]> 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 #14229 from yuwata/nspawn-network-interface-14223
[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 "fs-util.h"
16 #include "label.h"
17 #include "loop-util.h"
18 #include "loopback-setup.h"
19 #include "mkdir.h"
20 #include "mount-util.h"
21 #include "mountpoint-util.h"
22 #include "namespace-util.h"
23 #include "namespace.h"
24 #include "nulstr-util.h"
25 #include "path-util.h"
26 #include "selinux-util.h"
27 #include "socket-util.h"
28 #include "sort-util.h"
29 #include "stat-util.h"
30 #include "string-table.h"
31 #include "string-util.h"
32 #include "strv.h"
33 #include "tmpfile-util.h"
34 #include "umask-util.h"
35 #include "user-util.h"
36
37 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
38
39 typedef enum MountMode {
40 /* This is ordered by priority! */
41 INACCESSIBLE,
42 BIND_MOUNT,
43 BIND_MOUNT_RECURSIVE,
44 PRIVATE_TMP,
45 PRIVATE_DEV,
46 BIND_DEV,
47 EMPTY_DIR,
48 SYSFS,
49 PROCFS,
50 READONLY,
51 READWRITE,
52 TMPFS,
53 READWRITE_IMPLICIT, /* Should have the lowest priority. */
54 _MOUNT_MODE_MAX,
55 } MountMode;
56
57 typedef struct MountEntry {
58 const char *path_const; /* Memory allocated on stack or static */
59 MountMode mode:5;
60 bool ignore:1; /* Ignore if path does not exist? */
61 bool has_prefix:1; /* Already is prefixed by the root dir? */
62 bool read_only:1; /* Shall this mount point be read-only? */
63 bool nosuid:1; /* Shall set MS_NOSUID on the mount itself */
64 bool applied:1; /* Already applied */
65 char *path_malloc; /* Use this instead of 'path_const' if we had to allocate memory */
66 const char *source_const; /* The source path, for bind mounts */
67 char *source_malloc;
68 const char *options_const;/* Mount options for tmpfs */
69 char *options_malloc;
70 unsigned long flags; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
71 unsigned n_followed;
72 } MountEntry;
73
74 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
75 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
76 static const MountEntry apivfs_table[] = {
77 { "/proc", PROCFS, false },
78 { "/dev", BIND_DEV, false },
79 { "/sys", SYSFS, false },
80 };
81
82 /* ProtectKernelTunables= option and the related filesystem APIs */
83 static const MountEntry protect_kernel_tunables_table[] = {
84 { "/proc/acpi", READONLY, true },
85 { "/proc/apm", READONLY, true }, /* Obsolete API, there's no point in permitting access to this, ever */
86 { "/proc/asound", READONLY, true },
87 { "/proc/bus", READONLY, true },
88 { "/proc/fs", READONLY, true },
89 { "/proc/irq", READONLY, true },
90 { "/proc/kallsyms", INACCESSIBLE, true },
91 { "/proc/kcore", INACCESSIBLE, true },
92 { "/proc/latency_stats", READONLY, true },
93 { "/proc/mtrr", READONLY, true },
94 { "/proc/scsi", READONLY, true },
95 { "/proc/sys", READONLY, false },
96 { "/proc/sysrq-trigger", READONLY, true },
97 { "/proc/timer_stats", READONLY, true },
98 { "/sys", READONLY, false },
99 { "/sys/fs/bpf", READONLY, true },
100 { "/sys/fs/cgroup", READWRITE_IMPLICIT, false }, /* READONLY is set by ProtectControlGroups= option */
101 { "/sys/fs/selinux", READWRITE_IMPLICIT, true },
102 { "/sys/kernel/debug", READONLY, true },
103 { "/sys/kernel/tracing", READONLY, true },
104 };
105
106 /* ProtectKernelModules= option */
107 static const MountEntry protect_kernel_modules_table[] = {
108 #if HAVE_SPLIT_USR
109 { "/lib/modules", INACCESSIBLE, true },
110 #endif
111 { "/usr/lib/modules", INACCESSIBLE, true },
112 };
113
114 /* ProtectKernelLogs= option */
115 static const MountEntry protect_kernel_logs_table[] = {
116 { "/proc/kmsg", INACCESSIBLE, true },
117 { "/dev/kmsg", INACCESSIBLE, true },
118 };
119
120 /*
121 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
122 * system should be protected by ProtectSystem=
123 */
124 static const MountEntry protect_home_read_only_table[] = {
125 { "/home", READONLY, true },
126 { "/run/user", READONLY, true },
127 { "/root", READONLY, true },
128 };
129
130 /* ProtectHome=tmpfs table */
131 static const MountEntry protect_home_tmpfs_table[] = {
132 { "/home", TMPFS, true, .read_only = true, .options_const = "mode=0755", .flags = MS_NODEV|MS_STRICTATIME },
133 { "/run/user", TMPFS, true, .read_only = true, .options_const = "mode=0755", .flags = MS_NODEV|MS_STRICTATIME },
134 { "/root", TMPFS, true, .read_only = true, .options_const = "mode=0700", .flags = MS_NODEV|MS_STRICTATIME },
135 };
136
137 /* ProtectHome=yes table */
138 static const MountEntry protect_home_yes_table[] = {
139 { "/home", INACCESSIBLE, true },
140 { "/run/user", INACCESSIBLE, true },
141 { "/root", INACCESSIBLE, true },
142 };
143
144 /* ProtectSystem=yes table */
145 static const MountEntry protect_system_yes_table[] = {
146 { "/usr", READONLY, false },
147 { "/boot", READONLY, true },
148 { "/efi", READONLY, true },
149 #if HAVE_SPLIT_USR
150 { "/lib", READONLY, true },
151 { "/lib64", READONLY, true },
152 { "/bin", READONLY, true },
153 # if HAVE_SPLIT_BIN
154 { "/sbin", READONLY, true },
155 # endif
156 #endif
157 };
158
159 /* ProtectSystem=full includes ProtectSystem=yes */
160 static const MountEntry protect_system_full_table[] = {
161 { "/usr", READONLY, false },
162 { "/boot", READONLY, true },
163 { "/efi", READONLY, true },
164 { "/etc", READONLY, false },
165 #if HAVE_SPLIT_USR
166 { "/lib", READONLY, true },
167 { "/lib64", READONLY, true },
168 { "/bin", READONLY, true },
169 # if HAVE_SPLIT_BIN
170 { "/sbin", READONLY, true },
171 # endif
172 #endif
173 };
174
175 /*
176 * ProtectSystem=strict table. In this strict mode, we mount everything
177 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
178 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
179 * protect those, and these options should be fully orthogonal.
180 * (And of course /home and friends are also left writable, as ProtectHome=
181 * shall manage those, orthogonally).
