]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
15ae422b LP |
2 | |
3 | #include <errno.h> | |
e08f94ac | 4 | #include <linux/loop.h> |
07630cea | 5 | #include <sched.h> |
15ae422b | 6 | #include <stdio.h> |
07630cea | 7 | #include <sys/mount.h> |
07630cea | 8 | #include <unistd.h> |
25e870b5 | 9 | #include <linux/fs.h> |
15ae422b | 10 | |
b5efdb8a | 11 | #include "alloc-util.h" |
10404d52 | 12 | #include "base-filesystem.h" |
7f112f50 | 13 | #include "dev-setup.h" |
93f59701 LB |
14 | #include "env-util.h" |
15 | #include "escape.h" | |
16 | #include "extension-release.h" | |
3ffd4af2 | 17 | #include "fd-util.h" |
e5f10caf | 18 | #include "format-util.h" |
d944dc95 | 19 | #include "fs-util.h" |
e908468b | 20 | #include "label.h" |
b3d13314 | 21 | #include "list.h" |
915e6d16 | 22 | #include "loop-util.h" |
07630cea | 23 | #include "loopback-setup.h" |
07630cea | 24 | #include "mkdir.h" |
4349cd7c | 25 | #include "mount-util.h" |
049af8ad | 26 | #include "mountpoint-util.h" |
0cb8e3d1 | 27 | #include "namespace-util.h" |
3ffd4af2 | 28 | #include "namespace.h" |
d8b4d14d | 29 | #include "nulstr-util.h" |
93f59701 | 30 | #include "os-util.h" |
07630cea | 31 | #include "path-util.h" |
d7b8eec7 | 32 | #include "selinux-util.h" |
2583fbea | 33 | #include "socket-util.h" |
760877e9 | 34 | #include "sort-util.h" |
36ce7110 | 35 | #include "stat-util.h" |
8b43440b | 36 | #include "string-table.h" |
07630cea LP |
37 | #include "string-util.h" |
38 | #include "strv.h" | |
a652f050 | 39 | #include "tmpfile-util.h" |
affb60b1 | 40 | #include "umask-util.h" |
ee104e11 | 41 | #include "user-util.h" |
15ae422b | 42 | |
737ba3c8 | 43 | #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC) |
44 | ||
c17ec25e | 45 | typedef enum MountMode { |
15ae422b LP |
46 | /* This is ordered by priority! */ |
47 | INACCESSIBLE, | |
93f59701 | 48 | OVERLAY_MOUNT, |
b3d13314 | 49 | MOUNT_IMAGES, |
d2d6c096 LP |
50 | BIND_MOUNT, |
51 | BIND_MOUNT_RECURSIVE, | |
ac0930c8 | 52 | PRIVATE_TMP, |
56a13a49 | 53 | PRIVATE_TMP_READONLY, |
7f112f50 | 54 | PRIVATE_DEV, |
5d997827 | 55 | BIND_DEV, |
6c47cd7d | 56 | EMPTY_DIR, |
5d997827 LP |
57 | SYSFS, |
58 | PROCFS, | |
94293d65 | 59 | RUN, |
5d997827 | 60 | READONLY, |
59eeb84b | 61 | READWRITE, |
ddc155b2 TM |
62 | NOEXEC, |
63 | EXEC, | |
2abd4e38 | 64 | TMPFS, |
93f59701 | 65 | EXTENSION_IMAGES, /* Mounted outside the root directory, and used by subsequent mounts */ |
1e05071d | 66 | READWRITE_IMPLICIT, /* Should have the lowest priority. */ |
5beb8688 | 67 | _MOUNT_MODE_MAX, |
c17ec25e | 68 | } MountMode; |
15ae422b | 69 | |
34de407a | 70 | typedef struct MountEntry { |
5327c910 | 71 | const char *path_const; /* Memory allocated on stack or static */ |
cfbeb4ef | 72 | MountMode mode:5; |
5327c910 LP |
73 | bool ignore:1; /* Ignore if path does not exist? */ |
74 | bool has_prefix:1; /* Already is prefixed by the root dir? */ | |
cfbeb4ef | 75 | bool read_only:1; /* Shall this mount point be read-only? */ |
9ce4e4b0 | 76 | bool nosuid:1; /* Shall set MS_NOSUID on the mount itself */ |
ddc155b2 TM |
77 | bool noexec:1; /* Shall set MS_NOEXEC on the mount itself */ |
78 | bool exec:1; /* Shall clear MS_NOEXEC on the mount itself */ | |
088696fe | 79 | bool applied:1; /* Already applied */ |
55fe7432 | 80 | char *path_malloc; /* Use this instead of 'path_const' if we had to allocate memory */ |
809ceb82 LB |
81 | const char *unprefixed_path_const; /* If the path was amended with a prefix, these will save the original */ |
82 | char *unprefixed_path_malloc; | |
b3d13314 | 83 | const char *source_const; /* The source path, for bind mounts or images */ |
d2d6c096 | 84 | char *source_malloc; |
2abd4e38 YW |
85 | const char *options_const;/* Mount options for tmpfs */ |
86 | char *options_malloc; | |
87 | unsigned long flags; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */ | |
088696fe | 88 | unsigned n_followed; |
427353f6 | 89 | LIST_HEAD(MountOptions, image_options); |
34de407a | 90 | } MountEntry; |
15ae422b | 91 | |
94293d65 | 92 | /* If MountAPIVFS= is used, let's mount /sys, /proc, /dev and /run into the it, but only as a fallback if the user hasn't mounted |
3fe91079 | 93 | * something there already. These mounts are hence overridden by any other explicitly configured mounts. */ |
5d997827 LP |
94 | static const MountEntry apivfs_table[] = { |
95 | { "/proc", PROCFS, false }, | |
96 | { "/dev", BIND_DEV, false }, | |
97 | { "/sys", SYSFS, false }, | |
94293d65 | 98 | { "/run", RUN, false, .options_const = "mode=755" TMPFS_LIMITS_RUN, .flags = MS_NOSUID|MS_NODEV|MS_STRICTATIME }, |
5d997827 | 99 | }; |
f471b2af | 100 | |
11a30cec | 101 | /* ProtectKernelTunables= option and the related filesystem APIs */ |
34de407a | 102 | static const MountEntry protect_kernel_tunables_table[] = { |
1e05071d YW |
103 | { "/proc/acpi", READONLY, true }, |
104 | { "/proc/apm", READONLY, true }, /* Obsolete API, there's no point in permitting access to this, ever */ | |
105 | { "/proc/asound", READONLY, true }, | |
106 | { "/proc/bus", READONLY, true }, | |
107 | { "/proc/fs", READONLY, true }, | |
108 | { "/proc/irq", READONLY, true }, | |
109 | { "/proc/kallsyms", INACCESSIBLE, true }, | |
110 | { "/proc/kcore", INACCESSIBLE, true }, | |
111 | { "/proc/latency_stats", READONLY, true }, | |
112 | { "/proc/mtrr", READONLY, true }, | |
113 | { "/proc/scsi", READONLY, true }, | |
4e399953 | 114 | { "/proc/sys", READONLY, true }, |
1e05071d YW |
115 | { "/proc/sysrq-trigger", READONLY, true }, |
116 | { "/proc/timer_stats", READONLY, true }, | |
117 | { "/sys", READONLY, false }, | |
118 | { "/sys/fs/bpf", READONLY, true }, | |
119 | { "/sys/fs/cgroup", READWRITE_IMPLICIT, false }, /* READONLY is set by ProtectControlGroups= option */ | |
120 | { "/sys/fs/selinux", READWRITE_IMPLICIT, true }, | |
121 | { "/sys/kernel/debug", READONLY, true }, | |
122 | { "/sys/kernel/tracing", READONLY, true }, | |
11a30cec DH |
123 | }; |
124 | ||
c575770b | 125 | /* ProtectKernelModules= option */ |
34de407a | 126 | static const MountEntry protect_kernel_modules_table[] = { |
349cc4a5 | 127 | #if HAVE_SPLIT_USR |
c6232fb0 | 128 | { "/lib/modules", INACCESSIBLE, true }, |
c575770b | 129 | #endif |
c6232fb0 | 130 | { "/usr/lib/modules", INACCESSIBLE, true }, |
c575770b DH |
131 | }; |
132 | ||
94a7b275 KK |
133 | /* ProtectKernelLogs= option */ |
134 | static const MountEntry protect_kernel_logs_table[] = { | |
135 | { "/proc/kmsg", INACCESSIBLE, true }, | |
136 | { "/dev/kmsg", INACCESSIBLE, true }, | |
137 | }; | |
138 | ||
b6c432ca DH |
139 | /* |
140 | * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of | |
141 | * system should be protected by ProtectSystem= | |
142 | */ | |
34de407a | 143 | static const MountEntry protect_home_read_only_table[] = { |
c6232fb0 LP |
144 | { "/home", READONLY, true }, |
145 | { "/run/user", READONLY, true }, | |
146 | { "/root", READONLY, true }, | |
b6c432ca DH |
147 | }; |
148 | ||
e4da7d8c YW |
149 | /* ProtectHome=tmpfs table */ |
150 | static const MountEntry protect_home_tmpfs_table[] = { | |
7d85383e TM |
151 | { "/home", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, |
152 | { "/run/user", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
153 | { "/root", TMPFS, true, .read_only = true, .options_const = "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
e4da7d8c YW |
154 | }; |
155 | ||
b6c432ca | 156 | /* ProtectHome=yes table */ |
34de407a | 157 | static const MountEntry protect_home_yes_table[] = { |
c6232fb0 LP |
158 | { "/home", INACCESSIBLE, true }, |
159 | { "/run/user", INACCESSIBLE, true }, | |
160 | { "/root", INACCESSIBLE, true }, | |
b6c432ca DH |
161 | }; |
162 | ||
f471b2af | 163 | /* ProtectSystem=yes table */ |
34de407a | 164 | static const MountEntry protect_system_yes_table[] = { |
c6232fb0 LP |
165 | { "/usr", READONLY, false }, |
166 | { "/boot", READONLY, true }, | |
167 | { "/efi", READONLY, true }, | |
7486f305 AB |
168 | #if HAVE_SPLIT_USR |
169 | { "/lib", READONLY, true }, | |
170 | { "/lib64", READONLY, true }, | |
171 | { "/bin", READONLY, true }, | |
671f0f8d | 172 | # if HAVE_SPLIT_BIN |
7486f305 | 173 | { "/sbin", READONLY, true }, |
671f0f8d | 174 | # endif |
7486f305 | 175 | #endif |
f471b2af DH |
176 | }; |
177 | ||
178 | /* ProtectSystem=full includes ProtectSystem=yes */ | |
34de407a | 179 | static const MountEntry protect_system_full_table[] = { |
c6232fb0 LP |
180 | { "/usr", READONLY, false }, |
181 | { "/boot", READONLY, true }, | |
182 | { "/efi", READONLY, true }, | |
183 | { "/etc", READONLY, false }, | |
7486f305 AB |
184 | #if HAVE_SPLIT_USR |
185 | { "/lib", READONLY, true }, | |
186 | { "/lib64", READONLY, true }, | |
187 | { "/bin", READONLY, true }, | |
671f0f8d | 188 | # if HAVE_SPLIT_BIN |
7486f305 | 189 | { "/sbin", READONLY, true }, |
671f0f8d | 190 | # endif |
7486f305 | 191 | #endif |
f471b2af DH |
192 | }; |
193 | ||
194 | /* | |
195 | * ProtectSystem=strict table. In this strict mode, we mount everything | |
196 | * read-only, except for /proc, /dev, /sys which are the kernel API VFS, | |
197 | * which are left writable, but PrivateDevices= + ProtectKernelTunables= | |
198 | * protect those, and these options should be fully orthogonal. | |
199 | * (And of course /home and friends are also left writable, as ProtectHome= | |
200 | * shall manage those, orthogonally). | |
201 | */ | |
34de407a | 202 | static const MountEntry protect_system_strict_table[] = { |
1e05071d YW |
203 | { "/", READONLY, false }, |
204 | { "/proc", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
205 | { "/sys", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
206 | { "/dev", READWRITE_IMPLICIT, false }, /* PrivateDevices= */ | |
207 | { "/home", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
208 | { "/run/user", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
209 | { "/root", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
f471b2af DH |
210 | }; |
211 | ||
5beb8688 YW |
212 | static const char * const mount_mode_table[_MOUNT_MODE_MAX] = { |
213 | [INACCESSIBLE] = "inaccessible", | |
93f59701 | 214 | [OVERLAY_MOUNT] = "overlay", |
5beb8688 YW |
215 | [BIND_MOUNT] = "bind", |
216 | [BIND_MOUNT_RECURSIVE] = "rbind", | |
217 | [PRIVATE_TMP] = "private-tmp", | |
218 | [PRIVATE_DEV] = "private-dev", | |
219 | [BIND_DEV] = "bind-dev", | |
220 | [EMPTY_DIR] = "empty", | |
221 | [SYSFS] = "sysfs", | |
222 | [PROCFS] = "procfs", | |
223 | [READONLY] = "read-only", | |
224 | [READWRITE] = "read-write", | |
225 | [TMPFS] = "tmpfs", | |
b3d13314 | 226 | [MOUNT_IMAGES] = "mount-images", |
5beb8688 | 227 | [READWRITE_IMPLICIT] = "rw-implicit", |
ddc155b2 TM |
228 | [EXEC] = "exec", |
229 | [NOEXEC] = "noexec", | |
5beb8688 YW |
230 | }; |
231 | ||
232 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode, MountMode); | |
233 | ||
34de407a | 234 | static const char *mount_entry_path(const MountEntry *p) { |
f0a4feb0 DH |
235 | assert(p); |
236 | ||
5327c910 LP |
237 | /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that, |
238 | * otherwise the stack/static ->path field is returned. */ | |
f0a4feb0 | 239 | |
5327c910 | 240 | return p->path_malloc ?: p->path_const; |
f0a4feb0 DH |
241 | } |
242 | ||
809ceb82 LB |
243 | static const char *mount_entry_unprefixed_path(const MountEntry *p) { |
244 | assert(p); | |
245 | ||
246 | /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */ | |
247 | ||
248 | return p->unprefixed_path_malloc ?: p->unprefixed_path_const ?: mount_entry_path(p); | |
249 | } | |
250 | ||
251 | static void mount_entry_consume_prefix(MountEntry *p, char *new_path) { | |
252 | assert(p); | |
253 | assert(p->path_malloc || p->path_const); | |
254 | assert(new_path); | |
255 | ||
256 | /* Saves current path in unprefixed_ variable, and takes over new_path */ | |
257 | ||
258 | free_and_replace(p->unprefixed_path_malloc, p->path_malloc); | |
259 | /* If we didn't have a path on the heap, then it's a static one */ | |
260 | if (!p->unprefixed_path_malloc) | |
261 | p->unprefixed_path_const = p->path_const; | |
262 | p->path_malloc = new_path; | |
263 | p->has_prefix = true; | |
264 | } | |
265 | ||
34de407a | 266 | static bool mount_entry_read_only(const MountEntry *p) { |
cfbeb4ef LP |
267 | assert(p); |
268 | ||
56a13a49 | 269 | return p->read_only || IN_SET(p->mode, READONLY, INACCESSIBLE, PRIVATE_TMP_READONLY); |
cfbeb4ef LP |
270 | } |
271 | ||
ddc155b2 TM |
272 | static bool mount_entry_noexec(const MountEntry *p) { |
273 | assert(p); | |
274 | ||
b33cd6b3 | 275 | return p->noexec || IN_SET(p->mode, NOEXEC, INACCESSIBLE); |
ddc155b2 TM |
276 | } |
277 | ||
278 | static bool mount_entry_exec(const MountEntry *p) { | |
279 | assert(p); | |
280 | ||
281 | return p->exec || p->mode == EXEC; | |
282 | } | |
283 | ||
d2d6c096 LP |
284 | static const char *mount_entry_source(const MountEntry *p) { |
285 | assert(p); | |
286 | ||
287 | return p->source_malloc ?: p->source_const; | |
288 | } | |
289 | ||
2abd4e38 YW |
290 | static const char *mount_entry_options(const MountEntry *p) { |
291 | assert(p); | |
292 | ||
293 | return p->options_malloc ?: p->options_const; | |
294 | } | |
295 | ||
1eb7e08e LP |
