : 0)));
1291
1292 return !!tmp_dir + !!var_tmp_dir +
1293 strv_length(read_write_paths) +
1294 strv_length(read_only_paths) +
1295 strv_length(inaccessible_paths) +
1296 strv_length(empty_directories) +
1297 n_bind_mounts +
1298 n_mount_images +
1299 n_temporary_filesystems +
1300 ns_info->private_dev +
1301 (ns_info->protect_kernel_tunables ? ELEMENTSOF(protect_kernel_tunables_table) : 0) +
1302 (ns_info->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) +
1303 (ns_info->protect_kernel_logs ? ELEMENTSOF(protect_kernel_logs_table) : 0) +
1304 (ns_info->protect_control_groups ? 1 : 0) +
1305 protect_home_cnt + protect_system_cnt +
1306 (ns_info->protect_hostname ? 2 : 0) +
1307 (namespace_info_mount_apivfs(ns_info) ? ELEMENTSOF(apivfs_table) : 0) +
1308 !!log_namespace;
1309 }
1310
1311 static void normalize_mounts(const char *root_directory, MountEntry *mounts, size_t *n_mounts) {
1312 assert(root_directory);
1313 assert(n_mounts);
1314 assert(mounts || *n_mounts == 0);
1315
1316 typesafe_qsort(mounts, *n_mounts, mount_path_compare);
1317
1318 drop_duplicates(mounts, n_mounts);
1319 drop_outside_root(root_directory, mounts, n_mounts);
1320 drop_inaccessible(mounts, n_mounts);
1321 drop_nop(mounts, n_mounts);
1322 }
1323
1324 static bool root_read_only(
1325 char **read_only_paths,
1326 ProtectSystem protect_system) {
1327
1328 /* Determine whether the root directory is going to be read-only given the configured settings. */
1329
1330 if (protect_system == PROTECT_SYSTEM_STRICT)
1331 return true;
1332
1333 if (prefixed_path_strv_contains(read_only_paths, "/"))
1334 return true;
1335
1336 return false;
1337 }
1338
1339 static bool home_read_only(
1340 char** read_only_paths,
1341 char** inaccessible_paths,
1342 char** empty_directories,
1343 const BindMount *bind_mounts,
1344 size_t n_bind_mounts,
1345 const TemporaryFileSystem *temporary_filesystems,
1346 size_t n_temporary_filesystems,
1347 ProtectHome protect_home) {
1348
1349 size_t i;
1350
1351 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1352 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1353 * settings. */
1354
1355 if (protect_home != PROTECT_HOME_NO)
1356 return true;
1357
1358 if (prefixed_path_strv_contains(read_only_paths, "/home") ||
1359 prefixed_path_strv_contains(inaccessible_paths, "/home") ||
1360 prefixed_path_strv_contains(empty_directories, "/home"))
1361 return true;
1362
1363 for (i = 0; i < n_temporary_filesystems; i++)
1364 if (path_equal(temporary_filesystems[i].path, "/home"))
1365 return true;
1366
1367 /* If /home is overmounted with some dir from the host it's not writable. */
1368 for (i = 0; i < n_bind_mounts; i++)
1369 if (path_equal(bind_mounts[i].destination, "/home"))
1370 return true;
1371
1372 return false;
1373 }
1374
1375 int setup_namespace(
1376 const char* root_directory,
1377 const char* root_image,
1378 const MountOptions *root_image_options,
1379 const NamespaceInfo *ns_info,
1380 char** read_write_paths,
1381 char** read_only_paths,
1382 char** inaccessible_paths,
1383 char** empty_directories,
1384 const BindMount *bind_mounts,
1385 size_t n_bind_mounts,
1386 const TemporaryFileSystem *temporary_filesystems,
1387 size_t n_temporary_filesystems,
