]>
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, |
73ff4d48 | 77 | READWRITE_IMPLICIT, /* Should have the lowest priority. */ |
5beb8688 | 78 | _MOUNT_MODE_MAX, |
c17ec25e | 79 | } MountMode; |
15ae422b | 80 | |
34de407a | 81 | typedef struct MountEntry { |
5327c910 | 82 | const char *path_const; /* Memory allocated on stack or static */ |
cfbeb4ef | 83 | MountMode mode:5; |
5327c910 LP |
84 | bool ignore:1; /* Ignore if path does not exist? */ |
85 | bool has_prefix:1; /* Already is prefixed by the root dir? */ | |
cfbeb4ef | 86 | bool read_only:1; /* Shall this mount point be read-only? */ |
9ce4e4b0 | 87 | bool nosuid:1; /* Shall set MS_NOSUID on the mount itself */ |
ddc155b2 TM |
88 | bool noexec:1; /* Shall set MS_NOEXEC on the mount itself */ |
89 | bool exec:1; /* Shall clear MS_NOEXEC on the mount itself */ | |
088696fe | 90 | bool applied:1; /* Already applied */ |
55fe7432 | 91 | char *path_malloc; /* Use this instead of 'path_const' if we had to allocate memory */ |
809ceb82 LB |
92 | const char *unprefixed_path_const; /* If the path was amended with a prefix, these will save the original */ |
93 | char *unprefixed_path_malloc; | |
b3d13314 | 94 | const char *source_const; /* The source path, for bind mounts or images */ |
d2d6c096 | 95 | char *source_malloc; |
2abd4e38 YW |
96 | const char *options_const;/* Mount options for tmpfs */ |
97 | char *options_malloc; | |
98 | unsigned long flags; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */ | |
088696fe | 99 | unsigned n_followed; |
427353f6 | 100 | LIST_HEAD(MountOptions, image_options); |
34de407a | 101 | } MountEntry; |
15ae422b | 102 | |
063c977a LP |
103 | typedef struct MountList { |
104 | MountEntry *mounts; | |
105 | size_t n_mounts; | |
106 | } MountList; | |
107 | ||
94293d65 | 108 | /* 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 | 109 | * something there already. These mounts are hence overridden by any other explicitly configured mounts. */ |
5d997827 LP |
110 | static const MountEntry apivfs_table[] = { |
111 | { "/proc", PROCFS, false }, | |
112 | { "/dev", BIND_DEV, false }, | |
10028263 | 113 | { "/sys", BIND_SYSFS, false }, |
9f563f27 | 114 | { "/run", RUN, false, .options_const = "mode=0755" TMPFS_LIMITS_RUN, .flags = MS_NOSUID|MS_NODEV|MS_STRICTATIME }, |
5d997827 | 115 | }; |
f471b2af | 116 | |
11a30cec | 117 | /* ProtectKernelTunables= option and the related filesystem APIs */ |
788e7201 | 118 | static const MountEntry protect_kernel_tunables_proc_table[] = { |
1e05071d YW |
119 | { "/proc/acpi", READONLY, true }, |
120 | { "/proc/apm", READONLY, true }, /* Obsolete API, there's no point in permitting access to this, ever */ | |
121 | { "/proc/asound", READONLY, true }, | |
122 | { "/proc/bus", READONLY, true }, | |
123 | { "/proc/fs", READONLY, true }, | |
124 | { "/proc/irq", READONLY, true }, | |
125 | { "/proc/kallsyms", INACCESSIBLE, true }, | |
126 | { "/proc/kcore", INACCESSIBLE, true }, | |
127 | { "/proc/latency_stats", READONLY, true }, | |
128 | { "/proc/mtrr", READONLY, true }, | |
129 | { "/proc/scsi", READONLY, true }, | |
4e399953 | 130 | { "/proc/sys", READONLY, true }, |
1e05071d YW |
131 | { "/proc/sysrq-trigger", READONLY, true }, |
132 | { "/proc/timer_stats", READONLY, true }, | |
788e7201 TM |
133 | }; |
134 | ||
135 | static const MountEntry protect_kernel_tunables_sys_table[] = { | |
1e05071d YW |
136 | { "/sys", READONLY, false }, |
137 | { "/sys/fs/bpf", READONLY, true }, | |
138 | { "/sys/fs/cgroup", READWRITE_IMPLICIT, false }, /* READONLY is set by ProtectControlGroups= option */ | |
139 | { "/sys/fs/selinux", READWRITE_IMPLICIT, true }, | |
140 | { "/sys/kernel/debug", READONLY, true }, | |
141 | { "/sys/kernel/tracing", READONLY, true }, | |
11a30cec DH |
142 | }; |
143 | ||
c575770b | 144 | /* ProtectKernelModules= option */ |
34de407a | 145 | static const MountEntry protect_kernel_modules_table[] = { |
c6232fb0 | 146 | { "/usr/lib/modules", INACCESSIBLE, true }, |
c575770b DH |
147 | }; |
148 | ||
94a7b275 | 149 | /* ProtectKernelLogs= option */ |
788e7201 | 150 | static const MountEntry protect_kernel_logs_proc_table[] = { |
94a7b275 | 151 | { "/proc/kmsg", INACCESSIBLE, true }, |
788e7201 TM |
152 | }; |
153 | ||
154 | static const MountEntry protect_kernel_logs_dev_table[] = { | |
94a7b275 KK |
155 | { "/dev/kmsg", INACCESSIBLE, true }, |
156 | }; | |
157 | ||
b6c432ca DH |
158 | /* |
159 | * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of | |
160 | * system should be protected by ProtectSystem= | |
161 | */ | |
34de407a | 162 | static const MountEntry protect_home_read_only_table[] = { |
c6232fb0 LP |
163 | { "/home", READONLY, true }, |
164 | { "/run/user", READONLY, true }, | |
165 | { "/root", READONLY, true }, | |
b6c432ca DH |
166 | }; |
167 | ||
e4da7d8c YW |
168 | /* ProtectHome=tmpfs table */ |
169 | static const MountEntry protect_home_tmpfs_table[] = { | |
7d85383e TM |
170 | { "/home", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, |
171 | { "/run/user", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
172 | { "/root", TMPFS, true, .read_only = true, .options_const = "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
e4da7d8c YW |
173 | }; |
174 | ||
b6c432ca | 175 | /* ProtectHome=yes table */ |
34de407a | 176 | static const MountEntry protect_home_yes_table[] = { |
c6232fb0 LP |
177 | { "/home", INACCESSIBLE, true }, |
178 | { "/run/user", INACCESSIBLE, true }, | |
179 | { "/root", INACCESSIBLE, true }, | |
b6c432ca DH |
180 | }; |
181 | ||
f471b2af | 182 | /* ProtectSystem=yes table */ |
34de407a | 183 | static const MountEntry protect_system_yes_table[] = { |
c6232fb0 LP |
184 | { "/usr", READONLY, false }, |
185 | { "/boot", READONLY, true }, | |
186 | { "/efi", READONLY, true }, | |
f471b2af DH |
187 | }; |
188 | ||
189 | /* ProtectSystem=full includes ProtectSystem=yes */ | |
34de407a | 190 | static const MountEntry protect_system_full_table[] = { |
c6232fb0 LP |
191 | { "/usr", READONLY, false }, |
192 | { "/boot", READONLY, true }, | |
193 | { "/efi", READONLY, true }, | |
194 | { "/etc", READONLY, false }, | |
f471b2af DH |
195 | }; |
196 | ||
3d1b999b LP |
197 | /* ProtectSystem=strict table. In this strict mode, we mount everything read-only, except for /proc, /dev, |
198 | * /sys which are the kernel API VFS, which are left writable, but PrivateDevices= + ProtectKernelTunables= | |
199 | * protect those, and these options should be fully orthogonal. (And of course /home and friends are also | |
200 | * left writable, as ProtectHome= shall manage those, orthogonally). | |
f471b2af | 201 | */ |
34de407a | 202 | static const MountEntry protect_system_strict_table[] = { |
1e05071d YW |
203 | { "/", READONLY, false }, |
204 | { "/proc", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
205 | { "/sys", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
206 | { "/dev", READWRITE_IMPLICIT, false }, /* PrivateDevices= */ | |
207 | { "/home", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
208 | { "/run/user", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
209 | { "/root", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
f471b2af DH |
210 | }; |
211 | ||
3d1b999b LP |
212 | /* ProtectHostname=yes able */ |
213 | static const MountEntry protect_hostname_table[] = { | |
214 | { "/proc/sys/kernel/hostname", READONLY, false }, | |
215 | { "/proc/sys/kernel/domainname", READONLY, false }, | |
216 | }; | |
217 | ||
5beb8688 | 218 | static const char * const mount_mode_table[_MOUNT_MODE_MAX] = { |
cc45ff17 YW |
219 | [INACCESSIBLE] = "inaccessible", |
220 | [OVERLAY_MOUNT] = "overlay", | |
221 | [MOUNT_IMAGES] = "mount-images", | |
222 | [BIND_MOUNT] = "bind", | |
223 | [BIND_MOUNT_RECURSIVE] = "rbind", | |
224 | [PRIVATE_TMP] = "private-tmp", | |
225 | [PRIVATE_TMP_READONLY] = "private-tmp-read-only", | |
226 | [PRIVATE_DEV] = "private-dev", | |
227 | [BIND_DEV] = "bind-dev", | |
228 | [EMPTY_DIR] = "empty", | |
229 | [PRIVATE_SYSFS] = "private-sysfs", | |
230 | [BIND_SYSFS] = "bind-sysfs", | |
231 | [PROCFS] = "procfs", | |
232 | [READONLY] = "read-only", | |
233 | [READWRITE] = "read-write", | |
234 | [NOEXEC] = "noexec", | |
235 | [EXEC] = "exec", | |
236 | [TMPFS] = "tmpfs", | |
237 | [RUN] = "run", | |
238 | [EXTENSION_DIRECTORIES] = "extension-directories", | |
239 | [EXTENSION_IMAGES] = "extension-images", | |
240 | [MQUEUEFS] = "mqueuefs", | |
241 | [READWRITE_IMPLICIT] = "read-write-implicit", | |
5beb8688 YW |
242 | }; |
243 | ||
55ea4ef0 MG |
244 | /* Helper struct for naming simplicity and reusability */ |
245 | static const struct { | |
246 | const char *level_env; | |
247 | const char *level_env_print; | |
248 | } image_class_info[_IMAGE_CLASS_MAX] = { | |
249 | [IMAGE_SYSEXT] = { | |
250 | .level_env = "SYSEXT_LEVEL", | |
251 | .level_env_print = " SYSEXT_LEVEL=", | |
252 | }, | |
253 | [IMAGE_CONFEXT] = { | |
254 | .level_env = "CONFEXT_LEVEL", | |
255 | .level_env_print = " CONFEXT_LEVEL=", | |
256 | } | |
257 | }; | |
258 | ||
5beb8688 YW |
259 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode, MountMode); |
260 | ||
34de407a | 261 | static const char *mount_entry_path(const MountEntry *p) { |
f0a4feb0 DH |
262 | assert(p); |
263 | ||
5327c910 LP |
264 | /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that, |
265 | * otherwise the stack/static ->path field is returned. */ | |
f0a4feb0 | 266 | |
5327c910 | 267 | return p->path_malloc ?: p->path_const; |
f0a4feb0 DH |
268 | } |
269 | ||
809ceb82 LB |
270 | static const char *mount_entry_unprefixed_path(const MountEntry *p) { |
271 | assert(p); | |
272 | ||
273 | /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */ | |
274 | ||
275 | return p->unprefixed_path_malloc ?: p->unprefixed_path_const ?: mount_entry_path(p); | |
276 | } | |
277 | ||
278 | static void mount_entry_consume_prefix(MountEntry *p, char *new_path) { | |
279 | assert(p); | |
280 | assert(p->path_malloc || p->path_const); | |
281 | assert(new_path); | |
282 | ||
283 | /* Saves current path in unprefixed_ variable, and takes over new_path */ | |
284 | ||
285 | free_and_replace(p->unprefixed_path_malloc, p->path_malloc); | |
286 | /* If we didn't have a path on the heap, then it's a static one */ | |
287 | if (!p->unprefixed_path_malloc) | |
288 | p->unprefixed_path_const = p->path_const; | |
289 | p->path_malloc = new_path; | |
290 | p->has_prefix = true; | |
291 | } | |
292 | ||
34de407a | 293 | static bool mount_entry_read_only(const MountEntry *p) { |
cfbeb4ef LP |
294 | assert(p); |
295 | ||
56a13a49 | 296 | return p->read_only || IN_SET(p->mode, READONLY, INACCESSIBLE, PRIVATE_TMP_READONLY); |
cfbeb4ef LP |
297 | } |
298 | ||
ddc155b2 TM |
299 | static bool mount_entry_noexec(const MountEntry *p) { |
300 | assert(p); | |
301 | ||
277b269e | 302 | return p->noexec || IN_SET(p->mode, NOEXEC, INACCESSIBLE, PRIVATE_SYSFS, BIND_SYSFS, PROCFS); |
ddc155b2 TM |
303 | } |
304 | ||
305 | static bool mount_entry_exec(const MountEntry *p) { | |
306 | assert(p); | |
307 | ||
308 | return p->exec || p->mode == EXEC; | |
309 | } | |
310 | ||
d2d6c096 LP |
311 | static const char *mount_entry_source(const MountEntry *p) { |
312 | assert(p); | |
313 | ||
314 | return p->source_malloc ?: p->source_const; | |
315 | } | |
316 | ||
2abd4e38 YW |
317 | static const char *mount_entry_options(const MountEntry *p) { |
318 | assert(p); | |
319 | ||
320 | return p->options_malloc ?: p->options_const; | |
321 | } | |
322 | ||
1eb7e08e LP |
323 | static void mount_entry_done(MountEntry *p) { |
324 | assert(p); | |
325 | ||
326 | p->path_malloc = mfree(p->path_malloc); | |
809ceb82 | 327 | p->unprefixed_path_malloc = mfree(p->unprefixed_path_malloc); |
1eb7e08e | 328 | p->source_malloc = mfree(p->source_malloc); |
2abd4e38 | 329 | p->options_malloc = mfree(p->options_malloc); |
427353f6 | 330 | p->image_options = mount_options_free_all(p->image_options); |
1eb7e08e LP |
331 | } |
332 | ||
063c977a LP |
333 | static void mount_list_done(MountList *ml) { |
334 | assert(ml); | |
335 | ||
336 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) | |
337 | mount_entry_done(m); | |
338 | ||
339 | ml->mounts = mfree(ml->mounts); | |
340 | ml->n_mounts = 0; | |
341 | } | |
342 | ||
343 | static MountEntry *mount_list_extend(MountList *ml) { | |
344 | assert(ml); | |
345 | ||
346 | if (!GREEDY_REALLOC0(ml->mounts, ml->n_mounts+1)) | |
347 | return NULL; | |
348 | ||
349 | return ml->mounts + ml->n_mounts++; | |
350 | } | |
351 | ||
352 | static int append_access_mounts(MountList *ml, char **strv, MountMode mode, bool forcibly_require_prefix) { | |
353 | assert(ml); | |
613b411c | 354 | |
1e05071d | 355 | /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */ |
5327c910 | 356 | |
15ae422b | 357 | STRV_FOREACH(i, strv) { |
5327c910 LP |
358 | bool ignore = false, needs_prefix = false; |
359 | const char *e = *i; | |
15ae422b | 360 | |
5327c910 LP |
361 | /* Look for any prefixes */ |
362 | if (startswith(e, "-")) { | |
363 | e++; | |
9c94d52e | 364 | ignore = true; |
ea92ae33 | 365 | } |
5327c910 LP |
366 | if (startswith(e, "+")) { |
367 | e++; | |
368 | needs_prefix = true; | |
369 | } | |
ea92ae33 | 370 | |
baaa35ad | 371 | if (!path_is_absolute(e)) |
063c977a | 372 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not absolute: %s", e); |
15ae422b | 373 | |
063c977a LP |
374 | MountEntry *me = mount_list_extend(ml); |
375 | if (!me) | |
376 | return log_oom_debug(); | |
377 | ||
378 | *me = (MountEntry) { | |
5327c910 LP |
379 | .path_const = e, |
380 | .mode = mode, | |
381 | .ignore = ignore, | |
d18aff04 | 382 | .has_prefix = !needs_prefix && !forcibly_require_prefix, |
5327c910 | 383 | }; |
15ae422b LP |
384 | } |
385 | ||
386 | return 0; | |
387 | } | |
388 | ||
063c977a LP |
389 | static int append_empty_dir_mounts(MountList *ml, char **strv) { |
390 | assert(ml); | |
6c47cd7d LP |
391 | |
392 | /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the | |
393 | * "/private/" boundary directories for DynamicUser=1. */ | |
394 | ||
395 | STRV_FOREACH(i, strv) { | |
063c977a LP |
396 | MountEntry *me = mount_list_extend(ml); |
397 | if (!me) | |
398 | return log_oom_debug(); | |
6c47cd7d | 399 | |
063c977a | 400 | *me = (MountEntry) { |
6c47cd7d LP |
401 | .path_const = *i, |
402 | .mode = EMPTY_DIR, | |
403 | .ignore = false, | |
6c47cd7d | 404 | .read_only = true, |
9f563f27 | 405 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, |
2abd4e38 | 406 | .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, |
6c47cd7d LP |
407 | }; |
408 | } | |
409 | ||
410 | return 0; | |
411 | } | |
412 | ||
063c977a LP |
413 | static int append_bind_mounts(MountList *ml, const BindMount *binds, size_t n) { |
414 | assert(ml); | |
415 | assert(binds || n == 0); | |
d2d6c096 | 416 | |
063c977a LP |
417 | FOREACH_ARRAY(b, binds, n) { |
418 | MountEntry *me = mount_list_extend(ml); | |
419 | if (!me) | |
420 | return log_oom_debug(); | |
d2d6c096 | 421 | |
