]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
e83bebef | 2 | |
4f5dd394 | 3 | #include <sys/mount.h> |
07630cea | 4 | #include <linux/magic.h> |
e83bebef | 5 | |
b5efdb8a | 6 | #include "alloc-util.h" |
f461a28d | 7 | #include "chase.h" |
4f5dd394 | 8 | #include "escape.h" |
0996ef00 | 9 | #include "fd-util.h" |
ca78ad1d | 10 | #include "format-util.h" |
f4f15635 | 11 | #include "fs-util.h" |
0690160e | 12 | #include "label-util.h" |
35cd0ba5 | 13 | #include "mkdir-label.h" |
4349cd7c | 14 | #include "mount-util.h" |
049af8ad | 15 | #include "mountpoint-util.h" |
b71a0192 | 16 | #include "namespace-util.h" |
6bedfcbb LP |
17 | #include "nspawn-mount.h" |
18 | #include "parse-util.h" | |
4f5dd394 LP |
19 | #include "path-util.h" |
20 | #include "rm-rf.h" | |
e83bebef | 21 | #include "set.h" |
760877e9 | 22 | #include "sort-util.h" |
8fcde012 | 23 | #include "stat-util.h" |
07630cea | 24 | #include "string-util.h" |
4f5dd394 | 25 | #include "strv.h" |
e4de7287 | 26 | #include "tmpfile-util.h" |
ee104e11 | 27 | #include "user-util.h" |
e83bebef | 28 | |
88614c8a | 29 | CustomMount* custom_mount_add(CustomMount **l, size_t *n, CustomMountType t) { |
e83bebef LP |
30 | CustomMount *c, *ret; |
31 | ||
32 | assert(l); | |
33 | assert(n); | |
34 | assert(t >= 0); | |
35 | assert(t < _CUSTOM_MOUNT_TYPE_MAX); | |
36 | ||
aa484f35 | 37 | c = reallocarray(*l, *n + 1, sizeof(CustomMount)); |
e83bebef LP |
38 | if (!c) |
39 | return NULL; | |
40 | ||
41 | *l = c; | |
42 | ret = *l + *n; | |
43 | (*n)++; | |
44 | ||
511a8cfe LP |
45 | *ret = (CustomMount) { |
46 | .type = t | |
47 | }; | |
e83bebef LP |
48 | |
49 | return ret; | |
50 | } | |
51 | ||
88614c8a | 52 | void custom_mount_free_all(CustomMount *l, size_t n) { |
d2b99ed7 | 53 | for (size_t i = 0; i < n; i++) { |
e83bebef LP |
54 | CustomMount *m = l + i; |
55 | ||
56 | free(m->source); | |
57 | free(m->destination); | |
58 | free(m->options); | |
59 | ||
60 | if (m->work_dir) { | |
61 | (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL); | |
62 | free(m->work_dir); | |
63 | } | |
64 | ||
c7a4890c LP |
65 | if (m->rm_rf_tmpdir) { |
66 | (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL); | |
67 | free(m->rm_rf_tmpdir); | |
68 | } | |
69 | ||
e83bebef | 70 | strv_free(m->lower); |
de40a303 | 71 | free(m->type_argument); |
e83bebef LP |
72 | } |
73 | ||
74 | free(l); | |
75 | } | |
76 | ||
93bab288 | 77 | static int custom_mount_compare(const CustomMount *a, const CustomMount *b) { |
e83bebef LP |
78 | int r; |
79 | ||
93bab288 | 80 | r = path_compare(a->destination, b->destination); |
e83bebef LP |
81 | if (r != 0) |
82 | return r; | |
83 | ||
93bab288 | 84 | return CMP(a->type, b->type); |
e83bebef LP |
85 | } |
86 | ||
448f7377 | 87 | static int source_path_parse(const char *p, char **ret) { |
86c0dd4a | 88 | assert(p); |
448f7377 | 89 | assert(ret); |
86c0dd4a | 90 | |
448f7377 DDM |
91 | if (isempty(p)) |
92 | return -EINVAL; | |
93 | ||
94 | if (*p == '+') { | |
95 | if (!path_is_absolute(p + 1)) | |
96 | return -EINVAL; | |
97 | ||
98 | char *s = strdup(p); | |
99 | if (!s) | |
100 | return -ENOMEM; | |
101 | ||
102 | *ret = TAKE_PTR(s); | |
103 | return 0; | |
104 | } | |
105 | ||
106 | return path_make_absolute_cwd(p, ret); | |
107 | } | |
108 | ||
109 | static int source_path_parse_nullable(const char *p, char **ret) { | |
110 | assert(p); | |
111 | assert(ret); | |
112 | ||
113 | if (isempty(p)) { | |
114 | *ret = NULL; | |
115 | return 0; | |
116 | } | |
86c0dd4a | 117 | |
448f7377 | 118 | return source_path_parse(p, ret); |
86c0dd4a LP |
119 | } |
120 | ||
121 | static char *resolve_source_path(const char *dest, const char *source) { | |
86c0dd4a LP |
122 | if (!source) |
123 | return NULL; | |
124 | ||
125 | if (source[0] == '+') | |
c6134d3e | 126 | return path_join(dest, source + 1); |
86c0dd4a LP |
127 | |
128 | return strdup(source); | |
129 | } | |
130 | ||
d0556c55 LP |
131 | static int allocate_temporary_source(CustomMount *m) { |
132 | assert(m); | |
133 | assert(!m->source); | |
134 | assert(!m->rm_rf_tmpdir); | |
135 | ||
136 | m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX"); | |
137 | if (!m->rm_rf_tmpdir) | |
138 | return log_oom(); | |
139 | ||
140 | if (!mkdtemp(m->rm_rf_tmpdir)) { | |
141 | m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir); | |
142 | return log_error_errno(errno, "Failed to acquire temporary directory: %m"); | |
143 | } | |
144 | ||
145 | m->source = path_join(m->rm_rf_tmpdir, "src"); | |
146 | if (!m->source) | |
147 | return log_oom(); | |
148 | ||
149 | if (mkdir(m->source, 0755) < 0) | |
150 | return log_error_errno(errno, "Failed to create %s: %m", m->source); | |
151 | ||
152 | return 0; | |
153 | } | |
154 | ||
88614c8a | 155 | int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n) { |
86c0dd4a LP |
156 | int r; |
157 | ||
158 | /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the | |
159 | * parent process, so that we know the temporary directories to remove on exit before we fork off the | |
160 | * children. */ | |
161 | ||
162 | assert(l || n == 0); | |
163 | ||
164 | /* Order the custom mounts, and make sure we have a working directory */ | |
93bab288 | 165 | typesafe_qsort(l, n, custom_mount_compare); |
86c0dd4a | 166 | |
d2b99ed7 | 167 | for (size_t i = 0; i < n; i++) { |
86c0dd4a LP |
168 | CustomMount *m = l + i; |
169 | ||
de40a303 LP |
170 | /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount |
171 | * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in | |
172 | * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in | |
173 | * the inner child, not the outer one. Determine this here. */ | |
174 | m->in_userns = path_startswith(m->destination, "/proc"); | |
86c0dd4a | 175 | |
de40a303 LP |
176 | if (m->type == CUSTOM_MOUNT_BIND) { |
177 | if (m->source) { | |
178 | char *s; | |
86c0dd4a | 179 | |
de40a303 LP |
180 | s = resolve_source_path(dest, m->source); |
181 | if (!s) | |
182 | return log_oom(); | |
c7a4890c | 183 | |
de40a303 LP |
184 | free_and_replace(m->source, s); |
185 | } else { | |
186 | /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */ | |
c7a4890c | 187 | |
d0556c55 LP |
188 | r = allocate_temporary_source(m); |
189 | if (r < 0) | |
190 | return r; | |
de40a303 | 191 | } |
86c0dd4a LP |
192 | } |
193 | ||
194 | if (m->type == CUSTOM_MOUNT_OVERLAY) { | |
86c0dd4a LP |
195 | STRV_FOREACH(j, m->lower) { |
196 | char *s; | |
197 | ||
198 | s = resolve_source_path(dest, *j); | |
199 | if (!s) | |
200 | return log_oom(); | |
201 | ||
10af01a5 | 202 | free_and_replace(*j, s); |
86c0dd4a LP |
203 | } |
204 | ||
d0556c55 LP |
205 | if (m->source) { |
206 | char *s; | |
207 | ||
208 | s = resolve_source_path(dest, m->source); | |
209 | if (!s) | |
210 | return log_oom(); | |
211 | ||
212 | free_and_replace(m->source, s); | |
213 | } else { | |
214 | r = allocate_temporary_source(m); | |
215 | if (r < 0) | |
216 | return r; | |
217 | } | |
218 | ||
86c0dd4a LP |
219 | if (m->work_dir) { |
220 | char *s; | |
221 | ||
222 | s = resolve_source_path(dest, m->work_dir); | |
223 | if (!s) | |
224 | return log_oom(); | |