182 */
183 static const MountEntry protect_system_strict_table[] = {
184 { "/", READONLY, false },
185 { "/proc", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */
186 { "/sys", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */
187 { "/dev", READWRITE_IMPLICIT, false }, /* PrivateDevices= */
188 { "/home", READWRITE_IMPLICIT, true }, /* ProtectHome= */
189 { "/run/user", READWRITE_IMPLICIT, true }, /* ProtectHome= */
190 { "/root", READWRITE_IMPLICIT, true }, /* ProtectHome= */
191 };
192
193 static const char * const mount_mode_table[_MOUNT_MODE_MAX] = {
194 [INACCESSIBLE] = "inaccessible",
195 [BIND_MOUNT] = "bind",
196 [BIND_MOUNT_RECURSIVE] = "rbind",
197 [PRIVATE_TMP] = "private-tmp",
198 [PRIVATE_DEV] = "private-dev",
199 [BIND_DEV] = "bind-dev",
200 [EMPTY_DIR] = "empty",
201 [SYSFS] = "sysfs",
202 [PROCFS] = "procfs",
203 [READONLY] = "read-only",
204 [READWRITE] = "read-write",
205 [TMPFS] = "tmpfs",
206 [READWRITE_IMPLICIT] = "rw-implicit",
207 };
208
209 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode, MountMode);
210
211 static const char *mount_entry_path(const MountEntry *p) {
212 assert(p);
213
214 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
215 * otherwise the stack/static ->path field is returned. */
216
217 return p->path_malloc ?: p->path_const;
218 }
219
220 static bool mount_entry_read_only(const MountEntry *p) {
221 assert(p);
222
223 return p->read_only || IN_SET(p->mode, READONLY, INACCESSIBLE);
224 }
225
226 static const char *mount_entry_source(const MountEntry *p) {
227 assert(p);
228
229 return p->source_malloc ?: p->source_const;
230 }
231
232 static const char *mount_entry_options(const MountEntry *p) {
233 assert(p);
234
235 return p->options_malloc ?: p->options_const;
236 }
237
238 static void mount_entry_done(MountEntry *p) {
239 assert(p);
240
241 p->path_malloc = mfree(p->path_malloc);
242 p->source_malloc = mfree(p->source_malloc);
243 p->options_malloc = mfree(p->options_malloc);
244 }
245
246 static int append_access_mounts(MountEntry **p, char **strv, MountMode mode, bool forcibly_require_prefix) {
247 char **i;
248
249 assert(p);
250
251 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
252
253 STRV_FOREACH(i, strv) {
254 bool ignore = false, needs_prefix = false;
255 const char *e = *i;
256
257 /* Look for any prefixes */
258 if (startswith(e, "-")) {
259 e++;
260 ignore = true;
261 }
262 if (startswith(e, "+")) {
263 e++;
264 needs_prefix = true;
265 }
266
267 if (!path_is_absolute(e))
268 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL),
269 "Path is not absolute: %s", e);
270
271 *((*p)++) = (MountEntry) {
272 .path_const = e,
273 .mode = mode,
274 .ignore = ignore,
275 .has_prefix = !needs_prefix && !forcibly_require_prefix,
276 };
277 }
278
279 return 0;
280 }
281
282 static int append_empty_dir_mounts(MountEntry **p, char **strv) {
283 char **i;
284
285 assert(p);
286
287 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
288 * "/private/" boundary directories for DynamicUser=1. */
289
290 STRV_FOREACH(i, strv) {
291
292 *((*p)++) = (MountEntry) {
293 .path_const = *i,
294 .mode = EMPTY_DIR,
295 .ignore = false,
296 .read_only = true,
297 .options_const = "mode=755",
298 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
299 };
300 }
301
302 return 0;
303 }
304
305 static int append_bind_mounts(MountEntry **p, const BindMount *binds, size_t n) {
306 size_t i;
307
308 assert(p);
309
310 for (i = 0; i < n; i++) {
311 const BindMount *b = binds + i;
312
313 *((*p)++) = (MountEntry) {
314 .path_const = b->destination,
315 .mode = b->recursive ? BIND_MOUNT_RECURSIVE : BIND_MOUNT,
316 .read_only = b->read_only,
317 .nosuid = b->nosuid,
318 .source_const = b->source,
319 .ignore = b->ignore_enoent,
320 };
321 }
322
323 return 0;
324 }
325
326 static int append_tmpfs_mounts(MountEntry **p, const TemporaryFileSystem *tmpfs, size_t n) {
327 size_t i;
328 int r;
329
330 assert(p);
331
332 for (i = 0; i < n; i++) {
333 const TemporaryFileSystem *t = tmpfs + i;
334 _cleanup_free_ char *o = NULL, *str = NULL;
335 unsigned long flags;
336 bool ro = false;
337
338 if (!path_is_absolute(t->path))
339 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL),
340 "Path is not absolute: %s",
341 t->path);
342
343 str = strjoin("mode=0755,", t->options);
344 if (!str)
345 return -ENOMEM;
346
347 r = mount_option_mangle(str, MS_NODEV|MS_STRICTATIME, &flags, &o);
348 if (r < 0)
349 return log_debug_errno(r, "Failed to parse mount option '%s': %m", str);
350
351 ro = flags & MS_RDONLY;
352 if (ro)
353 flags ^= MS_RDONLY;
354
355 *((*p)++) = (MountEntry) {
356 .path_const = t->path,
357 .mode = TMPFS,
358 .read_only = ro,
359 .options_malloc = TAKE_PTR(o),
360 .flags = flags,
361 };
362 }
363
364 return 0;
365 }
366
367 static int append_static_mounts(MountEntry **p, const MountEntry *mounts, size_t n, bool ignore_protect) {
368 size_t i;
369
370 assert(p);
371 assert(mounts);
372
373 /* Adds a list of static pre-defined entries */
374
375 for (i = 0; i < n; i++)
376 *((*p)++) = (MountEntry) {
377 .path_const = mount_entry_path(mounts+i),
378 .mode = mounts[i].mode,
379 .ignore = mounts[i].ignore || ignore_protect,
380 };
381
382 return 0;
383 }
384
385 static int append_protect_home(MountEntry **p, ProtectHome protect_home, bool ignore_protect) {
386 assert(p);
387
388 switch (protect_home) {
389
390 case PROTECT_HOME_NO:
391 return 0;
392
393 case PROTECT_HOME_READ_ONLY:
394 return append_static_mounts(p, protect_home_read_only_table, ELEMENTSOF(protect_home_read_only_table), ignore_protect);
395
396 case PROTECT_HOME_TMPFS:
397 return append_static_mounts(p, protect_home_tmpfs_table, ELEMENTSOF(protect_home_tmpfs_table), ignore_protect);
398
399 case PROTECT_HOME_YES:
400 return append_static_mounts(p, protect_home_yes_table, ELEMENTSOF(protect_home_yes_table), ignore_protect);
401
402 default:
403 assert_not_reached("Unexpected ProtectHome= value");
404 }
405 }
406
407 static int append_protect_system(MountEntry **p, ProtectSystem protect_system, bool ignore_protect) {
408 assert(p);
409
410 switch (protect_system) {
411
412 case PROTECT_SYSTEM_NO:
413 return 0;
414
415 case PROTECT_SYSTEM_STRICT:
416 return append_static_mounts(p, protect_system_strict_table, ELEMENTSOF(protect_system_strict_table), ignore_protect);
417
418 case PROTECT_SYSTEM_YES:
419 return append_static_mounts(p, protect_system_yes_table, ELEMENTSOF(protect_system_yes_table), ignore_protect);
420
421 case PROTECT_SYSTEM_FULL:
422 return append_static_mounts(p, protect_system_full_table, ELEMENTSOF(protect_system_full_table), ignore_protect);
423
424 default:
425 assert_not_reached("Unexpected ProtectSystem= value");
426 }
427 }
428
429 static int mount_path_compare(const MountEntry *a, const MountEntry *b) {
430 int d;
431
432 /* If the paths are not equal, then order prefixes first */
433 d = path_compare(mount_entry_path(a), mount_entry_path(b));
434 if (d != 0)
435 return d;
436
437 /* If the paths are equal, check the mode */
438 return CMP((int) a->mode, (int) b->mode);
439 }
440
441 static int prefix_where_needed(MountEntry *m, size_t n, const char *root_directory) {
442 size_t i;
443
444 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