063c977a | 422 | *me = (MountEntry) { |
d2d6c096 LP |
423 | .path_const = b->destination, |
424 | .mode = b->recursive ? BIND_MOUNT_RECURSIVE : BIND_MOUNT, | |
425 | .read_only = b->read_only, | |
9ce4e4b0 | 426 | .nosuid = b->nosuid, |
d2d6c096 | 427 | .source_const = b->source, |
4ca763a9 | 428 | .ignore = b->ignore_enoent, |
d2d6c096 LP |
429 | }; |
430 | } | |
431 | ||
432 | return 0; | |
433 | } | |
434 | ||
063c977a LP |
435 | static int append_mount_images(MountList *ml, const MountImage *mount_images, size_t n) { |
436 | assert(ml); | |
437 | assert(mount_images || n == 0); | |
b3d13314 | 438 | |
063c977a LP |
439 | FOREACH_ARRAY(m, mount_images, n) { |
440 | MountEntry *me = mount_list_extend(ml); | |
441 | if (!me) | |
442 | return log_oom_debug(); | |
b3d13314 | 443 | |
063c977a | 444 | *me = (MountEntry) { |
b3d13314 LB |
445 | .path_const = m->destination, |
446 | .mode = MOUNT_IMAGES, | |
447 | .source_const = m->source, | |
427353f6 | 448 | .image_options = m->mount_options, |
b3d13314 LB |
449 | .ignore = m->ignore_enoent, |
450 | }; | |
451 | } | |
452 | ||
453 | return 0; | |
454 | } | |
455 | ||
a07b9926 | 456 | static int append_extensions( |
063c977a | 457 | MountList *ml, |
93f59701 LB |
458 | const char *root, |
459 | const char *extension_dir, | |
460 | char **hierarchies, | |
461 | const MountImage *mount_images, | |
a07b9926 LB |
462 | size_t n, |
463 | char **extension_directories) { | |
93f59701 LB |
464 | |
465 | _cleanup_strv_free_ char **overlays = NULL; | |
93f59701 LB |
466 | int r; |
467 | ||
063c977a LP |
468 | assert(ml); |
469 | ||
a07b9926 | 470 | if (n == 0 && strv_isempty(extension_directories)) |
93f59701 LB |
471 | return 0; |
472 | ||
24759d8f LB |
473 | assert(extension_dir); |
474 | ||
93f59701 | 475 | /* Prepare a list of overlays, that will have as each element a string suitable for being |
7a7d2f16 | 476 | * passed as a lowerdir= parameter, so start with the hierarchy on the root. |
93f59701 LB |
477 | * The overlays vector will have the same number of elements and will correspond to the |
478 | * hierarchies vector, so they can be iterated upon together. */ | |
479 | STRV_FOREACH(hierarchy, hierarchies) { | |
480 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
481 | ||
482 | prefixed_hierarchy = path_join(root, *hierarchy); | |
483 | if (!prefixed_hierarchy) | |
484 | return -ENOMEM; | |
485 | ||
486 | r = strv_consume(&overlays, TAKE_PTR(prefixed_hierarchy)); | |
487 | if (r < 0) | |
488 | return r; | |
489 | } | |
490 | ||
491 | /* First, prepare a mount for each image, but these won't be visible to the unit, instead | |
492 | * they will be mounted in our propagate directory, and used as a source for the overlay. */ | |
493 | for (size_t i = 0; i < n; i++) { | |
494 | _cleanup_free_ char *mount_point = NULL; | |
495 | const MountImage *m = mount_images + i; | |
496 | ||
063c977a | 497 | if (asprintf(&mount_point, "%s/%zu", extension_dir, i) < 0) |
93f59701 LB |
498 | return -ENOMEM; |
499 | ||
500 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
501 | _cleanup_free_ char *prefixed_hierarchy = NULL, *escaped = NULL, *lowerdir = NULL; | |
502 | ||
503 | prefixed_hierarchy = path_join(mount_point, hierarchies[j]); | |
504 | if (!prefixed_hierarchy) | |
505 | return -ENOMEM; | |
506 | ||
507 | escaped = shell_escape(prefixed_hierarchy, ",:"); | |
508 | if (!escaped) | |
509 | return -ENOMEM; | |
510 | ||
511 | /* Note that lowerdir= parameters are in 'reverse' order, so the | |
512 | * top-most directory in the overlay comes first in the list. */ | |
513 | lowerdir = strjoin(escaped, ":", overlays[j]); | |
514 | if (!lowerdir) | |
515 | return -ENOMEM; | |
516 | ||
517 | free_and_replace(overlays[j], lowerdir); | |
518 | } | |
519 | ||
063c977a LP |
520 | MountEntry *me = mount_list_extend(ml); |
521 | if (!me) | |
522 | return log_oom_debug(); | |
523 | ||
524 | *me = (MountEntry) { | |
93f59701 LB |
525 | .path_malloc = TAKE_PTR(mount_point), |
526 | .image_options = m->mount_options, | |
527 | .ignore = m->ignore_enoent, | |
528 | .source_const = m->source, | |
529 | .mode = EXTENSION_IMAGES, | |
530 | .has_prefix = true, | |
531 | }; | |
532 | } | |
533 | ||
a07b9926 LB |
534 | /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory. |
535 | * Bind mount them in the same location as the ExtensionImages, so that we | |
536 | * can check that they are valid trees (extension-release.d). */ | |
537 | STRV_FOREACH(extension_directory, extension_directories) { | |
538 | _cleanup_free_ char *mount_point = NULL, *source = NULL; | |
539 | const char *e = *extension_directory; | |
540 | bool ignore_enoent = false; | |
541 | ||
542 | /* Pick up the counter where the ExtensionImages left it. */ | |
063c977a | 543 | if (asprintf(&mount_point, "%s/%zu", extension_dir, n++) < 0) |
a07b9926 LB |
544 | return -ENOMEM; |
545 | ||
546 | /* Look for any prefixes */ | |
547 | if (startswith(e, "-")) { | |
548 | e++; | |
549 | ignore_enoent = true; | |
550 | } | |
551 | /* Ignore this for now */ | |
552 | if (startswith(e, "+")) | |
553 | e++; | |
554 | ||
555 | source = strdup(e); | |
556 | if (!source) | |
557 | return -ENOMEM; | |
558 | ||
559 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
560 | _cleanup_free_ char *prefixed_hierarchy = NULL, *escaped = NULL, *lowerdir = NULL; | |
561 | ||
562 | prefixed_hierarchy = path_join(mount_point, hierarchies[j]); | |
563 | if (!prefixed_hierarchy) | |
564 | return -ENOMEM; | |
565 | ||
566 | escaped = shell_escape(prefixed_hierarchy, ",:"); | |
567 | if (!escaped) | |
568 | return -ENOMEM; | |
569 | ||
570 | /* Note that lowerdir= parameters are in 'reverse' order, so the | |
571 | * top-most directory in the overlay comes first in the list. */ | |
572 | lowerdir = strjoin(escaped, ":", overlays[j]); | |
573 | if (!lowerdir) | |
574 | return -ENOMEM; | |
575 | ||
576 | free_and_replace(overlays[j], lowerdir); | |
577 | } | |
578 | ||
063c977a LP |
579 | MountEntry *me = mount_list_extend(ml); |
580 | if (!me) | |
581 | return log_oom_debug(); | |
582 | ||
583 | *me = (MountEntry) { | |
a07b9926 | 584 | .path_malloc = TAKE_PTR(mount_point), |
c9423746 | 585 | .source_malloc = TAKE_PTR(source), |
a07b9926 LB |
586 | .mode = EXTENSION_DIRECTORIES, |
587 | .ignore = ignore_enoent, | |
588 | .has_prefix = true, | |
589 | .read_only = true, | |
590 | }; | |
591 | } | |
592 | ||
93f59701 LB |
593 | /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings |
594 | * set up earlier. */ | |
595 | for (size_t i = 0; hierarchies && hierarchies[i]; ++i) { | |
596 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
597 | ||
598 | prefixed_hierarchy = path_join(root, hierarchies[i]); | |
599 | if (!prefixed_hierarchy) | |
600 | return -ENOMEM; | |
601 | ||
063c977a LP |
602 | MountEntry *me = mount_list_extend(ml); |
603 | if (!me) | |
604 | return log_oom_debug(); | |
605 | ||
606 | *me = (MountEntry) { | |
93f59701 LB |
607 | .path_malloc = TAKE_PTR(prefixed_hierarchy), |
608 | .options_malloc = TAKE_PTR(overlays[i]), | |
609 | .mode = OVERLAY_MOUNT, | |
610 | .has_prefix = true, | |
611 | .ignore = true, /* If the source image doesn't set the ignore bit it will fail earlier. */ | |
612 | }; | |
613 | } | |
614 | ||
615 | return 0; | |
616 | } | |
617 | ||
063c977a LP |
618 | static int append_tmpfs_mounts(MountList *ml, const TemporaryFileSystem *tmpfs, size_t n) { |
619 | assert(ml); | |
620 | assert(tmpfs || n == 0); | |
2abd4e38 | 621 | |
063c977a | 622 | FOREACH_ARRAY(t, tmpfs, n) { |
2abd4e38 | 623 | _cleanup_free_ char *o = NULL, *str = NULL; |
ad8e66dc | 624 | unsigned long flags; |
2abd4e38 | 625 | bool ro = false; |
b67ec8e5 | 626 | int r; |
2abd4e38 | 627 | |
baaa35ad | 628 | if (!path_is_absolute(t->path)) |
063c977a | 629 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not absolute: %s", t->path); |
2abd4e38 | 630 | |
b67ec8e5 | 631 | str = strjoin("mode=0755" NESTED_TMPFS_LIMITS ",", t->options); |
ad8e66dc AJ |
632 | if (!str) |
633 | return -ENOMEM; | |
2abd4e38 | 634 | |
ad8e66dc AJ |
635 | r = mount_option_mangle(str, MS_NODEV|MS_STRICTATIME, &flags, &o); |
636 | if (r < 0) | |
637 | return log_debug_errno(r, "Failed to parse mount option '%s': %m", str); | |
2abd4e38 | 638 | |
ad8e66dc AJ |
639 | ro = flags & MS_RDONLY; |
640 | if (ro) | |
641 | flags ^= MS_RDONLY; | |
2abd4e38 | 642 | |
063c977a LP |
643 | MountEntry *me = mount_list_extend(ml); |
644 | if (!me) | |
645 | return log_oom_debug(); | |
646 | ||
647 | *me = (MountEntry) { | |
2abd4e38 YW |
648 | .path_const = t->path, |
649 | .mode = TMPFS, | |
650 | .read_only = ro, | |
ad8e66dc | 651 | .options_malloc = TAKE_PTR(o), |
2abd4e38 YW |
652 | .flags = flags, |
653 | }; | |
2abd4e38 YW |
654 | } |
655 | ||
656 | return 0; | |
657 | } | |
658 | ||
063c977a LP |
659 | static int append_static_mounts(MountList *ml, const MountEntry *mounts, size_t n, bool ignore_protect) { |
660 | assert(ml); | |
661 | assert(mounts || n == 0); | |
11a30cec | 662 | |
5327c910 | 663 | /* Adds a list of static pre-defined entries */ |
f471b2af | 664 | |
063c977a LP |
665 | FOREACH_ARRAY(m, mounts, n) { |
666 | MountEntry *me = mount_list_extend(ml); | |
667 | if (!me) | |
668 | return log_oom_debug(); | |
669 | ||
670 | *me = (MountEntry) { | |
671 | .path_const = mount_entry_path(m), | |
672 | .mode = m->mode, | |
673 | .ignore = m->ignore || ignore_protect, | |
5327c910 | 674 | }; |
063c977a | 675 | } |
f471b2af DH |
676 | |
677 | return 0; | |
678 | } | |
679 | ||
063c977a LP |
680 | static int append_protect_home(MountList *ml, ProtectHome protect_home, bool ignore_protect) { |
681 | assert(ml); | |
c575770b | 682 | |
5327c910 | 683 | switch (protect_home) { |
b6c432ca | 684 | |
5327c910 | 685 | case PROTECT_HOME_NO: |
b6c432ca DH |
686 | return 0; |
687 | ||
b6c432ca | 688 | case PROTECT_HOME_READ_ONLY: |
063c977a | 689 | return append_static_mounts(ml, protect_home_read_only_table, ELEMENTSOF(protect_home_read_only_table), ignore_protect); |
5327c910 | 690 | |
e4da7d8c | 691 | case PROTECT_HOME_TMPFS: |
063c977a | 692 | return append_static_mounts(ml, protect_home_tmpfs_table, ELEMENTSOF(protect_home_tmpfs_table), ignore_protect); |
e4da7d8c | 693 | |
b6c432ca | 694 | case PROTECT_HOME_YES: |
063c977a | 695 | return append_static_mounts(ml, protect_home_yes_table, ELEMENTSOF(protect_home_yes_table), ignore_protect); |
5327c910 | 696 | |
b6c432ca | 697 | default: |
04499a70 | 698 | assert_not_reached(); |
b6c432ca | 699 | } |
b6c432ca DH |
700 | } |
701 | ||
063c977a LP |
702 | static int append_protect_system(MountList *ml, ProtectSystem protect_system, bool ignore_protect) { |
703 | assert(ml); | |
f471b2af | 704 | |
5327c910 LP |
705 | switch (protect_system) { |
706 | ||
707 | case PROTECT_SYSTEM_NO: | |
f471b2af DH |
708 | return 0; |
709 | ||
f471b2af | 710 | case PROTECT_SYSTEM_STRICT: |
063c977a | 711 | return append_static_mounts(ml, protect_system_strict_table, ELEMENTSOF(protect_system_strict_table), ignore_protect); |
5327c910 | 712 | |
f471b2af | 713 | case PROTECT_SYSTEM_YES: |
063c977a | 714 | return append_static_mounts(ml, protect_system_yes_table, ELEMENTSOF(protect_system_yes_table), ignore_protect); |
5327c910 | 715 | |
f471b2af | 716 | case PROTECT_SYSTEM_FULL: |
063c977a | 717 | return append_static_mounts(ml, protect_system_full_table, ELEMENTSOF(protect_system_full_table), ignore_protect); |
5327c910 | 718 | |
f471b2af | 719 | default: |
04499a70 | 720 | assert_not_reached(); |
f471b2af | 721 | } |
11a30cec DH |
722 | } |
723 | ||
93bab288 | 724 | static int mount_path_compare(const MountEntry *a, const MountEntry *b) { |
a0827e2b | 725 | int d; |
15ae422b | 726 | |
a07b9926 | 727 | /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first |
93f59701 LB |
728 | * regardless of the prefix - they are set up in the propagate directory anyway */ |
729 | d = -CMP(a->mode == EXTENSION_IMAGES, b->mode == EXTENSION_IMAGES); | |
a07b9926 LB |
730 | if (d != 0) |
731 | return d; | |
732 | d = -CMP(a->mode == EXTENSION_DIRECTORIES, b->mode == EXTENSION_DIRECTORIES); | |
93f59701 LB |
733 | if (d != 0) |
734 | return d; | |
735 | ||
6ee1a919 | 736 | /* If the paths are not equal, then order prefixes first */ |
93bab288 | 737 | d = path_compare(mount_entry_path(a), mount_entry_path(b)); |
6ee1a919 LP |
738 | if (d != 0) |
739 | return d; | |
15ae422b | 740 | |
6ee1a919 | 741 | /* If the paths are equal, check the mode */ |
93bab288 | 742 | return CMP((int) a->mode, (int) b->mode); |
15ae422b LP |
743 | } |
744 | ||
063c977a | 745 | static int prefix_where_needed(MountList *ml, const char *root_directory) { |
4a756839 | 746 | /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */ |
5327c910 | 747 | |
063c977a | 748 | assert(ml); |
fe96c0f8 | 749 | |
063c977a | 750 | FOREACH_ARRAY(me, ml->mounts, ml->n_mounts) { |
5327c910 LP |
751 | char *s; |
752 | ||
063c977a | 753 | if (me->has_prefix) |
5327c910 LP |
754 | continue; |
755 | ||
063c977a | 756 | s = path_join(root_directory, mount_entry_path(me)); |
5327c910 LP |
757 | if (!s) |
758 | return -ENOMEM; | |
759 | ||
063c977a | 760 | mount_entry_consume_prefix(me, s); |
5327c910 LP |
761 | } |
762 | ||
763 | return 0; | |
764 | } | |
765 | ||
063c977a | 766 | static void drop_duplicates(MountList *ml) { |
34de407a | 767 | MountEntry *f, *t, *previous; |
15ae422b | 768 | |
063c977a | 769 | assert(ml); |
15ae422b | 770 | |
fe3c2583 LP |
771 | /* Drops duplicate entries. Expects that the array is properly ordered already. */ |
772 | ||
063c977a | 773 | for (f = ml->mounts, t = ml->mounts, previous = NULL; f < ml->mounts + ml->n_mounts; f++) { |
15ae422b | 774 | |
fe3c2583 | 775 | /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare() |
088696fe LP |
776 | * above. Note that we only drop duplicates that haven't been mounted yet. */ |
777 | if (previous && | |
778 | path_equal(mount_entry_path(f), mount_entry_path(previous)) && | |
779 | !f->applied && !previous->applied) { | |
5beb8688 | 780 | log_debug("%s (%s) is duplicate.", mount_entry_path(f), mount_mode_to_string(f->mode)); |
ddc155b2 TM |
781 | /* Propagate the flags to the remaining entry */ |
782 | previous->read_only = previous->read_only || mount_entry_read_only(f); | |
783 | previous->noexec = previous->noexec || mount_entry_noexec(f); | |
784 | previous->exec = previous->exec || mount_entry_exec(f); | |
1eb7e08e | 785 | mount_entry_done(f); |
15ae422b | 786 | continue; |
fe3c2583 | 787 | } |
15ae422b | 788 | |
e2d7c1a0 | 789 | *t = *f; |
15ae422b | 790 | previous = t; |
fe3c2583 LP |
791 | t++; |
792 | } | |
793 | ||
063c977a | 794 | ml->n_mounts = t - ml->mounts; |
fe3c2583 LP |
795 | } |
796 | ||
063c977a | 797 | static void drop_inaccessible(MountList *ml) { |
34de407a | 798 | MountEntry *f, *t; |
fe3c2583 LP |
799 | const char *clear = NULL; |
800 | ||
063c977a | 801 | assert(ml); |
fe3c2583 LP |
802 | |
803 | /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly | |
804 | * ordered already. */ | |