1388 const MountImage *mount_images,
1389 size_t n_mount_images,
1390 const char* tmp_dir,
1391 const char* var_tmp_dir,
1392 const char *log_namespace,
1393 unsigned long mount_flags,
1394 const void *root_hash,
1395 size_t root_hash_size,
1396 const char *root_hash_path,
1397 const void *root_hash_sig,
1398 size_t root_hash_sig_size,
1399 const char *root_hash_sig_path,
1400 const char *root_verity,
1401 DissectImageFlags dissect_image_flags,
1402 char **error_path) {
1403
1404 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
1405 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
1406 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
1407 _cleanup_free_ void *root_hash_decoded = NULL;
1408 _cleanup_free_ char *verity_data = NULL, *hash_sig_path = NULL;
1409 MountEntry *m = NULL, *mounts = NULL;
1410 size_t n_mounts;
1411 bool require_prefix = false;
1412 const char *root;
1413 int r = 0;
1414
1415 assert(ns_info);
1416
1417 if (mount_flags == 0)
1418 mount_flags = MS_SHARED;
1419
1420 if (root_image) {
1421 dissect_image_flags |= DISSECT_IMAGE_REQUIRE_ROOT;
1422
1423 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
1424 if (root_read_only(read_only_paths,
1425 ns_info->protect_system) &&
1426 home_read_only(read_only_paths, inaccessible_paths, empty_directories,
1427 bind_mounts, n_bind_mounts, temporary_filesystems, n_temporary_filesystems,
1428 ns_info->protect_home) &&
1429 strv_isempty(read_write_paths))
1430 dissect_image_flags |= DISSECT_IMAGE_READ_ONLY;
1431
1432 r = loop_device_make_by_path(root_image,
1433 FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */,
1434 LO_FLAGS_PARTSCAN,
1435 &loop_device);
1436 if (r < 0)
1437 return log_debug_errno(r, "Failed to create loop device for root image: %m");
1438
1439 r = verity_metadata_load(root_image,
1440 root_hash_path,
1441 root_hash ? NULL : &root_hash_decoded,
1442 root_hash ? NULL : &root_hash_size,
1443 root_verity ? NULL : &verity_data,
1444 root_hash_sig || root_hash_sig_path ? NULL : &hash_sig_path);
1445 if (r < 0)
1446 return log_debug_errno(r, "Failed to load root hash: %m");
1447 dissect_image_flags |= root_verity || verity_data ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0;
1448
1449 r = dissect_image(loop_device->fd,
1450 root_hash ?: root_hash_decoded,
1451 root_hash_size,
1452 root_verity ?: verity_data,
1453 root_image_options,
1454 dissect_image_flags,
1455 &dissected_image);
1456 if (r < 0)
1457 return log_debug_errno(r, "Failed to dissect image: %m");
1458
1459 r = dissected_image_decrypt(dissected_image,
1460 NULL,
1461 root_hash ?: root_hash_decoded,
1462 root_hash_size,
1463 root_verity ?: verity_data,
1464 root_hash_sig_path ?: hash_sig_path,
1465 root_hash_sig,
1466 root_hash_sig_size,
1467 dissect_image_flags,
1468 &decrypted_image);
1469 if (r < 0)
1470 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
1471 }
1472
1473 if (root_directory)
1474 root = root_directory;
1475 else {
1476 /* Always create the mount namespace in a temporary directory, instead of operating
1477 * directly in the root. The temporary directory prevents any mounts from being
1478 * potentially obscured my other mounts we already applied.