445
446 for (i = 0; i < n; i++) {
447 char *s;
448
449 if (m[i].has_prefix)
450 continue;
451
452 s = path_join(root_directory, mount_entry_path(m+i));
453 if (!s)
454 return -ENOMEM;
455
456 free_and_replace(m[i].path_malloc, s);
457 m[i].has_prefix = true;
458 }
459
460 return 0;
461 }
462
463 static void drop_duplicates(MountEntry *m, size_t *n) {
464 MountEntry *f, *t, *previous;
465
466 assert(m);
467 assert(n);
468
469 /* Drops duplicate entries. Expects that the array is properly ordered already. */
470
471 for (f = m, t = m, previous = NULL; f < m + *n; f++) {
472
473 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
474 * above. Note that we only drop duplicates that haven't been mounted yet. */
475 if (previous &&
476 path_equal(mount_entry_path(f), mount_entry_path(previous)) &&
477 !f->applied && !previous->applied) {
478 log_debug("%s (%s) is duplicate.", mount_entry_path(f), mount_mode_to_string(f->mode));
479 previous->read_only = previous->read_only || mount_entry_read_only(f); /* Propagate the read-only flag to the remaining entry */
480 mount_entry_done(f);
481 continue;
482 }
483
484 *t = *f;
485 previous = t;
486 t++;
487 }
488
489 *n = t - m;
490 }
491
492 static void drop_inaccessible(MountEntry *m, size_t *n) {
493 MountEntry *f, *t;
494 const char *clear = NULL;
495
496 assert(m);
497 assert(n);
498
499 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
500 * ordered already. */
501
502 for (f = m, t = m; f < m + *n; f++) {
503
504 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
505 * it, as inaccessible paths really should drop the entire subtree. */
506 if (clear && path_startswith(mount_entry_path(f), clear)) {
507 log_debug("%s is masked by %s.", mount_entry_path(f), clear);
508 mount_entry_done(f);
509 continue;
510 }
511
512 clear = f->mode == INACCESSIBLE ? mount_entry_path(f) : NULL;
513
514 *t = *f;
515 t++;
516 }
517
518 *n = t - m;
519 }
520
521 static void drop_nop(MountEntry *m, size_t *n) {
522 MountEntry *f, *t;
523
524 assert(m);
525 assert(n);
526
527 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
528 * list is ordered by prefixes. */
529
530 for (f = m, t = m; f < m + *n; f++) {
531
532 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
533 if (IN_SET(f->mode, READONLY, READWRITE, READWRITE_IMPLICIT)) {
534 MountEntry *p;
535 bool found = false;
536
537 /* Now let's find the first parent of the entry we are looking at. */
538 for (p = t-1; p >= m; p--) {
539 if (path_startswith(mount_entry_path(f), mount_entry_path(p))) {
540 found = true;
541 break;
542 }
543 }
544
545 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
546 if (found && p->mode == f->mode) {
547 log_debug("%s (%s) is made redundant by %s (%s)",
548 mount_entry_path(f), mount_mode_to_string(f->mode),
549 mount_entry_path(p), mount_mode_to_string(p->mode));
550 mount_entry_done(f);
551 continue;
552 }
553 }
554
555 *t = *f;
556 t++;
557 }
558
559 *n = t - m;
560 }
561
562 static void drop_outside_root(const char *root_directory, MountEntry *m, size_t *n) {
563 MountEntry *f, *t;
564
565 assert(m);
566 assert(n);
567
568 /* Nothing to do */
569 if (!root_directory)
570 return;
571
572 /* Drops all mounts that are outside of the root directory. */
573
574 for (f = m, t = m; f < m + *n; f++) {
575
576 if (!path_startswith(mount_entry_path(f), root_directory)) {
577 log_debug("%s is outside of root directory.", mount_entry_path(f));
578 mount_entry_done(f);
579 continue;
580 }
581
582 *t = *f;
583 t++;
584 }
585
586 *n = t - m;
587 }
588
589 static int clone_device_node(
590 const char *d,
591 const char *temporary_mount,
592 bool *make_devnode) {
593
594 _cleanup_free_ char *sl = NULL;
595 const char *dn, *bn, *t;
596 struct stat st;
597 int r;
598
599 if (stat(d, &st) < 0) {
600 if (errno == ENOENT) {
601 log_debug_errno(errno, "Device node '%s' to clone does not exist, ignoring.", d);
602 return -ENXIO;
603 }
604
605 return log_debug_errno(errno, "Failed to stat() device node '%s' to clone, ignoring: %m", d);
606 }
607
608 if (!S_ISBLK(st.st_mode) &&
609 !S_ISCHR(st.st_mode))
610 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL),
611 "Device node '%s' to clone is not a device node, ignoring.",
612 d);
613
614 dn = strjoina(temporary_mount, d);
615
616 /* First, try to create device node properly */
617 if (*make_devnode) {
618 mac_selinux_create_file_prepare(d, st.st_mode);
619 r = mknod(dn, st.st_mode, st.st_rdev);
620 mac_selinux_create_file_clear();
621 if (r >= 0)
622 goto add_symlink;
623 if (errno != EPERM)
624 return log_debug_errno(errno, "mknod failed for %s: %m", d);
625
626 /* This didn't work, let's not try this again for the next iterations. */
627 *make_devnode = false;
628 }
629
630 /* We're about to fallback to bind-mounting the device
631 * node. So create a dummy bind-mount target. */
632 mac_selinux_create_file_prepare(d, 0);
633 r = mknod(dn, S_IFREG, 0);
634 mac_selinux_create_file_clear();
635 if (r < 0 && errno != EEXIST)
636 return log_debug_errno(errno, "mknod() fallback failed for '%s': %m", d);
637
638 /* Fallback to bind-mounting:
639 * The assumption here is that all used device nodes carry standard
640 * properties. Specifically, the devices nodes we bind-mount should
641 * either be owned by root:root or root:tty (e.g. /dev/tty, /dev/ptmx)
642 * and should not carry ACLs. */
643 if (mount(d, dn, NULL, MS_BIND, NULL) < 0)
644 return log_debug_errno(errno, "Bind mounting failed for '%s': %m", d);
645
646 add_symlink:
647 bn = path_startswith(d, "/dev/");
648 if (!bn)
649 return 0;
650
651 /* Create symlinks like /dev/char/1:9 → ../urandom */
652 if (asprintf(&sl, "%s/dev/%s/%u:%u", temporary_mount, S_ISCHR(st.st_mode) ? "char" : "block", major(st.st_rdev), minor(st.st_rdev)) < 0)
653 return log_oom();
654
655 (void) mkdir_parents(sl, 0755);
656
657 t = strjoina("../", bn);
658
659 if (symlink(t, sl) < 0)
660 log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl);
661
662 return 0;
663 }
664
665 static int mount_private_dev(MountEntry *m) {
666 static const char devnodes[] =
667 "/dev/null\0"
668 "/dev/zero\0"
669 "/dev/full\0"
670 "/dev/random\0"
671 "/dev/urandom\0"
672 "/dev/tty\0";
673
674 char temporary_mount[] = "/tmp/namespace-dev-XXXXXX";
675 const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL;
676 bool can_mknod = true;
677 _cleanup_umask_ mode_t u;
678 int r;
679
680 assert(m);
681
682 u = umask(0000);
683
684 if (!mkdtemp(temporary_mount))
685 return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount);
686
687 dev = strjoina(temporary_mount, "/dev");
688 (void) mkdir(dev, 0755);
689 if (mount("tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=755") < 0) {
690 r = log_debug_errno(errno, "Failed to mount tmpfs on '%s': %m", dev);
691 goto fail;
692 }
693
694 devpts = strjoina(temporary_mount, "/dev/pts");
695 (void) mkdir(devpts, 0755);
696 if (mount("/dev/pts", devpts, NULL, MS_BIND, NULL) < 0) {
697 r = log_debug_errno(errno, "Failed to bind mount /dev/pts on '%s': %m", devpts);
698 goto fail;
699 }
700
701 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
702 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
703 * Thus, in that case make a clone.
