]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
cd61c3bf | 2 | |
a8fbdf54 | 3 | #include <errno.h> |
ebd93cb6 | 4 | #include <fcntl.h> |
e08f94ac LP |
5 | #include <linux/fs.h> |
6 | #include <linux/loop.h> | |
a8fbdf54 TA |
7 | #include <stdio.h> |
8 | #include <stdlib.h> | |
e306723e | 9 | #include <sys/file.h> |
204f52e3 | 10 | #include <sys/ioctl.h> |
a8fbdf54 TA |
11 | #include <sys/stat.h> |
12 | #include <unistd.h> | |
546dbec5 | 13 | |
b5efdb8a | 14 | #include "alloc-util.h" |
cd61c3bf | 15 | #include "btrfs-util.h" |
c8b3094d | 16 | #include "chattr-util.h" |
ebd93cb6 | 17 | #include "copy.h" |
a0956174 | 18 | #include "dirent-util.h" |
c7664c07 | 19 | #include "dissect-image.h" |
686d13b9 | 20 | #include "env-file.h" |
b6e953f2 | 21 | #include "env-util.h" |
3ffd4af2 | 22 | #include "fd-util.h" |
f4f15635 | 23 | #include "fs-util.h" |
a8fbdf54 | 24 | #include "hashmap.h" |
c7664c07 LP |
25 | #include "hostname-util.h" |
26 | #include "id128-util.h" | |
a8fbdf54 TA |
27 | #include "lockfile-util.h" |
28 | #include "log.h" | |
c7664c07 | 29 | #include "loop-util.h" |
3ffd4af2 | 30 | #include "machine-image.h" |
546dbec5 | 31 | #include "macro.h" |
30535c16 | 32 | #include "mkdir.h" |
d8b4d14d | 33 | #include "nulstr-util.h" |
d58ad743 | 34 | #include "os-util.h" |
8e0b6570 | 35 | #include "path-util.h" |
c6878637 | 36 | #include "rm-rf.h" |
8b43440b | 37 | #include "string-table.h" |
07630cea | 38 | #include "string-util.h" |
8e0b6570 | 39 | #include "strv.h" |
a8fbdf54 | 40 | #include "time-util.h" |
8e0b6570 | 41 | #include "utf8.h" |
89a5a90c | 42 | #include "xattr-util.h" |
cd61c3bf | 43 | |
5ef46e5f | 44 | static const char* const image_search_path[_IMAGE_CLASS_MAX] = { |
ace9ab19 LP |
45 | [IMAGE_MACHINE] = "/etc/machines\0" /* only place symlinks here */ |
46 | "/run/machines\0" /* and here too */ | |
47 | "/var/lib/machines\0" /* the main place for images */ | |
48 | "/var/lib/container\0" /* legacy */ | |
5ef46e5f LP |
49 | "/usr/local/lib/machines\0" |
50 | "/usr/lib/machines\0", | |
51 | ||
ace9ab19 LP |
52 | [IMAGE_PORTABLE] = "/etc/portables\0" /* only place symlinks here */ |
53 | "/run/portables\0" /* and here too */ | |
54 | "/var/lib/portables\0" /* the main place for images */ | |
5ef46e5f LP |
55 | "/usr/local/lib/portables\0" |
56 | "/usr/lib/portables\0", | |
57 | }; | |
c2ce6a3d | 58 | |
8301aa0b YW |
59 | static Image *image_free(Image *i) { |
60 | assert(i); | |
9614bb06 | 61 | |
cd61c3bf LP |
62 | free(i->name); |
63 | free(i->path); | |
c7664c07 LP |
64 | |
65 | free(i->hostname); | |
66 | strv_free(i->machine_info); | |
67 | strv_free(i->os_release); | |
68 | ||
6b430fdb | 69 | return mfree(i); |
cd61c3bf LP |
70 | } |
71 | ||
8301aa0b | 72 | DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free); |
b07ec5a1 YW |
73 | DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func, |
74 | Image, image_unref); | |
9614bb06 | 75 | |
8e0b6570 LP |
76 | static char **image_settings_path(Image *image) { |
77 | _cleanup_strv_free_ char **l = NULL; | |
8e0b6570 LP |
78 | const char *fn, *s; |
79 | unsigned i = 0; | |
80 | ||
81 | assert(image); | |
82 | ||
83 | l = new0(char*, 4); | |
84 | if (!l) | |
85 | return NULL; | |
86 | ||
87 | fn = strjoina(image->name, ".nspawn"); | |
88 | ||
b910cc72 LP |
89 | FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") { |
90 | l[i] = path_join(s, fn); | |
8e0b6570 LP |
