]>
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> |
19df770f | 7 | #include <sys/file.h> |
07630cea | 8 | #include <sys/mount.h> |
07630cea | 9 | #include <unistd.h> |
3657d3a0 | 10 | #if WANT_LINUX_FS_H |
25e870b5 | 11 | #include <linux/fs.h> |
3657d3a0 | 12 | #endif |
15ae422b | 13 | |
b5efdb8a | 14 | #include "alloc-util.h" |
10404d52 | 15 | #include "base-filesystem.h" |
f461a28d | 16 | #include "chase.h" |
7f112f50 | 17 | #include "dev-setup.h" |
ec61371f | 18 | #include "devnum-util.h" |
93f59701 LB |
19 | #include "env-util.h" |
20 | #include "escape.h" | |
d51f8eb3 | 21 | #include "extension-util.h" |
3ffd4af2 | 22 | #include "fd-util.h" |
e5f10caf | 23 | #include "format-util.h" |
e2341b6b | 24 | #include "glyph-util.h" |
0690160e | 25 | #include "label-util.h" |
b3d13314 | 26 | #include "list.h" |
13339577 | 27 | #include "lock-util.h" |
915e6d16 | 28 | #include "loop-util.h" |
07630cea | 29 | #include "loopback-setup.h" |
1c265fcd | 30 | #include "missing_syscall.h" |
35cd0ba5 | 31 | #include "mkdir-label.h" |
4349cd7c | 32 | #include "mount-util.h" |
049af8ad | 33 | #include "mountpoint-util.h" |
0cb8e3d1 | 34 | #include "namespace-util.h" |
3ffd4af2 | 35 | #include "namespace.h" |
54c2459d | 36 | #include "nsflags.h" |
d8b4d14d | 37 | #include "nulstr-util.h" |
93f59701 | 38 | #include "os-util.h" |
07630cea | 39 | #include "path-util.h" |
d7b8eec7 | 40 | #include "selinux-util.h" |
2583fbea | 41 | #include "socket-util.h" |
760877e9 | 42 | #include "sort-util.h" |
36ce7110 | 43 | #include "stat-util.h" |
8b43440b | 44 | #include "string-table.h" |
07630cea LP |
45 | #include "string-util.h" |
46 | #include "strv.h" | |
a652f050 | 47 | #include "tmpfile-util.h" |
affb60b1 | 48 | #include "umask-util.h" |
ee104e11 | 49 | #include "user-util.h" |
15ae422b | 50 | |
737ba3c8 | 51 | #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC) |
52 | ||
c17ec25e | 53 | typedef enum MountMode { |
15ae422b LP |
54 | /* This is ordered by priority! */ |
55 | INACCESSIBLE, | |
93f59701 | 56 | OVERLAY_MOUNT, |
b3d13314 | 57 | MOUNT_IMAGES, |
d2d6c096 LP |
58 | BIND_MOUNT, |
59 | BIND_MOUNT_RECURSIVE, | |
ac0930c8 | 60 | PRIVATE_TMP, |
56a13a49 | 61 | PRIVATE_TMP_READONLY, |
7f112f50 | 62 | PRIVATE_DEV, |
5d997827 | 63 | BIND_DEV, |
6c47cd7d | 64 | EMPTY_DIR, |
277b269e | 65 | PRIVATE_SYSFS, |
10028263 | 66 | BIND_SYSFS, |
5d997827 LP |
67 | PROCFS, |
68 | READONLY, | |
59eeb84b | 69 | READWRITE, |
ddc155b2 TM |
70 | NOEXEC, |
71 | EXEC, | |
2abd4e38 | 72 | TMPFS, |
12413985 | 73 | RUN, |
a07b9926 | 74 | EXTENSION_DIRECTORIES, /* Bind-mounted outside the root directory, and used by subsequent mounts */ |
93f59701 | 75 | EXTENSION_IMAGES, /* Mounted outside the root directory, and used by subsequent mounts */ |
80271a44 | 76 | MQUEUEFS, |
94fe4cf2 YW |
77 | READWRITE_IMPLICIT, /* Should have the 2nd lowest priority. */ |
78 | MKDIR, /* Should have the lowest priority. */ | |
5beb8688 | 79 | _MOUNT_MODE_MAX, |
c17ec25e | 80 | } MountMode; |
15ae422b | 81 | |
34de407a | 82 | typedef struct MountEntry { |
5327c910 | 83 | const char *path_const; /* Memory allocated on stack or static */ |
cfbeb4ef | 84 | MountMode mode:5; |
5327c910 LP |
85 | bool ignore:1; /* Ignore if path does not exist? */ |
86 | bool has_prefix:1; /* Already is prefixed by the root dir? */ | |
cfbeb4ef | 87 | bool read_only:1; /* Shall this mount point be read-only? */ |
9ce4e4b0 | 88 | bool nosuid:1; /* Shall set MS_NOSUID on the mount itself */ |
ddc155b2 TM |
89 | bool noexec:1; /* Shall set MS_NOEXEC on the mount itself */ |
90 | bool exec:1; /* Shall clear MS_NOEXEC on the mount itself */ | |
088696fe | 91 | bool applied:1; /* Already applied */ |
55fe7432 | 92 | char *path_malloc; /* Use this instead of 'path_const' if we had to allocate memory */ |
809ceb82 LB |
93 | const char *unprefixed_path_const; /* If the path was amended with a prefix, these will save the original */ |
94 | char *unprefixed_path_malloc; | |
b3d13314 | 95 | const char *source_const; /* The source path, for bind mounts or images */ |
d2d6c096 | 96 | char *source_malloc; |
2abd4e38 YW |
97 | const char *options_const;/* Mount options for tmpfs */ |
98 | char *options_malloc; | |
99 | unsigned long flags; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */ | |
088696fe | 100 | unsigned n_followed; |
427353f6 | 101 | LIST_HEAD(MountOptions, image_options); |
34de407a | 102 | } MountEntry; |
15ae422b | 103 | |
94293d65 | 104 | /* 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 | 105 | * something there already. These mounts are hence overridden by any other explicitly configured mounts. */ |
5d997827 LP |
106 | static const MountEntry apivfs_table[] = { |
107 | { "/proc", PROCFS, false }, | |
108 | { "/dev", BIND_DEV, false }, | |
10028263 | 109 | { "/sys", BIND_SYSFS, false }, |
9f563f27 | 110 | { "/run", RUN, false, .options_const = "mode=0755" TMPFS_LIMITS_RUN, .flags = MS_NOSUID|MS_NODEV|MS_STRICTATIME }, |
5d997827 | 111 | }; |
f471b2af | 112 | |
11a30cec | 113 | /* ProtectKernelTunables= option and the related filesystem APIs */ |
788e7201 | 114 | static const MountEntry protect_kernel_tunables_proc_table[] = { |
1e05071d YW |
115 | { "/proc/acpi", READONLY, true }, |
116 | { "/proc/apm", READONLY, true }, /* Obsolete API, there's no point in permitting access to this, ever */ | |
117 | { "/proc/asound", READONLY, true }, | |
118 | { "/proc/bus", READONLY, true }, | |
119 | { "/proc/fs", READONLY, true }, | |
120 | { "/proc/irq", READONLY, true }, | |
121 | { "/proc/kallsyms", INACCESSIBLE, true }, | |
122 | { "/proc/kcore", INACCESSIBLE, true }, | |
123 | { "/proc/latency_stats", READONLY, true }, | |
124 | { "/proc/mtrr", READONLY, true }, | |
125 | { "/proc/scsi", READONLY, true }, | |
4e399953 | 126 | { "/proc/sys", READONLY, true }, |
1e05071d YW |
127 | { "/proc/sysrq-trigger", READONLY, true }, |
128 | { "/proc/timer_stats", READONLY, true }, | |
788e7201 TM |
129 | }; |
130 | ||
131 | static const MountEntry protect_kernel_tunables_sys_table[] = { | |
1e05071d YW |
132 | { "/sys", READONLY, false }, |
133 | { "/sys/fs/bpf", READONLY, true }, | |
134 | { "/sys/fs/cgroup", READWRITE_IMPLICIT, false }, /* READONLY is set by ProtectControlGroups= option */ | |
135 | { "/sys/fs/selinux", READWRITE_IMPLICIT, true }, | |
136 | { "/sys/kernel/debug", READONLY, true }, | |
137 | { "/sys/kernel/tracing", READONLY, true }, | |
11a30cec DH |
138 | }; |
139 | ||
c575770b | 140 | /* ProtectKernelModules= option */ |
34de407a | 141 | static const MountEntry protect_kernel_modules_table[] = { |
c6232fb0 | 142 | { "/usr/lib/modules", INACCESSIBLE, true }, |
c575770b DH |
143 | }; |
144 | ||
94a7b275 | 145 | /* ProtectKernelLogs= option */ |
788e7201 | 146 | static const MountEntry protect_kernel_logs_proc_table[] = { |
94a7b275 | 147 | { "/proc/kmsg", INACCESSIBLE, true }, |
788e7201 TM |
148 | }; |
149 | ||
150 | static const MountEntry protect_kernel_logs_dev_table[] = { | |
94a7b275 KK |
151 | { "/dev/kmsg", INACCESSIBLE, true }, |
152 | }; | |
153 | ||
b6c432ca DH |
154 | /* |
155 | * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of | |
156 | * system should be protected by ProtectSystem= | |
157 | */ | |
34de407a | 158 | static const MountEntry protect_home_read_only_table[] = { |
c6232fb0 LP |
159 | { "/home", READONLY, true }, |
160 | { "/run/user", READONLY, true }, | |
161 | { "/root", READONLY, true }, | |
b6c432ca DH |
162 | }; |
163 | ||
e4da7d8c YW |
164 | /* ProtectHome=tmpfs table */ |
165 | static const MountEntry protect_home_tmpfs_table[] = { | |
7d85383e TM |
166 | { "/home", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, |
167 | { "/run/user", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
168 | { "/root", TMPFS, true, .read_only = true, .options_const = "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
e4da7d8c YW |
169 | }; |
170 | ||
b6c432ca | 171 | /* ProtectHome=yes table */ |
34de407a | 172 | static const MountEntry protect_home_yes_table[] = { |
c6232fb0 LP |
173 | { "/home", INACCESSIBLE, true }, |
174 | { "/run/user", INACCESSIBLE, true }, | |
175 | { "/root", INACCESSIBLE, true }, | |
b6c432ca DH |
176 | }; |
177 | ||
f471b2af | 178 | /* ProtectSystem=yes table */ |
34de407a | 179 | static const MountEntry protect_system_yes_table[] = { |
c6232fb0 LP |
180 | { "/usr", READONLY, false }, |
181 | { "/boot", READONLY, true }, | |
182 | { "/efi", READONLY, true }, | |
f471b2af DH |
183 | }; |
184 | ||
185 | /* ProtectSystem=full includes ProtectSystem=yes */ | |
34de407a | 186 | static const MountEntry protect_system_full_table[] = { |
c6232fb0 LP |
187 | { "/usr", READONLY, false }, |
188 | { "/boot", READONLY, true }, | |
189 | { "/efi", READONLY, true }, | |
190 | { "/etc", READONLY, false }, | |
f471b2af DH |
191 | }; |
192 | ||
193 | /* | |
194 | * ProtectSystem=strict table. In this strict mode, we mount everything | |
195 | * read-only, except for /proc, /dev, /sys which are the kernel API VFS, | |
196 | * which are left writable, but PrivateDevices= + ProtectKernelTunables= | |
197 | * protect those, and these options should be fully orthogonal. | |
198 | * (And of course /home and friends are also left writable, as ProtectHome= | |
199 | * shall manage those, orthogonally). | |
200 | */ | |
34de407a | 201 | static const MountEntry protect_system_strict_table[] = { |
1e05071d YW |
202 | { "/", READONLY, false }, |
203 | { "/proc", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
204 | { "/sys", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
205 | { "/dev", READWRITE_IMPLICIT, false }, /* PrivateDevices= */ | |
206 | { "/home", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
207 | { "/run/user", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
208 | { "/root", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
f471b2af DH |
209 | }; |
210 | ||
5beb8688 | 211 | static const char * const mount_mode_table[_MOUNT_MODE_MAX] = { |
cc45ff17 YW |
212 | [INACCESSIBLE] = "inaccessible", |
213 | [OVERLAY_MOUNT] = "overlay", | |
214 | [MOUNT_IMAGES] = "mount-images", | |
215 | [BIND_MOUNT] = "bind", | |
216 | [BIND_MOUNT_RECURSIVE] = "rbind", | |
217 | [PRIVATE_TMP] = "private-tmp", | |
218 | [PRIVATE_TMP_READONLY] = "private-tmp-read-only", | |
219 | [PRIVATE_DEV] = "private-dev", | |
220 | [BIND_DEV] = "bind-dev", | |
221 | [EMPTY_DIR] = "empty", | |
222 | [PRIVATE_SYSFS] = "private-sysfs", | |
223 | [BIND_SYSFS] = "bind-sysfs", | |
224 | [PROCFS] = "procfs", | |
225 | [READONLY] = "read-only", | |
226 | [READWRITE] = "read-write", | |
227 | [NOEXEC] = "noexec", | |
228 | [EXEC] = "exec", | |
229 | [TMPFS] = "tmpfs", | |
230 | [RUN] = "run", | |
231 | [EXTENSION_DIRECTORIES] = "extension-directories", | |
232 | [EXTENSION_IMAGES] = "extension-images", | |
233 | [MQUEUEFS] = "mqueuefs", | |
234 | [READWRITE_IMPLICIT] = "read-write-implicit", | |
235 | [MKDIR] = "mkdir", | |
5beb8688 YW |
236 | }; |
237 | ||
55ea4ef0 MG |
238 | /* Helper struct for naming simplicity and reusability */ |
239 | static const struct { | |
240 | const char *level_env; | |
241 | const char *level_env_print; | |
242 | } image_class_info[_IMAGE_CLASS_MAX] = { | |
243 | [IMAGE_SYSEXT] = { | |
244 | .level_env = "SYSEXT_LEVEL", | |
245 | .level_env_print = " SYSEXT_LEVEL=", | |
246 | }, | |
247 | [IMAGE_CONFEXT] = { | |
248 | .level_env = "CONFEXT_LEVEL", | |
249 | .level_env_print = " CONFEXT_LEVEL=", | |
250 | } | |
251 | }; | |
252 | ||
5beb8688 YW |
253 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode, MountMode); |
254 | ||
34de407a | 255 | static const char *mount_entry_path(const MountEntry *p) { |
f0a4feb0 DH |
256 | assert(p); |
257 | ||
5327c910 LP |
258 | /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that, |
259 | * otherwise the stack/static ->path field is returned. */ | |
f0a4feb0 | 260 | |
5327c910 | 261 | return p->path_malloc ?: p->path_const; |
f0a4feb0 DH |
262 | } |
263 | ||
809ceb82 LB |
264 | static const char *mount_entry_unprefixed_path(const MountEntry *p) { |
265 | assert(p); | |
266 | ||
267 | /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */ | |
268 | ||
269 | return p->unprefixed_path_malloc ?: p->unprefixed_path_const ?: mount_entry_path(p); | |
270 | } | |
271 | ||
272 | static void mount_entry_consume_prefix(MountEntry *p, char *new_path) { | |
273 | assert(p); | |
274 | assert(p->path_malloc || p->path_const); | |
275 | assert(new_path); | |
276 | ||
277 | /* Saves current path in unprefixed_ variable, and takes over new_path */ | |
278 | ||
279 | free_and_replace(p->unprefixed_path_malloc, p->path_malloc); | |
280 | /* If we didn't have a path on the heap, then it's a static one */ | |
281 | if (!p->unprefixed_path_malloc) | |
282 | p->unprefixed_path_const = p->path_const; | |
283 | p->path_malloc = new_path; | |
284 | p->has_prefix = true; | |
285 | } | |
286 | ||
34de407a | 287 | static bool mount_entry_read_only(const MountEntry *p) { |
cfbeb4ef LP |
288 | assert(p); |
289 | ||
56a13a49 | 290 | return p->read_only || IN_SET(p->mode, READONLY, INACCESSIBLE, PRIVATE_TMP_READONLY); |
cfbeb4ef LP |
291 | } |
292 | ||
ddc155b2 TM |
293 | static bool mount_entry_noexec(const MountEntry *p) { |
294 | assert(p); | |
295 | ||
277b269e | 296 | return p->noexec || IN_SET(p->mode, NOEXEC, INACCESSIBLE, PRIVATE_SYSFS, BIND_SYSFS, PROCFS); |
ddc155b2 TM |
297 | } |
298 | ||
299 | static bool mount_entry_exec(const MountEntry *p) { | |
300 | assert(p); | |
301 | ||
302 | return p->exec || p->mode == EXEC; | |
303 | } | |
304 | ||
d2d6c096 LP |
305 | static const char *mount_entry_source(const MountEntry *p) { |
306 | assert(p); | |
307 | ||
308 | return p->source_malloc ?: p->source_const; | |
309 | } | |
310 | ||
2abd4e38 YW |
311 | static const char *mount_entry_options(const MountEntry *p) { |
312 | assert(p); | |
313 | ||
314 | return p->options_malloc ?: p->options_const; | |
315 | } | |
316 | ||
1eb7e08e LP |
317 | static void mount_entry_done(MountEntry *p) { |
318 | assert(p); | |
319 | ||
320 | p->path_malloc = mfree(p->path_malloc); | |
809ceb82 | 321 | p->unprefixed_path_malloc = mfree(p->unprefixed_path_malloc); |
1eb7e08e | 322 | p->source_malloc = mfree(p->source_malloc); |
2abd4e38 | 323 | p->options_malloc = mfree(p->options_malloc); |
427353f6 | 324 | p->image_options = mount_options_free_all(p->image_options); |
1eb7e08e LP |
325 | } |
326 | ||
d18aff04 | 327 | static int append_access_mounts(MountEntry **p, char **strv, MountMode mode, bool forcibly_require_prefix) { |
613b411c LP |
328 | assert(p); |
329 | ||
1e05071d | 330 | /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */ |
5327c910 | 331 | |
15ae422b | 332 | STRV_FOREACH(i, strv) { |
5327c910 LP |
333 | bool ignore = false, needs_prefix = false; |
334 | const char *e = *i; | |
15ae422b | 335 | |
5327c910 LP |
336 | /* Look for any prefixes */ |
337 | if (startswith(e, "-")) { | |
338 | e++; | |
9c94d52e | 339 | ignore = true; |
ea92ae33 | 340 | } |
5327c910 LP |
341 | if (startswith(e, "+")) { |
342 | e++; | |
343 | needs_prefix = true; | |
344 | } | |
ea92ae33 | 345 | |
baaa35ad ZJS |
346 | if (!path_is_absolute(e)) |
347 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
348 | "Path is not absolute: %s", e); | |
15ae422b | 349 | |
34de407a | 350 | *((*p)++) = (MountEntry) { |
5327c910 LP |
351 | .path_const = e, |
352 | .mode = mode, | |
353 | .ignore = ignore, | |
d18aff04 | 354 | .has_prefix = !needs_prefix && !forcibly_require_prefix, |