296 | static void mount_entry_done(MountEntry *p) { |
297 | assert(p); | |
298 | ||
299 | p->path_malloc = mfree(p->path_malloc); | |
809ceb82 | 300 | p->unprefixed_path_malloc = mfree(p->unprefixed_path_malloc); |
1eb7e08e | 301 | p->source_malloc = mfree(p->source_malloc); |
2abd4e38 | 302 | p->options_malloc = mfree(p->options_malloc); |
427353f6 | 303 | p->image_options = mount_options_free_all(p->image_options); |
1eb7e08e LP |
304 | } |
305 | ||
d18aff04 | 306 | static int append_access_mounts(MountEntry **p, char **strv, MountMode mode, bool forcibly_require_prefix) { |
15ae422b LP |
307 | char **i; |
308 | ||
613b411c LP |
309 | assert(p); |
310 | ||
1e05071d | 311 | /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */ |
5327c910 | 312 | |
15ae422b | 313 | STRV_FOREACH(i, strv) { |
5327c910 LP |
314 | bool ignore = false, needs_prefix = false; |
315 | const char *e = *i; | |
15ae422b | 316 | |
5327c910 LP |
317 | /* Look for any prefixes */ |
318 | if (startswith(e, "-")) { | |
319 | e++; | |
9c94d52e | 320 | ignore = true; |
ea92ae33 | 321 | } |
5327c910 LP |
322 | if (startswith(e, "+")) { |
323 | e++; | |
324 | needs_prefix = true; | |
325 | } | |
ea92ae33 | 326 | |
baaa35ad ZJS |
327 | if (!path_is_absolute(e)) |
328 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
329 | "Path is not absolute: %s", e); | |
15ae422b | 330 | |
34de407a | 331 | *((*p)++) = (MountEntry) { |
5327c910 LP |
332 | .path_const = e, |
333 | .mode = mode, | |
334 | .ignore = ignore, | |
d18aff04 | 335 | .has_prefix = !needs_prefix && !forcibly_require_prefix, |
5327c910 | 336 | }; |
15ae422b LP |
337 | } |
338 | ||
339 | return 0; | |
340 | } | |
341 | ||
6c47cd7d LP |
342 | static int append_empty_dir_mounts(MountEntry **p, char **strv) { |
343 | char **i; | |
344 | ||
345 | assert(p); | |
346 | ||
347 | /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the | |
348 | * "/private/" boundary directories for DynamicUser=1. */ | |
349 | ||
350 | STRV_FOREACH(i, strv) { | |
351 | ||
352 | *((*p)++) = (MountEntry) { | |
353 | .path_const = *i, | |
354 | .mode = EMPTY_DIR, | |
355 | .ignore = false, | |
6c47cd7d | 356 | .read_only = true, |
7d85383e | 357 | .options_const = "mode=755" TMPFS_LIMITS_EMPTY_OR_ALMOST, |
2abd4e38 | 358 | .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, |
6c47cd7d LP |
359 | }; |
360 | } | |
361 | ||
362 | return 0; | |
363 | } | |
364 | ||
da6053d0 | 365 | static int append_bind_mounts(MountEntry **p, const BindMount *binds, size_t n) { |
d2d6c096 LP |
366 | assert(p); |
367 | ||
fe96c0f8 | 368 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
369 | const BindMount *b = binds + i; |
370 | ||
371 | *((*p)++) = (MountEntry) { | |
372 | .path_const = b->destination, | |
373 | .mode = b->recursive ? BIND_MOUNT_RECURSIVE : BIND_MOUNT, | |
374 | .read_only = b->read_only, | |
9ce4e4b0 | 375 | .nosuid = b->nosuid, |
d2d6c096 | 376 | .source_const = b->source, |
4ca763a9 | 377 | .ignore = b->ignore_enoent, |
d2d6c096 LP |
378 | }; |
379 | } | |
380 | ||
381 | return 0; | |
382 | } | |
383 | ||
b3d13314 LB |
384 | static int append_mount_images(MountEntry **p, const MountImage *mount_images, size_t n) { |
385 | assert(p); | |
386 | ||
387 | for (size_t i = 0; i < n; i++) { | |
388 | const MountImage *m = mount_images + i; | |
389 | ||
390 | *((*p)++) = (MountEntry) { | |
391 | .path_const = m->destination, | |
392 | .mode = MOUNT_IMAGES, | |
393 | .source_const = m->source, | |
427353f6 | 394 | .image_options = m->mount_options, |
b3d13314 LB |
395 | .ignore = m->ignore_enoent, |
396 | }; | |
397 | } | |
398 | ||
399 | return 0; | |
400 | } | |
401 | ||
93f59701 LB |
402 | static int append_extension_images( |
403 | MountEntry **p, | |
404 | const char *root, | |
405 | const char *extension_dir, | |
406 | char **hierarchies, | |
407 | const MountImage *mount_images, | |
408 | size_t n) { | |
409 | ||
410 | _cleanup_strv_free_ char **overlays = NULL; | |
411 | char **hierarchy; | |
412 | int r; | |
413 | ||
414 | assert(p); | |
415 | assert(extension_dir); | |
416 | ||
417 | if (n == 0) | |
418 | return 0; | |
419 | ||
420 | /* Prepare a list of overlays, that will have as each element a string suitable for being | |
7a7d2f16 | 421 | * passed as a lowerdir= parameter, so start with the hierarchy on the root. |
93f59701 LB |
422 | * The overlays vector will have the same number of elements and will correspond to the |
423 | * hierarchies vector, so they can be iterated upon together. */ | |
424 | STRV_FOREACH(hierarchy, hierarchies) { | |
425 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
426 | ||
427 | prefixed_hierarchy = path_join(root, *hierarchy); | |
428 | if (!prefixed_hierarchy) | |
429 | return -ENOMEM; | |
430 | ||
431 | r = strv_consume(&overlays, TAKE_PTR(prefixed_hierarchy)); | |
432 | if (r < 0) | |
433 | return r; | |
434 | } | |
435 | ||
436 | /* First, prepare a mount for each image, but these won't be visible to the unit, instead | |
437 | * they will be mounted in our propagate directory, and used as a source for the overlay. */ | |
438 | for (size_t i = 0; i < n; i++) { | |
439 | _cleanup_free_ char *mount_point = NULL; | |
440 | const MountImage *m = mount_images + i; | |
441 | ||
442 | r = asprintf(&mount_point, "%s/%zu", extension_dir, i); | |
443 | if (r < 0) | |
444 | return -ENOMEM; | |
445 | ||
446 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
447 | _cleanup_free_ char *prefixed_hierarchy = NULL, *escaped = NULL, *lowerdir = NULL; | |
448 | ||
449 | prefixed_hierarchy = path_join(mount_point, hierarchies[j]); | |
450 | if (!prefixed_hierarchy) | |
451 | return -ENOMEM; | |
452 | ||
453 | escaped = shell_escape(prefixed_hierarchy, ",:"); | |
454 | if (!escaped) | |
455 | return -ENOMEM; | |
456 | ||
457 | /* Note that lowerdir= parameters are in 'reverse' order, so the | |
458 | * top-most directory in the overlay comes first in the list. */ | |
459 | lowerdir = strjoin(escaped, ":", overlays[j]); | |
460 | if (!lowerdir) | |
461 | return -ENOMEM; | |
462 | ||
463 | free_and_replace(overlays[j], lowerdir); | |
464 | } | |
465 | ||
466 | *((*p)++) = (MountEntry) { | |
467 | .path_malloc = TAKE_PTR(mount_point), | |
468 | .image_options = m->mount_options, | |
469 | .ignore = m->ignore_enoent, | |
470 | .source_const = m->source, | |
471 | .mode = EXTENSION_IMAGES, | |
472 | .has_prefix = true, | |
473 | }; | |
474 | } | |
475 | ||
476 | /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings | |
477 | * set up earlier. */ | |
478 | for (size_t i = 0; hierarchies && hierarchies[i]; ++i) { | |
479 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
480 | ||
481 | prefixed_hierarchy = path_join(root, hierarchies[i]); | |
482 | if (!prefixed_hierarchy) | |
483 | return -ENOMEM; | |
484 | ||
485 | *((*p)++) = (MountEntry) { | |
486 | .path_malloc = TAKE_PTR(prefixed_hierarchy), | |
487 | .options_malloc = TAKE_PTR(overlays[i]), | |
488 | .mode = OVERLAY_MOUNT, | |
489 | .has_prefix = true, | |
490 | .ignore = true, /* If the source image doesn't set the ignore bit it will fail earlier. */ | |
491 | }; | |
492 | } | |
493 | ||
494 | return 0; | |
495 | } | |
496 | ||
da6053d0 | 497 | static int append_tmpfs_mounts(MountEntry **p, const TemporaryFileSystem *tmpfs, size_t n) { |
2abd4e38 YW |
498 | assert(p); |
499 | ||
b67ec8e5 | 500 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
501 | const TemporaryFileSystem *t = tmpfs + i; |
502 | _cleanup_free_ char *o = NULL, *str = NULL; | |
ad8e66dc | 503 | unsigned long flags; |
2abd4e38 | 504 | bool ro = false; |
b67ec8e5 | 505 | int r; |
2abd4e38 | 506 | |
baaa35ad ZJS |
507 | if (!path_is_absolute(t->path)) |
508 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
509 | "Path is not absolute: %s", | |
510 | t->path); | |
2abd4e38 | 511 | |
b67ec8e5 | 512 | str = strjoin("mode=0755" NESTED_TMPFS_LIMITS ",", t->options); |
ad8e66dc AJ |
513 | if (!str) |
514 | return -ENOMEM; | |
2abd4e38 | 515 | |
ad8e66dc AJ |
516 | r = mount_option_mangle(str, MS_NODEV|MS_STRICTATIME, &flags, &o); |
517 | if (r < 0) | |
518 | return log_debug_errno(r, "Failed to parse mount option '%s': %m", str); | |
2abd4e38 | 519 | |
ad8e66dc AJ |
520 | ro = flags & MS_RDONLY; |
521 | if (ro) | |
522 | flags ^= MS_RDONLY; | |
2abd4e38 YW |
523 | |
524 | *((*p)++) = (MountEntry) { | |
525 | .path_const = t->path, | |
526 | .mode = TMPFS, | |
527 | .read_only = ro, | |
ad8e66dc | 528 | .options_malloc = TAKE_PTR(o), |
2abd4e38 YW |
529 | .flags = flags, |
530 | }; | |
2abd4e38 YW |
531 | } |
532 | ||
533 | return 0; | |
534 | } | |
535 | ||
da6053d0 | 536 | static int append_static_mounts(MountEntry **p, const MountEntry *mounts, size_t n, bool ignore_protect) { |
11a30cec | 537 | assert(p); |
f471b2af | 538 | assert(mounts); |
11a30cec | 539 | |
5327c910 | 540 | /* Adds a list of static pre-defined entries */ |
f471b2af | 541 | |
fe96c0f8 | 542 | for (size_t i = 0; i < n; i++) |
34de407a LP |
543 | *((*p)++) = (MountEntry) { |
544 | .path_const = mount_entry_path(mounts+i), | |
5327c910 LP |
545 | .mode = mounts[i].mode, |
546 | .ignore = mounts[i].ignore || ignore_protect, | |
547 | }; | |
f471b2af DH |
548 | |
549 | return 0; | |
550 | } | |
551 | ||
34de407a | 552 | static int append_protect_home(MountEntry **p, ProtectHome protect_home, bool ignore_protect) { |
c575770b DH |
553 | assert(p); |
554 | ||
5327c910 | 555 | switch (protect_home) { |
b6c432ca | 556 | |
5327c910 | 557 | case PROTECT_HOME_NO: |
b6c432ca DH |
558 | return 0; |
559 | ||
b6c432ca | 560 | case PROTECT_HOME_READ_ONLY: |
5327c910 LP |
561 | return append_static_mounts(p, protect_home_read_only_table, ELEMENTSOF(protect_home_read_only_table), ignore_protect); |
562 | ||
e4da7d8c YW |
563 | case PROTECT_HOME_TMPFS: |
564 | return append_static_mounts(p, protect_home_tmpfs_table, ELEMENTSOF(protect_home_tmpfs_table), ignore_protect); | |
565 | ||
b6c432ca | 566 | case PROTECT_HOME_YES: |
5327c910 LP |
567 | return append_static_mounts(p, protect_home_yes_table, ELEMENTSOF(protect_home_yes_table), ignore_protect); |
568 | ||
b6c432ca | 569 | default: |
5327c910 | 570 | assert_not_reached("Unexpected ProtectHome= value"); |
b6c432ca | 571 | } |
b6c432ca DH |
572 | } |
573 | ||
34de407a | 574 | static int append_protect_system(MountEntry **p, ProtectSystem protect_system, bool ignore_protect) { |
f471b2af DH |
575 | assert(p); |
576 | ||
5327c910 LP |
577 | switch (protect_system) { |
578 | ||
579 | case PROTECT_SYSTEM_NO: | |
f471b2af DH |
580 | return 0; |
581 | ||
f471b2af | 582 | case PROTECT_SYSTEM_STRICT: |
5327c910 LP |
583 | return append_static_mounts(p, protect_system_strict_table, ELEMENTSOF(protect_system_strict_table), ignore_protect); |
584 | ||
f471b2af | 585 | case PROTECT_SYSTEM_YES: |
5327c910 LP |
586 | return append_static_mounts(p, protect_system_yes_table, ELEMENTSOF(protect_system_yes_table), ignore_protect); |
587 | ||
f471b2af | 588 | case PROTECT_SYSTEM_FULL: |
5327c910 LP |
589 | return append_static_mounts(p, protect_system_full_table, ELEMENTSOF(protect_system_full_table), ignore_protect); |
590 | ||
f471b2af | 591 | default: |
5327c910 | 592 | assert_not_reached("Unexpected ProtectSystem= value"); |
f471b2af | 593 | } |
11a30cec DH |
594 | } |
595 | ||
93bab288 | 596 | static int mount_path_compare(const MountEntry *a, const MountEntry *b) { |
a0827e2b | 597 | int d; |
15ae422b | 598 | |
93f59701 LB |
599 | /* EXTENSION_IMAGES will be used by other mounts as a base, so sort them first |
600 | * regardless of the prefix - they are set up in the propagate directory anyway */ | |
601 | d = -CMP(a->mode == EXTENSION_IMAGES, b->mode == EXTENSION_IMAGES); | |
602 | if (d != 0) | |
603 | return d; | |
604 | ||
6ee1a919 | 605 | /* If the paths are not equal, then order prefixes first */ |
93bab288 | 606 | d = path_compare(mount_entry_path(a), mount_entry_path(b)); |
6ee1a919 LP |
607 | if (d != 0) |
608 | return d; | |
15ae422b | 609 | |
6ee1a919 | 610 | /* If the paths are equal, check the mode */ |
93bab288 | 611 | return CMP((int) a->mode, (int) b->mode); |
15ae422b LP |
612 | } |
613 | ||
da6053d0 | 614 | static int prefix_where_needed(MountEntry *m, size_t n, const char *root_directory) { |
4a756839 | 615 | /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */ |
5327c910 | 616 | |
fe96c0f8 SS |
617 | assert(m || n == 0); |
618 | ||
619 | for (size_t i = 0; i < n; i++) { | |
5327c910 LP |
620 | char *s; |
621 | ||
622 | if (m[i].has_prefix) | |
623 | continue; | |
624 | ||
c6134d3e | 625 | s = path_join(root_directory, mount_entry_path(m+i)); |
5327c910 LP |
626 | if (!s) |
627 | return -ENOMEM; | |
628 | ||
809ceb82 | 629 | mount_entry_consume_prefix(&m[i], s); |
5327c910 LP |
630 | } |
631 | ||
632 | return 0; | |
633 | } | |
634 | ||
da6053d0 | 635 | static void drop_duplicates(MountEntry *m, size_t *n) { |
34de407a | 636 | MountEntry *f, *t, *previous; |
15ae422b | 637 | |
c17ec25e | 638 | assert(m); |
15ae422b | 639 | assert(n); |
15ae422b | 640 | |
fe3c2583 LP |
641 | /* Drops duplicate entries. Expects that the array is properly ordered already. */ |
642 | ||
1d54cd5d | 643 | for (f = m, t = m, previous = NULL; f < m + *n; f++) { |
15ae422b | 644 | |