704 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
705 r = is_symlink("/dev/ptmx");
706 if (r < 0) {
707 log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
708 goto fail;
709 } else if (r > 0) {
710 devptmx = strjoina(temporary_mount, "/dev/ptmx");
711 if (symlink("pts/ptmx", devptmx) < 0) {
712 r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx);
713 goto fail;
714 }
715 } else {
716 r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod);
717 if (r < 0)
718 goto fail;
719 }
720
721 devshm = strjoina(temporary_mount, "/dev/shm");
722 (void) mkdir(devshm, 0755);
723 r = mount("/dev/shm", devshm, NULL, MS_BIND, NULL);
724 if (r < 0) {
725 r = log_debug_errno(errno, "Failed to bind mount /dev/shm on '%s': %m", devshm);
726 goto fail;
727 }
728
729 devmqueue = strjoina(temporary_mount, "/dev/mqueue");
730 (void) mkdir(devmqueue, 0755);
731 if (mount("/dev/mqueue", devmqueue, NULL, MS_BIND, NULL) < 0)
732 log_debug_errno(errno, "Failed to bind mount /dev/mqueue on '%s', ignoring: %m", devmqueue);
733
734 devhugepages = strjoina(temporary_mount, "/dev/hugepages");
735 (void) mkdir(devhugepages, 0755);
736 if (mount("/dev/hugepages", devhugepages, NULL, MS_BIND, NULL) < 0)
737 log_debug_errno(errno, "Failed to bind mount /dev/hugepages on '%s', ignoring: %m", devhugepages);
738
739 devlog = strjoina(temporary_mount, "/dev/log");
740 if (symlink("/run/systemd/journal/dev-log", devlog) < 0)
741 log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog);
742
743 NULSTR_FOREACH(d, devnodes) {
744 r = clone_device_node(d, temporary_mount, &can_mknod);
745 /* ENXIO means the the *source* is not a device file, skip creation in that case */
746 if (r < 0 && r != -ENXIO)
747 goto fail;
748 }
749
750 r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID);
751 if (r < 0)
752 log_debug_errno(r, "Failed to setup basic device tree at '%s', ignoring: %m", temporary_mount);
753
754 /* Create the /dev directory if missing. It is more likely to be
755 * missing when the service is started with RootDirectory. This is
756 * consistent with mount units creating the mount points when missing.
757 */
758 (void) mkdir_p_label(mount_entry_path(m), 0755);
759
760 /* Unmount everything in old /dev */
761 r = umount_recursive(mount_entry_path(m), 0);
762 if (r < 0)
763 log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m));
764
765 if (mount(dev, mount_entry_path(m), NULL, MS_MOVE, NULL) < 0) {
766 r = log_debug_errno(errno, "Failed to move mount point '%s' to '%s': %m", dev, mount_entry_path(m));
767 goto fail;
768 }
769
770 (void) rmdir(dev);
771 (void) rmdir(temporary_mount);
772
773 return 0;
774
775 fail:
776 if (devpts)
777 (void) umount(devpts);
778
779 if (devshm)
780 (void) umount(devshm);
781
782 if (devhugepages)
783 (void) umount(devhugepages);
784
785 if (devmqueue)
786 (void) umount(devmqueue);
787
788 (void) umount(dev);
789 (void) rmdir(dev);
790 (void) rmdir(temporary_mount);
791
792 return r;
793 }
794
795 static int mount_bind_dev(const MountEntry *m) {
796 int r;
797
798 assert(m);
799
800 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
801 * /dev. This is only used when RootDirectory= is set. */
802
803 (void) mkdir_p_label(mount_entry_path(m), 0755);
804
805 r = path_is_mount_point(mount_entry_path(m), NULL, 0);
806 if (r < 0)
807 return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m");
808 if (r > 0) /* make this a NOP if /dev is already a mount point */
809 return 0;
810
811 if (mount("/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL) < 0)
812 return log_debug_errno(errno, "Failed to bind mount %s: %m", mount_entry_path(m));
813
814 return 1;
815 }
816
817 static int mount_sysfs(const MountEntry *m) {
818 int r;
819
820 assert(m);
821
822 (void) mkdir_p_label(mount_entry_path(m), 0755);
823
824 r = path_is_mount_point(mount_entry_path(m), NULL, 0);
825 if (r < 0)
826 return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m");
827 if (r > 0) /* make this a NOP if /sys is already a mount point */
828 return 0;
829
830 /* Bind mount the host's version so that we get all child mounts of it, too. */
831 if (mount("/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL) < 0)
832 return log_debug_errno(errno, "Failed to mount %s: %m", mount_entry_path(m));
833
834 return 1;
835 }
836
837 static int mount_procfs(const MountEntry *m) {
838 int r;
839
840 assert(m);
841
842 (void) mkdir_p_label(mount_entry_path(m), 0755);
843
844 r = path_is_mount_point(mount_entry_path(m), NULL, 0);
845 if (r < 0)
846 return log_debug_errno(r, "Unable to determine whether /proc is already mounted: %m");
847 if (r > 0) /* make this a NOP if /proc is already a mount point */
848 return 0;
849
850 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in one */
851 if (mount("proc", mount_entry_path(m), "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0)
852 return log_debug_errno(errno, "Failed to mount %s: %m", mount_entry_path(m));
853
854 return 1;
855 }
856
857 static int mount_tmpfs(const MountEntry *m) {
858 assert(m);
859
860 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
861
862 (void) mkdir_p_label(mount_entry_path(m), 0755);
863 (void) umount_recursive(mount_entry_path(m), 0);
864
865 if (mount("tmpfs", mount_entry_path(m), "tmpfs", m->flags, mount_entry_options(m)) < 0)
866 return log_debug_errno(errno, "Failed to mount %s: %m", mount_entry_path(m));
867
868 return 1;
869 }
870
871 static int follow_symlink(
872 const char *root_directory,
873 MountEntry *m) {
874
875 _cleanup_free_ char *target = NULL;
876 int r;
877
878 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
879 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
880 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
881 * end and already have a fully normalized name. */
882
883 r = chase_symlinks(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL);
884 if (r < 0)
885 return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m));
886 if (r > 0) /* Reached the end, nothing more to resolve */
887 return 1;
888
889 if (m->n_followed >= CHASE_SYMLINKS_MAX) /* put a boundary on things */
890 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP),
891 "Symlink loop on '%s'.",
892 mount_entry_path(m));
893
894 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m), target);
895
896 free_and_replace(m->path_malloc, target);
897 m->has_prefix = true;
898
899 m->n_followed ++;
900
901 return 0;
902 }
903
904 static int apply_mount(
905 const char *root_directory,
906 MountEntry *m) {
907
908 bool rbind = true, make = false;
909 const char *what;
910 int r;
911
912 assert(m);
913
914 log_debug("Applying namespace mount on %s", mount_entry_path(m));
915
916 switch (m->mode) {
917
918 case INACCESSIBLE: {
919 struct stat target;
920
921 /* First, get rid of everything that is below if there
