]>
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" |
f4351959 | 16 | #include "chase-symlinks.h" |
7f112f50 | 17 | #include "dev-setup.h" |
ec61371f | 18 | #include "devnum-util.h" |
93f59701 LB |
19 | #include "env-util.h" |
20 | #include "escape.h" | |
21 | #include "extension-release.h" | |
3ffd4af2 | 22 | #include "fd-util.h" |
e5f10caf | 23 | #include "format-util.h" |
e2341b6b | 24 | #include "glyph-util.h" |
e908468b | 25 | #include "label.h" |
b3d13314 | 26 | #include "list.h" |
915e6d16 | 27 | #include "loop-util.h" |
07630cea | 28 | #include "loopback-setup.h" |
1c265fcd | 29 | #include "missing_syscall.h" |
35cd0ba5 | 30 | #include "mkdir-label.h" |
4349cd7c | 31 | #include "mount-util.h" |
049af8ad | 32 | #include "mountpoint-util.h" |
0cb8e3d1 | 33 | #include "namespace-util.h" |
3ffd4af2 | 34 | #include "namespace.h" |
54c2459d | 35 | #include "nsflags.h" |
d8b4d14d | 36 | #include "nulstr-util.h" |
93f59701 | 37 | #include "os-util.h" |
07630cea | 38 | #include "path-util.h" |
d7b8eec7 | 39 | #include "selinux-util.h" |
2583fbea | 40 | #include "socket-util.h" |
760877e9 | 41 | #include "sort-util.h" |
36ce7110 | 42 | #include "stat-util.h" |
8b43440b | 43 | #include "string-table.h" |
07630cea LP |
44 | #include "string-util.h" |
45 | #include "strv.h" | |
a652f050 | 46 | #include "tmpfile-util.h" |
affb60b1 | 47 | #include "umask-util.h" |
ee104e11 | 48 | #include "user-util.h" |
15ae422b | 49 | |
737ba3c8 | 50 | #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC) |
51 | ||
c17ec25e | 52 | typedef enum MountMode { |
15ae422b LP |
53 | /* This is ordered by priority! */ |
54 | INACCESSIBLE, | |
93f59701 | 55 | OVERLAY_MOUNT, |
b3d13314 | 56 | MOUNT_IMAGES, |
d2d6c096 LP |
57 | BIND_MOUNT, |
58 | BIND_MOUNT_RECURSIVE, | |
ac0930c8 | 59 | PRIVATE_TMP, |
56a13a49 | 60 | PRIVATE_TMP_READONLY, |
7f112f50 | 61 | PRIVATE_DEV, |
5d997827 | 62 | BIND_DEV, |
6c47cd7d | 63 | EMPTY_DIR, |
5d997827 LP |
64 | SYSFS, |
65 | PROCFS, | |
66 | READONLY, | |
59eeb84b | 67 | READWRITE, |
ddc155b2 TM |
68 | NOEXEC, |
69 | EXEC, | |
2abd4e38 | 70 | TMPFS, |
12413985 | 71 | RUN, |
a07b9926 | 72 | EXTENSION_DIRECTORIES, /* Bind-mounted outside the root directory, and used by subsequent mounts */ |
93f59701 | 73 | EXTENSION_IMAGES, /* Mounted outside the root directory, and used by subsequent mounts */ |
80271a44 | 74 | MQUEUEFS, |
1e05071d | 75 | READWRITE_IMPLICIT, /* Should have the lowest priority. */ |
5beb8688 | 76 | _MOUNT_MODE_MAX, |
c17ec25e | 77 | } MountMode; |
15ae422b | 78 | |
34de407a | 79 | typedef struct MountEntry { |
5327c910 | 80 | const char *path_const; /* Memory allocated on stack or static */ |
cfbeb4ef | 81 | MountMode mode:5; |
5327c910 LP |
82 | bool ignore:1; /* Ignore if path does not exist? */ |
83 | bool has_prefix:1; /* Already is prefixed by the root dir? */ | |
cfbeb4ef | 84 | bool read_only:1; /* Shall this mount point be read-only? */ |
9ce4e4b0 | 85 | bool nosuid:1; /* Shall set MS_NOSUID on the mount itself */ |
ddc155b2 TM |
86 | bool noexec:1; /* Shall set MS_NOEXEC on the mount itself */ |
87 | bool exec:1; /* Shall clear MS_NOEXEC on the mount itself */ | |
088696fe | 88 | bool applied:1; /* Already applied */ |
55fe7432 | 89 | char *path_malloc; /* Use this instead of 'path_const' if we had to allocate memory */ |
809ceb82 LB |
90 | const char *unprefixed_path_const; /* If the path was amended with a prefix, these will save the original */ |
91 | char *unprefixed_path_malloc; | |
b3d13314 | 92 | const char *source_const; /* The source path, for bind mounts or images */ |
d2d6c096 | 93 | char *source_malloc; |
2abd4e38 YW |
94 | const char *options_const;/* Mount options for tmpfs */ |
95 | char *options_malloc; | |
96 | unsigned long flags; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */ | |
088696fe | 97 | unsigned n_followed; |
427353f6 | 98 | LIST_HEAD(MountOptions, image_options); |
34de407a | 99 | } MountEntry; |
15ae422b | 100 | |
94293d65 | 101 | /* 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 | 102 | * something there already. These mounts are hence overridden by any other explicitly configured mounts. */ |
5d997827 LP |
103 | static const MountEntry apivfs_table[] = { |
104 | { "/proc", PROCFS, false }, | |
105 | { "/dev", BIND_DEV, false }, | |
106 | { "/sys", SYSFS, false }, | |
9f563f27 | 107 | { "/run", RUN, false, .options_const = "mode=0755" TMPFS_LIMITS_RUN, .flags = MS_NOSUID|MS_NODEV|MS_STRICTATIME }, |
5d997827 | 108 | }; |
f471b2af | 109 | |
11a30cec | 110 | /* ProtectKernelTunables= option and the related filesystem APIs */ |
788e7201 | 111 | static const MountEntry protect_kernel_tunables_proc_table[] = { |
1e05071d YW |
112 | { "/proc/acpi", READONLY, true }, |
113 | { "/proc/apm", READONLY, true }, /* Obsolete API, there's no point in permitting access to this, ever */ | |
114 | { "/proc/asound", READONLY, true }, | |
115 | { "/proc/bus", READONLY, true }, | |
116 | { "/proc/fs", READONLY, true }, | |
117 | { "/proc/irq", READONLY, true }, | |
118 | { "/proc/kallsyms", INACCESSIBLE, true }, | |
119 | { "/proc/kcore", INACCESSIBLE, true }, | |
120 | { "/proc/latency_stats", READONLY, true }, | |
121 | { "/proc/mtrr", READONLY, true }, | |
122 | { "/proc/scsi", READONLY, true }, | |
4e399953 | 123 | { "/proc/sys", READONLY, true }, |
1e05071d YW |
124 | { "/proc/sysrq-trigger", READONLY, true }, |
125 | { "/proc/timer_stats", READONLY, true }, | |
788e7201 TM |
126 | }; |
127 | ||
128 | static const MountEntry protect_kernel_tunables_sys_table[] = { | |
1e05071d YW |
129 | { "/sys", READONLY, false }, |
130 | { "/sys/fs/bpf", READONLY, true }, | |
131 | { "/sys/fs/cgroup", READWRITE_IMPLICIT, false }, /* READONLY is set by ProtectControlGroups= option */ | |
132 | { "/sys/fs/selinux", READWRITE_IMPLICIT, true }, | |
133 | { "/sys/kernel/debug", READONLY, true }, | |
134 | { "/sys/kernel/tracing", READONLY, true }, | |
11a30cec DH |
135 | }; |
136 | ||
c575770b | 137 | /* ProtectKernelModules= option */ |
34de407a | 138 | static const MountEntry protect_kernel_modules_table[] = { |
349cc4a5 | 139 | #if HAVE_SPLIT_USR |
c6232fb0 | 140 | { "/lib/modules", INACCESSIBLE, true }, |
c575770b | 141 | #endif |
c6232fb0 | 142 | { "/usr/lib/modules", INACCESSIBLE, true }, |
c575770b DH |
143 | }; |
144 | ||
94a7b275 | 145 | /* ProtectKernelLogs= option */ |
788e7201 | 146 | static const MountEntry protect_kernel_logs_proc_table[] = { |
94a7b275 | 147 | { "/proc/kmsg", INACCESSIBLE, true }, |
788e7201 TM |
148 | }; |
149 | ||
150 | static const MountEntry protect_kernel_logs_dev_table[] = { | |
94a7b275 KK |
151 | { "/dev/kmsg", INACCESSIBLE, true }, |
152 | }; | |
153 | ||
b6c432ca DH |
154 | /* |
155 | * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of | |
156 | * system should be protected by ProtectSystem= | |
157 | */ | |
34de407a | 158 | static const MountEntry protect_home_read_only_table[] = { |
c6232fb0 LP |
159 | { "/home", READONLY, true }, |
160 | { "/run/user", READONLY, true }, | |
161 | { "/root", READONLY, true }, | |
b6c432ca DH |
162 | }; |
163 | ||
e4da7d8c YW |
164 | /* ProtectHome=tmpfs table */ |
165 | static const MountEntry protect_home_tmpfs_table[] = { | |
7d85383e TM |
166 | { "/home", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, |
167 | { "/run/user", TMPFS, true, .read_only = true, .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
168 | { "/root", TMPFS, true, .read_only = true, .options_const = "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST, .flags = MS_NODEV|MS_STRICTATIME }, | |
e4da7d8c YW |
169 | }; |
170 | ||
b6c432ca | 171 | /* ProtectHome=yes table */ |
34de407a | 172 | static const MountEntry protect_home_yes_table[] = { |
c6232fb0 LP |
173 | { "/home", INACCESSIBLE, true }, |
174 | { "/run/user", INACCESSIBLE, true }, | |
175 | { "/root", INACCESSIBLE, true }, | |
b6c432ca DH |
176 | }; |
177 | ||
f471b2af | 178 | /* ProtectSystem=yes table */ |
34de407a | 179 | static const MountEntry protect_system_yes_table[] = { |
c6232fb0 LP |
180 | { "/usr", READONLY, false }, |
181 | { "/boot", READONLY, true }, | |
182 | { "/efi", READONLY, true }, | |
7486f305 AB |
183 | #if HAVE_SPLIT_USR |
184 | { "/lib", READONLY, true }, | |
185 | { "/lib64", READONLY, true }, | |
186 | { "/bin", READONLY, true }, | |
671f0f8d | 187 | # if HAVE_SPLIT_BIN |
7486f305 | 188 | { "/sbin", READONLY, true }, |
671f0f8d | 189 | # endif |
7486f305 | 190 | #endif |
f471b2af DH |
191 | }; |
192 | ||
193 | /* ProtectSystem=full includes ProtectSystem=yes */ | |
34de407a | 194 | static const MountEntry protect_system_full_table[] = { |
c6232fb0 LP |
195 | { "/usr", READONLY, false }, |
196 | { "/boot", READONLY, true }, | |
197 | { "/efi", READONLY, true }, | |
198 | { "/etc", READONLY, false }, | |
7486f305 AB |
199 | #if HAVE_SPLIT_USR |
200 | { "/lib", READONLY, true }, | |
201 | { "/lib64", READONLY, true }, | |
202 | { "/bin", READONLY, true }, | |
671f0f8d | 203 | # if HAVE_SPLIT_BIN |
7486f305 | 204 | { "/sbin", READONLY, true }, |
671f0f8d | 205 | # endif |
7486f305 | 206 | #endif |
f471b2af DH |
207 | }; |
208 | ||
209 | /* | |
210 | * ProtectSystem=strict table. In this strict mode, we mount everything | |
211 | * read-only, except for /proc, /dev, /sys which are the kernel API VFS, | |
212 | * which are left writable, but PrivateDevices= + ProtectKernelTunables= | |
213 | * protect those, and these options should be fully orthogonal. | |
214 | * (And of course /home and friends are also left writable, as ProtectHome= | |
215 | * shall manage those, orthogonally). | |
216 | */ | |
34de407a | 217 | static const MountEntry protect_system_strict_table[] = { |
1e05071d YW |
218 | { "/", READONLY, false }, |
219 | { "/proc", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
220 | { "/sys", READWRITE_IMPLICIT, false }, /* ProtectKernelTunables= */ | |
221 | { "/dev", READWRITE_IMPLICIT, false }, /* PrivateDevices= */ | |
222 | { "/home", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
223 | { "/run/user", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
224 | { "/root", READWRITE_IMPLICIT, true }, /* ProtectHome= */ | |
f471b2af DH |
225 | }; |
226 | ||
5beb8688 YW |
227 | static const char * const mount_mode_table[_MOUNT_MODE_MAX] = { |
228 | [INACCESSIBLE] = "inaccessible", | |
93f59701 | 229 | [OVERLAY_MOUNT] = "overlay", |
5beb8688 YW |
230 | [BIND_MOUNT] = "bind", |
231 | [BIND_MOUNT_RECURSIVE] = "rbind", | |
232 | [PRIVATE_TMP] = "private-tmp", | |
233 | [PRIVATE_DEV] = "private-dev", | |
234 | [BIND_DEV] = "bind-dev", | |
235 | [EMPTY_DIR] = "empty", | |
236 | [SYSFS] = "sysfs", | |
237 | [PROCFS] = "procfs", | |
238 | [READONLY] = "read-only", | |
239 | [READWRITE] = "read-write", | |
240 | [TMPFS] = "tmpfs", | |
b3d13314 | 241 | [MOUNT_IMAGES] = "mount-images", |
5beb8688 | 242 | [READWRITE_IMPLICIT] = "rw-implicit", |
ddc155b2 TM |
243 | [EXEC] = "exec", |
244 | [NOEXEC] = "noexec", | |
80271a44 | 245 | [MQUEUEFS] = "mqueuefs", |
5beb8688 YW |
246 | }; |
247 | ||
248 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode, MountMode); | |
249 | ||
34de407a | 250 | static const char *mount_entry_path(const MountEntry *p) { |
f0a4feb0 DH |
251 | assert(p); |
252 | ||
5327c910 LP |
253 | /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that, |
254 | * otherwise the stack/static ->path field is returned. */ | |
f0a4feb0 | 255 | |
5327c910 | 256 | return p->path_malloc ?: p->path_const; |
f0a4feb0 DH |
257 | } |
258 | ||
809ceb82 LB |
259 | static const char *mount_entry_unprefixed_path(const MountEntry *p) { |
260 | assert(p); | |
261 | ||
262 | /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */ | |
263 | ||
264 | return p->unprefixed_path_malloc ?: p->unprefixed_path_const ?: mount_entry_path(p); | |
265 | } | |
266 | ||
267 | static void mount_entry_consume_prefix(MountEntry *p, char *new_path) { | |
268 | assert(p); | |
269 | assert(p->path_malloc || p->path_const); | |
270 | assert(new_path); | |
271 | ||
272 | /* Saves current path in unprefixed_ variable, and takes over new_path */ | |
273 | ||
274 | free_and_replace(p->unprefixed_path_malloc, p->path_malloc); | |
275 | /* If we didn't have a path on the heap, then it's a static one */ | |
276 | if (!p->unprefixed_path_malloc) | |
277 | p->unprefixed_path_const = p->path_const; | |
278 | p->path_malloc = new_path; | |
279 | p->has_prefix = true; | |
280 | } | |
281 | ||
34de407a | 282 | static bool mount_entry_read_only(const MountEntry *p) { |
cfbeb4ef LP |
283 | assert(p); |
284 | ||
56a13a49 | 285 | return p->read_only || IN_SET(p->mode, READONLY, INACCESSIBLE, PRIVATE_TMP_READONLY); |
cfbeb4ef LP |
286 | } |
287 | ||
ddc155b2 TM |
288 | static bool mount_entry_noexec(const MountEntry *p) { |
289 | assert(p); | |
290 | ||
988bcc7c | 291 | return p->noexec || IN_SET(p->mode, NOEXEC, INACCESSIBLE, SYSFS, PROCFS); |
ddc155b2 TM |
292 | } |
293 | ||
294 | static bool mount_entry_exec(const MountEntry *p) { | |
295 | assert(p); | |
296 | ||
297 | return p->exec || p->mode == EXEC; | |
298 | } | |
299 | ||
d2d6c096 LP |
300 | static const char *mount_entry_source(const MountEntry *p) { |
301 | assert(p); | |
302 | ||
303 | return p->source_malloc ?: p->source_const; | |
304 | } | |
305 | ||
2abd4e38 YW |
306 | static const char *mount_entry_options(const MountEntry *p) { |
307 | assert(p); | |
308 | ||
309 | return p->options_malloc ?: p->options_const; | |
310 | } | |
311 | ||
1eb7e08e LP |
312 | static void mount_entry_done(MountEntry *p) { |
313 | assert(p); | |
314 | ||
315 | p->path_malloc = mfree(p->path_malloc); | |
809ceb82 | 316 | p->unprefixed_path_malloc = mfree(p->unprefixed_path_malloc); |
1eb7e08e | 317 | p->source_malloc = mfree(p->source_malloc); |
2abd4e38 | 318 | p->options_malloc = mfree(p->options_malloc); |
427353f6 | 319 | p->image_options = mount_options_free_all(p->image_options); |
1eb7e08e LP |
320 | } |
321 | ||
d18aff04 | 322 | static int append_access_mounts(MountEntry **p, char **strv, MountMode mode, bool forcibly_require_prefix) { |
613b411c LP |
323 | assert(p); |
324 | ||
1e05071d | 325 | /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */ |
5327c910 | 326 | |
15ae422b | 327 | STRV_FOREACH(i, strv) { |
5327c910 LP |
328 | bool ignore = false, needs_prefix = false; |
329 | const char *e = *i; | |
15ae422b | 330 | |
5327c910 LP |
331 | /* Look for any prefixes */ |
332 | if (startswith(e, "-")) { | |
333 | e++; | |
9c94d52e | 334 | ignore = true; |
ea92ae33 | 335 | } |
5327c910 LP |
336 | if (startswith(e, "+")) { |
337 | e++; | |
338 | needs_prefix = true; | |
339 | } | |
ea92ae33 | 340 | |
baaa35ad ZJS |
341 | if (!path_is_absolute(e)) |
342 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
343 | "Path is not absolute: %s", e); | |
15ae422b | 344 | |
34de407a | 345 | *((*p)++) = (MountEntry) { |
5327c910 LP |
346 | .path_const = e, |
347 | .mode = mode, | |
348 | .ignore = ignore, | |
d18aff04 | 349 | .has_prefix = !needs_prefix && !forcibly_require_prefix, |
5327c910 | 350 | }; |
15ae422b LP |
351 | } |
352 | ||
353 | return 0; | |
354 | } | |
355 | ||
6c47cd7d | 356 | static int append_empty_dir_mounts(MountEntry **p, char **strv) { |
6c47cd7d LP |
357 | assert(p); |
358 | ||
359 | /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the | |