fe3c2583 | 645 | /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare() |
088696fe LP |
646 | * above. Note that we only drop duplicates that haven't been mounted yet. */ |
647 | if (previous && | |
648 | path_equal(mount_entry_path(f), mount_entry_path(previous)) && | |
649 | !f->applied && !previous->applied) { | |
5beb8688 | 650 | log_debug("%s (%s) is duplicate.", mount_entry_path(f), mount_mode_to_string(f->mode)); |
ddc155b2 TM |
651 | /* Propagate the flags to the remaining entry */ |
652 | previous->read_only = previous->read_only || mount_entry_read_only(f); | |
653 | previous->noexec = previous->noexec || mount_entry_noexec(f); | |
654 | previous->exec = previous->exec || mount_entry_exec(f); | |
1eb7e08e | 655 | mount_entry_done(f); |
15ae422b | 656 | continue; |
fe3c2583 | 657 | } |
15ae422b | 658 | |
e2d7c1a0 | 659 | *t = *f; |
15ae422b | 660 | previous = t; |
fe3c2583 LP |
661 | t++; |
662 | } | |
663 | ||
664 | *n = t - m; | |
665 | } | |
666 | ||
da6053d0 | 667 | static void drop_inaccessible(MountEntry *m, size_t *n) { |
34de407a | 668 | MountEntry *f, *t; |
fe3c2583 LP |
669 | const char *clear = NULL; |
670 | ||
671 | assert(m); | |
672 | assert(n); | |
673 | ||
674 | /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly | |
675 | * ordered already. */ | |
676 | ||
1d54cd5d | 677 | for (f = m, t = m; f < m + *n; f++) { |
fe3c2583 LP |
678 | |
679 | /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop | |
680 | * it, as inaccessible paths really should drop the entire subtree. */ | |
34de407a LP |
681 | if (clear && path_startswith(mount_entry_path(f), clear)) { |
682 | log_debug("%s is masked by %s.", mount_entry_path(f), clear); | |
1eb7e08e | 683 | mount_entry_done(f); |
fe3c2583 LP |
684 | continue; |
685 | } | |
15ae422b | 686 | |
34de407a | 687 | clear = f->mode == INACCESSIBLE ? mount_entry_path(f) : NULL; |
fe3c2583 LP |
688 | |
689 | *t = *f; | |
15ae422b LP |
690 | t++; |
691 | } | |
692 | ||
c17ec25e | 693 | *n = t - m; |
15ae422b LP |
694 | } |
695 | ||
da6053d0 | 696 | static void drop_nop(MountEntry *m, size_t *n) { |
34de407a | 697 | MountEntry *f, *t; |
7648a565 LP |
698 | |
699 | assert(m); | |
700 | assert(n); | |
701 | ||
702 | /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the | |
703 | * list is ordered by prefixes. */ | |
704 | ||
1d54cd5d | 705 | for (f = m, t = m; f < m + *n; f++) { |
7648a565 | 706 | |
1e05071d YW |
707 | /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */ |
708 | if (IN_SET(f->mode, READONLY, READWRITE, READWRITE_IMPLICIT)) { | |
34de407a | 709 | MountEntry *p; |
7648a565 LP |
710 | bool found = false; |
711 | ||
712 | /* Now let's find the first parent of the entry we are looking at. */ | |
713 | for (p = t-1; p >= m; p--) { | |
34de407a | 714 | if (path_startswith(mount_entry_path(f), mount_entry_path(p))) { |
7648a565 LP |
715 | found = true; |
716 | break; | |
717 | } | |
718 | } | |
719 | ||
720 | /* We found it, let's see if it's the same mode, if so, we can drop this entry */ | |
721 | if (found && p->mode == f->mode) { | |
5beb8688 YW |
722 | log_debug("%s (%s) is made redundant by %s (%s)", |
723 | mount_entry_path(f), mount_mode_to_string(f->mode), | |
724 | mount_entry_path(p), mount_mode_to_string(p->mode)); | |
1eb7e08e | 725 | mount_entry_done(f); |
7648a565 LP |
726 | continue; |
727 | } | |
728 | } | |
729 | ||
730 | *t = *f; | |
731 | t++; | |
732 | } | |
733 | ||
734 | *n = t - m; | |
735 | } | |
736 | ||
da6053d0 | 737 | static void drop_outside_root(const char *root_directory, MountEntry *m, size_t *n) { |
34de407a | 738 | MountEntry *f, *t; |
cd2902c9 LP |
739 | |
740 | assert(m); | |
741 | assert(n); | |
742 | ||
1d54cd5d | 743 | /* Nothing to do */ |
cd2902c9 LP |
744 | if (!root_directory) |
745 | return; | |
746 | ||
747 | /* Drops all mounts that are outside of the root directory. */ | |
748 | ||
1d54cd5d | 749 | for (f = m, t = m; f < m + *n; f++) { |
cd2902c9 | 750 | |
93f59701 LB |
751 | /* ExtensionImages bases are opened in /run/systemd/unit-extensions on the host */ |
752 | if (f->mode != EXTENSION_IMAGES && !path_startswith(mount_entry_path(f), root_directory)) { | |
34de407a | 753 | log_debug("%s is outside of root directory.", mount_entry_path(f)); |
1eb7e08e | 754 | mount_entry_done(f); |
cd2902c9 LP |
755 | continue; |
756 | } | |
757 | ||
758 | *t = *f; | |
759 | t++; | |
760 | } | |
761 | ||
762 | *n = t - m; | |
763 | } | |
764 | ||
b2a60844 LP |
765 | static int clone_device_node( |
766 | const char *d, | |
767 | const char *temporary_mount, | |
768 | bool *make_devnode) { | |
769 | ||
770 | _cleanup_free_ char *sl = NULL; | |
771 | const char *dn, *bn, *t; | |
b5e99f23 ДГ |
772 | struct stat st; |
773 | int r; | |
774 | ||
414b304b | 775 | if (stat(d, &st) < 0) { |
b2a60844 LP |
776 | if (errno == ENOENT) { |
777 | log_debug_errno(errno, "Device node '%s' to clone does not exist, ignoring.", d); | |
af984e13 | 778 | return -ENXIO; |
b2a60844 LP |
779 | } |
780 | ||
781 | return log_debug_errno(errno, "Failed to stat() device node '%s' to clone, ignoring: %m", d); | |
b5e99f23 ДГ |
782 | } |
783 | ||
784 | if (!S_ISBLK(st.st_mode) && | |
baaa35ad ZJS |
785 | !S_ISCHR(st.st_mode)) |
786 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
787 | "Device node '%s' to clone is not a device node, ignoring.", | |
788 | d); | |
b5e99f23 | 789 | |
6f7f3a33 | 790 | dn = strjoina(temporary_mount, d); |
b5e99f23 | 791 | |
b2a60844 | 792 | /* First, try to create device node properly */ |
16498617 CB |
793 | if (*make_devnode) { |
794 | mac_selinux_create_file_prepare(d, st.st_mode); | |
795 | r = mknod(dn, st.st_mode, st.st_rdev); | |
796 | mac_selinux_create_file_clear(); | |
b2a60844 LP |
797 | if (r >= 0) |
798 | goto add_symlink; | |
16498617 CB |
799 | if (errno != EPERM) |
800 | return log_debug_errno(errno, "mknod failed for %s: %m", d); | |
801 | ||
b2a60844 | 802 | /* This didn't work, let's not try this again for the next iterations. */ |
16498617 CB |
803 | *make_devnode = false; |
804 | } | |
805 | ||
2aed63f4 | 806 | /* We're about to fall back to bind-mounting the device |
1acf344d CG |
807 | * node. So create a dummy bind-mount target. |
808 | * Do not prepare device-node SELinux label (see issue 13762) */ | |
16498617 | 809 | r = mknod(dn, S_IFREG, 0); |
16498617 | 810 | if (r < 0 && errno != EEXIST) |
b2a60844 | 811 | return log_debug_errno(errno, "mknod() fallback failed for '%s': %m", d); |
16498617 | 812 | |
21935150 LP |
813 | /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard |
814 | * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or | |
815 | * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */ | |
816 | r = mount_nofollow_verbose(LOG_DEBUG, d, dn, NULL, MS_BIND, NULL); | |
817 | if (r < 0) | |
818 | return r; | |
b2a60844 LP |
819 | |
820 | add_symlink: | |
821 | bn = path_startswith(d, "/dev/"); | |
822 | if (!bn) | |
823 | return 0; | |
824 | ||
825 | /* Create symlinks like /dev/char/1:9 → ../urandom */ | |
cbc056c8 ZJS |
826 | if (asprintf(&sl, "%s/dev/%s/%u:%u", |
827 | temporary_mount, | |
828 | S_ISCHR(st.st_mode) ? "char" : "block", | |
829 | major(st.st_rdev), minor(st.st_rdev)) < 0) | |
b2a60844 LP |
830 | return log_oom(); |
831 | ||
832 | (void) mkdir_parents(sl, 0755); | |
833 | ||
834 | t = strjoina("../", bn); | |
b2a60844 | 835 | if (symlink(t, sl) < 0) |
2e4a4fae | 836 | log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl); |
b5e99f23 | 837 | |
af984e13 | 838 | return 0; |
b5e99f23 ДГ |
839 | } |
840 | ||
5d997827 | 841 | static int mount_private_dev(MountEntry *m) { |
7f112f50 LP |
842 | static const char devnodes[] = |
843 | "/dev/null\0" | |
844 | "/dev/zero\0" | |
845 | "/dev/full\0" | |
846 | "/dev/random\0" | |
847 | "/dev/urandom\0" | |
848 | "/dev/tty\0"; | |
849 | ||
2b85f4e1 | 850 | char temporary_mount[] = "/tmp/namespace-dev-XXXXXX"; |
63cc4c31 | 851 | const char *d, *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL; |
16498617 | 852 | bool can_mknod = true; |
7f112f50 LP |
853 | _cleanup_umask_ mode_t u; |
854 | int r; | |
855 | ||
856 | assert(m); | |
857 | ||
858 | u = umask(0000); | |
859 | ||
2b85f4e1 | 860 | if (!mkdtemp(temporary_mount)) |
2e4a4fae | 861 | return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount); |
2b85f4e1 | 862 | |
63c372cb | 863 | dev = strjoina(temporary_mount, "/dev"); |
dc751688 | 864 | (void) mkdir(dev, 0755); |
21935150 LP |
865 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=755" TMPFS_LIMITS_DEV); |
866 | if (r < 0) | |
2b85f4e1 | 867 | goto fail; |
21935150 | 868 | |
c3151977 TM |
869 | r = label_fix_container(dev, "/dev", 0); |
870 | if (r < 0) { | |
871 | log_debug_errno(errno, "Failed to fix label of '%s' as /dev: %m", dev); | |
872 | goto fail; | |
873 | } | |
2b85f4e1 | 874 | |
63c372cb | 875 | devpts = strjoina(temporary_mount, "/dev/pts"); |
dc751688 | 876 | (void) mkdir(devpts, 0755); |
21935150 LP |
877 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/pts", devpts, NULL, MS_BIND, NULL); |
878 | if (r < 0) | |
2b85f4e1 | 879 | goto fail; |
2b85f4e1 | 880 | |
2e4a4fae YW |
881 | /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx. |
882 | * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible. | |
883 | * Thus, in that case make a clone. | |
884 | * In nspawn and other containers it will be a symlink, in that case make it a symlink. */ | |
36ce7110 | 885 | r = is_symlink("/dev/ptmx"); |
2e4a4fae YW |
886 | if (r < 0) { |
887 | log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m"); | |
3164e3cb | 888 | goto fail; |
2e4a4fae | 889 | } else if (r > 0) { |
414b304b ДГ |
890 | devptmx = strjoina(temporary_mount, "/dev/ptmx"); |
891 | if (symlink("pts/ptmx", devptmx) < 0) { | |
2e4a4fae | 892 | r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx); |
414b304b ДГ |
893 | goto fail; |
894 | } | |
895 | } else { | |
16498617 | 896 | r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod); |
152c475f LP |
897 | if (r < 0) |
898 | goto fail; | |
414b304b | 899 | } |
e06b6479 | 900 | |
63c372cb | 901 | devshm = strjoina(temporary_mount, "/dev/shm"); |
8d953682 | 902 | (void) mkdir(devshm, 0755); |
21935150 LP |
903 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/shm", devshm, NULL, MS_BIND, NULL); |
904 | if (r < 0) | |
2b85f4e1 | 905 | goto fail; |
2b85f4e1 | 906 | |
63c372cb | 907 | devmqueue = strjoina(temporary_mount, "/dev/mqueue"); |
dc751688 | 908 | (void) mkdir(devmqueue, 0755); |
21935150 | 909 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/mqueue", devmqueue, NULL, MS_BIND, NULL); |
2b85f4e1 | 910 | |
63c372cb | 911 | devhugepages = strjoina(temporary_mount, "/dev/hugepages"); |
dc751688 | 912 | (void) mkdir(devhugepages, 0755); |
21935150 | 913 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/hugepages", devhugepages, NULL, MS_BIND, NULL); |
2b85f4e1 | 914 | |
63c372cb | 915 | devlog = strjoina(temporary_mount, "/dev/log"); |
2e4a4fae YW |
916 | if (symlink("/run/systemd/journal/dev-log", devlog) < 0) |
917 | log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog); | |
82d25240 | 918 | |
7f112f50 | 919 | NULSTR_FOREACH(d, devnodes) { |
16498617 | 920 | r = clone_device_node(d, temporary_mount, &can_mknod); |
37b22b3b | 921 | /* ENXIO means the *source* is not a device file, skip creation in that case */ |
af984e13 | 922 | if (r < 0 && r != -ENXIO) |
2b85f4e1 | 923 | goto fail; |
7f112f50 LP |
924 | } |
925 | ||
2e4a4fae YW |
926 | r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID); |
927 | if (r < 0) | |
105a1a36 | 928 | log_debug_errno(r, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount); |
7f112f50 | 929 | |
ee818b89 AC |
930 | /* Create the /dev directory if missing. It is more likely to be |
931 | * missing when the service is started with RootDirectory. This is | |
932 | * consistent with mount units creating the mount points when missing. | |
933 | */ | |
34de407a | 934 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
ee818b89 | 935 | |
9e5f8252 | 936 | /* Unmount everything in old /dev */ |
2e4a4fae YW |
937 | r = umount_recursive(mount_entry_path(m), 0); |
938 | if (r < 0) | |
939 | log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m)); | |
940 | ||
21935150 LP |
941 | r = mount_nofollow_verbose(LOG_DEBUG, dev, mount_entry_path(m), NULL, MS_MOVE, NULL); |
942 | if (r < 0) | |
2b85f4e1 | 943 | goto fail; |
7f112f50 | 944 | |
1019a48f LP |
945 | (void) rmdir(dev); |
946 | (void) rmdir(temporary_mount); | |
7f112f50 | 947 | |
2b85f4e1 | 948 | return 0; |
7f112f50 | 949 | |
2b85f4e1 LP |
950 | fail: |
951 | if (devpts) | |
21935150 | 952 | (void) umount_verbose(LOG_DEBUG, devpts, UMOUNT_NOFOLLOW); |
7f112f50 | 953 | |
2b85f4e1 | 954 | if (devshm) |
21935150 | 955 | (void) umount_verbose(LOG_DEBUG, devshm, UMOUNT_NOFOLLOW); |
7f112f50 | 956 | |
2b85f4e1 | 957 | if (devhugepages) |
21935150 | 958 | (void) umount_verbose(LOG_DEBUG, devhugepages, UMOUNT_NOFOLLOW); |
7f112f50 | 959 | |
2b85f4e1 | 960 | if (devmqueue) |
21935150 | 961 | (void) umount_verbose(LOG_DEBUG, devmqueue, UMOUNT_NOFOLLOW); |
7f112f50 | 962 | |
21935150 | 963 | (void) umount_verbose(LOG_DEBUG, dev, UMOUNT_NOFOLLOW); |