225 | ||
10af01a5 | 226 | free_and_replace(m->work_dir, s); |
86c0dd4a | 227 | } else { |
86c0dd4a LP |
228 | r = tempfn_random(m->source, NULL, &m->work_dir); |
229 | if (r < 0) | |
230 | return log_error_errno(r, "Failed to acquire working directory: %m"); | |
231 | } | |
232 | ||
233 | (void) mkdir_label(m->work_dir, 0700); | |
234 | } | |
235 | } | |
236 | ||
237 | return 0; | |
238 | } | |
239 | ||
88614c8a | 240 | int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) { |
448f7377 | 241 | _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL, *p = NULL; |
e83bebef LP |
242 | CustomMount *m; |
243 | int r; | |
244 | ||
245 | assert(l); | |
246 | assert(n); | |
247 | ||
a50947c5 | 248 | r = extract_many_words(&s, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL); |
e83bebef LP |
249 | if (r < 0) |
250 | return r; | |
251 | if (r == 0) | |
252 | return -EINVAL; | |
e83bebef | 253 | if (r == 1) { |
86c0dd4a | 254 | destination = strdup(source[0] == '+' ? source+1 : source); |
e83bebef LP |
255 | if (!destination) |
256 | return -ENOMEM; | |
257 | } | |
a50947c5 DDM |
258 | if (r == 2 && !isempty(s)) { |
259 | opts = strdup(s); | |
e83bebef LP |
260 | if (!opts) |
261 | return -ENOMEM; | |
262 | } | |
263 | ||
448f7377 DDM |
264 | r = source_path_parse_nullable(source, &p); |
265 | if (r < 0) | |
266 | return r; | |
c7a4890c | 267 | |
e83bebef LP |
268 | if (!path_is_absolute(destination)) |
269 | return -EINVAL; | |
270 | ||
271 | m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND); | |
272 | if (!m) | |
48cbe5f8 | 273 | return -ENOMEM; |
e83bebef | 274 | |
448f7377 | 275 | m->source = TAKE_PTR(p); |
0e636bf5 | 276 | m->destination = TAKE_PTR(destination); |
e83bebef | 277 | m->read_only = read_only; |
0e636bf5 | 278 | m->options = TAKE_PTR(opts); |
de40a303 | 279 | |
e83bebef LP |
280 | return 0; |
281 | } | |
282 | ||
88614c8a | 283 | int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s) { |
e83bebef | 284 | _cleanup_free_ char *path = NULL, *opts = NULL; |
99534007 | 285 | const char *p = ASSERT_PTR(s); |
e83bebef LP |
286 | CustomMount *m; |
287 | int r; | |
288 | ||
289 | assert(l); | |
290 | assert(n); | |
e83bebef LP |
291 | |
292 | r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS); | |
293 | if (r < 0) | |
294 | return r; | |
295 | if (r == 0) | |
296 | return -EINVAL; | |
297 | ||
298 | if (isempty(p)) | |
299 | opts = strdup("mode=0755"); | |
300 | else | |
301 | opts = strdup(p); | |
302 | if (!opts) | |
303 | return -ENOMEM; | |
304 | ||
305 | if (!path_is_absolute(path)) | |
306 | return -EINVAL; | |
307 | ||
308 | m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS); | |
309 | if (!m) | |
310 | return -ENOMEM; | |
311 | ||
1cc6c93a YW |
312 | m->destination = TAKE_PTR(path); |
313 | m->options = TAKE_PTR(opts); | |
e83bebef | 314 | |
e83bebef LP |
315 | return 0; |
316 | } | |
317 | ||
88614c8a | 318 | int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) { |
ad85779a LP |
319 | _cleanup_free_ char *upper = NULL, *destination = NULL; |
320 | _cleanup_strv_free_ char **lower = NULL; | |
321 | CustomMount *m; | |
448f7377 | 322 | int r, k; |
ad85779a | 323 | |
90e30d76 | 324 | k = strv_split_full(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS); |
86c0dd4a LP |
325 | if (k < 0) |
326 | return k; | |
ad85779a LP |
327 | if (k < 2) |
328 | return -EADDRNOTAVAIL; | |
329 | if (k == 2) { | |
448f7377 DDM |
330 | _cleanup_free_ char *p = NULL; |
331 | ||
86c0dd4a LP |
332 | /* If two parameters are specified, the first one is the lower, the second one the upper directory. And |
333 | * we'll also define the destination mount point the same as the upper. */ | |
334 | ||
448f7377 DDM |
335 | r = source_path_parse(lower[0], &p); |
336 | if (r < 0) | |
337 | return r; | |
338 | ||
339 | free_and_replace(lower[0], p); | |
340 | ||
341 | r = source_path_parse(lower[1], &p); | |
342 | if (r < 0) | |
343 | return r; | |
344 | ||
345 | free_and_replace(lower[1], p); | |
86c0dd4a | 346 | |
ae2a15bc | 347 | upper = TAKE_PTR(lower[1]); |
ad85779a | 348 | |
86c0dd4a | 349 | destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */ |
ad85779a LP |
350 | if (!destination) |
351 | return -ENOMEM; | |
ad85779a | 352 | } else { |
448f7377 DDM |
353 | _cleanup_free_ char *p = NULL; |
354 | ||
86c0dd4a LP |
355 | /* If more than two parameters are specified, the last one is the destination, the second to last one |
356 | * the "upper", and all before that the "lower" directories. */ | |
357 | ||
ad85779a | 358 | destination = lower[k - 1]; |
ae2a15bc | 359 | upper = TAKE_PTR(lower[k - 2]); |
86c0dd4a | 360 | |
448f7377 DDM |
361 | STRV_FOREACH(i, lower) { |
362 | r = source_path_parse(*i, &p); | |
363 | if (r < 0) | |
364 | return r; | |
365 | ||
366 | free_and_replace(*i, p); | |
367 | } | |
c7a4890c LP |
368 | |
369 | /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory | |
370 | * in /var/tmp */ | |
448f7377 DDM |
371 | r = source_path_parse_nullable(upper, &p); |
372 | if (r < 0) | |
373 | return r; | |
374 | ||
375 | free_and_replace(upper, p); | |
c7a4890c | 376 | |
86c0dd4a LP |
377 | if (!path_is_absolute(destination)) |
378 | return -EINVAL; | |
ad85779a LP |
379 | } |
380 | ||
381 | m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY); | |
382 | if (!m) | |
383 | return -ENOMEM; | |
384 | ||
1cc6c93a YW |
385 | m->destination = TAKE_PTR(destination); |
386 | m->source = TAKE_PTR(upper); | |
387 | m->lower = TAKE_PTR(lower); | |
ad85779a LP |
388 | m->read_only = read_only; |
389 | ||
ad85779a LP |
390 | return 0; |
391 | } | |
392 | ||
de40a303 LP |
393 | int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s) { |
394 | _cleanup_free_ char *path = NULL; | |
395 | CustomMount *m; | |
396 | ||
397 | assert(l); | |
398 | assert(n); | |
399 | assert(s); | |
400 | ||
401 | if (!path_is_absolute(s)) | |
402 | return -EINVAL; | |
403 | ||
404 | path = strdup(s); | |
405 | if (!path) | |
406 | return -ENOMEM; | |
407 | ||
408 | m = custom_mount_add(l, n, CUSTOM_MOUNT_INACCESSIBLE); | |
409 | if (!m) | |
410 | return -ENOMEM; | |
411 | ||
412 | m->destination = TAKE_PTR(path); | |
413 | return 0; | |
414 | } | |
415 | ||
04029482 | 416 | int tmpfs_patch_options( |
e83bebef | 417 | const char *options, |
2fa017f1 | 418 | uid_t uid_shift, |
e83bebef LP |
419 | const char *selinux_apifs_context, |
420 | char **ret) { | |
421 | ||
cfea7618 | 422 | _cleanup_free_ char *buf = NULL; |
e83bebef | 423 | |
cfea7618 | 424 | assert(ret); |
e83bebef | 425 | |
cfea7618 YW |
426 | if (options) { |
427 | buf = strdup(options); | |
428 | if (!buf) | |
429 | return -ENOMEM; | |
e83bebef LP |
430 | } |
431 | ||
cfea7618 YW |
432 | if (uid_shift != UID_INVALID) |
433 | if (strextendf_with_separator(&buf, ",", "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift) < 0) | |
e83bebef | 434 | return -ENOMEM; |
e83bebef | 435 | |
cfea7618 YW |
436 | #if HAVE_SELINUX |
437 | if (selinux_apifs_context) | |
d3d15594 | 438 | if (strextendf_with_separator(&buf, ",", "context=\"%s\"", selinux_apifs_context) < 0) |
0996ef00 | 439 | return -ENOMEM; |