360 | * "/private/" boundary directories for DynamicUser=1. */ | |
361 | ||
362 | STRV_FOREACH(i, strv) { | |
363 | ||
364 | *((*p)++) = (MountEntry) { | |
365 | .path_const = *i, | |
366 | .mode = EMPTY_DIR, | |
367 | .ignore = false, | |
6c47cd7d | 368 | .read_only = true, |
9f563f27 | 369 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, |
2abd4e38 | 370 | .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, |
6c47cd7d LP |
371 | }; |
372 | } | |
373 | ||
374 | return 0; | |
375 | } | |
376 | ||
da6053d0 | 377 | static int append_bind_mounts(MountEntry **p, const BindMount *binds, size_t n) { |
d2d6c096 LP |
378 | assert(p); |
379 | ||
fe96c0f8 | 380 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
381 | const BindMount *b = binds + i; |
382 | ||
383 | *((*p)++) = (MountEntry) { | |
384 | .path_const = b->destination, | |
385 | .mode = b->recursive ? BIND_MOUNT_RECURSIVE : BIND_MOUNT, | |
386 | .read_only = b->read_only, | |
9ce4e4b0 | 387 | .nosuid = b->nosuid, |
d2d6c096 | 388 | .source_const = b->source, |
4ca763a9 | 389 | .ignore = b->ignore_enoent, |
d2d6c096 LP |
390 | }; |
391 | } | |
392 | ||
393 | return 0; | |
394 | } | |
395 | ||
b3d13314 LB |
396 | static int append_mount_images(MountEntry **p, const MountImage *mount_images, size_t n) { |
397 | assert(p); | |
398 | ||
399 | for (size_t i = 0; i < n; i++) { | |
400 | const MountImage *m = mount_images + i; | |
401 | ||
402 | *((*p)++) = (MountEntry) { | |
403 | .path_const = m->destination, | |
404 | .mode = MOUNT_IMAGES, | |
405 | .source_const = m->source, | |
427353f6 | 406 | .image_options = m->mount_options, |
b3d13314 LB |
407 | .ignore = m->ignore_enoent, |
408 | }; | |
409 | } | |
410 | ||
411 | return 0; | |
412 | } | |
413 | ||
a07b9926 | 414 | static int append_extensions( |
93f59701 LB |
415 | MountEntry **p, |
416 | const char *root, | |
417 | const char *extension_dir, | |
418 | char **hierarchies, | |
419 | const MountImage *mount_images, | |
a07b9926 LB |
420 | size_t n, |
421 | char **extension_directories) { | |
93f59701 LB |
422 | |
423 | _cleanup_strv_free_ char **overlays = NULL; | |
93f59701 LB |
424 | int r; |
425 | ||
a07b9926 | 426 | if (n == 0 && strv_isempty(extension_directories)) |
93f59701 LB |
427 | return 0; |
428 | ||
24759d8f LB |
429 | assert(p); |
430 | assert(extension_dir); | |
431 | ||
93f59701 | 432 | /* Prepare a list of overlays, that will have as each element a string suitable for being |
7a7d2f16 | 433 | * passed as a lowerdir= parameter, so start with the hierarchy on the root. |
93f59701 LB |
434 | * The overlays vector will have the same number of elements and will correspond to the |
435 | * hierarchies vector, so they can be iterated upon together. */ | |
436 | STRV_FOREACH(hierarchy, hierarchies) { | |
437 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
438 | ||
439 | prefixed_hierarchy = path_join(root, *hierarchy); | |
440 | if (!prefixed_hierarchy) | |
441 | return -ENOMEM; | |
442 | ||
443 | r = strv_consume(&overlays, TAKE_PTR(prefixed_hierarchy)); | |
444 | if (r < 0) | |
445 | return r; | |
446 | } | |
447 | ||
448 | /* First, prepare a mount for each image, but these won't be visible to the unit, instead | |
449 | * they will be mounted in our propagate directory, and used as a source for the overlay. */ | |
450 | for (size_t i = 0; i < n; i++) { | |
451 | _cleanup_free_ char *mount_point = NULL; | |
452 | const MountImage *m = mount_images + i; | |
453 | ||
454 | r = asprintf(&mount_point, "%s/%zu", extension_dir, i); | |
455 | if (r < 0) | |
456 | return -ENOMEM; | |
457 | ||
458 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
459 | _cleanup_free_ char *prefixed_hierarchy = NULL, *escaped = NULL, *lowerdir = NULL; | |
460 | ||
461 | prefixed_hierarchy = path_join(mount_point, hierarchies[j]); | |
462 | if (!prefixed_hierarchy) | |
463 | return -ENOMEM; | |
464 | ||
465 | escaped = shell_escape(prefixed_hierarchy, ",:"); | |
466 | if (!escaped) | |
467 | return -ENOMEM; | |
468 | ||
469 | /* Note that lowerdir= parameters are in 'reverse' order, so the | |
470 | * top-most directory in the overlay comes first in the list. */ | |
471 | lowerdir = strjoin(escaped, ":", overlays[j]); | |
472 | if (!lowerdir) | |
473 | return -ENOMEM; | |
474 | ||
475 | free_and_replace(overlays[j], lowerdir); | |
476 | } | |
477 | ||
478 | *((*p)++) = (MountEntry) { | |
479 | .path_malloc = TAKE_PTR(mount_point), | |
480 | .image_options = m->mount_options, | |
481 | .ignore = m->ignore_enoent, | |
482 | .source_const = m->source, | |
483 | .mode = EXTENSION_IMAGES, | |
484 | .has_prefix = true, | |
485 | }; | |
486 | } | |
487 | ||
a07b9926 LB |
488 | /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory. |
489 | * Bind mount them in the same location as the ExtensionImages, so that we | |
490 | * can check that they are valid trees (extension-release.d). */ | |
491 | STRV_FOREACH(extension_directory, extension_directories) { | |
492 | _cleanup_free_ char *mount_point = NULL, *source = NULL; | |
493 | const char *e = *extension_directory; | |
494 | bool ignore_enoent = false; | |
495 | ||
496 | /* Pick up the counter where the ExtensionImages left it. */ | |
497 | r = asprintf(&mount_point, "%s/%zu", extension_dir, n++); | |
498 | if (r < 0) | |
499 | return -ENOMEM; | |
500 | ||
501 | /* Look for any prefixes */ | |
502 | if (startswith(e, "-")) { | |
503 | e++; | |
504 | ignore_enoent = true; | |
505 | } | |
506 | /* Ignore this for now */ | |
507 | if (startswith(e, "+")) | |
508 | e++; | |
509 | ||
510 | source = strdup(e); | |
511 | if (!source) | |
512 | return -ENOMEM; | |
513 | ||
514 | for (size_t j = 0; hierarchies && hierarchies[j]; ++j) { | |
515 | _cleanup_free_ char *prefixed_hierarchy = NULL, *escaped = NULL, *lowerdir = NULL; | |
516 | ||
517 | prefixed_hierarchy = path_join(mount_point, hierarchies[j]); | |
518 | if (!prefixed_hierarchy) | |
519 | return -ENOMEM; | |
520 | ||
521 | escaped = shell_escape(prefixed_hierarchy, ",:"); | |
522 | if (!escaped) | |
523 | return -ENOMEM; | |
524 | ||
525 | /* Note that lowerdir= parameters are in 'reverse' order, so the | |
526 | * top-most directory in the overlay comes first in the list. */ | |
527 | lowerdir = strjoin(escaped, ":", overlays[j]); | |
528 | if (!lowerdir) | |
529 | return -ENOMEM; | |
530 | ||
531 | free_and_replace(overlays[j], lowerdir); | |
532 | } | |
533 | ||
534 | *((*p)++) = (MountEntry) { | |
535 | .path_malloc = TAKE_PTR(mount_point), | |
536 | .source_const = TAKE_PTR(source), | |
537 | .mode = EXTENSION_DIRECTORIES, | |
538 | .ignore = ignore_enoent, | |
539 | .has_prefix = true, | |
540 | .read_only = true, | |
541 | }; | |
542 | } | |
543 | ||
93f59701 LB |
544 | /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings |
545 | * set up earlier. */ | |
546 | for (size_t i = 0; hierarchies && hierarchies[i]; ++i) { | |
547 | _cleanup_free_ char *prefixed_hierarchy = NULL; | |
548 | ||
549 | prefixed_hierarchy = path_join(root, hierarchies[i]); | |
550 | if (!prefixed_hierarchy) | |
551 | return -ENOMEM; | |
552 | ||
553 | *((*p)++) = (MountEntry) { | |
554 | .path_malloc = TAKE_PTR(prefixed_hierarchy), | |
555 | .options_malloc = TAKE_PTR(overlays[i]), | |
556 | .mode = OVERLAY_MOUNT, | |
557 | .has_prefix = true, | |
558 | .ignore = true, /* If the source image doesn't set the ignore bit it will fail earlier. */ | |
559 | }; | |
560 | } | |
561 | ||
562 | return 0; | |
563 | } | |
564 | ||
da6053d0 | 565 | static int append_tmpfs_mounts(MountEntry **p, const TemporaryFileSystem *tmpfs, size_t n) { |
2abd4e38 YW |
566 | assert(p); |
567 | ||
b67ec8e5 | 568 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
569 | const TemporaryFileSystem *t = tmpfs + i; |
570 | _cleanup_free_ char *o = NULL, *str = NULL; | |
ad8e66dc | 571 | unsigned long flags; |
2abd4e38 | 572 | bool ro = false; |
b67ec8e5 | 573 | int r; |
2abd4e38 | 574 | |
baaa35ad ZJS |
575 | if (!path_is_absolute(t->path)) |
576 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
577 | "Path is not absolute: %s", | |
578 | t->path); | |
2abd4e38 | 579 | |
b67ec8e5 | 580 | str = strjoin("mode=0755" NESTED_TMPFS_LIMITS ",", t->options); |
ad8e66dc AJ |
581 | if (!str) |
582 | return -ENOMEM; | |
2abd4e38 | 583 | |
ad8e66dc AJ |
584 | r = mount_option_mangle(str, MS_NODEV|MS_STRICTATIME, &flags, &o); |
585 | if (r < 0) | |
586 | return log_debug_errno(r, "Failed to parse mount option '%s': %m", str); | |
2abd4e38 | 587 | |
ad8e66dc AJ |
588 | ro = flags & MS_RDONLY; |
589 | if (ro) | |
590 | flags ^= MS_RDONLY; | |
2abd4e38 YW |
591 | |
592 | *((*p)++) = (MountEntry) { | |
593 | .path_const = t->path, | |
594 | .mode = TMPFS, | |
595 | .read_only = ro, | |
ad8e66dc | 596 | .options_malloc = TAKE_PTR(o), |
2abd4e38 YW |
597 | .flags = flags, |
598 | }; | |
2abd4e38 YW |
599 | } |
600 | ||
601 | return 0; | |
602 | } | |
603 | ||
da6053d0 | 604 | static int append_static_mounts(MountEntry **p, const MountEntry *mounts, size_t n, bool ignore_protect) { |
11a30cec | 605 | assert(p); |
f471b2af | 606 | assert(mounts); |
11a30cec | 607 | |
5327c910 | 608 | /* Adds a list of static pre-defined entries */ |
f471b2af | 609 | |
fe96c0f8 | 610 | for (size_t i = 0; i < n; i++) |
34de407a LP |
611 | *((*p)++) = (MountEntry) { |
612 | .path_const = mount_entry_path(mounts+i), | |
5327c910 LP |
613 | .mode = mounts[i].mode, |
614 | .ignore = mounts[i].ignore || ignore_protect, | |
615 | }; | |
f471b2af DH |
616 | |
617 | return 0; | |
618 | } | |
619 | ||
34de407a | 620 | static int append_protect_home(MountEntry **p, ProtectHome protect_home, bool ignore_protect) { |
c575770b DH |
621 | assert(p); |
622 | ||
5327c910 | 623 | switch (protect_home) { |
b6c432ca | 624 | |
5327c910 | 625 | case PROTECT_HOME_NO: |
b6c432ca DH |
626 | return 0; |
627 | ||
b6c432ca | 628 | case PROTECT_HOME_READ_ONLY: |
5327c910 LP |
629 | return append_static_mounts(p, protect_home_read_only_table, ELEMENTSOF(protect_home_read_only_table), ignore_protect); |
630 | ||
e4da7d8c YW |
631 | case PROTECT_HOME_TMPFS: |
632 | return append_static_mounts(p, protect_home_tmpfs_table, ELEMENTSOF(protect_home_tmpfs_table), ignore_protect); | |
633 | ||
b6c432ca | 634 | case PROTECT_HOME_YES: |
5327c910 LP |
635 | return append_static_mounts(p, protect_home_yes_table, ELEMENTSOF(protect_home_yes_table), ignore_protect); |
636 | ||
b6c432ca | 637 | default: |
04499a70 | 638 | assert_not_reached(); |
b6c432ca | 639 | } |
b6c432ca DH |
640 | } |
641 | ||
34de407a | 642 | static int append_protect_system(MountEntry **p, ProtectSystem protect_system, bool ignore_protect) { |
f471b2af DH |
643 | assert(p); |
644 | ||
5327c910 LP |
645 | switch (protect_system) { |
646 | ||
647 | case PROTECT_SYSTEM_NO: | |
f471b2af DH |
648 | return 0; |
649 | ||
f471b2af | 650 | case PROTECT_SYSTEM_STRICT: |
5327c910 LP |
651 | return append_static_mounts(p, protect_system_strict_table, ELEMENTSOF(protect_system_strict_table), ignore_protect); |
652 | ||
f471b2af | 653 | case PROTECT_SYSTEM_YES: |
5327c910 LP |
654 | return append_static_mounts(p, protect_system_yes_table, ELEMENTSOF(protect_system_yes_table), ignore_protect); |
655 | ||
f471b2af | 656 | case PROTECT_SYSTEM_FULL: |
5327c910 LP |
657 | return append_static_mounts(p, protect_system_full_table, ELEMENTSOF(protect_system_full_table), ignore_protect); |
658 | ||
f471b2af | 659 | default: |
04499a70 | 660 | assert_not_reached(); |
f471b2af | 661 | } |
11a30cec DH |
662 | } |
663 | ||
93bab288 | 664 | static int mount_path_compare(const MountEntry *a, const MountEntry *b) { |
a0827e2b | 665 | int d; |
15ae422b | 666 | |
a07b9926 | 667 | /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first |
93f59701 LB |
668 | * regardless of the prefix - they are set up in the propagate directory anyway */ |
669 | d = -CMP(a->mode == EXTENSION_IMAGES, b->mode == EXTENSION_IMAGES); | |
a07b9926 LB |
670 | if (d != 0) |
671 | return d; | |
672 | d = -CMP(a->mode == EXTENSION_DIRECTORIES, b->mode == EXTENSION_DIRECTORIES); | |
93f59701 LB |
673 | if (d != 0) |
674 | return d; | |
675 | ||
6ee1a919 | 676 | /* If the paths are not equal, then order prefixes first */ |
93bab288 | 677 | d = path_compare(mount_entry_path(a), mount_entry_path(b)); |
6ee1a919 LP |
678 | if (d != 0) |
679 | return d; | |
15ae422b | 680 | |
6ee1a919 | 681 | /* If the paths are equal, check the mode */ |
93bab288 | 682 | return CMP((int) a->mode, (int) b->mode); |
15ae422b LP |
683 | } |
684 | ||
da6053d0 | 685 | static int prefix_where_needed(MountEntry *m, size_t n, const char *root_directory) { |
4a756839 | 686 | /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */ |
5327c910 | 687 | |
fe96c0f8 SS |
688 | assert(m || n == 0); |
689 | ||
690 | for (size_t i = 0; i < n; i++) { | |
5327c910 LP |
691 | char *s; |
692 | ||
693 | if (m[i].has_prefix) | |
694 | continue; | |
695 | ||
c6134d3e | 696 | s = path_join(root_directory, mount_entry_path(m+i)); |
5327c910 LP |
697 | if (!s) |
698 | return -ENOMEM; | |
699 | ||
809ceb82 | 700 | mount_entry_consume_prefix(&m[i], s); |
5327c910 LP |
701 | } |
702 | ||
703 | return 0; | |
704 | } | |
705 | ||
da6053d0 | 706 | static void drop_duplicates(MountEntry *m, size_t *n) { |
34de407a | 707 | MountEntry *f, *t, *previous; |
15ae422b | 708 | |
c17ec25e | 709 | assert(m); |
15ae422b | 710 | assert(n); |
15ae422b | 711 | |
fe3c2583 LP |
712 | /* Drops duplicate entries. Expects that the array is properly ordered already. */ |
713 | ||
1d54cd5d | 714 | for (f = m, t = m, previous = NULL; f < m + *n; f++) { |
15ae422b | 715 | |
fe3c2583 | 716 | /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare() |
088696fe LP |
717 | * above. Note that we only drop duplicates that haven't been mounted yet. */ |
718 | if (previous && | |
719 | path_equal(mount_entry_path(f), mount_entry_path(previous)) && | |
720 | !f->applied && !previous->applied) { | |
5beb8688 | 721 | log_debug("%s (%s) is duplicate.", mount_entry_path(f), mount_mode_to_string(f->mode)); |
ddc155b2 TM |
722 | /* Propagate the flags to the remaining entry */ |
723 | previous->read_only = previous->read_only || mount_entry_read_only(f); | |
724 | previous->noexec = previous->noexec || mount_entry_noexec(f); | |
725 | previous->exec = previous->exec || mount_entry_exec(f); | |
1eb7e08e | 726 | mount_entry_done(f); |
15ae422b | 727 | continue; |
fe3c2583 | 728 | } |
15ae422b | 729 | |
e2d7c1a0 | 730 | *t = *f; |
15ae422b | 731 | previous = t; |
fe3c2583 LP |
732 | t++; |
733 | } | |
734 | ||
735 | *n = t - m; | |
736 | } | |
737 | ||
da6053d0 | 738 | static void drop_inaccessible(MountEntry *m, size_t *n) { |
34de407a | 739 | MountEntry *f, *t; |
fe3c2583 LP |
740 | const char *clear = NULL; |
741 | ||
742 | assert(m); | |
743 | assert(n); | |
744 | ||
745 | /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly | |
746 | * ordered already. */ | |
747 | ||
1d54cd5d | 748 | for (f = m, t = m; f < m + *n; f++) { |
fe3c2583 LP |
749 | |
750 | /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop | |
751 | * it, as inaccessible paths really should drop the entire subtree. */ | |
34de407a LP |
752 | if (clear && path_startswith(mount_entry_path(f), clear)) { |
753 | log_debug("%s is masked by %s.", mount_entry_path(f), clear); | |
1eb7e08e | 754 | mount_entry_done(f); |