1019a48f LP |
964 | (void) rmdir(dev); |
965 | (void) rmdir(temporary_mount); | |
7f112f50 | 966 | |
2b85f4e1 | 967 | return r; |
7f112f50 LP |
968 | } |
969 | ||
2a2969fd | 970 | static int mount_bind_dev(const MountEntry *m) { |
5d997827 LP |
971 | int r; |
972 | ||
973 | assert(m); | |
974 | ||
975 | /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's | |
976 | * /dev. This is only used when RootDirectory= is set. */ | |
977 | ||
645767d6 LP |
978 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
979 | ||
5d997827 LP |
980 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
981 | if (r < 0) | |
982 | return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m"); | |
983 | if (r > 0) /* make this a NOP if /dev is already a mount point */ | |
984 | return 0; | |
985 | ||
21935150 LP |
986 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
987 | if (r < 0) | |
988 | return r; | |
5d997827 LP |
989 | |
990 | return 1; | |
991 | } | |
992 | ||
2a2969fd | 993 | static int mount_sysfs(const MountEntry *m) { |
5d997827 LP |
994 | int r; |
995 | ||
996 | assert(m); | |
997 | ||
645767d6 LP |
998 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
999 | ||
5d997827 LP |
1000 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1001 | if (r < 0) | |
1002 | return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m"); | |
1003 | if (r > 0) /* make this a NOP if /sys is already a mount point */ | |
1004 | return 0; | |
1005 | ||
1006 | /* Bind mount the host's version so that we get all child mounts of it, too. */ | |
21935150 LP |
1007 | r = mount_nofollow_verbose(LOG_DEBUG, "/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
1008 | if (r < 0) | |
1009 | return r; | |
5d997827 LP |
1010 | |
1011 | return 1; | |
1012 | } | |
1013 | ||
4e399953 | 1014 | static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) { |
61f8a7bd | 1015 | _cleanup_free_ char *opts = NULL; |
4e399953 | 1016 | const char *entry_path; |
61f8a7bd | 1017 | int r, n; |
5d997827 LP |
1018 | |
1019 | assert(m); | |
4e399953 | 1020 | assert(ns_info); |
5d997827 | 1021 | |
4e399953 LP |
1022 | if (ns_info->protect_proc != PROTECT_PROC_DEFAULT || |
1023 | ns_info->proc_subset != PROC_SUBSET_ALL) { | |
4e399953 LP |
1024 | |
1025 | /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it | |
1026 | * pretended to be per-instance but actually was per-namespace), hence let's make use of it | |
1027 | * if requested. To make sure this logic succeeds only on kernels where hidepid= is | |
1028 | * per-instance, we'll exclusively use the textual value for hidepid=, since support was | |
1029 | * added in the same commit: if it's supported it is thus also per-instance. */ | |
1030 | ||
1031 | opts = strjoin("hidepid=", | |
1032 | ns_info->protect_proc == PROTECT_PROC_DEFAULT ? "off" : | |
1033 | protect_proc_to_string(ns_info->protect_proc), | |
1034 | ns_info->proc_subset == PROC_SUBSET_PID ? ",subset=pid" : ""); | |
1035 | if (!opts) | |
1036 | return -ENOMEM; | |
4e399953 LP |
1037 | } |
1038 | ||
61f8a7bd YW |
1039 | entry_path = mount_entry_path(m); |
1040 | (void) mkdir_p_label(entry_path, 0755); | |
1041 | ||
1042 | /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in | |
1043 | * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by | |
1044 | * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything | |
1045 | * mounted on /proc/ first. */ | |
1046 | ||
1047 | n = umount_recursive(entry_path, 0); | |
1048 | ||
1049 | r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, opts); | |
1050 | if (r == -EINVAL && opts) | |
1051 | /* If this failed with EINVAL then this likely means the textual hidepid= stuff is | |
1052 | * not supported by the kernel, and thus the per-instance hidepid= neither, which | |
1053 | * means we really don't want to use it, since it would affect our host's /proc | |
1054 | * mount. Hence let's gracefully fallback to a classic, unrestricted version. */ | |
1055 | r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL); | |
1056 | if (r == -EPERM) { | |
24ee0f9d | 1057 | /* When we do not have enough privileges to mount /proc, fallback to use existing /proc. */ |
61f8a7bd YW |
1058 | |
1059 | if (n > 0) | |
1060 | /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree. | |
1061 | * Propagate the original error code returned by mount() in the above. */ | |
1062 | return -EPERM; | |
1063 | ||
1064 | r = path_is_mount_point(entry_path, NULL, 0); | |
1065 | if (r < 0) | |
1066 | return log_debug_errno(r, "Unable to determine whether /proc is already mounted: %m"); | |
1067 | if (r == 0) | |
1068 | /* /proc is not mounted. Propagate the original error code. */ | |
1069 | return -EPERM; | |
ad74f28a YW |
1070 | } else if (r < 0) |
1071 | return r; | |
5d997827 LP |
1072 | |
1073 | return 1; | |
1074 | } | |
1075 | ||
2abd4e38 | 1076 | static int mount_tmpfs(const MountEntry *m) { |
df6b900a | 1077 | const char *entry_path, *inner_path; |
abad72be | 1078 | int r; |
abad72be | 1079 | |
6c47cd7d LP |
1080 | assert(m); |
1081 | ||
df6b900a | 1082 | entry_path = mount_entry_path(m); |
809ceb82 | 1083 | inner_path = mount_entry_unprefixed_path(m); |
df6b900a | 1084 | |
2abd4e38 | 1085 | /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */ |
6c47cd7d | 1086 | |
abad72be CG |
1087 | (void) mkdir_p_label(entry_path, 0755); |
1088 | (void) umount_recursive(entry_path, 0); | |
6c47cd7d | 1089 | |
21935150 LP |
1090 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", entry_path, "tmpfs", m->flags, mount_entry_options(m)); |
1091 | if (r < 0) | |
1092 | return r; | |
abad72be | 1093 | |
df6b900a | 1094 | r = label_fix_container(entry_path, inner_path, 0); |
abad72be | 1095 | if (r < 0) |
df6b900a | 1096 | return log_debug_errno(r, "Failed to fix label of '%s' as '%s': %m", entry_path, inner_path); |
6c47cd7d LP |
1097 | |
1098 | return 1; | |
1099 | } | |
1100 | ||
94293d65 LB |
1101 | static int mount_run(const MountEntry *m) { |
1102 | int r; | |
1103 | ||
1104 | assert(m); | |
1105 | ||
1106 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); | |
1107 | if (r < 0 && r != -ENOENT) | |
1108 | return log_debug_errno(r, "Unable to determine whether /run is already mounted: %m"); | |
1109 | if (r > 0) /* make this a NOP if /run is already a mount point */ | |
1110 | return 0; | |
1111 | ||
1112 | return mount_tmpfs(m); | |
1113 | } | |
1114 | ||
93f59701 LB |
1115 | static int mount_image(const MountEntry *m, const char *root_directory) { |
1116 | ||
1117 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
1118 | *host_os_release_sysext_level = NULL; | |
b3d13314 LB |
1119 | int r; |
1120 | ||
89e62e0b LP |
1121 | assert(m); |
1122 | ||
93f59701 LB |
1123 | if (m->mode == EXTENSION_IMAGES) { |
1124 | r = parse_os_release( | |
1125 | empty_to_root(root_directory), | |
1126 | "ID", &host_os_release_id, | |
1127 | "VERSION_ID", &host_os_release_version_id, | |
1128 | "SYSEXT_LEVEL", &host_os_release_sysext_level, | |
1129 | NULL); | |
1130 | if (r < 0) | |
1131 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
1132 | } | |
1133 | ||
1134 | r = verity_dissect_and_mount( | |
1135 | mount_entry_source(m), mount_entry_path(m), m->image_options, | |
1136 | host_os_release_id, host_os_release_version_id, host_os_release_sysext_level); | |
b850a9b2 LB |
1137 | if (r == -ENOENT && m->ignore) |
1138 | return 0; | |
b3d13314 | 1139 | if (r < 0) |
4beda316 | 1140 | return log_debug_errno(r, "Failed to mount image %s on %s: %m", mount_entry_source(m), mount_entry_path(m)); |
b3d13314 LB |
1141 | |
1142 | return 1; | |
1143 | } | |
1144 | ||
93f59701 LB |
1145 | static int mount_overlay(const MountEntry *m) { |
1146 | const char *options; | |
1147 | int r; | |
1148 | ||
1149 | assert(m); | |
1150 | ||
1151 | options = strjoina("lowerdir=", mount_entry_options(m)); | |
1152 | ||
1153 | (void) mkdir_p_label(mount_entry_path(m), 0755); | |
1154 | ||
1155 | r = mount_nofollow_verbose(LOG_DEBUG, "overlay", mount_entry_path(m), "overlay", MS_RDONLY, options); | |
1156 | if (r == -ENOENT && m->ignore) | |
1157 | return 0; | |
1158 | if (r < 0) | |
1159 | return r; | |
1160 | ||
1161 | return 1; | |
1162 | } | |
1163 | ||
088696fe | 1164 | static int follow_symlink( |
d2d6c096 | 1165 | const char *root_directory, |
088696fe | 1166 | MountEntry *m) { |
d2d6c096 | 1167 | |
088696fe | 1168 | _cleanup_free_ char *target = NULL; |
8fceda93 LP |
1169 | int r; |
1170 | ||
088696fe LP |
1171 | /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we |
1172 | * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at | |
1173 | * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the | |
1174 | * end and already have a fully normalized name. */ | |
8fceda93 | 1175 | |
a5648b80 | 1176 | r = chase_symlinks(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL); |
088696fe LP |
1177 | if (r < 0) |
1178 | return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m)); | |
1179 | if (r > 0) /* Reached the end, nothing more to resolve */ | |
1180 | return 1; | |
8fceda93 | 1181 | |
baaa35ad ZJS |
1182 | if (m->n_followed >= CHASE_SYMLINKS_MAX) /* put a boundary on things */ |
1183 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), | |
1184 | "Symlink loop on '%s'.", | |
1185 | mount_entry_path(m)); | |
8fceda93 | 1186 | |
088696fe | 1187 | log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m), target); |
8fceda93 | 1188 | |
809ceb82 | 1189 | mount_entry_consume_prefix(m, TAKE_PTR(target)); |
8fceda93 | 1190 | |
088696fe LP |
1191 | m->n_followed ++; |
1192 | ||
1193 | return 0; | |
8fceda93 LP |
1194 | } |
1195 | ||
82fb2da2 | 1196 | static int apply_one_mount( |
8fceda93 | 1197 | const char *root_directory, |
4e399953 LP |
1198 | MountEntry *m, |
1199 | const NamespaceInfo *ns_info) { | |
ac0930c8 | 1200 | |
e5f10caf | 1201 | _cleanup_free_ char *inaccessible = NULL; |
a227a4be | 1202 | bool rbind = true, make = false; |
15ae422b | 1203 | const char *what; |
15ae422b | 1204 | int r; |
15ae422b | 1205 | |
c17ec25e | 1206 | assert(m); |
4e399953 | 1207 | assert(ns_info); |
15ae422b | 1208 | |
34de407a | 1209 | log_debug("Applying namespace mount on %s", mount_entry_path(m)); |
fe3c2583 | 1210 | |
c17ec25e | 1211 | switch (m->mode) { |
15ae422b | 1212 | |
160cfdbe | 1213 | case INACCESSIBLE: { |
e5f10caf AZ |
1214 | _cleanup_free_ char *tmp = NULL; |
1215 | const char *runtime_dir; | |
160cfdbe | 1216 | struct stat target; |
6d313367 LP |
1217 | |
1218 | /* First, get rid of everything that is below if there | |
1219 | * is anything... Then, overmount it with an | |
c4b41707 | 1220 | * inaccessible path. */ |
34de407a | 1221 | (void) umount_recursive(mount_entry_path(m), 0); |
6d313367 | 1222 | |
088696fe LP |
1223 | if (lstat(mount_entry_path(m), &target) < 0) { |
1224 | if (errno == ENOENT && m->ignore) | |
1225 | return 0; | |
1226 | ||
cbc056c8 ZJS |
1227 | return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m", |
1228 | mount_entry_path(m)); | |
088696fe | 1229 | } |
15ae422b | 1230 | |
e5f10caf | 1231 | if (geteuid() == 0) |
48b747fa | 1232 | runtime_dir = "/run"; |
e5f10caf | 1233 | else { |
48b747fa LP |
1234 | if (asprintf(&tmp, "/run/user/" UID_FMT, geteuid()) < 0) |
1235 | return -ENOMEM; | |
e5f10caf AZ |
1236 | |
1237 | runtime_dir = tmp; | |
1238 | } | |
1239 | ||
1240 | r = mode_to_inaccessible_node(runtime_dir, target.st_mode, &inaccessible); | |
1241 | if (r < 0) | |
baaa35ad ZJS |
1242 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1243 | "File type not supported for inaccessible mounts. Note that symlinks are not allowed"); | |
e5f10caf | 1244 | what = inaccessible; |
c4b41707 | 1245 | break; |
160cfdbe | 1246 | } |
fe3c2583 | 1247 | |
15ae422b | 1248 | case READONLY: |
15ae422b | 1249 | case READWRITE: |
1e05071d | 1250 | case READWRITE_IMPLICIT: |
ddc155b2 TM |
1251 | case EXEC: |
1252 | case NOEXEC: | |
8fceda93 | 1253 | r = path_is_mount_point(mount_entry_path(m), root_directory, 0); |
088696fe LP |
1254 | if (r == -ENOENT && m->ignore) |
1255 | return 0; | |
d944dc95 | 1256 | if (r < 0) |
cbc056c8 ZJS |
1257 | return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m", |
1258 | mount_entry_path(m)); | |
1259 | if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY | |
ddc155b2 | 1260 | * and MS_NOEXEC bits for the mount point if needed. */ |
6b7c9f8b | 1261 | return 0; |
6b7c9f8b | 1262 | /* This isn't a mount point yet, let's make it one. */ |
34de407a | 1263 | what = mount_entry_path(m); |
6b7c9f8b | 1264 | break; |
15ae422b | 1265 | |
d2d6c096 LP |
1266 | case BIND_MOUNT: |
1267 | rbind = false; | |
d2d6c096 | 1268 | |
4831981d | 1269 | _fallthrough_; |
088696fe LP |
1270 | case BIND_MOUNT_RECURSIVE: { |
1271 | _cleanup_free_ char *chased = NULL; | |
5d997827 | 1272 | |
cbc056c8 ZJS |
1273 | /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note |
1274 | * that bind mount source paths are always relative to the host root, hence we pass NULL as | |
1275 | * root directory to chase_symlinks() here. */ | |
088696fe | 1276 | |
a5648b80 | 1277 | r = chase_symlinks(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL); |
088696fe LP |
1278 | if (r == -ENOENT && m->ignore) { |