805 | ||
063c977a | 806 | for (f = ml->mounts, t = ml->mounts; f < ml->mounts + ml->n_mounts; f++) { |
fe3c2583 LP |
807 | |
808 | /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop | |
809 | * it, as inaccessible paths really should drop the entire subtree. */ | |
34de407a LP |
810 | if (clear && path_startswith(mount_entry_path(f), clear)) { |
811 | log_debug("%s is masked by %s.", mount_entry_path(f), clear); | |
1eb7e08e | 812 | mount_entry_done(f); |
fe3c2583 LP |
813 | continue; |
814 | } | |
15ae422b | 815 | |
34de407a | 816 | clear = f->mode == INACCESSIBLE ? mount_entry_path(f) : NULL; |
fe3c2583 LP |
817 | |
818 | *t = *f; | |
15ae422b LP |
819 | t++; |
820 | } | |
821 | ||
063c977a | 822 | ml->n_mounts = t - ml->mounts; |
15ae422b LP |
823 | } |
824 | ||
063c977a | 825 | static void drop_nop(MountList *ml) { |
34de407a | 826 | MountEntry *f, *t; |
7648a565 | 827 | |
063c977a | 828 | assert(ml); |
7648a565 LP |
829 | |
830 | /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the | |
831 | * list is ordered by prefixes. */ | |
832 | ||
063c977a | 833 | for (f = ml->mounts, t = ml->mounts; f < ml->mounts + ml->n_mounts; f++) { |
7648a565 | 834 | |
1e05071d YW |
835 | /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */ |
836 | if (IN_SET(f->mode, READONLY, READWRITE, READWRITE_IMPLICIT)) { | |
e7bf2fca | 837 | MountEntry *found = NULL; |
7648a565 LP |
838 | |
839 | /* Now let's find the first parent of the entry we are looking at. */ | |
063c977a | 840 | for (MountEntry *p = PTR_SUB1(t, ml->mounts); p; p = PTR_SUB1(p, ml->mounts)) |
34de407a | 841 | if (path_startswith(mount_entry_path(f), mount_entry_path(p))) { |
e7bf2fca | 842 | found = p; |
7648a565 LP |
843 | break; |
844 | } | |
7648a565 LP |
845 | |
846 | /* We found it, let's see if it's the same mode, if so, we can drop this entry */ | |
e7bf2fca | 847 | if (found && found->mode == f->mode) { |
5beb8688 YW |
848 | log_debug("%s (%s) is made redundant by %s (%s)", |
849 | mount_entry_path(f), mount_mode_to_string(f->mode), | |
e7bf2fca | 850 | mount_entry_path(found), mount_mode_to_string(found->mode)); |
1eb7e08e | 851 | mount_entry_done(f); |
7648a565 LP |
852 | continue; |
853 | } | |
854 | } | |
855 | ||
856 | *t = *f; | |
857 | t++; | |
858 | } | |
859 | ||
063c977a | 860 | ml->n_mounts = t - ml->mounts; |
7648a565 LP |
861 | } |
862 | ||
063c977a | 863 | static void drop_outside_root(MountList *ml, const char *root_directory) { |
34de407a | 864 | MountEntry *f, *t; |
cd2902c9 | 865 | |
063c977a | 866 | assert(ml); |
cd2902c9 | 867 | |
1d54cd5d | 868 | /* Nothing to do */ |
cd2902c9 LP |
869 | if (!root_directory) |
870 | return; | |
871 | ||
872 | /* Drops all mounts that are outside of the root directory. */ | |
873 | ||
063c977a | 874 | for (f = ml->mounts, t = ml->mounts; f < ml->mounts + ml->n_mounts; f++) { |
cd2902c9 | 875 | |
a07b9926 LB |
876 | /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */ |
877 | if (!IN_SET(f->mode, EXTENSION_IMAGES, EXTENSION_DIRECTORIES) && !path_startswith(mount_entry_path(f), root_directory)) { | |
34de407a | 878 | log_debug("%s is outside of root directory.", mount_entry_path(f)); |
1eb7e08e | 879 | mount_entry_done(f); |
cd2902c9 LP |
880 | continue; |
881 | } | |
882 | ||
883 | *t = *f; | |
884 | t++; | |
885 | } | |
886 | ||
063c977a | 887 | ml->n_mounts = t - ml->mounts; |
cd2902c9 LP |
888 | } |
889 | ||
b2a60844 LP |
890 | static int clone_device_node( |
891 | const char *d, | |
892 | const char *temporary_mount, | |
893 | bool *make_devnode) { | |
894 | ||
895 | _cleanup_free_ char *sl = NULL; | |
896 | const char *dn, *bn, *t; | |
b5e99f23 ДГ |
897 | struct stat st; |
898 | int r; | |
899 | ||
414b304b | 900 | if (stat(d, &st) < 0) { |
b2a60844 LP |
901 | if (errno == ENOENT) { |
902 | log_debug_errno(errno, "Device node '%s' to clone does not exist, ignoring.", d); | |
af984e13 | 903 | return -ENXIO; |
b2a60844 LP |
904 | } |
905 | ||
906 | return log_debug_errno(errno, "Failed to stat() device node '%s' to clone, ignoring: %m", d); | |
b5e99f23 ДГ |
907 | } |
908 | ||
909 | if (!S_ISBLK(st.st_mode) && | |
baaa35ad ZJS |
910 | !S_ISCHR(st.st_mode)) |
911 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
912 | "Device node '%s' to clone is not a device node, ignoring.", | |
913 | d); | |
b5e99f23 | 914 | |
6f7f3a33 | 915 | dn = strjoina(temporary_mount, d); |
b5e99f23 | 916 | |
b2a60844 | 917 | /* First, try to create device node properly */ |
16498617 CB |
918 | if (*make_devnode) { |
919 | mac_selinux_create_file_prepare(d, st.st_mode); | |
920 | r = mknod(dn, st.st_mode, st.st_rdev); | |
921 | mac_selinux_create_file_clear(); | |
b2a60844 LP |
922 | if (r >= 0) |
923 | goto add_symlink; | |
16498617 CB |
924 | if (errno != EPERM) |
925 | return log_debug_errno(errno, "mknod failed for %s: %m", d); | |
926 | ||
b2a60844 | 927 | /* This didn't work, let's not try this again for the next iterations. */ |
16498617 CB |
928 | *make_devnode = false; |
929 | } | |
930 | ||
d73020f2 | 931 | /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target. |
1acf344d | 932 | * Do not prepare device-node SELinux label (see issue 13762) */ |
16498617 | 933 | r = mknod(dn, S_IFREG, 0); |
16498617 | 934 | if (r < 0 && errno != EEXIST) |
b2a60844 | 935 | return log_debug_errno(errno, "mknod() fallback failed for '%s': %m", d); |
16498617 | 936 | |
21935150 LP |
937 | /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard |
938 | * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or | |
939 | * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */ | |
940 | r = mount_nofollow_verbose(LOG_DEBUG, d, dn, NULL, MS_BIND, NULL); | |
941 | if (r < 0) | |
942 | return r; | |
b2a60844 LP |
943 | |
944 | add_symlink: | |
945 | bn = path_startswith(d, "/dev/"); | |
946 | if (!bn) | |
947 | return 0; | |
948 | ||
949 | /* Create symlinks like /dev/char/1:9 → ../urandom */ | |
ec61371f | 950 | if (asprintf(&sl, "%s/dev/%s/" DEVNUM_FORMAT_STR, |
cbc056c8 ZJS |
951 | temporary_mount, |
952 | S_ISCHR(st.st_mode) ? "char" : "block", | |
ec61371f | 953 | DEVNUM_FORMAT_VAL(st.st_rdev)) < 0) |
d4f0878e | 954 | return log_oom_debug(); |
b2a60844 LP |
955 | |
956 | (void) mkdir_parents(sl, 0755); | |
957 | ||
958 | t = strjoina("../", bn); | |
b2a60844 | 959 | if (symlink(t, sl) < 0) |
2e4a4fae | 960 | log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl); |
b5e99f23 | 961 | |
af984e13 | 962 | return 0; |
b5e99f23 ДГ |
963 | } |
964 | ||
cd7f3702 DDM |
965 | static char *settle_runtime_dir(RuntimeScope scope) { |
966 | char *runtime_dir; | |
967 | ||
968 | if (scope != RUNTIME_SCOPE_USER) | |
969 | return strdup("/run/"); | |
970 | ||
971 | if (asprintf(&runtime_dir, "/run/user/" UID_FMT, geteuid()) < 0) | |
972 | return NULL; | |
973 | ||
974 | return runtime_dir; | |
975 | } | |
976 | ||
0c3d606c YW |
977 | static int create_temporary_mount_point(RuntimeScope scope, char **ret) { |
978 | _cleanup_free_ char *runtime_dir = NULL, *temporary_mount = NULL; | |
979 | ||
980 | assert(ret); | |
981 | ||
982 | runtime_dir = settle_runtime_dir(scope); | |
983 | if (!runtime_dir) | |
984 | return log_oom_debug(); | |
985 | ||
986 | temporary_mount = path_join(runtime_dir, "systemd/namespace-XXXXXX"); | |
987 | if (!temporary_mount) | |
988 | return log_oom_debug(); | |
989 | ||
990 | if (!mkdtemp(temporary_mount)) | |
991 | return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount); | |
992 | ||
993 | *ret = TAKE_PTR(temporary_mount); | |
994 | return 0; | |
995 | } | |
996 | ||
cd7f3702 | 997 | static int mount_private_dev(MountEntry *m, RuntimeScope scope) { |
7f112f50 LP |
998 | static const char devnodes[] = |
999 | "/dev/null\0" | |
1000 | "/dev/zero\0" | |
1001 | "/dev/full\0" | |
1002 | "/dev/random\0" | |
1003 | "/dev/urandom\0" | |
1004 | "/dev/tty\0"; | |
1005 | ||
0c3d606c | 1006 | _cleanup_free_ char *temporary_mount = NULL; |
12e2b70f | 1007 | const char *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL; |
16498617 | 1008 | bool can_mknod = true; |
7f112f50 LP |
1009 | int r; |
1010 | ||
1011 | assert(m); | |
1012 | ||
0c3d606c YW |
1013 | r = create_temporary_mount_point(scope, &temporary_mount); |
1014 | if (r < 0) | |
1015 | return r; | |
2b85f4e1 | 1016 | |
63c372cb | 1017 | dev = strjoina(temporary_mount, "/dev"); |
dc751688 | 1018 | (void) mkdir(dev, 0755); |
9f563f27 | 1019 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV); |
21935150 | 1020 | if (r < 0) |
2b85f4e1 | 1021 | goto fail; |
21935150 | 1022 | |
03bc11d1 | 1023 | r = label_fix_full(AT_FDCWD, dev, "/dev", 0); |
c3151977 | 1024 | if (r < 0) { |
1ce268c7 | 1025 | log_debug_errno(r, "Failed to fix label of '%s' as /dev: %m", dev); |
c3151977 TM |
1026 | goto fail; |
1027 | } | |
2b85f4e1 | 1028 | |
63c372cb | 1029 | devpts = strjoina(temporary_mount, "/dev/pts"); |
dc751688 | 1030 | (void) mkdir(devpts, 0755); |
21935150 LP |
1031 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/pts", devpts, NULL, MS_BIND, NULL); |
1032 | if (r < 0) | |
2b85f4e1 | 1033 | goto fail; |
2b85f4e1 | 1034 | |
2e4a4fae YW |
1035 | /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx. |
1036 | * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible. | |
1037 | * Thus, in that case make a clone. | |
1038 | * In nspawn and other containers it will be a symlink, in that case make it a symlink. */ | |
36ce7110 | 1039 | r = is_symlink("/dev/ptmx"); |
2e4a4fae YW |
1040 | if (r < 0) { |
1041 | log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m"); | |
3164e3cb | 1042 | goto fail; |
2e4a4fae | 1043 | } else if (r > 0) { |
414b304b ДГ |
1044 | devptmx = strjoina(temporary_mount, "/dev/ptmx"); |
1045 | if (symlink("pts/ptmx", devptmx) < 0) { | |
2e4a4fae | 1046 | r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx); |
414b304b ДГ |
1047 | goto fail; |
1048 | } | |
1049 | } else { | |
16498617 | 1050 | r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod); |
152c475f LP |
1051 | if (r < 0) |
1052 | goto fail; | |
414b304b | 1053 | } |
e06b6479 | 1054 | |
63c372cb | 1055 | devshm = strjoina(temporary_mount, "/dev/shm"); |
8d953682 | 1056 | (void) mkdir(devshm, 0755); |
21935150 LP |
1057 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/shm", devshm, NULL, MS_BIND, NULL); |
1058 | if (r < 0) | |
2b85f4e1 | 1059 | goto fail; |
2b85f4e1 | 1060 | |
63c372cb | 1061 | devmqueue = strjoina(temporary_mount, "/dev/mqueue"); |
dc751688 | 1062 | (void) mkdir(devmqueue, 0755); |
21935150 | 1063 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/mqueue", devmqueue, NULL, MS_BIND, NULL); |
2b85f4e1 | 1064 | |
63c372cb | 1065 | devhugepages = strjoina(temporary_mount, "/dev/hugepages"); |
dc751688 | 1066 | (void) mkdir(devhugepages, 0755); |
21935150 | 1067 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/hugepages", devhugepages, NULL, MS_BIND, NULL); |
2b85f4e1 | 1068 | |
63c372cb | 1069 | devlog = strjoina(temporary_mount, "/dev/log"); |
2e4a4fae YW |
1070 | if (symlink("/run/systemd/journal/dev-log", devlog) < 0) |
1071 | log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog); | |
82d25240 | 1072 | |
7f112f50 | 1073 | NULSTR_FOREACH(d, devnodes) { |
16498617 | 1074 | r = clone_device_node(d, temporary_mount, &can_mknod); |
37b22b3b | 1075 | /* ENXIO means the *source* is not a device file, skip creation in that case */ |
af984e13 | 1076 | if (r < 0 && r != -ENXIO) |
2b85f4e1 | 1077 | goto fail; |
7f112f50 LP |
1078 | } |
1079 | ||
2e4a4fae YW |
1080 | r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID); |
1081 | if (r < 0) | |
105a1a36 | 1082 | log_debug_errno(r, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount); |
7f112f50 | 1083 | |
4a9e03aa DDM |
1084 | /* Make the bind mount read-only. */ |
1085 | r = mount_nofollow_verbose(LOG_DEBUG, NULL, dev, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY, NULL); | |
1086 | if (r < 0) | |
1087 | return r; | |
1088 | ||
d73020f2 LP |
1089 | /* Create the /dev directory if missing. It is more likely to be missing when the service is started |
1090 | * with RootDirectory. This is consistent with mount units creating the mount points when missing. */ | |
34de407a | 1091 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
ee818b89 | 1092 | |
9e5f8252 | 1093 | /* Unmount everything in old /dev */ |
2e4a4fae YW |
1094 | r = umount_recursive(mount_entry_path(m), 0); |
1095 | if (r < 0) | |
1096 | log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m)); | |
1097 | ||
21935150 LP |
1098 | r = mount_nofollow_verbose(LOG_DEBUG, dev, mount_entry_path(m), NULL, MS_MOVE, NULL); |
1099 | if (r < 0) | |
2b85f4e1 | 1100 | goto fail; |
7f112f50 | 1101 | |
1019a48f LP |
1102 | (void) rmdir(dev); |
1103 | (void) rmdir(temporary_mount); | |
7f112f50 | 1104 | |
2b85f4e1 | 1105 | return 0; |
7f112f50 | 1106 | |
2b85f4e1 LP |
1107 | fail: |
1108 | if (devpts) | |
21935150 | 1109 | (void) umount_verbose(LOG_DEBUG, devpts, UMOUNT_NOFOLLOW); |
7f112f50 | 1110 | |
2b85f4e1 | 1111 | if (devshm) |
21935150 | 1112 | (void) umount_verbose(LOG_DEBUG, devshm, UMOUNT_NOFOLLOW); |
7f112f50 | 1113 | |
2b85f4e1 | 1114 | if (devhugepages) |
21935150 | 1115 | (void) umount_verbose(LOG_DEBUG, devhugepages, UMOUNT_NOFOLLOW); |
7f112f50 | 1116 | |
2b85f4e1 | 1117 | if (devmqueue) |
21935150 | 1118 | (void) umount_verbose(LOG_DEBUG, devmqueue, UMOUNT_NOFOLLOW); |
7f112f50 | 1119 | |
21935150 | 1120 | (void) umount_verbose(LOG_DEBUG, dev, UMOUNT_NOFOLLOW); |
1019a48f LP |
1121 | (void) rmdir(dev); |
1122 | (void) rmdir(temporary_mount); | |
7f112f50 | 1123 | |
2b85f4e1 | 1124 | return r; |
7f112f50 LP |
1125 | } |
1126 | ||
2a2969fd | 1127 | static int mount_bind_dev(const MountEntry *m) { |
5d997827 LP |
1128 | int r; |
1129 | ||
1130 | assert(m); | |
1131 | ||
d73020f2 LP |
1132 | /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the |
1133 | * service's /dev. This is only used when RootDirectory= is set. */ | |
5d997827 | 1134 | |
645767d6 LP |
1135 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1136 | ||
5d997827 LP |
1137 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1138 | if (r < 0) | |
1139 | return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m"); | |
1140 | if (r > 0) /* make this a NOP if /dev is already a mount point */ | |
1141 | return 0; | |
1142 | ||
51bf9b1a | 1143 | return mount_nofollow_verbose(LOG_DEBUG, "/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
5d997827 LP |
1144 | } |
1145 | ||
10028263 | 1146 | static int mount_bind_sysfs(const MountEntry *m) { |
5d997827 LP |
1147 | int r; |
1148 | ||
1149 | assert(m); | |
1150 | ||
645767d6 LP |
1151 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1152 | ||
5d997827 LP |
1153 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1154 | if (r < 0) | |
1155 | return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m"); | |
1156 | if (r > 0) /* make this a NOP if /sys is already a mount point */ | |
1157 | return 0; | |
1158 | ||