1479 * We use the same mount point for all images, which is safe, since they all live
1480 * in their own namespaces after all, and hence won't see each other. */
1481
1482 root = "/run/systemd/unit-root";
1483 (void) mkdir_label(root, 0700);
1484 require_prefix = true;
1485 }
1486
1487 n_mounts = namespace_calculate_mounts(
1488 ns_info,
1489 read_write_paths,
1490 read_only_paths,
1491 inaccessible_paths,
1492 empty_directories,
1493 n_bind_mounts,
1494 n_temporary_filesystems,
1495 n_mount_images,
1496 tmp_dir, var_tmp_dir,
1497 log_namespace);
1498
1499 if (n_mounts > 0) {
1500 m = mounts = new0(MountEntry, n_mounts);
1501 if (!mounts)
1502 return -ENOMEM;
1503
1504 r = append_access_mounts(&m, read_write_paths, READWRITE, require_prefix);
1505 if (r < 0)
1506 goto finish;
1507
1508 r = append_access_mounts(&m, read_only_paths, READONLY, require_prefix);
1509 if (r < 0)
1510 goto finish;
1511
1512 r = append_access_mounts(&m, inaccessible_paths, INACCESSIBLE, require_prefix);
1513 if (r < 0)
1514 goto finish;
1515
1516 r = append_empty_dir_mounts(&m, empty_directories);
1517 if (r < 0)
1518 goto finish;
1519
1520 r = append_bind_mounts(&m, bind_mounts, n_bind_mounts);
1521 if (r < 0)
1522 goto finish;
1523
1524 r = append_tmpfs_mounts(&m, temporary_filesystems, n_temporary_filesystems);
1525 if (r < 0)
1526 goto finish;
1527
1528 if (tmp_dir) {
1529 bool ro = streq(tmp_dir, RUN_SYSTEMD_EMPTY);
1530
1531 *(m++) = (MountEntry) {
1532 .path_const = "/tmp",
1533 .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP,
1534 .source_const = tmp_dir,
1535 };
1536 }
1537
1538 if (var_tmp_dir) {
1539 bool ro = streq(var_tmp_dir, RUN_SYSTEMD_EMPTY);
1540
1541 *(m++) = (MountEntry) {
1542 .path_const = "/var/tmp",
1543 .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP,
1544 .source_const = var_tmp_dir,
1545 };
1546 }
1547
1548 r = append_mount_images(&m, mount_images, n_mount_images);
1549 if (r < 0)
1550 goto finish;
1551
1552 if (ns_info->private_dev) {
1553 *(m++) = (MountEntry) {
1554 .path_const = "/dev",
1555 .mode = PRIVATE_DEV,
1556 .flags = DEV_MOUNT_OPTIONS,
1557 };
1558 }
1559
1560 if (ns_info->protect_kernel_tunables) {
1561 r = append_static_mounts(&m,
1562 protect_kernel_tunables_table,
1563 ELEMENTSOF(protect_kernel_tunables_table),
1564 ns_info->ignore_protect_paths);
1565 if (r < 0)
1566 goto finish;
1567 }
1568
1569 if (ns_info->protect_kernel_modules) {
1570 r = append_static_mounts(&m,
1571 protect_kernel_modules_table,
1572 ELEMENTSOF(protect_kernel_modules_table),
1573 ns_info->ignore_protect_paths);
1574 if (r < 0)
1575 goto finish;
1576 }
1577
1578 if (ns_info->protect_kernel_logs) {
1579 r = append_static_mounts(&m,
1580 protect_kernel_logs_table,
1581 ELEMENTSOF(protect_kernel_logs_table),
1582 ns_info->ignore_protect_paths);
1583 if (r < 0)
1584 goto finish;
1585 }
1586
1587 if (ns_info->protect_control_groups) {
1588 *(m++) = (MountEntry) {
1589 .path_const = "/sys/fs/cgroup",
1590 .mode = READONLY,
1591 };
1592 }
1593
1594 r = append_protect_home(&m, ns_info->protect_home, ns_info->ignore_protect_paths);
1595 if (r < 0)
1596 goto finish;
1597
1598 r = append_protect_system(&m, ns_info->protect_system, false);
1599 if (r < 0)
1600 goto finish;
1601
1602 if (namespace_info_mount_apivfs(ns_info)) {
1603 r = append_static_mounts(&m,