1279 | log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m)); | |
1280 | return 0; | |
1281 | } | |
1282 | if (r < 0) | |
1283 | return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m)); | |
1284 | ||
1285 | log_debug("Followed source symlinks %s → %s.", mount_entry_source(m), chased); | |
1286 | ||
1287 | free_and_replace(m->source_malloc, chased); | |
d2d6c096 LP |
1288 | |
1289 | what = mount_entry_source(m); | |
a227a4be | 1290 | make = true; |
d2d6c096 | 1291 | break; |
088696fe | 1292 | } |
d2d6c096 | 1293 | |
6c47cd7d | 1294 | case EMPTY_DIR: |
2abd4e38 YW |
1295 | case TMPFS: |
1296 | return mount_tmpfs(m); | |
6c47cd7d | 1297 | |
ac0930c8 | 1298 | case PRIVATE_TMP: |
56a13a49 | 1299 | case PRIVATE_TMP_READONLY: |
89bd586c | 1300 | what = mount_entry_source(m); |
a227a4be | 1301 | make = true; |
15ae422b | 1302 | break; |
e364ad06 | 1303 | |
d6797c92 | 1304 | case PRIVATE_DEV: |
5d997827 LP |
1305 | return mount_private_dev(m); |
1306 | ||
1307 | case BIND_DEV: | |
1308 | return mount_bind_dev(m); | |
1309 | ||
1310 | case SYSFS: | |
1311 | return mount_sysfs(m); | |
1312 | ||
1313 | case PROCFS: | |
4e399953 | 1314 | return mount_procfs(m, ns_info); |
d6797c92 | 1315 | |
94293d65 LB |
1316 | case RUN: |
1317 | return mount_run(m); | |
1318 | ||
b3d13314 | 1319 | case MOUNT_IMAGES: |
93f59701 LB |
1320 | return mount_image(m, NULL); |
1321 | ||
1322 | case EXTENSION_IMAGES: | |
1323 | return mount_image(m, root_directory); | |
1324 | ||
1325 | case OVERLAY_MOUNT: | |
1326 | return mount_overlay(m); | |
b3d13314 | 1327 | |
e364ad06 LP |
1328 | default: |
1329 | assert_not_reached("Unknown mode"); | |
15ae422b LP |
1330 | } |
1331 | ||
ac0930c8 | 1332 | assert(what); |
15ae422b | 1333 | |
21935150 LP |
1334 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); |
1335 | if (r < 0) { | |
a227a4be | 1336 | bool try_again = false; |
a227a4be LP |
1337 | |
1338 | if (r == -ENOENT && make) { | |
8bab8029 | 1339 | int q; |
a227a4be | 1340 | |
cbc056c8 ZJS |
1341 | /* Hmm, either the source or the destination are missing. Let's see if we can create |
1342 | the destination, then try again. */ | |
a227a4be | 1343 | |
8bab8029 | 1344 | (void) mkdir_parents(mount_entry_path(m), 0755); |
a227a4be | 1345 | |
8bab8029 LB |
1346 | q = make_mount_point_inode_from_path(what, mount_entry_path(m), 0755); |
1347 | if (q < 0) | |
1348 | log_error_errno(q, "Failed to create destination mount point node '%s': %m", | |
1349 | mount_entry_path(m)); | |
1350 | else | |
1351 | try_again = true; | |
a227a4be LP |
1352 | } |
1353 | ||
21935150 LP |
1354 | if (try_again) |
1355 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); | |
a227a4be | 1356 | if (r < 0) |
5dc60faa | 1357 | return log_error_errno(r, "Failed to mount %s to %s: %m", what, mount_entry_path(m)); |
a227a4be | 1358 | } |
6b7c9f8b | 1359 | |
34de407a | 1360 | log_debug("Successfully mounted %s to %s", what, mount_entry_path(m)); |
6b7c9f8b | 1361 | return 0; |
ac0930c8 | 1362 | } |
15ae422b | 1363 | |
6b000af4 | 1364 | static int make_read_only(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
9ce4e4b0 | 1365 | unsigned long new_flags = 0, flags_mask = 0; |
763a260a | 1366 | bool submounts = false; |
6b7c9f8b | 1367 | int r = 0; |
15ae422b | 1368 | |
c17ec25e | 1369 | assert(m); |
ac9de0b3 | 1370 | assert(proc_self_mountinfo); |
ac0930c8 | 1371 | |
9ce4e4b0 LP |
1372 | if (mount_entry_read_only(m) || m->mode == PRIVATE_DEV) { |
1373 | new_flags |= MS_RDONLY; | |
1374 | flags_mask |= MS_RDONLY; | |
1375 | } | |
1376 | ||
1377 | if (m->nosuid) { | |
1378 | new_flags |= MS_NOSUID; | |
1379 | flags_mask |= MS_NOSUID; | |
1380 | } | |
1381 | ||
1382 | if (flags_mask == 0) /* No Change? */ | |
6b7c9f8b LP |
1383 | return 0; |
1384 | ||
9ce4e4b0 LP |
1385 | /* We generally apply these changes recursively, except for /dev, and the cases we know there's |
1386 | * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the | |
1387 | * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace | |
1388 | * and running Linux <= 4.17. */ | |
1389 | submounts = | |
1390 | mount_entry_read_only(m) && | |
1391 | !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1392 | if (submounts) | |
6b000af4 | 1393 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); |
9ce4e4b0 | 1394 | else |
7cce68e1 | 1395 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); |
9ce4e4b0 | 1396 | |
ddc155b2 | 1397 | /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked |
867189b5 LP |
1398 | * read-only already stays this way. This improves compatibility with container managers, where we |
1399 | * won't attempt to undo read-only mounts already applied. */ | |
ac0930c8 | 1400 | |
8fceda93 | 1401 | if (r == -ENOENT && m->ignore) |
867189b5 | 1402 | return 0; |
763a260a | 1403 | if (r < 0) |
9ce4e4b0 | 1404 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), |
763a260a | 1405 | submounts ? " and its submounts" : ""); |
763a260a | 1406 | return 0; |
d944dc95 LP |
1407 | } |
1408 | ||
ddc155b2 TM |
1409 | static int make_noexec(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
1410 | unsigned long new_flags = 0, flags_mask = 0; | |
1411 | bool submounts = false; | |
1412 | int r = 0; | |
1413 | ||
1414 | assert(m); | |
1415 | assert(proc_self_mountinfo); | |
1416 | ||
1417 | if (mount_entry_noexec(m)) { | |
1418 | new_flags |= MS_NOEXEC; | |
1419 | flags_mask |= MS_NOEXEC; | |
1420 | } else if (mount_entry_exec(m)) { | |
1421 | new_flags &= ~MS_NOEXEC; | |
1422 | flags_mask |= MS_NOEXEC; | |
1423 | } | |
1424 | ||
1425 | if (flags_mask == 0) /* No Change? */ | |
1426 | return 0; | |
1427 | ||
1428 | submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1429 | ||
1430 | if (submounts) | |
1431 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); | |
1432 | else | |
1433 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); | |
1434 | ||
1435 | if (r == -ENOENT && m->ignore) | |
1436 | return 0; | |
1437 | if (r < 0) | |
1438 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1439 | submounts ? " and its submounts" : ""); | |
1440 | return 0; | |
1441 | } | |
1442 | ||
9b68367b | 1443 | static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) { |
5d997827 LP |
1444 | assert(ns_info); |
1445 | ||
9c988f93 DH |
1446 | /* |
1447 | * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=, | |
1448 | * since to protect the API VFS mounts, they need to be around in the | |
9b68367b | 1449 | * first place... |
9c988f93 | 1450 | */ |
5d997827 | 1451 | |
9b68367b YW |
1452 | return ns_info->mount_apivfs || |
1453 | ns_info->protect_control_groups || | |
4e399953 LP |
1454 | ns_info->protect_kernel_tunables || |
1455 | ns_info->protect_proc != PROTECT_PROC_DEFAULT || | |
1456 | ns_info->proc_subset != PROC_SUBSET_ALL; | |
5d997827 LP |
1457 | } |
1458 | ||
da6053d0 | 1459 | static size_t namespace_calculate_mounts( |
bb0ff3fb | 1460 | const NamespaceInfo *ns_info, |
2652c6c1 DH |
1461 | char** read_write_paths, |
1462 | char** read_only_paths, | |
1463 | char** inaccessible_paths, | |
ddc155b2 TM |
1464 | char** exec_paths, |
1465 | char** no_exec_paths, | |
6c47cd7d | 1466 | char** empty_directories, |
da6053d0 LP |
1467 | size_t n_bind_mounts, |
1468 | size_t n_temporary_filesystems, | |
b3d13314 | 1469 | size_t n_mount_images, |
93f59701 LB |
1470 | size_t n_extension_images, |
1471 | size_t n_hierarchies, | |
2652c6c1 DH |
1472 | const char* tmp_dir, |
1473 | const char* var_tmp_dir, | |
bbb4e7f3 | 1474 | const char *creds_path, |
5e8deb94 | 1475 | const char* log_namespace, |
3bdc25a4 LP |
1476 | bool setup_propagate, |
1477 | const char* notify_socket) { | |
2652c6c1 | 1478 | |
da6053d0 LP |
1479 | size_t protect_home_cnt; |
1480 | size_t protect_system_cnt = | |
52b3d652 | 1481 | (ns_info->protect_system == PROTECT_SYSTEM_STRICT ? |
f471b2af | 1482 | ELEMENTSOF(protect_system_strict_table) : |
52b3d652 | 1483 | ((ns_info->protect_system == PROTECT_SYSTEM_FULL) ? |
f471b2af | 1484 | ELEMENTSOF(protect_system_full_table) : |
52b3d652 | 1485 | ((ns_info->protect_system == PROTECT_SYSTEM_YES) ? |
f471b2af DH |
1486 | ELEMENTSOF(protect_system_yes_table) : 0))); |
1487 | ||
b6c432ca | 1488 | protect_home_cnt = |
52b3d652 | 1489 | (ns_info->protect_home == PROTECT_HOME_YES ? |
b6c432ca | 1490 | ELEMENTSOF(protect_home_yes_table) : |
52b3d652 | 1491 | ((ns_info->protect_home == PROTECT_HOME_READ_ONLY) ? |
e4da7d8c | 1492 | ELEMENTSOF(protect_home_read_only_table) : |
52b3d652 | 1493 | ((ns_info->protect_home == PROTECT_HOME_TMPFS) ? |
e4da7d8c | 1494 | ELEMENTSOF(protect_home_tmpfs_table) : 0))); |
b6c432ca | 1495 | |
2652c6c1 DH |
1496 | return !!tmp_dir + !!var_tmp_dir + |
1497 | strv_length(read_write_paths) + | |
1498 | strv_length(read_only_paths) + | |
1499 | strv_length(inaccessible_paths) + | |
ddc155b2 TM |
1500 | strv_length(exec_paths) + |
1501 | strv_length(no_exec_paths) + | |
6c47cd7d | 1502 | strv_length(empty_directories) + |
d2d6c096 | 1503 | n_bind_mounts + |
b3d13314 | 1504 | n_mount_images + |
93f59701 | 1505 | (n_extension_images > 0 ? n_hierarchies + n_extension_images : 0) + /* Mount each image plus an overlay per hierarchy */ |
2abd4e38 | 1506 | n_temporary_filesystems + |
c575770b DH |
1507 | ns_info->private_dev + |
1508 | (ns_info->protect_kernel_tunables ? ELEMENTSOF(protect_kernel_tunables_table) : 0) + | |
c575770b | 1509 | (ns_info->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) + |
94a7b275 KK |
1510 | (ns_info->protect_kernel_logs ? ELEMENTSOF(protect_kernel_logs_table) : 0) + |
1511 | (ns_info->protect_control_groups ? 1 : 0) + | |
5d997827 | 1512 | protect_home_cnt + protect_system_cnt + |
aecd5ac6 | 1513 | (ns_info->protect_hostname ? 2 : 0) + |
91dd5f7c | 1514 | (namespace_info_mount_apivfs(ns_info) ? ELEMENTSOF(apivfs_table) : 0) + |
bbb4e7f3 | 1515 | (creds_path ? 2 : 1) + |
3bdc25a4 LP |
1516 | !!log_namespace + |
1517 | setup_propagate + /* /run/systemd/incoming */ | |
1518 | !!notify_socket; | |
2652c6c1 DH |
1519 | } |
1520 | ||
da6053d0 | 1521 | static void normalize_mounts(const char *root_directory, MountEntry *mounts, size_t *n_mounts) { |
9b68367b | 1522 | assert(root_directory); |
f8b64b57 LP |
1523 | assert(n_mounts); |
1524 | assert(mounts || *n_mounts == 0); | |
1525 | ||
93bab288 | 1526 | typesafe_qsort(mounts, *n_mounts, mount_path_compare); |
f8b64b57 LP |
1527 | |
1528 | drop_duplicates(mounts, n_mounts); | |
1529 | drop_outside_root(root_directory, mounts, n_mounts); | |
1530 | drop_inaccessible(mounts, n_mounts); | |
1531 | drop_nop(mounts, n_mounts); | |
1532 | } | |
1533 | ||
82fb2da2 LB |
1534 | static int apply_mounts( |
1535 | const char *root, | |
1536 | const NamespaceInfo *ns_info, | |
1537 | MountEntry *mounts, | |
1538 | size_t *n_mounts, | |
1539 | char **error_path) { | |
1540 | ||
1541 | _cleanup_fclose_ FILE *proc_self_mountinfo = NULL; | |
1542 | _cleanup_free_ char **deny_list = NULL; | |
82fb2da2 LB |
1543 | int r; |
1544 | ||
1545 | if (n_mounts == 0) /* Shortcut: nothing to do */ | |
1546 | return 0; | |
1547 | ||
1548 | assert(root); | |
1549 | assert(mounts); | |
1550 | assert(n_mounts); | |
1551 | ||
1552 | /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of | |
1553 | * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */ | |
1554 | proc_self_mountinfo = fopen("/proc/self/mountinfo", "re"); | |
1555 | if (!proc_self_mountinfo) { | |
d60e3b40 LP |
1556 | r = -errno; |
1557 | ||
82fb2da2 LB |
1558 | if (error_path) |
1559 | *error_path = strdup("/proc/self/mountinfo"); | |
d60e3b40 LP |
1560 | |
1561 | return log_debug_errno(r, "Failed to open /proc/self/mountinfo: %m"); | |
82fb2da2 LB |
1562 | } |
1563 | ||
1564 | /* First round, establish all mounts we need */ | |
1565 | for (;;) { | |
1566 | bool again = false; | |
1567 | ||
1568 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1569 | ||
1570 | if (m->applied) | |
1571 | continue; | |
1572 | ||
93f59701 LB |
1573 | /* ExtensionImages are first opened in the propagate directory, not in the root_directory */ |
1574 | r = follow_symlink(m->mode != EXTENSION_IMAGES ? root : NULL, m); | |
82fb2da2 LB |
1575 | if (r < 0) { |
1576 | if (error_path && mount_entry_path(m)) | |
1577 | *error_path = strdup(mount_entry_path(m)); | |
1578 | return r; | |
1579 | } | |
1580 | if (r == 0) { | |
1581 | /* We hit a symlinked mount point. The entry got rewritten and might | |
1582 | * point to a very different place now. Let's normalize the changed | |
1583 | * list, and start from the beginning. After all to mount the entry | |
1584 | * at the new location we might need some other mounts first */ | |
1585 | again = true; | |
1586 | break; | |
1587 | } | |
1588 | ||
1589 | r = apply_one_mount(root, m, ns_info); | |
1590 | if (r < 0) { | |
1591 | if (error_path && mount_entry_path(m)) | |
1592 | *error_path = strdup(mount_entry_path(m)); | |
1593 | return r; | |
1594 | } | |
1595 | ||
1596 | m->applied = true; | |
1597 | } | |
1598 | ||
1599 | if (!again) | |
1600 | break; | |
1601 | ||
1602 | normalize_mounts(root, mounts, n_mounts); | |
1603 | } | |
1604 | ||
1605 | /* Create a deny list we can pass to bind_mount_recursive() */ | |
1606 | deny_list = new(char*, (*n_mounts)+1); | |