fe3c2583 LP |
755 | continue; |
756 | } | |
15ae422b | 757 | |
34de407a | 758 | clear = f->mode == INACCESSIBLE ? mount_entry_path(f) : NULL; |
fe3c2583 LP |
759 | |
760 | *t = *f; | |
15ae422b LP |
761 | t++; |
762 | } | |
763 | ||
c17ec25e | 764 | *n = t - m; |
15ae422b LP |
765 | } |
766 | ||
da6053d0 | 767 | static void drop_nop(MountEntry *m, size_t *n) { |
34de407a | 768 | MountEntry *f, *t; |
7648a565 LP |
769 | |
770 | assert(m); | |
771 | assert(n); | |
772 | ||
773 | /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the | |
774 | * list is ordered by prefixes. */ | |
775 | ||
1d54cd5d | 776 | for (f = m, t = m; f < m + *n; f++) { |
7648a565 | 777 | |
1e05071d YW |
778 | /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */ |
779 | if (IN_SET(f->mode, READONLY, READWRITE, READWRITE_IMPLICIT)) { | |
e7bf2fca | 780 | MountEntry *found = NULL; |
7648a565 LP |
781 | |
782 | /* Now let's find the first parent of the entry we are looking at. */ | |
e7bf2fca | 783 | for (MountEntry *p = PTR_SUB1(t, m); p; p = PTR_SUB1(p, m)) |
34de407a | 784 | if (path_startswith(mount_entry_path(f), mount_entry_path(p))) { |
e7bf2fca | 785 | found = p; |
7648a565 LP |
786 | break; |
787 | } | |
7648a565 LP |
788 | |
789 | /* We found it, let's see if it's the same mode, if so, we can drop this entry */ | |
e7bf2fca | 790 | if (found && found->mode == f->mode) { |
5beb8688 YW |
791 | log_debug("%s (%s) is made redundant by %s (%s)", |
792 | mount_entry_path(f), mount_mode_to_string(f->mode), | |
e7bf2fca | 793 | mount_entry_path(found), mount_mode_to_string(found->mode)); |
1eb7e08e | 794 | mount_entry_done(f); |
7648a565 LP |
795 | continue; |
796 | } | |
797 | } | |
798 | ||
799 | *t = *f; | |
800 | t++; | |
801 | } | |
802 | ||
803 | *n = t - m; | |
804 | } | |
805 | ||
da6053d0 | 806 | static void drop_outside_root(const char *root_directory, MountEntry *m, size_t *n) { |
34de407a | 807 | MountEntry *f, *t; |
cd2902c9 LP |
808 | |
809 | assert(m); | |
810 | assert(n); | |
811 | ||
1d54cd5d | 812 | /* Nothing to do */ |
cd2902c9 LP |
813 | if (!root_directory) |
814 | return; | |
815 | ||
816 | /* Drops all mounts that are outside of the root directory. */ | |
817 | ||
1d54cd5d | 818 | for (f = m, t = m; f < m + *n; f++) { |
cd2902c9 | 819 | |
a07b9926 LB |
820 | /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */ |
821 | if (!IN_SET(f->mode, EXTENSION_IMAGES, EXTENSION_DIRECTORIES) && !path_startswith(mount_entry_path(f), root_directory)) { | |
34de407a | 822 | log_debug("%s is outside of root directory.", mount_entry_path(f)); |
1eb7e08e | 823 | mount_entry_done(f); |
cd2902c9 LP |
824 | continue; |
825 | } | |
826 | ||
827 | *t = *f; | |
828 | t++; | |
829 | } | |
830 | ||
831 | *n = t - m; | |
832 | } | |
833 | ||
b2a60844 LP |
834 | static int clone_device_node( |
835 | const char *d, | |
836 | const char *temporary_mount, | |
837 | bool *make_devnode) { | |
838 | ||
839 | _cleanup_free_ char *sl = NULL; | |
840 | const char *dn, *bn, *t; | |
b5e99f23 ДГ |
841 | struct stat st; |
842 | int r; | |
843 | ||
414b304b | 844 | if (stat(d, &st) < 0) { |
b2a60844 LP |
845 | if (errno == ENOENT) { |
846 | log_debug_errno(errno, "Device node '%s' to clone does not exist, ignoring.", d); | |
af984e13 | 847 | return -ENXIO; |
b2a60844 LP |
848 | } |
849 | ||
850 | return log_debug_errno(errno, "Failed to stat() device node '%s' to clone, ignoring: %m", d); | |
b5e99f23 ДГ |
851 | } |
852 | ||
853 | if (!S_ISBLK(st.st_mode) && | |
baaa35ad ZJS |
854 | !S_ISCHR(st.st_mode)) |
855 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
856 | "Device node '%s' to clone is not a device node, ignoring.", | |
857 | d); | |
b5e99f23 | 858 | |
6f7f3a33 | 859 | dn = strjoina(temporary_mount, d); |
b5e99f23 | 860 | |
b2a60844 | 861 | /* First, try to create device node properly */ |
16498617 CB |
862 | if (*make_devnode) { |
863 | mac_selinux_create_file_prepare(d, st.st_mode); | |
864 | r = mknod(dn, st.st_mode, st.st_rdev); | |
865 | mac_selinux_create_file_clear(); | |
b2a60844 LP |
866 | if (r >= 0) |
867 | goto add_symlink; | |
16498617 CB |
868 | if (errno != EPERM) |
869 | return log_debug_errno(errno, "mknod failed for %s: %m", d); | |
870 | ||
b2a60844 | 871 | /* This didn't work, let's not try this again for the next iterations. */ |
16498617 CB |
872 | *make_devnode = false; |
873 | } | |
874 | ||
d73020f2 | 875 | /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target. |
1acf344d | 876 | * Do not prepare device-node SELinux label (see issue 13762) */ |
16498617 | 877 | r = mknod(dn, S_IFREG, 0); |
16498617 | 878 | if (r < 0 && errno != EEXIST) |
b2a60844 | 879 | return log_debug_errno(errno, "mknod() fallback failed for '%s': %m", d); |
16498617 | 880 | |
21935150 LP |
881 | /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard |
882 | * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or | |
883 | * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */ | |
884 | r = mount_nofollow_verbose(LOG_DEBUG, d, dn, NULL, MS_BIND, NULL); | |
885 | if (r < 0) | |
886 | return r; | |
b2a60844 LP |
887 | |
888 | add_symlink: | |
889 | bn = path_startswith(d, "/dev/"); | |
890 | if (!bn) | |
891 | return 0; | |
892 | ||
893 | /* Create symlinks like /dev/char/1:9 → ../urandom */ | |
ec61371f | 894 | if (asprintf(&sl, "%s/dev/%s/" DEVNUM_FORMAT_STR, |
cbc056c8 ZJS |
895 | temporary_mount, |
896 | S_ISCHR(st.st_mode) ? "char" : "block", | |
ec61371f | 897 | DEVNUM_FORMAT_VAL(st.st_rdev)) < 0) |
b2a60844 LP |
898 | return log_oom(); |
899 | ||
900 | (void) mkdir_parents(sl, 0755); | |
901 | ||
902 | t = strjoina("../", bn); | |
b2a60844 | 903 | if (symlink(t, sl) < 0) |
2e4a4fae | 904 | log_debug_errno(errno, "Failed to symlink '%s' to '%s', ignoring: %m", t, sl); |
b5e99f23 | 905 | |
af984e13 | 906 | return 0; |
b5e99f23 ДГ |
907 | } |
908 | ||
5d997827 | 909 | static int mount_private_dev(MountEntry *m) { |
7f112f50 LP |
910 | static const char devnodes[] = |
911 | "/dev/null\0" | |
912 | "/dev/zero\0" | |
913 | "/dev/full\0" | |
914 | "/dev/random\0" | |
915 | "/dev/urandom\0" | |
916 | "/dev/tty\0"; | |
917 | ||
2b85f4e1 | 918 | char temporary_mount[] = "/tmp/namespace-dev-XXXXXX"; |
12e2b70f | 919 | const char *dev = NULL, *devpts = NULL, *devshm = NULL, *devhugepages = NULL, *devmqueue = NULL, *devlog = NULL, *devptmx = NULL; |
16498617 | 920 | bool can_mknod = true; |
7f112f50 LP |
921 | int r; |
922 | ||
923 | assert(m); | |
924 | ||
2b85f4e1 | 925 | if (!mkdtemp(temporary_mount)) |
2e4a4fae | 926 | return log_debug_errno(errno, "Failed to create temporary directory '%s': %m", temporary_mount); |
2b85f4e1 | 927 | |
63c372cb | 928 | dev = strjoina(temporary_mount, "/dev"); |
dc751688 | 929 | (void) mkdir(dev, 0755); |
9f563f27 | 930 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", dev, "tmpfs", DEV_MOUNT_OPTIONS, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV); |
21935150 | 931 | if (r < 0) |
2b85f4e1 | 932 | goto fail; |
21935150 | 933 | |
03bc11d1 | 934 | r = label_fix_full(AT_FDCWD, dev, "/dev", 0); |
c3151977 | 935 | if (r < 0) { |
1ce268c7 | 936 | log_debug_errno(r, "Failed to fix label of '%s' as /dev: %m", dev); |
c3151977 TM |
937 | goto fail; |
938 | } | |
2b85f4e1 | 939 | |
63c372cb | 940 | devpts = strjoina(temporary_mount, "/dev/pts"); |
dc751688 | 941 | (void) mkdir(devpts, 0755); |
21935150 LP |
942 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/pts", devpts, NULL, MS_BIND, NULL); |
943 | if (r < 0) | |
2b85f4e1 | 944 | goto fail; |
2b85f4e1 | 945 | |
2e4a4fae YW |
946 | /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx. |
947 | * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible. | |
948 | * Thus, in that case make a clone. | |
949 | * In nspawn and other containers it will be a symlink, in that case make it a symlink. */ | |
36ce7110 | 950 | r = is_symlink("/dev/ptmx"); |
2e4a4fae YW |
951 | if (r < 0) { |
952 | log_debug_errno(r, "Failed to detect whether /dev/ptmx is a symlink or not: %m"); | |
3164e3cb | 953 | goto fail; |
2e4a4fae | 954 | } else if (r > 0) { |
414b304b ДГ |
955 | devptmx = strjoina(temporary_mount, "/dev/ptmx"); |
956 | if (symlink("pts/ptmx", devptmx) < 0) { | |
2e4a4fae | 957 | r = log_debug_errno(errno, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx); |
414b304b ДГ |
958 | goto fail; |
959 | } | |
960 | } else { | |
16498617 | 961 | r = clone_device_node("/dev/ptmx", temporary_mount, &can_mknod); |
152c475f LP |
962 | if (r < 0) |
963 | goto fail; | |
414b304b | 964 | } |
e06b6479 | 965 | |
63c372cb | 966 | devshm = strjoina(temporary_mount, "/dev/shm"); |
8d953682 | 967 | (void) mkdir(devshm, 0755); |
21935150 LP |
968 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev/shm", devshm, NULL, MS_BIND, NULL); |
969 | if (r < 0) | |
2b85f4e1 | 970 | goto fail; |
2b85f4e1 | 971 | |
63c372cb | 972 | devmqueue = strjoina(temporary_mount, "/dev/mqueue"); |
dc751688 | 973 | (void) mkdir(devmqueue, 0755); |
21935150 | 974 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/mqueue", devmqueue, NULL, MS_BIND, NULL); |
2b85f4e1 | 975 | |
63c372cb | 976 | devhugepages = strjoina(temporary_mount, "/dev/hugepages"); |
dc751688 | 977 | (void) mkdir(devhugepages, 0755); |
21935150 | 978 | (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/hugepages", devhugepages, NULL, MS_BIND, NULL); |
2b85f4e1 | 979 | |
63c372cb | 980 | devlog = strjoina(temporary_mount, "/dev/log"); |
2e4a4fae YW |
981 | if (symlink("/run/systemd/journal/dev-log", devlog) < 0) |
982 | log_debug_errno(errno, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog); | |
82d25240 | 983 | |
7f112f50 | 984 | NULSTR_FOREACH(d, devnodes) { |
16498617 | 985 | r = clone_device_node(d, temporary_mount, &can_mknod); |
37b22b3b | 986 | /* ENXIO means the *source* is not a device file, skip creation in that case */ |
af984e13 | 987 | if (r < 0 && r != -ENXIO) |
2b85f4e1 | 988 | goto fail; |
7f112f50 LP |
989 | } |
990 | ||
2e4a4fae YW |
991 | r = dev_setup(temporary_mount, UID_INVALID, GID_INVALID); |
992 | if (r < 0) | |
105a1a36 | 993 | log_debug_errno(r, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount); |
7f112f50 | 994 | |
d73020f2 LP |
995 | /* Create the /dev directory if missing. It is more likely to be missing when the service is started |
996 | * with RootDirectory. This is consistent with mount units creating the mount points when missing. */ | |
34de407a | 997 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
ee818b89 | 998 | |
9e5f8252 | 999 | /* Unmount everything in old /dev */ |
2e4a4fae YW |
1000 | r = umount_recursive(mount_entry_path(m), 0); |
1001 | if (r < 0) | |
1002 | log_debug_errno(r, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m)); | |
1003 | ||
21935150 LP |
1004 | r = mount_nofollow_verbose(LOG_DEBUG, dev, mount_entry_path(m), NULL, MS_MOVE, NULL); |
1005 | if (r < 0) | |
2b85f4e1 | 1006 | goto fail; |
7f112f50 | 1007 | |
1019a48f LP |
1008 | (void) rmdir(dev); |
1009 | (void) rmdir(temporary_mount); | |
7f112f50 | 1010 | |
2b85f4e1 | 1011 | return 0; |
7f112f50 | 1012 | |
2b85f4e1 LP |
1013 | fail: |
1014 | if (devpts) | |
21935150 | 1015 | (void) umount_verbose(LOG_DEBUG, devpts, UMOUNT_NOFOLLOW); |
7f112f50 | 1016 | |
2b85f4e1 | 1017 | if (devshm) |
21935150 | 1018 | (void) umount_verbose(LOG_DEBUG, devshm, UMOUNT_NOFOLLOW); |
7f112f50 | 1019 | |
2b85f4e1 | 1020 | if (devhugepages) |
21935150 | 1021 | (void) umount_verbose(LOG_DEBUG, devhugepages, UMOUNT_NOFOLLOW); |
7f112f50 | 1022 | |
2b85f4e1 | 1023 | if (devmqueue) |
21935150 | 1024 | (void) umount_verbose(LOG_DEBUG, devmqueue, UMOUNT_NOFOLLOW); |
7f112f50 | 1025 | |
21935150 | 1026 | (void) umount_verbose(LOG_DEBUG, dev, UMOUNT_NOFOLLOW); |
1019a48f LP |
1027 | (void) rmdir(dev); |
1028 | (void) rmdir(temporary_mount); | |
7f112f50 | 1029 | |
2b85f4e1 | 1030 | return r; |
7f112f50 LP |
1031 | } |
1032 | ||
2a2969fd | 1033 | static int mount_bind_dev(const MountEntry *m) { |
5d997827 LP |
1034 | int r; |
1035 | ||
1036 | assert(m); | |
1037 | ||
d73020f2 LP |
1038 | /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the |
1039 | * service's /dev. This is only used when RootDirectory= is set. */ | |
5d997827 | 1040 | |
645767d6 LP |
1041 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1042 | ||
5d997827 LP |
1043 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1044 | if (r < 0) | |
1045 | return log_debug_errno(r, "Unable to determine whether /dev is already mounted: %m"); | |
1046 | if (r > 0) /* make this a NOP if /dev is already a mount point */ | |
1047 | return 0; | |
1048 | ||
21935150 LP |
1049 | r = mount_nofollow_verbose(LOG_DEBUG, "/dev", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
1050 | if (r < 0) | |
1051 | return r; | |
5d997827 LP |
1052 | |
1053 | return 1; | |
1054 | } | |
1055 | ||
2a2969fd | 1056 | static int mount_sysfs(const MountEntry *m) { |
5d997827 LP |
1057 | int r; |
1058 | ||
1059 | assert(m); | |
1060 | ||
645767d6 LP |
1061 | (void) mkdir_p_label(mount_entry_path(m), 0755); |
1062 | ||
5d997827 LP |
1063 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); |
1064 | if (r < 0) | |
1065 | return log_debug_errno(r, "Unable to determine whether /sys is already mounted: %m"); | |
1066 | if (r > 0) /* make this a NOP if /sys is already a mount point */ | |
1067 | return 0; | |
1068 | ||
1069 | /* Bind mount the host's version so that we get all child mounts of it, too. */ | |
21935150 LP |
1070 | r = mount_nofollow_verbose(LOG_DEBUG, "/sys", mount_entry_path(m), NULL, MS_BIND|MS_REC, NULL); |
1071 | if (r < 0) | |
1072 | return r; | |
5d997827 LP |
1073 | |
1074 | return 1; | |
1075 | } | |
1076 | ||
1c265fcd | 1077 | static bool mount_option_supported(const char *fstype, const char *key, const char *value) { |
254d1313 | 1078 | _cleanup_close_ int fd = -EBADF; |
1c265fcd DDM |
1079 | int r; |
1080 | ||
1081 | /* This function assumes support by default. Only if the fsconfig() call fails with -EINVAL/-EOPNOTSUPP | |
1082 | * will it report that the option/value is not supported. */ | |
1083 | ||
1084 | fd = fsopen(fstype, FSOPEN_CLOEXEC); | |
1085 | if (fd < 0) { | |
1086 | if (errno != ENOSYS) | |
1087 | log_debug_errno(errno, "Failed to open superblock context for '%s': %m", fstype); | |
1088 | return true; /* If fsopen() fails for whatever reason, assume the value is supported. */ | |
1089 | } | |
1090 | ||
1091 | r = fsconfig(fd, FSCONFIG_SET_STRING, key, value, 0); | |
1092 | if (r < 0 && !IN_SET(errno, EINVAL, EOPNOTSUPP, ENOSYS)) | |
1093 | log_debug_errno(errno, "Failed to set '%s=%s' on '%s' superblock context: %m", key, value, fstype); | |
1094 | ||
1095 | return r >= 0 || !IN_SET(errno, EINVAL, EOPNOTSUPP); | |
1096 | } | |
1097 | ||
4e399953 | 1098 | static int mount_procfs(const MountEntry *m, const NamespaceInfo *ns_info) { |
61f8a7bd | 1099 | _cleanup_free_ char *opts = NULL; |
4e399953 | 1100 | const char *entry_path; |
61f8a7bd | 1101 | int r, n; |
5d997827 LP |
1102 | |
1103 | assert(m); | |
4e399953 | 1104 | assert(ns_info); |
5d997827 | 1105 | |
4e399953 LP |
1106 | if (ns_info->protect_proc != PROTECT_PROC_DEFAULT || |
1107 | ns_info->proc_subset != PROC_SUBSET_ALL) { | |
4e399953 LP |
1108 | |
1109 | /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it | |