5327c910 | 355 | }; |
15ae422b LP |
356 | } |
357 | ||
358 | return 0; | |
359 | } | |
360 | ||
6c47cd7d | 361 | static int append_empty_dir_mounts(MountEntry **p, char **strv) { |
6c47cd7d LP |
362 | assert(p); |
363 | ||
364 | /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the | |
365 | * "/private/" boundary directories for DynamicUser=1. */ | |
366 | ||
367 | STRV_FOREACH(i, strv) { | |
368 | ||
369 | *((*p)++) = (MountEntry) { | |
370 | .path_const = *i, | |
371 | .mode = EMPTY_DIR, | |
372 | .ignore = false, | |
6c47cd7d | 373 | .read_only = true, |
9f563f27 | 374 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, |
2abd4e38 | 375 | .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, |
6c47cd7d LP |
376 | }; |
377 | } | |
378 | ||
379 | return 0; | |
380 | } | |
381 | ||
da6053d0 | 382 | static int append_bind_mounts(MountEntry **p, const BindMount *binds, size_t n) { |
d2d6c096 LP |
383 | assert(p); |
384 | ||
fe96c0f8 | 385 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
386 | const BindMount *b = binds + i; |
387 | ||
388 | *((*p)++) = (MountEntry) { | |
389 | .path_const = b->destination, | |
390 | .mode = b->recursive ? BIND_MOUNT_RECURSIVE : BIND_MOUNT, | |
391 | .read_only = b->read_only, | |
9ce4e4b0 | 392 | .nosuid = b->nosuid, |
d2d6c096 | 393 | .source_const = b->source, |
4ca763a9 | 394 | .ignore = b->ignore_enoent, |
d2d6c096 LP |
395 | }; |
396 | } | |
397 | ||
398 | return 0; | |
399 | } | |
400 | ||
b3d13314 LB |
401 | static int append_mount_images(MountEntry **p, const MountImage *mount_images, size_t n) { |
402 | assert(p); | |
403 | ||
404 | for (size_t i = 0; i < n; i++) { | |
405 | const MountImage *m = mount_images + i; | |
406 | ||
407 | *((*p)++) = (MountEntry) { | |
408 | .path_const = m->destination, | |
409 | .mode = MOUNT_IMAGES, | |
410 | .source_const = m->source, | |
427353f6 | 411 | .image_options = m->mount_options, |
b3d13314 LB |
412 | .ignore = m->ignore_enoent, |
413 | }; | |
414 | } | |
415 | ||
416 | return 0; | |
417 | } | |
418 | ||
a07b9926 | 419 | static int append_extensions( |
93f59701 LB |
420 | MountEntry **p, |
421 | const char *root, | |
422 | const char *extension_dir, | |
423 | char **hierarchies, | |
424 | const MountImage *mount_images, | |
a07b9926 LB |
425 | size_t n, |
426 | char **extension_directories) { | |
93f59701 LB |
427 | |
428 | _cleanup_strv_free_ char **overlays = NULL; | |
93f59701 LB |
429 | int r; |
430 | ||
a07b9926 | 431 | if (n == 0 && strv_isempty(extension_directories)) |
93f59701 LB |
432 | return 0; |
433 | ||
24759d8f LB |
434 | assert(p); |
435 | assert(extension_dir); | |
436 | ||
93f59701 | 437 | /* Prepare a list of overlays, that will have as each element a string suitable for being |
7a7d2f16 | 438 | * passed as a lowerdir= parameter, so start with the hierarchy on the root. |
93f59701 LB |
439 | * The overlays vector will have the same number of elements and will correspond to the |
440 | * hierarchies vector, so they can be iterated upon together. */ | |
441 | STRV_FOREACH(hierarchy, hierarchies) { | |
442 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
443 | ||
444 | prefixed_hierarchy = path_join(root, *hierarchy); | |
445 | if (!prefixed_hierarchy) | |
446 | return -ENOMEM; | |
447 | ||
448 | r = strv_consume(&overlays, TAKE_PTR(prefixed_hierarchy)); | |
449 | if (r < 0) | |
450 | return r; | |
451 | } | |
452 | ||
453 | /* First, prepare a mount for each image, but these won't be visible to the unit, instead | |
454 | * they will be mounted in our propagate directory, and used as a source for the overlay. */ | |
455 | for (size_t i = 0; i < n; i++) { | |
456 | _cleanup_free_ char *mount_point = NULL; | |
457 | const MountImage *m = mount_images + i; | |
458 | ||
459 | r = asprintf(&mount_point, "%s/%zu", extension_dir, i); | |
460 | if (r < 0) | |
461 | return -ENOMEM; | |
462 | ||
463 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
464 | _cleanup_free_ char *prefixed_hierarchy = NULL, *escaped = NULL, *lowerdir = NULL; | |
465 | ||
466 | prefixed_hierarchy = path_join(mount_point, hierarchies[j]); | |
467 | if (!prefixed_hierarchy) | |
468 | return -ENOMEM; | |
469 | ||
470 | escaped = shell_escape(prefixed_hierarchy, ",:"); | |
471 | if (!escaped) | |
472 | return -ENOMEM; | |
473 | ||
474 | /* Note that lowerdir= parameters are in 'reverse' order, so the | |
475 | * top-most directory in the overlay comes first in the list. */ | |
476 | lowerdir = strjoin(escaped, ":", overlays[j]); | |
477 | if (!lowerdir) | |
478 | return -ENOMEM; | |
479 | ||
480 | free_and_replace(overlays[j], lowerdir); | |
481 | } | |
482 | ||
483 | *((*p)++) = (MountEntry) { | |
484 | .path_malloc = TAKE_PTR(mount_point), | |
485 | .image_options = m->mount_options, | |
486 | .ignore = m->ignore_enoent, | |
487 | .source_const = m->source, | |
488 | .mode = EXTENSION_IMAGES, | |
489 | .has_prefix = true, | |
490 | }; | |
491 | } | |
492 | ||
a07b9926 LB |
493 | /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory. |
494 | * Bind mount them in the same location as the ExtensionImages, so that we | |
495 | * can check that they are valid trees (extension-release.d). */ | |
496 | STRV_FOREACH(extension_directory, extension_directories) { | |
497 | _cleanup_free_ char *mount_point = NULL, *source = NULL; | |
498 | const char *e = *extension_directory; | |
499 | bool ignore_enoent = false; | |
500 | ||
501 | /* Pick up the counter where the ExtensionImages left it. */ | |
502 | r = asprintf(&mount_point, "%s/%zu", extension_dir, n++); | |
503 | if (r < 0) | |
504 | return -ENOMEM; | |
505 | ||
506 | /* Look for any prefixes */ | |
507 | if (startswith(e, "-")) { | |
508 | e++; | |
509 | ignore_enoent = true; | |
510 | } | |
511 | /* Ignore this for now */ | |
512 | if (startswith(e, "+")) | |
513 | e++; | |
514 | ||
515 | source = strdup(e); | |
516 | if (!source) | |
517 | return -ENOMEM; | |
518 | ||
519 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
520 | _cleanup_free_ char *prefixed_hierarchy = NULL, *escaped = NULL, *lowerdir = NULL; | |
521 | ||
522 | prefixed_hierarchy = path_join(mount_point, hierarchies[j]); | |
523 | if (!prefixed_hierarchy) | |
524 | return -ENOMEM; | |
525 | ||
526 | escaped = shell_escape(prefixed_hierarchy, ",:"); | |
527 | if (!escaped) | |
528 | return -ENOMEM; | |
529 | ||
530 | /* Note that lowerdir= parameters are in 'reverse' order, so the | |
531 | * top-most directory in the overlay comes first in the list. */ | |
532 | lowerdir = strjoin(escaped, ":", overlays[j]); | |
533 | if (!lowerdir) | |
534 | return -ENOMEM; | |
535 | ||
536 | free_and_replace(overlays[j], lowerdir); | |
537 | } | |
538 | ||
539 | *((*p)++) = (MountEntry) { | |
540 | .path_malloc = TAKE_PTR(mount_point), | |
c9423746 | 541 | .source_malloc = TAKE_PTR(source), |
a07b9926 LB |
542 | .mode = EXTENSION_DIRECTORIES, |
543 | .ignore = ignore_enoent, | |
544 | .has_prefix = true, | |
545 | .read_only = true, | |
546 | }; | |
547 | } | |
548 | ||
93f59701 LB |
549 | /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings |
550 | * set up earlier. */ | |
551 | for (size_t i = 0; hierarchies && hierarchies[i]; ++i) { | |
552 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
553 | ||
554 | prefixed_hierarchy = path_join(root, hierarchies[i]); | |
555 | if (!prefixed_hierarchy) | |
556 | return -ENOMEM; | |
557 | ||
558 | *((*p)++) = (MountEntry) { | |
559 | .path_malloc = TAKE_PTR(prefixed_hierarchy), | |
560 | .options_malloc = TAKE_PTR(overlays[i]), | |
561 | .mode = OVERLAY_MOUNT, | |
562 | .has_prefix = true, | |
563 | .ignore = true, /* If the source image doesn't set the ignore bit it will fail earlier. */ | |
564 | }; | |
565 | } | |
566 | ||
567 | return 0; | |
568 | } | |
569 | ||
da6053d0 | 570 | static int append_tmpfs_mounts(MountEntry **p, const TemporaryFileSystem *tmpfs, size_t n) { |
2abd4e38 YW |
571 | assert(p); |
572 | ||
b67ec8e5 | 573 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
574 | const TemporaryFileSystem *t = tmpfs + i; |
575 | _cleanup_free_ char *o = NULL, *str = NULL; | |
ad8e66dc | 576 | unsigned long flags; |
2abd4e38 | 577 | bool ro = false; |
b67ec8e5 | 578 | int r; |
2abd4e38 | 579 | |
baaa35ad ZJS |
580 | if (!path_is_absolute(t->path)) |
581 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
582 | "Path is not absolute: %s", | |
583 | t->path); | |
2abd4e38 | 584 | |
b67ec8e5 | 585 | str = strjoin("mode=0755" NESTED_TMPFS_LIMITS ",", t->options); |
ad8e66dc AJ |
586 | if (!str) |
587 | return -ENOMEM; | |
2abd4e38 | 588 | |
ad8e66dc AJ |
589 | r = mount_option_mangle(str, MS_NODEV|MS_STRICTATIME, &flags, &o); |
590 | if (r < 0) | |
591 | return log_debug_errno(r, "Failed to parse mount option '%s': %m", str); | |
2abd4e38 | 592 | |
ad8e66dc AJ |
593 | ro = flags & MS_RDONLY; |
594 | if (ro) | |
595 | flags ^= MS_RDONLY; | |
2abd4e38 YW |
596 | |
597 | *((*p)++) = (MountEntry) { | |
598 | .path_const = t->path, | |
599 | .mode = TMPFS, | |
600 | .read_only = ro, | |
ad8e66dc | 601 | .options_malloc = TAKE_PTR(o), |
2abd4e38 YW |
602 | .flags = flags, |
603 | }; | |
2abd4e38 YW |
604 | } |
605 | ||
606 | return 0; | |
607 | } | |
608 | ||
da6053d0 | 609 | static int append_static_mounts(MountEntry **p, const MountEntry *mounts, size_t n, bool ignore_protect) { |
11a30cec | 610 | assert(p); |
f471b2af | 611 | assert(mounts); |
11a30cec | 612 | |
5327c910 | 613 | /* Adds a list of static pre-defined entries */ |
f471b2af | 614 | |
fe96c0f8 | 615 | for (size_t i = 0; i < n; i++) |
34de407a LP |
616 | *((*p)++) = (MountEntry) { |
617 | .path_const = mount_entry_path(mounts+i), | |
5327c910 LP |
618 | .mode = mounts[i].mode, |
619 | .ignore = mounts[i].ignore || ignore_protect, | |
620 | }; | |
f471b2af DH |
621 | |
622 | return 0; | |
623 | } | |
624 | ||
34de407a | 625 | static int append_protect_home(MountEntry **p, ProtectHome protect_home, bool ignore_protect) { |
c575770b DH |
626 | assert(p); |
627 | ||
5327c910 | 628 | switch (protect_home) { |
b6c432ca | 629 | |
5327c910 | 630 | case PROTECT_HOME_NO: |
b6c432ca DH |
631 | return 0; |
632 | ||
b6c432ca | 633 | case PROTECT_HOME_READ_ONLY: |
5327c910 LP |
634 | return append_static_mounts(p, protect_home_read_only_table, ELEMENTSOF(protect_home_read_only_table), ignore_protect); |
635 | ||
e4da7d8c YW |
636 | case PROTECT_HOME_TMPFS: |
637 | return append_static_mounts(p, protect_home_tmpfs_table, ELEMENTSOF(protect_home_tmpfs_table), ignore_protect); | |
638 | ||
b6c432ca | 639 | case PROTECT_HOME_YES: |
5327c910 LP |
640 | return append_static_mounts(p, protect_home_yes_table, ELEMENTSOF(protect_home_yes_table), ignore_protect); |
641 | ||
b6c432ca | 642 | default: |
04499a70 | 643 | assert_not_reached(); |
b6c432ca | 644 | } |
b6c432ca DH |
645 | } |
646 | ||
34de407a | 647 | static int append_protect_system(MountEntry **p, ProtectSystem protect_system, bool ignore_protect) { |
f471b2af DH |
648 | assert(p); |
649 | ||
5327c910 LP |
650 | switch (protect_system) { |
651 | ||
652 | case PROTECT_SYSTEM_NO: | |
f471b2af DH |
653 | return 0; |
654 | ||
f471b2af | 655 | case PROTECT_SYSTEM_STRICT: |
5327c910 LP |
656 | return append_static_mounts(p, protect_system_strict_table, ELEMENTSOF(protect_system_strict_table), ignore_protect); |
657 | ||
f471b2af | 658 | case PROTECT_SYSTEM_YES: |
5327c910 LP |
659 | return append_static_mounts(p, protect_system_yes_table, ELEMENTSOF(protect_system_yes_table), ignore_protect); |
660 | ||
f471b2af | 661 | case PROTECT_SYSTEM_FULL: |
5327c910 LP |
662 | return append_static_mounts(p, protect_system_full_table, ELEMENTSOF(protect_system_full_table), ignore_protect); |
663 | ||
f471b2af | 664 | default: |
04499a70 | 665 | assert_not_reached(); |
f471b2af | 666 | } |
11a30cec DH |
667 | } |
668 | ||
93bab288 | 669 | static int mount_path_compare(const MountEntry *a, const MountEntry *b) { |
a0827e2b | 670 | int d; |
15ae422b | 671 | |
a07b9926 | 672 | /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first |
93f59701 LB |
673 | * regardless of the prefix - they are set up in the propagate directory anyway */ |
674 | d = -CMP(a->mode == EXTENSION_IMAGES, b->mode == EXTENSION_IMAGES); | |
a07b9926 LB |
675 | if (d != 0) |
676 | return d; | |
677 | d = -CMP(a->mode == EXTENSION_DIRECTORIES, b->mode == EXTENSION_DIRECTORIES); | |
93f59701 LB |
678 | if (d != 0) |
679 | return d; | |
680 | ||
6ee1a919 | 681 | /* If the paths are not equal, then order prefixes first */ |
93bab288 | 682 | d = path_compare(mount_entry_path(a), mount_entry_path(b)); |
6ee1a919 LP |
683 | if (d != 0) |
684 | return d; | |
15ae422b | 685 | |
6ee1a919 | 686 | /* If the paths are equal, check the mode */ |
93bab288 | 687 | return CMP((int) a->mode, (int) b->mode); |
15ae422b LP |
688 | } |
689 | ||
da6053d0 | 690 | static int prefix_where_needed(MountEntry *m, size_t n, const char *root_directory) { |
4a756839 | 691 | /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */ |
5327c910 | 692 | |
fe96c0f8 SS |
693 | assert(m || n == 0); |
694 | ||
695 | for (size_t i = 0; i < n; i++) { | |
5327c910 LP |
696 | char *s; |
697 | ||
698 | if (m[i].has_prefix) | |
699 | continue; | |
700 | ||
c6134d3e | 701 | s = path_join(root_directory, mount_entry_path(m+i)); |
5327c910 LP |
702 | if (!s) |
703 | return -ENOMEM; | |
704 | ||
809ceb82 | 705 | mount_entry_consume_prefix(&m[i], s); |
5327c910 LP |
706 | } |
707 | ||
708 | return 0; | |
709 | } | |
710 | ||
da6053d0 | 711 | static void drop_duplicates(MountEntry *m, size_t *n) { |
34de407a | 712 | MountEntry *f, *t, *previous; |
15ae422b | 713 | |
c17ec25e | 714 | assert(m); |
15ae422b | 715 | assert(n); |
15ae422b | 716 | |
fe3c2583 LP |
717 | /* Drops duplicate entries. Expects that the array is properly ordered already. */ |
718 | ||
1d54cd5d | 719 | for (f = m, t = m, previous = NULL; f < m + *n; f++) { |
15ae422b | 720 | |
fe3c2583 | 721 | /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare() |
088696fe LP |
722 | * above. Note that we only drop duplicates that haven't been mounted yet. */ |
723 | if (previous && | |
724 | path_equal(mount_entry_path(f), mount_entry_path(previous)) && | |
725 | !f->applied && !previous->applied) { | |
5beb8688 | 726 | log_debug("%s (%s) is duplicate.", mount_entry_path(f), mount_mode_to_string(f->mode)); |
ddc155b2 TM |
727 | /* Propagate the flags to the remaining entry */ |
728 | previous->read_only = previous->read_only || mount_entry_read_only(f); | |
729 | previous->noexec = previous->noexec || mount_entry_noexec(f); | |
730 | previous->exec = previous->exec || mount_entry_exec(f); | |
1eb7e08e | 731 | mount_entry_done(f); |
15ae422b | 732 | continue; |
fe3c2583 | 733 | } |
15ae422b | 734 | |
e2d7c1a0 | 735 | *t = *f; |
15ae422b | 736 | previous = t; |
fe3c2583 LP |
737 | t++; |
738 | } | |
739 | ||
740 | *n = t - m; | |
741 | } | |
742 | ||
da6053d0 | 743 | static void drop_inaccessible(MountEntry *m, size_t *n) { |
34de407a | 744 | MountEntry *f, *t; |
fe3c2583 LP |
745 | const char *clear = NULL; |
746 | ||
747 | assert(m); | |
748 | assert(n); | |
749 | ||
750 | /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly | |
751 | * ordered already. */ | |
752 | ||
1d54cd5d | 753 | for (f = m, t = m; f < m + *n; f++) { |
fe3c2583 LP |
754 | |
755 | /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop | |
756 | * it, as inaccessible paths really should drop the entire subtree. */ | |
34de407a LP |