1159 | /* Bind mount the host's version so that we get all child mounts of it, too. */ | |
51bf9b1a | 1160 | return mount_nofollow_verbose(LOG_DEBUG, "/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
5d997827 LP |
1161 | } |
1162 | ||
b18f4eed YW |
1163 | static int mount_private_apivfs( |
1164 | const char *fstype, | |
1165 | const char *entry_path, | |
1166 | const char *bind_source, | |
1167 | const char *opts) { | |
1168 | ||
553e3660 YW |
1169 | int r, n; |
1170 | ||
b18f4eed YW |
1171 | assert(fstype); |
1172 | assert(entry_path); | |
1173 | assert(bind_source); | |
553e3660 | 1174 | |
b18f4eed YW |
1175 | (void) mkdir_p_label(entry_path, 0755); |
1176 | n = umount_recursive(entry_path, /* flags = */ 0); | |
553e3660 | 1177 | |
b18f4eed YW |
1178 | r = mount_nofollow_verbose(LOG_DEBUG, fstype, entry_path, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts); |
1179 | if (r == -EINVAL && opts) | |
1180 | /* If this failed with EINVAL then this likely means the textual hidepid= stuff for procfs is | |
1181 | * not supported by the kernel, and thus the per-instance hidepid= neither, which means we | |
1182 | * really don't want to use it, since it would affect our host's /proc mount. Hence let's | |
1183 | * gracefully fallback to a classic, unrestricted version. */ | |
1184 | r = mount_nofollow_verbose(LOG_DEBUG, fstype, entry_path, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, /* opts = */ NULL); | |
553e3660 | 1185 | if (ERRNO_IS_NEG_PRIVILEGE(r)) { |
b18f4eed YW |
1186 | /* When we do not have enough privileges to mount a new instance, fall back to use an |
1187 | * existing mount. */ | |
553e3660 YW |
1188 | |
1189 | if (n > 0) | |
b18f4eed | 1190 | /* The mount or some of sub-mounts are umounted in the above. Refuse incomplete tree. |
553e3660 YW |
1191 | * Propagate the original error code returned by mount() in the above. */ |
1192 | return r; | |
1193 | ||
b18f4eed YW |
1194 | r = path_is_mount_point(entry_path, /* root = */ NULL, /* flags = */ 0); |
1195 | if (r < 0) | |
1196 | return log_debug_errno(r, "Unable to determine whether '%s' is already mounted: %m", entry_path); | |
1197 | if (r > 0) | |
1198 | return 0; /* Use the current mount as is. */ | |
1199 | ||
1200 | /* We lack permissions to mount a new instance, and it is not already mounted. But we can | |
1201 | * access the host's, so as a final fallback bind-mount it to the destination, as most likely | |
1202 | * we are inside a user manager in an unprivileged user namespace. */ | |
1203 | return mount_nofollow_verbose(LOG_DEBUG, bind_source, entry_path, /* fstype = */ NULL, MS_BIND|MS_REC, /* opts = */ NULL); | |
553e3660 YW |
1204 | |
1205 | } else if (r < 0) | |
1206 | return r; | |
1207 | ||
b18f4eed YW |
1208 | /* We mounted a new instance now. Let's bind mount the children over now. This matters for nspawn |
1209 | * where a bunch of files are overmounted, in particular the boot id. */ | |
1210 | (void) bind_mount_submounts(bind_source, entry_path); | |
553e3660 YW |
1211 | return 0; |
1212 | } | |
1213 | ||
b18f4eed YW |
1214 | static int mount_private_sysfs(const MountEntry *m) { |
1215 | assert(m); | |
1216 | return mount_private_apivfs("sysfs", mount_entry_path(m), "/sys", /* opts = */ NULL); | |
1217 | } | |
1218 | ||
79d956db | 1219 | static int mount_procfs(const MountEntry *m, const NamespaceParameters *p) { |
61f8a7bd | 1220 | _cleanup_free_ char *opts = NULL; |
5d997827 LP |
1221 | |
1222 | assert(m); | |
79d956db | 1223 | assert(p); |
5d997827 | 1224 | |
79d956db LP |
1225 | if (p->protect_proc != PROTECT_PROC_DEFAULT || |
1226 | p->proc_subset != PROC_SUBSET_ALL) { | |
4e399953 LP |
1227 | |
1228 | /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it | |
1229 | * pretended to be per-instance but actually was per-namespace), hence let's make use of it | |
1230 | * if requested. To make sure this logic succeeds only on kernels where hidepid= is | |
1231 | * per-instance, we'll exclusively use the textual value for hidepid=, since support was | |
1232 | * added in the same commit: if it's supported it is thus also per-instance. */ | |
1233 | ||
79d956db | 1234 | const char *hpv = p->protect_proc == PROTECT_PROC_DEFAULT ? |
7c76e181 | 1235 | "off" : |
79d956db | 1236 | protect_proc_to_string(p->protect_proc); |
1c265fcd DDM |
1237 | |
1238 | /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in | |
1239 | * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when | |
1240 | * trying to use hidepid= on systems where it isn't supported. The same applies for subset=. | |
1241 | * fsopen()/fsconfig() was also backported on some distros which allows us to detect | |
1242 | * hidepid=/subset= support in even more scenarios. */ | |
1243 | ||
117e7034 | 1244 | if (mount_option_supported("proc", "hidepid", hpv) != 0) { |
1c265fcd DDM |
1245 | opts = strjoin("hidepid=", hpv); |
1246 | if (!opts) | |
1247 | return -ENOMEM; | |
1248 | } | |
1249 | ||
79d956db | 1250 | if (p->proc_subset == PROC_SUBSET_PID && |
117e7034 | 1251 | mount_option_supported("proc", "subset", "pid") != 0) |
1c265fcd DDM |
1252 | if (!strextend_with_separator(&opts, ",", "subset=pid")) |
1253 | return -ENOMEM; | |
4e399953 LP |
1254 | } |
1255 | ||
61f8a7bd YW |
1256 | /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in |
1257 | * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by | |
1258 | * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything | |
1259 | * mounted on /proc/ first. */ | |
b18f4eed | 1260 | return mount_private_apivfs("proc", mount_entry_path(m), "/proc", opts); |
5d997827 LP |
1261 | } |
1262 | ||
2abd4e38 | 1263 | static int mount_tmpfs(const MountEntry *m) { |
df6b900a | 1264 | const char *entry_path, *inner_path; |
abad72be | 1265 | int r; |
abad72be | 1266 | |
6c47cd7d LP |
1267 | assert(m); |
1268 | ||
df6b900a | 1269 | entry_path = mount_entry_path(m); |
809ceb82 | 1270 | inner_path = mount_entry_unprefixed_path(m); |
df6b900a | 1271 | |
d73020f2 LP |
1272 | /* First, get rid of everything that is below if there is anything. Then, overmount with our new |
1273 | * tmpfs */ | |
6c47cd7d | 1274 | |
abad72be CG |
1275 | (void) mkdir_p_label(entry_path, 0755); |
1276 | (void) umount_recursive(entry_path, 0); | |
6c47cd7d | 1277 | |
21935150 LP |
1278 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", entry_path, "tmpfs", m->flags, mount_entry_options(m)); |
1279 | if (r < 0) | |
1280 | return r; | |
abad72be | 1281 | |
03bc11d1 | 1282 | r = label_fix_full(AT_FDCWD, entry_path, inner_path, 0); |
abad72be | 1283 | if (r < 0) |
df6b900a | 1284 | return log_debug_errno(r, "Failed to fix label of '%s' as '%s': %m", entry_path, inner_path); |
6c47cd7d | 1285 | |
51bf9b1a | 1286 | return 0; |
6c47cd7d LP |
1287 | } |
1288 | ||
94293d65 LB |
1289 | static int mount_run(const MountEntry *m) { |
1290 | int r; | |
1291 | ||
1292 | assert(m); | |
1293 | ||
1294 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); | |
1295 | if (r < 0 && r != -ENOENT) | |
1296 | return log_debug_errno(r, "Unable to determine whether /run is already mounted: %m"); | |
1297 | if (r > 0) /* make this a NOP if /run is already a mount point */ | |
1298 | return 0; | |
1299 | ||
1300 | return mount_tmpfs(m); | |
1301 | } | |
1302 | ||
80271a44 XR |
1303 | static int mount_mqueuefs(const MountEntry *m) { |
1304 | int r; | |
1305 | const char *entry_path; | |
1306 | ||
1307 | assert(m); | |
1308 | ||
1309 | entry_path = mount_entry_path(m); | |
1310 | ||
1311 | (void) mkdir_p_label(entry_path, 0755); | |
1312 | (void) umount_recursive(entry_path, 0); | |
1313 | ||
1314 | r = mount_nofollow_verbose(LOG_DEBUG, "mqueue", entry_path, "mqueue", m->flags, mount_entry_options(m)); | |
1315 | if (r < 0) | |
1316 | return r; | |
1317 | ||
1318 | return 0; | |
1319 | } | |
1320 | ||
84be0c71 LP |
1321 | static int mount_image( |
1322 | const MountEntry *m, | |
1323 | const char *root_directory, | |
1324 | const ImagePolicy *image_policy) { | |
93f59701 LB |
1325 | |
1326 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
5897469a LB |
1327 | *host_os_release_sysext_level = NULL, *host_os_release_confext_level = NULL, |
1328 | *extension_name = NULL; | |
b3d13314 LB |
1329 | int r; |
1330 | ||
89e62e0b LP |
1331 | assert(m); |
1332 | ||
55ea4ef0 MG |
1333 | r = path_extract_filename(mount_entry_source(m), &extension_name); |
1334 | if (r < 0) | |
1335 | return log_debug_errno(r, "Failed to extract extension name from %s: %m", mount_entry_source(m)); | |
1336 | ||
93f59701 LB |
1337 | if (m->mode == EXTENSION_IMAGES) { |
1338 | r = parse_os_release( | |
1339 | empty_to_root(root_directory), | |
1340 | "ID", &host_os_release_id, | |
1341 | "VERSION_ID", &host_os_release_version_id, | |
5897469a LB |
1342 | image_class_info[IMAGE_SYSEXT].level_env, &host_os_release_sysext_level, |
1343 | image_class_info[IMAGE_CONFEXT].level_env, &host_os_release_confext_level, | |
93f59701 LB |
1344 | NULL); |
1345 | if (r < 0) | |
1346 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
78ab2b50 LB |
1347 | if (isempty(host_os_release_id)) |
1348 | 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 |
1349 | } |
1350 | ||
1351 | r = verity_dissect_and_mount( | |
84be0c71 LP |
1352 | /* src_fd= */ -1, |
1353 | mount_entry_source(m), | |
1354 | mount_entry_path(m), | |
1355 | m->image_options, | |
1356 | image_policy, | |
1357 | host_os_release_id, | |
1358 | host_os_release_version_id, | |
5897469a LB |
1359 | host_os_release_sysext_level, |
1360 | host_os_release_confext_level, | |
3e107272 LB |
1361 | /* required_sysext_scope= */ NULL, |
1362 | /* ret_image= */ NULL); | |
b850a9b2 LB |
1363 | if (r == -ENOENT && m->ignore) |
1364 | return 0; | |
db4c8a25 | 1365 | if (r == -ESTALE && host_os_release_id) |
7fcd1343 | 1366 | return log_error_errno(r, // FIXME: this should not be logged ad LOG_ERR, as it will result in duplicate logging. |
5897469a | 1367 | "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s%s%s", |
db4c8a25 LB |
1368 | mount_entry_source(m), |
1369 | host_os_release_id, | |
1370 | host_os_release_version_id ? " VERSION_ID=" : "", | |
1371 | strempty(host_os_release_version_id), | |
5897469a LB |
1372 | host_os_release_sysext_level ? image_class_info[IMAGE_SYSEXT].level_env_print : "", |
1373 | strempty(host_os_release_sysext_level), | |
1374 | host_os_release_confext_level ? image_class_info[IMAGE_CONFEXT].level_env_print : "", | |
1375 | strempty(host_os_release_confext_level)); | |
b3d13314 | 1376 | if (r < 0) |
4beda316 | 1377 | return log_debug_errno(r, "Failed to mount image %s on %s: %m", mount_entry_source(m), mount_entry_path(m)); |
b3d13314 | 1378 | |
51bf9b1a | 1379 | return 0; |
b3d13314 LB |
1380 | } |
1381 | ||
93f59701 LB |
1382 | static int mount_overlay(const MountEntry *m) { |
1383 | const char *options; | |
1384 | int r; | |
1385 | ||
1386 | assert(m); | |
1387 | ||
1388 | options = strjoina("lowerdir=", mount_entry_options(m)); | |
1389 | ||
1390 | (void) mkdir_p_label(mount_entry_path(m), 0755); | |
1391 | ||
1392 | r = mount_nofollow_verbose(LOG_DEBUG, "overlay", mount_entry_path(m), "overlay", MS_RDONLY, options); | |
1393 | if (r == -ENOENT && m->ignore) | |
1394 | return 0; | |
93f59701 | 1395 | |
51bf9b1a | 1396 | return r; |
93f59701 LB |
1397 | } |
1398 | ||
088696fe | 1399 | static int follow_symlink( |
d2d6c096 | 1400 | const char *root_directory, |
088696fe | 1401 | MountEntry *m) { |
d2d6c096 | 1402 | |
088696fe | 1403 | _cleanup_free_ char *target = NULL; |
8fceda93 LP |
1404 | int r; |
1405 | ||
088696fe LP |
1406 | /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we |
1407 | * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at | |
1408 | * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the | |
1409 | * end and already have a fully normalized name. */ | |
8fceda93 | 1410 | |
f461a28d | 1411 | r = chase(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL); |
088696fe LP |
1412 | if (r < 0) |
1413 | return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m)); | |
1414 | if (r > 0) /* Reached the end, nothing more to resolve */ | |
1415 | return 1; | |
8fceda93 | 1416 | |
f461a28d | 1417 | if (m->n_followed >= CHASE_MAX) /* put a boundary on things */ |
baaa35ad ZJS |
1418 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1419 | "Symlink loop on '%s'.", | |
1420 | mount_entry_path(m)); | |
8fceda93 | 1421 | |
e2341b6b DT |
1422 | log_debug("Followed mount entry path symlink %s %s %s.", |
1423 | mount_entry_path(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), target); | |
8fceda93 | 1424 | |
809ceb82 | 1425 | mount_entry_consume_prefix(m, TAKE_PTR(target)); |
8fceda93 | 1426 | |
088696fe LP |
1427 | m->n_followed ++; |
1428 | ||
1429 | return 0; | |
8fceda93 LP |
1430 | } |
1431 | ||
82fb2da2 | 1432 | static int apply_one_mount( |
8fceda93 | 1433 | const char *root_directory, |
4e399953 | 1434 | MountEntry *m, |
79d956db | 1435 | const NamespaceParameters *p) { |
ac0930c8 | 1436 | |
e5f10caf | 1437 | _cleanup_free_ char *inaccessible = NULL; |
a227a4be | 1438 | bool rbind = true, make = false; |
15ae422b | 1439 | const char *what; |
15ae422b | 1440 | int r; |
15ae422b | 1441 | |
c17ec25e | 1442 | assert(m); |
79d956db | 1443 | assert(p); |
15ae422b | 1444 | |
34de407a | 1445 | log_debug("Applying namespace mount on %s", mount_entry_path(m)); |
fe3c2583 | 1446 | |
c17ec25e | 1447 | switch (m->mode) { |
15ae422b | 1448 | |
160cfdbe | 1449 | case INACCESSIBLE: { |
cd7f3702 | 1450 | _cleanup_free_ char *runtime_dir = NULL; |
160cfdbe | 1451 | struct stat target; |
6d313367 LP |
1452 | |
1453 | /* First, get rid of everything that is below if there | |
1454 | * is anything... Then, overmount it with an | |
c4b41707 | 1455 | * inaccessible path. */ |
34de407a | 1456 | (void) umount_recursive(mount_entry_path(m), 0); |
6d313367 | 1457 | |
088696fe LP |
1458 | if (lstat(mount_entry_path(m), &target) < 0) { |
1459 | if (errno == ENOENT && m->ignore) | |
1460 | return 0; | |
1461 | ||
cbc056c8 ZJS |
1462 | return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m", |
1463 | mount_entry_path(m)); | |
088696fe | 1464 | } |
15ae422b | 1465 | |
cd7f3702 DDM |
1466 | /* We don't pass the literal runtime scope through here but one based purely on our UID. This |
1467 | * means that the root user's --user services will use the host's inaccessible inodes rather | |
1468 | * then root's private ones. This is preferable since it means device nodes that are | |
1469 | * overmounted to make them inaccessible will be overmounted with a device node, rather than | |
1470 | * an AF_UNIX socket inode. */ | |
1471 | runtime_dir = settle_runtime_dir(geteuid() == 0 ? RUNTIME_SCOPE_SYSTEM : RUNTIME_SCOPE_USER); | |
1472 | if (!runtime_dir) | |
1473 | return log_oom_debug(); | |
e5f10caf AZ |
1474 | |
1475 | r = mode_to_inaccessible_node(runtime_dir, target.st_mode, &inaccessible); | |
1476 | if (r < 0) | |
baaa35ad ZJS |