1110 | * pretended to be per-instance but actually was per-namespace), hence let's make use of it | |
1111 | * if requested. To make sure this logic succeeds only on kernels where hidepid= is | |
1112 | * per-instance, we'll exclusively use the textual value for hidepid=, since support was | |
1113 | * added in the same commit: if it's supported it is thus also per-instance. */ | |
1114 | ||
1c265fcd | 1115 | const char *hpv = ns_info->protect_proc == PROTECT_PROC_DEFAULT ? |
7c76e181 ZJS |
1116 | "off" : |
1117 | protect_proc_to_string(ns_info->protect_proc); | |
1c265fcd DDM |
1118 | |
1119 | /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in | |
1120 | * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when | |
1121 | * trying to use hidepid= on systems where it isn't supported. The same applies for subset=. | |
1122 | * fsopen()/fsconfig() was also backported on some distros which allows us to detect | |
1123 | * hidepid=/subset= support in even more scenarios. */ | |
1124 | ||
1125 | if (mount_option_supported("proc", "hidepid", hpv)) { | |
1126 | opts = strjoin("hidepid=", hpv); | |
1127 | if (!opts) | |
1128 | return -ENOMEM; | |
1129 | } | |
1130 | ||
1131 | if (ns_info->proc_subset == PROC_SUBSET_PID && mount_option_supported("proc", "subset", "pid")) | |
1132 | if (!strextend_with_separator(&opts, ",", "subset=pid")) | |
1133 | return -ENOMEM; | |
4e399953 LP |
1134 | } |
1135 | ||
61f8a7bd YW |
1136 | entry_path = mount_entry_path(m); |
1137 | (void) mkdir_p_label(entry_path, 0755); | |
1138 | ||
1139 | /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in | |
1140 | * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by | |
1141 | * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything | |
1142 | * mounted on /proc/ first. */ | |
1143 | ||
1144 | n = umount_recursive(entry_path, 0); | |
1145 | ||
1146 | r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, opts); | |
1147 | if (r == -EINVAL && opts) | |
1148 | /* If this failed with EINVAL then this likely means the textual hidepid= stuff is | |
1149 | * not supported by the kernel, and thus the per-instance hidepid= neither, which | |
1150 | * means we really don't want to use it, since it would affect our host's /proc | |
1151 | * mount. Hence let's gracefully fallback to a classic, unrestricted version. */ | |
1152 | r = mount_nofollow_verbose(LOG_DEBUG, "proc", entry_path, "proc", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL); | |
1153 | if (r == -EPERM) { | |
24ee0f9d | 1154 | /* When we do not have enough privileges to mount /proc, fallback to use existing /proc. */ |
61f8a7bd YW |
1155 | |
1156 | if (n > 0) | |
1157 | /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree. | |
1158 | * Propagate the original error code returned by mount() in the above. */ | |
1159 | return -EPERM; | |
1160 | ||
1161 | r = path_is_mount_point(entry_path, NULL, 0); | |
1162 | if (r < 0) | |
1163 | return log_debug_errno(r, "Unable to determine whether /proc is already mounted: %m"); | |
ea63a260 LB |
1164 | if (r == 0) { |
1165 | /* We lack permissions to mount a new instance of /proc, and it is not already | |
1166 | * mounted. But we can access the host's, so as a final fallback bind-mount it to | |
1167 | * the destination, as most likely we are inside a user manager in an unprivileged | |
1168 | * user namespace. */ | |
1169 | r = mount_nofollow_verbose(LOG_DEBUG, "/proc", entry_path, NULL, MS_BIND|MS_REC, NULL); | |
1170 | if (r < 0) | |
1171 | return -EPERM; | |
1172 | } | |
ad74f28a YW |
1173 | } else if (r < 0) |
1174 | return r; | |
5d997827 LP |
1175 | |
1176 | return 1; | |
1177 | } | |
1178 | ||
2abd4e38 | 1179 | static int mount_tmpfs(const MountEntry *m) { |
df6b900a | 1180 | const char *entry_path, *inner_path; |
abad72be | 1181 | int r; |
abad72be | 1182 | |
6c47cd7d LP |
1183 | assert(m); |
1184 | ||
df6b900a | 1185 | entry_path = mount_entry_path(m); |
809ceb82 | 1186 | inner_path = mount_entry_unprefixed_path(m); |
df6b900a | 1187 | |
d73020f2 LP |
1188 | /* First, get rid of everything that is below if there is anything. Then, overmount with our new |
1189 | * tmpfs */ | |
6c47cd7d | 1190 | |
abad72be CG |
1191 | (void) mkdir_p_label(entry_path, 0755); |
1192 | (void) umount_recursive(entry_path, 0); | |
6c47cd7d | 1193 | |
21935150 LP |
1194 | r = mount_nofollow_verbose(LOG_DEBUG, "tmpfs", entry_path, "tmpfs", m->flags, mount_entry_options(m)); |
1195 | if (r < 0) | |
1196 | return r; | |
abad72be | 1197 | |
03bc11d1 | 1198 | r = label_fix_full(AT_FDCWD, entry_path, inner_path, 0); |
abad72be | 1199 | if (r < 0) |
df6b900a | 1200 | return log_debug_errno(r, "Failed to fix label of '%s' as '%s': %m", entry_path, inner_path); |
6c47cd7d LP |
1201 | |
1202 | return 1; | |
1203 | } | |
1204 | ||
94293d65 LB |
1205 | static int mount_run(const MountEntry *m) { |
1206 | int r; | |
1207 | ||
1208 | assert(m); | |
1209 | ||
1210 | r = path_is_mount_point(mount_entry_path(m), NULL, 0); | |
1211 | if (r < 0 && r != -ENOENT) | |
1212 | return log_debug_errno(r, "Unable to determine whether /run is already mounted: %m"); | |
1213 | if (r > 0) /* make this a NOP if /run is already a mount point */ | |
1214 | return 0; | |
1215 | ||
1216 | return mount_tmpfs(m); | |
1217 | } | |
1218 | ||
80271a44 XR |
1219 | static int mount_mqueuefs(const MountEntry *m) { |
1220 | int r; | |
1221 | const char *entry_path; | |
1222 | ||
1223 | assert(m); | |
1224 | ||
1225 | entry_path = mount_entry_path(m); | |
1226 | ||
1227 | (void) mkdir_p_label(entry_path, 0755); | |
1228 | (void) umount_recursive(entry_path, 0); | |
1229 | ||
1230 | r = mount_nofollow_verbose(LOG_DEBUG, "mqueue", entry_path, "mqueue", m->flags, mount_entry_options(m)); | |
1231 | if (r < 0) | |
1232 | return r; | |
1233 | ||
1234 | return 0; | |
1235 | } | |
1236 | ||
93f59701 LB |
1237 | static int mount_image(const MountEntry *m, const char *root_directory) { |
1238 | ||
1239 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
1240 | *host_os_release_sysext_level = NULL; | |
b3d13314 LB |
1241 | int r; |
1242 | ||
89e62e0b LP |
1243 | assert(m); |
1244 | ||
93f59701 LB |
1245 | if (m->mode == EXTENSION_IMAGES) { |
1246 | r = parse_os_release( | |
1247 | empty_to_root(root_directory), | |
1248 | "ID", &host_os_release_id, | |
1249 | "VERSION_ID", &host_os_release_version_id, | |
1250 | "SYSEXT_LEVEL", &host_os_release_sysext_level, | |
1251 | NULL); | |
1252 | if (r < 0) | |
1253 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
78ab2b50 LB |
1254 | if (isempty(host_os_release_id)) |
1255 | 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 |
1256 | } |
1257 | ||
1258 | r = verity_dissect_and_mount( | |
7c76e181 ZJS |
1259 | /* src_fd= */ -1, mount_entry_source(m), mount_entry_path(m), m->image_options, |
1260 | host_os_release_id, host_os_release_version_id, host_os_release_sysext_level, NULL); | |
b850a9b2 LB |
1261 | if (r == -ENOENT && m->ignore) |
1262 | return 0; | |
db4c8a25 LB |
1263 | if (r == -ESTALE && host_os_release_id) |
1264 | return log_error_errno(r, | |
1265 | "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s", | |
1266 | mount_entry_source(m), | |
1267 | host_os_release_id, | |
1268 | host_os_release_version_id ? " VERSION_ID=" : "", | |
1269 | strempty(host_os_release_version_id), | |
1270 | host_os_release_sysext_level ? " SYSEXT_LEVEL=" : "", | |
1271 | strempty(host_os_release_sysext_level)); | |
b3d13314 | 1272 | if (r < 0) |
4beda316 | 1273 | return log_debug_errno(r, "Failed to mount image %s on %s: %m", mount_entry_source(m), mount_entry_path(m)); |
b3d13314 LB |
1274 | |
1275 | return 1; | |
1276 | } | |
1277 | ||
93f59701 LB |
1278 | static int mount_overlay(const MountEntry *m) { |
1279 | const char *options; | |
1280 | int r; | |
1281 | ||
1282 | assert(m); | |
1283 | ||
1284 | options = strjoina("lowerdir=", mount_entry_options(m)); | |
1285 | ||
1286 | (void) mkdir_p_label(mount_entry_path(m), 0755); | |
1287 | ||
1288 | r = mount_nofollow_verbose(LOG_DEBUG, "overlay", mount_entry_path(m), "overlay", MS_RDONLY, options); | |
1289 | if (r == -ENOENT && m->ignore) | |
1290 | return 0; | |
1291 | if (r < 0) | |
1292 | return r; | |
1293 | ||
1294 | return 1; | |
1295 | } | |
1296 | ||
088696fe | 1297 | static int follow_symlink( |
d2d6c096 | 1298 | const char *root_directory, |
088696fe | 1299 | MountEntry *m) { |
d2d6c096 | 1300 | |
088696fe | 1301 | _cleanup_free_ char *target = NULL; |
8fceda93 LP |
1302 | int r; |
1303 | ||
088696fe LP |
1304 | /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we |
1305 | * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at | |
1306 | * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the | |
1307 | * end and already have a fully normalized name. */ | |
8fceda93 | 1308 | |
a5648b80 | 1309 | r = chase_symlinks(mount_entry_path(m), root_directory, CHASE_STEP|CHASE_NONEXISTENT, &target, NULL); |
088696fe LP |
1310 | if (r < 0) |
1311 | return log_debug_errno(r, "Failed to chase symlinks '%s': %m", mount_entry_path(m)); | |
1312 | if (r > 0) /* Reached the end, nothing more to resolve */ | |
1313 | return 1; | |
8fceda93 | 1314 | |
baaa35ad ZJS |
1315 | if (m->n_followed >= CHASE_SYMLINKS_MAX) /* put a boundary on things */ |
1316 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), | |
1317 | "Symlink loop on '%s'.", | |
1318 | mount_entry_path(m)); | |
8fceda93 | 1319 | |
e2341b6b DT |
1320 | log_debug("Followed mount entry path symlink %s %s %s.", |
1321 | mount_entry_path(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), target); | |
8fceda93 | 1322 | |
809ceb82 | 1323 | mount_entry_consume_prefix(m, TAKE_PTR(target)); |
8fceda93 | 1324 | |
088696fe LP |
1325 | m->n_followed ++; |
1326 | ||
1327 | return 0; | |
8fceda93 LP |
1328 | } |
1329 | ||
82fb2da2 | 1330 | static int apply_one_mount( |
8fceda93 | 1331 | const char *root_directory, |
4e399953 LP |
1332 | MountEntry *m, |
1333 | const NamespaceInfo *ns_info) { | |
ac0930c8 | 1334 | |
e5f10caf | 1335 | _cleanup_free_ char *inaccessible = NULL; |
a227a4be | 1336 | bool rbind = true, make = false; |
15ae422b | 1337 | const char *what; |
15ae422b | 1338 | int r; |
15ae422b | 1339 | |
c17ec25e | 1340 | assert(m); |
4e399953 | 1341 | assert(ns_info); |
15ae422b | 1342 | |
34de407a | 1343 | log_debug("Applying namespace mount on %s", mount_entry_path(m)); |
fe3c2583 | 1344 | |
c17ec25e | 1345 | switch (m->mode) { |
15ae422b | 1346 | |
160cfdbe | 1347 | case INACCESSIBLE: { |
e5f10caf AZ |
1348 | _cleanup_free_ char *tmp = NULL; |
1349 | const char *runtime_dir; | |
160cfdbe | 1350 | struct stat target; |
6d313367 LP |
1351 | |
1352 | /* First, get rid of everything that is below if there | |
1353 | * is anything... Then, overmount it with an | |
c4b41707 | 1354 | * inaccessible path. */ |
34de407a | 1355 | (void) umount_recursive(mount_entry_path(m), 0); |
6d313367 | 1356 | |
088696fe LP |
1357 | if (lstat(mount_entry_path(m), &target) < 0) { |
1358 | if (errno == ENOENT && m->ignore) | |
1359 | return 0; | |
1360 | ||
cbc056c8 ZJS |
1361 | return log_debug_errno(errno, "Failed to lstat() %s to determine what to mount over it: %m", |
1362 | mount_entry_path(m)); | |
088696fe | 1363 | } |
15ae422b | 1364 | |
e5f10caf | 1365 | if (geteuid() == 0) |
48b747fa | 1366 | runtime_dir = "/run"; |
e5f10caf | 1367 | else { |
48b747fa LP |
1368 | if (asprintf(&tmp, "/run/user/" UID_FMT, geteuid()) < 0) |
1369 | return -ENOMEM; | |
e5f10caf AZ |
1370 | |
1371 | runtime_dir = tmp; | |
1372 | } | |
1373 | ||
1374 | r = mode_to_inaccessible_node(runtime_dir, target.st_mode, &inaccessible); | |
1375 | if (r < 0) | |
baaa35ad ZJS |
1376 | return log_debug_errno(SYNTHETIC_ERRNO(ELOOP), |
1377 | "File type not supported for inaccessible mounts. Note that symlinks are not allowed"); | |
e5f10caf | 1378 | what = inaccessible; |
c4b41707 | 1379 | break; |
160cfdbe | 1380 | } |
fe3c2583 | 1381 | |
15ae422b | 1382 | case READONLY: |
15ae422b | 1383 | case READWRITE: |
1e05071d | 1384 | case READWRITE_IMPLICIT: |
ddc155b2 TM |
1385 | case EXEC: |
1386 | case NOEXEC: | |
8fceda93 | 1387 | r = path_is_mount_point(mount_entry_path(m), root_directory, 0); |
088696fe LP |
1388 | if (r == -ENOENT && m->ignore) |
1389 | return 0; | |
d944dc95 | 1390 | if (r < 0) |
cbc056c8 ZJS |
1391 | return log_debug_errno(r, "Failed to determine whether %s is already a mount point: %m", |
1392 | mount_entry_path(m)); | |
1393 | if (r > 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY | |
ddc155b2 | 1394 | * and MS_NOEXEC bits for the mount point if needed. */ |
6b7c9f8b | 1395 | return 0; |
6b7c9f8b | 1396 | /* This isn't a mount point yet, let's make it one. */ |
34de407a | 1397 | what = mount_entry_path(m); |
6b7c9f8b | 1398 | break; |
15ae422b | 1399 | |
a07b9926 LB |
1400 | case EXTENSION_DIRECTORIES: { |
1401 | _cleanup_free_ char *host_os_release_id = NULL, *host_os_release_version_id = NULL, | |
1402 | *host_os_release_sysext_level = NULL, *extension_name = NULL; | |
1403 | _cleanup_strv_free_ char **extension_release = NULL; | |
1404 | ||
1405 | r = path_extract_filename(mount_entry_source(m), &extension_name); | |
1406 | if (r < 0) | |
1407 | return log_debug_errno(r, "Failed to extract extension name from %s: %m", mount_entry_source(m)); | |
1408 | ||
1409 | r = parse_os_release( | |
1410 | empty_to_root(root_directory), | |
1411 | "ID", &host_os_release_id, | |
1412 | "VERSION_ID", &host_os_release_version_id, | |
1413 | "SYSEXT_LEVEL", &host_os_release_sysext_level, | |
1414 | NULL); | |
1415 | if (r < 0) | |
1416 | return log_debug_errno(r, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory)); | |
1417 | if (isempty(host_os_release_id)) | |
1418 | 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)); | |
1419 | ||
06768b90 | 1420 | r = load_extension_release_pairs(mount_entry_source(m), extension_name, /* relax_extension_release_check= */ false, &extension_release); |
a07b9926 LB |
1421 | if (r == -ENOENT && m->ignore) |
1422 | return 0; | |
1423 | if (r < 0) | |
1424 | return log_debug_errno(r, "Failed to parse directory %s extension-release metadata: %m", extension_name); | |
1425 | ||
1426 | r = extension_release_validate( | |
1427 | extension_name, | |
1428 | host_os_release_id, | |
1429 | host_os_release_version_id, | |
1430 | host_os_release_sysext_level, | |
1431 | /* host_sysext_scope */ NULL, /* Leave empty, we need to accept both system and portable */ | |
1432 | extension_release); | |
1433 | if (r == 0) | |
1434 | return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Directory %s extension-release metadata does not match the root's", extension_name); | |
1435 | if (r < 0) | |
1436 | return log_debug_errno(r, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name); | |
1437 | ||
1438 | _fallthrough_; | |
1439 | } | |
1440 | ||
d2d6c096 LP |
1441 | case BIND_MOUNT: |
1442 | rbind = false; | |
d2d6c096 | 1443 | |
4831981d | 1444 | _fallthrough_; |
088696fe LP |
1445 | case BIND_MOUNT_RECURSIVE: { |
1446 | _cleanup_free_ char *chased = NULL; | |
5d997827 | 1447 | |
cbc056c8 ZJS |
1448 | /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note |
1449 | * that bind mount source paths are always relative to the host root, hence we pass NULL as | |
1450 | * root directory to chase_symlinks() here. */ | |
088696fe | 1451 | |