cfea7618 | 440 | #endif |
0996ef00 | 441 | |
cfea7618 YW |
442 | *ret = TAKE_PTR(buf); |
443 | return !!*ret; | |
e83bebef LP |
444 | } |
445 | ||
4f086aab | 446 | int mount_sysfs(const char *dest, MountSettingsMask mount_settings) { |
5980d463 | 447 | const char *full, *top; |
d1678248 | 448 | int r; |
4f086aab | 449 | unsigned long extra_flags = 0; |
d8fc6a00 LP |
450 | |
451 | top = prefix_roota(dest, "/sys"); | |
40fd52f2 | 452 | r = path_is_fs_type(top, SYSFS_MAGIC); |
d1678248 ILG |
453 | if (r < 0) |
454 | return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top); | |
455 | /* /sys might already be mounted as sysfs by the outer child in the | |
456 | * !netns case. In this case, it's all good. Don't touch it because we | |
457 | * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555. | |
458 | */ | |
459 | if (r > 0) | |
460 | return 0; | |
461 | ||
d8fc6a00 LP |
462 | full = prefix_roota(top, "/full"); |
463 | ||
464 | (void) mkdir(full, 0755); | |
465 | ||
bd6609eb | 466 | if (FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO)) |
4f086aab SU |
467 | extra_flags |= MS_RDONLY; |
468 | ||
511a8cfe LP |
469 | r = mount_nofollow_verbose(LOG_ERR, "sysfs", full, "sysfs", |
470 | MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL); | |
60e76d48 ZJS |
471 | if (r < 0) |
472 | return r; | |
d8fc6a00 LP |
473 | |
474 | FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") { | |
475 | _cleanup_free_ char *from = NULL, *to = NULL; | |
476 | ||
c6134d3e | 477 | from = path_join(full, x); |
d8fc6a00 LP |
478 | if (!from) |
479 | return log_oom(); | |
480 | ||
c6134d3e | 481 | to = path_join(top, x); |
d8fc6a00 LP |
482 | if (!to) |
483 | return log_oom(); | |
484 | ||
485 | (void) mkdir(to, 0755); | |
486 | ||
511a8cfe | 487 | r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL); |
60e76d48 ZJS |
488 | if (r < 0) |
489 | return r; | |
d8fc6a00 | 490 | |
511a8cfe LP |
491 | r = mount_nofollow_verbose(LOG_ERR, NULL, to, NULL, |
492 | MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL); | |
60e76d48 ZJS |
493 | if (r < 0) |
494 | return r; | |
d8fc6a00 LP |
495 | } |
496 | ||
30f5d104 | 497 | r = umount_verbose(LOG_ERR, full, UMOUNT_NOFOLLOW); |
60e76d48 ZJS |
498 | if (r < 0) |
499 | return r; | |
d8fc6a00 LP |
500 | |
501 | if (rmdir(full) < 0) | |
502 | return log_error_errno(errno, "Failed to remove %s: %m", full); | |
503 | ||
0996ef00 CB |
504 | /* Create mountpoint for cgroups. Otherwise we are not allowed since we |
505 | * remount /sys read-only. | |
506 | */ | |
5980d463 | 507 | const char *x = prefix_roota(top, "/fs/cgroup"); |
677a72cd | 508 | (void) mkdir_p(x, 0755); |
d8fc6a00 | 509 | |
511a8cfe LP |
510 | return mount_nofollow_verbose(LOG_ERR, NULL, top, NULL, |
511 | MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL); | |
d8fc6a00 LP |
512 | } |
513 | ||
b71a0192 CB |
514 | #define PROC_DEFAULT_MOUNT_FLAGS (MS_NOSUID|MS_NOEXEC|MS_NODEV) |
515 | #define SYS_DEFAULT_MOUNT_FLAGS (MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV) | |
516 | ||
e83bebef | 517 | int mount_all(const char *dest, |
4f086aab | 518 | MountSettingsMask mount_settings, |
2fa017f1 | 519 | uid_t uid_shift, |
e83bebef LP |
520 | const char *selinux_apifs_context) { |
521 | ||
de40a303 LP |
522 | #define PROC_INACCESSIBLE_REG(path) \ |
523 | { "/run/systemd/inaccessible/reg", (path), NULL, NULL, MS_BIND, \ | |
524 | MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \ | |
d4b653c5 LP |
525 | { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \ |
526 | MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */ | |
527 | ||
528 | #define PROC_READ_ONLY(path) \ | |
529 | { (path), (path), NULL, NULL, MS_BIND, \ | |
530 | MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \ | |
531 | { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \ | |
532 | MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */ | |
533 | ||
e83bebef LP |
534 | typedef struct MountPoint { |
535 | const char *what; | |
536 | const char *where; | |
537 | const char *type; | |
538 | const char *options; | |
539 | unsigned long flags; | |
4f086aab | 540 | MountSettingsMask mount_settings; |
e83bebef LP |
541 | } MountPoint; |
542 | ||
543 | static const MountPoint mount_table[] = { | |
d4b653c5 | 544 | /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing) */ |
b71a0192 | 545 | { "proc", "/proc", "proc", NULL, PROC_DEFAULT_MOUNT_FLAGS, |
511a8cfe | 546 | MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_MKDIR|MOUNT_FOLLOW_SYMLINKS }, /* we follow symlinks here since not following them requires /proc/ already being mounted, which we don't have here. */ |
d4b653c5 LP |
547 | |
548 | { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND, | |
549 | MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ | |
550 | ||
551 | { "/proc/sys/net", "/proc/sys/net", NULL, NULL, MS_BIND, | |
552 | MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */ | |
553 | ||
554 | { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, | |
555 | MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* ... then, make it r/o */ | |
556 | ||
557 | /* Make these files inaccessible to container payloads: they potentially leak information about kernel | |
558 | * internals or the host's execution environment to the container */ | |
de40a303 LP |
559 | PROC_INACCESSIBLE_REG("/proc/kallsyms"), |
560 | PROC_INACCESSIBLE_REG("/proc/kcore"), | |
561 | PROC_INACCESSIBLE_REG("/proc/keys"), | |
562 | PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"), | |
563 | PROC_INACCESSIBLE_REG("/proc/timer_list"), | |
d4b653c5 LP |
564 | |
565 | /* Make these directories read-only to container payloads: they show hardware information, and in some | |
566 | * cases contain tunables the container really shouldn't have access to. */ | |
567 | PROC_READ_ONLY("/proc/acpi"), | |
568 | PROC_READ_ONLY("/proc/apm"), | |
569 | PROC_READ_ONLY("/proc/asound"), | |
570 | PROC_READ_ONLY("/proc/bus"), | |
571 | PROC_READ_ONLY("/proc/fs"), | |
572 | PROC_READ_ONLY("/proc/irq"), | |
573 | PROC_READ_ONLY("/proc/scsi"), | |
574 | ||
e1bb4b0d | 575 | { "mqueue", "/dev/mqueue", "mqueue", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, |
dcff2fa5 | 576 | MOUNT_IN_USERNS|MOUNT_MKDIR }, |
849b9b85 | 577 | |
d4b653c5 | 578 | /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */ |
9f563f27 | 579 | { "tmpfs", "/tmp", "tmpfs", "mode=01777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME, |
dcff2fa5 | 580 | MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP|MOUNT_MKDIR }, |
9f563f27 | 581 | { "tmpfs", "/sys", "tmpfs", "mode=0555" TMPFS_LIMITS_SYS, MS_NOSUID|MS_NOEXEC|MS_NODEV, |
dcff2fa5 | 582 | MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS|MOUNT_MKDIR }, |
b71a0192 | 583 | { "sysfs", "/sys", "sysfs", NULL, SYS_DEFAULT_MOUNT_FLAGS, |
dcff2fa5 | 584 | MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO|MOUNT_MKDIR }, /* skipped if above was mounted */ |
e1bb4b0d | 585 | { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, |
dcff2fa5 | 586 | MOUNT_FATAL|MOUNT_MKDIR }, /* skipped if above was mounted */ |