1477 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1478 | "File type not supported for inaccessible mounts. Note that symlinks are not allowed"); | |
e5f10caf | 1479 | what = inaccessible; |
c4b41707 | 1480 | break; |
160cfdbe | 1481 | } |
fe3c2583 | 1482 | |
15ae422b | 1483 | case READONLY: |
15ae422b | 1484 | case READWRITE: |
1e05071d | 1485 | case READWRITE_IMPLICIT: |
ddc155b2 TM |
1486 | case EXEC: |
1487 | case NOEXEC: | |
8fceda93 | 1488 | r = path_is_mount_point(mount_entry_path(m), root_directory, 0); |
088696fe LP |
1489 | if (r == -ENOENT && m->ignore) |
1490 | return 0; | |
d944dc95 | 1491 | if (r < 0) |
cbc056c8 ZJS |
1492 | return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m", |
1493 | mount_entry_path(m)); | |
1494 | if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY | |
ddc155b2 | 1495 | * and MS_NOEXEC bits for the mount point if needed. */ |
6b7c9f8b | 1496 | return 0; |
6b7c9f8b | 1497 | /* This isn't a mount point yet, let's make it one. */ |
34de407a | 1498 | what = mount_entry_path(m); |
6b7c9f8b | 1499 | break; |
15ae422b | 1500 | |
a07b9926 LB |
1501 | case EXTENSION_DIRECTORIES: { |
1502 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
55ea4ef0 | 1503 | *host_os_release_level = NULL, *extension_name = NULL; |
a07b9926 | 1504 | _cleanup_strv_free_ char **extension_release = NULL; |
55ea4ef0 | 1505 | ImageClass class = IMAGE_SYSEXT; |
a07b9926 LB |
1506 | |
1507 | r = path_extract_filename(mount_entry_source(m), &extension_name); | |
1508 | if (r < 0) | |
1509 | return log_debug_errno(r, "Failed to extract extension name from %s: %m", mount_entry_source(m)); | |
1510 | ||
55ea4ef0 MG |
1511 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_SYSEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); |
1512 | if (r == -ENOENT) { | |
1513 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_CONFEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); | |
1514 | if (r >= 0) | |
1515 | class = IMAGE_CONFEXT; | |
1516 | } | |
5897469a LB |
1517 | if (r < 0) |
1518 | return log_debug_errno(r, "Failed to acquire 'extension-release' data of extension tree %s: %m", mount_entry_source(m)); | |
55ea4ef0 | 1519 | |
a07b9926 LB |
1520 | r = parse_os_release( |
1521 | empty_to_root(root_directory), | |
1522 | "ID", &host_os_release_id, | |
1523 | "VERSION_ID", &host_os_release_version_id, | |
55ea4ef0 | 1524 | image_class_info[class].level_env, &host_os_release_level, |
a07b9926 LB |
1525 | NULL); |
1526 | if (r < 0) | |
1527 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
1528 | if (isempty(host_os_release_id)) | |
1529 | 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)); | |
1530 | ||
55ea4ef0 | 1531 | r = load_extension_release_pairs(mount_entry_source(m), class, extension_name, /* relax_extension_release_check= */ false, &extension_release); |
a07b9926 LB |
1532 | if (r == -ENOENT && m->ignore) |
1533 | return 0; | |
1534 | if (r < 0) | |
1535 | return log_debug_errno(r, "Failed to parse directory %s extension-release metadata: %m", extension_name); | |
1536 | ||
1537 | r = extension_release_validate( | |
1538 | extension_name, | |
1539 | host_os_release_id, | |
1540 | host_os_release_version_id, | |
55ea4ef0 MG |
1541 | host_os_release_level, |
1542 | /* host_extension_scope */ NULL, /* Leave empty, we need to accept both system and portable */ | |
30dfe035 | 1543 | extension_release, |
55ea4ef0 | 1544 | class); |
a07b9926 LB |
1545 | if (r == 0) |
1546 | return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Directory %s extension-release metadata does not match the root's", extension_name); | |
1547 | if (r < 0) | |
1548 | return log_debug_errno(r, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name); | |
1549 | ||
1550 | _fallthrough_; | |
1551 | } | |
1552 | ||
d2d6c096 LP |
1553 | case BIND_MOUNT: |
1554 | rbind = false; | |
d2d6c096 | 1555 | |
4831981d | 1556 | _fallthrough_; |
088696fe LP |
1557 | case BIND_MOUNT_RECURSIVE: { |
1558 | _cleanup_free_ char *chased = NULL; | |
5d997827 | 1559 | |
cbc056c8 ZJS |
1560 | /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note |
1561 | * that bind mount source paths are always relative to the host root, hence we pass NULL as | |
f461a28d | 1562 | * root directory to chase() here. */ |
088696fe | 1563 | |
f461a28d | 1564 | r = chase(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL); |
088696fe LP |
1565 | if (r == -ENOENT && m->ignore) { |
1566 | log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m)); | |
1567 | return 0; | |
1568 | } | |
1569 | if (r < 0) | |
1570 | return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m)); | |
1571 | ||
e2341b6b DT |
1572 | log_debug("Followed source symlinks %s %s %s.", |
1573 | mount_entry_source(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), chased); | |
088696fe LP |
1574 | |
1575 | free_and_replace(m->source_malloc, chased); | |
d2d6c096 LP |
1576 | |
1577 | what = mount_entry_source(m); | |
a227a4be | 1578 | make = true; |
d2d6c096 | 1579 | break; |
088696fe | 1580 | } |
d2d6c096 | 1581 | |
6c47cd7d | 1582 | case EMPTY_DIR: |
2abd4e38 YW |
1583 | case TMPFS: |
1584 | return mount_tmpfs(m); | |
6c47cd7d | 1585 | |
ac0930c8 | 1586 | case PRIVATE_TMP: |
56a13a49 | 1587 | case PRIVATE_TMP_READONLY: |
89bd586c | 1588 | what = mount_entry_source(m); |
a227a4be | 1589 | make = true; |
15ae422b | 1590 | break; |
e364ad06 | 1591 | |
d6797c92 | 1592 | case PRIVATE_DEV: |
79d956db | 1593 | return mount_private_dev(m, p->runtime_scope); |
5d997827 LP |
1594 | |
1595 | case BIND_DEV: | |
1596 | return mount_bind_dev(m); | |
1597 | ||
277b269e YW |
1598 | case PRIVATE_SYSFS: |
1599 | return mount_private_sysfs(m); | |
1600 | ||
10028263 YW |
1601 | case BIND_SYSFS: |
1602 | return mount_bind_sysfs(m); | |
5d997827 LP |
1603 | |
1604 | case PROCFS: | |
79d956db | 1605 | return mount_procfs(m, p); |
d6797c92 | 1606 | |
94293d65 LB |
1607 | case RUN: |
1608 | return mount_run(m); | |
1609 | ||
80271a44 XR |
1610 | case MQUEUEFS: |
1611 | return mount_mqueuefs(m); | |
1612 | ||
b3d13314 | 1613 | case MOUNT_IMAGES: |
79d956db | 1614 | return mount_image(m, NULL, p->mount_image_policy); |
93f59701 LB |
1615 | |
1616 | case EXTENSION_IMAGES: | |
79d956db | 1617 | return mount_image(m, root_directory, p->extension_image_policy); |
93f59701 LB |
1618 | |
1619 | case OVERLAY_MOUNT: | |
1620 | return mount_overlay(m); | |
b3d13314 | 1621 | |
e364ad06 | 1622 | default: |
04499a70 | 1623 | assert_not_reached(); |
15ae422b LP |
1624 | } |
1625 | ||
ac0930c8 | 1626 | assert(what); |
15ae422b | 1627 | |
21935150 LP |
1628 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); |
1629 | if (r < 0) { | |
a227a4be | 1630 | bool try_again = false; |
a227a4be LP |
1631 | |
1632 | if (r == -ENOENT && make) { | |
8bab8029 | 1633 | int q; |
a227a4be | 1634 | |
cbc056c8 ZJS |
1635 | /* Hmm, either the source or the destination are missing. Let's see if we can create |
1636 | the destination, then try again. */ | |
a227a4be | 1637 | |
8bab8029 | 1638 | (void) mkdir_parents(mount_entry_path(m), 0755); |
a227a4be | 1639 | |
8bab8029 | 1640 | q = make_mount_point_inode_from_path(what, mount_entry_path(m), 0755); |
283eb4fd | 1641 | if (q < 0) { |
7fcd1343 | 1642 | if (q != -EEXIST) // FIXME: this shouldn't be logged at LOG_WARNING, but be bubbled up, and logged there to avoid duplicate logging |
283eb4fd LP |
1643 | log_warning_errno(q, "Failed to create destination mount point node '%s', ignoring: %m", |
1644 | mount_entry_path(m)); | |
1645 | } else | |
8bab8029 | 1646 | try_again = true; |
a227a4be LP |
1647 | } |
1648 | ||
21935150 LP |
1649 | if (try_again) |
1650 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); | |
a227a4be | 1651 | if (r < 0) |
7fcd1343 | 1652 | return log_error_errno(r, "Failed to mount %s to %s: %m", what, mount_entry_path(m)); // FIXME: this should not be logged here, but be bubbled up, to avoid duplicate logging |
a227a4be | 1653 | } |
6b7c9f8b | 1654 | |
34de407a | 1655 | log_debug("Successfully mounted %s to %s", what, mount_entry_path(m)); |
6b7c9f8b | 1656 | return 0; |
ac0930c8 | 1657 | } |
15ae422b | 1658 | |
6b000af4 | 1659 | static int make_read_only(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
9ce4e4b0 | 1660 | unsigned long new_flags = 0, flags_mask = 0; |
57ccd9f6 YW |
1661 | bool submounts; |
1662 | int r; | |
15ae422b | 1663 | |
c17ec25e | 1664 | assert(m); |
ac9de0b3 | 1665 | assert(proc_self_mountinfo); |
ac0930c8 | 1666 | |
9ce4e4b0 LP |
1667 | if (mount_entry_read_only(m) || m->mode == PRIVATE_DEV) { |
1668 | new_flags |= MS_RDONLY; | |
1669 | flags_mask |= MS_RDONLY; | |
1670 | } | |
1671 | ||
1672 | if (m->nosuid) { | |
1673 | new_flags |= MS_NOSUID; | |
1674 | flags_mask |= MS_NOSUID; | |
1675 | } | |
1676 | ||
1677 | if (flags_mask == 0) /* No Change? */ | |
6b7c9f8b LP |
1678 | return 0; |
1679 | ||
9ce4e4b0 LP |
1680 | /* We generally apply these changes recursively, except for /dev, and the cases we know there's |
1681 | * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the | |
1682 | * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace | |
1683 | * and running Linux <= 4.17. */ | |
1684 | submounts = | |
1685 | mount_entry_read_only(m) && | |
1686 | !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1687 | if (submounts) | |
6b000af4 | 1688 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); |
9ce4e4b0 | 1689 | else |
7cce68e1 | 1690 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); |
9ce4e4b0 | 1691 | |
ddc155b2 | 1692 | /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked |
867189b5 LP |
1693 | * read-only already stays this way. This improves compatibility with container managers, where we |
1694 | * won't attempt to undo read-only mounts already applied. */ | |
ac0930c8 | 1695 | |
8fceda93 | 1696 | if (r == -ENOENT && m->ignore) |
867189b5 | 1697 | return 0; |
763a260a | 1698 | if (r < 0) |
9ce4e4b0 | 1699 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), |
763a260a | 1700 | submounts ? " and its submounts" : ""); |
763a260a | 1701 | return 0; |
d944dc95 LP |
1702 | } |
1703 | ||
ddc155b2 TM |
1704 | static int make_noexec(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
1705 | unsigned long new_flags = 0, flags_mask = 0; | |
57ccd9f6 YW |
1706 | bool submounts; |
1707 | int r; | |
ddc155b2 TM |
1708 | |
1709 | assert(m); | |
1710 | assert(proc_self_mountinfo); | |
1711 | ||
1712 | if (mount_entry_noexec(m)) { | |
1713 | new_flags |= MS_NOEXEC; | |
1714 | flags_mask |= MS_NOEXEC; | |
1715 | } else if (mount_entry_exec(m)) { | |
1716 | new_flags &= ~MS_NOEXEC; | |
1717 | flags_mask |= MS_NOEXEC; | |
1718 | } | |
1719 | ||
1720 | if (flags_mask == 0) /* No Change? */ | |
1721 | return 0; | |
1722 | ||
1723 | submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1724 | ||
1725 | if (submounts) | |
1726 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); | |
1727 | else | |
1728 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); | |
1729 | ||
1730 | if (r == -ENOENT && m->ignore) | |
1731 | return 0; | |
1732 | if (r < 0) | |
1733 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1734 | submounts ? " and its submounts" : ""); | |
1735 | return 0; | |
1736 | } | |
1737 | ||
6720e356 | 1738 | static int make_nosuid(const MountEntry *m, FILE *proc_self_mountinfo) { |
57ccd9f6 YW |
1739 | bool submounts; |
1740 | int r; | |
6720e356 YW |
1741 | |
1742 | assert(m); | |
1743 | assert(proc_self_mountinfo); | |
1744 | ||
1745 | submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1746 | ||
1747 | if (submounts) | |
1748 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, NULL, proc_self_mountinfo); | |
1749 | else | |
1750 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, proc_self_mountinfo); | |
1751 | if (r == -ENOENT && m->ignore) | |
1752 | return 0; | |
1753 | if (r < 0) | |
1754 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1755 | submounts ? " and its submounts" : ""); | |
1756 | return 0; | |
1757 | } | |
1758 | ||
79d956db LP |
1759 | static bool namespace_parameters_mount_apivfs(const NamespaceParameters *p) { |
1760 | assert(p); | |
5d997827 | 1761 | |
9c988f93 DH |
1762 | /* |
1763 | * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=, | |
1764 | * since to protect the API VFS mounts, they need to be around in the | |
9b68367b | 1765 | * first place... |
9c988f93 | 1766 | */ |
5d997827 | 1767 | |
79d956db LP |
1768 | return p->mount_apivfs || |
1769 | p->protect_control_groups || | |
1770 | p->protect_kernel_tunables || | |
1771 | p->protect_proc != PROTECT_PROC_DEFAULT || | |
1772 | p->proc_subset != PROC_SUBSET_ALL; | |
5d997827 LP |
1773 | } |
1774 | ||
fbf90c0d CB |
1775 | /* Walk all mount entries and dropping any unused mounts. This affects all |
1776 | * mounts: | |
1777 | * - that are implicitly protected by a path that has been rendered inaccessible | |
1778 | * - whose immediate parent requests the same protection mode as the mount itself | |
1779 | * - that are outside of the relevant root directory | |
1780 | * - which are duplicates | |
1781 | */ | |
063c977a LP |
1782 | static void drop_unused_mounts(MountList *ml, const char *root_directory) { |
1783 | assert(ml); | |
9b68367b | 1784 | assert(root_directory); |
f8b64b57 | 1785 | |
063c977a LP |
1786 | assert(ml->mounts || ml->n_mounts == 0); |
1787 | ||
1788 | typesafe_qsort(ml->mounts, ml->n_mounts, mount_path_compare); | |
f8b64b57 | 1789 | |
063c977a LP |
1790 | drop_duplicates(ml); |
1791 | drop_outside_root(ml, root_directory); | |
1792 | drop_inaccessible(ml); | |
1793 | drop_nop(ml); | |
f8b64b57 LP |
1794 | } |
1795 | ||
df61e79a | 1796 | static int create_symlinks_from_tuples(const char *root, char **strv_symlinks) { |
df61e79a LB |
1797 | int r; |
1798 | ||
1799 | STRV_FOREACH_PAIR(src, dst, strv_symlinks) { | |
1800 | _cleanup_free_ char *src_abs = NULL, *dst_abs = NULL; | |
1801 | ||
1802 | src_abs = path_join(root, *src); | |
1803 | dst_abs = path_join(root, *dst); | |
1804 | if (!src_abs || !dst_abs) | |
1805 | return -ENOMEM; | |
1806 | ||
1807 | r = mkdir_parents_label(dst_abs, 0755); | |
1808 | if (r < 0) | |
15461b7f LB |
1809 | return log_debug_errno( |
1810 | r, | |
1811 | "Failed to create parent directory for symlink '%s': %m", | |
1812 | dst_abs); | |
df61e79a LB |
1813 | |
1814 | r = symlink_idempotent(src_abs, dst_abs, true); | |
1815 | if (r < 0) | |
15461b7f LB |
1816 | return log_debug_errno( |
1817 | r, | |
1818 | "Failed to create symlink from '%s' to '%s': %m", | |
1819 | src_abs, | |
1820 | dst_abs); | |
df61e79a LB |
1821 | } |
1822 | ||
1823 | return 0; | |
1824 | } | |
1825 | ||
13b3af4a LB |
1826 | static void mount_entry_path_debug_string(const char *root, MountEntry *m, char **error_path) { |
1827 | assert(m); | |
1828 | ||
1829 | /* Create a string suitable for debugging logs, stripping for example the local working directory. | |
1830 | * For example, with a BindPaths=/var/bar that does not exist on the host: | |
1831 | * | |