a5648b80 | 1452 | r = chase_symlinks(mount_entry_source(m), NULL, CHASE_TRAIL_SLASH, &chased, NULL); |
088696fe LP |
1453 | if (r == -ENOENT && m->ignore) { |
1454 | log_debug_errno(r, "Path %s does not exist, ignoring.", mount_entry_source(m)); | |
1455 | return 0; | |
1456 | } | |
1457 | if (r < 0) | |
1458 | return log_debug_errno(r, "Failed to follow symlinks on %s: %m", mount_entry_source(m)); | |
1459 | ||
e2341b6b DT |
1460 | log_debug("Followed source symlinks %s %s %s.", |
1461 | mount_entry_source(m), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), chased); | |
088696fe LP |
1462 | |
1463 | free_and_replace(m->source_malloc, chased); | |
d2d6c096 LP |
1464 | |
1465 | what = mount_entry_source(m); | |
a227a4be | 1466 | make = true; |
d2d6c096 | 1467 | break; |
088696fe | 1468 | } |
d2d6c096 | 1469 | |
6c47cd7d | 1470 | case EMPTY_DIR: |
2abd4e38 YW |
1471 | case TMPFS: |
1472 | return mount_tmpfs(m); | |
6c47cd7d | 1473 | |
ac0930c8 | 1474 | case PRIVATE_TMP: |
56a13a49 | 1475 | case PRIVATE_TMP_READONLY: |
89bd586c | 1476 | what = mount_entry_source(m); |
a227a4be | 1477 | make = true; |
15ae422b | 1478 | break; |
e364ad06 | 1479 | |
d6797c92 | 1480 | case PRIVATE_DEV: |
5d997827 LP |
1481 | return mount_private_dev(m); |
1482 | ||
1483 | case BIND_DEV: | |
1484 | return mount_bind_dev(m); | |
1485 | ||
1486 | case SYSFS: | |
1487 | return mount_sysfs(m); | |
1488 | ||
1489 | case PROCFS: | |
4e399953 | 1490 | return mount_procfs(m, ns_info); |
d6797c92 | 1491 | |
94293d65 LB |
1492 | case RUN: |
1493 | return mount_run(m); | |
1494 | ||
80271a44 XR |
1495 | case MQUEUEFS: |
1496 | return mount_mqueuefs(m); | |
1497 | ||
b3d13314 | 1498 | case MOUNT_IMAGES: |
93f59701 LB |
1499 | return mount_image(m, NULL); |
1500 | ||
1501 | case EXTENSION_IMAGES: | |
1502 | return mount_image(m, root_directory); | |
1503 | ||
1504 | case OVERLAY_MOUNT: | |
1505 | return mount_overlay(m); | |
b3d13314 | 1506 | |
e364ad06 | 1507 | default: |
04499a70 | 1508 | assert_not_reached(); |
15ae422b LP |
1509 | } |
1510 | ||
ac0930c8 | 1511 | assert(what); |
15ae422b | 1512 | |
21935150 LP |
1513 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); |
1514 | if (r < 0) { | |
a227a4be | 1515 | bool try_again = false; |
a227a4be LP |
1516 | |
1517 | if (r == -ENOENT && make) { | |
8bab8029 | 1518 | int q; |
a227a4be | 1519 | |
cbc056c8 ZJS |
1520 | /* Hmm, either the source or the destination are missing. Let's see if we can create |
1521 | the destination, then try again. */ | |
a227a4be | 1522 | |
8bab8029 | 1523 | (void) mkdir_parents(mount_entry_path(m), 0755); |
a227a4be | 1524 | |
8bab8029 | 1525 | q = make_mount_point_inode_from_path(what, mount_entry_path(m), 0755); |
9d6d4c30 | 1526 | if (q < 0 && q != -EEXIST) |
8bab8029 LB |
1527 | log_error_errno(q, "Failed to create destination mount point node '%s': %m", |
1528 | mount_entry_path(m)); | |
1529 | else | |
1530 | try_again = true; | |
a227a4be LP |
1531 | } |
1532 | ||
21935150 LP |
1533 | if (try_again) |
1534 | r = mount_nofollow_verbose(LOG_DEBUG, what, mount_entry_path(m), NULL, MS_BIND|(rbind ? MS_REC : 0), NULL); | |
a227a4be | 1535 | if (r < 0) |
5dc60faa | 1536 | return log_error_errno(r, "Failed to mount %s to %s: %m", what, mount_entry_path(m)); |
a227a4be | 1537 | } |
6b7c9f8b | 1538 | |
34de407a | 1539 | log_debug("Successfully mounted %s to %s", what, mount_entry_path(m)); |
6b7c9f8b | 1540 | return 0; |
ac0930c8 | 1541 | } |
15ae422b | 1542 | |
6b000af4 | 1543 | static int make_read_only(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
9ce4e4b0 | 1544 | unsigned long new_flags = 0, flags_mask = 0; |
57ccd9f6 YW |
1545 | bool submounts; |
1546 | int r; | |
15ae422b | 1547 | |
c17ec25e | 1548 | assert(m); |
ac9de0b3 | 1549 | assert(proc_self_mountinfo); |
ac0930c8 | 1550 | |
9ce4e4b0 LP |
1551 | if (mount_entry_read_only(m) || m->mode == PRIVATE_DEV) { |
1552 | new_flags |= MS_RDONLY; | |
1553 | flags_mask |= MS_RDONLY; | |
1554 | } | |
1555 | ||
1556 | if (m->nosuid) { | |
1557 | new_flags |= MS_NOSUID; | |
1558 | flags_mask |= MS_NOSUID; | |
1559 | } | |
1560 | ||
1561 | if (flags_mask == 0) /* No Change? */ | |
6b7c9f8b LP |
1562 | return 0; |
1563 | ||
9ce4e4b0 LP |
1564 | /* We generally apply these changes recursively, except for /dev, and the cases we know there's |
1565 | * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the | |
1566 | * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace | |
1567 | * and running Linux <= 4.17. */ | |
1568 | submounts = | |
1569 | mount_entry_read_only(m) && | |
1570 | !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1571 | if (submounts) | |
6b000af4 | 1572 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); |
9ce4e4b0 | 1573 | else |
7cce68e1 | 1574 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); |
9ce4e4b0 | 1575 | |
ddc155b2 | 1576 | /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked |
867189b5 LP |
1577 | * read-only already stays this way. This improves compatibility with container managers, where we |
1578 | * won't attempt to undo read-only mounts already applied. */ | |
ac0930c8 | 1579 | |
8fceda93 | 1580 | if (r == -ENOENT && m->ignore) |
867189b5 | 1581 | return 0; |
763a260a | 1582 | if (r < 0) |
9ce4e4b0 | 1583 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), |
763a260a | 1584 | submounts ? " and its submounts" : ""); |
763a260a | 1585 | return 0; |
d944dc95 LP |
1586 | } |
1587 | ||
ddc155b2 TM |
1588 | static int make_noexec(const MountEntry *m, char **deny_list, FILE *proc_self_mountinfo) { |
1589 | unsigned long new_flags = 0, flags_mask = 0; | |
57ccd9f6 YW |
1590 | bool submounts; |
1591 | int r; | |
ddc155b2 TM |
1592 | |
1593 | assert(m); | |
1594 | assert(proc_self_mountinfo); | |
1595 | ||
1596 | if (mount_entry_noexec(m)) { | |
1597 | new_flags |= MS_NOEXEC; | |
1598 | flags_mask |= MS_NOEXEC; | |
1599 | } else if (mount_entry_exec(m)) { | |
1600 | new_flags &= ~MS_NOEXEC; | |
1601 | flags_mask |= MS_NOEXEC; | |
1602 | } | |
1603 | ||
1604 | if (flags_mask == 0) /* No Change? */ | |
1605 | return 0; | |
1606 | ||
1607 | submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1608 | ||
1609 | if (submounts) | |
1610 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, deny_list, proc_self_mountinfo); | |
1611 | else | |
1612 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), new_flags, flags_mask, proc_self_mountinfo); | |
1613 | ||
1614 | if (r == -ENOENT && m->ignore) | |
1615 | return 0; | |
1616 | if (r < 0) | |
1617 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1618 | submounts ? " and its submounts" : ""); | |
1619 | return 0; | |
1620 | } | |
1621 | ||
6720e356 | 1622 | static int make_nosuid(const MountEntry *m, FILE *proc_self_mountinfo) { |
57ccd9f6 YW |
1623 | bool submounts; |
1624 | int r; | |
6720e356 YW |
1625 | |
1626 | assert(m); | |
1627 | assert(proc_self_mountinfo); | |
1628 | ||
1629 | submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS); | |
1630 | ||
1631 | if (submounts) | |
1632 | r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, NULL, proc_self_mountinfo); | |
1633 | else | |
1634 | r = bind_remount_one_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, proc_self_mountinfo); | |
1635 | if (r == -ENOENT && m->ignore) | |
1636 | return 0; | |
1637 | if (r < 0) | |
1638 | return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m), | |
1639 | submounts ? " and its submounts" : ""); | |
1640 | return 0; | |
1641 | } | |
1642 | ||
9b68367b | 1643 | static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) { |
5d997827 LP |
1644 | assert(ns_info); |
1645 | ||
9c988f93 DH |
1646 | /* |
1647 | * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=, | |
1648 | * since to protect the API VFS mounts, they need to be around in the | |
9b68367b | 1649 | * first place... |
9c988f93 | 1650 | */ |
5d997827 | 1651 | |
9b68367b YW |
1652 | return ns_info->mount_apivfs || |
1653 | ns_info->protect_control_groups || | |
4e399953 LP |
1654 | ns_info->protect_kernel_tunables || |
1655 | ns_info->protect_proc != PROTECT_PROC_DEFAULT || | |
1656 | ns_info->proc_subset != PROC_SUBSET_ALL; | |
5d997827 LP |
1657 | } |
1658 | ||
da6053d0 | 1659 | static size_t namespace_calculate_mounts( |
bb0ff3fb | 1660 | const NamespaceInfo *ns_info, |
2652c6c1 DH |
1661 | char** read_write_paths, |
1662 | char** read_only_paths, | |
1663 | char** inaccessible_paths, | |
ddc155b2 TM |
1664 | char** exec_paths, |
1665 | char** no_exec_paths, | |
6c47cd7d | 1666 | char** empty_directories, |
da6053d0 LP |
1667 | size_t n_bind_mounts, |
1668 | size_t n_temporary_filesystems, | |
b3d13314 | 1669 | size_t n_mount_images, |
93f59701 | 1670 | size_t n_extension_images, |
a07b9926 | 1671 | size_t n_extension_directories, |
93f59701 | 1672 | size_t n_hierarchies, |
2652c6c1 DH |
1673 | const char* tmp_dir, |
1674 | const char* var_tmp_dir, | |
bbb4e7f3 | 1675 | const char *creds_path, |
5e8deb94 | 1676 | const char* log_namespace, |
3bdc25a4 LP |
1677 | bool setup_propagate, |
1678 | const char* notify_socket) { | |
2652c6c1 | 1679 | |
da6053d0 LP |
1680 | size_t protect_home_cnt; |
1681 | size_t protect_system_cnt = | |
52b3d652 | 1682 | (ns_info->protect_system == PROTECT_SYSTEM_STRICT ? |
f471b2af | 1683 | ELEMENTSOF(protect_system_strict_table) : |
52b3d652 | 1684 | ((ns_info->protect_system == PROTECT_SYSTEM_FULL) ? |
f471b2af | 1685 | ELEMENTSOF(protect_system_full_table) : |
52b3d652 | 1686 | ((ns_info->protect_system == PROTECT_SYSTEM_YES) ? |
f471b2af DH |
1687 | ELEMENTSOF(protect_system_yes_table) : 0))); |
1688 | ||
b6c432ca | 1689 | protect_home_cnt = |
52b3d652 | 1690 | (ns_info->protect_home == PROTECT_HOME_YES ? |
b6c432ca | 1691 | ELEMENTSOF(protect_home_yes_table) : |
52b3d652 | 1692 | ((ns_info->protect_home == PROTECT_HOME_READ_ONLY) ? |
e4da7d8c | 1693 | ELEMENTSOF(protect_home_read_only_table) : |
52b3d652 | 1694 | ((ns_info->protect_home == PROTECT_HOME_TMPFS) ? |
e4da7d8c | 1695 | ELEMENTSOF(protect_home_tmpfs_table) : 0))); |
b6c432ca | 1696 | |
2652c6c1 DH |
1697 | return !!tmp_dir + !!var_tmp_dir + |
1698 | strv_length(read_write_paths) + | |
1699 | strv_length(read_only_paths) + | |
1700 | strv_length(inaccessible_paths) + | |
ddc155b2 TM |
1701 | strv_length(exec_paths) + |
1702 | strv_length(no_exec_paths) + | |
6c47cd7d | 1703 | strv_length(empty_directories) + |
d2d6c096 | 1704 | n_bind_mounts + |
b3d13314 | 1705 | n_mount_images + |
a07b9926 | 1706 | (n_extension_images > 0 || n_extension_directories > 0 ? /* Mount each image and directory plus an overlay per hierarchy */ |
7c76e181 | 1707 | n_hierarchies + n_extension_images + n_extension_directories: 0) + |
2abd4e38 | 1708 | n_temporary_filesystems + |
c575770b | 1709 | ns_info->private_dev + |
788e7201 TM |
1710 | (ns_info->protect_kernel_tunables ? |
1711 | ELEMENTSOF(protect_kernel_tunables_proc_table) + ELEMENTSOF(protect_kernel_tunables_sys_table) : 0) + | |
c575770b | 1712 | (ns_info->protect_kernel_modules ? ELEMENTSOF(protect_kernel_modules_table) : 0) + |
788e7201 TM |
1713 | (ns_info->protect_kernel_logs ? |
1714 | ELEMENTSOF(protect_kernel_logs_proc_table) + ELEMENTSOF(protect_kernel_logs_dev_table) : 0) + | |
94a7b275 | 1715 | (ns_info->protect_control_groups ? 1 : 0) + |
5d997827 | 1716 | protect_home_cnt + protect_system_cnt + |
aecd5ac6 | 1717 | (ns_info->protect_hostname ? 2 : 0) + |
91dd5f7c | 1718 | (namespace_info_mount_apivfs(ns_info) ? ELEMENTSOF(apivfs_table) : 0) + |
bbb4e7f3 | 1719 | (creds_path ? 2 : 1) + |
3bdc25a4 LP |
1720 | !!log_namespace + |
1721 | setup_propagate + /* /run/systemd/incoming */ | |
80271a44 XR |
1722 | !!notify_socket + |
1723 | ns_info->private_ipc; /* /dev/mqueue */ | |
2652c6c1 DH |
1724 | } |
1725 | ||
fbf90c0d CB |
1726 | /* Walk all mount entries and dropping any unused mounts. This affects all |
1727 | * mounts: | |
1728 | * - that are implicitly protected by a path that has been rendered inaccessible | |
1729 | * - whose immediate parent requests the same protection mode as the mount itself | |
1730 | * - that are outside of the relevant root directory | |
1731 | * - which are duplicates | |
1732 | */ | |
1733 | static void drop_unused_mounts(const char *root_directory, MountEntry *mounts, size_t *n_mounts) { | |
9b68367b | 1734 | assert(root_directory); |
f8b64b57 LP |
1735 | assert(n_mounts); |
1736 | assert(mounts || *n_mounts == 0); | |
1737 | ||
93bab288 | 1738 | typesafe_qsort(mounts, *n_mounts, mount_path_compare); |
f8b64b57 LP |
1739 | |
1740 | drop_duplicates(mounts, n_mounts); | |
1741 | drop_outside_root(root_directory, mounts, n_mounts); | |
1742 | drop_inaccessible(mounts, n_mounts); | |
1743 | drop_nop(mounts, n_mounts); | |
1744 | } | |
1745 | ||
df61e79a | 1746 | static int create_symlinks_from_tuples(const char *root, char **strv_symlinks) { |
df61e79a LB |
1747 | int r; |
1748 | ||
1749 | STRV_FOREACH_PAIR(src, dst, strv_symlinks) { | |
1750 | _cleanup_free_ char *src_abs = NULL, *dst_abs = NULL; | |
1751 | ||
1752 | src_abs = path_join(root, *src); | |
1753 | dst_abs = path_join(root, *dst); | |
1754 | if (!src_abs || !dst_abs) | |
1755 | return -ENOMEM; | |
1756 | ||
1757 | r = mkdir_parents_label(dst_abs, 0755); | |
1758 | if (r < 0) | |
1759 | return r; | |
1760 | ||
1761 | r = symlink_idempotent(src_abs, dst_abs, true); | |
1762 | if (r < 0) | |
1763 | return r; | |
1764 | } | |
1765 | ||
1766 | return 0; | |
1767 | } | |
1768 | ||
82fb2da2 LB |
1769 | static int apply_mounts( |
1770 | const char *root, | |
1771 | const NamespaceInfo *ns_info, | |
1772 | MountEntry *mounts, | |
1773 | size_t *n_mounts, | |
df61e79a | 1774 | char **exec_dir_symlinks, |
82fb2da2 LB |
1775 | char **error_path) { |
1776 | ||
1777 | _cleanup_fclose_ FILE *proc_self_mountinfo = NULL; | |
1778 | _cleanup_free_ char **deny_list = NULL; | |
82fb2da2 LB |
1779 | int r; |
1780 | ||
1781 | if (n_mounts == 0) /* Shortcut: nothing to do */ | |
1782 | return 0; | |
1783 | ||
1784 | assert(root); | |
1785 | assert(mounts); | |
1786 | assert(n_mounts); | |
1787 | ||
1788 | /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of | |
1789 | * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */ | |
1790 | proc_self_mountinfo = fopen("/proc/self/mountinfo", "re"); | |
1791 | if (!proc_self_mountinfo) { | |
d60e3b40 LP |
1792 | r = -errno; |
1793 | ||
82fb2da2 LB |
1794 | if (error_path) |
1795 | *error_path = strdup("/proc/self/mountinfo"); | |
d60e3b40 LP |
1796 | |
1797 | return log_debug_errno(r, "Failed to open /proc/self/mountinfo: %m"); | |
82fb2da2 LB |
1798 | } |
1799 | ||
1800 | /* First round, establish all mounts we need */ | |
1801 | for (;;) { | |
1802 | bool again = false; | |
1803 | ||
1804 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1805 | ||
1806 | if (m->applied) | |
1807 | continue; | |
1808 | ||
a07b9926 LB |
1809 | /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */ |
1810 | r = follow_symlink(!IN_SET(m->mode, EXTENSION_IMAGES, EXTENSION_DIRECTORIES) ? root : NULL, m); | |
82fb2da2 LB |
1811 | if (r < 0) { |
1812 | if (error_path && mount_entry_path(m)) | |
1813 | *error_path = strdup(mount_entry_path(m)); | |
1814 | return r; | |
1815 | } | |
1816 | if (r == 0) { | |
1817 | /* We hit a symlinked mount point. The entry got rewritten and might | |