1604 apivfs_table,
1605 ELEMENTSOF(apivfs_table),
1606 ns_info->ignore_protect_paths);
1607 if (r < 0)
1608 goto finish;
1609 }
1610
1611 if (ns_info->protect_hostname) {
1612 *(m++) = (MountEntry) {
1613 .path_const = "/proc/sys/kernel/hostname",
1614 .mode = READONLY,
1615 };
1616 *(m++) = (MountEntry) {
1617 .path_const = "/proc/sys/kernel/domainname",
1618 .mode = READONLY,
1619 };
1620 }
1621
1622 if (log_namespace) {
1623 _cleanup_free_ char *q;
1624
1625 q = strjoin("/run/systemd/journal.", log_namespace);
1626 if (!q) {
1627 r = -ENOMEM;
1628 goto finish;
1629 }
1630
1631 *(m++) = (MountEntry) {
1632 .path_const = "/run/systemd/journal",
1633 .mode = BIND_MOUNT_RECURSIVE,
1634 .read_only = true,
1635 .source_malloc = TAKE_PTR(q),
1636 };
1637 }
1638
1639 assert(mounts + n_mounts == m);
1640
1641 /* Prepend the root directory where that's necessary */
1642 r = prefix_where_needed(mounts, n_mounts, root);
1643 if (r < 0)
1644 goto finish;
1645
1646 normalize_mounts(root, mounts, &n_mounts);
1647 }
1648
1649 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
1650
1651 if (unshare(CLONE_NEWNS) < 0) {
1652 r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m");
1653 if (IN_SET(r, -EACCES, -EPERM, -EOPNOTSUPP, -ENOSYS))
1654 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
1655 * in place that doesn't allow us to create namespaces (or a missing cap), then
1656 * propagate a recognizable error back, which the caller can use to detect this case
1657 * (and only this) and optionally continue without namespacing applied. */
1658 r = -ENOANO;
1659
1660 goto finish;
1661 }
1662
1663 /* Remount / as SLAVE so that nothing now mounted in the namespace
1664 * shows up in the parent */
1665 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
1666 r = log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m");
1667 goto finish;
1668 }
1669
1670 if (root_image) {
1671 /* A root image is specified, mount it to the right place */
1672 r = dissected_image_mount(dissected_image, root, UID_INVALID, dissect_image_flags);
1673 if (r < 0) {
1674 log_debug_errno(r, "Failed to mount root image: %m");
1675 goto finish;
1676 }
1677
1678 if (decrypted_image) {
1679 r = decrypted_image_relinquish(decrypted_image);
1680 if (r < 0) {
1681 log_debug_errno(r, "Failed to relinquish decrypted image: %m");
1682 goto finish;
1683 }
1684 }
1685
1686 loop_device_relinquish(loop_device);
1687
1688 } else if (root_directory) {
1689
1690 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1691 r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW);
1692 if (r < 0) {
1693 log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root);
1694 goto finish;
1695 }
1696 if (r == 0) {
1697 if (mount(root, root, NULL, MS_BIND|MS_REC, NULL) < 0) {
1698 r = log_debug_errno(errno, "Failed to bind mount '%s': %m", root);
1699 goto finish;
1700 }
1701 }
1702
1703 } else {
1704
1705 /* Let's mount the main root directory to the root directory to use */
1706 if (mount("/", root, NULL, MS_BIND|MS_REC, NULL) < 0) {
1707 r = log_debug_errno(errno, "Failed to bind mount '/' on '%s': %m", root);
1708 goto finish;
1709 }
1710 }
1711
1712 /* Try to set up the new root directory before mounting anything else there. */
1713 if (root_image || root_directory)