91 | if (!l[i]) |
92 | return NULL; | |
93 | ||
94 | i++; | |
95 | } | |
96 | ||
97 | l[i] = file_in_same_dir(image->path, fn); | |
98 | if (!l[i]) | |
99 | return NULL; | |
100 | ||
ae2a15bc | 101 | return TAKE_PTR(l); |
8e0b6570 LP |
102 | } |
103 | ||
bafbac4e LP |
104 | static char *image_roothash_path(Image *image) { |
105 | const char *fn; | |
106 | ||
107 | assert(image); | |
108 | ||
109 | fn = strjoina(image->name, ".roothash"); | |
110 | ||
111 | return file_in_same_dir(image->path, fn); | |
112 | } | |
113 | ||
c2ce6a3d | 114 | static int image_new( |
cd61c3bf | 115 | ImageType t, |
5fc7f358 | 116 | const char *pretty, |
cd61c3bf | 117 | const char *path, |
5fc7f358 | 118 | const char *filename, |
cd61c3bf | 119 | bool read_only, |
10f9c755 | 120 | usec_t crtime, |
cd61c3bf | 121 | usec_t mtime, |
c2ce6a3d | 122 | Image **ret) { |
cd61c3bf LP |
123 | |
124 | _cleanup_(image_unrefp) Image *i = NULL; | |
cd61c3bf | 125 | |
cd61c3bf LP |
126 | assert(t >= 0); |
127 | assert(t < _IMAGE_TYPE_MAX); | |
5fc7f358 LP |
128 | assert(pretty); |
129 | assert(filename); | |
c2ce6a3d | 130 | assert(ret); |
cd61c3bf | 131 | |
c2ce6a3d | 132 | i = new0(Image, 1); |
cd61c3bf LP |
133 | if (!i) |
134 | return -ENOMEM; | |
135 | ||
9614bb06 | 136 | i->n_ref = 1; |
cd61c3bf LP |
137 | i->type = t; |
138 | i->read_only = read_only; | |
10f9c755 | 139 | i->crtime = crtime; |
cd61c3bf | 140 | i->mtime = mtime; |
c19de711 | 141 | i->usage = i->usage_exclusive = (uint64_t) -1; |
b6b18498 | 142 | i->limit = i->limit_exclusive = (uint64_t) -1; |
cd61c3bf | 143 | |
5fc7f358 | 144 | i->name = strdup(pretty); |
cd61c3bf LP |
145 | if (!i->name) |
146 | return -ENOMEM; | |
147 | ||
657ee2d8 | 148 | i->path = path_join(path, filename); |
5fc7f358 LP |
149 | if (!i->path) |
150 | return -ENOMEM; | |
151 | ||
858d36c1 | 152 | path_simplify(i->path, false); |
cd61c3bf | 153 | |
1cc6c93a | 154 | *ret = TAKE_PTR(i); |
c2ce6a3d | 155 | |
cd61c3bf LP |
156 | return 0; |
157 | } | |
158 | ||
4756c94e LP |
159 | static int extract_pretty(const char *path, const char *suffix, char **ret) { |
160 | _cleanup_free_ char *name = NULL; | |
161 | const char *p; | |
162 | size_t n; | |
163 | ||
164 | assert(path); | |
165 | assert(ret); | |
166 | ||
167 | p = last_path_component(path); | |
168 | n = strcspn(p, "/"); | |
169 | ||
170 | name = strndup(p, n); | |
171 | if (!name) | |
172 | return -ENOMEM; | |
173 | ||
174 | if (suffix) { | |
175 | char *e; | |
176 | ||
177 | e = endswith(name, suffix); | |
178 | if (!e) | |
179 | return -EINVAL; | |
180 | ||
181 | *e = 0; | |
182 | } | |
183 | ||
184 | if (!image_name_is_valid(name)) | |
185 | return -EINVAL; | |
186 | ||
187 | *ret = TAKE_PTR(name); | |
188 | return 0; | |
189 | } | |
190 | ||
5fc7f358 LP |
191 | static int image_make( |
192 | const char *pretty, | |
193 | int dfd, | |
194 | const char *path, | |
195 | const char *filename, | |
3a6ce860 | 196 | const struct stat *st, |
5fc7f358 LP |
197 | Image **ret) { |
198 | ||
2570578d | 199 | _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL; |
3a6ce860 | 200 | struct stat stbuf; |
5fc7f358 | 201 | bool read_only; |
cd61c3bf LP |
202 | int r; |
203 | ||
3a6ce860 | 204 | assert(dfd >= 0 || dfd == AT_FDCWD); |
bcb846f3 | 205 | assert(path || dfd == AT_FDCWD); |
5fc7f358 | 206 | assert(filename); |
cd61c3bf | 207 | |
eb38edce | 208 | /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block |