1818 | * point to a very different place now. Let's normalize the changed | |
1819 | * list, and start from the beginning. After all to mount the entry | |
1820 | * at the new location we might need some other mounts first */ | |
1821 | again = true; | |
1822 | break; | |
1823 | } | |
1824 | ||
1825 | r = apply_one_mount(root, m, ns_info); | |
1826 | if (r < 0) { | |
1827 | if (error_path && mount_entry_path(m)) | |
1828 | *error_path = strdup(mount_entry_path(m)); | |
1829 | return r; | |
1830 | } | |
1831 | ||
1832 | m->applied = true; | |
1833 | } | |
1834 | ||
1835 | if (!again) | |
1836 | break; | |
1837 | ||
fbf90c0d | 1838 | drop_unused_mounts(root, mounts, n_mounts); |
82fb2da2 LB |
1839 | } |
1840 | ||
df61e79a LB |
1841 | /* Now that all filesystems have been set up, but before the |
1842 | * read-only switches are flipped, create the exec dirs symlinks. | |
1843 | * Note that when /var/lib is not empty/tmpfs, these symlinks will already | |
1844 | * exist, which means this will be a no-op. */ | |
1845 | r = create_symlinks_from_tuples(root, exec_dir_symlinks); | |
1846 | if (r < 0) | |
6d7c999a | 1847 | return log_debug_errno(r, "Failed to set up ExecDirectories symlinks inside mount namespace: %m"); |
df61e79a | 1848 | |
82fb2da2 LB |
1849 | /* Create a deny list we can pass to bind_mount_recursive() */ |
1850 | deny_list = new(char*, (*n_mounts)+1); | |
1851 | if (!deny_list) | |
1852 | return -ENOMEM; | |
8794e6db | 1853 | for (size_t j = 0; j < *n_mounts; j++) |
82fb2da2 | 1854 | deny_list[j] = (char*) mount_entry_path(mounts+j); |
8794e6db | 1855 | deny_list[*n_mounts] = NULL; |
82fb2da2 LB |
1856 | |
1857 | /* Second round, flip the ro bits if necessary. */ | |
1858 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1859 | r = make_read_only(m, deny_list, proc_self_mountinfo); | |
1860 | if (r < 0) { | |
1861 | if (error_path && mount_entry_path(m)) | |
1862 | *error_path = strdup(mount_entry_path(m)); | |
1863 | return r; | |
1864 | } | |
1865 | } | |
1866 | ||
1867 | /* Third round, flip the noexec bits with a simplified deny list. */ | |
8794e6db | 1868 | for (size_t j = 0; j < *n_mounts; j++) |
82fb2da2 LB |
1869 | if (IN_SET((mounts+j)->mode, EXEC, NOEXEC)) |
1870 | deny_list[j] = (char*) mount_entry_path(mounts+j); | |
8794e6db | 1871 | deny_list[*n_mounts] = NULL; |
82fb2da2 LB |
1872 | |
1873 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1874 | r = make_noexec(m, deny_list, proc_self_mountinfo); | |
1875 | if (r < 0) { | |
1876 | if (error_path && mount_entry_path(m)) | |
1877 | *error_path = strdup(mount_entry_path(m)); | |
1878 | return r; | |
1879 | } | |
1880 | } | |
1881 | ||
6720e356 YW |
1882 | /* Fourth round, flip the nosuid bits without a deny list. */ |
1883 | if (ns_info->mount_nosuid) | |
1884 | for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) { | |
1885 | r = make_nosuid(m, proc_self_mountinfo); | |
1886 | if (r < 0) { | |
1887 | if (error_path && mount_entry_path(m)) | |
1888 | *error_path = strdup(mount_entry_path(m)); | |
1889 | return r; | |
1890 | } | |
1891 | } | |
1892 | ||
82fb2da2 LB |
1893 | return 1; |
1894 | } | |
1895 | ||
c8c535d5 LP |
1896 | static bool root_read_only( |
1897 | char **read_only_paths, | |
1898 | ProtectSystem protect_system) { | |
1899 | ||
1900 | /* Determine whether the root directory is going to be read-only given the configured settings. */ | |
1901 | ||
1902 | if (protect_system == PROTECT_SYSTEM_STRICT) | |
1903 | return true; | |
1904 | ||
de46b2be | 1905 | if (prefixed_path_strv_contains(read_only_paths, "/")) |
c8c535d5 LP |
1906 | return true; |
1907 | ||
1908 | return false; | |
1909 | } | |
1910 | ||
1911 | static bool home_read_only( | |
1912 | char** read_only_paths, | |
1913 | char** inaccessible_paths, | |
1914 | char** empty_directories, | |
1915 | const BindMount *bind_mounts, | |
1916 | size_t n_bind_mounts, | |
1917 | const TemporaryFileSystem *temporary_filesystems, | |
1918 | size_t n_temporary_filesystems, | |
1919 | ProtectHome protect_home) { | |
1920 | ||
c8c535d5 LP |
1921 | /* Determine whether the /home directory is going to be read-only given the configured settings. Yes, |
1922 | * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple | |
1923 | * settings. */ | |
1924 | ||
1925 | if (protect_home != PROTECT_HOME_NO) | |
1926 | return true; | |
1927 | ||
de46b2be TM |
1928 | if (prefixed_path_strv_contains(read_only_paths, "/home") || |
1929 | prefixed_path_strv_contains(inaccessible_paths, "/home") || | |
1930 | prefixed_path_strv_contains(empty_directories, "/home")) | |
c8c535d5 LP |
1931 | return true; |
1932 | ||
8794e6db | 1933 | for (size_t i = 0; i < n_temporary_filesystems; i++) |
c8c535d5 LP |
1934 | if (path_equal(temporary_filesystems[i].path, "/home")) |
1935 | return true; | |
1936 | ||
1937 | /* If /home is overmounted with some dir from the host it's not writable. */ | |
8794e6db | 1938 | for (size_t i = 0; i < n_bind_mounts; i++) |
c8c535d5 LP |
1939 | if (path_equal(bind_mounts[i].destination, "/home")) |
1940 | return true; | |
1941 | ||
1942 | return false; | |
1943 | } | |
1944 | ||
89e62e0b LP |
1945 | static int verity_settings_prepare( |
1946 | VeritySettings *verity, | |
1947 | const char *root_image, | |
1948 | const void *root_hash, | |
1949 | size_t root_hash_size, | |
1950 | const char *root_hash_path, | |
1951 | const void *root_hash_sig, | |
1952 | size_t root_hash_sig_size, | |
1953 | const char *root_hash_sig_path, | |
1954 | const char *verity_data_path) { | |
1955 | ||
1956 | int r; | |
1957 | ||
1958 | assert(verity); | |
1959 | ||
1960 | if (root_hash) { | |
1961 | void *d; | |
1962 | ||
1963 | d = memdup(root_hash, root_hash_size); | |
1964 | if (!d) | |
1965 | return -ENOMEM; | |
1966 | ||
1967 | free_and_replace(verity->root_hash, d); | |
1968 | verity->root_hash_size = root_hash_size; | |
aee36b4e | 1969 | verity->designator = PARTITION_ROOT; |
89e62e0b LP |
1970 | } |
1971 | ||
1972 | if (root_hash_sig) { | |
1973 | void *d; | |
1974 | ||
1975 | d = memdup(root_hash_sig, root_hash_sig_size); | |
1976 | if (!d) | |
1977 | return -ENOMEM; | |
1978 | ||
1979 | free_and_replace(verity->root_hash_sig, d); | |
1980 | verity->root_hash_sig_size = root_hash_sig_size; | |
aee36b4e | 1981 | verity->designator = PARTITION_ROOT; |
89e62e0b LP |
1982 | } |
1983 | ||
1984 | if (verity_data_path) { | |
1985 | r = free_and_strdup(&verity->data_path, verity_data_path); | |
1986 | if (r < 0) | |
1987 | return r; | |
1988 | } | |
1989 | ||
1990 | r = verity_settings_load( | |
1991 | verity, | |
1992 | root_image, | |
1993 | root_hash_path, | |
1994 | root_hash_sig_path); | |
1995 | if (r < 0) | |
1996 | return log_debug_errno(r, "Failed to load root hash: %m"); | |
1997 | ||
1998 | return 0; | |
1999 | } | |
2000 | ||
613b411c | 2001 | int setup_namespace( |
ee818b89 | 2002 | const char* root_directory, |
915e6d16 | 2003 | const char* root_image, |
18d73705 | 2004 | const MountOptions *root_image_options, |
bb0ff3fb | 2005 | const NamespaceInfo *ns_info, |
2a624c36 AP |
2006 | char** read_write_paths, |
2007 | char** read_only_paths, | |
2008 | char** inaccessible_paths, | |
ddc155b2 TM |
2009 | char** exec_paths, |
2010 | char** no_exec_paths, | |
6c47cd7d | 2011 | char** empty_directories, |
df61e79a | 2012 | char** exec_dir_symlinks, |
d2d6c096 | 2013 | const BindMount *bind_mounts, |
da6053d0 | 2014 | size_t n_bind_mounts, |
2abd4e38 | 2015 | const TemporaryFileSystem *temporary_filesystems, |
da6053d0 | 2016 | size_t n_temporary_filesystems, |
b3d13314 LB |
2017 | const MountImage *mount_images, |
2018 | size_t n_mount_images, | |
a004cb4c LP |
2019 | const char* tmp_dir, |
2020 | const char* var_tmp_dir, | |
bbb4e7f3 | 2021 | const char *creds_path, |
91dd5f7c | 2022 | const char *log_namespace, |
915e6d16 | 2023 | unsigned long mount_flags, |
0389f4fa LB |
2024 | const void *root_hash, |
2025 | size_t root_hash_size, | |
2026 | const char *root_hash_path, | |
d4d55b0d LB |
2027 | const void *root_hash_sig, |
2028 | size_t root_hash_sig_size, | |
2029 | const char *root_hash_sig_path, | |
89e62e0b | 2030 | const char *verity_data_path, |
93f59701 LB |
2031 | const MountImage *extension_images, |
2032 | size_t n_extension_images, | |
a07b9926 | 2033 | char **extension_directories, |
5e8deb94 LB |
2034 | const char *propagate_dir, |
2035 | const char *incoming_dir, | |
24759d8f | 2036 | const char *extension_dir, |
3bdc25a4 | 2037 | const char *notify_socket, |
7cc5ef5f | 2038 | char **error_path) { |
15ae422b | 2039 | |
915e6d16 LP |
2040 | _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL; |
2041 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; | |
aee36b4e | 2042 | _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT; |
93f59701 | 2043 | _cleanup_strv_free_ char **hierarchies = NULL; |
5f7a690a | 2044 | MountEntry *m = NULL, *mounts = NULL; |
5e8deb94 | 2045 | bool require_prefix = false, setup_propagate = false; |
24759d8f | 2046 | const char *root; |
9cb1709b LP |
2047 | DissectImageFlags dissect_image_flags = |
2048 | DISSECT_IMAGE_GENERIC_ROOT | | |
2049 | DISSECT_IMAGE_REQUIRE_ROOT | | |
2050 | DISSECT_IMAGE_DISCARD_ON_LOOP | | |
2051 | DISSECT_IMAGE_RELAX_VAR_CHECK | | |
2052 | DISSECT_IMAGE_FSCK | | |
c65f854a | 2053 | DISSECT_IMAGE_USR_NO_ROOT | |
73d88b80 LP |
2054 | DISSECT_IMAGE_GROWFS | |
2055 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | | |
2056 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
89e62e0b LP |
2057 | size_t n_mounts; |
2058 | int r; | |
15ae422b | 2059 | |
915e6d16 LP |
2060 | assert(ns_info); |
2061 | ||
cdf42f9b LP |
2062 | /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes |
2063 | * we configure take effect */ | |
2064 | BLOCK_WITH_UMASK(0000); | |
2065 | ||
5e8deb94 LB |
2066 | if (!isempty(propagate_dir) && !isempty(incoming_dir)) |
2067 | setup_propagate = true; | |
2068 | ||
613b411c | 2069 | if (mount_flags == 0) |
c17ec25e | 2070 | mount_flags = MS_SHARED; |
ac0930c8 | 2071 | |
915e6d16 | 2072 | if (root_image) { |
c8c535d5 LP |
2073 | /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */ |
2074 | if (root_read_only(read_only_paths, | |
52b3d652 | 2075 | ns_info->protect_system) && |
c8c535d5 LP |
2076 | home_read_only(read_only_paths, inaccessible_paths, empty_directories, |
2077 | bind_mounts, n_bind_mounts, temporary_filesystems, n_temporary_filesystems, | |
52b3d652 | 2078 | ns_info->protect_home) && |
c9ef8573 | 2079 | strv_isempty(read_write_paths)) |
915e6d16 LP |
2080 | dissect_image_flags |= DISSECT_IMAGE_READ_ONLY; |
2081 | ||
89e62e0b LP |
2082 | r = verity_settings_prepare( |
2083 | &verity, | |
2084 | root_image, | |
2085 | root_hash, root_hash_size, root_hash_path, | |
2086 | root_hash_sig, root_hash_sig_size, root_hash_sig_path, | |
2087 | verity_data_path); | |
915e6d16 | 2088 | if (r < 0) |
89e62e0b LP |
2089 | return r; |
2090 | ||
2091 | SET_FLAG(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE, verity.data_path); | |
915e6d16 | 2092 | |
89e62e0b LP |
2093 | r = loop_device_make_by_path( |
2094 | root_image, | |
ef9c184d | 2095 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : -1 /* < 0 means writable if possible, read-only as fallback */, |
89e62e0b | 2096 | FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 2097 | LOCK_SH, |
89e62e0b | 2098 | &loop_device); |
78ebe980 | 2099 | if (r < 0) |
89e62e0b LP |
2100 | return log_debug_errno(r, "Failed to create loop device for root image: %m"); |
2101 | ||
bad31660 YW |
2102 | r = dissect_loop_device( |
2103 | loop_device, | |
89e62e0b LP |
2104 | &verity, |
2105 | root_image_options, | |
2106 | dissect_image_flags, | |
2107 | &dissected_image); | |
78ebe980 | 2108 | if (r < 0) |
763a260a | 2109 | return log_debug_errno(r, "Failed to dissect image: %m"); |
78ebe980 | 2110 | |
88b3300f LP |
2111 | r = dissected_image_load_verity_sig_partition( |
2112 | dissected_image, | |
2113 | loop_device->fd, | |
2114 | &verity); | |
2115 | if (r < 0) | |
2116 | return r; | |
2117 | ||
89e62e0b LP |
2118 | r = dissected_image_decrypt( |
2119 | dissected_image, | |
2120 | NULL, | |
2121 | &verity, | |
e330f97a | 2122 | dissect_image_flags); |
915e6d16 | 2123 | if (r < 0) |
763a260a | 2124 | return log_debug_errno(r, "Failed to decrypt dissected image: %m"); |
915e6d16 LP |
2125 | } |
2126 | ||
e908468b LP |
2127 | if (root_directory) |
2128 | root = root_directory; | |
0722b359 | 2129 | else { |
77f16dbd DDM |
2130 | /* /run/systemd should have been created by PID 1 early on already, but in some cases, like |
2131 | * when running tests (test-execute), it might not have been created yet so let's make sure | |
2132 | * we create it if it doesn't already exist. */ | |
2133 | (void) mkdir_p_label("/run/systemd", 0755); | |
2134 | ||
d73020f2 LP |
2135 | /* Always create the mount namespace in a temporary directory, instead of operating directly |
2136 | * in the root. The temporary directory prevents any mounts from being potentially obscured | |
2137 | * my other mounts we already applied. We use the same mount point for all images, which is | |
2138 | * safe, since they all live in their own namespaces after all, and hence won't see each | |
2139 | * other. */ | |
e908468b LP |
2140 | |
2141 | root = "/run/systemd/unit-root"; | |
2142 | (void) mkdir_label(root, 0700); | |
d18aff04 | 2143 | require_prefix = true; |
0722b359 | 2144 | } |
e908468b | 2145 | |
a07b9926 | 2146 | if (n_extension_images > 0 || !strv_isempty(extension_directories)) { |
93f59701 LB |
2147 | r = parse_env_extension_hierarchies(&hierarchies); |
2148 | if (r < 0) | |
2149 | return r; | |
2150 | } | |
2151 | ||
cfbeb4ef LP |
2152 | n_mounts = namespace_calculate_mounts( |
2153 | ns_info, | |
2154 | read_write_paths, | |
2155 | read_only_paths, | |
2156 | inaccessible_paths, | |
ddc155b2 TM |
2157 | exec_paths, |
2158 | no_exec_paths, | |
6c47cd7d | 2159 | empty_directories, |
f5c52a77 | 2160 | n_bind_mounts, |
2abd4e38 | 2161 | n_temporary_filesystems, |
b3d13314 | 2162 | n_mount_images, |
93f59701 | 2163 | n_extension_images, |
a07b9926 | 2164 | strv_length(extension_directories), |
93f59701 | 2165 | strv_length(hierarchies), |
cfbeb4ef | 2166 | tmp_dir, var_tmp_dir, |
bbb4e7f3 | 2167 | creds_path, |
5e8deb94 | 2168 | log_namespace, |
3bdc25a4 LP |
2169 | setup_propagate, |
2170 | notify_socket); | |
613b411c | 2171 | |
f0a4feb0 | 2172 | if (n_mounts > 0) { |
5f7a690a LP |
2173 | m = mounts = new0(MountEntry, n_mounts); |
2174 | if (!mounts) | |
2175 | return -ENOMEM; | |
2176 | ||
d18aff04 | 2177 | r = append_access_mounts(&m, read_write_paths, READWRITE, require_prefix); |
613b411c | 2178 | if (r < 0) |
f0a4feb0 | 2179 | goto finish; |
613b411c | 2180 | |
d18aff04 | 2181 | r = append_access_mounts(&m, read_only_paths, READONLY, require_prefix); |
613b411c | 2182 | if (r < 0) |
f0a4feb0 | 2183 | goto finish; |
613b411c | 2184 | |
d18aff04 | 2185 | r = append_access_mounts(&m, inaccessible_paths, INACCESSIBLE, require_prefix); |
613b411c | 2186 | if (r < 0) |
f0a4feb0 | 2187 | goto finish; |
7ff7394d | 2188 | |
ddc155b2 TM |
2189 | r = append_access_mounts(&m, exec_paths, EXEC, require_prefix); |
2190 | if (r < 0) | |
2191 | goto finish; | |
2192 | ||
2193 | r = append_access_mounts(&m, no_exec_paths, NOEXEC, require_prefix); | |
2194 | if (r < 0) | |