1832 | * Before: | |
1833 | * foo.service: Failed to set up mount namespacing: /run/systemd/unit-root/var/bar: No such file or directory | |
1834 | * After: | |
1835 | * foo.service: Failed to set up mount namespacing: /var/bar: No such file or directory | |
1836 | * | |
1837 | * Note that this is an error path, so no OOM check is done on purpose. */ | |
1838 | ||
1839 | if (!error_path) | |
1840 | return; | |
1841 | ||
1842 | if (!mount_entry_path(m)) { | |
1843 | *error_path = NULL; | |
1844 | return; | |
1845 | } | |
1846 | ||
1847 | if (root) { | |
1848 | const char *e = startswith(mount_entry_path(m), root); | |
1849 | if (e) { | |
1850 | *error_path = strdup(e); | |
1851 | return; | |
1852 | } | |
1853 | } | |
1854 | ||
1855 | *error_path = strdup(mount_entry_path(m)); | |
1856 | return; | |
1857 | } | |
1858 | ||
82fb2da2 | 1859 | static int apply_mounts( |
063c977a | 1860 | MountList *ml, |
82fb2da2 | 1861 | const char *root, |
79d956db | 1862 | const NamespaceParameters *p, |
82fb2da2 LB |
1863 | char **error_path) { |
1864 | ||
1865 | _cleanup_fclose_ FILE *proc_self_mountinfo = NULL; | |
1866 | _cleanup_free_ char **deny_list = NULL; | |
82fb2da2 LB |
1867 | int r; |
1868 | ||
063c977a | 1869 | assert(ml); |
82fb2da2 | 1870 | assert(root); |
063c977a LP |
1871 | assert(p); |
1872 | ||
1873 | if (ml->n_mounts == 0) /* Shortcut: nothing to do */ | |
1874 | return 0; | |
82fb2da2 LB |
1875 | |
1876 | /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of | |
1877 | * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */ | |
1878 | proc_self_mountinfo = fopen("/proc/self/mountinfo", "re"); | |
1879 | if (!proc_self_mountinfo) { | |
d60e3b40 LP |
1880 | r = -errno; |
1881 | ||
82fb2da2 LB |
1882 | if (error_path) |
1883 | *error_path = strdup("/proc/self/mountinfo"); | |
d60e3b40 LP |
1884 | |
1885 | return log_debug_errno(r, "Failed to open /proc/self/mountinfo: %m"); | |
82fb2da2 LB |
1886 | } |
1887 | ||
1888 | /* First round, establish all mounts we need */ | |
1889 | for (;;) { | |
1890 | bool again = false; | |
1891 | ||
063c977a | 1892 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
82fb2da2 LB |
1893 | |
1894 | if (m->applied) | |
1895 | continue; | |
1896 | ||
a07b9926 LB |
1897 | /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */ |
1898 | r = follow_symlink(!IN_SET(m->mode, EXTENSION_IMAGES, EXTENSION_DIRECTORIES) ? root : NULL, m); | |
82fb2da2 | 1899 | if (r < 0) { |
13b3af4a | 1900 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
1901 | return r; |
1902 | } | |
1903 | if (r == 0) { | |
1904 | /* We hit a symlinked mount point. The entry got rewritten and might | |
1905 | * point to a very different place now. Let's normalize the changed | |
1906 | * list, and start from the beginning. After all to mount the entry | |
1907 | * at the new location we might need some other mounts first */ | |
1908 | again = true; | |
1909 | break; | |
1910 | } | |
1911 | ||
79d956db | 1912 | r = apply_one_mount(root, m, p); |
82fb2da2 | 1913 | if (r < 0) { |
13b3af4a | 1914 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
1915 | return r; |
1916 | } | |
1917 | ||
1918 | m->applied = true; | |
1919 | } | |
1920 | ||
1921 | if (!again) | |
1922 | break; | |
1923 | ||
063c977a | 1924 | drop_unused_mounts(ml, root); |
82fb2da2 LB |
1925 | } |
1926 | ||
df61e79a | 1927 | /* Now that all filesystems have been set up, but before the |
663e2756 | 1928 | * read-only switches are flipped, create the exec dirs and other symlinks. |
df61e79a LB |
1929 | * Note that when /var/lib is not empty/tmpfs, these symlinks will already |
1930 | * exist, which means this will be a no-op. */ | |
79d956db | 1931 | r = create_symlinks_from_tuples(root, p->symlinks); |
df61e79a | 1932 | if (r < 0) |
663e2756 | 1933 | return log_debug_errno(r, "Failed to set up symlinks inside mount namespace: %m"); |
df61e79a | 1934 | |
82fb2da2 | 1935 | /* Create a deny list we can pass to bind_mount_recursive() */ |
063c977a | 1936 | deny_list = new(char*, ml->n_mounts+1); |
82fb2da2 LB |
1937 | if (!deny_list) |
1938 | return -ENOMEM; | |
063c977a LP |
1939 | for (size_t j = 0; j < ml->n_mounts; j++) |
1940 | deny_list[j] = (char*) mount_entry_path(ml->mounts+j); | |
1941 | deny_list[ml->n_mounts] = NULL; | |
82fb2da2 LB |
1942 | |
1943 | /* Second round, flip the ro bits if necessary. */ | |
063c977a | 1944 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
82fb2da2 LB |
1945 | r = make_read_only(m, deny_list, proc_self_mountinfo); |
1946 | if (r < 0) { | |
13b3af4a | 1947 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
1948 | return r; |
1949 | } | |
1950 | } | |
1951 | ||
1952 | /* Third round, flip the noexec bits with a simplified deny list. */ | |
063c977a LP |
1953 | for (size_t j = 0; j < ml->n_mounts; j++) |
1954 | if (IN_SET((ml->mounts+j)->mode, EXEC, NOEXEC)) | |
1955 | deny_list[j] = (char*) mount_entry_path(ml->mounts+j); | |
1956 | deny_list[ml->n_mounts] = NULL; | |
82fb2da2 | 1957 | |
063c977a | 1958 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
82fb2da2 LB |
1959 | r = make_noexec(m, deny_list, proc_self_mountinfo); |
1960 | if (r < 0) { | |
13b3af4a | 1961 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
1962 | return r; |
1963 | } | |
1964 | } | |
1965 | ||
6720e356 | 1966 | /* Fourth round, flip the nosuid bits without a deny list. */ |
79d956db | 1967 | if (p->mount_nosuid) |
063c977a | 1968 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
6720e356 YW |
1969 | r = make_nosuid(m, proc_self_mountinfo); |
1970 | if (r < 0) { | |
13b3af4a | 1971 | mount_entry_path_debug_string(root, m, error_path); |
6720e356 YW |
1972 | return r; |
1973 | } | |
1974 | } | |
1975 | ||
82fb2da2 LB |
1976 | return 1; |
1977 | } | |
1978 | ||
c8c535d5 LP |
1979 | static bool root_read_only( |
1980 | char **read_only_paths, | |
1981 | ProtectSystem protect_system) { | |
1982 | ||
1983 | /* Determine whether the root directory is going to be read-only given the configured settings. */ | |
1984 | ||
1985 | if (protect_system == PROTECT_SYSTEM_STRICT) | |
1986 | return true; | |
1987 | ||
de46b2be | 1988 | if (prefixed_path_strv_contains(read_only_paths, "/")) |
c8c535d5 LP |
1989 | return true; |
1990 | ||
1991 | return false; | |
1992 | } | |
1993 | ||
1994 | static bool home_read_only( | |
1995 | char** read_only_paths, | |
1996 | char** inaccessible_paths, | |
1997 | char** empty_directories, | |
1998 | const BindMount *bind_mounts, | |
1999 | size_t n_bind_mounts, | |
2000 | const TemporaryFileSystem *temporary_filesystems, | |
2001 | size_t n_temporary_filesystems, | |
2002 | ProtectHome protect_home) { | |
2003 | ||
c8c535d5 LP |
2004 | /* Determine whether the /home directory is going to be read-only given the configured settings. Yes, |
2005 | * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple | |
2006 | * settings. */ | |
2007 | ||
2008 | if (protect_home != PROTECT_HOME_NO) | |
2009 | return true; | |
2010 | ||
de46b2be TM |
2011 | if (prefixed_path_strv_contains(read_only_paths, "/home") || |
2012 | prefixed_path_strv_contains(inaccessible_paths, "/home") || | |
2013 | prefixed_path_strv_contains(empty_directories, "/home")) | |
c8c535d5 LP |
2014 | return true; |
2015 | ||
8794e6db | 2016 | for (size_t i = 0; i < n_temporary_filesystems; i++) |
c8c535d5 LP |
2017 | if (path_equal(temporary_filesystems[i].path, "/home")) |
2018 | return true; | |
2019 | ||
2020 | /* If /home is overmounted with some dir from the host it's not writable. */ | |
8794e6db | 2021 | for (size_t i = 0; i < n_bind_mounts; i++) |
c8c535d5 LP |
2022 | if (path_equal(bind_mounts[i].destination, "/home")) |
2023 | return true; | |
2024 | ||
2025 | return false; | |
2026 | } | |
2027 | ||
79d956db | 2028 | int setup_namespace(const NamespaceParameters *p, char **error_path) { |
15ae422b | 2029 | |
915e6d16 LP |
2030 | _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL; |
2031 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; | |
93f59701 | 2032 | _cleanup_strv_free_ char **hierarchies = NULL; |
063c977a | 2033 | _cleanup_(mount_list_done) MountList ml = {}; |
79d956db | 2034 | bool require_prefix = false; |
24759d8f | 2035 | const char *root; |
9cb1709b LP |
2036 | DissectImageFlags dissect_image_flags = |
2037 | DISSECT_IMAGE_GENERIC_ROOT | | |
2038 | DISSECT_IMAGE_REQUIRE_ROOT | | |
2039 | DISSECT_IMAGE_DISCARD_ON_LOOP | | |
2040 | DISSECT_IMAGE_RELAX_VAR_CHECK | | |
2041 | DISSECT_IMAGE_FSCK | | |
c65f854a | 2042 | DISSECT_IMAGE_USR_NO_ROOT | |
73d88b80 LP |
2043 | DISSECT_IMAGE_GROWFS | |
2044 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | | |
2045 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
89e62e0b | 2046 | int r; |
15ae422b | 2047 | |
79d956db | 2048 | assert(p); |
915e6d16 | 2049 | |
cdf42f9b LP |
2050 | /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes |
2051 | * we configure take effect */ | |
2052 | BLOCK_WITH_UMASK(0000); | |
2053 | ||
79d956db LP |
2054 | bool setup_propagate = !isempty(p->propagate_dir) && !isempty(p->incoming_dir); |
2055 | unsigned long mount_propagation_flag = p->mount_propagation_flag != 0 ? p->mount_propagation_flag : MS_SHARED; | |
ac0930c8 | 2056 | |
79d956db | 2057 | if (p->root_image) { |
c8c535d5 | 2058 | /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */ |
79d956db LP |
2059 | if (root_read_only(p->read_only_paths, |
2060 | p->protect_system) && | |
2061 | home_read_only(p->read_only_paths, p->inaccessible_paths, p->empty_directories, | |
2062 | p->bind_mounts, p->n_bind_mounts, p->temporary_filesystems, p->n_temporary_filesystems, | |
2063 | p->protect_home) && | |
2064 | strv_isempty(p->read_write_paths)) | |
915e6d16 LP |
2065 | dissect_image_flags |= DISSECT_IMAGE_READ_ONLY; |
2066 | ||
79d956db | 2067 | SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, p->verity && p->verity->data_path); |
915e6d16 | 2068 | |
89e62e0b | 2069 | r = loop_device_make_by_path( |
79d956db | 2070 | p->root_image, |
ef9c184d | 2071 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */, |
22ee78a8 | 2072 | /* sector_size= */ UINT32_MAX, |
89e62e0b | 2073 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 2074 | LOCK_SH, |
89e62e0b | 2075 | &loop_device); |
78ebe980 | 2076 | if (r < 0) |
89e62e0b LP |
2077 | return log_debug_errno(r, "Failed to create loop device for root image: %m"); |
2078 | ||
bad31660 YW |
2079 | r = dissect_loop_device( |
2080 | loop_device, | |
79d956db LP |
2081 | p->verity, |
2082 | p->root_image_options, | |
2083 | p->root_image_policy, | |
89e62e0b LP |
2084 | dissect_image_flags, |
2085 | &dissected_image); | |
78ebe980 | 2086 | if (r < 0) |
763a260a | 2087 | return log_debug_errno(r, "Failed to dissect image: %m"); |
78ebe980 | 2088 | |
88b3300f LP |
2089 | r = dissected_image_load_verity_sig_partition( |
2090 | dissected_image, | |
2091 | loop_device->fd, | |
79d956db | 2092 | p->verity); |
88b3300f LP |
2093 | if (r < 0) |
2094 | return r; | |
2095 | ||
89e62e0b LP |
2096 | r = dissected_image_decrypt( |
2097 | dissected_image, | |
2098 | NULL, | |
79d956db | 2099 | p->verity, |
e330f97a | 2100 | dissect_image_flags); |
915e6d16 | 2101 | if (r < 0) |
763a260a | 2102 | return log_debug_errno(r, "Failed to decrypt dissected image: %m"); |
915e6d16 LP |
2103 | } |
2104 | ||
79d956db LP |
2105 | if (p->root_directory) |
2106 | root = p->root_directory; | |
0722b359 | 2107 | else { |
77f16dbd DDM |
2108 | /* /run/systemd should have been created by PID 1 early on already, but in some cases, like |
2109 | * when running tests (test-execute), it might not have been created yet so let's make sure | |
2110 | * we create it if it doesn't already exist. */ | |
2111 | (void) mkdir_p_label("/run/systemd", 0755); | |
2112 | ||
d73020f2 LP |
2113 | /* Always create the mount namespace in a temporary directory, instead of operating directly |
2114 | * in the root. The temporary directory prevents any mounts from being potentially obscured | |
2115 | * my other mounts we already applied. We use the same mount point for all images, which is | |
2116 | * safe, since they all live in their own namespaces after all, and hence won't see each | |
be79c235 LP |
2117 | * other. (Note: this directory is also created by PID 1 early on, we create it here for |
2118 | * similar reasons as /run/systemd/ first.) */ | |
2119 | root = "/run/systemd/mount-rootfs"; | |
2120 | (void) mkdir_label(root, 0555); | |
e908468b | 2121 | |
d18aff04 | 2122 | require_prefix = true; |
0722b359 | 2123 | } |
e908468b | 2124 | |
79d956db | 2125 | if (p->n_extension_images > 0 || !strv_isempty(p->extension_directories)) { |
55ea4ef0 MG |
2126 | /* Hierarchy population needs to be done for sysext and confext extension images */ |
2127 | r = parse_env_extension_hierarchies(&hierarchies, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES"); | |
93f59701 LB |
2128 | if (r < 0) |
2129 | return r; | |
2130 | } | |
2131 | ||
063c977a LP |
2132 | r = append_access_mounts(&ml, p->read_write_paths, READWRITE, require_prefix); |
2133 | if (r < 0) | |
2134 | return r; | |
613b411c | 2135 | |
063c977a LP |
2136 | r = append_access_mounts(&ml, p->read_only_paths, READONLY, require_prefix); |
2137 | if (r < 0) | |
2138 | return r; | |
5f7a690a | 2139 | |
063c977a LP |
2140 | r = append_access_mounts(&ml, p->inaccessible_paths, INACCESSIBLE, require_prefix); |
2141 | if (r < 0) | |
2142 | return r; | |
613b411c | 2143 | |
063c977a LP |
2144 | r = append_access_mounts(&ml, p->exec_paths, EXEC, require_prefix); |
2145 | if (r < 0) | |
2146 | return r; | |
613b411c | 2147 | |
063c977a LP |
2148 | r = append_access_mounts(&ml, p->no_exec_paths, NOEXEC, require_prefix); |
2149 | if (r < 0) | |
2150 | return r; | |
2151 | ||
2152 | r = append_empty_dir_mounts(&ml, p->empty_directories); | |
2153 | if (r < 0) | |
2154 | return r; | |
2155 | ||
2156 | r = append_bind_mounts(&ml, p->bind_mounts, p->n_bind_mounts); | |
2157 | if (r < 0) | |
2158 | return r; | |
2159 | ||
2160 | r = append_tmpfs_mounts(&ml, p->temporary_filesystems, p->n_temporary_filesystems); | |
2161 | if (r < 0) | |
2162 | return r; | |
7ff7394d | 2163 | |
063c977a LP |
2164 | if (p->tmp_dir) { |
2165 | bool ro = streq(p->tmp_dir, RUN_SYSTEMD_EMPTY); | |
2166 | ||
2167 | MountEntry *me = mount_list_extend(&ml); | |
2168 | if (!me) | |
2169 | return log_oom_debug(); | |
2170 | ||
2171 | *me = (MountEntry) { | |
2172 | .path_const = "/tmp", | |
2173 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, | |
2174 | .source_const = p->tmp_dir, | |
2175 | }; | |
2176 | } | |
2177 | ||
2178 | if (p->var_tmp_dir) { | |
2179 | bool ro = streq(p->var_tmp_dir, RUN_SYSTEMD_EMPTY); | |
2180 | ||
2181 | MountEntry *me = mount_list_extend(&ml); | |
2182 | if (!me) | |
2183 | return log_oom_debug(); | |
2184 | ||
2185 | *me = (MountEntry) { | |
2186 | .path_const = "/var/tmp", | |
2187 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, | |
2188 | .source_const = p->var_tmp_dir, | |