757 | if (clear && path_startswith(mount_entry_path(f), clear)) { |
758 | log_debug("%s is masked by %s.", mount_entry_path(f), clear); | |
1eb7e08e | 759 | mount_entry_done(f); |
fe3c2583 LP |
760 | continue; |
761 | } | |
15ae422b | 762 | |
34de407a | 763 | clear = f->mode == INACCESSIBLE ? mount_entry_path(f) : NULL; |
fe3c2583 LP |
764 | |
765 | *t = *f; | |
15ae422b LP |
766 | t++; |
767 | } | |
768 | ||
c17ec25e | 769 | *n = t - m; |
15ae422b LP |
770 | } |
771 | ||
da6053d0 | 772 | static void drop_nop(MountEntry *m, size_t *n) { |
34de407a | 773 | MountEntry *f, *t; |
7648a565 LP |
774 | |
775 | assert(m); | |
776 | assert(n); | |
777 | ||
778 | /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the | |
779 | * list is ordered by prefixes. */ | |
780 | ||
1d54cd5d | 781 | for (f = m, t = m; f < m + *n; f++) { |
7648a565 | 782 | |
1e05071d YW |
783 | /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */ |
784 | if (IN_SET(f->mode, READONLY, READWRITE, READWRITE_IMPLICIT)) { | |
e7bf2fca | 785 | MountEntry *found = NULL; |
7648a565 LP |
786 | |
787 | /* Now let's find the first parent of the entry we are looking at. */ | |
e7bf2fca | 788 | for (MountEntry *p = PTR_SUB1(t, m); p; p = PTR_SUB1(p, m)) |
34de407a | 789 | if (path_startswith(mount_entry_path(f), mount_entry_path(p))) { |
e7bf2fca | 790 | found = p; |
7648a565 LP |
791 | break; |
792 | } | |
7648a565 LP |
793 | |
794 | /* We found it, let's see if it's the same mode, if so, we can drop this entry */ | |
e7bf2fca | 795 | if (found && found->mode == f->mode) { |
5beb8688 YW |
796 | log_debug("%s (%s) is made redundant by %s (%s)", |
797 | mount_entry_path(f), mount_mode_to_string(f->mode), | |
e7bf2fca | 798 | mount_entry_path(found), mount_mode_to_string(found->mode)); |
1eb7e08e | 799 | mount_entry_done(f); |
7648a565 LP |
800 | continue; |
801 | } | |
802 | } | |
803 | ||
804 | *t = *f; | |
805 | t++; | |
806 | } | |
807 | ||
808 | *n = t - m; | |
809 | } | |
810 | ||
da6053d0 | 811 | static void drop_outside_root(const char *root_directory, MountEntry *m, size_t *n) { |
34de407a | 812 | MountEntry *f, *t; |
cd2902c9 LP |
813 | |
814 | assert(m); | |
815 | assert(n); | |
816 | ||
1d54cd5d | 817 | /* Nothing to do */ |
cd2902c9 LP |
818 | if (!root_directory) |
819 | return; | |
820 | ||
821 | /* Drops all mounts that are outside of the root directory. */ | |
822 | ||
1d54cd5d | 823 | for (f = m, t = m; f < m + *n; f++) { |
cd2902c9 | 824 | |
a07b9926 LB |
825 | /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */ |
826 | if (!IN_SET(f->mode, EXTENSION_IMAGES, EXTENSION_DIRECTORIES) && !path_startswith(mount_entry_path(f), root_directory)) { | |
34de407a | 827 | log_debug("%s is outside of root directory.", mount_entry_path(f)); |
1eb7e08e | 828 | mount_entry_done(f); |
cd2902c9 LP |
829 | continue; |
830 | } | |
831 | ||
832 | *t = *f; | |
833 | t++; | |
834 | } | |
835 | ||
836 | *n = t - m; | |
837 | } | |
838 | ||
b2a60844 LP |
839 | static int clone_device_node( |
840 | const char *d, | |
841 | const char *temporary_mount, | |
842 | bool *make_devnode) { | |
843 | ||
844 | _cleanup_free_ char *sl = NULL; | |
845 | const char *dn, *bn, *t; | |
b5e99f23 ДГ |
846 | struct stat st; |
847 | int r; | |
848 | ||
414b304b | 849 | if (stat(d, &st) < 0) { |
b2a60844 LP |
850 | if (errno == ENOENT) { |
851 | log_debug_errno(errno, "Device node '%s' to clone does not exist, ignoring.", d); | |
af984e13 | 852 | return -ENXIO; |
b2a60844 LP |
853 | } |
854 | ||
855 | return log_debug_errno(errno, "Failed to stat() device node '%s' to clone, ignoring: %m", d); | |
b5e99f23 ДГ |
856 | } |
857 | ||
858 | if (!S_ISBLK(st.st_mode) && | |
baaa35ad ZJS |
859 | !S_ISCHR(st.st_mode)) |
860 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
861 | "Device node '%s' to clone is not a device node, ignoring.", | |
862 | d); | |
b5e99f23 | 863 | |
6f7f3a33 | 864 | dn = strjoina(temporary_mount, d); |
b5e99f23 | 865 | |
b2a60844 | 866 | /* First, try to create device node properly */ |
16498617 CB |
867 | if (*make_devnode) { |
868 | mac_selinux_create_file_prepare(d, st.st_mode); | |
869 | r = mknod(dn, st.st_mode, st.st_rdev); | |
870 | mac_selinux_create_file_clear(); | |
b2a60844 LP |
871 | if (r >= 0) |
872 | goto add_symlink; | |
16498617 CB |
873 | if (errno != EPERM) |
874 | return log_debug_errno(errno, "mknod failed for %s: %m", d); | |
875 | ||
b2a60844 | 876 | /* This didn't work, let's not try this again for the next iterations. */ |
16498617 CB |
877 | *make_devnode = false; |
878 | } | |
879 | ||
d73020f2 | 880 | /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target. |
1acf344d | 881 | * Do not prepare device-node SELinux label (see issue 13762) */ |
16498617 | 882 | r = mknod(dn, S_IFREG, 0); |
16498617 | 883 | if (r < 0 && errno != EEXIST) |
b2a60844 | 884 | return log_debug_errno(errno, "mknod() fallback failed for '%s': %m", d); |
16498617 | 885 | |
21935150 LP |
886 | /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard |
887 | * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or | |
888 | * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */ | |
889 | r = mount_nofollow_verbose(LOG_DEBUG, d, dn, NULL, MS_BIND, NULL); | |
890 | if (r < 0) | |
891 | return r; | |
b2a60844 LP |
892 | |
893 | add_symlink: | |
894 | bn = path_startswith(d, "/dev/"); | |
895 | if (!bn) | |
896 | return 0; | |
897 | ||
898 | /* Create symlinks like /dev/char/1:9 → ../urandom */ | |
ec61371f | 899 | if (asprintf(&sl, "%s/dev/%s/" DEVNUM_FORMAT_STR, |
cbc056c8 ZJS |
900 | temporary_mount, |
901 | S_ISCHR(st.st_mode) ? "char" : "block", | |
ec61371f | 902 | DEVNUM_FORMAT_VAL(st.st_rdev)) < 0) |
b2a60844 LP |
903 | return log_oom(); |
904 | ||
905 | (void) mkdir_parents(sl, 0755); | |
906 | ||
907 | t = strjoina("../", bn); | |
b2a60844 | 908 | if (symlink(t, sl) < 0) |
2e4a4fae | 909 | log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl); |
b5e99f23 | 910 | |
af984e13 | 911 | return 0; |
b5e99f23 ДГ |
912 | } |
913 | ||
5d997827 | 914 | static int mount_private_dev(MountEntry *m) { |
7f112f50 LP |
915 | static const char devnodes[] = |
916 | "/dev/null\0" | |
917 | "/dev/zero\0" | |
918 | "/dev/full\0" | |
919 | "/dev/random\0" | |
920 | "/dev/urandom\0" | |
921 | "/dev/tty\0"; | |
922 | ||
2b85f4e1 | 923 | char temporary_mount[] = "/tmp/namespace-dev-XXXXXX"; |
12e2b70f | 924 | const char *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL; |
16498617 | 925 | bool can_mknod = true; |
7f112f50 LP |
926 | int r; |
927 | ||
928 | assert(m); | |
929 | ||
2b85f4e1 | 930 | if (!mkdtemp(temporary_mount)) |
2e4a4fae | 931 | return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount); |
2b85f4e1 | 932 | |
63c372cb | 933 | dev = strjoina(temporary_mount, "/dev"); |
dc751688 | 934 | (void) mkdir(dev, 0755); |
9f563f27 | 935 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV); |
21935150 | 936 | if (r < 0) |
2b85f4e1 | 937 | goto fail; |
21935150 | 938 | |
03bc11d1 | 939 | r = label_fix_full(AT_FDCWD, dev, "/dev", 0); |
c3151977 | 940 | if (r < 0) { |
1ce268c7 | 941 | log_debug_errno(r, "Failed to fix label of '%s' as /dev: %m", dev); |
c3151977 TM |
942 | goto fail; |
943 | } | |
2b85f4e1 | 944 | |
63c372cb | 945 | devpts = strjoina(temporary_mount, "/dev/pts"); |
dc751688 | 946 | (void) mkdir(devpts, 0755); |
21935150 LP |
947 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/pts", devpts, NULL, MS_BIND, NULL); |
948 | if (r < 0) | |
2b85f4e1 | 949 | goto fail; |
2b85f4e1 | 950 | |
2e4a4fae YW |
951 | /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx. |
952 | * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible. | |
953 | * Thus, in that case make a clone. | |
954 | * In nspawn and other containers it will be a symlink, in that case make it a symlink. */ | |
36ce7110 | 955 | r = is_symlink("/dev/ptmx"); |
2e4a4fae YW |
956 | if (r < 0) { |
957 | log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m"); | |
3164e3cb | 958 | goto fail; |
2e4a4fae | 959 | } else if (r > 0) { |
414b304b ДГ |
960 | devptmx = strjoina(temporary_mount, "/dev/ptmx"); |
961 | if (symlink("pts/ptmx", devptmx) < 0) { | |
2e4a4fae | 962 | r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx); |
414b304b ДГ |
963 | goto fail; |
964 | } | |
965 | } else { | |
16498617 | 966 | r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod); |
152c475f LP |
967 | if (r < 0) |
968 | goto fail; | |
414b304b | 969 | } |
e06b6479 | 970 | |
63c372cb | 971 | devshm = strjoina(temporary_mount, "/dev/shm"); |
8d953682 | 972 | (void) mkdir(devshm, 0755); |
21935150 LP |
973 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/shm", devshm, NULL, MS_BIND, NULL); |
974 | if (r < 0) | |
2b85f4e1 | 975 | goto fail; |
2b85f4e1 | 976 | |
63c372cb | 977 | devmqueue = strjoina(temporary_mount, "/dev/mqueue"); |
dc751688 | 978 | (void) mkdir(devmqueue, 0755); |
21935150 | 979 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/mqueue", devmqueue, NULL, MS_BIND, NULL); |
2b85f4e1 | 980 | |
63c372cb | 981 | devhugepages = strjoina(temporary_mount, "/dev/hugepages"); |
dc751688 | 982 | (void) mkdir(devhugepages, 0755); |
21935150 | 983 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/hugepages", devhugepages, NULL, MS_BIND, NULL); |
2b85f4e1 | 984 | |
63c372cb | 985 | devlog = strjoina(temporary_mount, "/dev/log"); |
2e4a4fae YW |
986 | if (symlink("/run/systemd/journal/dev-log", devlog) < 0) |
987 | log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog); | |
82d25240 | 988 | |
7f112f50 | 989 | NULSTR_FOREACH(d, devnodes) { |
16498617 | 990 | r = clone_device_node(d, temporary_mount, &can_mknod); |
37b22b3b | 991 | /* ENXIO means the *source* is not a device file, skip creation in that case */ |
af984e13 | 992 | if (r < 0 && r != -ENXIO) |
2b85f4e1 | 993 | goto fail; |
7f112f50 LP |
994 | } |
995 | ||
2e4a4fae YW |
996 | r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID); |
997 | if (r < 0) | |
105a1a36 | 998 | log_debug_errno(r, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount); |
7f112f50 | 999 | |
d73020f2 LP |
1000 | /* Create the /dev directory if missing. It is more likely to be missing when the service is started |
1001 | * with RootDirectory. This is consistent with mount units creating the mount points when missing. */ | |
34de407a | 1002 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
ee818b89 | 1003 | |
9e5f8252 | 1004 | /* Unmount everything in old /dev */ |
2e4a4fae YW |
1005 | r = umount_recursive(mount_entry_path(m), 0); |
1006 | if (r < 0) | |
1007 | log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m)); | |
1008 | ||
21935150 LP |
1009 | r = mount_nofollow_verbose(LOG_DEBUG, dev, mount_entry_path(m), NULL, MS_MOVE, NULL); |
1010 | if (r < 0) | |
2b85f4e1 | 1011 | goto fail; |
7f112f50 | 1012 | |
1019a48f LP |
1013 | (void) rmdir(dev); |
1014 | (void) rmdir(temporary_mount); | |
7f112f50 | 1015 | |
2b85f4e1 | 1016 | return 0; |
7f112f50 | 1017 | |
2b85f4e1 LP |
1018 | fail: |
1019 | if (devpts) | |
21935150 | 1020 | (void) umount_verbose(LOG_DEBUG, devpts, UMOUNT_NOFOLLOW); |
7f112f50 | 1021 | |
2b85f4e1 | 1022 | if (devshm) |
21935150 | 1023 | (void) umount_verbose(LOG_DEBUG, devshm, UMOUNT_NOFOLLOW); |
7f112f50 | 1024 | |
2b85f4e1 | 1025 | if (devhugepages) |
21935150 | 1026 | (void) umount_verbose(LOG_DEBUG, devhugepages, UMOUNT_NOFOLLOW); |
7f112f50 | 1027 | |
2b85f4e1 | 1028 | if (devmqueue) |
21935150 | 1029 | (void) umount_verbose(LOG_DEBUG, devmqueue, UMOUNT_NOFOLLOW); |
7f112f50 | 1030 | |
21935150 | 1031 | (void) umount_verbose(LOG_DEBUG, dev, UMOUNT_NOFOLLOW); |
1019a48f LP |
1032 | (void) rmdir(dev); |
1033 | (void) rmdir(temporary_mount); | |
7f112f50 | 1034 | |
2b85f4e1 | 1035 | return r; |
7f112f50 LP |
1036 | } |
1037 | ||
2a2969fd | 1038 | static int mount_bind_dev(const MountEntry *m) { |
5d997827 LP |
1039 | int r; |
1040 | ||
1041 | assert(m); | |
1042 | ||
d73020f2 LP |
1043 | /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the |
1044 | * service's /dev. This is only used when RootDirectory= is set. */ | |
5d997827 | 1045 | |
645767d6 LP |
1046 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1047 | ||
5d997827 LP |
1048 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1049 | if (r < 0) | |
1050 | return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m"); | |
1051 | if (r > 0) /* make this a NOP if /dev is already a mount point */ | |
1052 | return 0; | |
1053 | ||
51bf9b1a | 1054 | return mount_nofollow_verbose(LOG_DEBUG, "/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
5d997827 LP |
1055 | } |
1056 | ||
10028263 | 1057 | static int mount_bind_sysfs(const MountEntry *m) { |
5d997827 LP |
1058 | int r; |
1059 | ||
1060 | assert(m); | |
1061 | ||
645767d6 LP |
1062 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1063 | ||
5d997827 LP |
1064 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1065 | if (r < 0) | |
1066 | return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m"); | |
1067 | if (r > 0) /* make this a NOP if /sys is already a mount point */ | |
1068 | return 0; | |
1069 | ||
1070 | /* Bind mount the host's version so that we get all child mounts of it, too. */ | |
51bf9b1a | 1071 | return mount_nofollow_verbose(LOG_DEBUG, "/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
5d997827 LP |
1072 | } |
1073 | ||
553e3660 YW |
1074 | static int mount_private_sysfs(const MountEntry *m) { |
1075 | const char *entry_path = mount_entry_path(ASSERT_PTR(m)); | |
1076 | int r, n; | |
1077 | ||
1078 | (void) mkdir_p_label(entry_path, 0755); | |
1079 | ||
1080 | n = umount_recursive(entry_path, 0); | |
1081 | ||
1082 | r = mount_nofollow_verbose(LOG_DEBUG, "sysfs", entry_path, "sysfs", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL); | |
1083 | if (ERRNO_IS_NEG_PRIVILEGE(r)) { | |
1084 | /* When we do not have enough privileges to mount sysfs, fall back to use existing /sys. */ | |
1085 | ||
1086 | if (n > 0) | |
1087 | /* /sys or some of sub-mounts are umounted in the above. Refuse incomplete tree. | |
1088 | * Propagate the original error code returned by mount() in the above. */ | |
1089 | return r; | |
1090 | ||
1091 | return mount_bind_sysfs(m); | |
1092 | ||
1093 | } else if (r < 0) | |
1094 | return r; | |
1095 | ||
1096 | /* We mounted a new instance now. Let's bind mount the children over now. */ | |
1097 | (void) bind_mount_submounts("/sys", entry_path); | |
1098 | return 0; | |
1099 | } | |
1100 | ||
4e399953 | 1101 | static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) { |
61f8a7bd | 1102 | _cleanup_free_ char *opts = NULL; |
4e399953 | 1103 | const char *entry_path; |
61f8a7bd | 1104 | int r, n; |
5d997827 LP |
1105 | |
1106 | assert(m); | |
4e399953 | 1107 | assert(ns_info); |
5d997827 | 1108 | |
4e399953 LP |
1109 | if (ns_info->protect_proc != PROTECT_PROC_DEFAULT || |
1110 | ns_info->proc_subset != PROC_SUBSET_ALL) { | |
4e399953 LP |
1111 | |
1112 | /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it | |
1113 | * pretended to be per-instance but actually was per-namespace), hence let's make use of it | |
1114 | * if requested. To make sure this logic succeeds only on kernels where hidepid= is | |
1115 | * per-instance, we'll exclusively use the textual value for hidepid=, since support was | |
1116 | * added in the same commit: if it's supported it is thus also per-instance. */ | |
1117 | ||
1c265fcd | 1118 | const char *hpv = ns_info->protect_proc == PROTECT_PROC_DEFAULT ? |