922 * is anything... Then, overmount it with an
923 * inaccessible path. */
924 (void) umount_recursive(mount_entry_path(m), 0);
925
926 if (lstat(mount_entry_path(m), &target) < 0) {
927 if (errno == ENOENT && m->ignore)
928 return 0;
929
930 return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m", mount_entry_path(m));
931 }
932
933 what = mode_to_inaccessible_node(target.st_mode);
934 if (!what)
935 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP),
936 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
937 break;
938 }
939
940 case READONLY:
941 case READWRITE:
942 case READWRITE_IMPLICIT:
943 r = path_is_mount_point(mount_entry_path(m), root_directory, 0);
944 if (r == -ENOENT && m->ignore)
945 return 0;
946 if (r < 0)
947 return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m", mount_entry_path(m));
948 if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY bit for the mount point if needed. */
949 return 0;
950 /* This isn't a mount point yet, let's make it one. */
951 what = mount_entry_path(m);
952 break;
953
954 case BIND_MOUNT:
955 rbind = false;
956
957 _fallthrough_;
958 case BIND_MOUNT_RECURSIVE: {
959 _cleanup_free_ char *chased = NULL;
960
961 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note that bind
962 * mount source paths are always relative to the host root, hence we pass NULL as root directory to
963 * chase_symlinks() here. */
964
965 r = chase_symlinks(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL);
966 if (r == -ENOENT && m->ignore) {
967 log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m));
968 return 0;
969 }
970 if (r < 0)
971 return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m));
972
973 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m), chased);
974
975 free_and_replace(m->source_malloc, chased);
976
977 what = mount_entry_source(m);
978 make = true;
979 break;
980 }
981
982 case EMPTY_DIR:
983 case TMPFS:
984 return mount_tmpfs(m);
985
986 case PRIVATE_TMP:
987 what = mount_entry_source(m);
988 make = true;
989 break;
990
991 case PRIVATE_DEV:
992 return mount_private_dev(m);
993
994 case BIND_DEV:
995 return mount_bind_dev(m);
996
997 case SYSFS:
998 return mount_sysfs(m);
999
1000 case PROCFS:
1001 return mount_procfs(m);
1002
1003 default:
1004 assert_not_reached("Unknown mode");
1005 }
1006
1007 assert(what);
1008
1009 if (mount(what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL) < 0) {
1010 bool try_again = false;
1011 r = -errno;
1012
1013 if (r == -ENOENT && make) {
1014 struct stat st;
1015
1016 /* Hmm, either the source or the destination are missing. Let's see if we can create the destination, then try again */
1017
1018 if (stat(what, &st) < 0)
1019 log_debug_errno(errno, "Mount point source '%s' is not accessible: %m", what);
1020 else {
1021 int q;
1022
1023 (void) mkdir_parents(mount_entry_path(m), 0755);
1024
1025 if (S_ISDIR(st.st_mode))
1026 q = mkdir(mount_entry_path(m), 0755) < 0 ? -errno : 0;
1027 else
1028 q = touch(mount_entry_path(m));
1029
1030 if (q < 0)
1031 log_debug_errno(q, "Failed to create destination mount point node '%s': %m", mount_entry_path(m));
1032 else
1033 try_again = true;
1034 }
1035 }
1036
1037 if (try_again) {
1038 if (mount(what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL) < 0)
1039 r = -errno;
1040 else
1041 r = 0;
1042 }
1043
1044 if (r < 0)
1045 return log_debug_errno(r, "Failed to mount %s to %s: %m", what, mount_entry_path(m));
1046 }
1047
1048 log_debug("Successfully mounted %s to %s", what, mount_entry_path(m));
1049 return 0;
1050 }
1051
1052 /* Change per-mount flags on an existing mount */
1053 static int bind_remount_one(const char *path, unsigned long orig_flags, unsigned long new_flags, unsigned long flags_mask) {
1054 if (mount(NULL, path, NULL, (orig_flags & ~flags_mask) | MS_REMOUNT | MS_BIND | new_flags, NULL) < 0)
1055 return -errno;
1056
1057 return 0;
1058 }
1059
1060 static int make_read_only(const MountEntry *m, char **blacklist, FILE *proc_self_mountinfo) {
1061 unsigned long new_flags = 0, flags_mask = 0;
1062 bool submounts = false;
1063 int r = 0;
1064
1065 assert(m);
1066 assert(proc_self_mountinfo);
1067
1068 if (mount_entry_read_only(m) || m->mode == PRIVATE_DEV) {
1069 new_flags |= MS_RDONLY;
1070 flags_mask |= MS_RDONLY;
1071 }
1072
1073 if (m->nosuid) {
1074 new_flags |= MS_NOSUID;
1075 flags_mask |= MS_NOSUID;
1076 }
1077
1078 if (flags_mask == 0) /* No Change? */
1079 return 0;
1080
1081 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1082 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1083 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1084 * and running Linux <= 4.17. */
1085 submounts =
1086 mount_entry_read_only(m) &&
1087 !IN_SET(m->mode, EMPTY_DIR, TMPFS);
1088 if (submounts)
1089 r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, blacklist, proc_self_mountinfo);
1090 else
1091 r = bind_remount_one(mount_entry_path(m), m->flags, new_flags, flags_mask);
1092
1093 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1094 * read-only already stays this way. This improves compatibility with container managers, where we
1095 * won't attempt to undo read-only mounts already applied. */
1096
1097 if (r == -ENOENT && m->ignore)
1098 return 0;
1099 if (r < 0)
1100 return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m),
1101 submounts ? " and its submounts" : "");
1102 return 0;
1103 }
1104
1105 static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) {
1106 assert(ns_info);
1107
1108 /*
1109 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1110 * since to protect the API VFS mounts, they need to be around in the
1111 * first place...
1112 */
1113
1114 return ns_info->mount_apivfs ||
1115 ns_info->protect_control_groups ||
1116 ns_info->protect_kernel_tunables;
1117 }
1118
1119 static size_t namespace_calculate_mounts(
1120 const NamespaceInfo *ns_info,
1121 char** read_write_paths,
1122 char** read_only_paths,
1123 char** inaccessible_paths,
1124 char** empty_directories,
1125 size_t n_bind_mounts,
1126 size_t n_temporary_filesystems,
1127 const char* tmp_dir,
1128 const char* var_tmp_dir,
1129 ProtectHome protect_home,
1130 ProtectSystem protect_system) {
1131
1132 size_t protect_home_cnt;
1133 size_t protect_system_cnt =
1134 (protect_system == PROTECT_SYSTEM_STRICT ?
1135 ELEMENTSOF(protect_system_strict_table) :
1136 ((protect_system == PROTECT_SYSTEM_FULL) ?
1137 ELEMENTSOF(protect_system_full_table) :
1138 ((protect_system == PROTECT_SYSTEM_YES) ?
1139 ELEMENTSOF(protect_system_yes_table) : 0)));
1140
1141 protect_home_cnt =
1142 (protect_home == PROTECT_HOME_YES ?
1143 ELEMENTSOF(protect_home_yes_table) :
1144 ((protect_home == PROTECT_HOME_READ_ONLY) ?
1145 ELEMENTSOF(protect_home_read_only_table) :
1146 ((protect_home == PROTECT_HOME_TMPFS) ?