1607 | if (!deny_list) | |
1608 | return -ENOMEM; | |
8794e6db | 1609 | for (size_t j = 0; j < *n_mounts; j++) |
82fb2da2 | 1610 | deny_list[j] = (char*) mount_entry_path(mounts+j); |
8794e6db | 1611 | deny_list[*n_mounts] = NULL; |
82fb2da2 LB |
1612 | |
1613 | /* Second round, flip the ro bits if necessary. */ | |
1614 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1615 | r = make_read_only(m, deny_list, proc_self_mountinfo); | |
1616 | if (r < 0) { | |
1617 | if (error_path && mount_entry_path(m)) | |
1618 | *error_path = strdup(mount_entry_path(m)); | |
1619 | return r; | |
1620 | } | |
1621 | } | |
1622 | ||
1623 | /* Third round, flip the noexec bits with a simplified deny list. */ | |
8794e6db | 1624 | for (size_t j = 0; j < *n_mounts; j++) |
82fb2da2 LB |
1625 | if (IN_SET((mounts+j)->mode, EXEC, NOEXEC)) |
1626 | deny_list[j] = (char*) mount_entry_path(mounts+j); | |
8794e6db | 1627 | deny_list[*n_mounts] = NULL; |
82fb2da2 LB |
1628 | |
1629 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1630 | r = make_noexec(m, deny_list, proc_self_mountinfo); | |
1631 | if (r < 0) { | |
1632 | if (error_path && mount_entry_path(m)) | |
1633 | *error_path = strdup(mount_entry_path(m)); | |
1634 | return r; | |
1635 | } | |
1636 | } | |
1637 | ||
1638 | return 1; | |
1639 | } | |
1640 | ||
c8c535d5 LP |
1641 | static bool root_read_only( |
1642 | char **read_only_paths, | |
1643 | ProtectSystem protect_system) { | |
1644 | ||
1645 | /* Determine whether the root directory is going to be read-only given the configured settings. */ | |
1646 | ||
1647 | if (protect_system == PROTECT_SYSTEM_STRICT) | |
1648 | return true; | |
1649 | ||
de46b2be | 1650 | if (prefixed_path_strv_contains(read_only_paths, "/")) |
c8c535d5 LP |
1651 | return true; |
1652 | ||
1653 | return false; | |
1654 | } | |
1655 | ||
1656 | static bool home_read_only( | |
1657 | char** read_only_paths, | |
1658 | char** inaccessible_paths, | |
1659 | char** empty_directories, | |
1660 | const BindMount *bind_mounts, | |
1661 | size_t n_bind_mounts, | |
1662 | const TemporaryFileSystem *temporary_filesystems, | |
1663 | size_t n_temporary_filesystems, | |
1664 | ProtectHome protect_home) { | |
1665 | ||
c8c535d5 LP |
1666 | /* Determine whether the /home directory is going to be read-only given the configured settings. Yes, |
1667 | * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple | |
1668 | * settings. */ | |
1669 | ||
1670 | if (protect_home != PROTECT_HOME_NO) | |
1671 | return true; | |
1672 | ||
de46b2be TM |
1673 | if (prefixed_path_strv_contains(read_only_paths, "/home") || |
1674 | prefixed_path_strv_contains(inaccessible_paths, "/home") || | |
1675 | prefixed_path_strv_contains(empty_directories, "/home")) | |
c8c535d5 LP |
1676 | return true; |
1677 | ||
8794e6db | 1678 | for (size_t i = 0; i < n_temporary_filesystems; i++) |
c8c535d5 LP |
1679 | if (path_equal(temporary_filesystems[i].path, "/home")) |
1680 | return true; | |
1681 | ||
1682 | /* If /home is overmounted with some dir from the host it's not writable. */ | |
8794e6db | 1683 | for (size_t i = 0; i < n_bind_mounts; i++) |
c8c535d5 LP |
1684 | if (path_equal(bind_mounts[i].destination, "/home")) |
1685 | return true; | |
1686 | ||
1687 | return false; | |
1688 | } | |
1689 | ||
89e62e0b LP |
1690 | static int verity_settings_prepare( |
1691 | VeritySettings *verity, | |
1692 | const char *root_image, | |
1693 | const void *root_hash, | |
1694 | size_t root_hash_size, | |
1695 | const char *root_hash_path, | |
1696 | const void *root_hash_sig, | |
1697 | size_t root_hash_sig_size, | |
1698 | const char *root_hash_sig_path, | |
1699 | const char *verity_data_path) { | |
1700 | ||
1701 | int r; | |
1702 | ||
1703 | assert(verity); | |
1704 | ||
1705 | if (root_hash) { | |
1706 | void *d; | |
1707 | ||
1708 | d = memdup(root_hash, root_hash_size); | |
1709 | if (!d) | |
1710 | return -ENOMEM; | |
1711 | ||
1712 | free_and_replace(verity->root_hash, d); | |
1713 | verity->root_hash_size = root_hash_size; | |
aee36b4e | 1714 | verity->designator = PARTITION_ROOT; |
89e62e0b LP |
1715 | } |
1716 | ||
1717 | if (root_hash_sig) { | |
1718 | void *d; | |
1719 | ||
1720 | d = memdup(root_hash_sig, root_hash_sig_size); | |
1721 | if (!d) | |
1722 | return -ENOMEM; | |
1723 | ||
1724 | free_and_replace(verity->root_hash_sig, d); | |
1725 | verity->root_hash_sig_size = root_hash_sig_size; | |
aee36b4e | 1726 | verity->designator = PARTITION_ROOT; |
89e62e0b LP |
1727 | } |
1728 | ||
1729 | if (verity_data_path) { | |
1730 | r = free_and_strdup(&verity->data_path, verity_data_path); | |
1731 | if (r < 0) | |
1732 | return r; | |
1733 | } | |
1734 | ||
1735 | r = verity_settings_load( | |
1736 | verity, | |
1737 | root_image, | |
1738 | root_hash_path, | |
1739 | root_hash_sig_path); | |
1740 | if (r < 0) | |
1741 | return log_debug_errno(r, "Failed to load root hash: %m"); | |
1742 | ||
1743 | return 0; | |
1744 | } | |
1745 | ||
613b411c | 1746 | int setup_namespace( |
ee818b89 | 1747 | const char* root_directory, |
915e6d16 | 1748 | const char* root_image, |
18d73705 | 1749 | const MountOptions *root_image_options, |
bb0ff3fb | 1750 | const NamespaceInfo *ns_info, |
2a624c36 AP |
1751 | char** read_write_paths, |
1752 | char** read_only_paths, | |
1753 | char** inaccessible_paths, | |
ddc155b2 TM |
1754 | char** exec_paths, |
1755 | char** no_exec_paths, | |
6c47cd7d | 1756 | char** empty_directories, |
d2d6c096 | 1757 | const BindMount *bind_mounts, |
da6053d0 | 1758 | size_t n_bind_mounts, |
2abd4e38 | 1759 | const TemporaryFileSystem *temporary_filesystems, |
da6053d0 | 1760 | size_t n_temporary_filesystems, |
b3d13314 LB |
1761 | const MountImage *mount_images, |
1762 | size_t n_mount_images, | |
a004cb4c LP |
1763 | const char* tmp_dir, |
1764 | const char* var_tmp_dir, | |
bbb4e7f3 | 1765 | const char *creds_path, |
91dd5f7c | 1766 | const char *log_namespace, |
915e6d16 | 1767 | unsigned long mount_flags, |
0389f4fa LB |
1768 | const void *root_hash, |
1769 | size_t root_hash_size, | |
1770 | const char *root_hash_path, | |
d4d55b0d LB |
1771 | const void *root_hash_sig, |
1772 | size_t root_hash_sig_size, | |
1773 | const char *root_hash_sig_path, | |
89e62e0b | 1774 | const char *verity_data_path, |
93f59701 LB |
1775 | const MountImage *extension_images, |
1776 | size_t n_extension_images, | |
5e8deb94 LB |
1777 | const char *propagate_dir, |
1778 | const char *incoming_dir, | |
3bdc25a4 | 1779 | const char *notify_socket, |
7cc5ef5f ZJS |
1780 | DissectImageFlags dissect_image_flags, |
1781 | char **error_path) { | |
15ae422b | 1782 | |
915e6d16 | 1783 | _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL; |
78ebe980 | 1784 | _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL; |
915e6d16 | 1785 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; |
aee36b4e | 1786 | _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT; |
93f59701 | 1787 | _cleanup_strv_free_ char **hierarchies = NULL; |
5f7a690a | 1788 | MountEntry *m = NULL, *mounts = NULL; |
5e8deb94 | 1789 | bool require_prefix = false, setup_propagate = false; |
93f59701 | 1790 | const char *root, *extension_dir = "/run/systemd/unit-extensions"; |
89e62e0b LP |
1791 | size_t n_mounts; |
1792 | int r; | |
15ae422b | 1793 | |
915e6d16 LP |
1794 | assert(ns_info); |
1795 | ||
5e8deb94 LB |
1796 | if (!isempty(propagate_dir) && !isempty(incoming_dir)) |
1797 | setup_propagate = true; | |
1798 | ||
613b411c | 1799 | if (mount_flags == 0) |
c17ec25e | 1800 | mount_flags = MS_SHARED; |
ac0930c8 | 1801 | |
915e6d16 LP |
1802 | if (root_image) { |
1803 | dissect_image_flags |= DISSECT_IMAGE_REQUIRE_ROOT; | |
1804 | ||
c8c535d5 LP |
1805 | /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */ |
1806 | if (root_read_only(read_only_paths, | |
52b3d652 | 1807 | ns_info->protect_system) && |
c8c535d5 LP |
1808 | home_read_only(read_only_paths, inaccessible_paths, empty_directories, |
1809 | bind_mounts, n_bind_mounts, temporary_filesystems, n_temporary_filesystems, | |
52b3d652 | 1810 | ns_info->protect_home) && |
c9ef8573 | 1811 | strv_isempty(read_write_paths)) |
915e6d16 LP |
1812 | dissect_image_flags |= DISSECT_IMAGE_READ_ONLY; |
1813 | ||
89e62e0b LP |
1814 | r = verity_settings_prepare( |
1815 | &verity, | |
1816 | root_image, | |
1817 | root_hash, root_hash_size, root_hash_path, | |
1818 | root_hash_sig, root_hash_sig_size, root_hash_sig_path, | |
1819 | verity_data_path); | |
915e6d16 | 1820 | if (r < 0) |
89e62e0b LP |
1821 | return r; |
1822 | ||
1823 | SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, verity.data_path); | |
915e6d16 | 1824 | |
89e62e0b LP |
1825 | r = loop_device_make_by_path( |
1826 | root_image, | |
1827 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */, | |
1828 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN, | |
1829 | &loop_device); | |
78ebe980 | 1830 | if (r < 0) |
89e62e0b LP |
1831 | return log_debug_errno(r, "Failed to create loop device for root image: %m"); |
1832 | ||
1833 | r = dissect_image( | |
1834 | loop_device->fd, | |
1835 | &verity, | |
1836 | root_image_options, | |
1837 | dissect_image_flags, | |
1838 | &dissected_image); | |
78ebe980 | 1839 | if (r < 0) |
763a260a | 1840 | return log_debug_errno(r, "Failed to dissect image: %m"); |
78ebe980 | 1841 | |
89e62e0b LP |
1842 | r = dissected_image_decrypt( |
1843 | dissected_image, | |
1844 | NULL, | |
1845 | &verity, | |
1846 | dissect_image_flags, | |
1847 | &decrypted_image); | |
915e6d16 | 1848 | if (r < 0) |
763a260a | 1849 | return log_debug_errno(r, "Failed to decrypt dissected image: %m"); |
915e6d16 LP |
1850 | } |
1851 | ||
e908468b LP |
1852 | if (root_directory) |
1853 | root = root_directory; | |
0722b359 | 1854 | else { |
77f16dbd DDM |
1855 | /* /run/systemd should have been created by PID 1 early on already, but in some cases, like |
1856 | * when running tests (test-execute), it might not have been created yet so let's make sure | |
1857 | * we create it if it doesn't already exist. */ | |
1858 | (void) mkdir_p_label("/run/systemd", 0755); | |
1859 | ||
0722b359 JS |
1860 | /* Always create the mount namespace in a temporary directory, instead of operating |
1861 | * directly in the root. The temporary directory prevents any mounts from being | |
1862 | * potentially obscured my other mounts we already applied. | |
1863 | * We use the same mount point for all images, which is safe, since they all live | |
1864 | * in their own namespaces after all, and hence won't see each other. */ | |
e908468b LP |
1865 | |
1866 | root = "/run/systemd/unit-root"; | |
1867 | (void) mkdir_label(root, 0700); | |
d18aff04 | 1868 | require_prefix = true; |
0722b359 | 1869 | } |
e908468b | 1870 | |
93f59701 LB |
1871 | if (n_extension_images > 0) { |
1872 | r = parse_env_extension_hierarchies(&hierarchies); | |
1873 | if (r < 0) | |
1874 | return r; | |
1875 | } | |
1876 | ||
cfbeb4ef LP |
1877 | n_mounts = namespace_calculate_mounts( |
1878 | ns_info, | |
1879 | read_write_paths, | |
1880 | read_only_paths, | |
1881 | inaccessible_paths, | |
ddc155b2 TM |
1882 | exec_paths, |
1883 | no_exec_paths, | |
6c47cd7d | 1884 | empty_directories, |
f5c52a77 | 1885 | n_bind_mounts, |
2abd4e38 | 1886 | n_temporary_filesystems, |
b3d13314 | 1887 | n_mount_images, |
93f59701 LB |
1888 | n_extension_images, |
1889 | strv_length(hierarchies), | |
cfbeb4ef | 1890 | tmp_dir, var_tmp_dir, |
bbb4e7f3 | 1891 | creds_path, |
5e8deb94 | 1892 | log_namespace, |
3bdc25a4 LP |
1893 | setup_propagate, |
1894 | notify_socket); | |
613b411c | 1895 | |
f0a4feb0 | 1896 | if (n_mounts > 0) { |
5f7a690a LP |
1897 | m = mounts = new0(MountEntry, n_mounts); |
1898 | if (!mounts) | |
1899 | return -ENOMEM; | |
1900 | ||
d18aff04 | 1901 | r = append_access_mounts(&m, read_write_paths, READWRITE, require_prefix); |
613b411c | 1902 | if (r < 0) |
f0a4feb0 | 1903 | goto finish; |
613b411c | 1904 | |
d18aff04 | 1905 | r = append_access_mounts(&m, read_only_paths, READONLY, require_prefix); |
613b411c | 1906 | if (r < 0) |
f0a4feb0 | 1907 | goto finish; |
613b411c | 1908 | |
d18aff04 | 1909 | r = append_access_mounts(&m, inaccessible_paths, INACCESSIBLE, require_prefix); |
613b411c | 1910 | if (r < 0) |
f0a4feb0 | 1911 | goto finish; |
7ff7394d | 1912 | |
ddc155b2 TM |
1913 | r = append_access_mounts(&m, exec_paths, EXEC, require_prefix); |
1914 | if (r < 0) | |
1915 | goto finish; | |
1916 | ||
1917 | r = append_access_mounts(&m, no_exec_paths, NOEXEC, require_prefix); | |
1918 | if (r < 0) | |
1919 | goto finish; | |
1920 | ||
6c47cd7d LP |
1921 | r = append_empty_dir_mounts(&m, empty_directories); |
1922 | if (r < 0) | |
1923 | goto finish; | |
1924 | ||
d2d6c096 LP |
1925 | r = append_bind_mounts(&m, bind_mounts, n_bind_mounts); |
1926 | if (r < 0) | |
1927 | goto finish; | |
1928 | ||
2abd4e38 YW |
1929 | r = append_tmpfs_mounts(&m, temporary_filesystems, n_temporary_filesystems); |
1930 | if (r < 0) | |
1931 | goto finish; | |
1932 | ||
613b411c | 1933 | if (tmp_dir) { |
56a13a49 ZJS |
1934 | bool ro = streq(tmp_dir, RUN_SYSTEMD_EMPTY); |
1935 | ||
34de407a | 1936 | *(m++) = (MountEntry) { |
5327c910 | 1937 | .path_const = "/tmp", |