7c76e181 ZJS |
1119 | "off" : |
1120 | protect_proc_to_string(ns_info->protect_proc); | |
1c265fcd DDM |
1121 | |
1122 | /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in | |
1123 | * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when | |
1124 | * trying to use hidepid= on systems where it isn't supported. The same applies for subset=. | |
1125 | * fsopen()/fsconfig() was also backported on some distros which allows us to detect | |
1126 | * hidepid=/subset= support in even more scenarios. */ | |
1127 | ||
117e7034 | 1128 | if (mount_option_supported("proc", "hidepid", hpv) != 0) { |
1c265fcd DDM |
1129 | opts = strjoin("hidepid=", hpv); |
1130 | if (!opts) | |
1131 | return -ENOMEM; | |
1132 | } | |
1133 | ||
117e7034 LP |
1134 | if (ns_info->proc_subset == PROC_SUBSET_PID && |
1135 | mount_option_supported("proc", "subset", "pid") != 0) | |
1c265fcd DDM |
1136 | if (!strextend_with_separator(&opts, ",", "subset=pid")) |
1137 | return -ENOMEM; | |
4e399953 LP |
1138 | } |
1139 | ||
61f8a7bd YW |
1140 | entry_path = mount_entry_path(m); |
1141 | (void) mkdir_p_label(entry_path, 0755); | |
1142 | ||
1143 | /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in | |
1144 | * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by | |
1145 | * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything | |
1146 | * mounted on /proc/ first. */ | |
1147 | ||
1148 | n = umount_recursive(entry_path, 0); | |
1149 | ||
1150 | r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, opts); | |
1151 | if (r == -EINVAL && opts) | |
1152 | /* If this failed with EINVAL then this likely means the textual hidepid= stuff is | |
1153 | * not supported by the kernel, and thus the per-instance hidepid= neither, which | |
1154 | * means we really don't want to use it, since it would affect our host's /proc | |
1155 | * mount. Hence let's gracefully fallback to a classic, unrestricted version. */ | |
1156 | r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL); | |
cc7e2dff YW |
1157 | if (ERRNO_IS_NEG_PRIVILEGE(r)) { |
1158 | /* When we do not have enough privileges to mount /proc, fall back to use existing /proc. */ | |
61f8a7bd YW |
1159 | |
1160 | if (n > 0) | |
1161 | /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree. | |
1162 | * Propagate the original error code returned by mount() in the above. */ | |
cc7e2dff | 1163 | return r; |
61f8a7bd YW |
1164 | |
1165 | r = path_is_mount_point(entry_path, NULL, 0); | |
1166 | if (r < 0) | |
1167 | return log_debug_errno(r, "Unable to determine whether /proc is already mounted: %m"); | |
cc7e2dff YW |
1168 | if (r > 0) |
1169 | return 0; | |
1170 | ||
1171 | /* We lack permissions to mount a new instance of /proc, and it is not already mounted. But | |
1172 | * we can access the host's, so as a final fallback bind-mount it to the destination, as most | |
1173 | * likely we are inside a user manager in an unprivileged user namespace. */ | |
1174 | return mount_nofollow_verbose(LOG_DEBUG, "/proc", entry_path, NULL, MS_BIND|MS_REC, NULL); | |
1175 | ||
ad74f28a YW |
1176 | } else if (r < 0) |
1177 | return r; | |
5d997827 | 1178 | |
cc7e2dff YW |
1179 | /* We mounted a new instance now. Let's bind mount the children over now. This matters for nspawn |
1180 | * where a bunch of files are overmounted, in particular the boot id */ | |
1181 | (void) bind_mount_submounts("/proc", entry_path); | |
1182 | return 0; | |
5d997827 LP |
1183 | } |
1184 | ||
2abd4e38 | 1185 | static int mount_tmpfs(const MountEntry *m) { |
df6b900a | 1186 | const char *entry_path, *inner_path; |
abad72be | 1187 | int r; |
abad72be | 1188 | |
6c47cd7d LP |
1189 | assert(m); |
1190 | ||
df6b900a | 1191 | entry_path = mount_entry_path(m); |
809ceb82 | 1192 | inner_path = mount_entry_unprefixed_path(m); |
df6b900a | 1193 | |
d73020f2 LP |
1194 | /* First, get rid of everything that is below if there is anything. Then, overmount with our new |
1195 | * tmpfs */ | |
6c47cd7d | 1196 | |
abad72be CG |
1197 | (void) mkdir_p_label(entry_path, 0755); |
1198 | (void) umount_recursive(entry_path, 0); | |
6c47cd7d | 1199 | |
21935150 LP |
1200 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", entry_path, "tmpfs", m->flags, mount_entry_options(m)); |
1201 | if (r < 0) | |
1202 | return r; | |
abad72be | 1203 | |
03bc11d1 | 1204 | r = label_fix_full(AT_FDCWD, entry_path, inner_path, 0); |
abad72be | 1205 | if (r < 0) |
df6b900a | 1206 | return log_debug_errno(r, "Failed to fix label of '%s' as '%s': %m", entry_path, inner_path); |
6c47cd7d | 1207 | |
51bf9b1a | 1208 | return 0; |
6c47cd7d LP |
1209 | } |
1210 | ||
94293d65 LB |
1211 | static int mount_run(const MountEntry *m) { |
1212 | int r; | |
1213 | ||
1214 | assert(m); | |
1215 | ||
1216 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); | |
1217 | if (r < 0 && r != -ENOENT) | |
1218 | return log_debug_errno(r, "Unable to determine whether /run is already mounted: %m"); | |
1219 | if (r > 0) /* make this a NOP if /run is already a mount point */ | |
1220 | return 0; | |
1221 | ||
1222 | return mount_tmpfs(m); | |
1223 | } | |
1224 | ||
80271a44 XR |
1225 | static int mount_mqueuefs(const MountEntry *m) { |
1226 | int r; | |
1227 | const char *entry_path; | |
1228 | ||
1229 | assert(m); | |
1230 | ||
1231 | entry_path = mount_entry_path(m); | |
1232 | ||
1233 | (void) mkdir_p_label(entry_path, 0755); | |
1234 | (void) umount_recursive(entry_path, 0); | |
1235 | ||
1236 | r = mount_nofollow_verbose(LOG_DEBUG, "mqueue", entry_path, "mqueue", m->flags, mount_entry_options(m)); | |
1237 | if (r < 0) | |
1238 | return r; | |
1239 | ||
1240 | return 0; | |
1241 | } | |
1242 | ||
84be0c71 LP |
1243 | static int mount_image( |
1244 | const MountEntry *m, | |
1245 | const char *root_directory, | |
1246 | const ImagePolicy *image_policy) { | |
93f59701 LB |
1247 | |
1248 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
55ea4ef0 MG |
1249 | *host_os_release_level = NULL, *extension_name = NULL; |
1250 | _cleanup_strv_free_ char **extension_release = NULL; | |
1251 | ImageClass class = IMAGE_SYSEXT; | |
b3d13314 LB |
1252 | int r; |
1253 | ||
89e62e0b LP |
1254 | assert(m); |
1255 | ||
55ea4ef0 MG |
1256 | r = path_extract_filename(mount_entry_source(m), &extension_name); |
1257 | if (r < 0) | |
1258 | return log_debug_errno(r, "Failed to extract extension name from %s: %m", mount_entry_source(m)); | |
1259 | ||
93f59701 | 1260 | if (m->mode == EXTENSION_IMAGES) { |
55ea4ef0 MG |
1261 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_SYSEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); |
1262 | if (r == -ENOENT) { | |
1263 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_CONFEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); | |
1264 | if (r >= 0) | |
1265 | class = IMAGE_CONFEXT; | |
1266 | } | |
1267 | if (r == -ENOENT) | |
1268 | return r; | |
1269 | ||
93f59701 LB |
1270 | r = parse_os_release( |
1271 | empty_to_root(root_directory), | |
1272 | "ID", &host_os_release_id, | |
1273 | "VERSION_ID", &host_os_release_version_id, | |
55ea4ef0 | 1274 | image_class_info[class].level_env, &host_os_release_level, |
93f59701 LB |
1275 | NULL); |
1276 | if (r < 0) | |
1277 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
78ab2b50 LB |
1278 | if (isempty(host_os_release_id)) |
1279 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'ID' field not found or empty in 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
93f59701 LB |
1280 | } |
1281 | ||
1282 | r = verity_dissect_and_mount( | |
84be0c71 LP |
1283 | /* src_fd= */ -1, |
1284 | mount_entry_source(m), | |
1285 | mount_entry_path(m), | |
1286 | m->image_options, | |
1287 | image_policy, | |
1288 | host_os_release_id, | |
1289 | host_os_release_version_id, | |
55ea4ef0 | 1290 | host_os_release_level, |
84be0c71 | 1291 | NULL); |
b850a9b2 LB |
1292 | if (r == -ENOENT && m->ignore) |
1293 | return 0; | |
db4c8a25 LB |
1294 | if (r == -ESTALE && host_os_release_id) |
1295 | return log_error_errno(r, | |
1296 | "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s", | |
1297 | mount_entry_source(m), | |
1298 | host_os_release_id, | |
1299 | host_os_release_version_id ? " VERSION_ID=" : "", | |
1300 | strempty(host_os_release_version_id), | |
55ea4ef0 MG |
1301 | host_os_release_level ? image_class_info[class].level_env_print : "", |
1302 | strempty(host_os_release_level)); | |
b3d13314 | 1303 | if (r < 0) |
4beda316 | 1304 | return log_debug_errno(r, "Failed to mount image %s on %s: %m", mount_entry_source(m), mount_entry_path(m)); |
b3d13314 | 1305 | |
51bf9b1a | 1306 | return 0; |
b3d13314 LB |
1307 | } |
1308 | ||
93f59701 LB |
1309 | static int mount_overlay(const MountEntry *m) { |
1310 | const char *options; | |
1311 | int r; | |
1312 | ||
1313 | assert(m); | |
1314 | ||
1315 | options = strjoina("lowerdir=", mount_entry_options(m)); | |
1316 | ||
1317 | (void) mkdir_p_label(mount_entry_path(m), 0755); | |
1318 | ||
1319 | r = mount_nofollow_verbose(LOG_DEBUG, "overlay", mount_entry_path(m), "overlay", MS_RDONLY, options); | |
1320 | if (r == -ENOENT && m->ignore) | |
1321 | return 0; | |
93f59701 | 1322 | |
51bf9b1a | 1323 | return r; |
93f59701 LB |
1324 | } |
1325 | ||
088696fe | 1326 | static int follow_symlink( |
d2d6c096 | 1327 | const char *root_directory, |
088696fe | 1328 | MountEntry *m) { |
d2d6c096 | 1329 | |
088696fe | 1330 | _cleanup_free_ char *target = NULL; |
8fceda93 LP |
1331 | int r; |
1332 | ||
088696fe LP |
1333 | /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we |
1334 | * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at | |
1335 | * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the | |
1336 | * end and already have a fully normalized name. */ | |
8fceda93 | 1337 | |
f461a28d | 1338 | r = chase(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL); |
088696fe LP |
1339 | if (r < 0) |
1340 | return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m)); | |
1341 | if (r > 0) /* Reached the end, nothing more to resolve */ | |
1342 | return 1; | |
8fceda93 | 1343 | |
f461a28d | 1344 | if (m->n_followed >= CHASE_MAX) /* put a boundary on things */ |
baaa35ad ZJS |
1345 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1346 | "Symlink loop on '%s'.", | |
1347 | mount_entry_path(m)); | |
8fceda93 | 1348 | |
e2341b6b DT |
1349 | log_debug("Followed mount entry path symlink %s %s %s.", |
1350 | mount_entry_path(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), target); | |
8fceda93 | 1351 | |
809ceb82 | 1352 | mount_entry_consume_prefix(m, TAKE_PTR(target)); |
8fceda93 | 1353 | |
088696fe LP |
1354 | m->n_followed ++; |
1355 | ||
1356 | return 0; | |
8fceda93 LP |
1357 | } |
1358 | ||
82fb2da2 | 1359 | static int apply_one_mount( |
8fceda93 | 1360 | const char *root_directory, |
4e399953 | 1361 | MountEntry *m, |
84be0c71 LP |
1362 | const ImagePolicy *mount_image_policy, |
1363 | const ImagePolicy *extension_image_policy, | |
4e399953 | 1364 | const NamespaceInfo *ns_info) { |
ac0930c8 | 1365 | |
e5f10caf | 1366 | _cleanup_free_ char *inaccessible = NULL; |
a227a4be | 1367 | bool rbind = true, make = false; |
15ae422b | 1368 | const char *what; |
15ae422b | 1369 | int r; |
15ae422b | 1370 | |
c17ec25e | 1371 | assert(m); |
4e399953 | 1372 | assert(ns_info); |
15ae422b | 1373 | |
34de407a | 1374 | log_debug("Applying namespace mount on %s", mount_entry_path(m)); |
fe3c2583 | 1375 | |
c17ec25e | 1376 | switch (m->mode) { |
15ae422b | 1377 | |
160cfdbe | 1378 | case INACCESSIBLE: { |
e5f10caf AZ |
1379 | _cleanup_free_ char *tmp = NULL; |
1380 | const char *runtime_dir; | |
160cfdbe | 1381 | struct stat target; |
6d313367 LP |
1382 | |
1383 | /* First, get rid of everything that is below if there | |
1384 | * is anything... Then, overmount it with an | |
c4b41707 | 1385 | * inaccessible path. */ |
34de407a | 1386 | (void) umount_recursive(mount_entry_path(m), 0); |
6d313367 | 1387 | |
088696fe LP |
1388 | if (lstat(mount_entry_path(m), &target) < 0) { |
1389 | if (errno == ENOENT && m->ignore) | |
1390 | return 0; | |
1391 | ||
cbc056c8 ZJS |
1392 | return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m", |
1393 | mount_entry_path(m)); | |
088696fe | 1394 | } |
15ae422b | 1395 | |
e5f10caf | 1396 | if (geteuid() == 0) |
48b747fa | 1397 | runtime_dir = "/run"; |
e5f10caf | 1398 | else { |
48b747fa LP |
1399 | if (asprintf(&tmp, "/run/user/" UID_FMT, geteuid()) < 0) |
1400 | return -ENOMEM; | |
e5f10caf AZ |
1401 | |
1402 | runtime_dir = tmp; | |
1403 | } | |
1404 | ||
1405 | r = mode_to_inaccessible_node(runtime_dir, target.st_mode, &inaccessible); | |
1406 | if (r < 0) | |
baaa35ad ZJS |
1407 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1408 | "File type not supported for inaccessible mounts. Note that symlinks are not allowed"); | |
e5f10caf | 1409 | what = inaccessible; |
c4b41707 | 1410 | break; |
160cfdbe | 1411 | } |
fe3c2583 | 1412 | |
15ae422b | 1413 | case READONLY: |
15ae422b | 1414 | case READWRITE: |
1e05071d | 1415 | case READWRITE_IMPLICIT: |
ddc155b2 TM |
1416 | case EXEC: |
1417 | case NOEXEC: | |
8fceda93 | 1418 | r = path_is_mount_point(mount_entry_path(m), root_directory, 0); |
088696fe LP |
1419 | if (r == -ENOENT && m->ignore) |
1420 | return 0; | |
d944dc95 | 1421 | if (r < 0) |
cbc056c8 ZJS |
1422 | return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m", |
1423 | mount_entry_path(m)); | |
1424 | if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY | |
ddc155b2 | 1425 | * and MS_NOEXEC bits for the mount point if needed. */ |
6b7c9f8b | 1426 | return 0; |
6b7c9f8b | 1427 | /* This isn't a mount point yet, let's make it one. */ |
34de407a | 1428 | what = mount_entry_path(m); |
6b7c9f8b | 1429 | break; |
15ae422b | 1430 | |
a07b9926 LB |
1431 | case EXTENSION_DIRECTORIES: { |
1432 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
55ea4ef0 | 1433 | *host_os_release_level = NULL, *extension_name = NULL; |
a07b9926 | 1434 | _cleanup_strv_free_ char **extension_release = NULL; |
55ea4ef0 | 1435 | ImageClass class = IMAGE_SYSEXT; |
a07b9926 LB |
1436 | |
1437 | r = path_extract_filename(mount_entry_source(m), &extension_name); | |
1438 | if (r < 0) | |
1439 | return log_debug_errno(r, "Failed to extract extension name from %s: %m", mount_entry_source(m)); | |
1440 | ||
55ea4ef0 MG |
1441 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_SYSEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); |
1442 | if (r == -ENOENT) { | |
1443 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_CONFEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); | |
1444 | if (r >= 0) | |
1445 | class = IMAGE_CONFEXT; | |
1446 | } | |
1447 | if (r == -ENOENT) | |
1448 | return r; | |
1449 | ||
a07b9926 LB |
1450 | r = parse_os_release( |
1451 | empty_to_root(root_directory), | |
1452 | "ID", &host_os_release_id, | |
1453 | "VERSION_ID", &host_os_release_version_id, | |
55ea4ef0 | 1454 | image_class_info[class].level_env, &host_os_release_level, |
a07b9926 LB |
1455 | NULL); |
1456 | if (r < 0) | |
1457 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
1458 | if (isempty(host_os_release_id)) | |
1459 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'ID' field not found or empty in 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
1460 | ||
55ea4ef0 | 1461 | r = load_extension_release_pairs(mount_entry_source(m), class, extension_name, /* relax_extension_release_check= */ false, &extension_release); |
