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