9f563f27 | 587 | { "tmpfs", "/dev", "tmpfs", "mode=0755" TMPFS_LIMITS_PRIVATE_DEV, MS_NOSUID|MS_STRICTATIME, |
dcff2fa5 | 588 | MOUNT_FATAL|MOUNT_MKDIR }, |
9f563f27 | 589 | { "tmpfs", "/dev/shm", "tmpfs", "mode=01777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME, |
dcff2fa5 | 590 | MOUNT_FATAL|MOUNT_MKDIR }, |
9f563f27 | 591 | { "tmpfs", "/run", "tmpfs", "mode=0755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME, |
dcff2fa5 | 592 | MOUNT_FATAL|MOUNT_MKDIR }, |
d64e32c2 LP |
593 | { "/run/host", "/run/host", NULL, NULL, MS_BIND, |
594 | MOUNT_FATAL|MOUNT_MKDIR|MOUNT_PREFIX_ROOT }, /* Prepare this so that we can make it read-only when we are done */ | |
595 | { "/etc/os-release", "/run/host/os-release", NULL, NULL, MS_BIND, | |
596 | MOUNT_TOUCH }, /* As per kernel interface requirements, bind mount first (creating mount points) and make read-only later */ | |
597 | { "/usr/lib/os-release", "/run/host/os-release", NULL, NULL, MS_BIND, | |
598 | MOUNT_FATAL }, /* If /etc/os-release doesn't exist use the version in /usr/lib as fallback */ | |
599 | { NULL, "/run/host/os-release", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, | |
600 | MOUNT_FATAL }, | |
56339a10 LP |
601 | { NULL, "/run/host/os-release", NULL, NULL, MS_PRIVATE, |
602 | MOUNT_FATAL }, /* Turn off propagation (we only want that for the mount propagation tunnel dir) */ | |
d64e32c2 LP |
603 | { NULL, "/run/host", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, |
604 | MOUNT_FATAL|MOUNT_IN_USERNS }, | |
349cc4a5 | 605 | #if HAVE_SELINUX |
e1bb4b0d | 606 | { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND, |
6fe01ced | 607 | MOUNT_MKDIR }, /* Bind mount first (mkdir/chown the mount point in case /sys/ is mounted as minimal skeleton tmpfs) */ |
e1bb4b0d | 608 | { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, |
6fe01ced | 609 | 0 }, /* Then, make it r/o (don't mkdir/chown the mount point here, the previous entry already did that) */ |
56339a10 LP |
610 | { NULL, "/sys/fs/selinux", NULL, NULL, MS_PRIVATE, |
611 | 0 }, /* Turn off propagation (we only want that for the mount propagation tunnel dir) */ | |
e83bebef LP |
612 | #endif |
613 | }; | |
614 | ||
bd6609eb DDM |
615 | bool use_userns = FLAGS_SET(mount_settings, MOUNT_USE_USERNS); |
616 | bool netns = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_NETNS); | |
617 | bool ro = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO); | |
618 | bool in_userns = FLAGS_SET(mount_settings, MOUNT_IN_USERNS); | |
619 | bool tmpfs_tmp = FLAGS_SET(mount_settings, MOUNT_APPLY_TMPFS_TMP); | |
88614c8a | 620 | int r; |
e83bebef | 621 | |
d2b99ed7 | 622 | for (size_t k = 0; k < ELEMENTSOF(mount_table); k++) { |
d64e32c2 | 623 | _cleanup_free_ char *where = NULL, *options = NULL, *prefixed = NULL; |
bd6609eb | 624 | bool fatal = FLAGS_SET(mount_table[k].mount_settings, MOUNT_FATAL); |
d64e32c2 | 625 | const char *o; |
4f086aab | 626 | |
bd6609eb | 627 | if (in_userns != FLAGS_SET(mount_table[k].mount_settings, MOUNT_IN_USERNS)) |
4f086aab | 628 | continue; |
e83bebef | 629 | |
bd6609eb | 630 | if (!netns && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_NETNS)) |
d1678248 ILG |
631 | continue; |
632 | ||
bd6609eb | 633 | if (!ro && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_RO)) |
e83bebef LP |
634 | continue; |
635 | ||
bd6609eb | 636 | if (!tmpfs_tmp && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_TMPFS_TMP)) |
1099ceeb LP |
637 | continue; |
638 | ||
f461a28d | 639 | r = chase(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where, NULL); |
8ce48cf0 | 640 | if (r < 0) |
e85c644e | 641 | return log_error_errno(r, "Failed to resolve %s%s: %m", strempty(dest), mount_table[k].where); |
e83bebef | 642 | |
e83bebef | 643 | /* Skip this entry if it is not a remount. */ |
de40a303 | 644 | if (mount_table[k].what) { |
b409aacb | 645 | r = path_is_mount_point(where); |
de40a303 LP |
646 | if (r < 0 && r != -ENOENT) |
647 | return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where); | |
648 | if (r > 0) | |
649 | continue; | |
650 | } | |
e83bebef | 651 | |
d64e32c2 | 652 | if ((mount_table[k].mount_settings & (MOUNT_MKDIR|MOUNT_TOUCH)) != 0) { |
b3b1a08a | 653 | uid_t u = (use_userns && !in_userns) ? uid_shift : UID_INVALID; |
e1bb4b0d LB |
654 | |
655 | if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH)) | |
656 | r = mkdir_parents_safe(dest, where, 0755, u, u, 0); | |
657 | else | |
658 | r = mkdir_p_safe(dest, where, 0755, u, u, 0); | |
dcff2fa5 LP |
659 | if (r < 0 && r != -EEXIST) { |
660 | if (fatal && r != -EROFS) | |
661 | return log_error_errno(r, "Failed to create directory %s: %m", where); | |
e83bebef | 662 | |
dcff2fa5 LP |
663 | log_debug_errno(r, "Failed to create directory %s: %m", where); |
664 | ||
665 | /* If we failed mkdir() or chown() due to the root directory being read only, | |
666 | * attempt to mount this fs anyway and let mount_verbose log any errors */ | |
667 | if (r != -EROFS) | |
668 | continue; | |
669 | } | |
d64e32c2 LP |
670 | } |
671 | ||
672 | if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH)) { | |
673 | r = touch(where); | |
674 | if (r < 0 && r != -EEXIST) { | |
675 | if (fatal && r != -EROFS) | |
676 | return log_error_errno(r, "Failed to create file %s: %m", where); | |
677 | ||
678 | log_debug_errno(r, "Failed to create file %s: %m", where); | |
679 | if (r != -EROFS) | |
680 | continue; | |
e1bb4b0d | 681 | } |
e83bebef LP |
682 | } |
683 | ||
684 | o = mount_table[k].options; | |
685 | if (streq_ptr(mount_table[k].type, "tmpfs")) { | |
2fa017f1 | 686 | r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options); |
e83bebef LP |
687 | if (r < 0) |
688 | return log_oom(); | |
689 | if (r > 0) | |
690 | o = options; | |
691 | } | |
692 | ||
d64e32c2 LP |
693 | if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_PREFIX_ROOT)) { |
694 | /* Optionally prefix the mount source with the root dir. This is useful in bind | |
695 | * mounts to be created within the container image before we transition into it. Note | |
fcdd21ec | 696 | * that MOUNT_IN_USERNS is run after we transitioned hence prefixing is not necessary |
d64e32c2 | 697 | * for those. */ |
f461a28d | 698 | r = chase(mount_table[k].what, dest, CHASE_PREFIX_ROOT, &prefixed, NULL); |
d64e32c2 | 699 | if (r < 0) |
e85c644e | 700 | return log_error_errno(r, "Failed to resolve %s%s: %m", strempty(dest), mount_table[k].what); |
d64e32c2 LP |
701 | } |
702 | ||
511a8cfe LP |
703 | r = mount_verbose_full( |
704 | fatal ? LOG_ERR : LOG_DEBUG, | |
705 | prefixed ?: mount_table[k].what, | |
706 | where, | |
707 | mount_table[k].type, | |
708 | mount_table[k].flags, | |
709 | o, | |
710 | FLAGS_SET(mount_table[k].mount_settings, MOUNT_FOLLOW_SYMLINKS)); | |
4f086aab | 711 | if (r < 0 && fatal) |
60e76d48 | 712 | return r; |
e83bebef LP |
713 | } |
714 | ||
715 | return 0; | |
716 | } | |
717 | ||
1aa18710 | 718 | static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts, RemountIdmapping *idmapping) { |
a11fd406 ILG |
719 | unsigned long flags = *mount_flags; |
720 | char *opts = NULL; | |