a07b9926 LB |
1462 | if (r == -ENOENT && m->ignore) |
1463 | return 0; | |
1464 | if (r < 0) | |
1465 | return log_debug_errno(r, "Failed to parse directory %s extension-release metadata: %m", extension_name); | |
1466 | ||
1467 | r = extension_release_validate( | |
1468 | extension_name, | |
1469 | host_os_release_id, | |
1470 | host_os_release_version_id, | |
55ea4ef0 MG |
1471 | host_os_release_level, |
1472 | /* host_extension_scope */ NULL, /* Leave empty, we need to accept both system and portable */ | |
30dfe035 | 1473 | extension_release, |
55ea4ef0 | 1474 | class); |
a07b9926 LB |
1475 | if (r == 0) |
1476 | return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Directory %s extension-release metadata does not match the root's", extension_name); | |
1477 | if (r < 0) | |
1478 | return log_debug_errno(r, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name); | |
1479 | ||
1480 | _fallthrough_; | |
1481 | } | |
1482 | ||
d2d6c096 LP |
1483 | case BIND_MOUNT: |
1484 | rbind = false; | |
d2d6c096 | 1485 | |
4831981d | 1486 | _fallthrough_; |
088696fe LP |
1487 | case BIND_MOUNT_RECURSIVE: { |
1488 | _cleanup_free_ char *chased = NULL; | |
5d997827 | 1489 | |
cbc056c8 ZJS |
1490 | /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note |
1491 | * that bind mount source paths are always relative to the host root, hence we pass NULL as | |
f461a28d | 1492 | * root directory to chase() here. */ |
088696fe | 1493 | |
f461a28d | 1494 | r = chase(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL); |
088696fe LP |
1495 | if (r == -ENOENT && m->ignore) { |
1496 | log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m)); | |
1497 | return 0; | |
1498 | } | |
1499 | if (r < 0) | |
1500 | return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m)); | |
1501 | ||
e2341b6b DT |
1502 | log_debug("Followed source symlinks %s %s %s.", |
1503 | mount_entry_source(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), chased); | |
088696fe LP |
1504 | |
1505 | free_and_replace(m->source_malloc, chased); | |
d2d6c096 LP |
1506 | |
1507 | what = mount_entry_source(m); | |
a227a4be | 1508 | make = true; |
d2d6c096 | 1509 | break; |
088696fe | 1510 | } |
d2d6c096 | 1511 | |
6c47cd7d | 1512 | case EMPTY_DIR: |
2abd4e38 YW |
1513 | case TMPFS: |
1514 | return mount_tmpfs(m); | |
6c47cd7d | 1515 | |
ac0930c8 | 1516 | case PRIVATE_TMP: |
56a13a49 | 1517 | case PRIVATE_TMP_READONLY: |
89bd586c | 1518 | what = mount_entry_source(m); |
a227a4be | 1519 | make = true; |
15ae422b | 1520 | break; |
e364ad06 | 1521 | |
d6797c92 | 1522 | case PRIVATE_DEV: |
5d997827 LP |
1523 | return mount_private_dev(m); |
1524 | ||
1525 | case BIND_DEV: | |
1526 | return mount_bind_dev(m); | |
1527 | ||
277b269e YW |
1528 | case PRIVATE_SYSFS: |
1529 | return mount_private_sysfs(m); | |
1530 | ||
10028263 YW |
1531 | case BIND_SYSFS: |
1532 | return mount_bind_sysfs(m); | |
5d997827 LP |
1533 | |
1534 | case PROCFS: | |
4e399953 | 1535 | return mount_procfs(m, ns_info); |
d6797c92 | 1536 | |
94293d65 LB |
1537 | case RUN: |
1538 | return mount_run(m); | |
1539 | ||
80271a44 XR |
1540 | case MQUEUEFS: |
1541 | return mount_mqueuefs(m); | |
1542 | ||
b3d13314 | 1543 | case MOUNT_IMAGES: |
84be0c71 | 1544 | return mount_image(m, NULL, mount_image_policy); |
93f59701 LB |
1545 | |
1546 | case EXTENSION_IMAGES: | |
84be0c71 | 1547 | return mount_image(m, root_directory, extension_image_policy); |
93f59701 LB |
1548 | |
1549 | case OVERLAY_MOUNT: | |
1550 | return mount_overlay(m); | |
b3d13314 | 1551 | |
94fe4cf2 YW |
1552 | case MKDIR: |
1553 | r = mkdir_p_label(mount_entry_path(m), 0755); | |
1554 | if (r < 0) | |
1555 | return r; | |
1556 | return 1; | |
1557 | ||
e364ad06 | 1558 | default: |
04499a70 | 1559 | assert_not_reached(); |
15ae422b LP |
1560 | } |
1561 | ||
ac0930c8 | 1562 | assert(what); |
15ae422b | 1563 | |
21935150 LP |
1564 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); |
1565 | if (r < 0) { | |
a227a4be | 1566 | bool try_again = false; |
a227a4be LP |
1567 | |
1568 | if (r == -ENOENT && make) { | |
8bab8029 | 1569 | int q; |
a227a4be | 1570 | |
cbc056c8 ZJS |
1571 | /* Hmm, either the source or the destination are missing. Let's see if we can create |
1572 | the destination, then try again. */ | |
a227a4be | 1573 | |
8bab8029 | 1574 | (void) mkdir_parents(mount_entry_path(m), 0755); |
a227a4be | 1575 | |
8bab8029 | 1576 | q = make_mount_point_inode_from_path(what, mount_entry_path(m), 0755); |
9d6d4c30 | 1577 | if (q < 0 && q != -EEXIST) |
8bab8029 LB |
1578 | log_error_errno(q, "Failed to create destination mount point node '%s': %m", |
1579 | mount_entry_path(m)); | |
1580 | else | |
1581 | try_again = true; | |
a227a4be LP |
1582 | } |
1583 | ||
21935150 LP |
1584 | if (try_again) |
1585 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); | |
a227a4be | 1586 | if (r < 0) |
5dc60faa | 1587 | return log_error_errno(r, "Failed to mount %s to %s: %m", what, mount_entry_path(m)); |
a227a4be | 1588 | } |
6b7c9f8b | 1589 | |
34de407a | 1590 | log_debug("Successfully mounted %s to %s", what, mount_entry_path(m)); |
6b7c9f8b | 1591 | return 0; |
ac0930c8 | 1592 | } |
15ae422b | 1593 | |
6b000af4 | 1594 | static int make_read_only(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
9ce4e4b0 | 1595 | unsigned long new_flags = 0, flags_mask = 0; |
57ccd9f6 YW |
1596 | bool submounts; |
1597 | int r; | |
15ae422b | 1598 | |
c17ec25e | 1599 | assert(m); |
ac9de0b3 | 1600 | assert(proc_self_mountinfo); |
ac0930c8 | 1601 | |
9ce4e4b0 LP |
1602 | if (mount_entry_read_only(m) || m->mode == PRIVATE_DEV) { |
1603 | new_flags |= MS_RDONLY; | |
1604 | flags_mask |= MS_RDONLY; | |
1605 | } | |
1606 | ||
1607 | if (m->nosuid) { | |
1608 | new_flags |= MS_NOSUID; | |
1609 | flags_mask |= MS_NOSUID; | |
1610 | } | |
1611 | ||
1612 | if (flags_mask == 0) /* No Change? */ | |
6b7c9f8b LP |
1613 | return 0; |
1614 | ||
9ce4e4b0 LP |
1615 | /* We generally apply these changes recursively, except for /dev, and the cases we know there's |
1616 | * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the | |
1617 | * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace | |
1618 | * and running Linux <= 4.17. */ | |
1619 | submounts = | |
1620 | mount_entry_read_only(m) && | |
1621 | !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1622 | if (submounts) | |
6b000af4 | 1623 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); |
9ce4e4b0 | 1624 | else |
7cce68e1 | 1625 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); |
9ce4e4b0 | 1626 | |
ddc155b2 | 1627 | /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked |
867189b5 LP |
1628 | * read-only already stays this way. This improves compatibility with container managers, where we |
1629 | * won't attempt to undo read-only mounts already applied. */ | |
ac0930c8 | 1630 | |
8fceda93 | 1631 | if (r == -ENOENT && m->ignore) |
867189b5 | 1632 | return 0; |
763a260a | 1633 | if (r < 0) |
9ce4e4b0 | 1634 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), |
763a260a | 1635 | submounts ? " and its submounts" : ""); |
763a260a | 1636 | return 0; |
d944dc95 LP |
1637 | } |
1638 | ||
ddc155b2 TM |
1639 | static int make_noexec(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
1640 | unsigned long new_flags = 0, flags_mask = 0; | |
57ccd9f6 YW |
1641 | bool submounts; |
1642 | int r; | |
ddc155b2 TM |
1643 | |
1644 | assert(m); | |
1645 | assert(proc_self_mountinfo); | |
1646 | ||
1647 | if (mount_entry_noexec(m)) { | |
1648 | new_flags |= MS_NOEXEC; | |
1649 | flags_mask |= MS_NOEXEC; | |
1650 | } else if (mount_entry_exec(m)) { | |
1651 | new_flags &= ~MS_NOEXEC; | |
1652 | flags_mask |= MS_NOEXEC; | |
1653 | } | |
1654 | ||
1655 | if (flags_mask == 0) /* No Change? */ | |
1656 | return 0; | |
1657 | ||
1658 | submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1659 | ||
1660 | if (submounts) | |
1661 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); | |
1662 | else | |
1663 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); | |
1664 | ||
1665 | if (r == -ENOENT && m->ignore) | |
1666 | return 0; | |
1667 | if (r < 0) | |
1668 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1669 | submounts ? " and its submounts" : ""); | |
1670 | return 0; | |
1671 | } | |
1672 | ||
6720e356 | 1673 | static int make_nosuid(const MountEntry *m, FILE *proc_self_mountinfo) { |
57ccd9f6 YW |
1674 | bool submounts; |
1675 | int r; | |
6720e356 YW |
1676 | |
1677 | assert(m); | |
1678 | assert(proc_self_mountinfo); | |
1679 | ||
1680 | submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1681 | ||
1682 | if (submounts) | |
1683 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, NULL, proc_self_mountinfo); | |
1684 | else | |
1685 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, proc_self_mountinfo); | |
1686 | if (r == -ENOENT && m->ignore) | |
1687 | return 0; | |
1688 | if (r < 0) | |
1689 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1690 | submounts ? " and its submounts" : ""); | |
1691 | return 0; | |
1692 | } | |
1693 | ||
9b68367b | 1694 | static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) { |
5d997827 LP |
1695 | assert(ns_info); |
1696 | ||
9c988f93 DH |
1697 | /* |
1698 | * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=, | |
1699 | * since to protect the API VFS mounts, they need to be around in the | |
9b68367b | 1700 | * first place... |
9c988f93 | 1701 | */ |
5d997827 | 1702 | |
9b68367b YW |
1703 | return ns_info->mount_apivfs || |
1704 | ns_info->protect_control_groups || | |
4e399953 LP |
1705 | ns_info->protect_kernel_tunables || |
1706 | ns_info->protect_proc != PROTECT_PROC_DEFAULT || | |
1707 | ns_info->proc_subset != PROC_SUBSET_ALL; | |
5d997827 LP |
1708 | } |
1709 | ||
da6053d0 | 1710 | static size_t namespace_calculate_mounts( |
bb0ff3fb | 1711 | const NamespaceInfo *ns_info, |
2652c6c1 DH |
1712 | char** read_write_paths, |
1713 | char** read_only_paths, | |
1714 | char** inaccessible_paths, | |
ddc155b2 TM |
1715 | char** exec_paths, |
1716 | char** no_exec_paths, | |
6c47cd7d | 1717 | char** empty_directories, |
da6053d0 LP |
1718 | size_t n_bind_mounts, |
1719 | size_t n_temporary_filesystems, | |
b3d13314 | 1720 | size_t n_mount_images, |
93f59701 | 1721 | size_t n_extension_images, |
a07b9926 | 1722 | size_t n_extension_directories, |
93f59701 | 1723 | size_t n_hierarchies, |
2652c6c1 DH |
1724 | const char* tmp_dir, |
1725 | const char* var_tmp_dir, | |
bbb4e7f3 | 1726 | const char *creds_path, |
5e8deb94 | 1727 | const char* log_namespace, |
3bdc25a4 | 1728 | bool setup_propagate, |
3f37a825 LB |
1729 | const char* notify_socket, |
1730 | const char* host_os_release) { | |
2652c6c1 | 1731 | |
da6053d0 LP |
1732 | size_t protect_home_cnt; |
1733 | size_t protect_system_cnt = | |
52b3d652 | 1734 | (ns_info->protect_system == PROTECT_SYSTEM_STRICT ? |
f471b2af | 1735 | ELEMENTSOF(protect_system_strict_table) : |
52b3d652 | 1736 | ((ns_info->protect_system == PROTECT_SYSTEM_FULL) ? |
f471b2af | 1737 | ELEMENTSOF(protect_system_full_table) : |
52b3d652 | 1738 | ((ns_info->protect_system == PROTECT_SYSTEM_YES) ? |
f471b2af DH |
1739 | ELEMENTSOF(protect_system_yes_table) : 0))); |
1740 | ||
b6c432ca | 1741 | protect_home_cnt = |
52b3d652 | 1742 | (ns_info->protect_home == PROTECT_HOME_YES ? |
b6c432ca | 1743 | ELEMENTSOF(protect_home_yes_table) : |
52b3d652 | 1744 | ((ns_info->protect_home == PROTECT_HOME_READ_ONLY) ? |
e4da7d8c | 1745 | ELEMENTSOF(protect_home_read_only_table) : |
52b3d652 | 1746 | ((ns_info->protect_home == PROTECT_HOME_TMPFS) ? |
e4da7d8c | 1747 | ELEMENTSOF(protect_home_tmpfs_table) : 0))); |
b6c432ca | 1748 | |
2652c6c1 DH |
1749 | return !!tmp_dir + !!var_tmp_dir + |
1750 | strv_length(read_write_paths) + | |
1751 | strv_length(read_only_paths) + | |
1752 | strv_length(inaccessible_paths) + | |
ddc155b2 TM |
1753 | strv_length(exec_paths) + |
1754 | strv_length(no_exec_paths) + | |
6c47cd7d | 1755 | strv_length(empty_directories) + |
d2d6c096 | 1756 | n_bind_mounts + |
b3d13314 | 1757 | n_mount_images + |
a07b9926 | 1758 | (n_extension_images > 0 || n_extension_directories > 0 ? /* Mount each image and directory plus an overlay per hierarchy */ |
7c76e181 | 1759 | n_hierarchies + n_extension_images + n_extension_directories: 0) + |
2abd4e38 | 1760 | n_temporary_filesystems + |
c575770b | 1761 | ns_info->private_dev + |
788e7201 TM |
1762 | (ns_info->protect_kernel_tunables ? |
1763 | ELEMENTSOF(protect_kernel_tunables_proc_table) + ELEMENTSOF(protect_kernel_tunables_sys_table) : 0) + | |
c575770b | 1764 | (ns_info->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) + |
788e7201 TM |
1765 | (ns_info->protect_kernel_logs ? |
1766 | ELEMENTSOF(protect_kernel_logs_proc_table) + ELEMENTSOF(protect_kernel_logs_dev_table) : 0) + | |
94a7b275 | 1767 | (ns_info->protect_control_groups ? 1 : 0) + |
5d997827 | 1768 | protect_home_cnt + protect_system_cnt + |
aecd5ac6 | 1769 | (ns_info->protect_hostname ? 2 : 0) + |
91dd5f7c | 1770 | (namespace_info_mount_apivfs(ns_info) ? ELEMENTSOF(apivfs_table) : 0) + |
bbb4e7f3 | 1771 | (creds_path ? 2 : 1) + |
3bdc25a4 LP |
1772 | !!log_namespace + |
1773 | setup_propagate + /* /run/systemd/incoming */ | |
80271a44 | 1774 | !!notify_socket + |
3f37a825 | 1775 | !!host_os_release + |
c2da3bf2 | 1776 | ns_info->private_network + /* /sys */ |
80271a44 | 1777 | ns_info->private_ipc; /* /dev/mqueue */ |
2652c6c1 DH |
1778 | } |
1779 | ||
fbf90c0d CB |
1780 | /* Walk all mount entries and dropping any unused mounts. This affects all |
1781 | * mounts: | |
1782 | * - that are implicitly protected by a path that has been rendered inaccessible | |
1783 | * - whose immediate parent requests the same protection mode as the mount itself | |
1784 | * - that are outside of the relevant root directory | |
1785 | * - which are duplicates | |
1786 | */ | |
1787 | static void drop_unused_mounts(const char *root_directory, MountEntry *mounts, size_t *n_mounts) { | |
9b68367b | 1788 | assert(root_directory); |
f8b64b57 LP |
1789 | assert(n_mounts); |
1790 | assert(mounts || *n_mounts == 0); | |
1791 | ||
93bab288 | 1792 | typesafe_qsort(mounts, *n_mounts, mount_path_compare); |
f8b64b57 LP |
1793 | |
1794 | drop_duplicates(mounts, n_mounts); | |
1795 | drop_outside_root(root_directory, mounts, n_mounts); | |
1796 | drop_inaccessible(mounts, n_mounts); | |
1797 | drop_nop(mounts, n_mounts); | |
1798 | } | |
1799 | ||
df61e79a | 1800 | static int create_symlinks_from_tuples(const char *root, char **strv_symlinks) { |
df61e79a LB |
1801 | int r; |
1802 | ||
1803 | STRV_FOREACH_PAIR(src, dst, strv_symlinks) { | |
1804 | _cleanup_free_ char *src_abs = NULL, *dst_abs = NULL; | |
1805 | ||
1806 | src_abs = path_join(root, *src); | |
1807 | dst_abs = path_join(root, *dst); | |
1808 | if (!src_abs || !dst_abs) | |
1809 | return -ENOMEM; | |
1810 | ||
1811 | r = mkdir_parents_label(dst_abs, 0755); | |
1812 | if (r < 0) | |
1813 | return r; | |
1814 | ||
1815 | r = symlink_idempotent(src_abs, dst_abs, true); | |
1816 | if (r < 0) | |
1817 | return r; | |
1818 | } | |
1819 | ||
1820 | return 0; | |
1821 | } | |
1822 | ||
82fb2da2 LB |
1823 | static int apply_mounts( |
1824 | const char *root, | |
84be0c71 LP |
1825 | const ImagePolicy *mount_image_policy, |
1826 | const ImagePolicy *extension_image_policy, | |
82fb2da2 LB |
1827 | const NamespaceInfo *ns_info, |
1828 | MountEntry *mounts, | |
1829 | size_t *n_mounts, | |