2189 | }; | |
2190 | } | |
2191 | ||
2192 | r = append_mount_images(&ml, p->mount_images, p->n_mount_images); | |
2193 | if (r < 0) | |
2194 | return r; | |
2195 | ||
2196 | r = append_extensions(&ml, root, p->extension_dir, hierarchies, p->extension_images, p->n_extension_images, p->extension_directories); | |
2197 | if (r < 0) | |
2198 | return r; | |
2199 | ||
2200 | if (p->private_dev) { | |
2201 | MountEntry *me = mount_list_extend(&ml); | |
2202 | if (!me) | |
2203 | return log_oom_debug(); | |
2204 | ||
2205 | *me = (MountEntry) { | |
2206 | .path_const = "/dev", | |
2207 | .mode = PRIVATE_DEV, | |
2208 | .flags = DEV_MOUNT_OPTIONS, | |
2209 | }; | |
2210 | } | |
2211 | ||
2212 | /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the protective | |
2213 | mounts to non-pid /proc paths would fail. But the pid only option may have failed gracefully, so | |
2214 | let's try the mounts but it's not fatal if they don't succeed. */ | |
2215 | bool ignore_protect_proc = p->ignore_protect_paths || p->proc_subset == PROC_SUBSET_PID; | |
2216 | if (p->protect_kernel_tunables) { | |
2217 | r = append_static_mounts(&ml, | |
2218 | protect_kernel_tunables_proc_table, | |
2219 | ELEMENTSOF(protect_kernel_tunables_proc_table), | |
2220 | ignore_protect_proc); | |
ddc155b2 | 2221 | if (r < 0) |
063c977a | 2222 | return r; |
ddc155b2 | 2223 | |
063c977a LP |
2224 | r = append_static_mounts(&ml, |
2225 | protect_kernel_tunables_sys_table, | |
2226 | ELEMENTSOF(protect_kernel_tunables_sys_table), | |
2227 | p->ignore_protect_paths); | |
ddc155b2 | 2228 | if (r < 0) |
063c977a LP |
2229 | return r; |
2230 | } | |
ddc155b2 | 2231 | |
063c977a LP |
2232 | if (p->protect_kernel_modules) { |
2233 | r = append_static_mounts(&ml, | |
2234 | protect_kernel_modules_table, | |
2235 | ELEMENTSOF(protect_kernel_modules_table), | |
2236 | p->ignore_protect_paths); | |
6c47cd7d | 2237 | if (r < 0) |
063c977a LP |
2238 | return r; |
2239 | } | |
6c47cd7d | 2240 | |
063c977a LP |
2241 | if (p->protect_kernel_logs) { |
2242 | r = append_static_mounts(&ml, | |
2243 | protect_kernel_logs_proc_table, | |
2244 | ELEMENTSOF(protect_kernel_logs_proc_table), | |
2245 | ignore_protect_proc); | |
d2d6c096 | 2246 | if (r < 0) |
063c977a | 2247 | return r; |
d2d6c096 | 2248 | |
063c977a LP |
2249 | r = append_static_mounts(&ml, |
2250 | protect_kernel_logs_dev_table, | |
2251 | ELEMENTSOF(protect_kernel_logs_dev_table), | |
2252 | p->ignore_protect_paths); | |
2abd4e38 | 2253 | if (r < 0) |
063c977a LP |
2254 | return r; |
2255 | } | |
2abd4e38 | 2256 | |
063c977a LP |
2257 | if (p->protect_control_groups) { |
2258 | MountEntry *me = mount_list_extend(&ml); | |
2259 | if (!me) | |
2260 | return log_oom_debug(); | |
56a13a49 | 2261 | |
063c977a LP |
2262 | *me = (MountEntry) { |
2263 | .path_const = "/sys/fs/cgroup", | |
2264 | .mode = READONLY, | |
2265 | }; | |
2266 | } | |
7ff7394d | 2267 | |
063c977a LP |
2268 | r = append_protect_home(&ml, p->protect_home, p->ignore_protect_paths); |
2269 | if (r < 0) | |
2270 | return r; | |
56a13a49 | 2271 | |
063c977a LP |
2272 | r = append_protect_system(&ml, p->protect_system, false); |
2273 | if (r < 0) | |
2274 | return r; | |
ac0930c8 | 2275 | |
063c977a LP |
2276 | if (namespace_parameters_mount_apivfs(p)) { |
2277 | r = append_static_mounts(&ml, | |
2278 | apivfs_table, | |
2279 | ELEMENTSOF(apivfs_table), | |
2280 | p->ignore_protect_paths); | |
b3d13314 | 2281 | if (r < 0) |
063c977a LP |
2282 | return r; |
2283 | } | |
b3d13314 | 2284 | |
063c977a LP |
2285 | /* Note, if proc is mounted with subset=pid then neither of the two paths will exist, i.e. they are |
2286 | * implicitly protected by the mount option. */ | |
2287 | if (p->protect_hostname) { | |
3d1b999b LP |
2288 | r = append_static_mounts( |
2289 | &ml, | |
2290 | protect_hostname_table, | |
2291 | ELEMENTSOF(protect_hostname_table), | |
2292 | ignore_protect_proc); | |
2293 | if (r < 0) | |
2294 | return r; | |
063c977a | 2295 | } |
788e7201 | 2296 | |
063c977a LP |
2297 | if (p->private_network) { |
2298 | MountEntry *me = mount_list_extend(&ml); | |
2299 | if (!me) | |
2300 | return log_oom_debug(); | |
94a7b275 | 2301 | |
063c977a LP |
2302 | *me = (MountEntry) { |
2303 | .path_const = "/sys", | |
2304 | .mode = PRIVATE_SYSFS, | |
2305 | }; | |
2306 | } | |
59eeb84b | 2307 | |
063c977a LP |
2308 | if (p->private_ipc) { |
2309 | MountEntry *me = mount_list_extend(&ml); | |
2310 | if (!me) | |
2311 | return log_oom_debug(); | |
417116f2 | 2312 | |
063c977a LP |
2313 | *me = (MountEntry) { |
2314 | .path_const = "/dev/mqueue", | |
2315 | .mode = MQUEUEFS, | |
2316 | .flags = MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, | |
2317 | }; | |
2318 | } | |
417116f2 | 2319 | |
063c977a LP |
2320 | if (p->creds_path) { |
2321 | /* If our service has a credentials store configured, then bind that one in, but hide | |
2322 | * everything else. */ | |
5d997827 | 2323 | |
063c977a LP |
2324 | MountEntry *me = mount_list_extend(&ml); |
2325 | if (!me) | |
2326 | return log_oom_debug(); | |
aecd5ac6 | 2327 | |
063c977a LP |
2328 | *me = (MountEntry) { |
2329 | .path_const = "/run/credentials", | |
2330 | .mode = TMPFS, | |
2331 | .read_only = true, | |
2332 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, | |
2333 | .flags = MS_NODEV|MS_STRICTATIME|MS_NOSUID|MS_NOEXEC, | |
2334 | }; | |
bbb4e7f3 | 2335 | |
063c977a LP |
2336 | me = mount_list_extend(&ml); |
2337 | if (!me) | |
2338 | return log_oom_debug(); | |
91dd5f7c | 2339 | |
063c977a LP |
2340 | *me = (MountEntry) { |
2341 | .path_const = p->creds_path, | |
2342 | .mode = BIND_MOUNT, | |
2343 | .read_only = true, | |
2344 | .source_const = p->creds_path, | |
2345 | .ignore = true, | |
2346 | }; | |
2347 | } else { | |
2348 | /* If our service has no credentials store configured, then make the whole credentials tree | |
2349 | * inaccessible wholesale. */ | |
91dd5f7c | 2350 | |
063c977a LP |
2351 | MountEntry *me = mount_list_extend(&ml); |
2352 | if (!me) | |
2353 | return log_oom_debug(); | |
91dd5f7c | 2354 | |
063c977a LP |
2355 | *me = (MountEntry) { |
2356 | .path_const = "/run/credentials", | |
2357 | .mode = INACCESSIBLE, | |
2358 | .ignore = true, | |
2359 | }; | |
2360 | } | |
2361 | ||
2362 | if (p->log_namespace) { | |
2363 | _cleanup_free_ char *q = NULL; | |
2364 | ||
2365 | q = strjoin("/run/systemd/journal.", p->log_namespace); | |
2366 | if (!q) | |
d4f0878e | 2367 | return log_oom_debug(); |
5327c910 | 2368 | |
063c977a LP |
2369 | MountEntry *me = mount_list_extend(&ml); |
2370 | if (!me) | |
2371 | return log_oom_debug(); | |
2372 | ||
2373 | *me = (MountEntry) { | |
2374 | .path_const = "/run/systemd/journal", | |
2375 | .mode = BIND_MOUNT_RECURSIVE, | |
2376 | .read_only = true, | |
2377 | .source_malloc = TAKE_PTR(q), | |
2378 | }; | |
15ae422b LP |
2379 | } |
2380 | ||
063c977a LP |
2381 | /* Will be used to add bind mounts at runtime */ |
2382 | if (setup_propagate) { | |
2383 | MountEntry *me = mount_list_extend(&ml); | |
2384 | if (!me) | |
2385 | return log_oom_debug(); | |
2386 | ||
2387 | *me = (MountEntry) { | |
2388 | .source_const = p->propagate_dir, | |
2389 | .path_const = p->incoming_dir, | |
2390 | .mode = BIND_MOUNT, | |
2391 | .read_only = true, | |
2392 | }; | |
2393 | } | |
2394 | ||
2395 | if (p->notify_socket) { | |
2396 | MountEntry *me = mount_list_extend(&ml); | |
2397 | if (!me) | |
2398 | return log_oom_debug(); | |
2399 | ||
2400 | *me = (MountEntry) { | |
2401 | .path_const = p->notify_socket, | |
2402 | .source_const = p->notify_socket, | |
2403 | .mode = BIND_MOUNT, | |
2404 | .read_only = true, | |
2405 | }; | |
2406 | } | |
2407 | ||
2408 | if (p->host_os_release_stage) { | |
2409 | MountEntry *me = mount_list_extend(&ml); | |
2410 | if (!me) | |
2411 | return log_oom_debug(); | |
2412 | ||
2413 | *me = (MountEntry) { | |
2414 | .path_const = "/run/host/.os-release-stage/", | |
2415 | .source_const = p->host_os_release_stage, | |
2416 | .mode = BIND_MOUNT, | |
2417 | .read_only = true, | |
2418 | .ignore = true, /* Live copy, don't hard-fail if it goes missing */ | |
2419 | }; | |
2420 | } | |
2421 | ||
2422 | /* Prepend the root directory where that's necessary */ | |
2423 | r = prefix_where_needed(&ml, root); | |
2424 | if (r < 0) | |
2425 | return r; | |
2426 | ||
2427 | drop_unused_mounts(&ml, root); | |
2428 | ||
1beab8b0 LP |
2429 | /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */ |
2430 | ||
d944dc95 | 2431 | if (unshare(CLONE_NEWNS) < 0) { |
763a260a | 2432 | r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m"); |
063c977a | 2433 | |
eacfbd89 LP |
2434 | if (ERRNO_IS_PRIVILEGE(r) || |
2435 | ERRNO_IS_NOT_SUPPORTED(r)) | |
cbc056c8 ZJS |
2436 | /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter |
2437 | * in place that doesn't allow us to create namespaces (or a missing cap), then | |
2438 | * propagate a recognizable error back, which the caller can use to detect this case | |
2439 | * (and only this) and optionally continue without namespacing applied. */ | |
063c977a | 2440 | return -ENOANO; |
1beab8b0 | 2441 | |
063c977a | 2442 | return r; |
d944dc95 | 2443 | } |
1e4e94c8 | 2444 | |
5e8deb94 LB |
2445 | /* Create the source directory to allow runtime propagation of mounts */ |
2446 | if (setup_propagate) | |
79d956db | 2447 | (void) mkdir_p(p->propagate_dir, 0600); |
5e8deb94 | 2448 | |
79d956db | 2449 | if (p->n_extension_images > 0 || !strv_isempty(p->extension_directories)) |
3ebf0b0b LB |
2450 | /* ExtensionImages/Directories mountpoint directories will be created while parsing the |
2451 | * mounts to create, so have the parent ready */ | |
79d956db | 2452 | (void) mkdir_p(p->extension_dir, 0600); |
93f59701 | 2453 | |
9b68367b YW |
2454 | /* Remount / as SLAVE so that nothing now mounted in the namespace |
2455 | * shows up in the parent */ | |
063c977a LP |
2456 | if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) |
2457 | return log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m"); | |
ee818b89 | 2458 | |
79d956db | 2459 | if (p->root_image) { |
e908468b | 2460 | /* A root image is specified, mount it to the right place */ |
8d9a1d59 LP |
2461 | r = dissected_image_mount( |
2462 | dissected_image, | |
2463 | root, | |
2464 | /* uid_shift= */ UID_INVALID, | |
2465 | /* uid_range= */ UID_INVALID, | |
2466 | /* userns_fd= */ -EBADF, | |
2467 | dissect_image_flags); | |
063c977a LP |
2468 | if (r < 0) |
2469 | return log_debug_errno(r, "Failed to mount root image: %m"); | |
41bc4849 LP |
2470 | |
2471 | /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device | |
2472 | * if it likes. */ | |
2473 | r = loop_device_flock(loop_device, LOCK_UN); | |
063c977a LP |
2474 | if (r < 0) |
2475 | return log_debug_errno(r, "Failed to release lock on loopback block device: %m"); | |
915e6d16 | 2476 | |
3044d343 | 2477 | r = dissected_image_relinquish(dissected_image); |
063c977a LP |
2478 | if (r < 0) |
2479 | return log_debug_errno(r, "Failed to relinquish dissected image: %m"); | |
78ebe980 | 2480 | |
79d956db | 2481 | } else if (p->root_directory) { |
915e6d16 | 2482 | |
e908468b LP |
2483 | /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */ |
2484 | r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW); | |
063c977a LP |
2485 | if (r < 0) |
2486 | return log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root); | |
8f1ad200 | 2487 | if (r == 0) { |
21935150 LP |
2488 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); |
2489 | if (r < 0) | |
063c977a | 2490 | return r; |
d944dc95 | 2491 | } |
e908468b | 2492 | |
9b68367b | 2493 | } else { |
e908468b | 2494 | /* Let's mount the main root directory to the root directory to use */ |
21935150 LP |
2495 | r = mount_nofollow_verbose(LOG_DEBUG, "/", root, NULL, MS_BIND|MS_REC, NULL); |
2496 | if (r < 0) | |
063c977a | 2497 | return r; |
ee818b89 | 2498 | } |
c2c13f2d | 2499 | |
4e0c20de | 2500 | /* Try to set up the new root directory before mounting anything else there. */ |
79d956db | 2501 | if (p->root_image || p->root_directory) |
4e0c20de LP |
2502 | (void) base_filesystem_create(root, UID_INVALID, GID_INVALID); |
2503 | ||
82fb2da2 | 2504 | /* Now make the magic happen */ |
063c977a | 2505 | r = apply_mounts(&ml, root, p, error_path); |
82fb2da2 | 2506 | if (r < 0) |
063c977a | 2507 | return r; |
15ae422b | 2508 | |
9b68367b | 2509 | /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */ |
9d50f850 | 2510 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
79d956db | 2511 | if (r == -EINVAL && p->root_directory) { |
ea63a260 LB |
2512 | /* If we are using root_directory and we don't have privileges (ie: user manager in a user |
2513 | * namespace) and the root_directory is already a mount point in the parent namespace, | |
2514 | * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than | |
2515 | * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a | |
2516 | * mount point) and try again. */ | |
2517 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); | |
2518 | if (r < 0) | |
063c977a | 2519 | return r; |
9d50f850 | 2520 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
ea63a260 | 2521 | } |
063c977a LP |
2522 | if (r < 0) |
2523 | return log_debug_errno(r, "Failed to mount root with MS_MOVE: %m"); | |
ee818b89 | 2524 | |
874cdcbc LP |
2525 | /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to |
2526 | * the host, since what's disconnected is disconnected. */ | |
063c977a LP |
2527 | if (mount(NULL, "/", NULL, mount_propagation_flag | MS_REC, NULL) < 0) |
2528 | return log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m"); | |
15ae422b | 2529 | |
5e8deb94 LB |
2530 | /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only |
2531 | * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */ | |
2532 | if (setup_propagate) { | |
063c977a LP |
2533 | if (mount(NULL, p->incoming_dir, NULL, MS_SLAVE, NULL) < 0) |
2534 | return log_debug_errno(errno, "Failed to remount %s with MS_SLAVE: %m", p->incoming_dir); | |
5e8deb94 LB |
2535 | } |
2536 | ||
063c977a | 2537 | return 0; |
613b411c LP |
2538 | } |
2539 | ||
da6053d0 | 2540 | void bind_mount_free_many(BindMount *b, size_t n) { |
d2d6c096 LP |
2541 | assert(b || n == 0); |
2542 | ||
fe96c0f8 | 2543 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
2544 | free(b[i].source); |
2545 | free(b[i].destination); | |
2546 | } | |
2547 | ||
2548 | free(b); | |
2549 | } | |
2550 | ||
da6053d0 | 2551 | int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) { |
d2d6c096 LP |
2552 | _cleanup_free_ char *s = NULL, *d = NULL; |
2553 | BindMount *c; | |
2554 | ||
2555 | assert(b); | |
2556 | assert(n); | |
2557 | assert(item); | |
2558 | ||
2559 | s = strdup(item->source); | |
2560 | if (!s) | |
2561 | return -ENOMEM; | |
2562 | ||
2563 | d = strdup(item->destination); | |
2564 | if (!d) | |
2565 | return -ENOMEM; | |
2566 | ||
aa484f35 | 2567 | c = reallocarray(*b, *n + 1, sizeof(BindMount)); |
d2d6c096 LP |