1714 (void) base_filesystem_create(root, UID_INVALID, GID_INVALID);
1715
1716 if (n_mounts > 0) {
1717 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
1718 _cleanup_free_ char **deny_list = NULL;
1719 size_t j;
1720
1721 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1722 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1723 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
1724 if (!proc_self_mountinfo) {
1725 r = log_debug_errno(errno, "Failed to open /proc/self/mountinfo: %m");
1726 if (error_path)
1727 *error_path = strdup("/proc/self/mountinfo");
1728 goto finish;
1729 }
1730
1731 /* First round, establish all mounts we need */
1732 for (;;) {
1733 bool again = false;
1734
1735 for (m = mounts; m < mounts + n_mounts; ++m) {
1736
1737 if (m->applied)
1738 continue;
1739
1740 r = follow_symlink(root, m);
1741 if (r < 0) {
1742 if (error_path && mount_entry_path(m))
1743 *error_path = strdup(mount_entry_path(m));
1744 goto finish;
1745 }
1746 if (r == 0) {
1747 /* We hit a symlinked mount point. The entry got rewritten and might
1748 * point to a very different place now. Let's normalize the changed
1749 * list, and start from the beginning. After all to mount the entry
1750 * at the new location we might need some other mounts first */
1751 again = true;
1752 break;
1753 }
1754
1755 r = apply_mount(root, m, ns_info);
1756 if (r < 0) {
1757 if (error_path && mount_entry_path(m))
1758 *error_path = strdup(mount_entry_path(m));
1759 goto finish;
1760 }
1761
1762 m->applied = true;
1763 }
1764
1765 if (!again)
1766 break;
1767
1768 normalize_mounts(root, mounts, &n_mounts);
1769 }
1770
1771 /* Create a deny list we can pass to bind_mount_recursive() */
1772 deny_list = new(char*, n_mounts+1);
1773 if (!deny_list) {
1774 r = -ENOMEM;
1775 goto finish;
1776 }
1777 for (j = 0; j < n_mounts; j++)
1778 deny_list[j] = (char*) mount_entry_path(mounts+j);
1779 deny_list[j] = NULL;
1780
1781 /* Second round, flip the ro bits if necessary. */
1782 for (m = mounts; m < mounts + n_mounts; ++m) {
1783 r = make_read_only(m, deny_list, proc_self_mountinfo);
1784 if (r < 0) {
1785 if (error_path && mount_entry_path(m))
1786 *error_path = strdup(mount_entry_path(m));
1787 goto finish;
1788 }
1789 }
1790 }
1791
1792 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1793 r = mount_move_root(root);
1794 if (r < 0) {
1795 log_debug_errno(r, "Failed to mount root with MS_MOVE: %m");
1796 goto finish;
1797 }
1798
1799 /* Remount / as the desired mode. Note that this will not
1800 * reestablish propagation from our side to the host, since
1801 * what's disconnected is disconnected. */
1802 if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
1803 r = log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m");
1804 goto finish;
1805 }
1806
1807 r = 0;
1808
1809 finish:
1810 if (n_mounts > 0)
1811 for (m = mounts; m < mounts + n_mounts; m++)
1812 mount_entry_done(m);
1813
1814 free(mounts);
1815
1816 return r;
1817 }
1818
1819 void bind_mount_free_many(BindMount *b, size_t n) {
1820 size_t i;
1821
1822 assert(b || n == 0);
1823
1824 for (i = 0; i < n; i++) {
1825 free(b[i].source);
1826 free(b[i].destination);
1827 }
1828
1829 free(b);
1830 }
1831
1832 int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) {
1833 _cleanup_free_ char *s = NULL, *d = NULL;
1834 BindMount *c;