663e2756 | 1830 | char **symlinks, |
82fb2da2 LB |
1831 | char **error_path) { |
1832 | ||
1833 | _cleanup_fclose_ FILE *proc_self_mountinfo = NULL; | |
1834 | _cleanup_free_ char **deny_list = NULL; | |
82fb2da2 LB |
1835 | int r; |
1836 | ||
1837 | if (n_mounts == 0) /* Shortcut: nothing to do */ | |
1838 | return 0; | |
1839 | ||
1840 | assert(root); | |
1841 | assert(mounts); | |
1842 | assert(n_mounts); | |
1843 | ||
1844 | /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of | |
1845 | * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */ | |
1846 | proc_self_mountinfo = fopen("/proc/self/mountinfo", "re"); | |
1847 | if (!proc_self_mountinfo) { | |
d60e3b40 LP |
1848 | r = -errno; |
1849 | ||
82fb2da2 LB |
1850 | if (error_path) |
1851 | *error_path = strdup("/proc/self/mountinfo"); | |
d60e3b40 LP |
1852 | |
1853 | return log_debug_errno(r, "Failed to open /proc/self/mountinfo: %m"); | |
82fb2da2 LB |
1854 | } |
1855 | ||
1856 | /* First round, establish all mounts we need */ | |
1857 | for (;;) { | |
1858 | bool again = false; | |
1859 | ||
1860 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1861 | ||
1862 | if (m->applied) | |
1863 | continue; | |
1864 | ||
a07b9926 LB |
1865 | /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */ |
1866 | r = follow_symlink(!IN_SET(m->mode, EXTENSION_IMAGES, EXTENSION_DIRECTORIES) ? root : NULL, m); | |
82fb2da2 LB |
1867 | if (r < 0) { |
1868 | if (error_path && mount_entry_path(m)) | |
1869 | *error_path = strdup(mount_entry_path(m)); | |
1870 | return r; | |
1871 | } | |
1872 | if (r == 0) { | |
1873 | /* We hit a symlinked mount point. The entry got rewritten and might | |
1874 | * point to a very different place now. Let's normalize the changed | |
1875 | * list, and start from the beginning. After all to mount the entry | |
1876 | * at the new location we might need some other mounts first */ | |
1877 | again = true; | |
1878 | break; | |
1879 | } | |
1880 | ||
84be0c71 | 1881 | r = apply_one_mount(root, m, mount_image_policy, extension_image_policy, ns_info); |
82fb2da2 LB |
1882 | if (r < 0) { |
1883 | if (error_path && mount_entry_path(m)) | |
1884 | *error_path = strdup(mount_entry_path(m)); | |
1885 | return r; | |
1886 | } | |
1887 | ||
1888 | m->applied = true; | |
1889 | } | |
1890 | ||
1891 | if (!again) | |
1892 | break; | |
1893 | ||
fbf90c0d | 1894 | drop_unused_mounts(root, mounts, n_mounts); |
82fb2da2 LB |
1895 | } |
1896 | ||
df61e79a | 1897 | /* Now that all filesystems have been set up, but before the |
663e2756 | 1898 | * read-only switches are flipped, create the exec dirs and other symlinks. |
df61e79a LB |
1899 | * Note that when /var/lib is not empty/tmpfs, these symlinks will already |
1900 | * exist, which means this will be a no-op. */ | |
663e2756 | 1901 | r = create_symlinks_from_tuples(root, symlinks); |
df61e79a | 1902 | if (r < 0) |
663e2756 | 1903 | return log_debug_errno(r, "Failed to set up symlinks inside mount namespace: %m"); |
df61e79a | 1904 | |
82fb2da2 LB |
1905 | /* Create a deny list we can pass to bind_mount_recursive() */ |
1906 | deny_list = new(char*, (*n_mounts)+1); | |
1907 | if (!deny_list) | |
1908 | return -ENOMEM; | |
8794e6db | 1909 | for (size_t j = 0; j < *n_mounts; j++) |
82fb2da2 | 1910 | deny_list[j] = (char*) mount_entry_path(mounts+j); |
8794e6db | 1911 | deny_list[*n_mounts] = NULL; |
82fb2da2 LB |
1912 | |
1913 | /* Second round, flip the ro bits if necessary. */ | |
1914 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1915 | r = make_read_only(m, deny_list, proc_self_mountinfo); | |
1916 | if (r < 0) { | |
1917 | if (error_path && mount_entry_path(m)) | |
1918 | *error_path = strdup(mount_entry_path(m)); | |
1919 | return r; | |
1920 | } | |
1921 | } | |
1922 | ||
1923 | /* Third round, flip the noexec bits with a simplified deny list. */ | |
8794e6db | 1924 | for (size_t j = 0; j < *n_mounts; j++) |
82fb2da2 LB |
1925 | if (IN_SET((mounts+j)->mode, EXEC, NOEXEC)) |
1926 | deny_list[j] = (char*) mount_entry_path(mounts+j); | |
8794e6db | 1927 | deny_list[*n_mounts] = NULL; |
82fb2da2 LB |
1928 | |
1929 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1930 | r = make_noexec(m, deny_list, proc_self_mountinfo); | |
1931 | if (r < 0) { | |
1932 | if (error_path && mount_entry_path(m)) | |
1933 | *error_path = strdup(mount_entry_path(m)); | |
1934 | return r; | |
1935 | } | |
1936 | } | |
1937 | ||
6720e356 YW |
1938 | /* Fourth round, flip the nosuid bits without a deny list. */ |
1939 | if (ns_info->mount_nosuid) | |
1940 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1941 | r = make_nosuid(m, proc_self_mountinfo); | |
1942 | if (r < 0) { | |
1943 | if (error_path && mount_entry_path(m)) | |
1944 | *error_path = strdup(mount_entry_path(m)); | |
1945 | return r; | |
1946 | } | |
1947 | } | |
1948 | ||
82fb2da2 LB |
1949 | return 1; |
1950 | } | |
1951 | ||
c8c535d5 LP |
1952 | static bool root_read_only( |
1953 | char **read_only_paths, | |
1954 | ProtectSystem protect_system) { | |
1955 | ||
1956 | /* Determine whether the root directory is going to be read-only given the configured settings. */ | |
1957 | ||
1958 | if (protect_system == PROTECT_SYSTEM_STRICT) | |
1959 | return true; | |
1960 | ||
de46b2be | 1961 | if (prefixed_path_strv_contains(read_only_paths, "/")) |
c8c535d5 LP |
1962 | return true; |
1963 | ||
1964 | return false; | |
1965 | } | |
1966 | ||
1967 | static bool home_read_only( | |
1968 | char** read_only_paths, | |
1969 | char** inaccessible_paths, | |
1970 | char** empty_directories, | |
1971 | const BindMount *bind_mounts, | |
1972 | size_t n_bind_mounts, | |
1973 | const TemporaryFileSystem *temporary_filesystems, | |
1974 | size_t n_temporary_filesystems, | |
1975 | ProtectHome protect_home) { | |
1976 | ||
c8c535d5 LP |
1977 | /* Determine whether the /home directory is going to be read-only given the configured settings. Yes, |
1978 | * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple | |
1979 | * settings. */ | |
1980 | ||
1981 | if (protect_home != PROTECT_HOME_NO) | |
1982 | return true; | |
1983 | ||
de46b2be TM |
1984 | if (prefixed_path_strv_contains(read_only_paths, "/home") || |
1985 | prefixed_path_strv_contains(inaccessible_paths, "/home") || | |
1986 | prefixed_path_strv_contains(empty_directories, "/home")) | |
c8c535d5 LP |
1987 | return true; |
1988 | ||
8794e6db | 1989 | for (size_t i = 0; i < n_temporary_filesystems; i++) |
c8c535d5 LP |
1990 | if (path_equal(temporary_filesystems[i].path, "/home")) |
1991 | return true; | |
1992 | ||
1993 | /* If /home is overmounted with some dir from the host it's not writable. */ | |
8794e6db | 1994 | for (size_t i = 0; i < n_bind_mounts; i++) |
c8c535d5 LP |
1995 | if (path_equal(bind_mounts[i].destination, "/home")) |
1996 | return true; | |
1997 | ||
1998 | return false; | |
1999 | } | |
2000 | ||
613b411c | 2001 | int setup_namespace( |
ee818b89 | 2002 | const char* root_directory, |
915e6d16 | 2003 | const char* root_image, |
84be0c71 LP |
2004 | const MountOptions *root_image_mount_options, |
2005 | const ImagePolicy *root_image_policy, | |
bb0ff3fb | 2006 | const NamespaceInfo *ns_info, |
2a624c36 AP |
2007 | char** read_write_paths, |
2008 | char** read_only_paths, | |
2009 | char** inaccessible_paths, | |
ddc155b2 TM |
2010 | char** exec_paths, |
2011 | char** no_exec_paths, | |
6c47cd7d | 2012 | char** empty_directories, |
663e2756 | 2013 | char** symlinks, |
d2d6c096 | 2014 | const BindMount *bind_mounts, |
da6053d0 | 2015 | size_t n_bind_mounts, |
2abd4e38 | 2016 | const TemporaryFileSystem *temporary_filesystems, |
da6053d0 | 2017 | size_t n_temporary_filesystems, |
b3d13314 LB |
2018 | const MountImage *mount_images, |
2019 | size_t n_mount_images, | |
84be0c71 | 2020 | const ImagePolicy *mount_image_policy, |
a004cb4c LP |
2021 | const char* tmp_dir, |
2022 | const char* var_tmp_dir, | |
bbb4e7f3 | 2023 | const char *creds_path, |
94fe4cf2 | 2024 | int creds_fd, |
91dd5f7c | 2025 | const char *log_namespace, |
874cdcbc | 2026 | unsigned long mount_propagation_flag, |
66130f0a | 2027 | VeritySettings *verity, |
93f59701 LB |
2028 | const MountImage *extension_images, |
2029 | size_t n_extension_images, | |
84be0c71 | 2030 | const ImagePolicy *extension_image_policy, |
a07b9926 | 2031 | char **extension_directories, |
5e8deb94 LB |
2032 | const char *propagate_dir, |
2033 | const char *incoming_dir, | |
24759d8f | 2034 | const char *extension_dir, |
3bdc25a4 | 2035 | const char *notify_socket, |
663e2756 | 2036 | const char *host_os_release_stage, |
7cc5ef5f | 2037 | char **error_path) { |
15ae422b | 2038 | |
915e6d16 LP |
2039 | _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL; |
2040 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; | |
93f59701 | 2041 | _cleanup_strv_free_ char **hierarchies = NULL; |
5f7a690a | 2042 | MountEntry *m = NULL, *mounts = NULL; |
5e8deb94 | 2043 | bool require_prefix = false, setup_propagate = false; |
24759d8f | 2044 | const char *root; |
9cb1709b LP |
2045 | DissectImageFlags dissect_image_flags = |
2046 | DISSECT_IMAGE_GENERIC_ROOT | | |
2047 | DISSECT_IMAGE_REQUIRE_ROOT | | |
2048 | DISSECT_IMAGE_DISCARD_ON_LOOP | | |
2049 | DISSECT_IMAGE_RELAX_VAR_CHECK | | |
2050 | DISSECT_IMAGE_FSCK | | |
c65f854a | 2051 | DISSECT_IMAGE_USR_NO_ROOT | |
73d88b80 LP |
2052 | DISSECT_IMAGE_GROWFS | |
2053 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | | |
2054 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
89e62e0b LP |
2055 | size_t n_mounts; |
2056 | int r; | |
15ae422b | 2057 | |
915e6d16 LP |
2058 | assert(ns_info); |
2059 | ||
cdf42f9b LP |
2060 | /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes |
2061 | * we configure take effect */ | |
2062 | BLOCK_WITH_UMASK(0000); | |
2063 | ||
5e8deb94 LB |
2064 | if (!isempty(propagate_dir) && !isempty(incoming_dir)) |
2065 | setup_propagate = true; | |
2066 | ||
874cdcbc LP |
2067 | if (mount_propagation_flag == 0) |
2068 | mount_propagation_flag = MS_SHARED; | |
ac0930c8 | 2069 | |
915e6d16 | 2070 | if (root_image) { |
c8c535d5 LP |
2071 | /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */ |
2072 | if (root_read_only(read_only_paths, | |
52b3d652 | 2073 | ns_info->protect_system) && |
c8c535d5 LP |
2074 | home_read_only(read_only_paths, inaccessible_paths, empty_directories, |
2075 | bind_mounts, n_bind_mounts, temporary_filesystems, n_temporary_filesystems, | |
52b3d652 | 2076 | ns_info->protect_home) && |
c9ef8573 | 2077 | strv_isempty(read_write_paths)) |
915e6d16 LP |
2078 | dissect_image_flags |= DISSECT_IMAGE_READ_ONLY; |
2079 | ||
66130f0a | 2080 | SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, verity && verity->data_path); |
915e6d16 | 2081 | |
89e62e0b LP |
2082 | r = loop_device_make_by_path( |
2083 | root_image, | |
ef9c184d | 2084 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */, |
22ee78a8 | 2085 | /* sector_size= */ UINT32_MAX, |
89e62e0b | 2086 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 2087 | LOCK_SH, |
89e62e0b | 2088 | &loop_device); |
78ebe980 | 2089 | if (r < 0) |
89e62e0b LP |
2090 | return log_debug_errno(r, "Failed to create loop device for root image: %m"); |
2091 | ||
bad31660 YW |
2092 | r = dissect_loop_device( |
2093 | loop_device, | |
66130f0a | 2094 | verity, |
84be0c71 LP |
2095 | root_image_mount_options, |
2096 | root_image_policy, | |
89e62e0b LP |
2097 | dissect_image_flags, |
2098 | &dissected_image); | |
78ebe980 | 2099 | if (r < 0) |
763a260a | 2100 | return log_debug_errno(r, "Failed to dissect image: %m"); |
78ebe980 | 2101 | |
88b3300f LP |
2102 | r = dissected_image_load_verity_sig_partition( |
2103 | dissected_image, | |
2104 | loop_device->fd, | |
66130f0a | 2105 | verity); |
88b3300f LP |
2106 | if (r < 0) |
2107 | return r; | |
2108 | ||
89e62e0b LP |
2109 | r = dissected_image_decrypt( |
2110 | dissected_image, | |
2111 | NULL, | |
66130f0a | 2112 | verity, |
e330f97a | 2113 | dissect_image_flags); |
915e6d16 | 2114 | if (r < 0) |
763a260a | 2115 | return log_debug_errno(r, "Failed to decrypt dissected image: %m"); |
915e6d16 LP |
2116 | } |
2117 | ||
e908468b LP |
2118 | if (root_directory) |
2119 | root = root_directory; | |
0722b359 | 2120 | else { |
77f16dbd DDM |
2121 | /* /run/systemd should have been created by PID 1 early on already, but in some cases, like |
2122 | * when running tests (test-execute), it might not have been created yet so let's make sure | |
2123 | * we create it if it doesn't already exist. */ | |
2124 | (void) mkdir_p_label("/run/systemd", 0755); | |
2125 | ||
d73020f2 LP |
2126 | /* Always create the mount namespace in a temporary directory, instead of operating directly |
2127 | * in the root. The temporary directory prevents any mounts from being potentially obscured | |
2128 | * my other mounts we already applied. We use the same mount point for all images, which is | |
2129 | * safe, since they all live in their own namespaces after all, and hence won't see each | |
be79c235 LP |
2130 | * other. (Note: this directory is also created by PID 1 early on, we create it here for |
2131 | * similar reasons as /run/systemd/ first.) */ | |
2132 | root = "/run/systemd/mount-rootfs"; | |
2133 | (void) mkdir_label(root, 0555); | |
e908468b | 2134 | |
d18aff04 | 2135 | require_prefix = true; |
0722b359 | 2136 | } |
e908468b | 2137 | |
a07b9926 | 2138 | if (n_extension_images > 0 || !strv_isempty(extension_directories)) { |
55ea4ef0 MG |
2139 | /* Hierarchy population needs to be done for sysext and confext extension images */ |
2140 | r = parse_env_extension_hierarchies(&hierarchies, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES"); | |
93f59701 LB |
2141 | if (r < 0) |
2142 | return r; | |
2143 | } | |
2144 | ||
cfbeb4ef LP |
2145 | n_mounts = namespace_calculate_mounts( |
2146 | ns_info, | |
2147 | read_write_paths, | |
2148 | read_only_paths, | |
2149 | inaccessible_paths, | |
ddc155b2 TM |
2150 | exec_paths, |
2151 | no_exec_paths, | |
6c47cd7d | 2152 | empty_directories, |
f5c52a77 | 2153 | n_bind_mounts, |
2abd4e38 | 2154 | n_temporary_filesystems, |
b3d13314 | 2155 | n_mount_images, |
93f59701 | 2156 | n_extension_images, |
a07b9926 | 2157 | strv_length(extension_directories), |
93f59701 | 2158 | strv_length(hierarchies), |
cfbeb4ef | 2159 | tmp_dir, var_tmp_dir, |
bbb4e7f3 | 2160 | creds_path, |
5e8deb94 | 2161 | log_namespace, |
3bdc25a4 | 2162 | setup_propagate, |
3f37a825 | 2163 | notify_socket, |
663e2756 | 2164 | host_os_release_stage); |
613b411c | 2165 | |
f0a4feb0 | 2166 | if (n_mounts > 0) { |
5f7a690a LP |
2167 | m = mounts = new0(MountEntry, n_mounts); |
2168 | if (!mounts) | |
2169 | return -ENOMEM; | |
2170 | ||
d18aff04 | 2171 | r = append_access_mounts(&m, read_write_paths, READWRITE, require_prefix); |
613b411c | 2172 | if (r < 0) |
f0a4feb0 | 2173 | goto finish; |
613b411c | 2174 | |
d18aff04 | 2175 | r = append_access_mounts(&m, read_only_paths, READONLY, require_prefix); |
613b411c | 2176 | if (r < 0) |
f0a4feb0 | 2177 | goto finish; |
613b411c | 2178 | |
d18aff04 | 2179 | r = append_access_mounts(&m, inaccessible_paths, INACCESSIBLE, require_prefix); |
613b411c | 2180 | if (r < 0) |
f0a4feb0 | 2181 | goto finish; |
7ff7394d | 2182 | |
ddc155b2 TM |
2183 | r = append_access_mounts(&m, exec_paths, EXEC, require_prefix); |
2184 | if (r < 0) | |
2185 | goto finish; | |
2186 | ||
2187 | r = append_access_mounts(&m, no_exec_paths, NOEXEC, require_prefix); | |
2188 | if (r < 0) | |