56a13a49 | 1938 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, |
89bd586c | 1939 | .source_const = tmp_dir, |
5327c910 | 1940 | }; |
613b411c | 1941 | } |
7ff7394d | 1942 | |
613b411c | 1943 | if (var_tmp_dir) { |
56a13a49 ZJS |
1944 | bool ro = streq(var_tmp_dir, RUN_SYSTEMD_EMPTY); |
1945 | ||
34de407a | 1946 | *(m++) = (MountEntry) { |
5327c910 | 1947 | .path_const = "/var/tmp", |
56a13a49 | 1948 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, |
89bd586c | 1949 | .source_const = var_tmp_dir, |
5327c910 | 1950 | }; |
7ff7394d | 1951 | } |
ac0930c8 | 1952 | |
b3d13314 LB |
1953 | r = append_mount_images(&m, mount_images, n_mount_images); |
1954 | if (r < 0) | |
1955 | goto finish; | |
1956 | ||
93f59701 LB |
1957 | r = append_extension_images(&m, root, extension_dir, hierarchies, extension_images, n_extension_images); |
1958 | if (r < 0) | |
1959 | goto finish; | |
1960 | ||
d46b79bb | 1961 | if (ns_info->private_dev) |
34de407a | 1962 | *(m++) = (MountEntry) { |
5327c910 LP |
1963 | .path_const = "/dev", |
1964 | .mode = PRIVATE_DEV, | |
9ce4e4b0 | 1965 | .flags = DEV_MOUNT_OPTIONS, |
5327c910 | 1966 | }; |
7f112f50 | 1967 | |
c575770b | 1968 | if (ns_info->protect_kernel_tunables) { |
cbc056c8 ZJS |
1969 | r = append_static_mounts(&m, |
1970 | protect_kernel_tunables_table, | |
1971 | ELEMENTSOF(protect_kernel_tunables_table), | |
1972 | ns_info->ignore_protect_paths); | |
c575770b | 1973 | if (r < 0) |
f0a4feb0 | 1974 | goto finish; |
c575770b DH |
1975 | } |
1976 | ||
1977 | if (ns_info->protect_kernel_modules) { | |
cbc056c8 ZJS |
1978 | r = append_static_mounts(&m, |
1979 | protect_kernel_modules_table, | |
1980 | ELEMENTSOF(protect_kernel_modules_table), | |
1981 | ns_info->ignore_protect_paths); | |
c575770b | 1982 | if (r < 0) |
f0a4feb0 | 1983 | goto finish; |
c575770b | 1984 | } |
59eeb84b | 1985 | |
94a7b275 | 1986 | if (ns_info->protect_kernel_logs) { |
cbc056c8 ZJS |
1987 | r = append_static_mounts(&m, |
1988 | protect_kernel_logs_table, | |
1989 | ELEMENTSOF(protect_kernel_logs_table), | |
1990 | ns_info->ignore_protect_paths); | |
94a7b275 KK |
1991 | if (r < 0) |
1992 | goto finish; | |
1993 | } | |
1994 | ||
d46b79bb | 1995 | if (ns_info->protect_control_groups) |
34de407a | 1996 | *(m++) = (MountEntry) { |
5327c910 LP |
1997 | .path_const = "/sys/fs/cgroup", |
1998 | .mode = READONLY, | |
1999 | }; | |
59eeb84b | 2000 | |
52b3d652 | 2001 | r = append_protect_home(&m, ns_info->protect_home, ns_info->ignore_protect_paths); |
b6c432ca | 2002 | if (r < 0) |
f0a4feb0 | 2003 | goto finish; |
417116f2 | 2004 | |
52b3d652 | 2005 | r = append_protect_system(&m, ns_info->protect_system, false); |
f471b2af | 2006 | if (r < 0) |
f0a4feb0 | 2007 | goto finish; |
417116f2 | 2008 | |
9b68367b | 2009 | if (namespace_info_mount_apivfs(ns_info)) { |
cbc056c8 ZJS |
2010 | r = append_static_mounts(&m, |
2011 | apivfs_table, | |
2012 | ELEMENTSOF(apivfs_table), | |
2013 | ns_info->ignore_protect_paths); | |
5d997827 LP |
2014 | if (r < 0) |
2015 | goto finish; | |
2016 | } | |
2017 | ||
aecd5ac6 TM |
2018 | if (ns_info->protect_hostname) { |
2019 | *(m++) = (MountEntry) { | |
2020 | .path_const = "/proc/sys/kernel/hostname", | |
2021 | .mode = READONLY, | |
2022 | }; | |
2023 | *(m++) = (MountEntry) { | |
2024 | .path_const = "/proc/sys/kernel/domainname", | |
2025 | .mode = READONLY, | |
2026 | }; | |
2027 | } | |
2028 | ||
bbb4e7f3 LP |
2029 | if (creds_path) { |
2030 | /* If our service has a credentials store configured, then bind that one in, but hide | |
2031 | * everything else. */ | |
2032 | ||
2033 | *(m++) = (MountEntry) { | |
2034 | .path_const = "/run/credentials", | |
2035 | .mode = TMPFS, | |
2036 | .read_only = true, | |
2037 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, | |
2038 | .flags = MS_NODEV|MS_STRICTATIME|MS_NOSUID|MS_NOEXEC, | |
2039 | }; | |
2040 | ||
2041 | *(m++) = (MountEntry) { | |
2042 | .path_const = creds_path, | |
2043 | .mode = BIND_MOUNT, | |
2044 | .read_only = true, | |
2045 | .source_const = creds_path, | |
2046 | }; | |
2047 | } else { | |
2048 | /* If our service has no credentials store configured, then make the whole | |
2049 | * credentials tree inaccessible wholesale. */ | |
2050 | ||
2051 | *(m++) = (MountEntry) { | |
2052 | .path_const = "/run/credentials", | |
2053 | .mode = INACCESSIBLE, | |
2054 | .ignore = true, | |
2055 | }; | |
2056 | } | |
2057 | ||
91dd5f7c LP |
2058 | if (log_namespace) { |
2059 | _cleanup_free_ char *q; | |
2060 | ||
2061 | q = strjoin("/run/systemd/journal.", log_namespace); | |
2062 | if (!q) { | |
2063 | r = -ENOMEM; | |
2064 | goto finish; | |
2065 | } | |
2066 | ||
2067 | *(m++) = (MountEntry) { | |
2068 | .path_const = "/run/systemd/journal", | |
2069 | .mode = BIND_MOUNT_RECURSIVE, | |
2070 | .read_only = true, | |
2071 | .source_malloc = TAKE_PTR(q), | |
2072 | }; | |
2073 | } | |
2074 | ||
5e8deb94 LB |
2075 | /* Will be used to add bind mounts at runtime */ |
2076 | if (setup_propagate) | |
2077 | *(m++) = (MountEntry) { | |
2078 | .source_const = propagate_dir, | |
2079 | .path_const = incoming_dir, | |
2080 | .mode = BIND_MOUNT, | |
2081 | .read_only = true, | |
2082 | }; | |
2083 | ||
3bdc25a4 LP |
2084 | if (notify_socket) |
2085 | *(m++) = (MountEntry) { | |
2086 | .path_const = notify_socket, | |
2087 | .source_const = notify_socket, | |
2088 | .mode = BIND_MOUNT, | |
2089 | .read_only = true, | |
2090 | }; | |
2091 | ||
f0a4feb0 | 2092 | assert(mounts + n_mounts == m); |
ac0930c8 | 2093 | |
5327c910 | 2094 | /* Prepend the root directory where that's necessary */ |
e908468b | 2095 | r = prefix_where_needed(mounts, n_mounts, root); |
5327c910 LP |
2096 | if (r < 0) |
2097 | goto finish; | |
2098 | ||
839f1877 | 2099 | normalize_mounts(root, mounts, &n_mounts); |
15ae422b LP |
2100 | } |
2101 | ||
1beab8b0 LP |
2102 | /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */ |
2103 | ||
d944dc95 | 2104 | if (unshare(CLONE_NEWNS) < 0) { |
763a260a | 2105 | r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m"); |
1beab8b0 | 2106 | if (IN_SET(r, -EACCES, -EPERM, -EOPNOTSUPP, -ENOSYS)) |
cbc056c8 ZJS |
2107 | /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter |
2108 | * in place that doesn't allow us to create namespaces (or a missing cap), then | |
2109 | * propagate a recognizable error back, which the caller can use to detect this case | |
2110 | * (and only this) and optionally continue without namespacing applied. */ | |
1beab8b0 LP |
2111 | r = -ENOANO; |
2112 | ||
d944dc95 LP |
2113 | goto finish; |
2114 | } | |
1e4e94c8 | 2115 | |
5e8deb94 LB |
2116 | /* Create the source directory to allow runtime propagation of mounts */ |
2117 | if (setup_propagate) | |
2118 | (void) mkdir_p(propagate_dir, 0600); | |
2119 | ||
93f59701 LB |
2120 | if (n_extension_images > 0) { |
2121 | /* ExtensionImages mountpoint directories will be created | |
2122 | * while parsing the mounts to create, so have the parent ready */ | |
2123 | (void) mkdir_p(extension_dir, 0600); | |
2124 | } | |
2125 | ||
9b68367b YW |
2126 | /* Remount / as SLAVE so that nothing now mounted in the namespace |
2127 | * shows up in the parent */ | |
2128 | if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) { | |
763a260a | 2129 | r = log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m"); |
9b68367b | 2130 | goto finish; |
ee818b89 AC |
2131 | } |
2132 | ||
915e6d16 | 2133 | if (root_image) { |
e908468b | 2134 | /* A root image is specified, mount it to the right place */ |
2d3a5a73 | 2135 | r = dissected_image_mount(dissected_image, root, UID_INVALID, dissect_image_flags); |
763a260a YW |
2136 | if (r < 0) { |
2137 | log_debug_errno(r, "Failed to mount root image: %m"); | |
915e6d16 | 2138 | goto finish; |
763a260a | 2139 | } |
915e6d16 | 2140 | |
07ce7407 TM |
2141 | if (decrypted_image) { |
2142 | r = decrypted_image_relinquish(decrypted_image); | |
763a260a YW |
2143 | if (r < 0) { |
2144 | log_debug_errno(r, "Failed to relinquish decrypted image: %m"); | |
07ce7407 | 2145 | goto finish; |
763a260a | 2146 | } |
07ce7407 | 2147 | } |
78ebe980 | 2148 | |
915e6d16 LP |
2149 | loop_device_relinquish(loop_device); |
2150 | ||
2151 | } else if (root_directory) { | |
2152 | ||
e908468b LP |
2153 | /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */ |
2154 | r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW); | |
763a260a YW |
2155 | if (r < 0) { |
2156 | log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root); | |
d944dc95 | 2157 | goto finish; |
763a260a | 2158 | } |
8f1ad200 | 2159 | if (r == 0) { |
21935150 LP |
2160 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); |
2161 | if (r < 0) | |
8f1ad200 | 2162 | goto finish; |
d944dc95 | 2163 | } |
e908468b | 2164 | |
9b68367b | 2165 | } else { |
e908468b | 2166 | /* Let's mount the main root directory to the root directory to use */ |
21935150 LP |
2167 | r = mount_nofollow_verbose(LOG_DEBUG, "/", root, NULL, MS_BIND|MS_REC, NULL); |
2168 | if (r < 0) | |
e908468b | 2169 | goto finish; |
ee818b89 | 2170 | } |
c2c13f2d | 2171 | |
4e0c20de LP |
2172 | /* Try to set up the new root directory before mounting anything else there. */ |
2173 | if (root_image || root_directory) | |
2174 | (void) base_filesystem_create(root, UID_INVALID, GID_INVALID); | |
2175 | ||
82fb2da2 LB |
2176 | /* Now make the magic happen */ |
2177 | r = apply_mounts(root, ns_info, mounts, &n_mounts, error_path); | |
2178 | if (r < 0) | |
2179 | goto finish; | |
15ae422b | 2180 | |
9b68367b YW |
2181 | /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */ |
2182 | r = mount_move_root(root); | |
763a260a YW |
2183 | if (r < 0) { |
2184 | log_debug_errno(r, "Failed to mount root with MS_MOVE: %m"); | |
9b68367b | 2185 | goto finish; |
763a260a | 2186 | } |
ee818b89 | 2187 | |
55fe7432 | 2188 | /* Remount / as the desired mode. Note that this will not |
c2c13f2d LP |
2189 | * reestablish propagation from our side to the host, since |
2190 | * what's disconnected is disconnected. */ | |
d944dc95 | 2191 | if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) { |
763a260a | 2192 | r = log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m"); |
d944dc95 LP |
2193 | goto finish; |
2194 | } | |
15ae422b | 2195 | |
5e8deb94 LB |
2196 | /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only |
2197 | * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */ | |
2198 | if (setup_propagate) { | |
2199 | r = mount(NULL, incoming_dir, NULL, MS_SLAVE, NULL); | |
2200 | if (r < 0) { | |
2201 | log_error_errno(r, "Failed to remount %s with MS_SLAVE: %m", incoming_dir); | |
2202 | goto finish; | |
2203 | } | |
2204 | } | |
2205 | ||
d944dc95 | 2206 | r = 0; |
15ae422b | 2207 | |
d944dc95 | 2208 | finish: |
0cd41757 LP |
2209 | if (n_mounts > 0) |
2210 | for (m = mounts; m < mounts + n_mounts; m++) | |
2211 | mount_entry_done(m); | |
613b411c | 2212 | |
5f7a690a LP |
2213 | free(mounts); |
2214 | ||
613b411c LP |
2215 | return r; |
2216 | } | |
2217 | ||
da6053d0 | 2218 | void bind_mount_free_many(BindMount *b, size_t n) { |
d2d6c096 LP |
2219 | assert(b || n == 0); |
2220 | ||
fe96c0f8 | 2221 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
2222 | free(b[i].source); |
2223 | free(b[i].destination); | |
2224 | } | |
2225 | ||
2226 | free(b); | |
2227 | } | |
2228 | ||
da6053d0 | 2229 | int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) { |
d2d6c096 LP |
2230 | _cleanup_free_ char *s = NULL, *d = NULL; |
2231 | BindMount *c; | |
2232 | ||
2233 | assert(b); | |
2234 | assert(n); | |
2235 | assert(item); | |
2236 | ||
2237 | s = strdup(item->source); | |
2238 | if (!s) | |
2239 | return -ENOMEM; | |
2240 | ||
2241 | d = strdup(item->destination); | |
2242 | if (!d) | |
2243 | return -ENOMEM; | |
2244 | ||
aa484f35 | 2245 | c = reallocarray(*b, *n + 1, sizeof(BindMount)); |
d2d6c096 LP |
2246 | if (!c) |
2247 | return -ENOMEM; | |
2248 | ||
2249 | *b = c; | |
2250 | ||
2251 | c[(*n) ++] = (BindMount) { | |
1cc6c93a YW |
2252 | .source = TAKE_PTR(s), |
2253 | .destination = TAKE_PTR(d), | |
d2d6c096 | 2254 | .read_only = item->read_only, |
9ce4e4b0 | 2255 | .nosuid = item->nosuid, |
d2d6c096 LP |
2256 | .recursive = item->recursive, |
2257 | .ignore_enoent = item->ignore_enoent, | |
2258 | }; | |
2259 | ||
d2d6c096 LP |
2260 | return 0; |
2261 | } | |
2262 | ||
b3d13314 | 2263 | MountImage* mount_image_free_many(MountImage *m, size_t *n) { |
b3d13314 LB |
2264 | assert(n); |
2265 | assert(m || *n == 0); | |
2266 | ||
fe96c0f8 | 2267 | for (size_t i = 0; i < *n; i++) { |
b3d13314 LB |
2268 | free(m[i].source); |
2269 | free(m[i].destination); | |
427353f6 | 2270 | mount_options_free_all(m[i].mount_options); |
b3d13314 LB |
2271 | } |
2272 | ||
2273 | free(m); | |
2274 | *n = 0; | |
2275 | return NULL; | |
2276 | } | |
2277 | ||
2278 | int mount_image_add(MountImage **m, size_t *n, const MountImage *item) { | |
2279 | _cleanup_free_ char *s = NULL, *d = NULL; | |
427353f6 LB |
2280 | _cleanup_(mount_options_free_allp) MountOptions *options = NULL; |
2281 | MountOptions *i; | |
b3d13314 LB |
2282 | MountImage *c; |
2283 | ||
2284 | assert(m); | |
2285 | assert(n); | |
2286 | assert(item); | |
2287 | ||
