]>
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 | 54 | /* This is ordered by priority! */ |
a868e437 LP |
55 | MOUNT_INACCESSIBLE, |
56 | MOUNT_OVERLAY, | |
57 | MOUNT_IMAGE, | |
58 | MOUNT_BIND, | |
59 | MOUNT_BIND_RECURSIVE, | |
60 | MOUNT_PRIVATE_TMP, | |
61 | MOUNT_PRIVATE_TMP_READ_ONLY, | |
62 | MOUNT_PRIVATE_DEV, | |
63 | MOUNT_BIND_DEV, | |
64 | MOUNT_EMPTY_DIR, | |
65 | MOUNT_PRIVATE_SYSFS, | |
66 | MOUNT_BIND_SYSFS, | |
67 | MOUNT_PROCFS, | |
68 | MOUNT_READ_ONLY, | |
69 | MOUNT_READ_WRITE, | |
70 | MOUNT_NOEXEC, | |
71 | MOUNT_EXEC, | |
72 | MOUNT_TMPFS, | |
73 | MOUNT_RUN, | |
74 | MOUNT_EXTENSION_DIRECTORY, /* Bind-mounted outside the root directory, and used by subsequent mounts */ | |
75 | MOUNT_EXTENSION_IMAGE, /* Mounted outside the root directory, and used by subsequent mounts */ | |
76 | MOUNT_MQUEUEFS, | |
77 | MOUNT_READ_WRITE_IMPLICIT, /* Should have the lowest priority. */ | |
5beb8688 | 78 | _MOUNT_MODE_MAX, |
a868e437 | 79 | _MOUNT_MODE_INVALID = -EINVAL, |
c17ec25e | 80 | } MountMode; |
15ae422b | 81 | |
63862de4 | 82 | typedef enum MountEntryState { |
67248bbd LP |
83 | MOUNT_PENDING, |
84 | MOUNT_APPLIED, | |
85 | MOUNT_SKIPPED, | |
86 | _MOUNT_ENTRY_STATE_MAX, | |
87 | _MOUNT_ENTRY_STATE_INVALID = -EINVAL, | |
63862de4 LB |
88 | } MountEntryState; |
89 | ||
34de407a | 90 | typedef struct MountEntry { |
5327c910 | 91 | const char *path_const; /* Memory allocated on stack or static */ |
a868e437 | 92 | MountMode mode; |
5327c910 LP |
93 | bool ignore:1; /* Ignore if path does not exist? */ |
94 | bool has_prefix:1; /* Already is prefixed by the root dir? */ | |
cfbeb4ef | 95 | bool read_only:1; /* Shall this mount point be read-only? */ |
9ce4e4b0 | 96 | bool nosuid:1; /* Shall set MS_NOSUID on the mount itself */ |
ddc155b2 TM |
97 | bool noexec:1; /* Shall set MS_NOEXEC on the mount itself */ |
98 | bool exec:1; /* Shall clear MS_NOEXEC on the mount itself */ | |
63862de4 | 99 | MountEntryState state; /* Whether it was already processed or skipped */ |
55fe7432 | 100 | char *path_malloc; /* Use this instead of 'path_const' if we had to allocate memory */ |
809ceb82 LB |
101 | const char *unprefixed_path_const; /* If the path was amended with a prefix, these will save the original */ |
102 | char *unprefixed_path_malloc; | |
b3d13314 | 103 | const char *source_const; /* The source path, for bind mounts or images */ |
d2d6c096 | 104 | char *source_malloc; |
2abd4e38 YW |
105 | const char *options_const;/* Mount options for tmpfs */ |
106 | char *options_malloc; | |
107 | unsigned long flags; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */ | |
088696fe | 108 | unsigned n_followed; |
9dc6a6af | 109 | LIST_HEAD(MountOptions, image_options_const); |
f0304df6 | 110 | char **overlay_layers; |
34de407a | 111 | } MountEntry; |
15ae422b | 112 | |
063c977a LP |
113 | typedef struct MountList { |
114 | MountEntry *mounts; | |
115 | size_t n_mounts; | |
116 | } MountList; | |
117 | ||
94293d65 | 118 | /* 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 | 119 | * something there already. These mounts are hence overridden by any other explicitly configured mounts. */ |
5d997827 | 120 | static const MountEntry apivfs_table[] = { |
a868e437 LP |
121 | { "/proc", MOUNT_PROCFS, false }, |
122 | { "/dev", MOUNT_BIND_DEV, false }, | |
123 | { "/sys", MOUNT_BIND_SYSFS, false }, | |
124 | { "/run", MOUNT_RUN, false, .options_const = "mode=0755" TMPFS_LIMITS_RUN, .flags = MS_NOSUID|MS_NODEV|MS_STRICTATIME }, | |
5d997827 | 125 | }; |
f471b2af | 126 | |
11a30cec | 127 | /* ProtectKernelTunables= option and the related filesystem APIs */ |
788e7201 | 128 | static const MountEntry protect_kernel_tunables_proc_table[] = { |
a868e437 LP |
129 | { "/proc/acpi", MOUNT_READ_ONLY, true }, |
130 | { "/proc/apm", MOUNT_READ_ONLY, true }, /* Obsolete API, there's no point in permitting access to this, ever */ | |
131 | { "/proc/asound", MOUNT_READ_ONLY, true }, | |
132 | { "/proc/bus", MOUNT_READ_ONLY, true }, | |
133 | { "/proc/fs", MOUNT_READ_ONLY, true }, | |
134 | { "/proc/irq", MOUNT_READ_ONLY, true }, | |
135 | { "/proc/kallsyms", MOUNT_INACCESSIBLE, true }, | |
136 | { "/proc/kcore", MOUNT_INACCESSIBLE, true }, | |
137 | { "/proc/latency_stats", MOUNT_READ_ONLY, true }, | |
138 | { "/proc/mtrr", MOUNT_READ_ONLY, true }, | |
139 | { "/proc/scsi", MOUNT_READ_ONLY, true }, | |
140 | { "/proc/sys", MOUNT_READ_ONLY, true }, | |
141 | { "/proc/sysrq-trigger", MOUNT_READ_ONLY, true }, | |
142 | { "/proc/timer_stats", MOUNT_READ_ONLY, true }, | |
788e7201 TM |
143 | }; |
144 | ||
145 | static const MountEntry protect_kernel_tunables_sys_table[] = { | |
a868e437 LP |
146 | { "/sys", MOUNT_READ_ONLY, false }, |
147 | { "/sys/fs/bpf", MOUNT_READ_ONLY, true }, | |
148 | { "/sys/fs/cgroup", MOUNT_READ_WRITE_IMPLICIT, false }, /* READ_ONLY is set by ProtectControlGroups= option */ | |
149 | { "/sys/fs/selinux", MOUNT_READ_WRITE_IMPLICIT, true }, | |
150 | { "/sys/kernel/debug", MOUNT_READ_ONLY, true }, | |
151 | { "/sys/kernel/tracing", MOUNT_READ_ONLY, true }, | |
11a30cec DH |
152 | }; |
153 | ||
c575770b | 154 | /* ProtectKernelModules= option */ |
34de407a | 155 | static const MountEntry protect_kernel_modules_table[] = { |
a868e437 | 156 | { "/usr/lib/modules", MOUNT_INACCESSIBLE, true }, |
c575770b DH |
157 | }; |
158 | ||
94a7b275 | 159 | /* ProtectKernelLogs= option */ |
788e7201 | 160 | static const MountEntry protect_kernel_logs_proc_table[] = { |
a868e437 | 161 | { "/proc/kmsg", MOUNT_INACCESSIBLE, true }, |
788e7201 TM |
162 | }; |
163 | ||
164 | static const MountEntry protect_kernel_logs_dev_table[] = { | |
a868e437 | 165 | { "/dev/kmsg", MOUNT_INACCESSIBLE, true }, |
94a7b275 KK |
166 | }; |
167 | ||
b6c432ca DH |
168 | /* |
169 | * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of | |
170 | * system should be protected by ProtectSystem= | |
171 | */ | |
34de407a | 172 | static const MountEntry protect_home_read_only_table[] = { |
a868e437 LP |
173 | { "/home", MOUNT_READ_ONLY, true }, |
174 | { "/run/user", MOUNT_READ_ONLY, true }, | |
175 | { "/root", MOUNT_READ_ONLY, true }, | |
b6c432ca DH |
176 | }; |
177 | ||
e4da7d8c YW |
178 | /* ProtectHome=tmpfs table */ |
179 | static const MountEntry protect_home_tmpfs_table[] = { | |
a868e437 LP |
180 | { "/home", MOUNT_TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, |
181 | { "/run/user", MOUNT_TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
182 | { "/root", MOUNT_TMPFS, true, .read_only = true, .options_const = "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
e4da7d8c YW |
183 | }; |
184 | ||
b6c432ca | 185 | /* ProtectHome=yes table */ |
34de407a | 186 | static const MountEntry protect_home_yes_table[] = { |
a868e437 LP |
187 | { "/home", MOUNT_INACCESSIBLE, true }, |
188 | { "/run/user", MOUNT_INACCESSIBLE, true }, | |
189 | { "/root", MOUNT_INACCESSIBLE, true }, | |
b6c432ca DH |
190 | }; |
191 | ||
f471b2af | 192 | /* ProtectSystem=yes table */ |
34de407a | 193 | static const MountEntry protect_system_yes_table[] = { |
a868e437 LP |
194 | { "/usr", MOUNT_READ_ONLY, false }, |
195 | { "/boot", MOUNT_READ_ONLY, true }, | |
196 | { "/efi", MOUNT_READ_ONLY, true }, | |
f471b2af DH |
197 | }; |
198 | ||
199 | /* ProtectSystem=full includes ProtectSystem=yes */ | |
34de407a | 200 | static const MountEntry protect_system_full_table[] = { |
a868e437 LP |
201 | { "/usr", MOUNT_READ_ONLY, false }, |
202 | { "/boot", MOUNT_READ_ONLY, true }, | |
203 | { "/efi", MOUNT_READ_ONLY, true }, | |
204 | { "/etc", MOUNT_READ_ONLY, false }, | |
f471b2af DH |
205 | }; |
206 | ||
3d1b999b LP |
207 | /* ProtectSystem=strict table. In this strict mode, we mount everything read-only, except for /proc, /dev, |
208 | * /sys which are the kernel API VFS, which are left writable, but PrivateDevices= + ProtectKernelTunables= | |
209 | * protect those, and these options should be fully orthogonal. (And of course /home and friends are also | |
210 | * left writable, as ProtectHome= shall manage those, orthogonally). | |
f471b2af | 211 | */ |
34de407a | 212 | static const MountEntry protect_system_strict_table[] = { |
a868e437 LP |
213 | { "/", MOUNT_READ_ONLY, false }, |
214 | { "/proc", MOUNT_READ_WRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
215 | { "/sys", MOUNT_READ_WRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
216 | { "/dev", MOUNT_READ_WRITE_IMPLICIT, false }, /* PrivateDevices= */ | |
217 | { "/home", MOUNT_READ_WRITE_IMPLICIT, true }, /* ProtectHome= */ | |
218 | { "/run/user", MOUNT_READ_WRITE_IMPLICIT, true }, /* ProtectHome= */ | |
219 | { "/root", MOUNT_READ_WRITE_IMPLICIT, true }, /* ProtectHome= */ | |
f471b2af DH |
220 | }; |
221 | ||
3d1b999b LP |
222 | /* ProtectHostname=yes able */ |
223 | static const MountEntry protect_hostname_table[] = { | |
a868e437 LP |
224 | { "/proc/sys/kernel/hostname", MOUNT_READ_ONLY, false }, |
225 | { "/proc/sys/kernel/domainname", MOUNT_READ_ONLY, false }, | |
3d1b999b LP |
226 | }; |
227 | ||
5beb8688 | 228 | static const char * const mount_mode_table[_MOUNT_MODE_MAX] = { |
a868e437 LP |
229 | [MOUNT_INACCESSIBLE] = "inaccessible", |
230 | [MOUNT_OVERLAY] = "overlay", | |
231 | [MOUNT_IMAGE] = "image", | |
232 | [MOUNT_BIND] = "bind", | |
233 | [MOUNT_BIND_RECURSIVE] = "bind-recursive", | |
234 | [MOUNT_PRIVATE_TMP] = "private-tmp", | |
235 | [MOUNT_PRIVATE_TMP_READ_ONLY] = "private-tmp-read-only", | |
236 | [MOUNT_PRIVATE_DEV] = "private-dev", | |
237 | [MOUNT_BIND_DEV] = "bind-dev", | |
238 | [MOUNT_EMPTY_DIR] = "empty-dir", | |
239 | [MOUNT_PRIVATE_SYSFS] = "private-sysfs", | |
240 | [MOUNT_BIND_SYSFS] = "bind-sysfs", | |
241 | [MOUNT_PROCFS] = "procfs", | |
242 | [MOUNT_READ_ONLY] = "read-only", | |
243 | [MOUNT_READ_WRITE] = "read-write", | |
244 | [MOUNT_NOEXEC] = "noexec", | |
245 | [MOUNT_EXEC] = "exec", | |
246 | [MOUNT_TMPFS] = "tmpfs", | |
247 | [MOUNT_RUN] = "run", | |
248 | [MOUNT_EXTENSION_DIRECTORY] = "extension-directory", | |
249 | [MOUNT_EXTENSION_IMAGE] = "extension-image", | |
250 | [MOUNT_MQUEUEFS] = "mqueuefs", | |
251 | [MOUNT_READ_WRITE_IMPLICIT] = "read-write-implicit", | |
5beb8688 YW |
252 | }; |
253 | ||
55ea4ef0 MG |
254 | /* Helper struct for naming simplicity and reusability */ |
255 | static const struct { | |
256 | const char *level_env; | |
257 | const char *level_env_print; | |
258 | } image_class_info[_IMAGE_CLASS_MAX] = { | |
259 | [IMAGE_SYSEXT] = { | |
260 | .level_env = "SYSEXT_LEVEL", | |
261 | .level_env_print = " SYSEXT_LEVEL=", | |
262 | }, | |
263 | [IMAGE_CONFEXT] = { | |
264 | .level_env = "CONFEXT_LEVEL", | |
265 | .level_env_print = " CONFEXT_LEVEL=", | |
266 | } | |
267 | }; | |
268 | ||
5beb8688 YW |
269 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode, MountMode); |
270 | ||
34de407a | 271 | static const char *mount_entry_path(const MountEntry *p) { |
f0a4feb0 DH |
272 | assert(p); |
273 | ||
5327c910 LP |
274 | /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that, |
275 | * otherwise the stack/static ->path field is returned. */ | |
f0a4feb0 | 276 | |
5327c910 | 277 | return p->path_malloc ?: p->path_const; |
f0a4feb0 DH |
278 | } |
279 | ||
809ceb82 LB |
280 | static const char *mount_entry_unprefixed_path(const MountEntry *p) { |
281 | assert(p); | |
282 | ||
283 | /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */ | |
284 | ||
285 | return p->unprefixed_path_malloc ?: p->unprefixed_path_const ?: mount_entry_path(p); | |
286 | } | |
287 | ||
288 | static void mount_entry_consume_prefix(MountEntry *p, char *new_path) { | |
289 | assert(p); | |
290 | assert(p->path_malloc || p->path_const); | |
291 | assert(new_path); | |
292 | ||
293 | /* Saves current path in unprefixed_ variable, and takes over new_path */ | |
294 | ||
295 | free_and_replace(p->unprefixed_path_malloc, p->path_malloc); | |
296 | /* If we didn't have a path on the heap, then it's a static one */ | |
297 | if (!p->unprefixed_path_malloc) | |
298 | p->unprefixed_path_const = p->path_const; | |
299 | p->path_malloc = new_path; | |
300 | p->has_prefix = true; | |
301 | } | |
302 | ||
34de407a | 303 | static bool mount_entry_read_only(const MountEntry *p) { |
cfbeb4ef LP |
304 | assert(p); |
305 | ||
a868e437 | 306 | return p->read_only || IN_SET(p->mode, MOUNT_READ_ONLY, MOUNT_INACCESSIBLE, MOUNT_PRIVATE_TMP_READ_ONLY); |
cfbeb4ef LP |
307 | } |
308 | ||
ddc155b2 TM |
309 | static bool mount_entry_noexec(const MountEntry *p) { |
310 | assert(p); | |
311 | ||
a868e437 | 312 | return p->noexec || IN_SET(p->mode, MOUNT_NOEXEC, MOUNT_INACCESSIBLE, MOUNT_PRIVATE_SYSFS, MOUNT_BIND_SYSFS, MOUNT_PROCFS); |
ddc155b2 TM |
313 | } |
314 | ||
315 | static bool mount_entry_exec(const MountEntry *p) { | |
316 | assert(p); | |
317 | ||
a868e437 | 318 | return p->exec || p->mode == MOUNT_EXEC; |
ddc155b2 TM |
319 | } |
320 | ||
d2d6c096 LP |
321 | static const char *mount_entry_source(const MountEntry *p) { |
322 | assert(p); | |
323 | ||
324 | return p->source_malloc ?: p->source_const; | |
325 | } | |
326 | ||
2abd4e38 YW |
327 | static const char *mount_entry_options(const MountEntry *p) { |
328 | assert(p); | |
329 | ||
330 | return p->options_malloc ?: p->options_const; | |
331 | } | |
332 | ||
1eb7e08e LP |
333 | static void mount_entry_done(MountEntry *p) { |
334 | assert(p); | |
335 | ||
336 | p->path_malloc = mfree(p->path_malloc); | |
809ceb82 | 337 | p->unprefixed_path_malloc = mfree(p->unprefixed_path_malloc); |
1eb7e08e | 338 | p->source_malloc = mfree(p->source_malloc); |
2abd4e38 | 339 | p->options_malloc = mfree(p->options_malloc); |
f0304df6 | 340 | p->overlay_layers = strv_free(p->overlay_layers); |
1eb7e08e LP |
341 | } |
342 | ||
063c977a LP |
343 | static void mount_list_done(MountList *ml) { |
344 | assert(ml); | |
345 | ||
346 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) | |
347 | mount_entry_done(m); | |
348 | ||
349 | ml->mounts = mfree(ml->mounts); | |
350 | ml->n_mounts = 0; | |
351 | } | |
352 | ||
353 | static MountEntry *mount_list_extend(MountList *ml) { | |
354 | assert(ml); | |
355 | ||
356 | if (!GREEDY_REALLOC0(ml->mounts, ml->n_mounts+1)) | |
357 | return NULL; | |
358 | ||
359 | return ml->mounts + ml->n_mounts++; | |
360 | } | |
361 | ||
362 | static int append_access_mounts(MountList *ml, char **strv, MountMode mode, bool forcibly_require_prefix) { | |
363 | assert(ml); | |
613b411c | 364 | |
a868e437 | 365 | /* Adds a list of user-supplied READ_WRITE/READ_WRITE_IMPLICIT/READ_ONLY/INACCESSIBLE entries */ |
5327c910 | 366 | |
15ae422b | 367 | STRV_FOREACH(i, strv) { |
5327c910 LP |
368 | bool ignore = false, needs_prefix = false; |
369 | const char *e = *i; | |
15ae422b | 370 | |
5327c910 LP |
371 | /* Look for any prefixes */ |
372 | if (startswith(e, "-")) { | |
373 | e++; | |
9c94d52e | 374 | ignore = true; |
ea92ae33 | 375 | } |
5327c910 LP |
376 | if (startswith(e, "+")) { |
377 | e++; | |
378 | needs_prefix = true; | |
379 | } | |
ea92ae33 | 380 | |
baaa35ad | 381 | if (!path_is_absolute(e)) |
063c977a | 382 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not absolute: %s", e); |
15ae422b | 383 | |
063c977a LP |
384 | MountEntry *me = mount_list_extend(ml); |
385 | if (!me) | |
386 | return log_oom_debug(); | |
387 | ||
388 | *me = (MountEntry) { | |
5327c910 LP |
389 | .path_const = e, |
390 | .mode = mode, | |
391 | .ignore = ignore, | |
d18aff04 | 392 | .has_prefix = !needs_prefix && !forcibly_require_prefix, |
5327c910 | 393 | }; |
15ae422b LP |
394 | } |
395 | ||
396 | return 0; | |
397 | } | |
398 | ||
063c977a LP |
399 | static int append_empty_dir_mounts(MountList *ml, char **strv) { |