2189 | goto finish; | |
2190 | ||
6c47cd7d LP |
2191 | r = append_empty_dir_mounts(&m, empty_directories); |
2192 | if (r < 0) | |
2193 | goto finish; | |
2194 | ||
d2d6c096 LP |
2195 | r = append_bind_mounts(&m, bind_mounts, n_bind_mounts); |
2196 | if (r < 0) | |
2197 | goto finish; | |
2198 | ||
2abd4e38 YW |
2199 | r = append_tmpfs_mounts(&m, temporary_filesystems, n_temporary_filesystems); |
2200 | if (r < 0) | |
2201 | goto finish; | |
2202 | ||
613b411c | 2203 | if (tmp_dir) { |
56a13a49 ZJS |
2204 | bool ro = streq(tmp_dir, RUN_SYSTEMD_EMPTY); |
2205 | ||
34de407a | 2206 | *(m++) = (MountEntry) { |
5327c910 | 2207 | .path_const = "/tmp", |
56a13a49 | 2208 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, |
89bd586c | 2209 | .source_const = tmp_dir, |
5327c910 | 2210 | }; |
613b411c | 2211 | } |
7ff7394d | 2212 | |
613b411c | 2213 | if (var_tmp_dir) { |
56a13a49 ZJS |
2214 | bool ro = streq(var_tmp_dir, RUN_SYSTEMD_EMPTY); |
2215 | ||
34de407a | 2216 | *(m++) = (MountEntry) { |
5327c910 | 2217 | .path_const = "/var/tmp", |
56a13a49 | 2218 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, |
89bd586c | 2219 | .source_const = var_tmp_dir, |
5327c910 | 2220 | }; |
7ff7394d | 2221 | } |
ac0930c8 | 2222 | |
b3d13314 LB |
2223 | r = append_mount_images(&m, mount_images, n_mount_images); |
2224 | if (r < 0) | |
2225 | goto finish; | |
2226 | ||
a07b9926 | 2227 | r = append_extensions(&m, root, extension_dir, hierarchies, extension_images, n_extension_images, extension_directories); |
93f59701 LB |
2228 | if (r < 0) |
2229 | goto finish; | |
2230 | ||
d46b79bb | 2231 | if (ns_info->private_dev) |
34de407a | 2232 | *(m++) = (MountEntry) { |
5327c910 LP |
2233 | .path_const = "/dev", |
2234 | .mode = PRIVATE_DEV, | |
9ce4e4b0 | 2235 | .flags = DEV_MOUNT_OPTIONS, |
5327c910 | 2236 | }; |
7f112f50 | 2237 | |
788e7201 TM |
2238 | /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the |
2239 | protective mounts to non-pid /proc paths would fail. But the pid only option may have | |
2240 | failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */ | |
2241 | bool ignore_protect_proc = ns_info->ignore_protect_paths || ns_info->proc_subset == PROC_SUBSET_PID; | |
c575770b | 2242 | if (ns_info->protect_kernel_tunables) { |
cbc056c8 | 2243 | r = append_static_mounts(&m, |
788e7201 TM |
2244 | protect_kernel_tunables_proc_table, |
2245 | ELEMENTSOF(protect_kernel_tunables_proc_table), | |
2246 | ignore_protect_proc); | |
2247 | if (r < 0) | |
2248 | goto finish; | |
2249 | ||
2250 | r = append_static_mounts(&m, | |
2251 | protect_kernel_tunables_sys_table, | |
2252 | ELEMENTSOF(protect_kernel_tunables_sys_table), | |
cbc056c8 | 2253 | ns_info->ignore_protect_paths); |
c575770b | 2254 | if (r < 0) |
f0a4feb0 | 2255 | goto finish; |
c575770b DH |
2256 | } |
2257 | ||
2258 | if (ns_info->protect_kernel_modules) { | |
cbc056c8 ZJS |
2259 | r = append_static_mounts(&m, |
2260 | protect_kernel_modules_table, | |
2261 | ELEMENTSOF(protect_kernel_modules_table), | |
2262 | ns_info->ignore_protect_paths); | |
c575770b | 2263 | if (r < 0) |
f0a4feb0 | 2264 | goto finish; |
c575770b | 2265 | } |
59eeb84b | 2266 | |
94a7b275 | 2267 | if (ns_info->protect_kernel_logs) { |
cbc056c8 | 2268 | r = append_static_mounts(&m, |
788e7201 TM |
2269 | protect_kernel_logs_proc_table, |
2270 | ELEMENTSOF(protect_kernel_logs_proc_table), | |
2271 | ignore_protect_proc); | |
2272 | if (r < 0) | |
2273 | goto finish; | |
2274 | ||
2275 | r = append_static_mounts(&m, | |
2276 | protect_kernel_logs_dev_table, | |
2277 | ELEMENTSOF(protect_kernel_logs_dev_table), | |
cbc056c8 | 2278 | ns_info->ignore_protect_paths); |
94a7b275 KK |
2279 | if (r < 0) |
2280 | goto finish; | |
2281 | } | |
2282 | ||
d46b79bb | 2283 | if (ns_info->protect_control_groups) |
34de407a | 2284 | *(m++) = (MountEntry) { |
5327c910 LP |
2285 | .path_const = "/sys/fs/cgroup", |
2286 | .mode = READONLY, | |
2287 | }; | |
59eeb84b | 2288 | |
52b3d652 | 2289 | r = append_protect_home(&m, ns_info->protect_home, ns_info->ignore_protect_paths); |
b6c432ca | 2290 | if (r < 0) |
f0a4feb0 | 2291 | goto finish; |
417116f2 | 2292 | |
52b3d652 | 2293 | r = append_protect_system(&m, ns_info->protect_system, false); |
f471b2af | 2294 | if (r < 0) |
f0a4feb0 | 2295 | goto finish; |
417116f2 | 2296 | |
9b68367b | 2297 | if (namespace_info_mount_apivfs(ns_info)) { |
cbc056c8 ZJS |
2298 | r = append_static_mounts(&m, |
2299 | apivfs_table, | |
2300 | ELEMENTSOF(apivfs_table), | |
2301 | ns_info->ignore_protect_paths); | |
5d997827 LP |
2302 | if (r < 0) |
2303 | goto finish; | |
2304 | } | |
2305 | ||
1361f015 CB |
2306 | /* Note, if proc is mounted with subset=pid then neither of the |
2307 | * two paths will exist, i.e. they are implicitly protected by | |
2308 | * the mount option. */ | |
aecd5ac6 TM |
2309 | if (ns_info->protect_hostname) { |
2310 | *(m++) = (MountEntry) { | |
2311 | .path_const = "/proc/sys/kernel/hostname", | |
2312 | .mode = READONLY, | |
1361f015 | 2313 | .ignore = ignore_protect_proc, |
aecd5ac6 TM |
2314 | }; |
2315 | *(m++) = (MountEntry) { | |
2316 | .path_const = "/proc/sys/kernel/domainname", | |
2317 | .mode = READONLY, | |
1361f015 | 2318 | .ignore = ignore_protect_proc, |
aecd5ac6 TM |
2319 | }; |
2320 | } | |
2321 | ||
c2da3bf2 YW |
2322 | if (ns_info->private_network) |
2323 | *(m++) = (MountEntry) { | |
2324 | .path_const = "/sys", | |
2325 | .mode = PRIVATE_SYSFS, | |
2326 | }; | |
2327 | ||
57ac6959 | 2328 | if (ns_info->private_ipc) |
80271a44 XR |
2329 | *(m++) = (MountEntry) { |
2330 | .path_const = "/dev/mqueue", | |
2331 | .mode = MQUEUEFS, | |
2332 | .flags = MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, | |
2333 | }; | |
80271a44 | 2334 | |
bbb4e7f3 LP |
2335 | if (creds_path) { |
2336 | /* If our service has a credentials store configured, then bind that one in, but hide | |
2337 | * everything else. */ | |
2338 | ||
2339 | *(m++) = (MountEntry) { | |
2340 | .path_const = "/run/credentials", | |
2341 | .mode = TMPFS, | |
2342 | .read_only = true, | |
2343 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, | |
2344 | .flags = MS_NODEV|MS_STRICTATIME|MS_NOSUID|MS_NOEXEC, | |
2345 | }; | |
2346 | ||
94fe4cf2 YW |
2347 | /* If we have mount fd for credentials directory, then it will be mounted after |
2348 | * namespace is set up. So, here we only create the mount point. */ | |
2349 | ||
2350 | if (creds_fd < 0) | |
2351 | *(m++) = (MountEntry) { | |
2352 | .path_const = creds_path, | |
2353 | .mode = BIND_MOUNT, | |
2354 | .read_only = true, | |
2355 | .source_const = creds_path, | |
2356 | .ignore = true, | |
2357 | }; | |
2358 | else | |
2359 | *(m++) = (MountEntry) { | |
2360 | .path_const = creds_path, | |
2361 | .mode = MKDIR, | |
2362 | }; | |
bbb4e7f3 LP |
2363 | } else { |
2364 | /* If our service has no credentials store configured, then make the whole | |
2365 | * credentials tree inaccessible wholesale. */ | |
2366 | ||
2367 | *(m++) = (MountEntry) { | |
2368 | .path_const = "/run/credentials", | |
2369 | .mode = INACCESSIBLE, | |
2370 | .ignore = true, | |
2371 | }; | |
2372 | } | |
2373 | ||
91dd5f7c | 2374 | if (log_namespace) { |
c2b2df60 | 2375 | _cleanup_free_ char *q = NULL; |
91dd5f7c LP |
2376 | |
2377 | q = strjoin("/run/systemd/journal.", log_namespace); | |
2378 | if (!q) { | |
2379 | r = -ENOMEM; | |
2380 | goto finish; | |
2381 | } | |
2382 | ||
2383 | *(m++) = (MountEntry) { | |
2384 | .path_const = "/run/systemd/journal", | |
2385 | .mode = BIND_MOUNT_RECURSIVE, | |
2386 | .read_only = true, | |
2387 | .source_malloc = TAKE_PTR(q), | |
2388 | }; | |
2389 | } | |
2390 | ||
5e8deb94 LB |
2391 | /* Will be used to add bind mounts at runtime */ |
2392 | if (setup_propagate) | |
2393 | *(m++) = (MountEntry) { | |
2394 | .source_const = propagate_dir, | |
2395 | .path_const = incoming_dir, | |
2396 | .mode = BIND_MOUNT, | |
2397 | .read_only = true, | |
2398 | }; | |
2399 | ||
3bdc25a4 LP |
2400 | if (notify_socket) |
2401 | *(m++) = (MountEntry) { | |
2402 | .path_const = notify_socket, | |
2403 | .source_const = notify_socket, | |
2404 | .mode = BIND_MOUNT, | |
2405 | .read_only = true, | |
2406 | }; | |
2407 | ||
663e2756 | 2408 | if (host_os_release_stage) |
3f37a825 | 2409 | *(m++) = (MountEntry) { |
663e2756 LB |
2410 | .path_const = "/run/host/.os-release-stage/", |
2411 | .source_const = host_os_release_stage, | |
3f37a825 LB |
2412 | .mode = BIND_MOUNT, |
2413 | .read_only = true, | |
2414 | .ignore = true, /* Live copy, don't hard-fail if it goes missing */ | |
2415 | }; | |
2416 | ||
f0a4feb0 | 2417 | assert(mounts + n_mounts == m); |
ac0930c8 | 2418 | |
5327c910 | 2419 | /* Prepend the root directory where that's necessary */ |
e908468b | 2420 | r = prefix_where_needed(mounts, n_mounts, root); |
5327c910 LP |
2421 | if (r < 0) |
2422 | goto finish; | |
2423 | ||
fbf90c0d | 2424 | drop_unused_mounts(root, mounts, &n_mounts); |
15ae422b LP |
2425 | } |
2426 | ||
1beab8b0 LP |
2427 | /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */ |
2428 | ||
d944dc95 | 2429 | if (unshare(CLONE_NEWNS) < 0) { |
763a260a | 2430 | r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m"); |
eacfbd89 LP |
2431 | if (ERRNO_IS_PRIVILEGE(r) || |
2432 | ERRNO_IS_NOT_SUPPORTED(r)) | |
cbc056c8 ZJS |
2433 | /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter |
2434 | * in place that doesn't allow us to create namespaces (or a missing cap), then | |
2435 | * propagate a recognizable error back, which the caller can use to detect this case | |
2436 | * (and only this) and optionally continue without namespacing applied. */ | |
1beab8b0 LP |
2437 | r = -ENOANO; |
2438 | ||
d944dc95 LP |
2439 | goto finish; |
2440 | } | |
1e4e94c8 | 2441 | |
5e8deb94 LB |
2442 | /* Create the source directory to allow runtime propagation of mounts */ |
2443 | if (setup_propagate) | |
2444 | (void) mkdir_p(propagate_dir, 0600); | |
2445 | ||
3ebf0b0b LB |
2446 | if (n_extension_images > 0 || !strv_isempty(extension_directories)) |
2447 | /* ExtensionImages/Directories mountpoint directories will be created while parsing the | |
2448 | * mounts to create, so have the parent ready */ | |
93f59701 | 2449 | (void) mkdir_p(extension_dir, 0600); |
93f59701 | 2450 | |
9b68367b YW |
2451 | /* Remount / as SLAVE so that nothing now mounted in the namespace |
2452 | * shows up in the parent */ | |
2453 | if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) { | |
763a260a | 2454 | r = log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m"); |
9b68367b | 2455 | goto finish; |
ee818b89 AC |
2456 | } |
2457 | ||
915e6d16 | 2458 | if (root_image) { |
e908468b | 2459 | /* A root image is specified, mount it to the right place */ |
21b61b1d | 2460 | r = dissected_image_mount(dissected_image, root, UID_INVALID, UID_INVALID, dissect_image_flags); |
763a260a YW |
2461 | if (r < 0) { |
2462 | log_debug_errno(r, "Failed to mount root image: %m"); | |
41bc4849 LP |
2463 | goto finish; |
2464 | } | |
2465 | ||
2466 | /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device | |
2467 | * if it likes. */ | |
2468 | r = loop_device_flock(loop_device, LOCK_UN); | |
2469 | if (r < 0) { | |
2470 | log_debug_errno(r, "Failed to release lock on loopback block device: %m"); | |
915e6d16 | 2471 | goto finish; |
763a260a | 2472 | } |
915e6d16 | 2473 | |
3044d343 YW |
2474 | r = dissected_image_relinquish(dissected_image); |
2475 | if (r < 0) { | |
2476 | log_debug_errno(r, "Failed to relinquish dissected image: %m"); | |
2477 | goto finish; | |
07ce7407 | 2478 | } |
78ebe980 | 2479 | |
915e6d16 LP |
2480 | } else if (root_directory) { |
2481 | ||
e908468b LP |
2482 | /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */ |
2483 | r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW); | |
763a260a YW |
2484 | if (r < 0) { |
2485 | log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root); | |
d944dc95 | 2486 | goto finish; |
763a260a | 2487 | } |
8f1ad200 | 2488 | if (r == 0) { |
21935150 LP |
2489 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); |
2490 | if (r < 0) | |
8f1ad200 | 2491 | goto finish; |
d944dc95 | 2492 | } |
e908468b | 2493 | |
9b68367b | 2494 | } else { |
e908468b | 2495 | /* Let's mount the main root directory to the root directory to use */ |
21935150 LP |
2496 | r = mount_nofollow_verbose(LOG_DEBUG, "/", root, NULL, MS_BIND|MS_REC, NULL); |
2497 | if (r < 0) | |
e908468b | 2498 | goto finish; |
ee818b89 | 2499 | } |
c2c13f2d | 2500 | |
4e0c20de LP |
2501 | /* Try to set up the new root directory before mounting anything else there. */ |
2502 | if (root_image || root_directory) | |
2503 | (void) base_filesystem_create(root, UID_INVALID, GID_INVALID); | |
2504 | ||
82fb2da2 | 2505 | /* Now make the magic happen */ |
663e2756 | 2506 | r = apply_mounts(root, mount_image_policy, extension_image_policy, ns_info, mounts, &n_mounts, symlinks, error_path); |
82fb2da2 LB |
2507 | if (r < 0) |
2508 | goto finish; | |
15ae422b | 2509 | |
9b68367b | 2510 | /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */ |
9d50f850 | 2511 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
ea63a260 LB |
2512 | if (r == -EINVAL && root_directory) { |
2513 | /* If we are using root_directory and we don't have privileges (ie: user manager in a user | |
2514 | * namespace) and the root_directory is already a mount point in the parent namespace, | |
2515 | * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than | |
2516 | * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a | |
2517 | * mount point) and try again. */ | |
2518 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); | |
2519 | if (r < 0) | |
2520 | goto finish; | |
9d50f850 | 2521 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
ea63a260 | 2522 | } |
763a260a YW |
2523 | if (r < 0) { |
2524 | log_debug_errno(r, "Failed to mount root with MS_MOVE: %m"); | |
9b68367b | 2525 | goto finish; |
763a260a | 2526 | } |
ee818b89 | 2527 | |
874cdcbc LP |
2528 | /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to |
2529 | * the host, since what's disconnected is disconnected. */ | |
2530 | if (mount(NULL, "/", NULL, mount_propagation_flag | MS_REC, NULL) < 0) { | |
763a260a | 2531 | r = log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m"); |
d944dc95 LP |
2532 | goto finish; |
2533 | } | |
15ae422b | 2534 | |
5e8deb94 LB |
2535 | /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only |
2536 | * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */ | |
2537 | if (setup_propagate) { | |
2538 | r = mount(NULL, incoming_dir, NULL, MS_SLAVE, NULL); | |
2539 | if (r < 0) { | |
2540 | log_error_errno(r, "Failed to remount %s with MS_SLAVE: %m", incoming_dir); | |
2541 | goto finish; | |
2542 | } | |
2543 | } | |
2544 | ||
d944dc95 | 2545 | r = 0; |
15ae422b | 2546 | |
d944dc95 | 2547 | finish: |
0cd41757 LP |
2548 | if (n_mounts > 0) |
2549 | for (m = mounts; m < mounts + n_mounts; m++) | |
2550 | mount_entry_done(m); | |
613b411c | 2551 | |
5f7a690a LP |
2552 | free(mounts); |
2553 | ||
613b411c LP |
2554 | return r; |
2555 | } | |
2556 | ||
da6053d0 | 2557 | void bind_mount_free_many(BindMount *b, size_t n) { |
d2d6c096 LP |
2558 | assert(b || n == 0); |
2559 | ||
fe96c0f8 | 2560 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
2561 | free(b[i].source); |
2562 | free(b[i].destination); | |
2563 | } | |
2564 | ||
2565 | free(b); | |
2566 | } | |
2567 | ||
da6053d0 | 2568 | int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) { |