2288 | s = strdup(item->source); | |
2289 | if (!s) | |
2290 | return -ENOMEM; | |
2291 | ||
93f59701 LB |
2292 | if (item->destination) { |
2293 | d = strdup(item->destination); | |
2294 | if (!d) | |
2295 | return -ENOMEM; | |
2296 | } | |
b3d13314 | 2297 | |
427353f6 LB |
2298 | LIST_FOREACH(mount_options, i, item->mount_options) { |
2299 | _cleanup_(mount_options_free_allp) MountOptions *o; | |
2300 | ||
2301 | o = new(MountOptions, 1); | |
2302 | if (!o) | |
2303 | return -ENOMEM; | |
2304 | ||
2305 | *o = (MountOptions) { | |
2306 | .partition_designator = i->partition_designator, | |
2307 | .options = strdup(i->options), | |
2308 | }; | |
2309 | if (!o->options) | |
2310 | return -ENOMEM; | |
2311 | ||
2312 | LIST_APPEND(mount_options, options, TAKE_PTR(o)); | |
2313 | } | |
2314 | ||
b3d13314 LB |
2315 | c = reallocarray(*m, *n + 1, sizeof(MountImage)); |
2316 | if (!c) | |
2317 | return -ENOMEM; | |
2318 | ||
2319 | *m = c; | |
2320 | ||
2321 | c[(*n) ++] = (MountImage) { | |
2322 | .source = TAKE_PTR(s), | |
2323 | .destination = TAKE_PTR(d), | |
427353f6 | 2324 | .mount_options = TAKE_PTR(options), |
b3d13314 | 2325 | .ignore_enoent = item->ignore_enoent, |
93f59701 | 2326 | .type = item->type, |
b3d13314 LB |
2327 | }; |
2328 | ||
2329 | return 0; | |
2330 | } | |
2331 | ||
da6053d0 | 2332 | void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) { |
2abd4e38 YW |
2333 | assert(t || n == 0); |
2334 | ||
fe96c0f8 | 2335 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
2336 | free(t[i].path); |
2337 | free(t[i].options); | |
2338 | } | |
2339 | ||
2340 | free(t); | |
2341 | } | |
2342 | ||
2343 | int temporary_filesystem_add( | |
2344 | TemporaryFileSystem **t, | |
da6053d0 | 2345 | size_t *n, |
2abd4e38 YW |
2346 | const char *path, |
2347 | const char *options) { | |
2348 | ||
2349 | _cleanup_free_ char *p = NULL, *o = NULL; | |
2350 | TemporaryFileSystem *c; | |
2351 | ||
2352 | assert(t); | |
2353 | assert(n); | |
2354 | assert(path); | |
2355 | ||
2356 | p = strdup(path); | |
2357 | if (!p) | |
2358 | return -ENOMEM; | |
2359 | ||
2360 | if (!isempty(options)) { | |
2361 | o = strdup(options); | |
2362 | if (!o) | |
2363 | return -ENOMEM; | |
2364 | } | |
2365 | ||
aa484f35 | 2366 | c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem)); |
2abd4e38 YW |
2367 | if (!c) |
2368 | return -ENOMEM; | |
2369 | ||
2370 | *t = c; | |
2371 | ||
2372 | c[(*n) ++] = (TemporaryFileSystem) { | |
1cc6c93a YW |
2373 | .path = TAKE_PTR(p), |
2374 | .options = TAKE_PTR(o), | |
2abd4e38 YW |
2375 | }; |
2376 | ||
2abd4e38 YW |
2377 | return 0; |
2378 | } | |
2379 | ||
a652f050 JR |
2380 | static int make_tmp_prefix(const char *prefix) { |
2381 | _cleanup_free_ char *t = NULL; | |
2382 | int r; | |
2383 | ||
2384 | /* Don't do anything unless we know the dir is actually missing */ | |
2385 | r = access(prefix, F_OK); | |
2386 | if (r >= 0) | |
2387 | return 0; | |
2388 | if (errno != ENOENT) | |
2389 | return -errno; | |
2390 | ||
2391 | r = mkdir_parents(prefix, 0755); | |
2392 | if (r < 0) | |
2393 | return r; | |
2394 | ||
2395 | r = tempfn_random(prefix, NULL, &t); | |
2396 | if (r < 0) | |
2397 | return r; | |
2398 | ||
2399 | if (mkdir(t, 0777) < 0) | |
2400 | return -errno; | |
2401 | ||
2402 | if (chmod(t, 01777) < 0) { | |
2403 | r = -errno; | |
2404 | (void) rmdir(t); | |
2405 | return r; | |
2406 | } | |
2407 | ||
2408 | if (rename(t, prefix) < 0) { | |
2409 | r = -errno; | |
2410 | (void) rmdir(t); | |
2411 | return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */ | |
2412 | } | |
2413 | ||
2414 | return 0; | |
2415 | ||
2416 | } | |
2417 | ||
56a13a49 | 2418 | static int setup_one_tmp_dir(const char *id, const char *prefix, char **path, char **tmp_path) { |
613b411c | 2419 | _cleanup_free_ char *x = NULL; |
19cd4e19 | 2420 | _cleanup_free_ char *y = NULL; |
6b46ea73 LP |
2421 | char bid[SD_ID128_STRING_MAX]; |
2422 | sd_id128_t boot_id; | |
56a13a49 | 2423 | bool rw = true; |
6b46ea73 | 2424 | int r; |
613b411c LP |
2425 | |
2426 | assert(id); | |
2427 | assert(prefix); | |
2428 | assert(path); | |
2429 | ||
6b46ea73 LP |
2430 | /* We include the boot id in the directory so that after a |
2431 | * reboot we can easily identify obsolete directories. */ | |
2432 | ||
2433 | r = sd_id128_get_boot(&boot_id); | |
2434 | if (r < 0) | |
2435 | return r; | |
2436 | ||
605405c6 | 2437 | x = strjoin(prefix, "/systemd-private-", sd_id128_to_string(boot_id, bid), "-", id, "-XXXXXX"); |
613b411c LP |
2438 | if (!x) |
2439 | return -ENOMEM; | |
2440 | ||
a652f050 JR |
2441 | r = make_tmp_prefix(prefix); |
2442 | if (r < 0) | |
2443 | return r; | |
2444 | ||
613b411c | 2445 | RUN_WITH_UMASK(0077) |
56a13a49 ZJS |
2446 | if (!mkdtemp(x)) { |
2447 | if (errno == EROFS || ERRNO_IS_DISK_SPACE(errno)) | |
2448 | rw = false; | |
2449 | else | |
2450 | return -errno; | |
2451 | } | |
613b411c | 2452 | |
56a13a49 | 2453 | if (rw) { |
19cd4e19 | 2454 | y = strjoin(x, "/tmp"); |
2455 | if (!y) | |
2456 | return -ENOMEM; | |
2457 | ||
2458 | RUN_WITH_UMASK(0000) { | |
2459 | if (mkdir(y, 0777 | S_ISVTX) < 0) | |
2460 | return -errno; | |
2461 | } | |
2462 | ||
2463 | r = label_fix_container(y, prefix, 0); | |
56a13a49 ZJS |
2464 | if (r < 0) |
2465 | return r; | |
19cd4e19 | 2466 | |
2467 | if (tmp_path) | |
2468 | *tmp_path = TAKE_PTR(y); | |
56a13a49 ZJS |
2469 | } else { |
2470 | /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being | |
2471 | * read-only. This way the service will get the EROFS result as if it was writing to the real | |
2472 | * file system. */ | |
2473 | r = mkdir_p(RUN_SYSTEMD_EMPTY, 0500); | |
2474 | if (r < 0) | |
2475 | return r; | |
613b411c | 2476 | |
3f181262 LP |
2477 | r = free_and_strdup(&x, RUN_SYSTEMD_EMPTY); |
2478 | if (r < 0) | |
2479 | return r; | |
c17ec25e | 2480 | } |
15ae422b | 2481 | |
1cc6c93a | 2482 | *path = TAKE_PTR(x); |
613b411c LP |
2483 | return 0; |
2484 | } | |
2485 | ||
2486 | int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) { | |
56a13a49 ZJS |
2487 | _cleanup_(namespace_cleanup_tmpdirp) char *a = NULL; |
2488 | _cleanup_(rmdir_and_freep) char *a_tmp = NULL; | |
2489 | char *b; | |
613b411c LP |
2490 | int r; |
2491 | ||
2492 | assert(id); | |
2493 | assert(tmp_dir); | |
2494 | assert(var_tmp_dir); | |
2495 | ||
56a13a49 | 2496 | r = setup_one_tmp_dir(id, "/tmp", &a, &a_tmp); |
613b411c LP |
2497 | if (r < 0) |
2498 | return r; | |
2499 | ||
56a13a49 ZJS |
2500 | r = setup_one_tmp_dir(id, "/var/tmp", &b, NULL); |
2501 | if (r < 0) | |
613b411c | 2502 | return r; |
613b411c | 2503 | |
56a13a49 ZJS |
2504 | a_tmp = mfree(a_tmp); /* avoid rmdir */ |
2505 | *tmp_dir = TAKE_PTR(a); | |
2506 | *var_tmp_dir = TAKE_PTR(b); | |
613b411c LP |
2507 | |
2508 | return 0; | |
2509 | } | |
2510 | ||
2caa38e9 | 2511 | int setup_netns(const int netns_storage_socket[static 2]) { |
613b411c | 2512 | _cleanup_close_ int netns = -1; |
3ee897d6 | 2513 | int r, q; |
613b411c LP |
2514 | |
2515 | assert(netns_storage_socket); | |
2516 | assert(netns_storage_socket[0] >= 0); | |
2517 | assert(netns_storage_socket[1] >= 0); | |
2518 | ||
2519 | /* We use the passed socketpair as a storage buffer for our | |
76cd584b LP |
2520 | * namespace reference fd. Whatever process runs this first |
2521 | * shall create a new namespace, all others should just join | |
2522 | * it. To serialize that we use a file lock on the socket | |
2523 | * pair. | |
613b411c LP |
2524 | * |
2525 | * It's a bit crazy, but hey, works great! */ | |
2526 | ||
2527 | if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0) | |
2528 | return -errno; | |
2529 | ||
3ee897d6 LP |
2530 | netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT); |
2531 | if (netns == -EAGAIN) { | |
44ffcbae | 2532 | /* Nothing stored yet, so let's create a new namespace. */ |
613b411c LP |
2533 | |
2534 | if (unshare(CLONE_NEWNET) < 0) { | |
2535 | r = -errno; | |
2536 | goto fail; | |
2537 | } | |
2538 | ||
44ffcbae | 2539 | (void) loopback_setup(); |
613b411c LP |
2540 | |
2541 | netns = open("/proc/self/ns/net", O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
2542 | if (netns < 0) { | |
2543 | r = -errno; | |
2544 | goto fail; | |
2545 | } | |
2546 | ||
2547 | r = 1; | |
613b411c | 2548 | |
3ee897d6 LP |
2549 | } else if (netns < 0) { |
2550 | r = netns; | |
2551 | goto fail; | |
613b411c | 2552 | |
3ee897d6 LP |
2553 | } else { |
2554 | /* Yay, found something, so let's join the namespace */ | |
613b411c LP |
2555 | if (setns(netns, CLONE_NEWNET) < 0) { |
2556 | r = -errno; | |
2557 | goto fail; | |
2558 | } | |
2559 | ||
2560 | r = 0; | |
2561 | } | |
2562 | ||
3ee897d6 LP |
2563 | q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT); |
2564 | if (q < 0) { | |
2565 | r = q; | |
613b411c LP |
2566 | goto fail; |
2567 | } | |
2568 | ||
2569 | fail: | |
fe048ce5 | 2570 | (void) lockf(netns_storage_socket[0], F_ULOCK, 0); |
15ae422b LP |
2571 | return r; |
2572 | } | |
417116f2 | 2573 | |
2caa38e9 | 2574 | int open_netns_path(const int netns_storage_socket[static 2], const char *path) { |
51af7fb2 LP |
2575 | _cleanup_close_ int netns = -1; |
2576 | int q, r; | |
2577 | ||
2578 | assert(netns_storage_socket); | |
2579 | assert(netns_storage_socket[0] >= 0); | |
2580 | assert(netns_storage_socket[1] >= 0); | |
2581 | assert(path); | |
2582 | ||
2583 | /* If the storage socket doesn't contain a netns fd yet, open one via the file system and store it in | |
2584 | * it. This is supposed to be called ahead of time, i.e. before setup_netns() which will allocate a | |
2585 | * new anonymous netns if needed. */ | |
2586 | ||
2587 | if (lockf(netns_storage_socket[0], F_LOCK, 0) < 0) | |
2588 | return -errno; | |
2589 | ||
2590 | netns = receive_one_fd(netns_storage_socket[0], MSG_DONTWAIT); | |
2591 | if (netns == -EAGAIN) { | |
2592 | /* Nothing stored yet. Open the file from the file system. */ | |
2593 | ||
2594 | netns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC); | |
2595 | if (netns < 0) { | |
2596 | r = -errno; | |
2597 | goto fail; | |
2598 | } | |
2599 | ||
2600 | r = fd_is_network_ns(netns); | |
2601 | if (r == 0) { /* Not a netns? Refuse early. */ | |
2602 | r = -EINVAL; | |
2603 | goto fail; | |
2604 | } | |
2605 | if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */ | |
2606 | goto fail; | |
2607 | ||
2608 | r = 1; | |
2609 | ||
2610 | } else if (netns < 0) { | |
2611 | r = netns; | |
2612 | goto fail; | |
2613 | } else | |
2614 | r = 0; /* Already allocated */ | |
2615 | ||
2616 | q = send_one_fd(netns_storage_socket[1], netns, MSG_DONTWAIT); | |
2617 | if (q < 0) { | |
2618 | r = q; | |
2619 | goto fail; | |
2620 | } | |
2621 | ||
2622 | fail: | |
2623 | (void) lockf(netns_storage_socket[0], F_ULOCK, 0); | |
2624 | return r; | |
2625 | } | |
2626 | ||
6e2d7c4f MS |
2627 | bool ns_type_supported(NamespaceType type) { |
2628 | const char *t, *ns_proc; | |
2629 | ||
0fa5b831 LP |
2630 | t = namespace_type_to_string(type); |
2631 | if (!t) /* Don't know how to translate this? Then it's not supported */ | |
6e2d7c4f MS |
2632 | return false; |
2633 | ||
6e2d7c4f | 2634 | ns_proc = strjoina("/proc/self/ns/", t); |
6e2d7c4f MS |
2635 | return access(ns_proc, F_OK) == 0; |
2636 | } | |
2637 | ||
1b8689f9 | 2638 | static const char *const protect_home_table[_PROTECT_HOME_MAX] = { |
cbc056c8 ZJS |
2639 | [PROTECT_HOME_NO] = "no", |
2640 | [PROTECT_HOME_YES] = "yes", | |
1b8689f9 | 2641 | [PROTECT_HOME_READ_ONLY] = "read-only", |
cbc056c8 | 2642 | [PROTECT_HOME_TMPFS] = "tmpfs", |
417116f2 LP |
2643 | }; |
2644 | ||
1e8c7bd5 | 2645 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES); |
5e1c6154 | 2646 | |
1b8689f9 | 2647 | static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = { |
cbc056c8 ZJS |
2648 | [PROTECT_SYSTEM_NO] = "no", |
2649 | [PROTECT_SYSTEM_YES] = "yes", | |
2650 | [PROTECT_SYSTEM_FULL] = "full", | |
3f815163 | 2651 | [PROTECT_SYSTEM_STRICT] = "strict", |
1b8689f9 LP |
2652 | }; |
2653 | ||
1e8c7bd5 | 2654 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES); |
03c791aa | 2655 | |
6e2d7c4f | 2656 | static const char* const namespace_type_table[] = { |
cbc056c8 | 2657 | [NAMESPACE_MOUNT] = "mnt", |
6e2d7c4f | 2658 | [NAMESPACE_CGROUP] = "cgroup", |
cbc056c8 ZJS |
2659 | [NAMESPACE_UTS] = "uts", |
2660 | [NAMESPACE_IPC] = "ipc", | |
2661 | [NAMESPACE_USER] = "user", | |
2662 | [NAMESPACE_PID] = "pid", | |
2663 | [NAMESPACE_NET] = "net", | |
6e2d7c4f MS |
2664 | }; |
2665 | ||
2666 | DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType); | |
4e399953 LP |
2667 | |
2668 | static const char* const protect_proc_table[_PROTECT_PROC_MAX] = { | |
2669 | [PROTECT_PROC_DEFAULT] = "default", | |
2670 | [PROTECT_PROC_NOACCESS] = "noaccess", | |
2671 | [PROTECT_PROC_INVISIBLE] = "invisible", | |
2672 | [PROTECT_PROC_PTRACEABLE] = "ptraceable", | |
2673 | }; | |
2674 | ||
2675 | DEFINE_STRING_TABLE_LOOKUP(protect_proc, ProtectProc); | |
2676 | ||
2677 | static const char* const proc_subset_table[_PROC_SUBSET_MAX] = { | |
2678 | [PROC_SUBSET_ALL] = "all", | |
2679 | [PROC_SUBSET_PID] = "pid", | |
2680 | }; | |
2681 | ||
2682 | DEFINE_STRING_TABLE_LOOKUP(proc_subset, ProcSubset); |