400 | assert(ml); | |
6c47cd7d LP |
401 | |
402 | /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the | |
403 | * "/private/" boundary directories for DynamicUser=1. */ | |
404 | ||
405 | STRV_FOREACH(i, strv) { | |
063c977a LP |
406 | MountEntry *me = mount_list_extend(ml); |
407 | if (!me) | |
408 | return log_oom_debug(); | |
6c47cd7d | 409 | |
063c977a | 410 | *me = (MountEntry) { |
6c47cd7d | 411 | .path_const = *i, |
a868e437 | 412 | .mode = MOUNT_EMPTY_DIR, |
6c47cd7d | 413 | .ignore = false, |
6c47cd7d | 414 | .read_only = true, |
9f563f27 | 415 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, |
2abd4e38 | 416 | .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, |
6c47cd7d LP |
417 | }; |
418 | } | |
419 | ||
420 | return 0; | |
421 | } | |
422 | ||
063c977a LP |
423 | static int append_bind_mounts(MountList *ml, const BindMount *binds, size_t n) { |
424 | assert(ml); | |
425 | assert(binds || n == 0); | |
d2d6c096 | 426 | |
063c977a LP |
427 | FOREACH_ARRAY(b, binds, n) { |
428 | MountEntry *me = mount_list_extend(ml); | |
429 | if (!me) | |
430 | return log_oom_debug(); | |
d2d6c096 | 431 | |
063c977a | 432 | *me = (MountEntry) { |
d2d6c096 | 433 | .path_const = b->destination, |
a868e437 | 434 | .mode = b->recursive ? MOUNT_BIND_RECURSIVE : MOUNT_BIND, |
d2d6c096 | 435 | .read_only = b->read_only, |
9ce4e4b0 | 436 | .nosuid = b->nosuid, |
d2d6c096 | 437 | .source_const = b->source, |
4ca763a9 | 438 | .ignore = b->ignore_enoent, |
d2d6c096 LP |
439 | }; |
440 | } | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
063c977a LP |
445 | static int append_mount_images(MountList *ml, const MountImage *mount_images, size_t n) { |
446 | assert(ml); | |
447 | assert(mount_images || n == 0); | |
b3d13314 | 448 | |
063c977a LP |
449 | FOREACH_ARRAY(m, mount_images, n) { |
450 | MountEntry *me = mount_list_extend(ml); | |
451 | if (!me) | |
452 | return log_oom_debug(); | |
b3d13314 | 453 | |
063c977a | 454 | *me = (MountEntry) { |
b3d13314 | 455 | .path_const = m->destination, |
a868e437 | 456 | .mode = MOUNT_IMAGE, |
b3d13314 | 457 | .source_const = m->source, |
9dc6a6af | 458 | .image_options_const = m->mount_options, |
b3d13314 LB |
459 | .ignore = m->ignore_enoent, |
460 | }; | |
461 | } | |
462 | ||
463 | return 0; | |
464 | } | |
465 | ||
a07b9926 | 466 | static int append_extensions( |
063c977a | 467 | MountList *ml, |
93f59701 LB |
468 | const char *root, |
469 | const char *extension_dir, | |
470 | char **hierarchies, | |
471 | const MountImage *mount_images, | |
a07b9926 LB |
472 | size_t n, |
473 | char **extension_directories) { | |
93f59701 | 474 | |
f0304df6 LB |
475 | char ***overlays = NULL; |
476 | size_t n_overlays = 0; | |
93f59701 LB |
477 | int r; |
478 | ||
063c977a LP |
479 | assert(ml); |
480 | ||
a07b9926 | 481 | if (n == 0 && strv_isempty(extension_directories)) |
93f59701 LB |
482 | return 0; |
483 | ||
24759d8f LB |
484 | assert(extension_dir); |
485 | ||
f0304df6 LB |
486 | n_overlays = strv_length(hierarchies); |
487 | if (n_overlays == 0) | |
488 | return 0; | |
489 | ||
490 | /* Prepare a list of overlays, that will have as each element a strv containing all the layers that | |
491 | * will later be concatenated as a lowerdir= parameter for the mount operation. | |
93f59701 LB |
492 | * The overlays vector will have the same number of elements and will correspond to the |
493 | * hierarchies vector, so they can be iterated upon together. */ | |
f0304df6 LB |
494 | overlays = new0(char**, n_overlays); |
495 | if (!overlays) | |
496 | return -ENOMEM; | |
93f59701 | 497 | |
f0304df6 | 498 | CLEANUP_ARRAY(overlays, n_overlays, strv_free_many); |
93f59701 LB |
499 | |
500 | /* First, prepare a mount for each image, but these won't be visible to the unit, instead | |
501 | * they will be mounted in our propagate directory, and used as a source for the overlay. */ | |
502 | for (size_t i = 0; i < n; i++) { | |
503 | _cleanup_free_ char *mount_point = NULL; | |
504 | const MountImage *m = mount_images + i; | |
505 | ||
063c977a | 506 | if (asprintf(&mount_point, "%s/%zu", extension_dir, i) < 0) |
93f59701 LB |
507 | return -ENOMEM; |
508 | ||
509 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
f0304df6 | 510 | char *prefixed_hierarchy = path_join(mount_point, hierarchies[j]); |
93f59701 LB |
511 | if (!prefixed_hierarchy) |
512 | return -ENOMEM; | |
513 | ||
f0304df6 LB |
514 | r = strv_consume(&overlays[j], TAKE_PTR(prefixed_hierarchy)); |
515 | if (r < 0) | |
516 | return r; | |
93f59701 LB |
517 | } |
518 | ||
063c977a LP |
519 | MountEntry *me = mount_list_extend(ml); |
520 | if (!me) | |
f0304df6 | 521 | return -ENOMEM; |
063c977a LP |
522 | |
523 | *me = (MountEntry) { | |
93f59701 | 524 | .path_malloc = TAKE_PTR(mount_point), |
9dc6a6af | 525 | .image_options_const = m->mount_options, |
93f59701 LB |
526 | .ignore = m->ignore_enoent, |
527 | .source_const = m->source, | |
a868e437 | 528 | .mode = MOUNT_EXTENSION_IMAGE, |
93f59701 LB |
529 | .has_prefix = true, |
530 | }; | |
531 | } | |
532 | ||
a07b9926 LB |
533 | /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory. |
534 | * Bind mount them in the same location as the ExtensionImages, so that we | |
535 | * can check that they are valid trees (extension-release.d). */ | |
536 | STRV_FOREACH(extension_directory, extension_directories) { | |
537 | _cleanup_free_ char *mount_point = NULL, *source = NULL; | |
538 | const char *e = *extension_directory; | |
539 | bool ignore_enoent = false; | |
540 | ||
541 | /* Pick up the counter where the ExtensionImages left it. */ | |
063c977a | 542 | if (asprintf(&mount_point, "%s/%zu", extension_dir, n++) < 0) |
a07b9926 LB |
543 | return -ENOMEM; |
544 | ||
545 | /* Look for any prefixes */ | |
546 | if (startswith(e, "-")) { | |
547 | e++; | |
548 | ignore_enoent = true; | |
549 | } | |
550 | /* Ignore this for now */ | |
551 | if (startswith(e, "+")) | |
552 | e++; | |
553 | ||
554 | source = strdup(e); | |
555 | if (!source) | |
556 | return -ENOMEM; | |
557 | ||
558 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
f0304df6 | 559 | char *prefixed_hierarchy = path_join(mount_point, hierarchies[j]); |
a07b9926 LB |
560 | if (!prefixed_hierarchy) |
561 | return -ENOMEM; | |
562 | ||
f0304df6 LB |
563 | r = strv_consume(&overlays[j], TAKE_PTR(prefixed_hierarchy)); |
564 | if (r < 0) | |
565 | return r; | |
a07b9926 LB |
566 | } |
567 | ||
063c977a LP |
568 | MountEntry *me = mount_list_extend(ml); |
569 | if (!me) | |
f0304df6 | 570 | return -ENOMEM; |
063c977a LP |
571 | |
572 | *me = (MountEntry) { | |
a07b9926 | 573 | .path_malloc = TAKE_PTR(mount_point), |
c9423746 | 574 | .source_malloc = TAKE_PTR(source), |
a868e437 | 575 | .mode = MOUNT_EXTENSION_DIRECTORY, |
a07b9926 LB |
576 | .ignore = ignore_enoent, |
577 | .has_prefix = true, | |
578 | .read_only = true, | |
579 | }; | |
580 | } | |
581 | ||
93f59701 LB |
582 | /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings |
583 | * set up earlier. */ | |
584 | for (size_t i = 0; hierarchies && hierarchies[i]; ++i) { | |
585 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
586 | ||
587 | prefixed_hierarchy = path_join(root, hierarchies[i]); | |
588 | if (!prefixed_hierarchy) | |
589 | return -ENOMEM; | |
590 | ||
063c977a LP |
591 | MountEntry *me = mount_list_extend(ml); |
592 | if (!me) | |
f0304df6 | 593 | return -ENOMEM; |
063c977a LP |
594 | |
595 | *me = (MountEntry) { | |
93f59701 | 596 | .path_malloc = TAKE_PTR(prefixed_hierarchy), |
f0304df6 | 597 | .overlay_layers = TAKE_PTR(overlays[i]), |
a868e437 | 598 | .mode = MOUNT_OVERLAY, |
93f59701 LB |
599 | .has_prefix = true, |
600 | .ignore = true, /* If the source image doesn't set the ignore bit it will fail earlier. */ | |
601 | }; | |
602 | } | |
603 | ||
604 | return 0; | |
605 | } | |
606 | ||
063c977a LP |
607 | static int append_tmpfs_mounts(MountList *ml, const TemporaryFileSystem *tmpfs, size_t n) { |
608 | assert(ml); | |
609 | assert(tmpfs || n == 0); | |
2abd4e38 | 610 | |
063c977a | 611 | FOREACH_ARRAY(t, tmpfs, n) { |
2abd4e38 | 612 | _cleanup_free_ char *o = NULL, *str = NULL; |
ad8e66dc | 613 | unsigned long flags; |
2abd4e38 | 614 | bool ro = false; |
b67ec8e5 | 615 | int r; |
2abd4e38 | 616 | |
baaa35ad | 617 | if (!path_is_absolute(t->path)) |
063c977a | 618 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not absolute: %s", t->path); |
2abd4e38 | 619 | |
b67ec8e5 | 620 | str = strjoin("mode=0755" NESTED_TMPFS_LIMITS ",", t->options); |
ad8e66dc AJ |
621 | if (!str) |
622 | return -ENOMEM; | |
2abd4e38 | 623 | |
ad8e66dc AJ |
624 | r = mount_option_mangle(str, MS_NODEV|MS_STRICTATIME, &flags, &o); |
625 | if (r < 0) | |
626 | return log_debug_errno(r, "Failed to parse mount option '%s': %m", str); | |
2abd4e38 | 627 | |
ad8e66dc AJ |
628 | ro = flags & MS_RDONLY; |
629 | if (ro) | |
630 | flags ^= MS_RDONLY; | |
2abd4e38 | 631 | |
063c977a LP |
632 | MountEntry *me = mount_list_extend(ml); |
633 | if (!me) | |
634 | return log_oom_debug(); | |
635 | ||
636 | *me = (MountEntry) { | |
2abd4e38 | 637 | .path_const = t->path, |
a868e437 | 638 | .mode = MOUNT_TMPFS, |
2abd4e38 | 639 | .read_only = ro, |
ad8e66dc | 640 | .options_malloc = TAKE_PTR(o), |
2abd4e38 YW |
641 | .flags = flags, |
642 | }; | |
2abd4e38 YW |
643 | } |
644 | ||
645 | return 0; | |
646 | } | |
647 | ||
063c977a LP |
648 | static int append_static_mounts(MountList *ml, const MountEntry *mounts, size_t n, bool ignore_protect) { |
649 | assert(ml); | |
650 | assert(mounts || n == 0); | |
11a30cec | 651 | |
5327c910 | 652 | /* Adds a list of static pre-defined entries */ |
f471b2af | 653 | |
063c977a LP |
654 | FOREACH_ARRAY(m, mounts, n) { |
655 | MountEntry *me = mount_list_extend(ml); | |
656 | if (!me) | |
657 | return log_oom_debug(); | |
658 | ||
659 | *me = (MountEntry) { | |
660 | .path_const = mount_entry_path(m), | |
661 | .mode = m->mode, | |
662 | .ignore = m->ignore || ignore_protect, | |
5327c910 | 663 | }; |
063c977a | 664 | } |
f471b2af DH |
665 | |
666 | return 0; | |
667 | } | |
668 | ||
063c977a LP |
669 | static int append_protect_home(MountList *ml, ProtectHome protect_home, bool ignore_protect) { |
670 | assert(ml); | |
c575770b | 671 | |
5327c910 | 672 | switch (protect_home) { |
b6c432ca | 673 | |
5327c910 | 674 | case PROTECT_HOME_NO: |
b6c432ca DH |
675 | return 0; |
676 | ||
b6c432ca | 677 | case PROTECT_HOME_READ_ONLY: |
063c977a | 678 | return append_static_mounts(ml, protect_home_read_only_table, ELEMENTSOF(protect_home_read_only_table), ignore_protect); |
5327c910 | 679 | |
e4da7d8c | 680 | case PROTECT_HOME_TMPFS: |
063c977a | 681 | return append_static_mounts(ml, protect_home_tmpfs_table, ELEMENTSOF(protect_home_tmpfs_table), ignore_protect); |
e4da7d8c | 682 | |
b6c432ca | 683 | case PROTECT_HOME_YES: |
063c977a | 684 | return append_static_mounts(ml, protect_home_yes_table, ELEMENTSOF(protect_home_yes_table), ignore_protect); |
5327c910 | 685 | |
b6c432ca | 686 | default: |
04499a70 | 687 | assert_not_reached(); |
b6c432ca | 688 | } |
b6c432ca DH |
689 | } |
690 | ||
063c977a LP |
691 | static int append_protect_system(MountList *ml, ProtectSystem protect_system, bool ignore_protect) { |
692 | assert(ml); | |
f471b2af | 693 | |
5327c910 LP |
694 | switch (protect_system) { |
695 | ||
696 | case PROTECT_SYSTEM_NO: | |
f471b2af DH |
697 | return 0; |
698 | ||
f471b2af | 699 | case PROTECT_SYSTEM_STRICT: |
063c977a | 700 | return append_static_mounts(ml, protect_system_strict_table, ELEMENTSOF(protect_system_strict_table), ignore_protect); |
5327c910 | 701 | |
f471b2af | 702 | case PROTECT_SYSTEM_YES: |
063c977a | 703 | return append_static_mounts(ml, protect_system_yes_table, ELEMENTSOF(protect_system_yes_table), ignore_protect); |
5327c910 | 704 | |
f471b2af | 705 | case PROTECT_SYSTEM_FULL: |
063c977a | 706 | return append_static_mounts(ml, protect_system_full_table, ELEMENTSOF(protect_system_full_table), ignore_protect); |
5327c910 | 707 | |
f471b2af | 708 | default: |
04499a70 | 709 | assert_not_reached(); |
f471b2af | 710 | } |
11a30cec DH |
711 | } |
712 | ||
93bab288 | 713 | static int mount_path_compare(const MountEntry *a, const MountEntry *b) { |
a0827e2b | 714 | int d; |
15ae422b | 715 | |
a07b9926 | 716 | /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first |
93f59701 | 717 | * regardless of the prefix - they are set up in the propagate directory anyway */ |
a868e437 | 718 | d = -CMP(a->mode == MOUNT_EXTENSION_IMAGE, b->mode == MOUNT_EXTENSION_IMAGE); |
a07b9926 LB |
719 | if (d != 0) |
720 | return d; | |
a868e437 | 721 | d = -CMP(a->mode == MOUNT_EXTENSION_DIRECTORY, b->mode == MOUNT_EXTENSION_DIRECTORY); |
93f59701 LB |
722 | if (d != 0) |
723 | return d; | |
724 | ||
6ee1a919 | 725 | /* If the paths are not equal, then order prefixes first */ |
93bab288 | 726 | d = path_compare(mount_entry_path(a), mount_entry_path(b)); |
6ee1a919 LP |
727 | if (d != 0) |
728 | return d; | |
15ae422b | 729 | |
6ee1a919 | 730 | /* If the paths are equal, check the mode */ |
93bab288 | 731 | return CMP((int) a->mode, (int) b->mode); |
15ae422b LP |
732 | } |
733 | ||
063c977a | 734 | static int prefix_where_needed(MountList *ml, const char *root_directory) { |
4a756839 | 735 | /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */ |
5327c910 | 736 | |
063c977a | 737 | assert(ml); |
fe96c0f8 | 738 | |
063c977a | 739 | FOREACH_ARRAY(me, ml->mounts, ml->n_mounts) { |
5327c910 LP |
740 | char *s; |
741 | ||
063c977a | 742 | if (me->has_prefix) |
5327c910 LP |
743 | continue; |
744 | ||
063c977a | 745 | s = path_join(root_directory, mount_entry_path(me)); |
5327c910 LP |
746 | if (!s) |
747 | return -ENOMEM; | |
748 | ||
063c977a | 749 | mount_entry_consume_prefix(me, s); |
5327c910 LP |
750 | } |
751 | ||
752 | return 0; | |
753 | } | |
754 | ||
063c977a | 755 | static void drop_duplicates(MountList *ml) { |
34de407a | 756 | MountEntry *f, *t, *previous; |
15ae422b | 757 | |
063c977a | 758 | assert(ml); |
15ae422b | 759 | |
fe3c2583 LP |
760 | /* Drops duplicate entries. Expects that the array is properly ordered already. */ |
761 | ||
063c977a | 762 | for (f = ml->mounts, t = ml->mounts, previous = NULL; f < ml->mounts + ml->n_mounts; f++) { |
15ae422b | 763 | |
fe3c2583 | 764 | /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare() |
088696fe LP |
765 | * above. Note that we only drop duplicates that haven't been mounted yet. */ |
766 | if (previous && | |
767 | path_equal(mount_entry_path(f), mount_entry_path(previous)) && | |
63862de4 | 768 | f->state == MOUNT_PENDING && previous->state == MOUNT_PENDING) { |
5beb8688 | 769 | log_debug("%s (%s) is duplicate.", mount_entry_path(f), mount_mode_to_string(f->mode)); |
ddc155b2 TM |
770 | /* Propagate the flags to the remaining entry */ |
771 | previous->read_only = previous->read_only || mount_entry_read_only(f); | |
772 | previous->noexec = previous->noexec || mount_entry_noexec(f); | |
773 | previous->exec = previous->exec || mount_entry_exec(f); | |
1eb7e08e | 774 | mount_entry_done(f); |
15ae422b | 775 | continue; |
fe3c2583 | 776 | } |
15ae422b | 777 | |
e2d7c1a0 | 778 | *t = *f; |
15ae422b | 779 | previous = t; |
fe3c2583 LP |
780 | t++; |
781 | } | |
782 | ||
063c977a | 783 | ml->n_mounts = t - ml->mounts; |
fe3c2583 LP |
784 | } |
785 | ||
063c977a | 786 | static void drop_inaccessible(MountList *ml) { |
34de407a | 787 | MountEntry *f, *t; |
fe3c2583 LP |
788 | const char *clear = NULL; |
789 | ||
063c977a | 790 | assert(ml); |
fe3c2583 LP |
791 | |
792 | /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly | |
793 | * ordered already. */ | |
794 | ||
063c977a | 795 | for (f = ml->mounts, t = ml->mounts; f < ml->mounts + ml->n_mounts; f++) { |
fe3c2583 LP |
796 | |
797 | /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop | |
798 | * it, as inaccessible paths really should drop the entire subtree. */ | |
34de407a LP |
799 | if (clear && path_startswith(mount_entry_path(f), clear)) { |
800 | log_debug("%s is masked by %s.", mount_entry_path(f), clear); | |
1eb7e08e | 801 | mount_entry_done(f); |
fe3c2583 LP |
802 | continue; |
803 | } | |
15ae422b | 804 | |
a868e437 | 805 | clear = f->mode == MOUNT_INACCESSIBLE ? mount_entry_path(f) : NULL; |
fe3c2583 LP |
806 | |
807 | *t = *f; | |
15ae422b LP |
808 | t++; |
809 | } | |
810 | ||
063c977a | 811 | ml->n_mounts = t - ml->mounts; |
15ae422b LP |
812 | } |
813 | ||
063c977a | 814 | static void drop_nop(MountList *ml) { |
34de407a | 815 | MountEntry *f, *t; |
7648a565 | 816 | |
063c977a | 817 | assert(ml); |
7648a565 LP |
818 | |
819 | /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the | |
820 | * list is ordered by prefixes. */ | |
821 | ||
063c977a | 822 | for (f = ml->mounts, t = ml->mounts; f < ml->mounts + ml->n_mounts; f++) { |
7648a565 | 823 | |
a868e437 LP |
824 | /* Only suppress such subtrees for READ_ONLY, READ_WRITE and READ_WRITE_IMPLICIT entries */ |
825 | if (IN_SET(f->mode, MOUNT_READ_ONLY, MOUNT_READ_WRITE, MOUNT_READ_WRITE_IMPLICIT)) { | |
e7bf2fca | 826 | MountEntry *found = NULL; |
7648a565 LP |
827 | |
828 | /* Now let's find the first parent of the entry we are looking at. */ | |
063c977a | 829 | for (MountEntry *p = PTR_SUB1(t, ml->mounts); p; p = PTR_SUB1(p, ml->mounts)) |
34de407a | 830 | if (path_startswith(mount_entry_path(f), mount_entry_path(p))) { |
e7bf2fca | 831 | found = p; |
7648a565 LP |
832 | break; |
833 | } | |
7648a565 LP |
834 | |
835 | /* We found it, let's see if it's the same mode, if so, we can drop this entry */ | |
e7bf2fca | 836 | if (found && found->mode == f->mode) { |
5beb8688 YW |
837 | log_debug("%s (%s) is made redundant by %s (%s)", |
838 | mount_entry_path(f), mount_mode_to_string(f->mode), | |
e7bf2fca | 839 | mount_entry_path(found), mount_mode_to_string(found->mode)); |
1eb7e08e | 840 | mount_entry_done(f); |
7648a565 LP |
841 | continue; |
842 | } | |
843 | } | |
844 | ||
845 | *t = *f; | |
846 | t++; | |
847 | } | |
848 | ||
063c977a | 849 | ml->n_mounts = t - ml->mounts; |
7648a565 LP |
850 | } |
851 | ||
063c977a | 852 | static void drop_outside_root(MountList *ml, const char *root_directory) { |
34de407a | 853 | MountEntry *f, *t; |
cd2902c9 | 854 | |
063c977a | 855 | assert(ml); |
cd2902c9 | 856 | |
1d54cd5d | 857 | /* Nothing to do */ |
cd2902c9 LP |
858 | if (!root_directory) |
859 | return; | |
860 | ||
861 | /* Drops all mounts that are outside of the root directory. */ | |
862 | ||
063c977a | 863 | for (f = ml->mounts, t = ml->mounts; f < ml->mounts + ml->n_mounts; f++) { |
cd2902c9 | 864 | |
a07b9926 | 865 | /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */ |
a868e437 | 866 | if (!IN_SET(f->mode, MOUNT_EXTENSION_IMAGE, MOUNT_EXTENSION_DIRECTORY) && !path_startswith(mount_entry_path(f), root_directory)) { |
34de407a | 867 | log_debug("%s is outside of root directory.", mount_entry_path(f)); |
1eb7e08e | 868 | mount_entry_done(f); |
cd2902c9 LP |
869 | continue; |
870 | } | |
871 | ||
872 | *t = *f; | |
873 | t++; | |
874 | } | |
875 | ||
063c977a | 876 | ml->n_mounts = t - ml->mounts; |
cd2902c9 LP |
877 | } |
878 | ||
b2a60844 LP |
879 | static int clone_device_node( |
880 | const char *d, | |
881 | const char *temporary_mount, | |
882 | bool *make_devnode) { | |
883 | ||
884 | _cleanup_free_ char *sl = NULL; | |
885 | const char *dn, *bn, *t; | |
b5e99f23 ДГ |
886 | struct stat st; |
887 | int r; | |
888 | ||
414b304b | 889 | if (stat(d, &st) < 0) { |
b2a60844 LP |
890 | if (errno == ENOENT) { |
891 | log_debug_errno(errno, "Device node '%s' to clone does not exist, ignoring.", d); | |
af984e13 | 892 | return -ENXIO; |
b2a60844 LP |
893 | } |
894 | ||
895 | return log_debug_errno(errno, "Failed to stat() device node '%s' to clone, ignoring: %m", d); | |
b5e99f23 ДГ |
896 | } |
897 | ||
898 | if (!S_ISBLK(st.st_mode) && | |
baaa35ad ZJS |
899 | !S_ISCHR(st.st_mode)) |
900 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
901 | "Device node '%s' to clone is not a device node, ignoring.", | |
902 | d); | |
b5e99f23 | 903 | |
6f7f3a33 | 904 | dn = strjoina(temporary_mount, d); |
b5e99f23 | 905 | |
b2a60844 | 906 | /* First, try to create device node properly */ |
16498617 CB |
907 | if (*make_devnode) { |
908 | mac_selinux_create_file_prepare(d, st.st_mode); | |
909 | r = mknod(dn, st.st_mode, st.st_rdev); | |
910 | mac_selinux_create_file_clear(); | |
b2a60844 LP |
911 | if (r >= 0) |
912 | goto add_symlink; | |
16498617 CB |
913 | if (errno != EPERM) |
914 | return log_debug_errno(errno, "mknod failed for %s: %m", d); | |
915 | ||
b2a60844 | 916 | /* This didn't work, let's not try this again for the next iterations. */ |
16498617 CB |
917 | *make_devnode = false; |
918 | } | |
919 | ||
d73020f2 | 920 | /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target. |
1acf344d | 921 | * Do not prepare device-node SELinux label (see issue 13762) */ |
16498617 | 922 | r = mknod(dn, S_IFREG, 0); |
16498617 | 923 | if (r < 0 && errno != EEXIST) |
b2a60844 | 924 | return log_debug_errno(errno, "mknod() fallback failed for '%s': %m", d); |
16498617 | 925 | |
21935150 LP |
926 | /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard |
927 | * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or | |
928 | * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */ | |
929 | r = mount_nofollow_verbose(LOG_DEBUG, d, dn, NULL, MS_BIND, NULL); | |
930 | if (r < 0) | |
931 | return r; | |
b2a60844 LP |
932 | |
933 | add_symlink: | |
934 | bn = path_startswith(d, "/dev/"); | |
935 | if (!bn) | |
936 | return 0; | |
937 | ||
938 | /* Create symlinks like /dev/char/1:9 → ../urandom */ | |
ec61371f | 939 | if (asprintf(&sl, "%s/dev/%s/" DEVNUM_FORMAT_STR, |
cbc056c8 ZJS |
940 | temporary_mount, |
941 | S_ISCHR(st.st_mode) ? "char" : "block", | |
ec61371f | 942 | DEVNUM_FORMAT_VAL(st.st_rdev)) < 0) |
d4f0878e | 943 | return log_oom_debug(); |
b2a60844 LP |
944 | |
945 | (void) mkdir_parents(sl, 0755); | |
946 | ||
947 | t = strjoina("../", bn); | |
b2a60844 | 948 | if (symlink(t, sl) < 0) |
2e4a4fae | 949 | log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl); |
b5e99f23 | 950 | |
af984e13 | 951 | return 0; |
b5e99f23 ДГ |
952 | } |
953 | ||
cd7f3702 DDM |
954 | static char *settle_runtime_dir(RuntimeScope scope) { |
955 | char *runtime_dir; | |
956 | ||
957 | if (scope != RUNTIME_SCOPE_USER) | |
958 | return strdup("/run/"); | |
959 | ||
960 | if (asprintf(&runtime_dir, "/run/user/" UID_FMT, geteuid()) < 0) | |
961 | return NULL; | |
962 | ||
963 | return runtime_dir; | |
964 | } | |
965 | ||
0c3d606c YW |
966 | static int create_temporary_mount_point(RuntimeScope scope, char **ret) { |
967 | _cleanup_free_ char *runtime_dir = NULL, *temporary_mount = NULL; | |
968 | ||
969 | assert(ret); | |
970 | ||
971 | runtime_dir = settle_runtime_dir(scope); | |
972 | if (!runtime_dir) | |
973 | return log_oom_debug(); | |
974 | ||
975 | temporary_mount = path_join(runtime_dir, "systemd/namespace-XXXXXX"); | |
976 | if (!temporary_mount) | |
977 | return log_oom_debug(); | |
978 | ||
979 | if (!mkdtemp(temporary_mount)) | |
980 | return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount); | |
981 | ||
982 | *ret = TAKE_PTR(temporary_mount); | |
983 | return 0; | |
984 | } | |
985 | ||
cd7f3702 | 986 | static int mount_private_dev(MountEntry *m, RuntimeScope scope) { |
7f112f50 LP |
987 | static const char devnodes[] = |
988 | "/dev/null\0" | |
989 | "/dev/zero\0" | |
990 | "/dev/full\0" | |
991 | "/dev/random\0" | |
992 | "/dev/urandom\0" | |
993 | "/dev/tty\0"; | |
994 | ||
0c3d606c | 995 | _cleanup_free_ char *temporary_mount = NULL; |
12e2b70f | 996 | const char *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL; |
16498617 | 997 | bool can_mknod = true; |
7f112f50 LP |
998 | int r; |
999 | ||
1000 | assert(m); | |
1001 | ||
0c3d606c YW |
1002 | r = create_temporary_mount_point(scope, &temporary_mount); |
1003 | if (r < 0) | |
1004 | return r; | |
2b85f4e1 | 1005 | |
63c372cb | 1006 | dev = strjoina(temporary_mount, "/dev"); |
dc751688 | 1007 | (void) mkdir(dev, 0755); |
9f563f27 | 1008 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV); |
21935150 | 1009 | if (r < 0) |
2b85f4e1 | 1010 | goto fail; |
21935150 | 1011 | |
03bc11d1 | 1012 | r = label_fix_full(AT_FDCWD, dev, "/dev", 0); |
c3151977 | 1013 | if (r < 0) { |
1ce268c7 | 1014 | log_debug_errno(r, "Failed to fix label of '%s' as /dev: %m", dev); |
c3151977 TM |
1015 | goto fail; |
1016 | } | |
2b85f4e1 | 1017 | |
63c372cb | 1018 | devpts = strjoina(temporary_mount, "/dev/pts"); |
dc751688 | 1019 | (void) mkdir(devpts, 0755); |
21935150 LP |
1020 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/pts", devpts, NULL, MS_BIND, NULL); |
1021 | if (r < 0) | |
2b85f4e1 | 1022 | goto fail; |
2b85f4e1 | 1023 | |
2e4a4fae YW |
1024 | /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx. |
1025 | * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible. | |
1026 | * Thus, in that case make a clone. | |
1027 | * In nspawn and other containers it will be a symlink, in that case make it a symlink. */ | |
36ce7110 | 1028 | r = is_symlink("/dev/ptmx"); |
2e4a4fae YW |
1029 | if (r < 0) { |
1030 | log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m"); | |
3164e3cb | 1031 | goto fail; |
2e4a4fae | 1032 | } else if (r > 0) { |
414b304b ДГ |
1033 | devptmx = strjoina(temporary_mount, "/dev/ptmx"); |
1034 | if (symlink("pts/ptmx", devptmx) < 0) { | |
2e4a4fae | 1035 | r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx); |
414b304b ДГ |
1036 | goto fail; |
1037 | } | |
1038 | } else { | |
16498617 | 1039 | r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod); |
152c475f LP |
1040 | if (r < 0) |
1041 | goto fail; | |
414b304b | 1042 | } |
e06b6479 | 1043 | |
63c372cb | 1044 | devshm = strjoina(temporary_mount, "/dev/shm"); |
8d953682 | 1045 | (void) mkdir(devshm, 0755); |
21935150 LP |
1046 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/shm", devshm, NULL, MS_BIND, NULL); |
1047 | if (r < 0) | |
2b85f4e1 | 1048 | goto fail; |
2b85f4e1 | 1049 | |
63c372cb | 1050 | devmqueue = strjoina(temporary_mount, "/dev/mqueue"); |
dc751688 | 1051 | (void) mkdir(devmqueue, 0755); |
21935150 | 1052 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/mqueue", devmqueue, NULL, MS_BIND, NULL); |
2b85f4e1 | 1053 | |
63c372cb | 1054 | devhugepages = strjoina(temporary_mount, "/dev/hugepages"); |
dc751688 | 1055 | (void) mkdir(devhugepages, 0755); |
21935150 | 1056 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/hugepages", devhugepages, NULL, MS_BIND, NULL); |
2b85f4e1 | 1057 | |
63c372cb | 1058 | devlog = strjoina(temporary_mount, "/dev/log"); |
2e4a4fae YW |
1059 | if (symlink("/run/systemd/journal/dev-log", devlog) < 0) |
1060 | log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog); | |
82d25240 | 1061 | |
7f112f50 | 1062 | NULSTR_FOREACH(d, devnodes) { |
16498617 | 1063 | r = clone_device_node(d, temporary_mount, &can_mknod); |
37b22b3b | 1064 | /* ENXIO means the *source* is not a device file, skip creation in that case */ |
af984e13 | 1065 | if (r < 0 && r != -ENXIO) |
2b85f4e1 | 1066 | goto fail; |
7f112f50 LP |
1067 | } |
1068 | ||
2e4a4fae YW |
1069 | r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID); |
1070 | if (r < 0) | |
105a1a36 | 1071 | log_debug_errno(r, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount); |
7f112f50 | 1072 | |
d73020f2 LP |
1073 | /* Create the /dev directory if missing. It is more likely to be missing when the service is started |
1074 | * with RootDirectory. This is consistent with mount units creating the mount points when missing. */ | |
34de407a | 1075 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
ee818b89 | 1076 | |
9e5f8252 | 1077 | /* Unmount everything in old /dev */ |
2e4a4fae YW |
1078 | r = umount_recursive(mount_entry_path(m), 0); |
1079 | if (r < 0) | |
1080 | log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m)); | |
1081 | ||
21935150 LP |
1082 | r = mount_nofollow_verbose(LOG_DEBUG, dev, mount_entry_path(m), NULL, MS_MOVE, NULL); |
1083 | if (r < 0) | |
2b85f4e1 | 1084 | goto fail; |
7f112f50 | 1085 | |
1019a48f LP |
1086 | (void) rmdir(dev); |
1087 | (void) rmdir(temporary_mount); | |
7f112f50 | 1088 | |
63862de4 | 1089 | return 1; |
7f112f50 | 1090 | |
2b85f4e1 LP |
1091 | fail: |
1092 | if (devpts) | |
21935150 | 1093 | (void) umount_verbose(LOG_DEBUG, devpts, UMOUNT_NOFOLLOW); |
7f112f50 | 1094 | |
2b85f4e1 | 1095 | if (devshm) |
21935150 | 1096 | (void) umount_verbose(LOG_DEBUG, devshm, UMOUNT_NOFOLLOW); |
7f112f50 | 1097 | |
2b85f4e1 | 1098 | if (devhugepages) |
21935150 | 1099 | (void) umount_verbose(LOG_DEBUG, devhugepages, UMOUNT_NOFOLLOW); |
7f112f50 | 1100 | |
2b85f4e1 | 1101 | if (devmqueue) |
21935150 | 1102 | (void) umount_verbose(LOG_DEBUG, devmqueue, UMOUNT_NOFOLLOW); |
7f112f50 | 1103 | |
21935150 | 1104 | (void) umount_verbose(LOG_DEBUG, dev, UMOUNT_NOFOLLOW); |
1019a48f LP |
1105 | (void) rmdir(dev); |
1106 | (void) rmdir(temporary_mount); | |
7f112f50 | 1107 | |
2b85f4e1 | 1108 | return r; |
7f112f50 LP |
1109 | } |
1110 | ||
2a2969fd | 1111 | static int mount_bind_dev(const MountEntry *m) { |
5d997827 LP |
1112 | int r; |
1113 | ||
1114 | assert(m); | |
1115 | ||
d73020f2 LP |
1116 | /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the |
1117 | * service's /dev. This is only used when RootDirectory= is set. */ | |
5d997827 | 1118 | |
645767d6 LP |
1119 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1120 | ||
5d997827 LP |
1121 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1122 | if (r < 0) | |
1123 | return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m"); | |
1124 | if (r > 0) /* make this a NOP if /dev is already a mount point */ | |
1125 | return 0; | |
1126 | ||
63862de4 LB |
1127 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
1128 | if (r < 0) | |
1129 | return r; | |
1130 | ||
1131 | return 1; | |
5d997827 LP |
1132 | } |
1133 | ||
10028263 | 1134 | static int mount_bind_sysfs(const MountEntry *m) { |
5d997827 LP |
1135 | int r; |
1136 | ||
1137 | assert(m); | |
1138 | ||
645767d6 LP |
1139 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1140 | ||
5d997827 LP |
1141 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1142 | if (r < 0) | |
1143 | return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m"); | |
1144 | if (r > 0) /* make this a NOP if /sys is already a mount point */ | |
1145 | return 0; | |
1146 | ||
1147 | /* Bind mount the host's version so that we get all child mounts of it, too. */ | |
63862de4 LB |
1148 | r = mount_nofollow_verbose(LOG_DEBUG, "/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
1149 | if (r < 0) | |
1150 | return r; | |
1151 | ||
1152 | return 1; | |
5d997827 LP |
1153 | } |
1154 | ||
b18f4eed YW |
1155 | static int mount_private_apivfs( |
1156 | const char *fstype, | |
1157 | const char *entry_path, | |
1158 | const char *bind_source, | |
4793605d YW |
1159 | const char *opts, |
1160 | RuntimeScope scope) { | |
b18f4eed | 1161 | |
4793605d YW |
1162 | _cleanup_(rmdir_and_freep) char *temporary_mount = NULL; |
1163 | int r; | |
553e3660 | 1164 | |
b18f4eed YW |
1165 | assert(fstype); |
1166 | assert(entry_path); | |
