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