2195 | goto finish; | |
2196 | ||
6c47cd7d LP |
2197 | r = append_empty_dir_mounts(&m, empty_directories); |
2198 | if (r < 0) | |
2199 | goto finish; | |
2200 | ||
d2d6c096 LP |
2201 | r = append_bind_mounts(&m, bind_mounts, n_bind_mounts); |
2202 | if (r < 0) | |
2203 | goto finish; | |
2204 | ||
2abd4e38 YW |
2205 | r = append_tmpfs_mounts(&m, temporary_filesystems, n_temporary_filesystems); |
2206 | if (r < 0) | |
2207 | goto finish; | |
2208 | ||
613b411c | 2209 | if (tmp_dir) { |
56a13a49 ZJS |
2210 | bool ro = streq(tmp_dir, RUN_SYSTEMD_EMPTY); |
2211 | ||
34de407a | 2212 | *(m++) = (MountEntry) { |
5327c910 | 2213 | .path_const = "/tmp", |
56a13a49 | 2214 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, |
89bd586c | 2215 | .source_const = tmp_dir, |
5327c910 | 2216 | }; |
613b411c | 2217 | } |
7ff7394d | 2218 | |
613b411c | 2219 | if (var_tmp_dir) { |
56a13a49 ZJS |
2220 | bool ro = streq(var_tmp_dir, RUN_SYSTEMD_EMPTY); |
2221 | ||
34de407a | 2222 | *(m++) = (MountEntry) { |
5327c910 | 2223 | .path_const = "/var/tmp", |
56a13a49 | 2224 | .mode = ro ? PRIVATE_TMP_READONLY : PRIVATE_TMP, |
89bd586c | 2225 | .source_const = var_tmp_dir, |
5327c910 | 2226 | }; |
7ff7394d | 2227 | } |
ac0930c8 | 2228 | |
b3d13314 LB |
2229 | r = append_mount_images(&m, mount_images, n_mount_images); |
2230 | if (r < 0) | |
2231 | goto finish; | |
2232 | ||
a07b9926 | 2233 | r = append_extensions(&m, root, extension_dir, hierarchies, extension_images, n_extension_images, extension_directories); |
93f59701 LB |
2234 | if (r < 0) |
2235 | goto finish; | |
2236 | ||
d46b79bb | 2237 | if (ns_info->private_dev) |
34de407a | 2238 | *(m++) = (MountEntry) { |
5327c910 LP |
2239 | .path_const = "/dev", |
2240 | .mode = PRIVATE_DEV, | |
9ce4e4b0 | 2241 | .flags = DEV_MOUNT_OPTIONS, |
5327c910 | 2242 | }; |
7f112f50 | 2243 | |
788e7201 TM |
2244 | /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the |
2245 | protective mounts to non-pid /proc paths would fail. But the pid only option may have | |
2246 | failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */ | |
2247 | bool ignore_protect_proc = ns_info->ignore_protect_paths || ns_info->proc_subset == PROC_SUBSET_PID; | |
c575770b | 2248 | if (ns_info->protect_kernel_tunables) { |
cbc056c8 | 2249 | r = append_static_mounts(&m, |
788e7201 TM |
2250 | protect_kernel_tunables_proc_table, |
2251 | ELEMENTSOF(protect_kernel_tunables_proc_table), | |
2252 | ignore_protect_proc); | |
2253 | if (r < 0) | |
2254 | goto finish; | |
2255 | ||
2256 | r = append_static_mounts(&m, | |
2257 | protect_kernel_tunables_sys_table, | |
2258 | ELEMENTSOF(protect_kernel_tunables_sys_table), | |
cbc056c8 | 2259 | ns_info->ignore_protect_paths); |
c575770b | 2260 | if (r < 0) |
f0a4feb0 | 2261 | goto finish; |
c575770b DH |
2262 | } |
2263 | ||
2264 | if (ns_info->protect_kernel_modules) { | |
cbc056c8 ZJS |
2265 | r = append_static_mounts(&m, |
2266 | protect_kernel_modules_table, | |
2267 | ELEMENTSOF(protect_kernel_modules_table), | |
2268 | ns_info->ignore_protect_paths); | |
c575770b | 2269 | if (r < 0) |
f0a4feb0 | 2270 | goto finish; |
c575770b | 2271 | } |
59eeb84b | 2272 | |
94a7b275 | 2273 | if (ns_info->protect_kernel_logs) { |
cbc056c8 | 2274 | r = append_static_mounts(&m, |
788e7201 TM |
2275 | protect_kernel_logs_proc_table, |
2276 | ELEMENTSOF(protect_kernel_logs_proc_table), | |
2277 | ignore_protect_proc); | |
2278 | if (r < 0) | |
2279 | goto finish; | |
2280 | ||
2281 | r = append_static_mounts(&m, | |
2282 | protect_kernel_logs_dev_table, | |
2283 | ELEMENTSOF(protect_kernel_logs_dev_table), | |
cbc056c8 | 2284 | ns_info->ignore_protect_paths); |
94a7b275 KK |
2285 | if (r < 0) |
2286 | goto finish; | |
2287 | } | |
2288 | ||
d46b79bb | 2289 | if (ns_info->protect_control_groups) |
34de407a | 2290 | *(m++) = (MountEntry) { |
5327c910 LP |
2291 | .path_const = "/sys/fs/cgroup", |
2292 | .mode = READONLY, | |
2293 | }; | |
59eeb84b | 2294 | |
52b3d652 | 2295 | r = append_protect_home(&m, ns_info->protect_home, ns_info->ignore_protect_paths); |
b6c432ca | 2296 | if (r < 0) |
f0a4feb0 | 2297 | goto finish; |
417116f2 | 2298 | |
52b3d652 | 2299 | r = append_protect_system(&m, ns_info->protect_system, false); |
f471b2af | 2300 | if (r < 0) |
f0a4feb0 | 2301 | goto finish; |
417116f2 | 2302 | |
9b68367b | 2303 | if (namespace_info_mount_apivfs(ns_info)) { |
cbc056c8 ZJS |
2304 | r = append_static_mounts(&m, |
2305 | apivfs_table, | |
2306 | ELEMENTSOF(apivfs_table), | |
2307 | ns_info->ignore_protect_paths); | |
5d997827 LP |
2308 | if (r < 0) |
2309 | goto finish; | |
2310 | } | |
2311 | ||
1361f015 CB |
2312 | /* Note, if proc is mounted with subset=pid then neither of the |
2313 | * two paths will exist, i.e. they are implicitly protected by | |
2314 | * the mount option. */ | |
aecd5ac6 TM |
2315 | if (ns_info->protect_hostname) { |
2316 | *(m++) = (MountEntry) { | |
2317 | .path_const = "/proc/sys/kernel/hostname", | |
2318 | .mode = READONLY, | |
1361f015 | 2319 | .ignore = ignore_protect_proc, |
aecd5ac6 TM |
2320 | }; |
2321 | *(m++) = (MountEntry) { | |
2322 | .path_const = "/proc/sys/kernel/domainname", | |
2323 | .mode = READONLY, | |
1361f015 | 2324 | .ignore = ignore_protect_proc, |
aecd5ac6 TM |
2325 | }; |
2326 | } | |
2327 | ||
57ac6959 | 2328 | if (ns_info->private_ipc) |
80271a44 XR |
2329 | *(m++) = (MountEntry) { |
2330 | .path_const = "/dev/mqueue", | |
2331 | .mode = MQUEUEFS, | |
2332 | .flags = MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, | |
2333 | }; | |
80271a44 | 2334 | |
bbb4e7f3 LP |
2335 | if (creds_path) { |
2336 | /* If our service has a credentials store configured, then bind that one in, but hide | |
2337 | * everything else. */ | |
2338 | ||
2339 | *(m++) = (MountEntry) { | |
2340 | .path_const = "/run/credentials", | |
2341 | .mode = TMPFS, | |
2342 | .read_only = true, | |
2343 | .options_const = "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST, | |
2344 | .flags = MS_NODEV|MS_STRICTATIME|MS_NOSUID|MS_NOEXEC, | |
2345 | }; | |
2346 | ||
2347 | *(m++) = (MountEntry) { | |
2348 | .path_const = creds_path, | |
2349 | .mode = BIND_MOUNT, | |
2350 | .read_only = true, | |
2351 | .source_const = creds_path, | |
2352 | }; | |
2353 | } else { | |
2354 | /* If our service has no credentials store configured, then make the whole | |
2355 | * credentials tree inaccessible wholesale. */ | |
2356 | ||
2357 | *(m++) = (MountEntry) { | |
2358 | .path_const = "/run/credentials", | |
2359 | .mode = INACCESSIBLE, | |
2360 | .ignore = true, | |
2361 | }; | |
2362 | } | |
2363 | ||
91dd5f7c | 2364 | if (log_namespace) { |
c2b2df60 | 2365 | _cleanup_free_ char *q = NULL; |
91dd5f7c LP |
2366 | |
2367 | q = strjoin("/run/systemd/journal.", log_namespace); | |
2368 | if (!q) { | |
2369 | r = -ENOMEM; | |
2370 | goto finish; | |
2371 | } | |
2372 | ||
2373 | *(m++) = (MountEntry) { | |
2374 | .path_const = "/run/systemd/journal", | |
2375 | .mode = BIND_MOUNT_RECURSIVE, | |
2376 | .read_only = true, | |
2377 | .source_malloc = TAKE_PTR(q), | |
2378 | }; | |
2379 | } | |
2380 | ||
5e8deb94 LB |
2381 | /* Will be used to add bind mounts at runtime */ |
2382 | if (setup_propagate) | |
2383 | *(m++) = (MountEntry) { | |
2384 | .source_const = propagate_dir, | |
2385 | .path_const = incoming_dir, | |
2386 | .mode = BIND_MOUNT, | |
2387 | .read_only = true, | |
2388 | }; | |
2389 | ||
3bdc25a4 LP |
2390 | if (notify_socket) |
2391 | *(m++) = (MountEntry) { | |
2392 | .path_const = notify_socket, | |
2393 | .source_const = notify_socket, | |
2394 | .mode = BIND_MOUNT, | |
2395 | .read_only = true, | |
2396 | }; | |
2397 | ||
f0a4feb0 | 2398 | assert(mounts + n_mounts == m); |
ac0930c8 | 2399 | |
5327c910 | 2400 | /* Prepend the root directory where that's necessary */ |
e908468b | 2401 | r = prefix_where_needed(mounts, n_mounts, root); |
5327c910 LP |
2402 | if (r < 0) |
2403 | goto finish; | |
2404 | ||
fbf90c0d | 2405 | drop_unused_mounts(root, mounts, &n_mounts); |
15ae422b LP |
2406 | } |
2407 | ||
1beab8b0 LP |
2408 | /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */ |
2409 | ||
d944dc95 | 2410 | if (unshare(CLONE_NEWNS) < 0) { |
763a260a | 2411 | r = log_debug_errno(errno, "Failed to unshare the mount namespace: %m"); |
1beab8b0 | 2412 | if (IN_SET(r, -EACCES, -EPERM, -EOPNOTSUPP, -ENOSYS)) |
cbc056c8 ZJS |
2413 | /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter |
2414 | * in place that doesn't allow us to create namespaces (or a missing cap), then | |
2415 | * propagate a recognizable error back, which the caller can use to detect this case | |
2416 | * (and only this) and optionally continue without namespacing applied. */ | |
1beab8b0 LP |
2417 | r = -ENOANO; |
2418 | ||
d944dc95 LP |
2419 | goto finish; |
2420 | } | |
1e4e94c8 | 2421 | |
5e8deb94 LB |
2422 | /* Create the source directory to allow runtime propagation of mounts */ |
2423 | if (setup_propagate) | |
2424 | (void) mkdir_p(propagate_dir, 0600); | |
2425 | ||
3ebf0b0b LB |
2426 | if (n_extension_images > 0 || !strv_isempty(extension_directories)) |
2427 | /* ExtensionImages/Directories mountpoint directories will be created while parsing the | |
2428 | * mounts to create, so have the parent ready */ | |
93f59701 | 2429 | (void) mkdir_p(extension_dir, 0600); |
93f59701 | 2430 | |
9b68367b YW |
2431 | /* Remount / as SLAVE so that nothing now mounted in the namespace |
2432 | * shows up in the parent */ | |
2433 | if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL) < 0) { | |
763a260a | 2434 | r = log_debug_errno(errno, "Failed to remount '/' as SLAVE: %m"); |
9b68367b | 2435 | goto finish; |
ee818b89 AC |
2436 | } |
2437 | ||
915e6d16 | 2438 | if (root_image) { |
e908468b | 2439 | /* A root image is specified, mount it to the right place */ |
21b61b1d | 2440 | r = dissected_image_mount(dissected_image, root, UID_INVALID, UID_INVALID, dissect_image_flags); |
763a260a YW |
2441 | if (r < 0) { |
2442 | log_debug_errno(r, "Failed to mount root image: %m"); | |
41bc4849 LP |
2443 | goto finish; |
2444 | } | |
2445 | ||
2446 | /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device | |
2447 | * if it likes. */ | |
2448 | r = loop_device_flock(loop_device, LOCK_UN); | |
2449 | if (r < 0) { | |
2450 | log_debug_errno(r, "Failed to release lock on loopback block device: %m"); | |
915e6d16 | 2451 | goto finish; |
763a260a | 2452 | } |
915e6d16 | 2453 | |
3044d343 YW |
2454 | r = dissected_image_relinquish(dissected_image); |
2455 | if (r < 0) { | |
2456 | log_debug_errno(r, "Failed to relinquish dissected image: %m"); | |
2457 | goto finish; | |
07ce7407 | 2458 | } |
78ebe980 | 2459 | |
915e6d16 LP |
2460 | } else if (root_directory) { |
2461 | ||
e908468b LP |
2462 | /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */ |
2463 | r = path_is_mount_point(root, NULL, AT_SYMLINK_FOLLOW); | |
763a260a YW |
2464 | if (r < 0) { |
2465 | log_debug_errno(r, "Failed to detect that %s is a mount point or not: %m", root); | |
d944dc95 | 2466 | goto finish; |
763a260a | 2467 | } |
8f1ad200 | 2468 | if (r == 0) { |
21935150 LP |
2469 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); |
2470 | if (r < 0) | |
8f1ad200 | 2471 | goto finish; |
d944dc95 | 2472 | } |
e908468b | 2473 | |
9b68367b | 2474 | } else { |
e908468b | 2475 | /* Let's mount the main root directory to the root directory to use */ |
21935150 LP |
2476 | r = mount_nofollow_verbose(LOG_DEBUG, "/", root, NULL, MS_BIND|MS_REC, NULL); |
2477 | if (r < 0) | |
e908468b | 2478 | goto finish; |
ee818b89 | 2479 | } |
c2c13f2d | 2480 | |
4e0c20de LP |
2481 | /* Try to set up the new root directory before mounting anything else there. */ |
2482 | if (root_image || root_directory) | |
2483 | (void) base_filesystem_create(root, UID_INVALID, GID_INVALID); | |
2484 | ||
82fb2da2 | 2485 | /* Now make the magic happen */ |
df61e79a | 2486 | r = apply_mounts(root, ns_info, mounts, &n_mounts, exec_dir_symlinks, error_path); |
82fb2da2 LB |
2487 | if (r < 0) |
2488 | goto finish; | |
15ae422b | 2489 | |
9b68367b | 2490 | /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */ |
9d50f850 | 2491 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
ea63a260 LB |
2492 | if (r == -EINVAL && root_directory) { |
2493 | /* If we are using root_directory and we don't have privileges (ie: user manager in a user | |
2494 | * namespace) and the root_directory is already a mount point in the parent namespace, | |
2495 | * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than | |
2496 | * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a | |
2497 | * mount point) and try again. */ | |
2498 | r = mount_nofollow_verbose(LOG_DEBUG, root, root, NULL, MS_BIND|MS_REC, NULL); | |
2499 | if (r < 0) | |
2500 | goto finish; | |
9d50f850 | 2501 | r = mount_switch_root(root, /* mount_propagation_flag = */ 0); |
ea63a260 | 2502 | } |
763a260a YW |
2503 | if (r < 0) { |
2504 | log_debug_errno(r, "Failed to mount root with MS_MOVE: %m"); | |
9b68367b | 2505 | goto finish; |
763a260a | 2506 | } |
ee818b89 | 2507 | |
55fe7432 | 2508 | /* Remount / as the desired mode. Note that this will not |
c2c13f2d LP |
2509 | * reestablish propagation from our side to the host, since |
2510 | * what's disconnected is disconnected. */ | |
d944dc95 | 2511 | if (mount(NULL, "/", NULL, mount_flags | MS_REC, NULL) < 0) { |
763a260a | 2512 | r = log_debug_errno(errno, "Failed to remount '/' with desired mount flags: %m"); |
d944dc95 LP |
2513 | goto finish; |
2514 | } | |
15ae422b | 2515 | |
5e8deb94 LB |
2516 | /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only |
2517 | * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */ | |
2518 | if (setup_propagate) { | |
2519 | r = mount(NULL, incoming_dir, NULL, MS_SLAVE, NULL); | |
2520 | if (r < 0) { | |
2521 | log_error_errno(r, "Failed to remount %s with MS_SLAVE: %m", incoming_dir); | |
2522 | goto finish; | |
2523 | } | |
2524 | } | |
2525 | ||
d944dc95 | 2526 | r = 0; |
15ae422b | 2527 | |
d944dc95 | 2528 | finish: |
0cd41757 LP |
2529 | if (n_mounts > 0) |
2530 | for (m = mounts; m < mounts + n_mounts; m++) | |
2531 | mount_entry_done(m); | |
613b411c | 2532 | |
5f7a690a LP |
2533 | free(mounts); |
2534 | ||
613b411c LP |
2535 | return r; |
2536 | } | |
2537 | ||
da6053d0 | 2538 | void bind_mount_free_many(BindMount *b, size_t n) { |
d2d6c096 LP |
2539 | assert(b || n == 0); |
2540 | ||
fe96c0f8 | 2541 | for (size_t i = 0; i < n; i++) { |
d2d6c096 LP |
2542 | free(b[i].source); |
2543 | free(b[i].destination); | |
2544 | } | |
2545 | ||
2546 | free(b); | |
2547 | } | |
2548 | ||
da6053d0 | 2549 | int bind_mount_add(BindMount **b, size_t *n, const BindMount *item) { |
d2d6c096 LP |
2550 | _cleanup_free_ char *s = NULL, *d = NULL; |
2551 | BindMount *c; | |
2552 | ||
2553 | assert(b); | |
2554 | assert(n); | |
2555 | assert(item); | |
2556 | ||
2557 | s = strdup(item->source); | |
2558 | if (!s) | |
2559 | return -ENOMEM; | |
2560 | ||
2561 | d = strdup(item->destination); | |
2562 | if (!d) | |
2563 | return -ENOMEM; | |
2564 | ||
aa484f35 | 2565 | c = reallocarray(*b, *n + 1, sizeof(BindMount)); |
d2d6c096 LP |
2566 | if (!c) |
2567 | return -ENOMEM; | |
2568 | ||
2569 | *b = c; | |
2570 | ||
2571 | c[(*n) ++] = (BindMount) { | |
1cc6c93a YW |
2572 | .source = TAKE_PTR(s), |