1167 | assert(bind_source); | |
553e3660 YW |
1168 | |
1169 | (void) mkdir_p_label(entry_path, 0755); | |
1170 | ||
4793605d YW |
1171 | /* First, check if we have enough privileges to mount a new instance. Note, a new sysfs instance |
1172 | * cannot be mounted on an already existing mount. Let's use a temporary place. */ | |
1173 | r = create_temporary_mount_point(scope, &temporary_mount); | |
1174 | if (r < 0) | |
1175 | return r; | |
553e3660 | 1176 | |
4793605d | 1177 | r = mount_nofollow_verbose(LOG_DEBUG, fstype, temporary_mount, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts); |
b18f4eed YW |
1178 | if (r == -EINVAL && opts) |
1179 | /* If this failed with EINVAL then this likely means the textual hidepid= stuff for procfs is | |
1180 | * not supported by the kernel, and thus the per-instance hidepid= neither, which means we | |
1181 | * really don't want to use it, since it would affect our host's /proc mount. Hence let's | |
1182 | * gracefully fallback to a classic, unrestricted version. */ | |
4793605d | 1183 | r = mount_nofollow_verbose(LOG_DEBUG, fstype, temporary_mount, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, /* opts = */ NULL); |
553e3660 | 1184 | if (ERRNO_IS_NEG_PRIVILEGE(r)) { |
b18f4eed YW |
1185 | /* When we do not have enough privileges to mount a new instance, fall back to use an |
1186 | * existing mount. */ | |
553e3660 | 1187 | |
b18f4eed YW |
1188 | r = path_is_mount_point(entry_path, /* root = */ NULL, /* flags = */ 0); |
1189 | if (r < 0) | |
1190 | return log_debug_errno(r, "Unable to determine whether '%s' is already mounted: %m", entry_path); | |
1191 | if (r > 0) | |
1192 | return 0; /* Use the current mount as is. */ | |
553e3660 | 1193 | |
b18f4eed YW |
1194 | /* We lack permissions to mount a new instance, and it is not already mounted. But we can |
1195 | * access the host's, so as a final fallback bind-mount it to the destination, as most likely | |
1196 | * we are inside a user manager in an unprivileged user namespace. */ | |
63862de4 LB |
1197 | r = mount_nofollow_verbose(LOG_DEBUG, bind_source, entry_path, /* fstype = */ NULL, MS_BIND|MS_REC, /* opts = */ NULL); |
1198 | if (r < 0) | |
1199 | return r; | |
1200 | ||
1201 | return 1; | |
553e3660 YW |
1202 | |
1203 | } else if (r < 0) | |
1204 | return r; | |
1205 | ||
4793605d YW |
1206 | /* OK. We have a new mount instance. Let's clear an existing mount and its submounts. */ |
1207 | r = umount_recursive(entry_path, /* flags = */ 0); | |
1208 | if (r < 0) | |
1209 | log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", entry_path); | |
1210 | ||
1211 | /* Then, move the new mount instance. */ | |
1212 | r = mount_nofollow_verbose(LOG_DEBUG, temporary_mount, entry_path, /* fstype = */ NULL, MS_MOVE, /* opts = */ NULL); | |
1213 | if (r < 0) | |
1214 | return r; | |
1215 | ||
b18f4eed YW |
1216 | /* We mounted a new instance now. Let's bind mount the children over now. This matters for nspawn |
1217 | * where a bunch of files are overmounted, in particular the boot id. */ | |
1218 | (void) bind_mount_submounts(bind_source, entry_path); | |
63862de4 | 1219 | return 1; |
553e3660 YW |
1220 | } |
1221 | ||
4793605d | 1222 | static int mount_private_sysfs(const MountEntry *m, const NamespaceParameters *p) { |
b18f4eed | 1223 | assert(m); |
4793605d YW |
1224 | assert(p); |
1225 | return mount_private_apivfs("sysfs", mount_entry_path(m), "/sys", /* opts = */ NULL, p->runtime_scope); | |
b18f4eed YW |
1226 | } |
1227 | ||
79d956db | 1228 | static int mount_procfs(const MountEntry *m, const NamespaceParameters *p) { |
61f8a7bd | 1229 | _cleanup_free_ char *opts = NULL; |
5d997827 LP |
1230 | |
1231 | assert(m); | |
79d956db | 1232 | assert(p); |
5d997827 | 1233 | |
79d956db LP |
1234 | if (p->protect_proc != PROTECT_PROC_DEFAULT || |
1235 | p->proc_subset != PROC_SUBSET_ALL) { | |
4e399953 LP |
1236 | |
1237 | /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it | |
1238 | * pretended to be per-instance but actually was per-namespace), hence let's make use of it | |
1239 | * if requested. To make sure this logic succeeds only on kernels where hidepid= is | |
1240 | * per-instance, we'll exclusively use the textual value for hidepid=, since support was | |
1241 | * added in the same commit: if it's supported it is thus also per-instance. */ | |
1242 | ||
79d956db | 1243 | const char *hpv = p->protect_proc == PROTECT_PROC_DEFAULT ? |
7c76e181 | 1244 | "off" : |
79d956db | 1245 | protect_proc_to_string(p->protect_proc); |
1c265fcd DDM |
1246 | |
1247 | /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in | |
1248 | * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when | |
1249 | * trying to use hidepid= on systems where it isn't supported. The same applies for subset=. | |
1250 | * fsopen()/fsconfig() was also backported on some distros which allows us to detect | |
1251 | * hidepid=/subset= support in even more scenarios. */ | |
1252 | ||
117e7034 | 1253 | if (mount_option_supported("proc", "hidepid", hpv) != 0) { |
1c265fcd DDM |
1254 | opts = strjoin("hidepid=", hpv); |
1255 | if (!opts) | |
1256 | return -ENOMEM; | |
1257 | } | |
1258 | ||
79d956db | 1259 | if (p->proc_subset == PROC_SUBSET_PID && |
117e7034 | 1260 | mount_option_supported("proc", "subset", "pid") != 0) |
1c265fcd DDM |
1261 | if (!strextend_with_separator(&opts, ",", "subset=pid")) |
1262 | return -ENOMEM; | |
4e399953 LP |
1263 | } |
1264 | ||
61f8a7bd YW |
1265 | /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in |
1266 | * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by | |
1267 | * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything | |
1268 | * mounted on /proc/ first. */ | |
4793605d | 1269 | return mount_private_apivfs("proc", mount_entry_path(m), "/proc", opts, p->runtime_scope); |
5d997827 LP |
1270 | } |
1271 | ||
2abd4e38 | 1272 | static int mount_tmpfs(const MountEntry *m) { |
df6b900a | 1273 | const char *entry_path, *inner_path; |
abad72be | 1274 | int r; |
abad72be | 1275 | |
6c47cd7d LP |
1276 | assert(m); |
1277 | ||
df6b900a | 1278 | entry_path = mount_entry_path(m); |
809ceb82 | 1279 | inner_path = mount_entry_unprefixed_path(m); |
df6b900a | 1280 | |
d73020f2 LP |
1281 | /* First, get rid of everything that is below if there is anything. Then, overmount with our new |
1282 | * tmpfs */ | |
6c47cd7d | 1283 | |
abad72be CG |
1284 | (void) mkdir_p_label(entry_path, 0755); |
1285 | (void) umount_recursive(entry_path, 0); | |
6c47cd7d | 1286 | |
21935150 LP |
1287 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", entry_path, "tmpfs", m->flags, mount_entry_options(m)); |
1288 | if (r < 0) | |
1289 | return r; | |
abad72be | 1290 | |
03bc11d1 | 1291 | r = label_fix_full(AT_FDCWD, entry_path, inner_path, 0); |
abad72be | 1292 | if (r < 0) |
df6b900a | 1293 | return log_debug_errno(r, "Failed to fix label of '%s' as '%s': %m", entry_path, inner_path); |
6c47cd7d | 1294 | |
63862de4 | 1295 | return 1; |
6c47cd7d LP |
1296 | } |
1297 | ||
94293d65 LB |
1298 | static int mount_run(const MountEntry *m) { |
1299 | int r; | |
1300 | ||
1301 | assert(m); | |
1302 | ||
1303 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); | |
1304 | if (r < 0 && r != -ENOENT) | |
1305 | return log_debug_errno(r, "Unable to determine whether /run is already mounted: %m"); | |
1306 | if (r > 0) /* make this a NOP if /run is already a mount point */ | |
1307 | return 0; | |
1308 | ||
1309 | return mount_tmpfs(m); | |
1310 | } | |
1311 | ||
80271a44 XR |
1312 | static int mount_mqueuefs(const MountEntry *m) { |
1313 | int r; | |
1314 | const char *entry_path; | |
1315 | ||
1316 | assert(m); | |
1317 | ||
1318 | entry_path = mount_entry_path(m); | |
1319 | ||
1320 | (void) mkdir_p_label(entry_path, 0755); | |
1321 | (void) umount_recursive(entry_path, 0); | |
1322 | ||
1323 | r = mount_nofollow_verbose(LOG_DEBUG, "mqueue", entry_path, "mqueue", m->flags, mount_entry_options(m)); | |
1324 | if (r < 0) | |
1325 | return r; | |
1326 | ||
63862de4 | 1327 | return 1; |
80271a44 XR |
1328 | } |
1329 | ||
84be0c71 LP |
1330 | static int mount_image( |
1331 | const MountEntry *m, | |
1332 | const char *root_directory, | |
1333 | const ImagePolicy *image_policy) { | |
93f59701 LB |
1334 | |
1335 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
5897469a LB |
1336 | *host_os_release_sysext_level = NULL, *host_os_release_confext_level = NULL, |
1337 | *extension_name = NULL; | |
b3d13314 LB |
1338 | int r; |
1339 | ||
89e62e0b LP |
1340 | assert(m); |
1341 | ||
55ea4ef0 MG |
1342 | r = path_extract_filename(mount_entry_source(m), &extension_name); |
1343 | if (r < 0) | |
1344 | return log_debug_errno(r, "Failed to extract extension name from %s: %m", mount_entry_source(m)); | |
1345 | ||
a868e437 | 1346 | if (m->mode == MOUNT_EXTENSION_IMAGE) { |
93f59701 LB |
1347 | r = parse_os_release( |
1348 | empty_to_root(root_directory), | |
1349 | "ID", &host_os_release_id, | |
1350 | "VERSION_ID", &host_os_release_version_id, | |
5897469a LB |
1351 | image_class_info[IMAGE_SYSEXT].level_env, &host_os_release_sysext_level, |
1352 | image_class_info[IMAGE_CONFEXT].level_env, &host_os_release_confext_level, | |
93f59701 LB |
1353 | NULL); |
1354 | if (r < 0) | |
1355 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
78ab2b50 LB |
1356 | if (isempty(host_os_release_id)) |
1357 | 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 |
1358 | } |
1359 | ||
1360 | r = verity_dissect_and_mount( | |
84be0c71 LP |
1361 | /* src_fd= */ -1, |
1362 | mount_entry_source(m), | |
1363 | mount_entry_path(m), | |
9dc6a6af | 1364 | m->image_options_const, |
84be0c71 LP |
1365 | image_policy, |
1366 | host_os_release_id, | |
1367 | host_os_release_version_id, | |
5897469a LB |
1368 | host_os_release_sysext_level, |
1369 | host_os_release_confext_level, | |
3e107272 LB |
1370 | /* required_sysext_scope= */ NULL, |
1371 | /* ret_image= */ NULL); | |
b850a9b2 LB |
1372 | if (r == -ENOENT && m->ignore) |
1373 | return 0; | |
db4c8a25 | 1374 | if (r == -ESTALE && host_os_release_id) |
7fcd1343 | 1375 | return log_error_errno(r, // FIXME: this should not be logged ad LOG_ERR, as it will result in duplicate logging. |
5897469a | 1376 | "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 |
1377 | mount_entry_source(m), |
1378 | host_os_release_id, | |
1379 | host_os_release_version_id ? " VERSION_ID=" : "", | |
1380 | strempty(host_os_release_version_id), | |
5897469a LB |
1381 | host_os_release_sysext_level ? image_class_info[IMAGE_SYSEXT].level_env_print : "", |
1382 | strempty(host_os_release_sysext_level), | |
1383 | host_os_release_confext_level ? image_class_info[IMAGE_CONFEXT].level_env_print : "", | |
1384 | strempty(host_os_release_confext_level)); | |
b3d13314 | 1385 | if (r < 0) |
4beda316 | 1386 | return log_debug_errno(r, "Failed to mount image %s on %s: %m", mount_entry_source(m), mount_entry_path(m)); |
b3d13314 | 1387 | |
63862de4 | 1388 | return 1; |
b3d13314 LB |
1389 | } |
1390 | ||
93f59701 | 1391 | static int mount_overlay(const MountEntry *m) { |
f0304df6 | 1392 | _cleanup_free_ char *options = NULL, *layers = NULL; |
93f59701 LB |
1393 | int r; |
1394 | ||
1395 | assert(m); | |
1396 | ||
f0304df6 LB |
1397 | /* Extension hierarchies are optional (e.g.: confext might not have /opt) so check if they actually |
1398 | * exist in an image before attempting to create an overlay with them, otherwise the mount will | |
1399 | * fail. We can't check before this, as the images will not be mounted until now. */ | |
1400 | ||
1401 | /* Note that lowerdir= parameters are in 'reverse' order, so the top-most directory in the overlay | |
1402 | * comes first in the list. */ | |
1403 | STRV_FOREACH_BACKWARDS(o, m->overlay_layers) { | |
1404 | _cleanup_free_ char *escaped = NULL; | |
1405 | ||
1406 | r = is_dir(*o, /* follow= */ false); | |
1407 | if (r <= 0) { | |
1408 | if (r != -ENOENT) | |
1409 | log_debug_errno(r, | |
1410 | "Failed to check whether overlay layer source path '%s' exists, ignoring: %m", | |
1411 | *o); | |
1412 | continue; | |
1413 | } | |
1414 | ||
1415 | escaped = shell_escape(*o, ",:"); | |
1416 | if (!escaped) | |
1417 | return log_oom_debug(); | |
1418 | ||
1419 | if (!strextend_with_separator(&layers, ":", escaped)) | |
1420 | return log_oom_debug(); | |
1421 | } | |
1422 | ||
1423 | if (!layers) { | |
1424 | log_debug("None of the overlays specified in '%s' exist at the source, skipping.", | |
1425 | mount_entry_options(m)); | |
1426 | return 0; /* Only the root is set? Then there's nothing to overlay */ | |
1427 | } | |
1428 | ||
1429 | options = strjoin("lowerdir=", layers, ":", mount_entry_path(m)); /* The root goes in last */ | |
1430 | if (!options) | |
1431 | return log_oom_debug(); | |
93f59701 LB |
1432 | |
1433 | (void) mkdir_p_label(mount_entry_path(m), 0755); | |
1434 | ||
1435 | r = mount_nofollow_verbose(LOG_DEBUG, "overlay", mount_entry_path(m), "overlay", MS_RDONLY, options); | |
1436 | if (r == -ENOENT && m->ignore) | |
1437 | return 0; | |
63862de4 LB |
1438 | if (r < 0) |
1439 | return r; | |
93f59701 | 1440 | |
63862de4 | 1441 | return 1; |
93f59701 LB |
1442 | } |
1443 | ||
088696fe | 1444 | static int follow_symlink( |
d2d6c096 | 1445 | const char *root_directory, |
088696fe | 1446 | MountEntry *m) { |
d2d6c096 | 1447 | |
088696fe | 1448 | _cleanup_free_ char *target = NULL; |
8fceda93 LP |
1449 | int r; |
1450 | ||
088696fe LP |
1451 | /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we |
1452 | * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at | |
1453 | * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the | |
1454 | * end and already have a fully normalized name. */ | |
8fceda93 | 1455 | |
f461a28d | 1456 | r = chase(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL); |
088696fe LP |
1457 | if (r < 0) |
1458 | return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m)); | |
1459 | if (r > 0) /* Reached the end, nothing more to resolve */ | |
1460 | return 1; | |
8fceda93 | 1461 | |
f461a28d | 1462 | if (m->n_followed >= CHASE_MAX) /* put a boundary on things */ |
baaa35ad ZJS |
1463 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1464 | "Symlink loop on '%s'.", | |
1465 | mount_entry_path(m)); | |
8fceda93 | 1466 | |
e2341b6b DT |
1467 | log_debug("Followed mount entry path symlink %s %s %s.", |
1468 | mount_entry_path(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), target); | |
8fceda93 | 1469 | |
809ceb82 | 1470 | mount_entry_consume_prefix(m, TAKE_PTR(target)); |
8fceda93 | 1471 | |
b3a9d980 | 1472 | m->n_followed++; |
088696fe LP |
1473 | |
1474 | return 0; | |
8fceda93 LP |
1475 | } |
1476 | ||
82fb2da2 | 1477 | static int apply_one_mount( |
8fceda93 | 1478 | const char *root_directory, |
4e399953 | 1479 | MountEntry *m, |
79d956db | 1480 | const NamespaceParameters *p) { |
ac0930c8 | 1481 | |
e5f10caf | 1482 | _cleanup_free_ char *inaccessible = NULL; |
a227a4be | 1483 | bool rbind = true, make = false; |
15ae422b | 1484 | const char *what; |
15ae422b | 1485 | int r; |
15ae422b | 1486 | |
63862de4 LB |
1487 | /* Return 1 when the mount should be post-processed (remounted r/o, etc.), 0 otherwise. In most |
1488 | * cases post-processing is the right thing, the typical exception is when the mount is gracefully | |
1489 | * skipped. */ | |
1490 | ||
c17ec25e | 1491 | assert(m); |
79d956db | 1492 | assert(p); |
15ae422b | 1493 | |
34de407a | 1494 | log_debug("Applying namespace mount on %s", mount_entry_path(m)); |
fe3c2583 | 1495 | |
c17ec25e | 1496 | switch (m->mode) { |
15ae422b | 1497 | |
a868e437 | 1498 | case MOUNT_INACCESSIBLE: { |