1147 ELEMENTSOF(protect_home_tmpfs_table) : 0)));
1148
1149 return !!tmp_dir + !!var_tmp_dir +
1150 strv_length(read_write_paths) +
1151 strv_length(read_only_paths) +
1152 strv_length(inaccessible_paths) +
1153 strv_length(empty_directories) +
1154 n_bind_mounts +
1155 n_temporary_filesystems +
1156 ns_info->private_dev +
1157 (ns_info->protect_kernel_tunables ? ELEMENTSOF(protect_kernel_tunables_table) : 0) +
1158 (ns_info->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) +
1159 (ns_info->protect_kernel_logs ? ELEMENTSOF(protect_kernel_logs_table) : 0) +
1160 (ns_info->protect_control_groups ? 1 : 0) +
1161 protect_home_cnt + protect_system_cnt +
1162 (ns_info->protect_hostname ? 2 : 0) +
1163 (namespace_info_mount_apivfs(ns_info) ? ELEMENTSOF(apivfs_table) : 0);
1164 }
1165
1166 static void normalize_mounts(const char *root_directory, MountEntry *mounts, size_t *n_mounts) {
1167 assert(root_directory);
1168 assert(n_mounts);
1169 assert(mounts || *n_mounts == 0);
1170
1171 typesafe_qsort(mounts, *n_mounts, mount_path_compare);
1172
1173 drop_duplicates(mounts, n_mounts);
1174 drop_outside_root(root_directory, mounts, n_mounts);
1175 drop_inaccessible(mounts, n_mounts);
1176 drop_nop(mounts, n_mounts);
1177 }
1178
1179 int setup_namespace(
1180 const char* root_directory,
1181 const char* root_image,
1182 const NamespaceInfo *ns_info,
1183 char** read_write_paths,
1184 char** read_only_paths,
1185 char** inaccessible_paths,
1186 char** empty_directories,
1187 const BindMount *bind_mounts,
1188 size_t n_bind_mounts,
1189 const TemporaryFileSystem *temporary_filesystems,
1190 size_t n_temporary_filesystems,
1191 const char* tmp_dir,
1192 const char* var_tmp_dir,
1193 ProtectHome protect_home,
1194 ProtectSystem protect_system,
1195 unsigned long mount_flags,
1196 DissectImageFlags dissect_image_flags,
1197 char **error_path) {
1198
1199 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
1200 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
1201 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
1202 _cleanup_free_ void *root_hash = NULL;
1203 MountEntry *m = NULL, *mounts = NULL;
1204 size_t n_mounts, root_hash_size = 0;
1205 bool require_prefix = false;
1206 const char *root;
1207 int r = 0;
1208
1209 assert(ns_info);
1210
1211 if (mount_flags == 0)
1212 mount_flags = MS_SHARED;
1213
1214 if (root_image) {
1215 dissect_image_flags |= DISSECT_IMAGE_REQUIRE_ROOT;
1216
1217 if (protect_system == PROTECT_SYSTEM_STRICT &&
1218 protect_home != PROTECT_HOME_NO &&
1219 strv_isempty(read_write_paths))
1220 dissect_image_flags |= DISSECT_IMAGE_READ_ONLY;
1221
1222 r = loop_device_make_by_path(root_image,
1223 dissect_image_flags & DISSECT_IMAGE_READ_ONLY ? O_RDONLY : O_RDWR,
1224 LO_FLAGS_PARTSCAN,
1225 &loop_device);
1226 if (r < 0)
1227 return log_debug_errno(r, "Failed to create loop device for root image: %m");
1228
1229 r = root_hash_load(root_image, &root_hash, &root_hash_size);
1230 if (r < 0)
1231 return log_debug_errno(r, "Failed to load root hash: %m");
1232
1233 r = dissect_image(loop_device->fd, root_hash, root_hash_size, dissect_image_flags, &dissected_image);
1234 if (r < 0)
1235 return log_debug_errno(r, "Failed to dissect image: %m");
1236
1237 r = dissected_image_decrypt(dissected_image, NULL, root_hash, root_hash_size, dissect_image_flags, &decrypted_image);
1238 if (r < 0)
1239 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
1240 }
1241
1242 if (root_directory)
1243 root = root_directory;
1244 else {
1245 /* Always create the mount namespace in a temporary directory, instead of operating
1246 * directly in the root. The temporary directory prevents any mounts from being
1247 * potentially obscured my other mounts we already applied.
1248 * We use the same mount point for all images, which is safe, since they all live
1249 * in their own namespaces after all, and hence won't see each other. */
1250
1251 root = "/run/systemd/unit-root";
1252 (void) mkdir_label(root, 0700);
1253 require_prefix = true;
1254 }
1255
1256 n_mounts = namespace_calculate_mounts(
1257 ns_info,
1258 read_write_paths,
1259 read_only_paths,
1260 inaccessible_paths,
1261 empty_directories,
1262 n_bind_mounts,
1263 n_temporary_filesystems,
1264 tmp_dir, var_tmp_dir,
1265 protect_home, protect_system);
1266
1267 if (n_mounts > 0) {
1268 m = mounts = new0(MountEntry, n_mounts);
1269 if (!mounts)
1270 return -ENOMEM;
1271
1272 r = append_access_mounts(&m, read_write_paths, READWRITE, require_prefix);
1273 if (r < 0)
1274 goto finish;
1275
1276 r = append_access_mounts(&m, read_only_paths, READONLY, require_prefix);
1277 if (r < 0)
1278 goto finish;
1279
1280 r = append_access_mounts(&m, inaccessible_paths, INACCESSIBLE, require_prefix);
1281 if (r < 0)
1282 goto finish;
1283
1284 r = append_empty_dir_mounts(&m, empty_directories);
1285 if (r < 0)
1286 goto finish;
1287
1288 r = append_bind_mounts(&m, bind_mounts, n_bind_mounts);
1289 if (r < 0)
1290 goto finish;
1291
1292 r = append_tmpfs_mounts(&m, temporary_filesystems, n_temporary_filesystems);
1293 if (r < 0)
1294 goto finish;
1295
1296 if (tmp_dir) {
1297 *(m++) = (MountEntry) {
1298 .path_const = "/tmp",
1299 .mode = PRIVATE_TMP,
1300 .source_const = tmp_dir,
1301 };
1302 }
1303
1304 if (var_tmp_dir) {
1305 *(m++) = (MountEntry) {
1306 .path_const = "/var/tmp",
1307 .mode = PRIVATE_TMP,
1308 .source_const = var_tmp_dir,
1309 };
1310 }
1311
1312 if (ns_info->private_dev) {
1313 *(m++) = (MountEntry) {
1314 .path_const = "/dev",
1315 .mode = PRIVATE_DEV,
1316 .flags = DEV_MOUNT_OPTIONS,
1317 };
1318 }
1319
1320 if (ns_info->protect_kernel_tunables) {
1321 r = append_static_mounts(&m, protect_kernel_tunables_table, ELEMENTSOF(protect_kernel_tunables_table), ns_info->ignore_protect_paths);
1322 if (r < 0)
1323 goto finish;
1324 }
1325
1326 if (ns_info->protect_kernel_modules) {
1327 r = append_static_mounts(&m, protect_kernel_modules_table, ELEMENTSOF(protect_kernel_modules_table), ns_info->ignore_protect_paths);
1328 if (r < 0)
1329 goto finish;
1330 }
1331
1332 if (ns_info->protect_kernel_logs) {
1333 r = append_static_mounts(&m, protect_kernel_logs_table, ELEMENTSOF(protect_kernel_logs_table), ns_info->ignore_protect_paths);
1334 if (r < 0)
1335 goto finish;
1336 }
1337
1338 if (ns_info->protect_control_groups) {
1339 *(m++) = (MountEntry) {
1340 .path_const = "/sys/fs/cgroup",
1341 .mode = READONLY,
1342 };
1343 }
1344
1345 r = append_protect_home(&m, protect_home, ns_info->ignore_protect_paths);
1346 if (r < 0)
1347 goto finish;
1348
1349 r = append_protect_system(&m, protect_system, false);
1350 if (r < 0)
1351 goto finish;
1352
1353 if (namespace_info_mount_apivfs(ns_info)) {
1354 r = append_static_mounts(&m, apivfs_table, ELEMENTSOF(apivfs_table), ns_info->ignore_protect_paths);
1355 if (r < 0)
1356 goto finish;
1357 }
1358
1359 if (ns_info->protect_hostname) {
1360 *(m++) = (MountEntry) {
1361 .path_const = "/proc/sys/kernel/hostname",
1362 .mode = READONLY,
1363 };
1364 *(m++) = (MountEntry) {
1365 .path_const = "/proc/sys/kernel/domainname",
1366 .mode = READONLY,
1367 };
1368 }
1369
1370 assert(mounts + n_mounts == m);
1371
1372 /* Prepend the root directory where that's necessary */
1373 r = prefix_where_needed(mounts, n_mounts, root);
1374 if (r < 0)
1375 goto finish;
1376
1377 normalize_mounts(root, mounts, &n_mounts);
1378 }
1379
1380 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
1381
1382 if (unshare(CLONE_NEWNS) < 0) {
1383 r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m");
1384 if (IN_SET(r, -EACCES, -EPERM, -EOPNOTSUPP, -ENOSYS))
1385 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter in place
1386 * that doesn't allow us to create namespaces (or a missing cap), then propagate a recognizable
1387 * error back, which the caller can use to detect this case (and only this) and optionally
1388 * continue without namespacing applied. */
1389 r = -ENOANO;
1390
1391 goto finish;
1392 }
1393
1394 /* Remount / as SLAVE so that nothing now mounted in the namespace
1395 * shows up in the parent */
1396 if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) {
1397 r = log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m");
1398 goto finish;
1399 }
1400
1401 if (root_image) {
1402 /* A root image is specified, mount it to the right place */
1403 r = dissected_image_mount(dissected_image, root, UID_INVALID, dissect_image_flags);
1404 if (r < 0) {
1405 log_debug_errno(r, "Failed to mount root image: %m");
1406 goto finish;
1407 }
1408
1409 if (decrypted_image) {
1410 r = decrypted_image_relinquish(decrypted_image);
1411 if (r < 0) {
1412 log_debug_errno(r, "Failed to relinquish decrypted image: %m");
1413 goto finish;
1414 }
1415 }
1416
1417 loop_device_relinquish(loop_device);
1418
1419 } else if (root_directory) {
1420
1421 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1422 r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW);
1423 if (r < 0) {
1424 log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root);
1425 goto finish;
1426 }
1427 if (r == 0) {
1428 if (mount(root, root, NULL, MS_BIND|MS_REC, NULL) < 0) {
1429 r = log_debug_errno(errno, "Failed to bind mount '%s': %m", root);
1430 goto finish;
1431 }
1432 }
1433
1434 } else {
1435
1436 /* Let's mount the main root directory to the root directory to use */
1437 if (mount("/", root, NULL, MS_BIND|MS_REC, NULL) < 0) {
1438 r = log_debug_errno(errno, "Failed to bind mount '/' on '%s': %m", root);
1439 goto finish;
1440 }
1441 }
1442
1443 /* Try to set up the new root directory before mounting anything else there. */
1444 if (root_image || root_directory)
1445 (void) base_filesystem_create(root, UID_INVALID, GID_INVALID);
1446
1447 if (n_mounts > 0) {
1448 _cleanup_fclose_ FILE *proc_self_mountinfo = NULL;
1449 _cleanup_free_ char **blacklist = NULL;
1450 size_t j;
1451
1452 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of /proc.