1835
1836 assert(b);
1837 assert(n);
1838 assert(item);
1839
1840 s = strdup(item->source);
1841 if (!s)
1842 return -ENOMEM;
1843
1844 d = strdup(item->destination);
1845 if (!d)
1846 return -ENOMEM;
1847
1848 c = reallocarray(*b, *n + 1, sizeof(BindMount));
1849 if (!c)
1850 return -ENOMEM;
1851
1852 *b = c;
1853
1854 c[(*n) ++] = (BindMount) {
1855 .source = TAKE_PTR(s),
1856 .destination = TAKE_PTR(d),
1857 .read_only = item->read_only,
1858 .nosuid = item->nosuid,
1859 .recursive = item->recursive,
1860 .ignore_enoent = item->ignore_enoent,
1861 };
1862
1863 return 0;
1864 }
1865
1866 MountImage* mount_image_free_many(MountImage *m, size_t *n) {
1867 size_t i;
1868
1869 assert(n);
1870 assert(m || *n == 0);
1871
1872 for (i = 0; i < *n; i++) {
1873 free(m[i].source);
1874 free(m[i].destination);
1875 mount_options_free_all(m[i].mount_options);
1876 }
1877
1878 free(m);
1879 *n = 0;
1880 return NULL;
1881 }
1882
1883 int mount_image_add(MountImage **m, size_t *n, const MountImage *item) {
1884 _cleanup_free_ char *s = NULL, *d = NULL;
1885 _cleanup_(mount_options_free_allp) MountOptions *options = NULL;
1886 MountOptions *i;
1887 MountImage *c;
1888
1889 assert(m);
1890 assert(n);
1891 assert(item);
1892
1893 s = strdup(item->source);
1894 if (!s)
1895 return -ENOMEM;
1896
1897 d = strdup(item->destination);
1898 if (!d)
1899 return -ENOMEM;
1900
1901 LIST_FOREACH(mount_options, i, item->mount_options) {
1902 _cleanup_(mount_options_free_allp) MountOptions *o;
1903
1904 o = new(MountOptions, 1);
1905 if (!o)
1906 return -ENOMEM;
1907
1908 *o = (MountOptions) {
1909 .partition_designator = i->partition_designator,
1910 .options = strdup(i->options),
1911 };
1912 if (!o->options)
1913 return -ENOMEM;
1914
1915 LIST_APPEND(mount_options, options, TAKE_PTR(o));
1916 }
1917
1918 c = reallocarray(*m, *n + 1, sizeof(MountImage));
1919 if (!c)
1920 return -ENOMEM;
1921
1922 *m = c;
1923
1924 c[(*n) ++] = (MountImage) {
1925 .source = TAKE_PTR(s),
1926 .destination = TAKE_PTR(d),
1927 .mount_options = TAKE_PTR(options),
1928 .ignore_enoent = item->ignore_enoent,
1929 };
1930
1931 return 0;
1932 }
1933
1934 void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) {
1935 size_t i;
1936
1937 assert(t || n == 0);
1938
1939 for (i = 0; i < n; i++) {
1940 free(t[i].path);
1941 free(t[i].options);
1942 }
1943
1944 free(t);
1945 }
1946
1947 int temporary_filesystem_add(
1948 TemporaryFileSystem **t,
1949 size_t *n,
1950 const char *path,
1951 const char *options) {
1952
1953 _cleanup_free_ char *p = NULL, *o = NULL;
1954 TemporaryFileSystem *c;
1955
1956 assert(t);
1957 assert(n);
1958 assert(path);
1959
1960 p = strdup(path);
1961 if (!p)
1962 return -ENOMEM;
1963
1964 if (!isempty(options)) {
1965 o = strdup(options);
1966 if (!o)
1967 return -ENOMEM;
1968 }
1969
1970 c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem));
1971 if (!c)
1972 return -ENOMEM;
1973
1974 *t = c;
1975
1976 c[(*n) ++] = (TemporaryFileSystem) {
1977 .path = TAKE_PTR(p),
1978 .options = TAKE_PTR(o),
1979 };
1980
1981 return 0;
1982 }
1983
1984 static int make_tmp_prefix(const char *prefix) {
1985 _cleanup_free_ char *t = NULL;
1986 int r;
1987
1988 /* Don't do anything unless we know the dir is actually missing */
1989 r = access(prefix, F_OK);
1990 if (r >= 0)
1991 return 0;