cd7f3702 | 1499 | _cleanup_free_ char *runtime_dir = NULL; |
160cfdbe | 1500 | struct stat target; |
6d313367 LP |
1501 | |
1502 | /* First, get rid of everything that is below if there | |
1503 | * is anything... Then, overmount it with an | |
c4b41707 | 1504 | * inaccessible path. */ |
34de407a | 1505 | (void) umount_recursive(mount_entry_path(m), 0); |
6d313367 | 1506 | |
088696fe LP |
1507 | if (lstat(mount_entry_path(m), &target) < 0) { |
1508 | if (errno == ENOENT && m->ignore) | |
1509 | return 0; | |
1510 | ||
cbc056c8 ZJS |
1511 | return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m", |
1512 | mount_entry_path(m)); | |
088696fe | 1513 | } |
15ae422b | 1514 | |
cd7f3702 DDM |
1515 | /* We don't pass the literal runtime scope through here but one based purely on our UID. This |
1516 | * means that the root user's --user services will use the host's inaccessible inodes rather | |
1517 | * then root's private ones. This is preferable since it means device nodes that are | |
1518 | * overmounted to make them inaccessible will be overmounted with a device node, rather than | |
1519 | * an AF_UNIX socket inode. */ | |
1520 | runtime_dir = settle_runtime_dir(geteuid() == 0 ? RUNTIME_SCOPE_SYSTEM : RUNTIME_SCOPE_USER); | |
1521 | if (!runtime_dir) | |
1522 | return log_oom_debug(); | |
e5f10caf AZ |
1523 | |
1524 | r = mode_to_inaccessible_node(runtime_dir, target.st_mode, &inaccessible); | |
1525 | if (r < 0) | |
baaa35ad ZJS |
1526 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1527 | "File type not supported for inaccessible mounts. Note that symlinks are not allowed"); | |
e5f10caf | 1528 | what = inaccessible; |
c4b41707 | 1529 | break; |
160cfdbe | 1530 | } |
fe3c2583 | 1531 | |
a868e437 LP |
1532 | case MOUNT_READ_ONLY: |
1533 | case MOUNT_READ_WRITE: | |
1534 | case MOUNT_READ_WRITE_IMPLICIT: | |
1535 | case MOUNT_EXEC: | |
1536 | case MOUNT_NOEXEC: | |
8fceda93 | 1537 | r = path_is_mount_point(mount_entry_path(m), root_directory, 0); |
088696fe LP |
1538 | if (r == -ENOENT && m->ignore) |
1539 | return 0; | |
d944dc95 | 1540 | if (r < 0) |
cbc056c8 ZJS |
1541 | return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m", |
1542 | mount_entry_path(m)); | |
1543 | if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY | |
ddc155b2 | 1544 | * and MS_NOEXEC bits for the mount point if needed. */ |
63862de4 | 1545 | return 1; |
6b7c9f8b | 1546 | /* This isn't a mount point yet, let's make it one. */ |
34de407a | 1547 | what = mount_entry_path(m); |
6b7c9f8b | 1548 | break; |
15ae422b | 1549 | |
a868e437 | 1550 | case MOUNT_EXTENSION_DIRECTORY: { |
a07b9926 | 1551 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, |
55ea4ef0 | 1552 | *host_os_release_level = NULL, *extension_name = NULL; |
a07b9926 | 1553 | _cleanup_strv_free_ char **extension_release = NULL; |
55ea4ef0 | 1554 | ImageClass class = IMAGE_SYSEXT; |
a07b9926 LB |
1555 | |
1556 | r = path_extract_filename(mount_entry_source(m), &extension_name); | |
1557 | if (r < 0) | |
1558 | return log_debug_errno(r, "Failed to extract extension name from %s: %m", mount_entry_source(m)); | |
1559 | ||
55ea4ef0 MG |
1560 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_SYSEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); |
1561 | if (r == -ENOENT) { | |
1562 | r = load_extension_release_pairs(mount_entry_source(m), IMAGE_CONFEXT, extension_name, /* relax_extension_release_check= */ false, &extension_release); | |
1563 | if (r >= 0) | |
1564 | class = IMAGE_CONFEXT; | |
1565 | } | |
5897469a LB |
1566 | if (r < 0) |
1567 | return log_debug_errno(r, "Failed to acquire 'extension-release' data of extension tree %s: %m", mount_entry_source(m)); | |
55ea4ef0 | 1568 | |
a07b9926 LB |
1569 | r = parse_os_release( |
1570 | empty_to_root(root_directory), | |
1571 | "ID", &host_os_release_id, | |
1572 | "VERSION_ID", &host_os_release_version_id, | |
55ea4ef0 | 1573 | image_class_info[class].level_env, &host_os_release_level, |
a07b9926 LB |
1574 | NULL); |
1575 | if (r < 0) | |
1576 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
1577 | if (isempty(host_os_release_id)) | |
1578 | 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)); | |
1579 | ||
55ea4ef0 | 1580 | r = load_extension_release_pairs(mount_entry_source(m), class, extension_name, /* relax_extension_release_check= */ false, &extension_release); |
a07b9926 LB |
1581 | if (r == -ENOENT && m->ignore) |
1582 | return 0; | |
1583 | if (r < 0) | |
1584 | return log_debug_errno(r, "Failed to parse directory %s extension-release metadata: %m", extension_name); | |
1585 | ||
1586 | r = extension_release_validate( | |
1587 | extension_name, | |
1588 | host_os_release_id, | |
1589 | host_os_release_version_id, | |
55ea4ef0 MG |
1590 | host_os_release_level, |
1591 | /* host_extension_scope */ NULL, /* Leave empty, we need to accept both system and portable */ | |
30dfe035 | 1592 | extension_release, |
55ea4ef0 | 1593 | class); |
a07b9926 LB |
1594 | if (r == 0) |
1595 | return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Directory %s extension-release metadata does not match the root's", extension_name); | |
1596 | if (r < 0) | |
1597 | return log_debug_errno(r, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name); | |
1598 | ||
1599 | _fallthrough_; | |
1600 | } | |
1601 | ||
a868e437 | 1602 | case MOUNT_BIND: |
d2d6c096 | 1603 | rbind = false; |
d2d6c096 | 1604 | |
4831981d | 1605 | _fallthrough_; |
a868e437 | 1606 | case MOUNT_BIND_RECURSIVE: { |
088696fe | 1607 | _cleanup_free_ char *chased = NULL; |
5d997827 | 1608 | |
cbc056c8 ZJS |
1609 | /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note |
1610 | * that bind mount source paths are always relative to the host root, hence we pass NULL as | |
f461a28d | 1611 | * root directory to chase() here. */ |
088696fe | 1612 | |
f461a28d | 1613 | r = chase(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL); |
088696fe LP |
1614 | if (r == -ENOENT && m->ignore) { |
1615 | log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m)); | |
1616 | return 0; | |
1617 | } | |
1618 | if (r < 0) | |
1619 | return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m)); | |
1620 | ||
e2341b6b DT |
1621 | log_debug("Followed source symlinks %s %s %s.", |
1622 | mount_entry_source(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), chased); | |
088696fe LP |
1623 | |
1624 | free_and_replace(m->source_malloc, chased); | |
d2d6c096 LP |
1625 | |
1626 | what = mount_entry_source(m); | |
a227a4be | 1627 | make = true; |
d2d6c096 | 1628 | break; |
088696fe | 1629 | } |
d2d6c096 | 1630 | |
a868e437 LP |
1631 | case MOUNT_EMPTY_DIR: |
1632 | case MOUNT_TMPFS: | |
2abd4e38 | 1633 | return mount_tmpfs(m); |
6c47cd7d | 1634 | |
a868e437 LP |
1635 | case MOUNT_PRIVATE_TMP: |
1636 | case MOUNT_PRIVATE_TMP_READ_ONLY: | |
89bd586c | 1637 | what = mount_entry_source(m); |
a227a4be | 1638 | make = true; |
15ae422b | 1639 | break; |
e364ad06 | 1640 | |
a868e437 | 1641 | case MOUNT_PRIVATE_DEV: |
79d956db | 1642 | return mount_private_dev(m, p->runtime_scope); |
5d997827 | 1643 | |
a868e437 | 1644 | case MOUNT_BIND_DEV: |
5d997827 LP |
1645 | return mount_bind_dev(m); |
1646 | ||
a868e437 | 1647 | case MOUNT_PRIVATE_SYSFS: |
4793605d | 1648 | return mount_private_sysfs(m, p); |
277b269e | 1649 | |
a868e437 | 1650 | case MOUNT_BIND_SYSFS: |
10028263 | 1651 | return mount_bind_sysfs(m); |
5d997827 | 1652 | |
a868e437 | 1653 | case MOUNT_PROCFS: |
79d956db | 1654 | return mount_procfs(m, p); |
d6797c92 | 1655 | |
a868e437 | 1656 | case MOUNT_RUN: |
94293d65 LB |
1657 | return mount_run(m); |
1658 | ||
a868e437 | 1659 | case MOUNT_MQUEUEFS: |
80271a44 XR |
1660 | return mount_mqueuefs(m); |
1661 | ||
a868e437 | 1662 | case MOUNT_IMAGE: |
79d956db | 1663 | return mount_image(m, NULL, p->mount_image_policy); |
93f59701 | 1664 | |
a868e437 | 1665 | case MOUNT_EXTENSION_IMAGE: |
79d956db | 1666 | return mount_image(m, root_directory, p->extension_image_policy); |
93f59701 | 1667 | |
a868e437 | 1668 | case MOUNT_OVERLAY: |
93f59701 | 1669 | return mount_overlay(m); |
b3d13314 | 1670 | |
e364ad06 | 1671 | default: |
04499a70 | 1672 | assert_not_reached(); |
15ae422b LP |
1673 | } |
1674 | ||
ac0930c8 | 1675 | assert(what); |
15ae422b | 1676 | |
21935150 LP |
1677 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); |
1678 | if (r < 0) { | |
a227a4be | 1679 | bool try_again = false; |
a227a4be LP |
1680 | |
1681 | if (r == -ENOENT && make) { | |
8bab8029 | 1682 | int q; |
a227a4be | 1683 | |
cbc056c8 ZJS |
1684 | /* Hmm, either the source or the destination are missing. Let's see if we can create |
1685 | the destination, then try again. */ | |
a227a4be | 1686 | |
8bab8029 | 1687 | (void) mkdir_parents(mount_entry_path(m), 0755); |
a227a4be | 1688 | |
8bab8029 | 1689 | q = make_mount_point_inode_from_path(what, mount_entry_path(m), 0755); |
283eb4fd | 1690 | if (q < 0) { |
7fcd1343 | 1691 | 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 |
1692 | log_warning_errno(q, "Failed to create destination mount point node '%s', ignoring: %m", |
1693 | mount_entry_path(m)); | |
1694 | } else | |
8bab8029 | 1695 | try_again = true; |
a227a4be LP |
1696 | } |
1697 | ||
21935150 LP |
1698 | if (try_again) |
1699 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); | |
a227a4be | 1700 | if (r < 0) |
7fcd1343 | 1701 | 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 | 1702 | } |
6b7c9f8b | 1703 | |
34de407a | 1704 | log_debug("Successfully mounted %s to %s", what, mount_entry_path(m)); |
63862de4 | 1705 | return 1; |
ac0930c8 | 1706 | } |
15ae422b | 1707 | |
6b000af4 | 1708 | static int make_read_only(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
9ce4e4b0 | 1709 | unsigned long new_flags = 0, flags_mask = 0; |
57ccd9f6 YW |
1710 | bool submounts; |
1711 | int r; | |
15ae422b | 1712 | |
c17ec25e | 1713 | assert(m); |
ac9de0b3 | 1714 | assert(proc_self_mountinfo); |
ac0930c8 | 1715 | |
63862de4 LB |
1716 | if (m->state != MOUNT_APPLIED) |
1717 | return 0; | |
1718 | ||
a868e437 | 1719 | if (mount_entry_read_only(m) || m->mode == MOUNT_PRIVATE_DEV) { |
9ce4e4b0 LP |
1720 | new_flags |= MS_RDONLY; |
1721 | flags_mask |= MS_RDONLY; | |
1722 | } | |
1723 | ||
1724 | if (m->nosuid) { | |
1725 | new_flags |= MS_NOSUID; | |
1726 | flags_mask |= MS_NOSUID; | |
1727 | } | |
1728 | ||
1729 | if (flags_mask == 0) /* No Change? */ | |
6b7c9f8b LP |
1730 | return 0; |
1731 | ||
9ce4e4b0 LP |
1732 | /* We generally apply these changes recursively, except for /dev, and the cases we know there's |
1733 | * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the | |
1734 | * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace | |
1735 | * and running Linux <= 4.17. */ | |
1736 | submounts = | |
1737 | mount_entry_read_only(m) && | |
a868e437 | 1738 | !IN_SET(m->mode, MOUNT_EMPTY_DIR, MOUNT_TMPFS); |
9ce4e4b0 | 1739 | if (submounts) |
6b000af4 | 1740 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); |
9ce4e4b0 | 1741 | else |
7cce68e1 | 1742 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); |
9ce4e4b0 | 1743 | |
ddc155b2 | 1744 | /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked |
867189b5 LP |
1745 | * read-only already stays this way. This improves compatibility with container managers, where we |
1746 | * won't attempt to undo read-only mounts already applied. */ | |
ac0930c8 | 1747 | |
8fceda93 | 1748 | if (r == -ENOENT && m->ignore) |
867189b5 | 1749 | return 0; |
763a260a | 1750 | if (r < 0) |
9ce4e4b0 | 1751 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), |
763a260a | 1752 | submounts ? " and its submounts" : ""); |
763a260a | 1753 | return 0; |
d944dc95 LP |
1754 | } |
1755 | ||
ddc155b2 TM |
1756 | static int make_noexec(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
1757 | unsigned long new_flags = 0, flags_mask = 0; | |
57ccd9f6 YW |
1758 | bool submounts; |
1759 | int r; | |
ddc155b2 TM |
1760 | |
1761 | assert(m); | |
1762 | assert(proc_self_mountinfo); | |
1763 | ||
63862de4 LB |
1764 | if (m->state != MOUNT_APPLIED) |
1765 | return 0; | |
1766 | ||
ddc155b2 TM |
1767 | if (mount_entry_noexec(m)) { |
1768 | new_flags |= MS_NOEXEC; | |
1769 | flags_mask |= MS_NOEXEC; | |
1770 | } else if (mount_entry_exec(m)) { | |
1771 | new_flags &= ~MS_NOEXEC; | |
1772 | flags_mask |= MS_NOEXEC; | |
1773 | } | |
1774 | ||
1775 | if (flags_mask == 0) /* No Change? */ | |
1776 | return 0; | |
1777 | ||
a868e437 | 1778 | submounts = !IN_SET(m->mode, MOUNT_EMPTY_DIR, MOUNT_TMPFS); |
ddc155b2 TM |
1779 | |
1780 | if (submounts) | |
1781 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); | |
1782 | else | |
1783 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); | |
1784 | ||
1785 | if (r == -ENOENT && m->ignore) | |
1786 | return 0; | |
1787 | if (r < 0) | |
1788 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1789 | submounts ? " and its submounts" : ""); | |
1790 | return 0; | |
1791 | } | |
1792 | ||
6720e356 | 1793 | static int make_nosuid(const MountEntry *m, FILE *proc_self_mountinfo) { |
57ccd9f6 YW |
1794 | bool submounts; |
1795 | int r; | |
6720e356 YW |
1796 | |
1797 | assert(m); | |
1798 | assert(proc_self_mountinfo); | |
1799 | ||
63862de4 LB |
1800 | if (m->state != MOUNT_APPLIED) |
1801 | return 0; | |
1802 | ||
a868e437 | 1803 | submounts = !IN_SET(m->mode, MOUNT_EMPTY_DIR, MOUNT_TMPFS); |
6720e356 YW |
1804 | if (submounts) |
1805 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, NULL, proc_self_mountinfo); | |
1806 | else | |
1807 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, proc_self_mountinfo); | |
1808 | if (r == -ENOENT && m->ignore) | |
1809 | return 0; | |
1810 | if (r < 0) | |
1811 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1812 | submounts ? " and its submounts" : ""); | |
1813 | return 0; | |
1814 | } | |
1815 | ||
79d956db LP |
1816 | static bool namespace_parameters_mount_apivfs(const NamespaceParameters *p) { |
1817 | assert(p); | |
5d997827 | 1818 | |
9c988f93 DH |
1819 | /* |
1820 | * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=, | |
1821 | * since to protect the API VFS mounts, they need to be around in the | |
9b68367b | 1822 | * first place... |
9c988f93 | 1823 | */ |
5d997827 | 1824 | |
79d956db LP |
1825 | return p->mount_apivfs || |
1826 | p->protect_control_groups || | |
1827 | p->protect_kernel_tunables || | |
1828 | p->protect_proc != PROTECT_PROC_DEFAULT || | |
1829 | p->proc_subset != PROC_SUBSET_ALL; | |
5d997827 LP |
1830 | } |
1831 | ||
fbf90c0d CB |
1832 | /* Walk all mount entries and dropping any unused mounts. This affects all |
1833 | * mounts: | |
1834 | * - that are implicitly protected by a path that has been rendered inaccessible | |
1835 | * - whose immediate parent requests the same protection mode as the mount itself | |
1836 | * - that are outside of the relevant root directory | |
1837 | * - which are duplicates | |
1838 | */ | |
063c977a LP |
1839 | static void drop_unused_mounts(MountList *ml, const char *root_directory) { |
1840 | assert(ml); | |
9b68367b | 1841 | assert(root_directory); |
f8b64b57 | 1842 | |
063c977a LP |
1843 | assert(ml->mounts || ml->n_mounts == 0); |
1844 | ||
1845 | typesafe_qsort(ml->mounts, ml->n_mounts, mount_path_compare); | |
f8b64b57 | 1846 | |
063c977a LP |
1847 | drop_duplicates(ml); |
1848 | drop_outside_root(ml, root_directory); | |
1849 | drop_inaccessible(ml); | |
1850 | drop_nop(ml); | |
f8b64b57 LP |
1851 | } |
1852 | ||
df61e79a | 1853 | static int create_symlinks_from_tuples(const char *root, char **strv_symlinks) { |
df61e79a LB |