1aa18710 | 721 | RemountIdmapping new_idmapping = *idmapping; |
a11fd406 ILG |
722 | int r; |
723 | ||
724 | assert(options); | |
725 | ||
726 | for (;;) { | |
727 | _cleanup_free_ char *word = NULL; | |
728 | ||
d2b99ed7 | 729 | r = extract_first_word(&options, &word, ",", 0); |
a11fd406 ILG |
730 | if (r < 0) |
731 | return log_error_errno(r, "Failed to extract mount option: %m"); | |
732 | if (r == 0) | |
733 | break; | |
734 | ||
735 | if (streq(word, "rbind")) | |
736 | flags |= MS_REC; | |
737 | else if (streq(word, "norbind")) | |
738 | flags &= ~MS_REC; | |
c0c8f718 | 739 | else if (streq(word, "idmap")) |
1aa18710 | 740 | new_idmapping = REMOUNT_IDMAPPING_HOST_ROOT; |
c0c8f718 | 741 | else if (streq(word, "noidmap")) |
1aa18710 | 742 | new_idmapping = REMOUNT_IDMAPPING_NONE; |
2b2777ed QD |
743 | else if (streq(word, "rootidmap")) |
744 | new_idmapping = REMOUNT_IDMAPPING_HOST_OWNER; | |
614d09a3 BF |
745 | else if (streq(word, "owneridmap")) |
746 | new_idmapping = REMOUNT_IDMAPPING_HOST_OWNER_TO_TARGET_OWNER; | |
d2b99ed7 | 747 | else |
38288f0b | 748 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), |
d2b99ed7 | 749 | "Invalid bind mount option: %s", word); |
a11fd406 ILG |
750 | } |
751 | ||
752 | *mount_flags = flags; | |
1aa18710 | 753 | *idmapping = new_idmapping; |
a11fd406 ILG |
754 | /* in the future mount_opts will hold string options for mount(2) */ |
755 | *mount_opts = opts; | |
756 | ||
757 | return 0; | |
758 | } | |
759 | ||
c0c8f718 | 760 | static int mount_bind(const char *dest, CustomMount *m, uid_t uid_shift, uid_t uid_range) { |
a11fd406 ILG |
761 | _cleanup_free_ char *mount_opts = NULL, *where = NULL; |
762 | unsigned long mount_flags = MS_BIND | MS_REC; | |
68cf43c3 | 763 | struct stat source_st, dest_st; |
614d09a3 | 764 | uid_t dest_uid = UID_INVALID; |
e83bebef | 765 | int r; |
1aa18710 | 766 | RemountIdmapping idmapping = REMOUNT_IDMAPPING_NONE; |
e83bebef | 767 | |
86c0dd4a | 768 | assert(dest); |
e83bebef LP |
769 | assert(m); |
770 | ||
a11fd406 | 771 | if (m->options) { |
1aa18710 | 772 | r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts, &idmapping); |
a11fd406 ILG |
773 | if (r < 0) |
774 | return r; | |
775 | } | |
776 | ||
07bca16f LP |
777 | /* If this is a bind mount from a temporary sources change ownership of the source to the container's |
778 | * root UID. Otherwise it would always show up as "nobody" if user namespacing is used. */ | |
779 | if (m->rm_rf_tmpdir && chown(m->source, uid_shift, uid_shift) < 0) | |
780 | return log_error_errno(errno, "Failed to chown %s: %m", m->source); | |
781 | ||
e83bebef LP |
782 | if (stat(m->source, &source_st) < 0) |
783 | return log_error_errno(errno, "Failed to stat %s: %m", m->source); | |
784 | ||
f461a28d | 785 | r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL); |
68cf43c3 | 786 | if (r < 0) |
ec57bd42 | 787 | return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination); |
8ce48cf0 LP |
788 | if (r > 0) { /* Path exists already? */ |
789 | ||
790 | if (stat(where, &dest_st) < 0) | |
791 | return log_error_errno(errno, "Failed to stat %s: %m", where); | |
e83bebef | 792 | |
614d09a3 BF |
793 | dest_uid = dest_st.st_uid; |
794 | ||
baaa35ad ZJS |
795 | if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) |
796 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), | |
797 | "Cannot bind mount directory %s on file %s.", | |
798 | m->source, where); | |
799 | ||
800 | if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode)) | |
801 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), | |
802 | "Cannot bind mount file %s on directory %s.", | |
803 | m->source, where); | |
e83bebef | 804 | |
8ce48cf0 | 805 | } else { /* Path doesn't exist yet? */ |
0a67965f | 806 | r = mkdir_parents_safe_label(dest, where, 0755, uid_shift, uid_shift, MKDIR_IGNORE_EXISTING); |
e83bebef LP |
807 | if (r < 0) |
808 | return log_error_errno(r, "Failed to make parents of %s: %m", where); | |
b97e83cb BN |
809 | |
810 | /* Create the mount point. Any non-directory file can be | |
811 | * mounted on any non-directory file (regular, fifo, socket, | |
812 | * char, block). | |
813 | */ | |
814 | if (S_ISDIR(source_st.st_mode)) | |
815 | r = mkdir_label(where, 0755); | |
816 | else | |
817 | r = touch(where); | |
818 | if (r < 0) | |
819 | return log_error_errno(r, "Failed to create mount point %s: %m", where); | |
0a67965f DDM |
820 | |
821 | if (chown(where, uid_shift, uid_shift) < 0) | |
822 | return log_error_errno(errno, "Failed to chown %s: %m", where); | |
614d09a3 BF |
823 | |
824 | dest_uid = uid_shift; | |
8ce48cf0 | 825 | } |
e83bebef | 826 | |
511a8cfe | 827 | r = mount_nofollow_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts); |
60e76d48 ZJS |
828 | if (r < 0) |
829 | return r; | |
e83bebef LP |
830 | |
831 | if (m->read_only) { | |
64e82c19 | 832 | r = bind_remount_recursive(where, MS_RDONLY, MS_RDONLY, NULL); |
e83bebef LP |
833 | if (r < 0) |
834 | return log_error_errno(r, "Read-only bind mount failed: %m"); | |
835 | } | |
836 | ||
1aa18710 | 837 | if (idmapping != REMOUNT_IDMAPPING_NONE) { |
614d09a3 | 838 | r = remount_idmap(STRV_MAKE(where), uid_shift, uid_range, source_st.st_uid, dest_uid, idmapping); |
c0c8f718 AV |
839 | if (r < 0) |
840 | return log_error_errno(r, "Failed to map ids for bind mount %s: %m", where); | |
841 | } | |
842 | ||
e83bebef LP |
843 | return 0; |
844 | } | |
845 | ||
e091a5df | 846 | static int mount_tmpfs(const char *dest, CustomMount *m, uid_t uid_shift, const char *selinux_apifs_context) { |
68cf43c3 LP |
847 | const char *options; |
848 | _cleanup_free_ char *buf = NULL, *where = NULL; | |
e83bebef LP |
849 | int r; |
850 | ||
851 | assert(dest); | |
852 | assert(m); | |
853 | ||
f461a28d | 854 | r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL); |
68cf43c3 | 855 | if (r < 0) |
ec57bd42 | 856 | return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination); |
8ce48cf0 LP |
857 | if (r == 0) { /* Doesn't exist yet? */ |
858 | r = mkdir_p_label(where, 0755); | |
859 | if (r < 0) | |
860 | return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where); | |
861 | } | |
e83bebef | 862 | |
2fa017f1 | 863 | r = tmpfs_patch_options(m->options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf); |
e83bebef LP |
864 | if (r < 0) |
865 | return log_oom(); | |
866 | options = r > 0 ? buf : m->options; | |
867 | ||
511a8cfe | 868 | return mount_nofollow_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options); |
e83bebef LP |
869 | } |
870 | ||
86c0dd4a | 871 | static char *joined_and_escaped_lower_dirs(char **lower) { |
e83bebef LP |
872 | _cleanup_strv_free_ char **sv = NULL; |
873 | ||
874 | sv = strv_copy(lower); | |
875 | if (!sv) | |
876 | return NULL; | |
877 | ||
878 | strv_reverse(sv); | |
879 | ||
880 | if (!strv_shell_escape(sv, ",:")) | |
881 | return NULL; | |
882 | ||
883 | return strv_join(sv, ":"); | |
884 | } | |
885 | ||
886 | static int mount_overlay(const char *dest, CustomMount *m) { | |
86c0dd4a | 887 | _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL; |
68cf43c3 | 888 | const char *options; |
e83bebef LP |