2568 | if (!c) |
2569 | return -ENOMEM; | |
2570 | ||
2571 | *b = c; | |
2572 | ||
2573 | c[(*n) ++] = (BindMount) { | |
1cc6c93a YW |
2574 | .source = TAKE_PTR(s), |
2575 | .destination = TAKE_PTR(d), | |
d2d6c096 | 2576 | .read_only = item->read_only, |
9ce4e4b0 | 2577 | .nosuid = item->nosuid, |
d2d6c096 LP |
2578 | .recursive = item->recursive, |
2579 | .ignore_enoent = item->ignore_enoent, | |
2580 | }; | |
2581 | ||
d2d6c096 LP |
2582 | return 0; |
2583 | } | |
2584 | ||
b3d13314 | 2585 | MountImage* mount_image_free_many(MountImage *m, size_t *n) { |
b3d13314 LB |
2586 | assert(n); |
2587 | assert(m || *n == 0); | |
2588 | ||
fe96c0f8 | 2589 | for (size_t i = 0; i < *n; i++) { |
b3d13314 LB |
2590 | free(m[i].source); |
2591 | free(m[i].destination); | |
427353f6 | 2592 | mount_options_free_all(m[i].mount_options); |
b3d13314 LB |
2593 | } |
2594 | ||
2595 | free(m); | |
2596 | *n = 0; | |
2597 | return NULL; | |
2598 | } | |
2599 | ||
2600 | int mount_image_add(MountImage **m, size_t *n, const MountImage *item) { | |
2601 | _cleanup_free_ char *s = NULL, *d = NULL; | |
427353f6 | 2602 | _cleanup_(mount_options_free_allp) MountOptions *options = NULL; |
b3d13314 LB |
2603 | MountImage *c; |
2604 | ||
2605 | assert(m); | |
2606 | assert(n); | |
2607 | assert(item); | |
2608 | ||
2609 | s = strdup(item->source); | |
2610 | if (!s) | |
2611 | return -ENOMEM; | |
2612 | ||
93f59701 LB |
2613 | if (item->destination) { |
2614 | d = strdup(item->destination); | |
2615 | if (!d) | |
2616 | return -ENOMEM; | |
2617 | } | |
b3d13314 | 2618 | |
427353f6 | 2619 | LIST_FOREACH(mount_options, i, item->mount_options) { |
c2b2df60 | 2620 | _cleanup_(mount_options_free_allp) MountOptions *o = NULL; |
427353f6 LB |
2621 | |
2622 | o = new(MountOptions, 1); | |
2623 | if (!o) | |
2624 | return -ENOMEM; | |
2625 | ||
2626 | *o = (MountOptions) { | |
2627 | .partition_designator = i->partition_designator, | |
2628 | .options = strdup(i->options), | |
2629 | }; | |
2630 | if (!o->options) | |
2631 | return -ENOMEM; | |
2632 | ||
2633 | LIST_APPEND(mount_options, options, TAKE_PTR(o)); | |
2634 | } | |
2635 | ||
b3d13314 LB |
2636 | c = reallocarray(*m, *n + 1, sizeof(MountImage)); |
2637 | if (!c) | |
2638 | return -ENOMEM; | |
2639 | ||
2640 | *m = c; | |
2641 | ||
2642 | c[(*n) ++] = (MountImage) { | |
2643 | .source = TAKE_PTR(s), | |
2644 | .destination = TAKE_PTR(d), | |
427353f6 | 2645 | .mount_options = TAKE_PTR(options), |
b3d13314 | 2646 | .ignore_enoent = item->ignore_enoent, |
93f59701 | 2647 | .type = item->type, |
b3d13314 LB |
2648 | }; |
2649 | ||
2650 | return 0; | |
2651 | } | |
2652 | ||
da6053d0 | 2653 | void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) { |
2abd4e38 YW |
2654 | assert(t || n == 0); |
2655 | ||
fe96c0f8 | 2656 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
2657 | free(t[i].path); |
2658 | free(t[i].options); | |
2659 | } | |
2660 | ||
2661 | free(t); | |
2662 | } | |
2663 | ||
2664 | int temporary_filesystem_add( | |
2665 | TemporaryFileSystem **t, | |
da6053d0 | 2666 | size_t *n, |
2abd4e38 YW |
2667 | const char *path, |
2668 | const char *options) { | |
2669 | ||
2670 | _cleanup_free_ char *p = NULL, *o = NULL; | |
2671 | TemporaryFileSystem *c; | |
2672 | ||
2673 | assert(t); | |
2674 | assert(n); | |
2675 | assert(path); | |
2676 | ||
2677 | p = strdup(path); | |
2678 | if (!p) | |
2679 | return -ENOMEM; | |
2680 | ||
2681 | if (!isempty(options)) { | |
2682 | o = strdup(options); | |
2683 | if (!o) | |
2684 | return -ENOMEM; | |
2685 | } | |
2686 | ||
aa484f35 | 2687 | c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem)); |
2abd4e38 YW |
2688 | if (!c) |
2689 | return -ENOMEM; | |
2690 | ||
2691 | *t = c; | |
2692 | ||
2693 | c[(*n) ++] = (TemporaryFileSystem) { | |
1cc6c93a YW |
2694 | .path = TAKE_PTR(p), |
2695 | .options = TAKE_PTR(o), | |
2abd4e38 YW |
2696 | }; |
2697 | ||
2abd4e38 YW |
2698 | return 0; |
2699 | } | |
2700 | ||
a652f050 JR |
2701 | static int make_tmp_prefix(const char *prefix) { |
2702 | _cleanup_free_ char *t = NULL; | |
254d1313 | 2703 | _cleanup_close_ int fd = -EBADF; |
a652f050 JR |
2704 | int r; |
2705 | ||
2706 | /* Don't do anything unless we know the dir is actually missing */ | |
2707 | r = access(prefix, F_OK); | |
2708 | if (r >= 0) | |
2709 | return 0; | |
2710 | if (errno != ENOENT) | |
2711 | return -errno; | |
2712 | ||
2053593f | 2713 | WITH_UMASK(000) |
30443439 | 2714 | r = mkdir_parents(prefix, 0755); |
a652f050 JR |
2715 | if (r < 0) |
2716 | return r; | |
2717 | ||
2718 | r = tempfn_random(prefix, NULL, &t); | |
2719 | if (r < 0) | |
2720 | return r; | |
2721 | ||
96603ea0 LP |
2722 | /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for |
2723 | * the suid bit, below. */ | |
2724 | fd = open_mkdir_at(AT_FDCWD, t, O_EXCL|O_CLOEXEC, 0777); | |
2725 | if (fd < 0) | |
2726 | return fd; | |
a652f050 | 2727 | |
96603ea0 LP |
2728 | r = RET_NERRNO(fchmod(fd, 01777)); |
2729 | if (r < 0) { | |
a652f050 JR |
2730 | (void) rmdir(t); |
2731 | return r; | |
2732 | } | |
2733 | ||
96603ea0 LP |
2734 | r = RET_NERRNO(rename(t, prefix)); |
2735 | if (r < 0) { | |
a652f050 JR |
2736 | (void) rmdir(t); |
2737 | return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */ | |
2738 | } | |
2739 | ||
2740 | return 0; | |
2741 | ||
2742 | } | |
2743 | ||
56a13a49 | 2744 | static int setup_one_tmp_dir(const char *id, const char *prefix, char **path, char **tmp_path) { |
613b411c | 2745 | _cleanup_free_ char *x = NULL; |
19cd4e19 | 2746 | _cleanup_free_ char *y = NULL; |
6b46ea73 | 2747 | sd_id128_t boot_id; |
56a13a49 | 2748 | bool rw = true; |
6b46ea73 | 2749 | int r; |
613b411c LP |
2750 | |
2751 | assert(id); | |
2752 | assert(prefix); | |
2753 | assert(path); | |
2754 | ||
6b46ea73 LP |
2755 | /* We include the boot id in the directory so that after a |
2756 | * reboot we can easily identify obsolete directories. */ | |
2757 | ||
2758 | r = sd_id128_get_boot(&boot_id); | |
2759 | if (r < 0) | |
2760 | return r; | |
2761 | ||
85b55869 | 2762 | x = strjoin(prefix, "/systemd-private-", SD_ID128_TO_STRING(boot_id), "-", id, "-XXXXXX"); |
613b411c LP |
2763 | if (!x) |
2764 | return -ENOMEM; | |
2765 | ||
a652f050 JR |
2766 | r = make_tmp_prefix(prefix); |
2767 | if (r < 0) | |
2768 | return r; | |
2769 | ||
2053593f | 2770 | WITH_UMASK(0077) |
56a13a49 ZJS |
2771 | if (!mkdtemp(x)) { |
2772 | if (errno == EROFS || ERRNO_IS_DISK_SPACE(errno)) | |
2773 | rw = false; | |
2774 | else | |
2775 | return -errno; | |
2776 | } | |
613b411c | 2777 | |
56a13a49 | 2778 | if (rw) { |
19cd4e19 | 2779 | y = strjoin(x, "/tmp"); |
2780 | if (!y) | |
2781 | return -ENOMEM; | |
2782 | ||
2053593f | 2783 | WITH_UMASK(0000) |
19cd4e19 | 2784 | if (mkdir(y, 0777 | S_ISVTX) < 0) |
7c76e181 | 2785 | return -errno; |
19cd4e19 | 2786 | |
03bc11d1 | 2787 | r = label_fix_full(AT_FDCWD, y, prefix, 0); |
56a13a49 ZJS |
2788 | if (r < 0) |
2789 | return r; | |
19cd4e19 | 2790 | |
2791 | if (tmp_path) | |
2792 | *tmp_path = TAKE_PTR(y); | |
56a13a49 ZJS |
2793 | } else { |
2794 | /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being | |
2795 | * read-only. This way the service will get the EROFS result as if it was writing to the real | |
2796 | * file system. */ | |
2053593f | 2797 | WITH_UMASK(0000) |
30443439 | 2798 | r = mkdir_p(RUN_SYSTEMD_EMPTY, 0500); |
56a13a49 ZJS |
2799 | if (r < 0) |
2800 | return r; | |
613b411c | 2801 | |
3f181262 LP |
2802 | r = free_and_strdup(&x, RUN_SYSTEMD_EMPTY); |
2803 | if (r < 0) | |
2804 | return r; | |
c17ec25e | 2805 | } |
15ae422b | 2806 | |
1cc6c93a | 2807 | *path = TAKE_PTR(x); |
613b411c LP |
2808 | return 0; |
2809 | } | |
2810 | ||
2811 | int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) { | |
56a13a49 ZJS |
2812 | _cleanup_(namespace_cleanup_tmpdirp) char *a = NULL; |
2813 | _cleanup_(rmdir_and_freep) char *a_tmp = NULL; | |
2814 | char *b; | |
613b411c LP |
2815 | int r; |
2816 | ||
2817 | assert(id); | |
2818 | assert(tmp_dir); | |
2819 | assert(var_tmp_dir); | |
2820 | ||
56a13a49 | 2821 | r = setup_one_tmp_dir(id, "/tmp", &a, &a_tmp); |
613b411c LP |
2822 | if (r < 0) |
2823 | return r; | |
2824 | ||
56a13a49 ZJS |
2825 | r = setup_one_tmp_dir(id, "/var/tmp", &b, NULL); |
2826 | if (r < 0) | |
613b411c | 2827 | return r; |
613b411c | 2828 | |
56a13a49 ZJS |
2829 | a_tmp = mfree(a_tmp); /* avoid rmdir */ |
2830 | *tmp_dir = TAKE_PTR(a); | |
2831 | *var_tmp_dir = TAKE_PTR(b); | |
613b411c LP |
2832 | |
2833 | return 0; | |
2834 | } | |
2835 | ||
13339577 | 2836 | int setup_shareable_ns(int ns_storage_socket[static 2], unsigned long nsflag) { |
5bb1d7fb | 2837 | _cleanup_close_ int ns = -EBADF; |
13339577 | 2838 | int r; |
54c2459d | 2839 | const char *ns_name, *ns_path; |
613b411c | 2840 | |
54c2459d XR |
2841 | assert(ns_storage_socket); |
2842 | assert(ns_storage_socket[0] >= 0); | |
2843 | assert(ns_storage_socket[1] >= 0); | |
2844 | ||
2845 | ns_name = namespace_single_flag_to_string(nsflag); | |
2846 | assert(ns_name); | |
613b411c LP |
2847 | |
2848 | /* We use the passed socketpair as a storage buffer for our | |
76cd584b LP |
2849 | * namespace reference fd. Whatever process runs this first |
2850 | * shall create a new namespace, all others should just join | |
2851 | * it. To serialize that we use a file lock on the socket | |
2852 | * pair. | |
613b411c LP |
2853 | * |
2854 | * It's a bit crazy, but hey, works great! */ | |
2855 | ||
13339577 DDM |
2856 | r = posix_lock(ns_storage_socket[0], LOCK_EX); |
2857 | if (r < 0) | |
2858 | return r; | |
613b411c | 2859 | |
13339577 | 2860 | CLEANUP_POSIX_UNLOCK(ns_storage_socket[0]); |
613b411c | 2861 | |
13339577 DDM |
2862 | ns = receive_one_fd(ns_storage_socket[0], MSG_PEEK|MSG_DONTWAIT); |
2863 | if (ns >= 0) { | |
2864 | /* Yay, found something, so let's join the namespace */ | |
2865 | r = RET_NERRNO(setns(ns, nsflag)); | |
2866 | if (r < 0) | |
2867 | return r; | |
613b411c | 2868 | |
13339577 DDM |
2869 | return 0; |
2870 | } | |
613b411c | 2871 | |
13339577 DDM |
2872 | if (ns != -EAGAIN) |
2873 | return ns; | |
613b411c | 2874 | |
13339577 | 2875 | /* Nothing stored yet, so let's create a new namespace. */ |
613b411c | 2876 | |
13339577 DDM |
2877 | if (unshare(nsflag) < 0) |
2878 | return -errno; | |
613b411c | 2879 | |
13339577 | 2880 | (void) loopback_setup(); |
613b411c | 2881 | |
13339577 DDM |
2882 | ns_path = strjoina("/proc/self/ns/", ns_name); |
2883 | ns = open(ns_path, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
2884 | if (ns < 0) | |
2885 | return -errno; | |
613b411c | 2886 | |
13339577 DDM |
2887 | r = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
2888 | if (r < 0) | |
2889 | return r; | |
613b411c | 2890 | |
13339577 | 2891 | return 1; |
15ae422b | 2892 | } |
417116f2 | 2893 | |
13339577 | 2894 | int open_shareable_ns_path(int ns_storage_socket[static 2], const char *path, unsigned long nsflag) { |
5bb1d7fb | 2895 | _cleanup_close_ int ns = -EBADF; |
13339577 | 2896 | int r; |
51af7fb2 | 2897 | |
54c2459d XR |
2898 | assert(ns_storage_socket); |
2899 | assert(ns_storage_socket[0] >= 0); | |
2900 | assert(ns_storage_socket[1] >= 0); | |
51af7fb2 LP |
2901 | assert(path); |
2902 | ||
54c2459d XR |
2903 | /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in |
2904 | * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will | |
2905 | * allocate a new anonymous ns if needed. */ | |
51af7fb2 | 2906 | |
13339577 DDM |
2907 | r = posix_lock(ns_storage_socket[0], LOCK_EX); |
2908 | if (r < 0) | |
2909 | return r; | |
51af7fb2 | 2910 | |
13339577 | 2911 | CLEANUP_POSIX_UNLOCK(ns_storage_socket[0]); |
51af7fb2 | 2912 | |
13339577 DDM |
2913 | ns = receive_one_fd(ns_storage_socket[0], MSG_PEEK|MSG_DONTWAIT); |
2914 | if (ns >= 0) | |
2915 | return 0; | |
2916 | if (ns != -EAGAIN) | |
2917 | return ns; | |
51af7fb2 | 2918 | |
13339577 | 2919 | /* Nothing stored yet. Open the file from the file system. */ |
51af7fb2 | 2920 | |
13339577 DDM |
2921 | ns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC); |
2922 | if (ns < 0) | |
2923 | return -errno; | |
51af7fb2 | 2924 | |
13339577 DDM |
2925 | r = fd_is_ns(ns, nsflag); |
2926 | if (r == 0) | |
2927 | return -EINVAL; | |
2928 | if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */ | |
2929 | return r; | |
51af7fb2 | 2930 | |
13339577 DDM |
2931 | r = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
2932 | if (r < 0) | |
2933 | return r; | |
51af7fb2 | 2934 | |
13339577 | 2935 | return 1; |
51af7fb2 LP |
2936 | } |
2937 | ||
6e2d7c4f MS |
2938 | bool ns_type_supported(NamespaceType type) { |
2939 | const char *t, *ns_proc; | |
2940 | ||
0fa5b831 LP |
2941 | t = namespace_type_to_string(type); |
2942 | if (!t) /* Don't know how to translate this? Then it's not supported */ | |
6e2d7c4f MS |
2943 | return false; |
2944 | ||
6e2d7c4f | 2945 | ns_proc = strjoina("/proc/self/ns/", t); |
6e2d7c4f MS |
2946 | return access(ns_proc, F_OK) == 0; |
2947 | } | |
2948 | ||
1b8689f9 | 2949 | static const char *const protect_home_table[_PROTECT_HOME_MAX] = { |
cbc056c8 ZJS |
2950 | [PROTECT_HOME_NO] = "no", |
2951 | [PROTECT_HOME_YES] = "yes", | |
1b8689f9 | 2952 | [PROTECT_HOME_READ_ONLY] = "read-only", |
cbc056c8 | 2953 | [PROTECT_HOME_TMPFS] = "tmpfs", |
417116f2 LP |
2954 | }; |
2955 | ||
1e8c7bd5 | 2956 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES); |
5e1c6154 | 2957 | |
1b8689f9 | 2958 | static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = { |
cbc056c8 ZJS |
2959 | [PROTECT_SYSTEM_NO] = "no", |
2960 | [PROTECT_SYSTEM_YES] = "yes", | |
2961 | [PROTECT_SYSTEM_FULL] = "full", | |
3f815163 | 2962 | [PROTECT_SYSTEM_STRICT] = "strict", |
1b8689f9 LP |
2963 | }; |
2964 | ||
1e8c7bd5 | 2965 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES); |
03c791aa | 2966 | |
6e2d7c4f | 2967 | static const char* const namespace_type_table[] = { |
cbc056c8 | 2968 | [NAMESPACE_MOUNT] = "mnt", |
6e2d7c4f | 2969 | [NAMESPACE_CGROUP] = "cgroup", |
cbc056c8 ZJS |
2970 | [NAMESPACE_UTS] = "uts", |
2971 | [NAMESPACE_IPC] = "ipc", | |
2972 | [NAMESPACE_USER] = "user", | |
2973 | [NAMESPACE_PID] = "pid", | |
2974 | [NAMESPACE_NET] = "net", | |
c3b9c418 | 2975 | [NAMESPACE_TIME] = "time", |
6e2d7c4f MS |
2976 | }; |
2977 | ||
2978 | DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType); | |
4e399953 LP |
2979 | |
2980 | static const char* const protect_proc_table[_PROTECT_PROC_MAX] = { | |
2981 | [PROTECT_PROC_DEFAULT] = "default", | |
2982 | [PROTECT_PROC_NOACCESS] = "noaccess", | |
2983 | [PROTECT_PROC_INVISIBLE] = "invisible", | |
2984 | [PROTECT_PROC_PTRACEABLE] = "ptraceable", | |
2985 | }; | |
2986 | ||
2987 | DEFINE_STRING_TABLE_LOOKUP(protect_proc, ProtectProc); | |
2988 | ||
2989 | static const char* const proc_subset_table[_PROC_SUBSET_MAX] = { | |
2990 | [PROC_SUBSET_ALL] = "all", | |
2991 | [PROC_SUBSET_PID] = "pid", | |
2992 | }; | |
2993 | ||
2994 | DEFINE_STRING_TABLE_LOOKUP(proc_subset, ProcSubset); |