1854 | int r; |
1855 | ||
1856 | STRV_FOREACH_PAIR(src, dst, strv_symlinks) { | |
1857 | _cleanup_free_ char *src_abs = NULL, *dst_abs = NULL; | |
1858 | ||
1859 | src_abs = path_join(root, *src); | |
1860 | dst_abs = path_join(root, *dst); | |
1861 | if (!src_abs || !dst_abs) | |
1862 | return -ENOMEM; | |
1863 | ||
1864 | r = mkdir_parents_label(dst_abs, 0755); | |
1865 | if (r < 0) | |
15461b7f LB |
1866 | return log_debug_errno( |
1867 | r, | |
1868 | "Failed to create parent directory for symlink '%s': %m", | |
1869 | dst_abs); | |
df61e79a LB |
1870 | |
1871 | r = symlink_idempotent(src_abs, dst_abs, true); | |
1872 | if (r < 0) | |
15461b7f LB |
1873 | return log_debug_errno( |
1874 | r, | |
1875 | "Failed to create symlink from '%s' to '%s': %m", | |
1876 | src_abs, | |
1877 | dst_abs); | |
df61e79a LB |
1878 | } |
1879 | ||
1880 | return 0; | |
1881 | } | |
1882 | ||
13b3af4a LB |
1883 | static void mount_entry_path_debug_string(const char *root, MountEntry *m, char **error_path) { |
1884 | assert(m); | |
1885 | ||
1886 | /* Create a string suitable for debugging logs, stripping for example the local working directory. | |
1887 | * For example, with a BindPaths=/var/bar that does not exist on the host: | |
1888 | * | |
1889 | * Before: | |
1890 | * foo.service: Failed to set up mount namespacing: /run/systemd/unit-root/var/bar: No such file or directory | |
1891 | * After: | |
1892 | * foo.service: Failed to set up mount namespacing: /var/bar: No such file or directory | |
1893 | * | |
1894 | * Note that this is an error path, so no OOM check is done on purpose. */ | |
1895 | ||
1896 | if (!error_path) | |
1897 | return; | |
1898 | ||
1899 | if (!mount_entry_path(m)) { | |
1900 | *error_path = NULL; | |
1901 | return; | |
1902 | } | |
1903 | ||
1904 | if (root) { | |
1905 | const char *e = startswith(mount_entry_path(m), root); | |
1906 | if (e) { | |
1907 | *error_path = strdup(e); | |
1908 | return; | |
1909 | } | |
1910 | } | |
1911 | ||
1912 | *error_path = strdup(mount_entry_path(m)); | |
1913 | return; | |
1914 | } | |
1915 | ||
82fb2da2 | 1916 | static int apply_mounts( |
063c977a | 1917 | MountList *ml, |
82fb2da2 | 1918 | const char *root, |
79d956db | 1919 | const NamespaceParameters *p, |
82fb2da2 LB |
1920 | char **error_path) { |
1921 | ||
1922 | _cleanup_fclose_ FILE *proc_self_mountinfo = NULL; | |
1923 | _cleanup_free_ char **deny_list = NULL; | |
82fb2da2 LB |
1924 | int r; |
1925 | ||
063c977a | 1926 | assert(ml); |
82fb2da2 | 1927 | assert(root); |
063c977a LP |
1928 | assert(p); |
1929 | ||
1930 | if (ml->n_mounts == 0) /* Shortcut: nothing to do */ | |
1931 | return 0; | |
82fb2da2 LB |
1932 | |
1933 | /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of | |
1934 | * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */ | |
1935 | proc_self_mountinfo = fopen("/proc/self/mountinfo", "re"); | |
1936 | if (!proc_self_mountinfo) { | |
d60e3b40 LP |
1937 | r = -errno; |
1938 | ||
82fb2da2 LB |
1939 | if (error_path) |
1940 | *error_path = strdup("/proc/self/mountinfo"); | |
d60e3b40 LP |
1941 | |
1942 | return log_debug_errno(r, "Failed to open /proc/self/mountinfo: %m"); | |
82fb2da2 LB |
1943 | } |
1944 | ||
1945 | /* First round, establish all mounts we need */ | |
1946 | for (;;) { | |
1947 | bool again = false; | |
1948 | ||
063c977a | 1949 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
82fb2da2 | 1950 | |
63862de4 | 1951 | if (m->state != MOUNT_PENDING) |
82fb2da2 LB |
1952 | continue; |
1953 | ||
a07b9926 | 1954 | /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */ |
a868e437 | 1955 | r = follow_symlink(!IN_SET(m->mode, MOUNT_EXTENSION_IMAGE, MOUNT_EXTENSION_DIRECTORY) ? root : NULL, m); |
82fb2da2 | 1956 | if (r < 0) { |
13b3af4a | 1957 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
1958 | return r; |
1959 | } | |
1960 | if (r == 0) { | |
1961 | /* We hit a symlinked mount point. The entry got rewritten and might | |
1962 | * point to a very different place now. Let's normalize the changed | |
1963 | * list, and start from the beginning. After all to mount the entry | |
1964 | * at the new location we might need some other mounts first */ | |
1965 | again = true; | |
1966 | break; | |
1967 | } | |
1968 | ||
63862de4 | 1969 | /* Returns 1 if the mount should be post-processed, 0 otherwise */ |
79d956db | 1970 | r = apply_one_mount(root, m, p); |
82fb2da2 | 1971 | if (r < 0) { |
13b3af4a | 1972 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
1973 | return r; |
1974 | } | |
63862de4 | 1975 | m->state = r == 0 ? MOUNT_SKIPPED : MOUNT_APPLIED; |
82fb2da2 LB |
1976 | } |
1977 | ||
1978 | if (!again) | |
1979 | break; | |
1980 | ||
063c977a | 1981 | drop_unused_mounts(ml, root); |
82fb2da2 LB |
1982 | } |
1983 | ||
df61e79a | 1984 | /* Now that all filesystems have been set up, but before the |
663e2756 | 1985 | * read-only switches are flipped, create the exec dirs and other symlinks. |
df61e79a LB |
1986 | * Note that when /var/lib is not empty/tmpfs, these symlinks will already |
1987 | * exist, which means this will be a no-op. */ | |
79d956db | 1988 | r = create_symlinks_from_tuples(root, p->symlinks); |
df61e79a | 1989 | if (r < 0) |
663e2756 | 1990 | return log_debug_errno(r, "Failed to set up symlinks inside mount namespace: %m"); |
df61e79a | 1991 | |
82fb2da2 | 1992 | /* Create a deny list we can pass to bind_mount_recursive() */ |
063c977a | 1993 | deny_list = new(char*, ml->n_mounts+1); |
82fb2da2 LB |
1994 | if (!deny_list) |
1995 | return -ENOMEM; | |
063c977a LP |
1996 | for (size_t j = 0; j < ml->n_mounts; j++) |
1997 | deny_list[j] = (char*) mount_entry_path(ml->mounts+j); | |
1998 | deny_list[ml->n_mounts] = NULL; | |
82fb2da2 LB |
1999 | |
2000 | /* Second round, flip the ro bits if necessary. */ | |
063c977a | 2001 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
82fb2da2 LB |
2002 | r = make_read_only(m, deny_list, proc_self_mountinfo); |
2003 | if (r < 0) { | |
13b3af4a | 2004 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
2005 | return r; |
2006 | } | |
2007 | } | |
2008 | ||
2009 | /* Third round, flip the noexec bits with a simplified deny list. */ | |
063c977a | 2010 | for (size_t j = 0; j < ml->n_mounts; j++) |
a868e437 | 2011 | if (IN_SET((ml->mounts+j)->mode, MOUNT_EXEC, MOUNT_NOEXEC)) |
063c977a LP |
2012 | deny_list[j] = (char*) mount_entry_path(ml->mounts+j); |
2013 | deny_list[ml->n_mounts] = NULL; | |
82fb2da2 | 2014 | |
063c977a | 2015 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
82fb2da2 LB |
2016 | r = make_noexec(m, deny_list, proc_self_mountinfo); |
2017 | if (r < 0) { | |
13b3af4a | 2018 | mount_entry_path_debug_string(root, m, error_path); |
82fb2da2 LB |
2019 | return r; |
2020 | } | |
2021 | } | |
2022 | ||
6720e356 | 2023 | /* Fourth round, flip the nosuid bits without a deny list. */ |
79d956db | 2024 | if (p->mount_nosuid) |
063c977a | 2025 | FOREACH_ARRAY(m, ml->mounts, ml->n_mounts) { |
6720e356 YW |
2026 | r = make_nosuid(m, proc_self_mountinfo); |
2027 | if (r < 0) { | |
13b3af4a | 2028 | mount_entry_path_debug_string(root, m, error_path); |
6720e356 YW |
2029 | return r; |
2030 | } | |
2031 | } | |
2032 | ||
82fb2da2 LB |
2033 | return 1; |
2034 | } | |
2035 | ||
c8c535d5 LP |
2036 | static bool root_read_only( |
2037 | char **read_only_paths, | |
2038 | ProtectSystem protect_system) { | |
2039 | ||
2040 | /* Determine whether the root directory is going to be read-only given the configured settings. */ | |
2041 | ||
2042 | if (protect_system == PROTECT_SYSTEM_STRICT) | |
2043 | return true; | |
2044 | ||
de46b2be | 2045 | if (prefixed_path_strv_contains(read_only_paths, "/")) |
c8c535d5 LP |
2046 | return true; |
2047 | ||
2048 | return false; | |
2049 | } | |
2050 | ||
2051 | static bool home_read_only( | |
2052 | char** read_only_paths, | |
2053 | char** inaccessible_paths, | |
2054 | char** empty_directories, | |
2055 | const BindMount *bind_mounts, | |
2056 | size_t n_bind_mounts, | |
2057 | const TemporaryFileSystem *temporary_filesystems, | |
2058 | size_t n_temporary_filesystems, | |
2059 | ProtectHome protect_home) { | |
2060 | ||
c8c535d5 LP |
2061 | /* Determine whether the /home directory is going to be read-only given the configured settings. Yes, |
2062 | * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple | |
2063 | * settings. */ | |
2064 | ||
2065 | if (protect_home != PROTECT_HOME_NO) | |
2066 | return true; | |
2067 | ||
de46b2be TM |
2068 | if (prefixed_path_strv_contains(read_only_paths, "/home") || |
2069 | prefixed_path_strv_contains(inaccessible_paths, "/home") || | |
2070 | prefixed_path_strv_contains(empty_directories, "/home")) | |
c8c535d5 LP |
2071 | return true; |
2072 | ||
8794e6db | 2073 | for (size_t i = 0; i < n_temporary_filesystems; i++) |
c8c535d5 LP |
2074 | if (path_equal(temporary_filesystems[i].path, "/home")) |
2075 | return true; | |
2076 | ||
2077 | /* If /home is overmounted with some dir from the host it's not writable. */ | |
8794e6db | 2078 | for (size_t i = 0; i < n_bind_mounts; i++) |
c8c535d5 LP |
2079 | if (path_equal(bind_mounts[i].destination, "/home")) |
2080 | return true; | |
2081 | ||
2082 | return false; | |
2083 | } | |
2084 | ||
79d956db | 2085 | int setup_namespace(const NamespaceParameters *p, char **error_path) { |
15ae422b | 2086 | |
915e6d16 LP |
2087 | _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL; |
2088 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; | |
93f59701 | 2089 | _cleanup_strv_free_ char **hierarchies = NULL; |
063c977a | 2090 | _cleanup_(mount_list_done) MountList ml = {}; |
79d956db | 2091 | bool require_prefix = false; |
24759d8f | 2092 | const char *root; |
9cb1709b LP |
2093 | DissectImageFlags dissect_image_flags = |
2094 | DISSECT_IMAGE_GENERIC_ROOT | | |
2095 | DISSECT_IMAGE_REQUIRE_ROOT | | |
2096 | DISSECT_IMAGE_DISCARD_ON_LOOP | | |
2097 | DISSECT_IMAGE_RELAX_VAR_CHECK | | |
2098 | DISSECT_IMAGE_FSCK | | |
c65f854a | 2099 | DISSECT_IMAGE_USR_NO_ROOT | |
73d88b80 LP |
2100 | DISSECT_IMAGE_GROWFS | |
2101 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | | |
2102 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
89e62e0b | 2103 | int r; |
15ae422b | 2104 | |
79d956db | 2105 | assert(p); |
915e6d16 | 2106 | |
cdf42f9b LP |
2107 | /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes |
2108 | * we configure take effect */ | |
2109 | BLOCK_WITH_UMASK(0000); | |
2110 | ||
79d956db LP |
2111 | bool setup_propagate = !isempty(p->propagate_dir) && !isempty(p->incoming_dir); |
2112 | unsigned long mount_propagation_flag = p->mount_propagation_flag != 0 ? p->mount_propagation_flag : MS_SHARED; | |
ac0930c8 | 2113 | |
79d956db | 2114 | if (p->root_image) { |
c8c535d5 | 2115 | /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */ |
79d956db LP |
2116 | if (root_read_only(p->read_only_paths, |
2117 | p->protect_system) && | |
2118 | home_read_only(p->read_only_paths, p->inaccessible_paths, p->empty_directories, | |
2119 | p->bind_mounts, p->n_bind_mounts, p->temporary_filesystems, p->n_temporary_filesystems, | |
2120 | p->protect_home) && | |
2121 | strv_isempty(p->read_write_paths)) | |
915e6d16 LP |
2122 | dissect_image_flags |= DISSECT_IMAGE_READ_ONLY; |
2123 | ||
79d956db | 2124 | SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, p->verity && p->verity->data_path); |
915e6d16 | 2125 | |
89e62e0b | 2126 | r = loop_device_make_by_path( |
79d956db | 2127 | p->root_image, |
ef9c184d | 2128 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */, |
22ee78a8 | 2129 | /* sector_size= */ UINT32_MAX, |
89e62e0b | 2130 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 2131 | LOCK_SH, |
89e62e0b | 2132 | &loop_device); |
78ebe980 | 2133 | if (r < 0) |
89e62e0b LP |
2134 | return log_debug_errno(r, "Failed to create loop device for root image: %m"); |
2135 | ||
bad31660 YW |
2136 | r = dissect_loop_device( |
2137 | loop_device, | |
79d956db LP |
2138 | p->verity, |
2139 | p->root_image_options, | |
2140 | p->root_image_policy, | |
89e62e0b LP |
2141 | dissect_image_flags, |
2142 | &dissected_image); | |
78ebe980 | 2143 | if (r < 0) |
763a260a | 2144 | return log_debug_errno(r, "Failed to dissect image: %m"); |
78ebe980 | 2145 | |
88b3300f LP |
2146 | r = dissected_image_load_verity_sig_partition( |
2147 | dissected_image, | |
2148 | loop_device->fd, | |
79d956db | 2149 | p->verity); |
88b3300f LP |
2150 | if (r < 0) |
2151 | return r; | |
2152 | ||
89e62e0b LP |
2153 | r = dissected_image_decrypt( |
2154 | dissected_image, | |
2155 | NULL, | |
79d956db | 2156 | p->verity, |
e330f97a | 2157 | dissect_image_flags); |
915e6d16 | 2158 | if (r < 0) |
763a260a | 2159 | return log_debug_errno(r, "Failed to decrypt dissected image: %m"); |
915e6d16 LP |
2160 | } |
2161 | ||
79d956db LP |
2162 | if (p->root_directory) |
2163 | root = p->root_directory; | |
0722b359 | 2164 | else { |
77f16dbd DDM |
2165 | /* /run/systemd should have been created by PID 1 early on already, but in some cases, like |
2166 | * when running tests (test-execute), it might not have been created yet so let's make sure | |
2167 | * we create it if it doesn't already exist. */ | |
2168 | (void) mkdir_p_label("/run/systemd", 0755); | |
2169 | ||
d73020f2 LP |
2170 | /* Always create the mount namespace in a temporary directory, instead of operating directly |
2171 | * in the root. The temporary directory prevents any mounts from being potentially obscured | |
2172 | * my other mounts we already applied. We use the same mount point for all images, which is | |
2173 | * safe, since they all live in their own namespaces after all, and hence won't see each | |
be79c235 LP |
2174 | * other. (Note: this directory is also created by PID 1 early on, we create it here for |
2175 | * similar reasons as /run/systemd/ first.) */ | |
2176 | root = "/run/systemd/mount-rootfs"; | |
2177 | (void) mkdir_label(root, 0555); | |
e908468b | 2178 | |
d18aff04 | 2179 | require_prefix = true; |
0722b359 | 2180 | } |
e908468b | 2181 | |
79d956db | 2182 | if (p->n_extension_images > 0 || !strv_isempty(p->extension_directories)) { |
55ea4ef0 MG |
2183 | /* Hierarchy population needs to be done for sysext and confext extension images */ |
2184 | r = parse_env_extension_hierarchies(&hierarchies, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES"); | |
93f59701 LB |
2185 | if (r < 0) |
2186 | return r; | |
2187 | } | |
2188 | ||
a868e437 | 2189 | r = append_access_mounts(&ml, p->read_write_paths, MOUNT_READ_WRITE, require_prefix); |
063c977a LP |
2190 | if (r < 0) |
2191 | return r; | |
613b411c | 2192 | |
a868e437 | 2193 | r = append_access_mounts(&ml, p->read_only_paths, MOUNT_READ_ONLY, require_prefix); |
063c977a LP |
2194 | if (r < 0) |
2195 | return r; | |
5f7a690a | 2196 | |
a868e437 | 2197 | r = append_access_mounts(&ml, p->inaccessible_paths, MOUNT_INACCESSIBLE, require_prefix); |
063c977a LP |
2198 | if (r < 0) |
2199 | return r; | |
613b411c | 2200 | |
a868e437 | 2201 | r = append_access_mounts(&ml, p->exec_paths, MOUNT_EXEC, require_prefix); |
063c977a LP |
2202 | if (r < 0) |
2203 | return r; | |
613b411c | 2204 | |
a868e437 | 2205 | r = append_access_mounts(&ml, p->no_exec_paths, MOUNT_NOEXEC, require_prefix); |
063c977a LP |
2206 | if (r < 0) |
2207 | return r; | |
2208 | ||
2209 | r = append_empty_dir_mounts(&ml, p->empty_directories); | |
2210 | if (r < 0) | |
2211 | return r; | |
2212 | ||
2213 | r = append_bind_mounts(&ml, p->bind_mounts, p->n_bind_mounts); | |
2214 | if (r < 0) | |
2215 | return r; | |
2216 | ||
2217 | r = append_tmpfs_mounts(&ml, p->temporary_filesystems, p->n_temporary_filesystems); | |