1992 if (errno != ENOENT)
1993 return -errno;
1994
1995 r = mkdir_parents(prefix, 0755);
1996 if (r < 0)
1997 return r;
1998
1999 r = tempfn_random(prefix, NULL, &t);
2000 if (r < 0)
2001 return r;
2002
2003 if (mkdir(t, 0777) < 0)
2004 return -errno;
2005
2006 if (chmod(t, 01777) < 0) {
2007 r = -errno;
2008 (void) rmdir(t);
2009 return r;
2010 }
2011
2012 if (rename(t, prefix) < 0) {
2013 r = -errno;
2014 (void) rmdir(t);
2015 return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */
2016 }
2017
2018 return 0;
2019
2020 }
2021
2022 static int make_tmp_subdir(const char *parent, char **ret) {
2023 _cleanup_free_ char *y = NULL;
2024
2025 y = path_join(parent, "/tmp");
2026 if (!y)
2027 return -ENOMEM;
2028
2029 RUN_WITH_UMASK(0000) {
2030 if (mkdir(y, 0777 | S_ISVTX) < 0)
2031 return -errno;
2032 }
2033
2034 if (ret)
2035 *ret = TAKE_PTR(y);
2036 return 0;
2037 }
2038
2039 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path, char **tmp_path) {
2040 _cleanup_free_ char *x = NULL;
2041 char bid[SD_ID128_STRING_MAX];
2042 sd_id128_t boot_id;
2043 bool rw = true;
2044 int r;
2045
2046 assert(id);
2047 assert(prefix);
2048 assert(path);
2049
2050 /* We include the boot id in the directory so that after a
2051 * reboot we can easily identify obsolete directories. */
2052
2053 r = sd_id128_get_boot(&boot_id);
2054 if (r < 0)
2055 return r;
2056
2057 x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX");
2058 if (!x)
2059 return -ENOMEM;
2060
2061 r = make_tmp_prefix(prefix);
2062 if (r < 0)
2063 return r;
2064
2065 RUN_WITH_UMASK(0077)
2066 if (!mkdtemp(x)) {
2067 if (errno == EROFS || ERRNO_IS_DISK_SPACE(errno))
2068 rw = false;
2069 else
2070 return -errno;
2071 }
2072
2073 if (rw) {
2074 r = make_tmp_subdir(x, tmp_path);
2075 if (r < 0)
2076 return r;
2077 } else {
2078 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2079 * read-only. This way the service will get the EROFS result as if it was writing to the real
2080 * file system. */
2081 r = mkdir_p(RUN_SYSTEMD_EMPTY, 0500);
2082 if (r < 0)
2083 return r;
2084
2085 r = free_and_strdup(&x, RUN_SYSTEMD_EMPTY);
2086 if (r < 0)
2087 return r;
2088 }
2089
2090 *path = TAKE_PTR(x);
2091 return 0;
2092 }
2093
2094 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
2095 _cleanup_(namespace_cleanup_tmpdirp) char *a = NULL;
2096 _cleanup_(rmdir_and_freep) char *a_tmp = NULL;
2097 char *b;
2098 int r;
2099
2100 assert(id);
2101 assert(tmp_dir);
2102 assert(var_tmp_dir);
2103
2104 r = setup_one_tmp_dir(id, "/tmp", &a, &a_tmp);
2105 if (r < 0)
2106 return r;
2107
2108 r = setup_one_tmp_dir(id, "/var/tmp", &b, NULL);
2109 if (r < 0)
2110 return r;
2111
2112 a_tmp = mfree(a_tmp); /* avoid rmdir */
2113 *tmp_dir = TAKE_PTR(a);
2114 *var_tmp_dir = TAKE_PTR(b);
2115
2116 return 0;
2117 }
2118
2119 int setup_netns(const int netns_storage_socket[static 2]) {
2120 _cleanup_close_ int netns = -1;
2121 int r, q;
2122
2123 assert(netns_storage_socket);
2124 assert(netns_storage_socket[0] >= 0);
2125 assert(netns_storage_socket[1] >= 0);
2126
2127 /* We use the passed socketpair as a storage buffer for our
2128 * namespace reference fd. Whatever process runs this first
2129 * shall create a new namespace, all others should just join
2130 * it. To serialize that we use a file lock on the socket
2131 * pair.