1453 * For example, this is the case with the option: 'InaccessiblePaths=/proc' */
1454 proc_self_mountinfo = fopen("/proc/self/mountinfo", "re");
1455 if (!proc_self_mountinfo) {
1456 r = log_debug_errno(errno, "Failed to open /proc/self/mountinfo: %m");
1457 if (error_path)
1458 *error_path = strdup("/proc/self/mountinfo");
1459 goto finish;
1460 }
1461
1462 /* First round, establish all mounts we need */
1463 for (;;) {
1464 bool again = false;
1465
1466 for (m = mounts; m < mounts + n_mounts; ++m) {
1467
1468 if (m->applied)
1469 continue;
1470
1471 r = follow_symlink(root, m);
1472 if (r < 0) {
1473 if (error_path && mount_entry_path(m))
1474 *error_path = strdup(mount_entry_path(m));
1475 goto finish;
1476 }
1477 if (r == 0) {
1478 /* We hit a symlinked mount point. The entry got rewritten and might point to a
1479 * very different place now. Let's normalize the changed list, and start from
1480 * the beginning. After all to mount the entry at the new location we might
1481 * need some other mounts first */
1482 again = true;
1483 break;
1484 }
1485
1486 r = apply_mount(root, m);
1487 if (r < 0) {
1488 if (error_path && mount_entry_path(m))
1489 *error_path = strdup(mount_entry_path(m));
1490 goto finish;
1491 }
1492
1493 m->applied = true;
1494 }
1495
1496 if (!again)
1497 break;
1498
1499 normalize_mounts(root, mounts, &n_mounts);
1500 }
1501
1502 /* Create a blacklist we can pass to bind_mount_recursive() */
1503 blacklist = new(char*, n_mounts+1);
1504 if (!blacklist) {
1505 r = -ENOMEM;
1506 goto finish;
1507 }
1508 for (j = 0; j < n_mounts; j++)
1509 blacklist[j] = (char*) mount_entry_path(mounts+j);
1510 blacklist[j] = NULL;
1511
1512 /* Second round, flip the ro bits if necessary. */
1513 for (m = mounts; m < mounts + n_mounts; ++m) {
1514 r = make_read_only(m, blacklist, proc_self_mountinfo);
1515 if (r < 0) {
1516 if (error_path && mount_entry_path(m))
1517 *error_path = strdup(mount_entry_path(m));
1518 goto finish;
1519 }
1520 }
1521 }
1522
1523 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1524 r = mount_move_root(root);
1525 if (r < 0) {
1526 log_debug_errno(r, "Failed to mount root with MS_MOVE: %m");
1527 goto finish;
1528 }
1529
1530 /* Remount / as the desired mode. Note that this will not
1531 * reestablish propagation from our side to the host, since
1532 * what's disconnected is disconnected. */
1533 if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) {
1534 r = log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m");
1535 goto finish;
1536 }
1537
1538 r = 0;
1539
1540 finish:
1541 for (m = mounts; m < mounts + n_mounts; m++)
1542 mount_entry_done(m);
1543
1544 free(mounts);
1545
1546 return r;
1547 }
1548
1549 void bind_mount_free_many(BindMount *b, size_t n) {
1550 size_t i;
1551
1552 assert(b || n == 0);
1553
1554 for (i = 0; i < n; i++) {
1555 free(b[i].source);
1556 free(b[i].destination);
1557 }
1558
1559 free(b);
1560 }
1561
1562 int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) {
1563 _cleanup_free_ char *s = NULL, *d = NULL;
1564 BindMount *c;
1565
1566 assert(b);
1567 assert(n);
1568 assert(item);
1569
1570 s = strdup(item->source);
1571 if (!s)
1572 return -ENOMEM;
1573
1574 d = strdup(item->destination);
1575 if (!d)
1576 return -ENOMEM;
1577
1578 c = reallocarray(*b, *n + 1, sizeof(BindMount));
1579 if (!c)
1580 return -ENOMEM;
1581
1582 *b = c;
1583
1584 c[(*n) ++] = (BindMount) {
1585 .source = TAKE_PTR(s),
1586 .destination = TAKE_PTR(d),
1587 .read_only = item->read_only,
1588 .nosuid = item->nosuid,
1589 .recursive = item->recursive,
1590 .ignore_enoent = item->ignore_enoent,
1591 };
1592
1593 return 0;
1594 }
1595
1596 void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) {
1597 size_t i;
1598
1599 assert(t || n == 0);
1600
1601 for (i = 0; i < n; i++) {
1602 free(t[i].path);
1603 free(t[i].options);
1604 }
1605
1606 free(t);
1607 }
1608
1609 int temporary_filesystem_add(
1610 TemporaryFileSystem **t,
1611 size_t *n,
1612 const char *path,
1613 const char *options) {
1614
1615 _cleanup_free_ char *p = NULL, *o = NULL;
1616 TemporaryFileSystem *c;
1617
1618 assert(t);
1619 assert(n);
1620 assert(path);
1621
1622 p = strdup(path);
1623 if (!p)
1624 return -ENOMEM;
1625
1626 if (!isempty(options)) {
1627 o = strdup(options);
1628 if (!o)
1629 return -ENOMEM;
1630 }
1631
1632 c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem));
1633 if (!c)
1634 return -ENOMEM;
1635
1636 *t = c;
1637
1638 c[(*n) ++] = (TemporaryFileSystem) {
1639 .path = TAKE_PTR(p),
1640 .options = TAKE_PTR(o),
1641 };
1642
1643 return 0;
1644 }
1645
1646 static int make_tmp_prefix(const char *prefix) {
1647 _cleanup_free_ char *t = NULL;
1648 int r;
1649
1650 /* Don't do anything unless we know the dir is actually missing */
1651 r = access(prefix, F_OK);
1652 if (r >= 0)
1653 return 0;
1654 if (errno != ENOENT)
1655 return -errno;
1656
1657 r = mkdir_parents(prefix, 0755);
1658 if (r < 0)
1659 return r;
1660
1661 r = tempfn_random(prefix, NULL, &t);
1662 if (r < 0)
1663 return r;
1664
1665 if (mkdir(t, 0777) < 0)
1666 return -errno;
1667
1668 if (chmod(t, 01777) < 0) {
1669 r = -errno;
1670 (void) rmdir(t);
1671 return r;
1672 }
1673
1674 if (rename(t, prefix) < 0) {
1675 r = -errno;
1676 (void) rmdir(t);
1677 return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */
1678 }
1679
1680 return 0;
1681
1682 }
1683
1684 static int setup_one_tmp_dir(const char *id, const char *prefix, char **path) {
1685 _cleanup_free_ char *x = NULL;
1686 char bid[SD_ID128_STRING_MAX];
1687 sd_id128_t boot_id;
1688 int r;
1689
1690 assert(id);
1691 assert(prefix);