2218 | if (r < 0) | |
2219 | return r; | |
7ff7394d | 2220 | |
063c977a LP |
2221 | if (p->tmp_dir) { |
2222 | bool ro = streq(p->tmp_dir, RUN_SYSTEMD_EMPTY); | |
2223 | ||
2224 | MountEntry *me = mount_list_extend(&ml); | |
2225 | if (!me) | |
2226 | return log_oom_debug(); | |
2227 | ||
2228 | *me = (MountEntry) { | |
2229 | .path_const = "/tmp", | |
a868e437 | 2230 | .mode = ro ? MOUNT_PRIVATE_TMP_READ_ONLY : MOUNT_PRIVATE_TMP, |
063c977a LP |
2231 | .source_const = p->tmp_dir, |
2232 | }; | |
2233 | } | |
2234 | ||
2235 | if (p->var_tmp_dir) { | |
2236 | bool ro = streq(p->var_tmp_dir, RUN_SYSTEMD_EMPTY); | |
2237 | ||
2238 | MountEntry *me = mount_list_extend(&ml); | |
2239 | if (!me) | |
2240 | return log_oom_debug(); | |
2241 | ||
2242 | *me = (MountEntry) { | |
2243 | .path_const = "/var/tmp", | |
a868e437 | 2244 | .mode = ro ? MOUNT_PRIVATE_TMP_READ_ONLY : MOUNT_PRIVATE_TMP, |
063c977a LP |
2245 | .source_const = p->var_tmp_dir, |
2246 | }; | |
2247 | } | |
2248 | ||
2249 | r = append_mount_images(&ml, p->mount_images, p->n_mount_images); | |
2250 | if (r < 0) | |
2251 | return r; | |
2252 | ||
2253 | r = append_extensions(&ml, root, p->extension_dir, hierarchies, p->extension_images, p->n_extension_images, p->extension_directories); | |
2254 | if (r < 0) | |
2255 | return r; | |
2256 | ||
2257 | if (p->private_dev) { | |
2258 | MountEntry *me = mount_list_extend(&ml); | |
2259 | if (!me) | |
2260 | return log_oom_debug(); | |
2261 | ||
2262 | *me = (MountEntry) { | |
2263 | .path_const = "/dev", | |
a868e437 | 2264 | .mode = MOUNT_PRIVATE_DEV, |
063c977a LP |
2265 | .flags = DEV_MOUNT_OPTIONS, |
2266 | }; | |
2267 | } | |
2268 | ||
2269 | /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the protective | |
2270 | mounts to non-pid /proc paths would fail. But the pid only option may have failed gracefully, so | |
2271 | let's try the mounts but it's not fatal if they don't succeed. */ | |
2272 | bool ignore_protect_proc = p->ignore_protect_paths || p->proc_subset == PROC_SUBSET_PID; | |
2273 | if (p->protect_kernel_tunables) { | |
2274 | r = append_static_mounts(&ml, | |
2275 | protect_kernel_tunables_proc_table, | |
2276 | ELEMENTSOF(protect_kernel_tunables_proc_table), | |
2277 | ignore_protect_proc); | |
ddc155b2 | 2278 | if (r < 0) |
063c977a | 2279 | return r; |
ddc155b2 | 2280 | |
063c977a LP |
2281 | r = append_static_mounts(&ml, |
2282 | protect_kernel_tunables_sys_table, | |
2283 | ELEMENTSOF(protect_kernel_tunables_sys_table), | |
2284 | p->ignore_protect_paths); | |
ddc155b2 | 2285 | if (r < 0) |
063c977a LP |
2286 | return r; |
2287 | } | |
ddc155b2 | 2288 | |
063c977a LP |
2289 | if (p->protect_kernel_modules) { |
2290 | r = append_static_mounts(&ml, | |
2291 | protect_kernel_modules_table, | |
2292 | ELEMENTSOF(protect_kernel_modules_table), | |
2293 | p->ignore_protect_paths); | |
6c47cd7d | 2294 | if (r < 0) |
063c977a LP |
2295 | return r; |
2296 | } | |
6c47cd7d | 2297 | |
063c977a LP |
2298 | if (p->protect_kernel_logs) { |
2299 | r = append_static_mounts(&ml, | |
2300 | protect_kernel_logs_proc_table, | |
2301 | ELEMENTSOF(protect_kernel_logs_proc_table), | |
2302 | ignore_protect_proc); | |
d2d6c096 | 2303 | if (r < 0) |
063c977a | 2304 | return r; |
d2d6c096 | 2305 | |
063c977a LP |
2306 | r = append_static_mounts(&ml, |
2307 | protect_kernel_logs_dev_table, | |
2308 | ELEMENTSOF(protect_kernel_logs_dev_table), | |
2309 | p->ignore_protect_paths); | |
2abd4e38 | 2310 | if (r < 0) |
063c977a LP |
2311 | return r; |
2312 | } | |
2abd4e38 | 2313 | |
063c977a LP |
2314 | if (p->protect_control_groups) { |
2315 | MountEntry *me = mount_list_extend(&ml); | |
2316 | if (!me) | |
2317 | return log_oom_debug(); | |
56a13a49 | 2318 | |
063c977a LP |
2319 | *me = (MountEntry) { |
2320 | .path_const = "/sys/fs/cgroup", | |
a868e437 | 2321 | .mode = MOUNT_READ_ONLY, |
063c977a LP |
2322 | }; |
2323 | } | |
7ff7394d | 2324 | |
063c977a LP |
2325 | r = append_protect_home(&ml, p->protect_home, p->ignore_protect_paths); |
2326 | if (r < 0) | |
2327 | return r; | |
56a13a49 | 2328 | |
063c977a LP |
2329 | r = append_protect_system(&ml, p->protect_system, false); |
2330 | if (r < 0) | |
2331 | return r; | |
ac0930c8 | 2332 | |
063c977a LP |
2333 | if (namespace_parameters_mount_apivfs(p)) { |
2334 | r = append_static_mounts(&ml, | |
2335 | apivfs_table, | |
2336 | ELEMENTSOF(apivfs_table), | |
2337 | p->ignore_protect_paths); | |
b3d13314 | 2338 | if (r < 0) |
063c977a LP |
2339 | return r; |
2340 | } | |
b3d13314 | 2341 | |
063c977a LP |
2342 | /* Note, if proc is mounted with subset=pid then neither of the two paths will exist, i.e. they are |
2343 | * implicitly protected by the mount option. */ | |
2344 | if (p->protect_hostname) { | |
3d1b999b LP |
2345 | r = append_static_mounts( |
2346 | &ml, | |
2347 | protect_hostname_table, | |
2348 | ELEMENTSOF(protect_hostname_table), | |
2349 | ignore_protect_proc); | |
2350 | if (r < 0) | |
2351 | return r; | |
063c977a | 2352 | } |
788e7201 | 2353 | |
063c977a LP |
2354 | if (p->private_network) { |
2355 | MountEntry *me = mount_list_extend(&ml); | |
2356 | if (!me) | |
2357 | return log_oom_debug(); | |
94a7b275 | 2358 | |
063c977a LP |
2359 | *me = (MountEntry) { |
2360 | .path_const = "/sys", | |
a868e437 | 2361 | .mode = MOUNT_PRIVATE_SYSFS, |
063c977a LP |
2362 | }; |
2363 | } | |
59eeb84b | 2364 | |
063c977a LP |
2365 | if (p->private_ipc) { |
2366 | MountEntry *me = mount_list_extend(&ml); | |
2367 | if (!me) | |
2368 | return log_oom_debug(); | |
417116f2 | 2369 | |
063c977a LP |
2370 | *me = (MountEntry) { |
2371 | .path_const = "/dev/mqueue", | |
a868e437 | 2372 | .mode = MOUNT_MQUEUEFS, |
063c977a LP |
2373 | .flags = MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, |
2374 | }; | |
2375 | } | |
417116f2 | 2376 | |
063c977a LP |
2377 | if (p->creds_path) { |
2378 | /* If our service has a credentials store configured, then bind that one in, but hide | |
2379 | * everything else. */ | |
5d997827 | 2380 | |
063c977a LP |
2381 | MountEntry *me = mount_list_extend(&ml); |
2382 | if (!me) | |
2383 | return log_oom_debug(); | |
aecd5ac6 | 2384 | |
063c977a LP |
2385 | *me = (MountEntry) { |
2386 | .path_const = "/run/credentials", | |
a868e437 | 2387 | .mode = MOUNT_TMPFS, |
063c977a LP |
2388 | .read_only = true, |
2389 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, | |
2390 | .flags = MS_NODEV|MS_STRICTATIME|MS_NOSUID|MS_NOEXEC, | |
2391 | }; | |
bbb4e7f3 | 2392 | |
063c977a LP |
2393 | me = mount_list_extend(&ml); |
2394 | if (!me) | |
2395 | return log_oom_debug(); | |
91dd5f7c | 2396 | |
063c977a LP |
2397 | *me = (MountEntry) { |
2398 | .path_const = p->creds_path, | |
a868e437 | 2399 | .mode = MOUNT_BIND, |
063c977a LP |
2400 | .read_only = true, |
2401 | .source_const = p->creds_path, | |
2402 | .ignore = true, | |
2403 | }; | |
2404 | } else { | |
2405 | /* If our service has no credentials store configured, then make the whole credentials tree | |
2406 | * inaccessible wholesale. */ | |
91dd5f7c | 2407 | |
063c977a LP |
2408 | MountEntry *me = mount_list_extend(&ml); |
2409 | if (!me) | |
2410 | return log_oom_debug(); | |
91dd5f7c | 2411 | |
063c977a LP |
2412 | *me = (MountEntry) { |
2413 | .path_const = "/run/credentials", | |
a868e437 | 2414 | .mode = MOUNT_INACCESSIBLE, |
063c977a LP |
2415 | .ignore = true, |
2416 | }; | |
2417 | } | |
2418 | ||
2419 | if (p->log_namespace) { | |
2420 | _cleanup_free_ char *q = NULL; | |
2421 | ||
2422 | q = strjoin("/run/systemd/journal.", p->log_namespace); | |
2423 | if (!q) | |
d4f0878e | 2424 | return log_oom_debug(); |
5327c910 | 2425 | |
063c977a LP |
2426 | MountEntry *me = mount_list_extend(&ml); |
2427 | if (!me) | |
2428 | return log_oom_debug(); | |
2429 | ||
2430 | *me = (MountEntry) { | |
2431 | .path_const = "/run/systemd/journal", | |
a868e437 | 2432 | .mode = MOUNT_BIND_RECURSIVE, |
063c977a LP |
2433 | .read_only = true, |
2434 | .source_malloc = TAKE_PTR(q), | |
2435 | }; | |
15ae422b LP |
2436 | } |
2437 | ||
063c977a LP |
2438 | /* Will be used to add bind mounts at runtime */ |
2439 | if (setup_propagate) { | |
2440 | MountEntry *me = mount_list_extend(&ml); | |
2441 | if (!me) | |
2442 | return log_oom_debug(); | |
2443 | ||
2444 | *me = (MountEntry) { | |
2445 | .source_const = p->propagate_dir, | |
2446 | .path_const = p->incoming_dir, | |
a868e437 | 2447 | .mode = MOUNT_BIND, |
063c977a LP |
2448 | .read_only = true, |
2449 | }; | |
2450 | } | |
2451 | ||
2452 | if (p->notify_socket) { | |
2453 | MountEntry *me = mount_list_extend(&ml); | |
2454 | if (!me) | |
2455 | return log_oom_debug(); | |
2456 | ||
2457 | *me = (MountEntry) { | |
2458 | .path_const = p->notify_socket, | |
2459 | .source_const = p->notify_socket, | |
a868e437 | 2460 | .mode = MOUNT_BIND, |
063c977a LP |
2461 | .read_only = true, |
2462 | }; | |
2463 | } | |
2464 | ||
2465 | if (p->host_os_release_stage) { | |
2466 | MountEntry *me = mount_list_extend(&ml); | |
2467 | if (!me) | |
2468 | return log_oom_debug(); | |
2469 | ||
2470 | *me = (MountEntry) { | |
2471 | .path_const = "/run/host/.os-release-stage/", | |
2472 | .source_const = p->host_os_release_stage, | |
a868e437 | 2473 | .mode = MOUNT_BIND, |
063c977a LP |
2474 | .read_only = true, |
2475 | .ignore = true, /* Live copy, don't hard-fail if it goes missing */ | |
2476 | }; | |
2477 | } | |
2478 | ||
2479 | /* Prepend the root directory where that's necessary */ | |
2480 | r = prefix_where_needed(&ml, root); | |
2481 | if (r < 0) | |
2482 | return r; | |
2483 | ||
2484 | drop_unused_mounts(&ml, root); | |
2485 | ||
1beab8b0 LP |
2486 | /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */ |
2487 | ||
d944dc95 | 2488 | if (unshare(CLONE_NEWNS) < 0) { |
763a260a | 2489 | r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m"); |
063c977a | 2490 | |
eacfbd89 LP |
2491 | if (ERRNO_IS_PRIVILEGE(r) || |
2492 | ERRNO_IS_NOT_SUPPORTED(r)) | |
cbc056c8 ZJS |
2493 | /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter |
2494 | * in place that doesn't allow us to create namespaces (or a missing cap), then | |
2495 | * propagate a recognizable error back, which the caller can use to detect this case | |
2496 | * (and only this) and optionally continue without namespacing applied. */ | |
063c977a | 2497 | return -ENOANO; |
1beab8b0 | 2498 | |
063c977a | 2499 | return r; |
d944dc95 | 2500 | } |
1e4e94c8 | 2501 | |
5e8deb94 LB |
2502 | /* Create the source directory to allow runtime propagation of mounts */ |
2503 | if (setup_propagate) | |
79d956db | 2504 | (void) mkdir_p(p->propagate_dir, 0600); |
5e8deb94 | 2505 | |
79d956db | 2506 | if (p->n_extension_images > 0 || !strv_isempty(p->extension_directories)) |
3ebf0b0b LB |
2507 | /* ExtensionImages/Directories mountpoint directories will be created while parsing the |
2508 | * mounts to create, so have the parent ready */ | |
79d956db | 2509 | (void) mkdir_p(p->extension_dir, 0600); |
93f59701 | 2510 | |
9b68367b YW |
2511 | /* Remount / as SLAVE so that nothing now mounted in the namespace |
2512 | * shows up in the parent */ | |
063c977a LP |
2513 | if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) |
2514 | return log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m"); | |
ee818b89 | 2515 | |
79d956db | 2516 | if (p->root_image) { |
e908468b | 2517 | /* A root image is specified, mount it to the right place */ |
8d9a1d59 LP |
2518 | r = dissected_image_mount( |
2519 | dissected_image, | |
2520 | root, | |
2521 | /* uid_shift= */ UID_INVALID, | |
2522 | /* uid_range= */ UID_INVALID, | |
2523 | /* userns_fd= */ -EBADF, | |
2524 | dissect_image_flags); | |
063c977a LP |
2525 | if (r < 0) |
2526 | return log_debug_errno(r, "Failed to mount root image: %m"); | |
41bc4849 LP |
2527 | |
2528 | /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device | |
2529 | * if it likes. */ | |
2530 | r = loop_device_flock(loop_device, LOCK_UN); | |
063c977a LP |
2531 | if (r < 0) |
2532 | return log_debug_errno(r, "Failed to release lock on loopback block device: %m"); | |
915e6d16 | 2533 | |
3044d343 | 2534 | r = dissected_image_relinquish(dissected_image); |
063c977a LP |
2535 | if (r < 0) |
2536 | return log_debug_errno(r, "Failed to relinquish dissected image: %m"); | |
78ebe980 | 2537 | |
79d956db | 2538 | } else if (p->root_directory) { |
915e6d16 | 2539 | |
e908468b LP |
2540 | /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */ |
2541 | r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW); | |
063c977a LP |
2542 | if (r < 0) |
2543 | return log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root); | |
8f1ad200 | 2544 | if (r == 0) { |
21935150 LP |
2545 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); |
2546 | if (r < 0) | |
063c977a | 2547 | return r; |
d944dc95 | 2548 | } |
e908468b | 2549 | |
9b68367b | 2550 | } else { |
e908468b | 2551 | /* Let's mount the main root directory to the root directory to use */ |
21935150 LP |
2552 | r = mount_nofollow_verbose(LOG_DEBUG, "/", root, NULL, MS_BIND|MS_REC, NULL); |
2553 | if (r < 0) | |
063c977a | 2554 | return r; |
ee818b89 | 2555 | } |
c2c13f2d | 2556 | |
4e0c20de | 2557 | /* Try to set up the new root directory before mounting anything else there. */ |
79d956db | 2558 | if (p->root_image || p->root_directory) |
4e0c20de LP |
2559 | (void) base_filesystem_create(root, UID_INVALID, GID_INVALID); |
2560 | ||
82fb2da2 | 2561 | /* Now make the magic happen */ |
063c977a | 2562 | r = apply_mounts(&ml, root, p, error_path); |
82fb2da2 | 2563 | if (r < 0) |
063c977a | 2564 | return r; |
15ae422b | 2565 | |
9b68367b | 2566 | /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */ |
9d50f850 | 2567 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
79d956db | 2568 | if (r == -EINVAL && p->root_directory) { |
ea63a260 LB |
2569 | /* If we are using root_directory and we don't have privileges (ie: user manager in a user |
2570 | * namespace) and the root_directory is already a mount point in the parent namespace, | |
2571 | * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than | |
2572 | * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a | |
2573 | * mount point) and try again. */ | |
2574 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); | |
2575 | if (r < 0) | |
063c977a | 2576 | return r; |
9d50f850 | 2577 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
ea63a260 | 2578 | } |
063c977a LP |
2579 | if (r < 0) |
2580 | return log_debug_errno(r, "Failed to mount root with MS_MOVE: %m"); | |
ee818b89 | 2581 | |
874cdcbc LP |
2582 | /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to |
2583 | * the host, since what's disconnected is disconnected. */ | |
063c977a LP |
2584 | if (mount(NULL, "/", NULL, mount_propagation_flag | MS_REC, NULL) < 0) |
2585 | return log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m"); | |
15ae422b | 2586 | |
bbe92ea5 MY |
2587 | /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only supported for |
2588 | * non-shared mounts. This needs to happen after remounting / or it will fail. */ | |
2589 | if (setup_propagate && mount(NULL, p->incoming_dir, NULL, MS_SLAVE, NULL) < 0) | |
2590 | return log_debug_errno(errno, "Failed to remount %s with MS_SLAVE: %m", p->incoming_dir); | |
5e8deb94 | 2591 | |
063c977a | 2592 | return 0; |
613b411c LP |
2593 | } |
2594 | ||
da6053d0 | 2595 | void bind_mount_free_many(BindMount *b, size_t n) { |
d2d6c096 LP |
2596 | assert(b || n == 0); |
2597 | ||
fe96c0f8 | 2598 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
2599 | free(b[i].source); |
2600 | free(b[i].destination); | |