2573 | .destination = TAKE_PTR(d), | |
d2d6c096 | 2574 | .read_only = item->read_only, |
9ce4e4b0 | 2575 | .nosuid = item->nosuid, |
d2d6c096 LP |
2576 | .recursive = item->recursive, |
2577 | .ignore_enoent = item->ignore_enoent, | |
2578 | }; | |
2579 | ||
d2d6c096 LP |
2580 | return 0; |
2581 | } | |
2582 | ||
b3d13314 | 2583 | MountImage* mount_image_free_many(MountImage *m, size_t *n) { |
b3d13314 LB |
2584 | assert(n); |
2585 | assert(m || *n == 0); | |
2586 | ||
fe96c0f8 | 2587 | for (size_t i = 0; i < *n; i++) { |
b3d13314 LB |
2588 | free(m[i].source); |
2589 | free(m[i].destination); | |
427353f6 | 2590 | mount_options_free_all(m[i].mount_options); |
b3d13314 LB |
2591 | } |
2592 | ||
2593 | free(m); | |
2594 | *n = 0; | |
2595 | return NULL; | |
2596 | } | |
2597 | ||
2598 | int mount_image_add(MountImage **m, size_t *n, const MountImage *item) { | |
2599 | _cleanup_free_ char *s = NULL, *d = NULL; | |
427353f6 | 2600 | _cleanup_(mount_options_free_allp) MountOptions *options = NULL; |
b3d13314 LB |
2601 | MountImage *c; |
2602 | ||
2603 | assert(m); | |
2604 | assert(n); | |
2605 | assert(item); | |
2606 | ||
2607 | s = strdup(item->source); | |
2608 | if (!s) | |
2609 | return -ENOMEM; | |
2610 | ||
93f59701 LB |
2611 | if (item->destination) { |
2612 | d = strdup(item->destination); | |
2613 | if (!d) | |
2614 | return -ENOMEM; | |
2615 | } | |
b3d13314 | 2616 | |
427353f6 | 2617 | LIST_FOREACH(mount_options, i, item->mount_options) { |
c2b2df60 | 2618 | _cleanup_(mount_options_free_allp) MountOptions *o = NULL; |
427353f6 LB |
2619 | |
2620 | o = new(MountOptions, 1); | |
2621 | if (!o) | |
2622 | return -ENOMEM; | |
2623 | ||
2624 | *o = (MountOptions) { | |
2625 | .partition_designator = i->partition_designator, | |
2626 | .options = strdup(i->options), | |
2627 | }; | |
2628 | if (!o->options) | |
2629 | return -ENOMEM; | |
2630 | ||
2631 | LIST_APPEND(mount_options, options, TAKE_PTR(o)); | |
2632 | } | |
2633 | ||
b3d13314 LB |
2634 | c = reallocarray(*m, *n + 1, sizeof(MountImage)); |
2635 | if (!c) | |
2636 | return -ENOMEM; | |
2637 | ||
2638 | *m = c; | |
2639 | ||
2640 | c[(*n) ++] = (MountImage) { | |
2641 | .source = TAKE_PTR(s), | |
2642 | .destination = TAKE_PTR(d), | |
427353f6 | 2643 | .mount_options = TAKE_PTR(options), |
b3d13314 | 2644 | .ignore_enoent = item->ignore_enoent, |
93f59701 | 2645 | .type = item->type, |
b3d13314 LB |
2646 | }; |
2647 | ||
2648 | return 0; | |
2649 | } | |
2650 | ||
da6053d0 | 2651 | void temporary_filesystem_free_many(TemporaryFileSystem *t, size_t n) { |
2abd4e38 YW |
2652 | assert(t || n == 0); |
2653 | ||
fe96c0f8 | 2654 | for (size_t i = 0; i < n; i++) { |
2abd4e38 YW |
2655 | free(t[i].path); |
2656 | free(t[i].options); | |
2657 | } | |
2658 | ||
2659 | free(t); | |
2660 | } | |
2661 | ||
2662 | int temporary_filesystem_add( | |
2663 | TemporaryFileSystem **t, | |
da6053d0 | 2664 | size_t *n, |
2abd4e38 YW |
2665 | const char *path, |
2666 | const char *options) { | |
2667 | ||
2668 | _cleanup_free_ char *p = NULL, *o = NULL; | |
2669 | TemporaryFileSystem *c; | |
2670 | ||
2671 | assert(t); | |
2672 | assert(n); | |
2673 | assert(path); | |
2674 | ||
2675 | p = strdup(path); | |
2676 | if (!p) | |
2677 | return -ENOMEM; | |
2678 | ||
2679 | if (!isempty(options)) { | |
2680 | o = strdup(options); | |
2681 | if (!o) | |
2682 | return -ENOMEM; | |
2683 | } | |
2684 | ||
aa484f35 | 2685 | c = reallocarray(*t, *n + 1, sizeof(TemporaryFileSystem)); |
2abd4e38 YW |
2686 | if (!c) |
2687 | return -ENOMEM; | |
2688 | ||
2689 | *t = c; | |
2690 | ||
2691 | c[(*n) ++] = (TemporaryFileSystem) { | |
1cc6c93a YW |
2692 | .path = TAKE_PTR(p), |
2693 | .options = TAKE_PTR(o), | |
2abd4e38 YW |
2694 | }; |
2695 | ||
2abd4e38 YW |
2696 | return 0; |
2697 | } | |
2698 | ||
a652f050 JR |
2699 | static int make_tmp_prefix(const char *prefix) { |
2700 | _cleanup_free_ char *t = NULL; | |
254d1313 | 2701 | _cleanup_close_ int fd = -EBADF; |
a652f050 JR |
2702 | int r; |
2703 | ||
2704 | /* Don't do anything unless we know the dir is actually missing */ | |
2705 | r = access(prefix, F_OK); | |
2706 | if (r >= 0) | |
2707 | return 0; | |
2708 | if (errno != ENOENT) | |
2709 | return -errno; | |
2710 | ||
2053593f | 2711 | WITH_UMASK(000) |
30443439 | 2712 | r = mkdir_parents(prefix, 0755); |
a652f050 JR |
2713 | if (r < 0) |
2714 | return r; | |
2715 | ||
2716 | r = tempfn_random(prefix, NULL, &t); | |
2717 | if (r < 0) | |
2718 | return r; | |
2719 | ||
96603ea0 LP |
2720 | /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for |
2721 | * the suid bit, below. */ | |
2722 | fd = open_mkdir_at(AT_FDCWD, t, O_EXCL|O_CLOEXEC, 0777); | |
2723 | if (fd < 0) | |
2724 | return fd; | |
a652f050 | 2725 | |
96603ea0 LP |
2726 | r = RET_NERRNO(fchmod(fd, 01777)); |
2727 | if (r < 0) { | |
a652f050 JR |
2728 | (void) rmdir(t); |
2729 | return r; | |
2730 | } | |
2731 | ||
96603ea0 LP |
2732 | r = RET_NERRNO(rename(t, prefix)); |
2733 | if (r < 0) { | |
a652f050 JR |
2734 | (void) rmdir(t); |
2735 | return r == -EEXIST ? 0 : r; /* it's fine if someone else created the dir by now */ | |
2736 | } | |
2737 | ||
2738 | return 0; | |
2739 | ||
2740 | } | |
2741 | ||
56a13a49 | 2742 | static int setup_one_tmp_dir(const char *id, const char *prefix, char **path, char **tmp_path) { |
613b411c | 2743 | _cleanup_free_ char *x = NULL; |
19cd4e19 | 2744 | _cleanup_free_ char *y = NULL; |
6b46ea73 | 2745 | sd_id128_t boot_id; |
56a13a49 | 2746 | bool rw = true; |
6b46ea73 | 2747 | int r; |
613b411c LP |
2748 | |
2749 | assert(id); | |
2750 | assert(prefix); | |
2751 | assert(path); | |
2752 | ||
6b46ea73 LP |
2753 | /* We include the boot id in the directory so that after a |
2754 | * reboot we can easily identify obsolete directories. */ | |
2755 | ||
2756 | r = sd_id128_get_boot(&boot_id); | |
2757 | if (r < 0) | |
2758 | return r; | |
2759 | ||
85b55869 | 2760 | x = strjoin(prefix, "/systemd-private-", SD_ID128_TO_STRING(boot_id), "-", id, "-XXXXXX"); |
613b411c LP |
2761 | if (!x) |
2762 | return -ENOMEM; | |
2763 | ||
a652f050 JR |
2764 | r = make_tmp_prefix(prefix); |
2765 | if (r < 0) | |
2766 | return r; | |
2767 | ||
2053593f | 2768 | WITH_UMASK(0077) |
56a13a49 ZJS |
2769 | if (!mkdtemp(x)) { |
2770 | if (errno == EROFS || ERRNO_IS_DISK_SPACE(errno)) | |
2771 | rw = false; | |
2772 | else | |
2773 | return -errno; | |
2774 | } | |
613b411c | 2775 | |
56a13a49 | 2776 | if (rw) { |
19cd4e19 | 2777 | y = strjoin(x, "/tmp"); |
2778 | if (!y) | |
2779 | return -ENOMEM; | |
2780 | ||
2053593f | 2781 | WITH_UMASK(0000) |
19cd4e19 | 2782 | if (mkdir(y, 0777 | S_ISVTX) < 0) |
7c76e181 | 2783 | return -errno; |
19cd4e19 | 2784 | |
03bc11d1 | 2785 | r = label_fix_full(AT_FDCWD, y, prefix, 0); |
56a13a49 ZJS |
2786 | if (r < 0) |
2787 | return r; | |
19cd4e19 | 2788 | |
2789 | if (tmp_path) | |
2790 | *tmp_path = TAKE_PTR(y); | |
56a13a49 ZJS |
2791 | } else { |
2792 | /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being | |
2793 | * read-only. This way the service will get the EROFS result as if it was writing to the real | |
2794 | * file system. */ | |
2053593f | 2795 | WITH_UMASK(0000) |
30443439 | 2796 | r = mkdir_p(RUN_SYSTEMD_EMPTY, 0500); |
56a13a49 ZJS |
2797 | if (r < 0) |
2798 | return r; | |
613b411c | 2799 | |
3f181262 LP |
2800 | r = free_and_strdup(&x, RUN_SYSTEMD_EMPTY); |
2801 | if (r < 0) | |
2802 | return r; | |
c17ec25e | 2803 | } |
15ae422b | 2804 | |
1cc6c93a | 2805 | *path = TAKE_PTR(x); |
613b411c LP |
2806 | return 0; |
2807 | } | |
2808 | ||
2809 | int setup_tmp_dirs(const char *id, char **tmp_dir, char **var_tmp_dir) { | |
56a13a49 ZJS |
2810 | _cleanup_(namespace_cleanup_tmpdirp) char *a = NULL; |
2811 | _cleanup_(rmdir_and_freep) char *a_tmp = NULL; | |
2812 | char *b; | |
613b411c LP |
2813 | int r; |
2814 | ||
2815 | assert(id); | |
2816 | assert(tmp_dir); | |
2817 | assert(var_tmp_dir); | |
2818 | ||
56a13a49 | 2819 | r = setup_one_tmp_dir(id, "/tmp", &a, &a_tmp); |
613b411c LP |
2820 | if (r < 0) |
2821 | return r; | |
2822 | ||
56a13a49 ZJS |
2823 | r = setup_one_tmp_dir(id, "/var/tmp", &b, NULL); |
2824 | if (r < 0) | |
613b411c | 2825 | return r; |
613b411c | 2826 | |
56a13a49 ZJS |
2827 | a_tmp = mfree(a_tmp); /* avoid rmdir */ |
2828 | *tmp_dir = TAKE_PTR(a); | |
2829 | *var_tmp_dir = TAKE_PTR(b); | |
613b411c LP |
2830 | |
2831 | return 0; | |
2832 | } | |
2833 | ||
54c2459d XR |
2834 | int setup_shareable_ns(const int ns_storage_socket[static 2], unsigned long nsflag) { |
2835 | _cleanup_close_ int ns = -1; | |
3ee897d6 | 2836 | int r, q; |
54c2459d | 2837 | const char *ns_name, *ns_path; |
613b411c | 2838 | |
54c2459d XR |
2839 | assert(ns_storage_socket); |
2840 | assert(ns_storage_socket[0] >= 0); | |
2841 | assert(ns_storage_socket[1] >= 0); | |
2842 | ||
2843 | ns_name = namespace_single_flag_to_string(nsflag); | |
2844 | assert(ns_name); | |
613b411c LP |
2845 | |
2846 | /* We use the passed socketpair as a storage buffer for our | |
76cd584b LP |
2847 | * namespace reference fd. Whatever process runs this first |
2848 | * shall create a new namespace, all others should just join | |
2849 | * it. To serialize that we use a file lock on the socket | |
2850 | * pair. | |
613b411c LP |
2851 | * |
2852 | * It's a bit crazy, but hey, works great! */ | |
2853 | ||
54c2459d | 2854 | if (lockf(ns_storage_socket[0], F_LOCK, 0) < 0) |
613b411c LP |
2855 | return -errno; |
2856 | ||
54c2459d XR |
2857 | ns = receive_one_fd(ns_storage_socket[0], MSG_DONTWAIT); |
2858 | if (ns == -EAGAIN) { | |
44ffcbae | 2859 | /* Nothing stored yet, so let's create a new namespace. */ |
613b411c | 2860 | |
54c2459d | 2861 | if (unshare(nsflag) < 0) { |
613b411c LP |
2862 | r = -errno; |
2863 | goto fail; | |
2864 | } | |
2865 | ||
44ffcbae | 2866 | (void) loopback_setup(); |
613b411c | 2867 | |
54c2459d XR |
2868 | ns_path = strjoina("/proc/self/ns/", ns_name); |
2869 | ns = open(ns_path, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
2870 | if (ns < 0) { | |
613b411c LP |
2871 | r = -errno; |
2872 | goto fail; | |
2873 | } | |
2874 | ||
2875 | r = 1; | |
613b411c | 2876 | |
54c2459d XR |
2877 | } else if (ns < 0) { |
2878 | r = ns; | |
3ee897d6 | 2879 | goto fail; |
613b411c | 2880 | |
3ee897d6 LP |
2881 | } else { |
2882 | /* Yay, found something, so let's join the namespace */ | |
54c2459d | 2883 | if (setns(ns, nsflag) < 0) { |
613b411c LP |
2884 | r = -errno; |
2885 | goto fail; | |
2886 | } | |
2887 | ||
2888 | r = 0; | |
2889 | } | |
2890 | ||
54c2459d | 2891 | q = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
3ee897d6 LP |
2892 | if (q < 0) { |
2893 | r = q; | |
613b411c LP |
2894 | goto fail; |
2895 | } | |
2896 | ||
2897 | fail: | |
54c2459d | 2898 | (void) lockf(ns_storage_socket[0], F_ULOCK, 0); |
15ae422b LP |
2899 | return r; |
2900 | } | |
417116f2 | 2901 | |
54c2459d XR |
2902 | int open_shareable_ns_path(const int ns_storage_socket[static 2], const char *path, unsigned long nsflag) { |
2903 | _cleanup_close_ int ns = -1; | |
51af7fb2 LP |
2904 | int q, r; |
2905 | ||
54c2459d XR |
2906 | assert(ns_storage_socket); |
2907 | assert(ns_storage_socket[0] >= 0); | |
2908 | assert(ns_storage_socket[1] >= 0); | |
51af7fb2 LP |
2909 | assert(path); |
2910 | ||
54c2459d XR |
2911 | /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in |
2912 | * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will | |
2913 | * allocate a new anonymous ns if needed. */ | |
51af7fb2 | 2914 | |
54c2459d | 2915 | if (lockf(ns_storage_socket[0], F_LOCK, 0) < 0) |
51af7fb2 LP |
2916 | return -errno; |
2917 | ||
54c2459d XR |
2918 | ns = receive_one_fd(ns_storage_socket[0], MSG_DONTWAIT); |
2919 | if (ns == -EAGAIN) { | |
51af7fb2 LP |
2920 | /* Nothing stored yet. Open the file from the file system. */ |
2921 | ||
54c2459d XR |
2922 | ns = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC); |
2923 | if (ns < 0) { | |
51af7fb2 LP |
2924 | r = -errno; |
2925 | goto fail; | |
2926 | } | |
2927 | ||
54c2459d XR |
2928 | r = fd_is_ns(ns, nsflag); |
2929 | if (r == 0) { /* Not a ns of our type? Refuse early. */ | |
51af7fb2 LP |
2930 | r = -EINVAL; |
2931 | goto fail; | |
2932 | } | |
2933 | if (r < 0 && r != -EUCLEAN) /* EUCLEAN: we don't know */ | |
2934 | goto fail; | |
2935 | ||
2936 | r = 1; | |
2937 | ||
54c2459d XR |
2938 | } else if (ns < 0) { |
2939 | r = ns; | |
51af7fb2 LP |
2940 | goto fail; |
2941 | } else | |
2942 | r = 0; /* Already allocated */ | |
2943 | ||
54c2459d | 2944 | q = send_one_fd(ns_storage_socket[1], ns, MSG_DONTWAIT); |
51af7fb2 LP |
2945 | if (q < 0) { |
2946 | r = q; | |
2947 | goto fail; | |
2948 | } | |
2949 | ||
2950 | fail: | |
54c2459d | 2951 | (void) lockf(ns_storage_socket[0], F_ULOCK, 0); |
51af7fb2 LP |
2952 | return r; |
2953 | } | |
2954 | ||
6e2d7c4f MS |
2955 | bool ns_type_supported(NamespaceType type) { |
2956 | const char *t, *ns_proc; | |
2957 | ||
0fa5b831 LP |
2958 | t = namespace_type_to_string(type); |
2959 | if (!t) /* Don't know how to translate this? Then it's not supported */ | |
6e2d7c4f MS |
2960 | return false; |
2961 | ||
6e2d7c4f | 2962 | ns_proc = strjoina("/proc/self/ns/", t); |
6e2d7c4f MS |
2963 | return access(ns_proc, F_OK) == 0; |
2964 | } | |
2965 | ||
1b8689f9 | 2966 | static const char *const protect_home_table[_PROTECT_HOME_MAX] = { |
cbc056c8 ZJS |
2967 | [PROTECT_HOME_NO] = "no", |
2968 | [PROTECT_HOME_YES] = "yes", | |
1b8689f9 | 2969 | [PROTECT_HOME_READ_ONLY] = "read-only", |
cbc056c8 | 2970 | [PROTECT_HOME_TMPFS] = "tmpfs", |
417116f2 LP |
2971 | }; |
2972 | ||
1e8c7bd5 | 2973 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home, ProtectHome, PROTECT_HOME_YES); |
5e1c6154 | 2974 | |
1b8689f9 | 2975 | static const char *const protect_system_table[_PROTECT_SYSTEM_MAX] = { |
cbc056c8 ZJS |
2976 | [PROTECT_SYSTEM_NO] = "no", |
2977 | [PROTECT_SYSTEM_YES] = "yes", | |
2978 | [PROTECT_SYSTEM_FULL] = "full", | |
3f815163 | 2979 | [PROTECT_SYSTEM_STRICT] = "strict", |
1b8689f9 LP |
2980 | }; |
2981 | ||
1e8c7bd5 | 2982 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system, ProtectSystem, PROTECT_SYSTEM_YES); |
03c791aa | 2983 | |
6e2d7c4f | 2984 | static const char* const namespace_type_table[] = { |
cbc056c8 | 2985 | [NAMESPACE_MOUNT] = "mnt", |
6e2d7c4f | 2986 | [NAMESPACE_CGROUP] = "cgroup", |
cbc056c8 ZJS |
2987 | [NAMESPACE_UTS] = "uts", |
2988 | [NAMESPACE_IPC] = "ipc", | |
2989 | [NAMESPACE_USER] = "user", | |
2990 | [NAMESPACE_PID] = "pid", | |
2991 | [NAMESPACE_NET] = "net", | |
c3b9c418 | 2992 | [NAMESPACE_TIME] = "time", |
6e2d7c4f MS |
2993 | }; |
2994 | ||
2995 | DEFINE_STRING_TABLE_LOOKUP(namespace_type, NamespaceType); | |
4e399953 LP |
2996 | |
2997 | static const char* const protect_proc_table[_PROTECT_PROC_MAX] = { | |
2998 | [PROTECT_PROC_DEFAULT] = "default", | |
2999 | [PROTECT_PROC_NOACCESS] = "noaccess", | |
3000 | [PROTECT_PROC_INVISIBLE] = "invisible", | |
3001 | [PROTECT_PROC_PTRACEABLE] = "ptraceable", | |
3002 | }; | |
3003 | ||
3004 | DEFINE_STRING_TABLE_LOOKUP(protect_proc, ProtectProc); | |
3005 | ||
3006 | static const char* const proc_subset_table[_PROC_SUBSET_MAX] = { | |
3007 | [PROC_SUBSET_ALL] = "all", | |
3008 | [PROC_SUBSET_PID] = "pid", | |
3009 | }; | |
3010 | ||
3011 | DEFINE_STRING_TABLE_LOOKUP(proc_subset, ProcSubset); |