1692 assert(path);
1693
1694 /* We include the boot id in the directory so that after a
1695 * reboot we can easily identify obsolete directories. */
1696
1697 r = sd_id128_get_boot(&boot_id);
1698 if (r < 0)
1699 return r;
1700
1701 x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX");
1702 if (!x)
1703 return -ENOMEM;
1704
1705 r = make_tmp_prefix(prefix);
1706 if (r < 0)
1707 return r;
1708
1709 RUN_WITH_UMASK(0077)
1710 if (!mkdtemp(x))
1711 return -errno;
1712
1713 RUN_WITH_UMASK(0000) {
1714 char *y;
1715
1716 y = strjoina(x, "/tmp");
1717
1718 if (mkdir(y, 0777 | S_ISVTX) < 0)
1719 return -errno;
1720 }
1721
1722 *path = TAKE_PTR(x);
1723
1724 return 0;
1725 }
1726
1727 int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) {
1728 char *a, *b;
1729 int r;
1730
1731 assert(id);
1732 assert(tmp_dir);
1733 assert(var_tmp_dir);
1734
1735 r = setup_one_tmp_dir(id, "/tmp", &a);
1736 if (r < 0)
1737 return r;
1738
1739 r = setup_one_tmp_dir(id, "/var/tmp", &b);
1740 if (r < 0) {
1741 char *t;
1742
1743 t = strjoina(a, "/tmp");
1744 (void) rmdir(t);
1745 (void) rmdir(a);
1746
1747 free(a);
1748 return r;
1749 }
1750
1751 *tmp_dir = a;
1752 *var_tmp_dir = b;
1753
1754 return 0;
1755 }
1756
1757 int setup_netns(const int netns_storage_socket[static 2]) {
1758 _cleanup_close_ int netns = -1;
1759 int r, q;
1760
1761 assert(netns_storage_socket);
1762 assert(netns_storage_socket[0] >= 0);
1763 assert(netns_storage_socket[1] >= 0);
1764
1765 /* We use the passed socketpair as a storage buffer for our
1766 * namespace reference fd. Whatever process runs this first
1767 * shall create a new namespace, all others should just join
1768 * it. To serialize that we use a file lock on the socket
1769 * pair.
1770 *
1771 * It's a bit crazy, but hey, works great! */
1772
1773 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
1774 return -errno;
1775
1776 netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
1777 if (netns == -EAGAIN) {
1778 /* Nothing stored yet, so let's create a new namespace. */
1779
1780 if (unshare(CLONE_NEWNET) < 0) {
1781 r = -errno;
1782 goto fail;
1783 }
1784
1785 (void) loopback_setup();
1786
1787 netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY);
1788 if (netns < 0) {
1789 r = -errno;
1790 goto fail;
1791 }
1792
1793 r = 1;
1794
1795 } else if (netns < 0) {
1796 r = netns;
1797 goto fail;
1798
1799 } else {
1800 /* Yay, found something, so let's join the namespace */
1801 if (setns(netns, CLONE_NEWNET) < 0) {
1802 r = -errno;
1803 goto fail;
1804 }
1805
1806 r = 0;
1807 }
1808
1809 q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
1810 if (q < 0) {
1811 r = q;
1812 goto fail;
1813 }
1814
1815 fail:
1816 (void) lockf(netns_storage_socket[0], F_ULOCK, 0);
1817 return r;
1818 }
1819
1820 int open_netns_path(const int netns_storage_socket[static 2], const char *path) {
1821 _cleanup_close_ int netns = -1;
1822 int q, r;
1823
1824 assert(netns_storage_socket);
1825 assert(netns_storage_socket[0] >= 0);
1826 assert(netns_storage_socket[1] >= 0);
1827 assert(path);
1828
1829 /* If the storage socket doesn't contain a netns fd yet, open one via the file system and store it in
1830 * it. This is supposed to be called ahead of time, i.e. before setup_netns() which will allocate a
1831 * new anonymous netns if needed. */
1832
1833 if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0)
1834 return -errno;
1835
1836 netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT);
1837 if (netns == -EAGAIN) {
1838 /* Nothing stored yet. Open the file from the file system. */
1839
1840 netns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
1841 if (netns < 0) {
1842 r = -errno;
1843 goto fail;
1844 }
1845
1846 r = fd_is_network_ns(netns);
1847 if (r == 0) { /* Not a netns? Refuse early. */
1848 r = -EINVAL;
1849 goto fail;
1850 }
1851 if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */
1852 goto fail;
1853
1854 r = 1;
1855
1856 } else if (netns < 0) {
1857 r = netns;
1858 goto fail;
1859 } else
1860 r = 0; /* Already allocated */
1861
1862 q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT);
1863 if (q < 0) {
1864 r = q;
1865 goto fail;
1866 }
1867
1868 fail:
1869 (void) lockf(netns_storage_socket[0], F_ULOCK, 0);
1870 return r;
1871 }
1872
1873 bool ns_type_supported(NamespaceType type) {
1874 const char *t, *ns_proc;
1875
1876 t = namespace_type_to_string(type);
1877 if (!t) /* Don't know how to translate this? Then it's not supported */
1878 return false;
1879
1880 ns_proc = strjoina("/proc/self/ns/", t);
1881 return access(ns_proc, F_OK) == 0;
1882 }
1883
1884 static const char *const protect_home_table[_PROTECT_HOME_MAX] = {
1885 [PROTECT_HOME_NO] = "no",
1886 [PROTECT_HOME_YES] = "yes",
1887 [PROTECT_HOME_READ_ONLY] = "read-only",
1888 [PROTECT_HOME_TMPFS] = "tmpfs",
1889 };
1890
1891 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES);
1892
1893 static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = {
1894 [PROTECT_SYSTEM_NO] = "no",
1895 [PROTECT_SYSTEM_YES] = "yes",
1896 [PROTECT_SYSTEM_FULL] = "full",
1897 [PROTECT_SYSTEM_STRICT] = "strict",
1898 };
1899
1900 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES);
1901
1902 static const char* const namespace_type_table[] = {
1903 [NAMESPACE_MOUNT] = "mnt",
1904 [NAMESPACE_CGROUP] = "cgroup",
1905 [NAMESPACE_UTS] = "uts",
1906 [NAMESPACE_IPC] = "ipc",
1907 [NAMESPACE_USER] = "user",
1908 [NAMESPACE_PID] = "pid",
1909 [NAMESPACE_NET] = "net",
1910 };
1911
1912 DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType);
Unnamed repository; edit this file 'description' to name the repository.