889 | int r; |
890 | ||
891 | assert(dest); | |
892 | assert(m); | |
893 | ||
f461a28d | 894 | r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL); |
68cf43c3 | 895 | if (r < 0) |
ec57bd42 | 896 | return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination); |
8ce48cf0 LP |
897 | if (r == 0) { /* Doesn't exist yet? */ |
898 | r = mkdir_label(where, 0755); | |
899 | if (r < 0) | |
900 | return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where); | |
901 | } | |
e83bebef LP |
902 | |
903 | (void) mkdir_p_label(m->source, 0755); | |
904 | ||
905 | lower = joined_and_escaped_lower_dirs(m->lower); | |
906 | if (!lower) | |
907 | return log_oom(); | |
908 | ||
86c0dd4a LP |
909 | escaped_source = shell_escape(m->source, ",:"); |
910 | if (!escaped_source) | |
911 | return log_oom(); | |
e83bebef | 912 | |
86c0dd4a | 913 | if (m->read_only) |
e83bebef | 914 | options = strjoina("lowerdir=", escaped_source, ":", lower); |
86c0dd4a LP |
915 | else { |
916 | _cleanup_free_ char *escaped_work_dir = NULL; | |
e83bebef | 917 | |
e83bebef LP |
918 | escaped_work_dir = shell_escape(m->work_dir, ",:"); |
919 | if (!escaped_work_dir) | |
920 | return log_oom(); | |
921 | ||
922 | options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir); | |
923 | } | |
924 | ||
511a8cfe | 925 | return mount_nofollow_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options); |
e83bebef LP |
926 | } |
927 | ||
de40a303 | 928 | static int mount_inaccessible(const char *dest, CustomMount *m) { |
e5f10caf | 929 | _cleanup_free_ char *where = NULL, *source = NULL; |
de40a303 LP |
930 | struct stat st; |
931 | int r; | |
932 | ||
933 | assert(dest); | |
934 | assert(m); | |
935 | ||
f461a28d | 936 | r = chase_and_stat(m->destination, dest, CHASE_PREFIX_ROOT, &where, &st); |
de40a303 LP |
937 | if (r < 0) { |
938 | log_full_errno(m->graceful ? LOG_DEBUG : LOG_ERR, r, "Failed to resolve %s/%s: %m", dest, m->destination); | |
939 | return m->graceful ? 0 : r; | |
940 | } | |
941 | ||
48b747fa | 942 | r = mode_to_inaccessible_node(NULL, st.st_mode, &source); |
e5f10caf AZ |
943 | if (r < 0) |
944 | return m->graceful ? 0 : r; | |
de40a303 | 945 | |
511a8cfe | 946 | r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, source, where, NULL, MS_BIND, NULL); |
de40a303 LP |
947 | if (r < 0) |
948 | return m->graceful ? 0 : r; | |
949 | ||
511a8cfe | 950 | r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, NULL, where, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, NULL); |
2c9b7a7e | 951 | if (r < 0) { |
30f5d104 | 952 | (void) umount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, where, UMOUNT_NOFOLLOW); |
de40a303 | 953 | return m->graceful ? 0 : r; |
2c9b7a7e | 954 | } |
de40a303 LP |
955 | |
956 | return 0; | |
957 | } | |
958 | ||
959 | static int mount_arbitrary(const char *dest, CustomMount *m) { | |
960 | _cleanup_free_ char *where = NULL; | |
961 | int r; | |
962 | ||
963 | assert(dest); | |
964 | assert(m); | |
965 | ||
f461a28d | 966 | r = chase(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL); |
de40a303 LP |
967 | if (r < 0) |
968 | return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination); | |
969 | if (r == 0) { /* Doesn't exist yet? */ | |
970 | r = mkdir_p_label(where, 0755); | |
971 | if (r < 0) | |
972 | return log_error_errno(r, "Creating mount point for mount %s failed: %m", where); | |
973 | } | |
974 | ||
511a8cfe | 975 | return mount_nofollow_verbose(LOG_ERR, m->source, where, m->type_argument, 0, m->options); |
de40a303 LP |
976 | } |
977 | ||
e83bebef LP |
978 | int mount_custom( |
979 | const char *dest, | |
88614c8a | 980 | CustomMount *mounts, size_t n, |
e091a5df | 981 | uid_t uid_shift, |
c0c8f718 | 982 | uid_t uid_range, |
de40a303 | 983 | const char *selinux_apifs_context, |
5f0a6347 | 984 | MountSettingsMask mount_settings) { |
e83bebef LP |
985 | int r; |
986 | ||
987 | assert(dest); | |
988 | ||
d2b99ed7 | 989 | for (size_t i = 0; i < n; i++) { |
e83bebef LP |
990 | CustomMount *m = mounts + i; |
991 | ||
bd6609eb | 992 | if (FLAGS_SET(mount_settings, MOUNT_IN_USERNS) != m->in_userns) |
5f0a6347 DDM |
993 | continue; |
994 | ||
bd6609eb | 995 | if (FLAGS_SET(mount_settings, MOUNT_ROOT_ONLY) && !path_equal(m->destination, "/")) |
5f0a6347 DDM |
996 | continue; |
997 | ||
bd6609eb | 998 | if (FLAGS_SET(mount_settings, MOUNT_NON_ROOT_ONLY) && path_equal(m->destination, "/")) |
de40a303 LP |
999 | continue; |
1000 | ||
e83bebef LP |
1001 | switch (m->type) { |
1002 | ||
1003 | case CUSTOM_MOUNT_BIND: | |
c0c8f718 | 1004 | r = mount_bind(dest, m, uid_shift, uid_range); |
e83bebef LP |
1005 | break; |
1006 | ||
1007 | case CUSTOM_MOUNT_TMPFS: | |
e091a5df | 1008 | r = mount_tmpfs(dest, m, uid_shift, selinux_apifs_context); |
e83bebef LP |
1009 | break; |
1010 | ||
1011 | case CUSTOM_MOUNT_OVERLAY: | |
1012 | r = mount_overlay(dest, m); | |
1013 | break; | |
1014 | ||
de40a303 LP |
1015 | case CUSTOM_MOUNT_INACCESSIBLE: |
1016 | r = mount_inaccessible(dest, m); | |
1017 | break; | |
1018 | ||
1019 | case CUSTOM_MOUNT_ARBITRARY: | |
1020 | r = mount_arbitrary(dest, m); | |
1021 | break; | |
1022 | ||
e83bebef | 1023 | default: |
04499a70 | 1024 | assert_not_reached(); |
e83bebef LP |
1025 | } |
1026 | ||
1027 | if (r < 0) | |
1028 | return r; | |
1029 | } | |
1030 | ||
1031 | return 0; | |
1032 | } | |
1033 | ||
bbd407ea | 1034 | bool has_custom_root_mount(const CustomMount *mounts, size_t n) { |
d2b99ed7 ZJS |
1035 | for (size_t i = 0; i < n; i++) |
1036 | if (path_equal(mounts[i].destination, "/")) | |
bbd407ea | 1037 | return true; |
bbd407ea DDM |
1038 | |
1039 | return false; | |
1040 | } | |
1041 | ||
e091a5df | 1042 | static int setup_volatile_state(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) { |
e83bebef LP |
1043 | _cleanup_free_ char *buf = NULL; |
1044 | const char *p, *options; | |
1045 | int r; | |
1046 | ||
1047 | assert(directory); | |
1048 | ||
e5b43a04 | 1049 | /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */ |
e83bebef | 1050 | |
64e82c19 | 1051 | r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL); |
e83bebef LP |
1052 | if (r < 0) |
1053 | return log_error_errno(r, "Failed to remount %s read-only: %m", directory); | |
1054 | ||
1055 | p = prefix_roota(directory, "/var"); | |
1056 | r = mkdir(p, 0755); | |
1057 | if (r < 0 && errno != EEXIST) | |
1058 | return log_error_errno(errno, "Failed to create %s: %m", directory); | |
1059 | ||
9f563f27 | 1060 | options = "mode=0755" TMPFS_LIMITS_VOLATILE_STATE; |
2fa017f1 | 1061 | r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf); |
e83bebef LP |
1062 | if (r < 0) |
1063 | return log_oom(); | |
1064 | if (r > 0) | |
1065 | options = buf; | |
1066 | ||
511a8cfe | 1067 | return mount_nofollow_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options); |
e83bebef LP |
1068 | } |
1069 | ||
e091a5df | 1070 | static int setup_volatile_yes(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) { |
e83bebef LP |
1071 | bool tmpfs_mounted = false, bind_mounted = false; |
1072 | char template[] = "/tmp/nspawn-volatile-XXXXXX"; | |