2601 | } | |
2602 | ||
2603 | free(b); | |
2604 | } | |
2605 | ||
da6053d0 | 2606 | int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) { |
d2d6c096 LP |
2607 | _cleanup_free_ char *s = NULL, *d = NULL; |
2608 | BindMount *c; | |
2609 | ||
2610 | assert(b); | |
2611 | assert(n); | |
2612 | assert(item); | |
2613 | ||
2614 | s = strdup(item->source); | |
2615 | if (!s) | |
2616 | return -ENOMEM; | |
2617 | ||
2618 | d = strdup(item->destination); | |
2619 | if (!d) | |
2620 | return -ENOMEM; | |
2621 | ||
aa484f35 | 2622 | c = reallocarray(*b, *n + 1, sizeof(BindMount)); |
d2d6c096 LP |
2623 | if (!c) |
2624 | return -ENOMEM; | |
2625 | ||
2626 | *b = c; | |
2627 | ||
b3a9d980 | 2628 | c[(*n)++] = (BindMount) { |
1cc6c93a YW |
2629 | .source = TAKE_PTR(s), |
2630 | .destination = TAKE_PTR(d), | |
d2d6c096 | 2631 | .read_only = item->read_only, |
9ce4e4b0 | 2632 | .nosuid = item->nosuid, |
d2d6c096 LP |
2633 | .recursive = item->recursive, |
2634 | .ignore_enoent = item->ignore_enoent, | |
2635 | }; | |
2636 | ||
d2d6c096 LP |
2637 | return 0; |
2638 | } | |
2639 | ||
b3d13314 | 2640 | MountImage* mount_image_free_many(MountImage *m, size_t *n) { |
b3d13314 LB |
2641 | assert(n); |
2642 | assert(m || *n == 0); | |
2643 | ||
fe96c0f8 | 2644 | for (size_t i = 0; i < *n; i++) { |
b3d13314 LB |
2645 | free(m[i].source); |
2646 | free(m[i].destination); | |
427353f6 | 2647 | mount_options_free_all(m[i].mount_options); |
b3d13314 LB |
2648 | } |
2649 | ||
2650 | free(m); | |
2651 | *n = 0; | |
2652 | return NULL; | |
2653 | } | |
2654 | ||
2655 | int mount_image_add(MountImage **m, size_t *n, const MountImage *item) { | |
2656 | _cleanup_free_ char *s = NULL, *d = NULL; | |
427353f6 | 2657 | _cleanup_(mount_options_free_allp) MountOptions *options = NULL; |
b3d13314 LB |
2658 | MountImage *c; |
2659 | ||
2660 | assert(m); | |
2661 | assert(n); | |
2662 | assert(item); | |
2663 | ||
2664 | s = strdup(item->source); | |
2665 | if (!s) | |
2666 | return -ENOMEM; | |
2667 | ||
93f59701 LB |
2668 | if (item->destination) { |
2669 | d = strdup(item->destination); | |
2670 | if (!d) | |
2671 | return -ENOMEM; | |
2672 | } | |
b3d13314 | 2673 | |
427353f6 | 2674 | LIST_FOREACH(mount_options, i, item->mount_options) { |
c2b2df60 | 2675 | _cleanup_(mount_options_free_allp) MountOptions *o = NULL; |
427353f6 LB |
2676 | |
2677 | o = new(MountOptions, 1); | |
2678 | if (!o) | |
2679 | return -ENOMEM; | |
2680 | ||
2681 | *o = (MountOptions) { | |
2682 | .partition_designator = i->partition_designator, | |
2683 | .options = strdup(i->options), | |
2684 | }; | |
2685 | if (!o->options) | |
2686 | return -ENOMEM; | |
2687 | ||
2688 | LIST_APPEND(mount_options, options, TAKE_PTR(o)); | |
2689 | } | |
2690 | ||
b3d13314 LB |
2691 | c = reallocarray(*m, *n + 1, sizeof(MountImage)); |
2692 | if (!c) | |
2693 | return -ENOMEM; | |
2694 | ||
2695 | *m = c; | |
2696 | ||
b3a9d980 | 2697 | c[(*n)++] = (MountImage) { |
b3d13314 LB |
2698 | .source = TAKE_PTR(s), |
2699 | .destination = TAKE_PTR(d), | |
427353f6 | 2700 | .mount_options = TAKE_PTR(options), |
b3d13314 | 2701 | .ignore_enoent = item->ignore_enoent, |
93f59701 | 2702 | .type = item->type, |
b3d13314 LB |
2703 | }; |
2704 | ||
2705 | return 0; | |
2706 | } | |
2707 | ||
da6053d0 | 2708 | void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) { |
2abd4e38 YW |
2709 | assert(t || n == 0); |
2710 | ||
fe96c0f8 | 2711 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
2712 | free(t[i].path); |
2713 | free(t[i].options); | |
2714 | } | |
2715 | ||
2716 | free(t); | |
2717 | } | |
2718 | ||
2719 | int temporary_filesystem_add( | |
2720 | TemporaryFileSystem **t, | |
da6053d0 | 2721 | size_t *n, |
2abd4e38 YW |
2722 | const char *path, |
2723 | const char *options) { | |
2724 | ||
2725 | _cleanup_free_ char *p = NULL, *o = NULL; | |
2726 | TemporaryFileSystem *c; | |
2727 | ||
2728 | assert(t); | |
2729 | assert(n); | |
2730 | assert(path); | |
2731 | ||
2732 | p = strdup(path); | |
2733 | if (!p) | |
2734 | return -ENOMEM; | |
2735 | ||
2736 | if (!isempty(options)) { | |
2737 | o = strdup(options); | |
2738 | if (!o) | |
2739 | return -ENOMEM; | |
2740 | } | |
2741 | ||
aa484f35 | 2742 | c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem)); |
2abd4e38 YW |
2743 | if (!c) |
2744 | return -ENOMEM; | |
2745 | ||
2746 | *t = c; | |
2747 | ||
b3a9d980 | 2748 | c[(*n)++] = (TemporaryFileSystem) { |
1cc6c93a YW |
2749 | .path = TAKE_PTR(p), |
2750 | .options = TAKE_PTR(o), | |
2abd4e38 YW |
2751 | }; |
2752 | ||
2abd4e38 YW |
2753 | return 0; |
2754 | } | |
2755 | ||
a652f050 JR |
2756 | static int make_tmp_prefix(const char *prefix) { |
2757 | _cleanup_free_ char *t = NULL; | |
254d1313 | 2758 | _cleanup_close_ int fd = -EBADF; |
a652f050 JR |
2759 | int r; |
2760 | ||
2761 | /* Don't do anything unless we know the dir is actually missing */ | |
2762 | r = access(prefix, F_OK); | |
2763 | if (r >= 0) | |
2764 | return 0; | |
2765 | if (errno != ENOENT) | |
2766 | return -errno; | |
2767 | ||
2053593f | 2768 | WITH_UMASK(000) |
30443439 | 2769 | r = mkdir_parents(prefix, 0755); |
a652f050 JR |
2770 | if (r < 0) |
2771 | return r; | |
2772 | ||
2773 | r = tempfn_random(prefix, NULL, &t); | |
2774 | if (r < 0) | |
2775 | return r; | |
2776 | ||
96603ea0 LP |
2777 | /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for |
2778 | * the suid bit, below. */ | |
2779 | fd = open_mkdir_at(AT_FDCWD, t, O_EXCL|O_CLOEXEC, 0777); | |
2780 | if (fd < 0) | |
2781 | return fd; | |
a652f050 | 2782 | |
96603ea0 LP |
2783 | r = RET_NERRNO(fchmod(fd, 01777)); |
2784 | if (r < 0) { | |
a652f050 JR |
2785 | (void) rmdir(t); |
2786 | return r; | |
2787 | } | |
2788 | ||
96603ea0 LP |
2789 | r = RET_NERRNO(rename(t, prefix)); |
2790 | if (r < 0) { | |
a652f050 JR |
2791 | (void) rmdir(t); |
2792 | return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */ | |
2793 | } | |
2794 | ||
2795 | return 0; | |
2796 | ||
2797 | } | |
2798 | ||
56a13a49 | 2799 | static int setup_one_tmp_dir(const char *id, const char *prefix, char **path, char **tmp_path) { |
613b411c | 2800 | _cleanup_free_ char *x = NULL; |
19cd4e19 | 2801 | _cleanup_free_ char *y = NULL; |
6b46ea73 | 2802 | sd_id128_t boot_id; |
56a13a49 | 2803 | bool rw = true; |
6b46ea73 | 2804 | int r; |
613b411c LP |
2805 | |
2806 | assert(id); | |
2807 | assert(prefix); | |
2808 | assert(path); | |
2809 | ||
6b46ea73 LP |
2810 | /* We include the boot id in the directory so that after a |
2811 | * reboot we can easily identify obsolete directories. */ | |
2812 | ||
2813 | r = sd_id128_get_boot(&boot_id); | |
2814 | if (r < 0) | |
2815 | return r; | |
2816 | ||
85b55869 | 2817 | x = strjoin(prefix, "/systemd-private-", SD_ID128_TO_STRING(boot_id), "-", id, "-XXXXXX"); |
613b411c LP |
2818 | if (!x) |
2819 | return -ENOMEM; | |
2820 | ||
a652f050 JR |
2821 | r = make_tmp_prefix(prefix); |
2822 | if (r < 0) | |
2823 | return r; | |
2824 | ||
2053593f | 2825 | WITH_UMASK(0077) |
56a13a49 ZJS |
2826 | if (!mkdtemp(x)) { |
2827 | if (errno == EROFS || ERRNO_IS_DISK_SPACE(errno)) | |
2828 | rw = false; | |
2829 | else | |
2830 | return -errno; | |
2831 | } | |
613b411c | 2832 | |
56a13a49 | 2833 | if (rw) { |
19cd4e19 | 2834 | y = strjoin(x, "/tmp"); |
2835 | if (!y) | |
2836 | return -ENOMEM; | |
2837 | ||
2053593f | 2838 | WITH_UMASK(0000) |
19cd4e19 | 2839 | if (mkdir(y, 0777 | S_ISVTX) < 0) |
7c76e181 | 2840 | return -errno; |
19cd4e19 | 2841 | |
03bc11d1 | 2842 | r = label_fix_full(AT_FDCWD, y, prefix, 0); |
56a13a49 ZJS |
2843 | if (r < 0) |
2844 | return r; | |
19cd4e19 | 2845 | |
2846 | if (tmp_path) | |
2847 | *tmp_path = TAKE_PTR(y); | |
56a13a49 ZJS |
2848 | } else { |
2849 | /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being | |
2850 | * read-only. This way the service will get the EROFS result as if it was writing to the real | |
2851 | * file system. */ | |
2053593f | 2852 | WITH_UMASK(0000) |
30443439 | 2853 | r = mkdir_p(RUN_SYSTEMD_EMPTY, 0500); |
56a13a49 ZJS |
2854 | if (r < 0) |
2855 | return r; | |
613b411c | 2856 | |
3f181262 LP |
2857 | r = free_and_strdup(&x, RUN_SYSTEMD_EMPTY); |
2858 | if (r < 0) | |
2859 | return r; | |
c17ec25e | 2860 | } |
15ae422b | 2861 | |
1cc6c93a | 2862 | *path = TAKE_PTR(x); |
613b411c LP |
2863 | return 0; |
2864 | } | |
2865 | ||
2866 | int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) { | |
56a13a49 ZJS |
2867 | _cleanup_(namespace_cleanup_tmpdirp) char *a = NULL; |
2868 | _cleanup_(rmdir_and_freep) char *a_tmp = NULL; | |
2869 | char *b; | |
613b411c LP |
2870 | int r; |
2871 | ||
2872 | assert(id); | |
2873 | assert(tmp_dir); | |
2874 | assert(var_tmp_dir); | |
2875 | ||
56a13a49 | 2876 | r = setup_one_tmp_dir(id, "/tmp", &a, &a_tmp); |
613b411c LP |
2877 | if (r < 0) |
2878 | return r; | |
2879 | ||
56a13a49 ZJS |
2880 | r = setup_one_tmp_dir(id, "/var/tmp", &b, NULL); |
2881 | if (r < 0) | |
613b411c | 2882 | return r; |
613b411c | 2883 | |
56a13a49 ZJS |
2884 | a_tmp = mfree(a_tmp); /* avoid rmdir */ |
2885 | *tmp_dir = TAKE_PTR(a); | |
2886 | *var_tmp_dir = TAKE_PTR(b); | |
613b411c LP |
2887 | |
2888 | return 0; | |
2889 | } | |
2890 | ||
13339577 | 2891 | int setup_shareable_ns(int ns_storage_socket[static 2], unsigned long nsflag) { |
5bb1d7fb | 2892 | _cleanup_close_ int ns = -EBADF; |
13339577 | 2893 | int r; |
54c2459d | 2894 | const char *ns_name, *ns_path; |
613b411c | 2895 | |
54c2459d XR |
2896 | assert(ns_storage_socket); |
2897 | assert(ns_storage_socket[0] >= 0); | |
2898 | assert(ns_storage_socket[1] >= 0); | |
2899 | ||
2900 | ns_name = namespace_single_flag_to_string(nsflag); | |
2901 | assert(ns_name); | |
613b411c LP |
2902 | |
2903 | /* We use the passed socketpair as a storage buffer for our | |
76cd584b LP |
2904 | * namespace reference fd. Whatever process runs this first |
2905 | * shall create a new namespace, all others should just join | |
2906 | * it. To serialize that we use a file lock on the socket | |
2907 | * pair. | |
613b411c LP |
2908 | * |
2909 | * It's a bit crazy, but hey, works great! */ | |
2910 | ||
13339577 DDM |
2911 | r = posix_lock(ns_storage_socket[0], LOCK_EX); |
2912 | if (r < 0) | |
2913 | return r; | |
613b411c | 2914 | |
13339577 | 2915 | CLEANUP_POSIX_UNLOCK(ns_storage_socket[0]); |
613b411c | 2916 | |
13339577 DDM |
2917 | ns = receive_one_fd(ns_storage_socket[0], MSG_PEEK|MSG_DONTWAIT); |
2918 | if (ns >= 0) { | |
2919 | /* Yay, found something, so let's join the namespace */ | |
2920 | r = RET_NERRNO(setns(ns, nsflag)); | |
2921 | if (r < 0) | |
2922 | return r; | |
613b411c | 2923 | |
13339577 DDM |
2924 | return 0; |
2925 | } | |
613b411c | 2926 | |
13339577 DDM |
2927 | if (ns != -EAGAIN) |
2928 | return ns; | |
613b411c | 2929 | |
13339577 | 2930 | /* Nothing stored yet, so let's create a new namespace. */ |
613b411c | 2931 | |
13339577 DDM |
2932 | if (unshare(nsflag) < 0) |
2933 | return -errno; | |
613b411c | 2934 | |
13339577 | 2935 | (void) loopback_setup(); |
613b411c | 2936 | |
13339577 DDM |
2937 | ns_path = strjoina("/proc/self/ns/", ns_name); |
2938 | ns = open(ns_path, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
2939 | if (ns < 0) | |
2940 | return -errno; | |
613b411c | 2941 | |
13339577 DDM |
2942 | r = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
2943 | if (r < 0) | |
2944 | return r; | |
613b411c | 2945 | |
13339577 | 2946 | return 1; |
15ae422b | 2947 | } |
417116f2 | 2948 | |
13339577 | 2949 | int open_shareable_ns_path(int ns_storage_socket[static 2], const char *path, unsigned long nsflag) { |
5bb1d7fb | 2950 | _cleanup_close_ int ns = -EBADF; |
13339577 | 2951 | int r; |
51af7fb2 | 2952 | |
54c2459d XR |
2953 | assert(ns_storage_socket); |
2954 | assert(ns_storage_socket[0] >= 0); | |
2955 | assert(ns_storage_socket[1] >= 0); | |
51af7fb2 LP |
2956 | assert(path); |
2957 | ||
54c2459d XR |
2958 | /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in |
2959 | * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will | |
2960 | * allocate a new anonymous ns if needed. */ | |
51af7fb2 | 2961 | |
13339577 DDM |
2962 | r = posix_lock(ns_storage_socket[0], LOCK_EX); |
2963 | if (r < 0) | |
2964 | return r; | |
51af7fb2 | 2965 | |
13339577 | 2966 | CLEANUP_POSIX_UNLOCK(ns_storage_socket[0]); |
51af7fb2 | 2967 | |
13339577 DDM |
2968 | ns = receive_one_fd(ns_storage_socket[0], MSG_PEEK|MSG_DONTWAIT); |
2969 | if (ns >= 0) | |
2970 | return 0; | |
2971 | if (ns != -EAGAIN) | |
2972 | return ns; | |
51af7fb2 | 2973 | |
13339577 | 2974 | /* Nothing stored yet. Open the file from the file system. */ |
51af7fb2 | 2975 | |
13339577 DDM |
2976 | ns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC); |
2977 | if (ns < 0) | |
2978 | return -errno; | |
51af7fb2 | 2979 | |
13339577 DDM |
2980 | r = fd_is_ns(ns, nsflag); |
2981 | if (r == 0) | |
2982 | return -EINVAL; | |
2983 | if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */ | |
2984 | return r; | |
51af7fb2 | 2985 | |
13339577 DDM |
2986 | r = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
2987 | if (r < 0) | |
2988 | return r; | |
51af7fb2 | 2989 | |
13339577 | 2990 | return 1; |
51af7fb2 LP |
2991 | } |
2992 | ||
6e2d7c4f MS |
2993 | bool ns_type_supported(NamespaceType type) { |
2994 | const char *t, *ns_proc; | |
2995 | ||
0fa5b831 LP |
2996 | t = namespace_type_to_string(type); |
2997 | if (!t) /* Don't know how to translate this? Then it's not supported */ | |
6e2d7c4f MS |
2998 | return false; |
2999 | ||
6e2d7c4f | 3000 | ns_proc = strjoina("/proc/self/ns/", t); |
6e2d7c4f MS |
3001 | return access(ns_proc, F_OK) == 0; |
3002 | } | |
3003 | ||
1b8689f9 | 3004 | static const char *const protect_home_table[_PROTECT_HOME_MAX] = { |
cbc056c8 ZJS |
3005 | [PROTECT_HOME_NO] = "no", |
3006 | [PROTECT_HOME_YES] = "yes", | |
1b8689f9 | 3007 | [PROTECT_HOME_READ_ONLY] = "read-only", |
cbc056c8 | 3008 | [PROTECT_HOME_TMPFS] = "tmpfs", |
417116f2 LP |
3009 | }; |
3010 | ||
1e8c7bd5 | 3011 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES); |
5e1c6154 | 3012 | |
1b8689f9 | 3013 | static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = { |
cbc056c8 ZJS |
3014 | [PROTECT_SYSTEM_NO] = "no", |
3015 | [PROTECT_SYSTEM_YES] = "yes", | |
3016 | [PROTECT_SYSTEM_FULL] = "full", | |
3f815163 | 3017 | [PROTECT_SYSTEM_STRICT] = "strict", |
1b8689f9 LP |
3018 | }; |
3019 | ||
1e8c7bd5 | 3020 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES); |
03c791aa | 3021 | |
6e2d7c4f | 3022 | static const char* const namespace_type_table[] = { |
cbc056c8 | 3023 | [NAMESPACE_MOUNT] = "mnt", |
6e2d7c4f | 3024 | [NAMESPACE_CGROUP] = "cgroup", |
cbc056c8 ZJS |
3025 | [NAMESPACE_UTS] = "uts", |
3026 | [NAMESPACE_IPC] = "ipc", | |
3027 | [NAMESPACE_USER] = "user", | |
3028 | [NAMESPACE_PID] = "pid", | |
3029 | [NAMESPACE_NET] = "net", | |
c3b9c418 | 3030 | [NAMESPACE_TIME] = "time", |
6e2d7c4f MS |
3031 | }; |
3032 | ||
3033 | DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType); | |
4e399953 LP |
3034 | |
3035 | static const char* const protect_proc_table[_PROTECT_PROC_MAX] = { | |
3036 | [PROTECT_PROC_DEFAULT] = "default", | |
3037 | [PROTECT_PROC_NOACCESS] = "noaccess", | |
3038 | [PROTECT_PROC_INVISIBLE] = "invisible", | |
3039 | [PROTECT_PROC_PTRACEABLE] = "ptraceable", | |
3040 | }; | |
3041 | ||
3042 | DEFINE_STRING_TABLE_LOOKUP(protect_proc, ProtectProc); | |
3043 | ||
3044 | static const char* const proc_subset_table[_PROC_SUBSET_MAX] = { | |
3045 | [PROC_SUBSET_ALL] = "all", | |
3046 | [PROC_SUBSET_PID] = "pid", | |
3047 | }; | |
3048 | ||
3049 | DEFINE_STRING_TABLE_LOOKUP(proc_subset, ProcSubset); |