3a6ce860 LP |
209 | * devices into /var/lib/machines/, and treat them normally. |
210 | * | |
211 | * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we | |
212 | * recognize. */ | |
cd61c3bf | 213 | |
3a6ce860 LP |
214 | if (!st) { |
215 | if (fstatat(dfd, filename, &stbuf, 0) < 0) | |
216 | return -errno; | |
217 | ||
218 | st = &stbuf; | |
219 | } | |
cd61c3bf | 220 | |
2570578d LP |
221 | if (!path) { |
222 | if (dfd == AT_FDCWD) | |
223 | (void) safe_getcwd(&parent); | |
224 | else | |
225 | (void) fd_get_path(dfd, &parent); | |
226 | } | |
bcb846f3 | 227 | |
5fc7f358 LP |
228 | read_only = |
229 | (path && path_startswith(path, "/usr")) || | |
08ff5529 | 230 | (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS); |
86e339c8 | 231 | |
3a6ce860 | 232 | if (S_ISDIR(st->st_mode)) { |
01b72568 LP |
233 | _cleanup_close_ int fd = -1; |
234 | unsigned file_attr = 0; | |
cd61c3bf | 235 | |
c2ce6a3d | 236 | if (!ret) |
3a6ce860 | 237 | return 0; |
cd61c3bf | 238 | |
4756c94e LP |
239 | if (!pretty) { |
240 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
241 | if (r < 0) | |
242 | return r; | |
243 | ||
244 | pretty = pretty_buffer; | |
245 | } | |
5fc7f358 | 246 | |
01b72568 LP |
247 | fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY); |
248 | if (fd < 0) | |
249 | return -errno; | |
250 | ||
c2ce6a3d | 251 | /* btrfs subvolumes have inode 256 */ |
3a6ce860 | 252 | if (st->st_ino == 256) { |
cd61c3bf | 253 | |
21222ea5 LP |
254 | r = btrfs_is_filesystem(fd); |
255 | if (r < 0) | |
256 | return r; | |
257 | if (r) { | |
10f9c755 | 258 | BtrfsSubvolInfo info; |
cd61c3bf | 259 | |
c2ce6a3d | 260 | /* It's a btrfs subvolume */ |
cd61c3bf | 261 | |
5bcd08db | 262 | r = btrfs_subvol_get_info_fd(fd, 0, &info); |
10f9c755 LP |
263 | if (r < 0) |
264 | return r; | |
c2ce6a3d LP |
265 | |
266 | r = image_new(IMAGE_SUBVOLUME, | |
5fc7f358 | 267 | pretty, |
c2ce6a3d | 268 | path, |
5fc7f358 LP |
269 | filename, |
270 | info.read_only || read_only, | |
10f9c755 | 271 | info.otime, |
c2ce6a3d | 272 | 0, |
c2ce6a3d LP |
273 | ret); |
274 | if (r < 0) | |
275 | return r; | |
276 | ||
5bcd08db LP |
277 | if (btrfs_quota_scan_ongoing(fd) == 0) { |
278 | BtrfsQuotaInfo quota; | |
b6b18498 | 279 | |
5bcd08db LP |
280 | r = btrfs_subvol_get_subtree_quota_fd(fd, 0, "a); |
281 | if (r >= 0) { | |
282 | (*ret)->usage = quota.referenced; | |
283 | (*ret)->usage_exclusive = quota.exclusive; | |
284 | ||
285 | (*ret)->limit = quota.referenced_max; | |
286 | (*ret)->limit_exclusive = quota.exclusive_max; | |
287 | } | |
b6b18498 LP |
288 | } |
289 | ||
3a6ce860 | 290 | return 0; |
cd61c3bf | 291 | } |
c2ce6a3d | 292 | } |
cd61c3bf | 293 | |
01b72568 LP |
294 | /* If the IMMUTABLE bit is set, we consider the |
295 | * directory read-only. Since the ioctl is not | |
296 | * supported everywhere we ignore failures. */ | |
297 | (void) read_attr_fd(fd, &file_attr); | |
cd61c3bf | 298 | |
01b72568 | 299 | /* It's just a normal directory. */ |
c2ce6a3d | 300 | r = image_new(IMAGE_DIRECTORY, |
5fc7f358 | 301 | pretty, |
c2ce6a3d | 302 | path, |
5fc7f358 | 303 | filename, |
01b72568 | 304 | read_only || (file_attr & FS_IMMUTABLE_FL), |
c2ce6a3d LP |
305 | 0, |
306 | 0, | |
307 | ret); | |
308 | if (r < 0) | |
309 | return r; | |
cd61c3bf | 310 | |
3a6ce860 | 311 | return 0; |
cd61c3bf | 312 | |