07b9f3f0 | 1073 | _cleanup_free_ char *buf = NULL, *bindir = NULL; |
e83bebef | 1074 | const char *f, *t, *options; |
07b9f3f0 | 1075 | struct stat st; |
e83bebef LP |
1076 | int r; |
1077 | ||
1078 | assert(directory); | |
1079 | ||
07b9f3f0 LP |
1080 | /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and |
1081 | * that read-only. Before we start setting this up let's validate if the image has the /usr merge | |
1082 | * implemented, and let's output a friendly log message if it hasn't. */ | |
1083 | ||
1084 | bindir = path_join(directory, "/bin"); | |
1085 | if (!bindir) | |
1086 | return log_oom(); | |
1087 | if (lstat(bindir, &st) < 0) { | |
1088 | if (errno != ENOENT) | |
1089 | return log_error_errno(errno, "Failed to stat /bin directory below image: %m"); | |
1090 | ||
1091 | /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the | |
1092 | * rest. */ | |
1093 | } else if (S_ISDIR(st.st_mode)) | |
1094 | return log_error_errno(SYNTHETIC_ERRNO(EISDIR), | |
1095 | "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. " | |
1096 | "Please work with your distribution and help them adopt the merged /usr scheme."); | |
1097 | else if (!S_ISLNK(st.st_mode)) | |
1098 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), | |
1099 | "Error starting image: if --volatile=yes is used /bin must be a symlink (for merged /usr support) or non-existent (in which case a symlink is created automatically)."); | |
e83bebef LP |
1100 | |
1101 | if (!mkdtemp(template)) | |
1102 | return log_error_errno(errno, "Failed to create temporary directory: %m"); | |
1103 | ||
9f563f27 | 1104 | options = "mode=0755" TMPFS_LIMITS_ROOTFS; |
2fa017f1 | 1105 | r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf); |
e83bebef | 1106 | if (r < 0) |
c55d0ae7 | 1107 | goto fail; |
e83bebef LP |
1108 | if (r > 0) |
1109 | options = buf; | |
1110 | ||
511a8cfe | 1111 | r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options); |
60e76d48 | 1112 | if (r < 0) |
e83bebef | 1113 | goto fail; |
e83bebef LP |
1114 | |
1115 | tmpfs_mounted = true; | |
1116 | ||
1117 | f = prefix_roota(directory, "/usr"); | |
1118 | t = prefix_roota(template, "/usr"); | |
1119 | ||
1120 | r = mkdir(t, 0755); | |
1121 | if (r < 0 && errno != EEXIST) { | |
1122 | r = log_error_errno(errno, "Failed to create %s: %m", t); | |
1123 | goto fail; | |
1124 | } | |
1125 | ||
511a8cfe | 1126 | r = mount_nofollow_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL); |
60e76d48 | 1127 | if (r < 0) |
e83bebef | 1128 | goto fail; |
e83bebef LP |
1129 | |
1130 | bind_mounted = true; | |
1131 | ||
64e82c19 | 1132 | r = bind_remount_recursive(t, MS_RDONLY, MS_RDONLY, NULL); |
e83bebef LP |
1133 | if (r < 0) { |
1134 | log_error_errno(r, "Failed to remount %s read-only: %m", t); | |
1135 | goto fail; | |
1136 | } | |
1137 | ||
511a8cfe | 1138 | r = mount_nofollow_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL); |
60e76d48 | 1139 | if (r < 0) |
e83bebef | 1140 | goto fail; |
e83bebef LP |
1141 | |
1142 | (void) rmdir(template); | |
1143 | ||
1144 | return 0; | |
1145 | ||
1146 | fail: | |
1147 | if (bind_mounted) | |
30f5d104 | 1148 | (void) umount_verbose(LOG_ERR, t, UMOUNT_NOFOLLOW); |
e83bebef LP |
1149 | |
1150 | if (tmpfs_mounted) | |
30f5d104 LP |
1151 | (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW); |
1152 | ||
e83bebef LP |
1153 | (void) rmdir(template); |
1154 | return r; | |
1155 | } | |
b53ede69 | 1156 | |
e091a5df | 1157 | static int setup_volatile_overlay(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) { |
6c610aca LP |
1158 | _cleanup_free_ char *buf = NULL, *escaped_directory = NULL, *escaped_upper = NULL, *escaped_work = NULL; |
1159 | char template[] = "/tmp/nspawn-volatile-XXXXXX"; | |
1160 | const char *upper, *work, *options; | |
1161 | bool tmpfs_mounted = false; | |
1162 | int r; | |
1163 | ||
1164 | assert(directory); | |
1165 | ||
1166 | /* --volatile=overlay means we mount an overlayfs to the root dir. */ | |
1167 | ||
1168 | if (!mkdtemp(template)) | |
1169 | return log_error_errno(errno, "Failed to create temporary directory: %m"); | |
1170 | ||
9f563f27 | 1171 | options = "mode=0755" TMPFS_LIMITS_ROOTFS; |
6c610aca LP |
1172 | r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf); |
1173 | if (r < 0) | |
1174 | goto finish; | |
1175 | if (r > 0) | |
1176 | options = buf; | |
1177 | ||
511a8cfe | 1178 | r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options); |
6c610aca LP |
1179 | if (r < 0) |
1180 | goto finish; | |
1181 | ||
1182 | tmpfs_mounted = true; | |
1183 | ||
1184 | upper = strjoina(template, "/upper"); | |
1185 | work = strjoina(template, "/work"); | |
1186 | ||
1187 | if (mkdir(upper, 0755) < 0) { | |
1188 | r = log_error_errno(errno, "Failed to create %s: %m", upper); | |
1189 | goto finish; | |
1190 | } | |
1191 | if (mkdir(work, 0755) < 0) { | |
1192 | r = log_error_errno(errno, "Failed to create %s: %m", work); | |
1193 | goto finish; | |
1194 | } | |
1195 | ||
1196 | /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice | |
1197 | * that the kernel allows us to do that without going through some mount point rearrangements. */ | |
1198 | ||
1199 | escaped_directory = shell_escape(directory, ",:"); | |
1200 | escaped_upper = shell_escape(upper, ",:"); | |
1201 | escaped_work = shell_escape(work, ",:"); | |
1202 | if (!escaped_directory || !escaped_upper || !escaped_work) { | |
1203 | r = -ENOMEM; | |
1204 | goto finish; | |
1205 | } | |
1206 | ||
1207 | options = strjoina("lowerdir=", escaped_directory, ",upperdir=", escaped_upper, ",workdir=", escaped_work); | |
511a8cfe | 1208 | r = mount_nofollow_verbose(LOG_ERR, "overlay", directory, "overlay", 0, options); |
6c610aca LP |
1209 | |
1210 | finish: | |
1211 | if (tmpfs_mounted) | |
30f5d104 | 1212 | (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW); |
6c610aca LP |
1213 | |
1214 | (void) rmdir(template); | |
1215 | return r; | |
1216 | } | |
1217 | ||
e5b43a04 LP |
1218 | int setup_volatile_mode( |
1219 | const char *directory, | |
1220 | VolatileMode mode, | |
e091a5df | 1221 | uid_t uid_shift, |
e5b43a04 LP |
1222 | const char *selinux_apifs_context) { |
1223 | ||
1224 | switch (mode) { | |
1225 | ||
1226 | case VOLATILE_YES: | |
e091a5df | 1227 | return setup_volatile_yes(directory, uid_shift, selinux_apifs_context); |
e5b43a04 LP |
1228 | |
1229 | case VOLATILE_STATE: | |
e091a5df | 1230 | return setup_volatile_state(directory, uid_shift, selinux_apifs_context); |
e5b43a04 | 1231 | |
6c610aca | 1232 | case VOLATILE_OVERLAY: |
e091a5df | 1233 | return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context); |
6c610aca | 1234 | |
e5b43a04 LP |
1235 | default: |
1236 | return 0; | |
1237 | } | |
1238 | } | |
1239 | ||
b53ede69 PW |
1240 | /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */ |
1241 | int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) { | |
1242 | _cleanup_free_ char *root_new = NULL, *root_old = NULL; | |
1243 | const char *p = s; | |
1244 | int r; | |
1245 | ||
1246 | assert(pivot_root_new); | |
1247 | assert(pivot_root_old); | |
1248 | ||
1249 | r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS); | |