d2d6c096 LP |
2569 | _cleanup_free_ char *s = NULL, *d = NULL; |
2570 | BindMount *c; | |
2571 | ||
2572 | assert(b); | |
2573 | assert(n); | |
2574 | assert(item); | |
2575 | ||
2576 | s = strdup(item->source); | |
2577 | if (!s) | |
2578 | return -ENOMEM; | |
2579 | ||
2580 | d = strdup(item->destination); | |
2581 | if (!d) | |
2582 | return -ENOMEM; | |
2583 | ||
aa484f35 | 2584 | c = reallocarray(*b, *n + 1, sizeof(BindMount)); |
d2d6c096 LP |
2585 | if (!c) |
2586 | return -ENOMEM; | |
2587 | ||
2588 | *b = c; | |
2589 | ||
2590 | c[(*n) ++] = (BindMount) { | |
1cc6c93a YW |
2591 | .source = TAKE_PTR(s), |
2592 | .destination = TAKE_PTR(d), | |
d2d6c096 | 2593 | .read_only = item->read_only, |
9ce4e4b0 | 2594 | .nosuid = item->nosuid, |
d2d6c096 LP |
2595 | .recursive = item->recursive, |
2596 | .ignore_enoent = item->ignore_enoent, | |
2597 | }; | |
2598 | ||
d2d6c096 LP |
2599 | return 0; |
2600 | } | |
2601 | ||
b3d13314 | 2602 | MountImage* mount_image_free_many(MountImage *m, size_t *n) { |
b3d13314 LB |
2603 | assert(n); |
2604 | assert(m || *n == 0); | |
2605 | ||
fe96c0f8 | 2606 | for (size_t i = 0; i < *n; i++) { |
b3d13314 LB |
2607 | free(m[i].source); |
2608 | free(m[i].destination); | |
427353f6 | 2609 | mount_options_free_all(m[i].mount_options); |
b3d13314 LB |
2610 | } |
2611 | ||
2612 | free(m); | |
2613 | *n = 0; | |
2614 | return NULL; | |
2615 | } | |
2616 | ||
2617 | int mount_image_add(MountImage **m, size_t *n, const MountImage *item) { | |
2618 | _cleanup_free_ char *s = NULL, *d = NULL; | |
427353f6 | 2619 | _cleanup_(mount_options_free_allp) MountOptions *options = NULL; |
b3d13314 LB |
2620 | MountImage *c; |
2621 | ||
2622 | assert(m); | |
2623 | assert(n); | |
2624 | assert(item); | |
2625 | ||
2626 | s = strdup(item->source); | |
2627 | if (!s) | |
2628 | return -ENOMEM; | |
2629 | ||
93f59701 LB |
2630 | if (item->destination) { |
2631 | d = strdup(item->destination); | |
2632 | if (!d) | |
2633 | return -ENOMEM; | |
2634 | } | |
b3d13314 | 2635 | |
427353f6 | 2636 | LIST_FOREACH(mount_options, i, item->mount_options) { |
c2b2df60 | 2637 | _cleanup_(mount_options_free_allp) MountOptions *o = NULL; |
427353f6 LB |
2638 | |
2639 | o = new(MountOptions, 1); | |
2640 | if (!o) | |
2641 | return -ENOMEM; | |
2642 | ||
2643 | *o = (MountOptions) { | |
2644 | .partition_designator = i->partition_designator, | |
2645 | .options = strdup(i->options), | |
2646 | }; | |
2647 | if (!o->options) | |
2648 | return -ENOMEM; | |
2649 | ||
2650 | LIST_APPEND(mount_options, options, TAKE_PTR(o)); | |
2651 | } | |
2652 | ||
b3d13314 LB |
2653 | c = reallocarray(*m, *n + 1, sizeof(MountImage)); |
2654 | if (!c) | |
2655 | return -ENOMEM; | |
2656 | ||
2657 | *m = c; | |
2658 | ||
2659 | c[(*n) ++] = (MountImage) { | |
2660 | .source = TAKE_PTR(s), | |
2661 | .destination = TAKE_PTR(d), | |
427353f6 | 2662 | .mount_options = TAKE_PTR(options), |
b3d13314 | 2663 | .ignore_enoent = item->ignore_enoent, |
93f59701 | 2664 | .type = item->type, |
b3d13314 LB |
2665 | }; |
2666 | ||
2667 | return 0; | |
2668 | } | |
2669 | ||
da6053d0 | 2670 | void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) { |
2abd4e38 YW |
2671 | assert(t || n == 0); |
2672 | ||
fe96c0f8 | 2673 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
2674 | free(t[i].path); |
2675 | free(t[i].options); | |
2676 | } | |
2677 | ||
2678 | free(t); | |
2679 | } | |
2680 | ||
2681 | int temporary_filesystem_add( | |
2682 | TemporaryFileSystem **t, | |
da6053d0 | 2683 | size_t *n, |
2abd4e38 YW |
2684 | const char *path, |
2685 | const char *options) { | |
2686 | ||
2687 | _cleanup_free_ char *p = NULL, *o = NULL; | |
2688 | TemporaryFileSystem *c; | |
2689 | ||
2690 | assert(t); | |
2691 | assert(n); | |
2692 | assert(path); | |
2693 | ||
2694 | p = strdup(path); | |
2695 | if (!p) | |
2696 | return -ENOMEM; | |
2697 | ||
2698 | if (!isempty(options)) { | |
2699 | o = strdup(options); | |
2700 | if (!o) | |
2701 | return -ENOMEM; | |
2702 | } | |
2703 | ||
aa484f35 | 2704 | c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem)); |
2abd4e38 YW |
2705 | if (!c) |
2706 | return -ENOMEM; | |
2707 | ||
2708 | *t = c; | |
2709 | ||
2710 | c[(*n) ++] = (TemporaryFileSystem) { | |
1cc6c93a YW |
2711 | .path = TAKE_PTR(p), |
2712 | .options = TAKE_PTR(o), | |
2abd4e38 YW |
2713 | }; |
2714 | ||
2abd4e38 YW |
2715 | return 0; |
2716 | } | |
2717 | ||
a652f050 JR |
2718 | static int make_tmp_prefix(const char *prefix) { |
2719 | _cleanup_free_ char *t = NULL; | |
254d1313 | 2720 | _cleanup_close_ int fd = -EBADF; |
a652f050 JR |
2721 | int r; |
2722 | ||
2723 | /* Don't do anything unless we know the dir is actually missing */ | |
2724 | r = access(prefix, F_OK); | |
2725 | if (r >= 0) | |
2726 | return 0; | |
2727 | if (errno != ENOENT) | |
2728 | return -errno; | |
2729 | ||
2053593f | 2730 | WITH_UMASK(000) |
30443439 | 2731 | r = mkdir_parents(prefix, 0755); |
a652f050 JR |
2732 | if (r < 0) |
2733 | return r; | |
2734 | ||
2735 | r = tempfn_random(prefix, NULL, &t); | |
2736 | if (r < 0) | |
2737 | return r; | |
2738 | ||
96603ea0 LP |
2739 | /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for |
2740 | * the suid bit, below. */ | |
2741 | fd = open_mkdir_at(AT_FDCWD, t, O_EXCL|O_CLOEXEC, 0777); | |
2742 | if (fd < 0) | |
2743 | return fd; | |
a652f050 | 2744 | |
96603ea0 LP |
2745 | r = RET_NERRNO(fchmod(fd, 01777)); |
2746 | if (r < 0) { | |
a652f050 JR |
2747 | (void) rmdir(t); |
2748 | return r; | |
2749 | } | |
2750 | ||
96603ea0 LP |
2751 | r = RET_NERRNO(rename(t, prefix)); |
2752 | if (r < 0) { | |
a652f050 JR |
2753 | (void) rmdir(t); |
2754 | return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */ | |
2755 | } | |
2756 | ||
2757 | return 0; | |
2758 | ||
2759 | } | |
2760 | ||
56a13a49 | 2761 | static int setup_one_tmp_dir(const char *id, const char *prefix, char **path, char **tmp_path) { |
613b411c | 2762 | _cleanup_free_ char *x = NULL; |
19cd4e19 | 2763 | _cleanup_free_ char *y = NULL; |
6b46ea73 | 2764 | sd_id128_t boot_id; |
56a13a49 | 2765 | bool rw = true; |
6b46ea73 | 2766 | int r; |
613b411c LP |
2767 | |
2768 | assert(id); | |
2769 | assert(prefix); | |
2770 | assert(path); | |
2771 | ||
6b46ea73 LP |
2772 | /* We include the boot id in the directory so that after a |
2773 | * reboot we can easily identify obsolete directories. */ | |
2774 | ||
2775 | r = sd_id128_get_boot(&boot_id); | |
2776 | if (r < 0) | |
2777 | return r; | |
2778 | ||
85b55869 | 2779 | x = strjoin(prefix, "/systemd-private-", SD_ID128_TO_STRING(boot_id), "-", id, "-XXXXXX"); |
613b411c LP |
2780 | if (!x) |
2781 | return -ENOMEM; | |
2782 | ||
a652f050 JR |
2783 | r = make_tmp_prefix(prefix); |
2784 | if (r < 0) | |
2785 | return r; | |
2786 | ||
2053593f | 2787 | WITH_UMASK(0077) |
56a13a49 ZJS |
2788 | if (!mkdtemp(x)) { |
2789 | if (errno == EROFS || ERRNO_IS_DISK_SPACE(errno)) | |
2790 | rw = false; | |
2791 | else | |
2792 | return -errno; | |
2793 | } | |
613b411c | 2794 | |
56a13a49 | 2795 | if (rw) { |
19cd4e19 | 2796 | y = strjoin(x, "/tmp"); |
2797 | if (!y) | |
2798 | return -ENOMEM; | |
2799 | ||
2053593f | 2800 | WITH_UMASK(0000) |
19cd4e19 | 2801 | if (mkdir(y, 0777 | S_ISVTX) < 0) |
7c76e181 | 2802 | return -errno; |
19cd4e19 | 2803 | |
03bc11d1 | 2804 | r = label_fix_full(AT_FDCWD, y, prefix, 0); |
56a13a49 ZJS |
2805 | if (r < 0) |
2806 | return r; | |
19cd4e19 | 2807 | |
2808 | if (tmp_path) | |
2809 | *tmp_path = TAKE_PTR(y); | |
56a13a49 ZJS |
2810 | } else { |
2811 | /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being | |
2812 | * read-only. This way the service will get the EROFS result as if it was writing to the real | |
2813 | * file system. */ | |
2053593f | 2814 | WITH_UMASK(0000) |
30443439 | 2815 | r = mkdir_p(RUN_SYSTEMD_EMPTY, 0500); |
56a13a49 ZJS |
2816 | if (r < 0) |
2817 | return r; | |
613b411c | 2818 | |
3f181262 LP |
2819 | r = free_and_strdup(&x, RUN_SYSTEMD_EMPTY); |
2820 | if (r < 0) | |
2821 | return r; | |
c17ec25e | 2822 | } |
15ae422b | 2823 | |
1cc6c93a | 2824 | *path = TAKE_PTR(x); |
613b411c LP |
2825 | return 0; |
2826 | } | |
2827 | ||
2828 | int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) { | |
56a13a49 ZJS |
2829 | _cleanup_(namespace_cleanup_tmpdirp) char *a = NULL; |
2830 | _cleanup_(rmdir_and_freep) char *a_tmp = NULL; | |
2831 | char *b; | |
613b411c LP |
2832 | int r; |
2833 | ||
2834 | assert(id); | |
2835 | assert(tmp_dir); | |
2836 | assert(var_tmp_dir); | |
2837 | ||
56a13a49 | 2838 | r = setup_one_tmp_dir(id, "/tmp", &a, &a_tmp); |
613b411c LP |
2839 | if (r < 0) |
2840 | return r; | |
2841 | ||
56a13a49 ZJS |
2842 | r = setup_one_tmp_dir(id, "/var/tmp", &b, NULL); |
2843 | if (r < 0) | |
613b411c | 2844 | return r; |
613b411c | 2845 | |
56a13a49 ZJS |
2846 | a_tmp = mfree(a_tmp); /* avoid rmdir */ |
2847 | *tmp_dir = TAKE_PTR(a); | |
2848 | *var_tmp_dir = TAKE_PTR(b); | |
613b411c LP |
2849 | |
2850 | return 0; | |
2851 | } | |
2852 | ||
13339577 | 2853 | int setup_shareable_ns(int ns_storage_socket[static 2], unsigned long nsflag) { |
5bb1d7fb | 2854 | _cleanup_close_ int ns = -EBADF; |
13339577 | 2855 | int r; |
54c2459d | 2856 | const char *ns_name, *ns_path; |
613b411c | 2857 | |
54c2459d XR |
2858 | assert(ns_storage_socket); |
2859 | assert(ns_storage_socket[0] >= 0); | |
2860 | assert(ns_storage_socket[1] >= 0); | |
2861 | ||
2862 | ns_name = namespace_single_flag_to_string(nsflag); | |
2863 | assert(ns_name); | |
613b411c LP |
2864 | |
2865 | /* We use the passed socketpair as a storage buffer for our | |
76cd584b LP |
2866 | * namespace reference fd. Whatever process runs this first |
2867 | * shall create a new namespace, all others should just join | |
2868 | * it. To serialize that we use a file lock on the socket | |
2869 | * pair. | |
613b411c LP |
2870 | * |
2871 | * It's a bit crazy, but hey, works great! */ | |
2872 | ||
13339577 DDM |
2873 | r = posix_lock(ns_storage_socket[0], LOCK_EX); |
2874 | if (r < 0) | |
2875 | return r; | |
613b411c | 2876 | |
13339577 | 2877 | CLEANUP_POSIX_UNLOCK(ns_storage_socket[0]); |
613b411c | 2878 | |
13339577 DDM |
2879 | ns = receive_one_fd(ns_storage_socket[0], MSG_PEEK|MSG_DONTWAIT); |
2880 | if (ns >= 0) { | |
2881 | /* Yay, found something, so let's join the namespace */ | |
2882 | r = RET_NERRNO(setns(ns, nsflag)); | |
2883 | if (r < 0) | |
2884 | return r; | |
613b411c | 2885 | |
13339577 DDM |
2886 | return 0; |
2887 | } | |
613b411c | 2888 | |
13339577 DDM |
2889 | if (ns != -EAGAIN) |
2890 | return ns; | |
613b411c | 2891 | |
13339577 | 2892 | /* Nothing stored yet, so let's create a new namespace. */ |
613b411c | 2893 | |
13339577 DDM |
2894 | if (unshare(nsflag) < 0) |
2895 | return -errno; | |
613b411c | 2896 | |
13339577 | 2897 | (void) loopback_setup(); |
613b411c | 2898 | |
13339577 DDM |
2899 | ns_path = strjoina("/proc/self/ns/", ns_name); |
2900 | ns = open(ns_path, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
2901 | if (ns < 0) | |
2902 | return -errno; | |
613b411c | 2903 | |
13339577 DDM |
2904 | r = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
2905 | if (r < 0) | |
2906 | return r; | |
613b411c | 2907 | |
13339577 | 2908 | return 1; |
15ae422b | 2909 | } |
417116f2 | 2910 | |
13339577 | 2911 | int open_shareable_ns_path(int ns_storage_socket[static 2], const char *path, unsigned long nsflag) { |
5bb1d7fb | 2912 | _cleanup_close_ int ns = -EBADF; |
13339577 | 2913 | int r; |
51af7fb2 | 2914 | |
54c2459d XR |
2915 | assert(ns_storage_socket); |
2916 | assert(ns_storage_socket[0] >= 0); | |
2917 | assert(ns_storage_socket[1] >= 0); | |
51af7fb2 LP |
2918 | assert(path); |
2919 | ||
54c2459d XR |
2920 | /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in |
2921 | * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will | |
2922 | * allocate a new anonymous ns if needed. */ | |
51af7fb2 | 2923 | |
13339577 DDM |
2924 | r = posix_lock(ns_storage_socket[0], LOCK_EX); |
2925 | if (r < 0) | |
2926 | return r; | |
51af7fb2 | 2927 | |
13339577 | 2928 | CLEANUP_POSIX_UNLOCK(ns_storage_socket[0]); |
51af7fb2 | 2929 | |
13339577 DDM |
2930 | ns = receive_one_fd(ns_storage_socket[0], MSG_PEEK|MSG_DONTWAIT); |
2931 | if (ns >= 0) | |
2932 | return 0; | |
2933 | if (ns != -EAGAIN) | |
2934 | return ns; | |
51af7fb2 | 2935 | |
13339577 | 2936 | /* Nothing stored yet. Open the file from the file system. */ |
51af7fb2 | 2937 | |
13339577 DDM |
2938 | ns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC); |
2939 | if (ns < 0) | |
2940 | return -errno; | |
51af7fb2 | 2941 | |
13339577 DDM |
2942 | r = fd_is_ns(ns, nsflag); |
2943 | if (r == 0) | |
2944 | return -EINVAL; | |
2945 | if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */ | |
2946 | return r; | |
51af7fb2 | 2947 | |
13339577 DDM |
2948 | r = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
2949 | if (r < 0) | |
2950 | return r; | |
51af7fb2 | 2951 | |
13339577 | 2952 | return 1; |
51af7fb2 LP |
2953 | } |
2954 | ||
6e2d7c4f MS |
2955 | bool ns_type_supported(NamespaceType type) { |
2956 | const char *t, *ns_proc; | |
2957 | ||
0fa5b831 LP |
2958 | t = namespace_type_to_string(type); |
2959 | if (!t) /* Don't know how to translate this? Then it's not supported */ | |
6e2d7c4f MS |
2960 | return false; |
2961 | ||
6e2d7c4f | 2962 | ns_proc = strjoina("/proc/self/ns/", t); |
6e2d7c4f MS |
2963 | return access(ns_proc, F_OK) == 0; |
2964 | } | |
2965 | ||
1b8689f9 | 2966 | static const char *const protect_home_table[_PROTECT_HOME_MAX] = { |
cbc056c8 ZJS |
2967 | [PROTECT_HOME_NO] = "no", |
2968 | [PROTECT_HOME_YES] = "yes", | |
1b8689f9 | 2969 | [PROTECT_HOME_READ_ONLY] = "read-only", |
cbc056c8 | 2970 | [PROTECT_HOME_TMPFS] = "tmpfs", |
417116f2 LP |
2971 | }; |
2972 | ||
1e8c7bd5 | 2973 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES); |
5e1c6154 | 2974 | |
1b8689f9 | 2975 | static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = { |
cbc056c8 ZJS |
2976 | [PROTECT_SYSTEM_NO] = "no", |
2977 | [PROTECT_SYSTEM_YES] = "yes", | |
2978 | [PROTECT_SYSTEM_FULL] = "full", | |
3f815163 | 2979 | [PROTECT_SYSTEM_STRICT] = "strict", |
1b8689f9 LP |
2980 | }; |
2981 | ||
1e8c7bd5 | 2982 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES); |
03c791aa | 2983 | |
6e2d7c4f | 2984 | static const char* const namespace_type_table[] = { |
cbc056c8 | 2985 | [NAMESPACE_MOUNT] = "mnt", |
6e2d7c4f | 2986 | [NAMESPACE_CGROUP] = "cgroup", |
cbc056c8 ZJS |
2987 | [NAMESPACE_UTS] = "uts", |
2988 | [NAMESPACE_IPC] = "ipc", | |
2989 | [NAMESPACE_USER] = "user", | |
2990 | [NAMESPACE_PID] = "pid", | |
2991 | [NAMESPACE_NET] = "net", | |
c3b9c418 | 2992 | [NAMESPACE_TIME] = "time", |
6e2d7c4f MS |
2993 | }; |
2994 | ||
2995 | DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType); | |
4e399953 LP |
2996 | |
2997 | static const char* const protect_proc_table[_PROTECT_PROC_MAX] = { | |
2998 | [PROTECT_PROC_DEFAULT] = "default", | |
2999 | [PROTECT_PROC_NOACCESS] = "noaccess", | |
3000 | [PROTECT_PROC_INVISIBLE] = "invisible", | |
3001 | [PROTECT_PROC_PTRACEABLE] = "ptraceable", | |
3002 | }; | |
3003 | ||
3004 | DEFINE_STRING_TABLE_LOOKUP(protect_proc, ProtectProc); | |
3005 | ||
3006 | static const char* const proc_subset_table[_PROC_SUBSET_MAX] = { | |
3007 | [PROC_SUBSET_ALL] = "all", | |
3008 | [PROC_SUBSET_PID] = "pid", | |
3009 | }; | |
3010 | ||
3011 | DEFINE_STRING_TABLE_LOOKUP(proc_subset, ProcSubset); |