2132 *
2133 * It's a bit crazy, but hey, works great! */
2134
2135 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
2136 return -errno;
2137
2138 netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
2139 if (netns == -EAGAIN) {
2140 /* Nothing stored yet, so let's create a new namespace. */
2141
2142 if (unshare(CLONE_NEWNET) < 0) {
2143 r = -errno;
2144 goto fail;
2145 }
2146
2147 (void) loopback_setup();
2148
2149 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
2150 if (netns < 0) {
2151 r = -errno;
2152 goto fail;
2153 }
2154
2155 r = 1;
2156
2157 } else if (netns < 0) {
2158 r = netns;
2159 goto fail;
2160
2161 } else {
2162 /* Yay, found something, so let's join the namespace */
2163 if (setns(netns, CLONE_NEWNET) < 0) {
2164 r = -errno;
2165 goto fail;
2166 }
2167
2168 r = 0;
2169 }
2170
2171 q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
2172 if (q < 0) {
2173 r = q;
2174 goto fail;
2175 }
2176
2177 fail:
2178 (void) lockf(netns_storage_socket[0], F_ULOCK, 0);
2179 return r;
2180 }
2181
2182 int open_netns_path(const int netns_storage_socket[static 2], const char *path) {
2183 _cleanup_close_ int netns = -1;
2184 int q, r;
2185
2186 assert(netns_storage_socket);
2187 assert(netns_storage_socket[0] >= 0);
2188 assert(netns_storage_socket[1] >= 0);
2189 assert(path);
2190
2191 /* If the storage socket doesn't contain a netns fd yet, open one via the file system and store it in
2192 * it. This is supposed to be called ahead of time, i.e. before setup_netns() which will allocate a
2193 * new anonymous netns if needed. */
2194
2195 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
2196 return -errno;
2197
2198 netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
2199 if (netns == -EAGAIN) {
2200 /* Nothing stored yet. Open the file from the file system. */
2201
2202 netns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
2203 if (netns < 0) {
2204 r = -errno;
2205 goto fail;
2206 }
2207
2208 r = fd_is_network_ns(netns);
2209 if (r == 0) { /* Not a netns? Refuse early. */
2210 r = -EINVAL;
2211 goto fail;
2212 }
2213 if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */
2214 goto fail;
2215
2216 r = 1;
2217
2218 } else if (netns < 0) {
2219 r = netns;
2220 goto fail;
2221 } else
2222 r = 0; /* Already allocated */
2223
2224 q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
2225 if (q < 0) {
2226 r = q;
2227 goto fail;
2228 }
2229
2230 fail:
2231 (void) lockf(netns_storage_socket[0], F_ULOCK, 0);
2232 return r;
2233 }
2234
2235 bool ns_type_supported(NamespaceType type) {
2236 const char *t, *ns_proc;
2237
2238 t = namespace_type_to_string(type);
2239 if (!t) /* Don't know how to translate this? Then it's not supported */
2240 return false;
2241
2242 ns_proc = strjoina("/proc/self/ns/", t);
2243 return access(ns_proc, F_OK) == 0;
2244 }
2245
2246 static const char *const protect_home_table[_PROTECT_HOME_MAX] = {
2247 [PROTECT_HOME_NO] = "no",
2248 [PROTECT_HOME_YES] = "yes",
2249 [PROTECT_HOME_READ_ONLY] = "read-only",
2250 [PROTECT_HOME_TMPFS] = "tmpfs",
2251 };
2252
2253 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES);
2254
2255 static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
2256 [PROTECT_SYSTEM_NO] = "no",
2257 [PROTECT_SYSTEM_YES] = "yes",
2258 [PROTECT_SYSTEM_FULL] = "full",
2259 [PROTECT_SYSTEM_STRICT] = "strict",
2260 };
2261
2262 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES);
2263
2264 static const char* const namespace_type_table[] = {
2265 [NAMESPACE_MOUNT] = "mnt",
2266 [NAMESPACE_CGROUP] = "cgroup",
2267 [NAMESPACE_UTS] = "uts",
2268 [NAMESPACE_IPC] = "ipc",
2269 [NAMESPACE_USER] = "user",
2270 [NAMESPACE_PID] = "pid",
2271 [NAMESPACE_NET] = "net",
2272 };
2273
2274 DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType);
2275
2276 static const char* const protect_proc_table[_PROTECT_PROC_MAX] = {
2277 [PROTECT_PROC_DEFAULT] = "default",
2278 [PROTECT_PROC_NOACCESS] = "noaccess",
2279 [PROTECT_PROC_INVISIBLE] = "invisible",
2280 [PROTECT_PROC_PTRACEABLE] = "ptraceable",
2281 };
2282
2283 DEFINE_STRING_TABLE_LOOKUP(protect_proc, ProtectProc);
2284
2285 static const char* const proc_subset_table[_PROC_SUBSET_MAX] = {
2286 [PROC_SUBSET_ALL] = "all",
2287 [PROC_SUBSET_PID] = "pid",
2288 };
2289
2290 DEFINE_STRING_TABLE_LOOKUP(proc_subset, ProcSubset);
Unnamed repository; edit this file 'description' to name the repository.