1250 | if (r < 0) | |
1251 | return r; | |
1252 | if (r == 0) | |
1253 | return -EINVAL; | |
1254 | ||
1255 | if (isempty(p)) | |
1256 | root_old = NULL; | |
1257 | else { | |
1258 | root_old = strdup(p); | |
1259 | if (!root_old) | |
1260 | return -ENOMEM; | |
1261 | } | |
1262 | ||
1263 | if (!path_is_absolute(root_new)) | |
1264 | return -EINVAL; | |
1265 | if (root_old && !path_is_absolute(root_old)) | |
1266 | return -EINVAL; | |
1267 | ||
1268 | free_and_replace(*pivot_root_new, root_new); | |
1269 | free_and_replace(*pivot_root_old, root_old); | |
1270 | ||
1271 | return 0; | |
1272 | } | |
1273 | ||
1274 | int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) { | |
1275 | _cleanup_free_ char *directory_pivot_root_new = NULL; | |
1276 | _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL; | |
1277 | char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX"; | |
1278 | bool remove_pivot_tmp = false; | |
1279 | int r; | |
1280 | ||
1281 | assert(directory); | |
1282 | ||
1283 | if (!pivot_root_new) | |
1284 | return 0; | |
1285 | ||
1286 | /* Pivot pivot_root_new to / and the existing / to pivot_root_old. | |
1287 | * If pivot_root_old is NULL, the existing / disappears. | |
1288 | * This requires a temporary directory, pivot_tmp, which is | |
1289 | * not a child of either. | |
1290 | * | |
32e27670 LP |
1291 | * This is typically used for OSTree-style containers, where the root partition contains several |
1292 | * sysroots which could be run. Normally, one would be chosen by the bootloader and pivoted to / by | |
1293 | * initrd. | |
b53ede69 PW |
1294 | * |
1295 | * For example, for an OSTree deployment, pivot_root_new | |
1296 | * would be: /ostree/deploy/$os/deploy/$checksum. Note that this | |
1297 | * code doesn’t do the /var mount which OSTree expects: use | |
1298 | * --bind +/sysroot/ostree/deploy/$os/var:/var for that. | |
1299 | * | |
1300 | * So in the OSTree case, we’ll end up with something like: | |
1301 | * - directory = /tmp/nspawn-root-123456 | |
1302 | * - pivot_root_new = /ostree/deploy/os/deploy/123abc | |
1303 | * - pivot_root_old = /sysroot | |
1304 | * - directory_pivot_root_new = | |
1305 | * /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc | |
1306 | * - pivot_tmp = /tmp/nspawn-pivot-123456 | |
1307 | * - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot | |
1308 | * | |
1309 | * Requires all file systems at directory and below to be mounted | |
1310 | * MS_PRIVATE or MS_SLAVE so they can be moved. | |
1311 | */ | |
c6134d3e LP |
1312 | directory_pivot_root_new = path_join(directory, pivot_root_new); |
1313 | if (!directory_pivot_root_new) | |
1314 | return log_oom(); | |
b53ede69 PW |
1315 | |
1316 | /* Remount directory_pivot_root_new to make it movable. */ | |
511a8cfe | 1317 | r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL); |
b53ede69 PW |
1318 | if (r < 0) |
1319 | goto done; | |
1320 | ||
1321 | if (pivot_root_old) { | |
1322 | if (!mkdtemp(pivot_tmp)) { | |
1323 | r = log_error_errno(errno, "Failed to create temporary directory: %m"); | |
1324 | goto done; | |
1325 | } | |
1326 | ||
1327 | remove_pivot_tmp = true; | |
c6134d3e LP |
1328 | pivot_tmp_pivot_root_old = path_join(pivot_tmp, pivot_root_old); |
1329 | if (!pivot_tmp_pivot_root_old) { | |
1330 | r = log_oom(); | |
1331 | goto done; | |
1332 | } | |
b53ede69 | 1333 | |
511a8cfe | 1334 | r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL); |
b53ede69 PW |
1335 | if (r < 0) |
1336 | goto done; | |
1337 | ||
511a8cfe | 1338 | r = mount_nofollow_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL); |
b53ede69 PW |
1339 | if (r < 0) |
1340 | goto done; | |
1341 | ||
511a8cfe | 1342 | r = mount_nofollow_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL); |
b53ede69 PW |
1343 | if (r < 0) |
1344 | goto done; | |
1345 | } else { | |
511a8cfe | 1346 | r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL); |
b53ede69 PW |
1347 | if (r < 0) |
1348 | goto done; | |
1349 | } | |
1350 | ||
1351 | done: | |
1352 | if (remove_pivot_tmp) | |
1353 | (void) rmdir(pivot_tmp); | |
1354 | ||
1355 | return r; | |
1356 | } | |
b71a0192 CB |
1357 | |
1358 | #define NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS "/run/host/proc" | |
1359 | #define NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS "/run/host/sys" | |
1360 | ||
1361 | int pin_fully_visible_fs(void) { | |
1362 | int r; | |
1363 | ||
1364 | (void) mkdir_p(NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS, 0755); | |
1365 | (void) mkdir_p(NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS, 0755); | |
1366 | ||
1367 | r = mount_follow_verbose(LOG_ERR, "proc", NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS, "proc", PROC_DEFAULT_MOUNT_FLAGS, NULL); | |
1368 | if (r < 0) | |
1369 | return r; | |
1370 | ||
1371 | r = mount_follow_verbose(LOG_ERR, "sysfs", NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS, "sysfs", SYS_DEFAULT_MOUNT_FLAGS, NULL); | |
1372 | if (r < 0) | |
1373 | return r; | |
1374 | ||
1375 | return 0; | |
1376 | } | |
1377 | ||
1378 | static int do_wipe_fully_visible_fs(void) { | |
1379 | if (umount2(NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS, MNT_DETACH) < 0) | |
1380 | return log_error_errno(errno, "Failed to unmount temporary proc: %m"); | |
1381 | ||
1382 | if (rmdir(NSPAWN_PRIVATE_FULLY_VISIBLE_PROCFS) < 0) | |
1383 | return log_error_errno(errno, "Failed to remove temporary proc mountpoint: %m"); | |
1384 | ||
1385 | if (umount2(NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS, MNT_DETACH) < 0) | |
1386 | return log_error_errno(errno, "Failed to unmount temporary sys: %m"); | |
1387 | ||
1388 | if (rmdir(NSPAWN_PRIVATE_FULLY_VISIBLE_SYSFS) < 0) | |
1389 | return log_error_errno(errno, "Failed to remove temporary sys mountpoint: %m"); | |
1390 | ||
1391 | return 0; | |
1392 | } | |
1393 | ||
1394 | int wipe_fully_visible_fs(int mntns_fd) { | |
1395 | _cleanup_close_ int orig_mntns_fd = -EBADF; | |
1396 | int r, rr; | |
1397 | ||
d2881ef9 YW |
1398 | r = namespace_open(0, |
1399 | /* ret_pidns_fd = */ NULL, | |
1400 | &orig_mntns_fd, | |
1401 | /* ret_netns_fd = */ NULL, | |
1402 | /* ret_userns_fd = */ NULL, | |
1403 | /* ret_root_fd = */ NULL); | |
b71a0192 CB |
1404 | if (r < 0) |
1405 | return log_error_errno(r, "Failed to pin originating mount namespace: %m"); | |
1406 | ||
d2881ef9 YW |
1407 | r = namespace_enter(/* pidns_fd = */ -EBADF, |
1408 | mntns_fd, | |
1409 | /* netns_fd = */ -EBADF, | |
1410 | /* userns_fd = */ -EBADF, | |
1411 | /* root_fd = */ -EBADF); | |
b71a0192 CB |
1412 | if (r < 0) |
1413 | return log_error_errno(r, "Failed to enter mount namespace: %m"); | |
1414 | ||
1415 | rr = do_wipe_fully_visible_fs(); | |
1416 | ||
d2881ef9 YW |
1417 | r = namespace_enter(/* pidns_fd = */ -EBADF, |
1418 | orig_mntns_fd, | |
1419 | /* netns_fd = */ -EBADF, | |
1420 | /* userns_fd = */ -EBADF, | |
1421 | /* root_fd = */ -EBADF); | |
b71a0192 CB |
1422 | if (r < 0) |
1423 | return log_error_errno(r, "Failed to enter original mount namespace: %m"); | |
1424 | ||
1425 | return rr; | |
1426 | } |