3a6ce860 | 313 | } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) { |
10f9c755 | 314 | usec_t crtime = 0; |
cd61c3bf | 315 | |
aceac2f0 | 316 | /* It's a RAW disk image */ |
cd61c3bf | 317 | |
c2ce6a3d | 318 | if (!ret) |
3a6ce860 | 319 | return 0; |
cd61c3bf | 320 | |
4756c94e | 321 | (void) fd_getcrtime_at(dfd, filename, &crtime, 0); |
10f9c755 | 322 | |
4756c94e LP |
323 | if (!pretty) { |
324 | r = extract_pretty(filename, ".raw", &pretty_buffer); | |
325 | if (r < 0) | |
326 | return r; | |
327 | ||
328 | pretty = pretty_buffer; | |
329 | } | |
10f9c755 | 330 | |
aceac2f0 | 331 | r = image_new(IMAGE_RAW, |
5fc7f358 | 332 | pretty, |
c2ce6a3d | 333 | path, |
5fc7f358 | 334 | filename, |
3a6ce860 | 335 | !(st->st_mode & 0222) || read_only, |
10f9c755 | 336 | crtime, |
3a6ce860 | 337 | timespec_load(&st->st_mtim), |
c2ce6a3d LP |
338 | ret); |
339 | if (r < 0) | |
340 | return r; | |
cd61c3bf | 341 | |
3a6ce860 LP |
342 | (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512; |
343 | (*ret)->limit = (*ret)->limit_exclusive = st->st_size; | |
b6b18498 | 344 | |
3a6ce860 | 345 | return 0; |
eb38edce | 346 | |
3a6ce860 | 347 | } else if (S_ISBLK(st->st_mode)) { |
eb38edce LP |
348 | _cleanup_close_ int block_fd = -1; |
349 | uint64_t size = UINT64_MAX; | |
350 | ||
351 | /* A block device */ | |
352 | ||
353 | if (!ret) | |
3a6ce860 | 354 | return 0; |
eb38edce | 355 | |
4756c94e LP |
356 | if (!pretty) { |
357 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
358 | if (r < 0) | |
359 | return r; | |
360 | ||
361 | pretty = pretty_buffer; | |
362 | } | |
eb38edce LP |
363 | |
364 | block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
365 | if (block_fd < 0) | |
2570578d | 366 | log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce | 367 | else { |
3a6ce860 LP |
368 | /* Refresh stat data after opening the node */ |
369 | if (fstat(block_fd, &stbuf) < 0) | |
eb38edce | 370 | return -errno; |
3a6ce860 LP |
371 | st = &stbuf; |
372 | ||
373 | if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */ | |
eb38edce LP |
374 | return -ENOTTY; |
375 | ||
376 | if (!read_only) { | |
377 | int state = 0; | |
378 | ||
379 | if (ioctl(block_fd, BLKROGET, &state) < 0) | |
2570578d | 380 | log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
381 | else if (state) |
382 | read_only = true; | |
383 | } | |
384 | ||
385 | if (ioctl(block_fd, BLKGETSIZE64, &size) < 0) | |
2570578d | 386 | log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
387 | |
388 | block_fd = safe_close(block_fd); | |
389 | } | |
390 | ||
391 | r = image_new(IMAGE_BLOCK, | |
392 | pretty, | |
393 | path, | |
394 | filename, | |
3a6ce860 | 395 | !(st->st_mode & 0222) || read_only, |
eb38edce LP |
396 | 0, |
397 | 0, | |
398 | ret); | |
399 | if (r < 0) | |
400 | return r; | |
401 | ||
ed0cb346 | 402 | if (!IN_SET(size, 0, UINT64_MAX)) |
eb38edce LP |
403 | (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size; |
404 | ||
3a6ce860 | 405 | return 0; |
c2ce6a3d | 406 | } |
cd61c3bf | 407 | |
3a6ce860 | 408 | return -EMEDIUMTYPE; |
c2ce6a3d | 409 | } |
cd61c3bf | 410 | |
5ef46e5f | 411 | int image_find(ImageClass class, const char *name, Image **ret) { |
c2ce6a3d LP |
412 | const char *path; |
413 | int r; | |
cd61c3bf | 414 | |
5ef46e5f LP |
415 | assert(class >= 0); |
416 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d | 417 | assert(name); |
cd61c3bf | 418 | |
c2ce6a3d LP |
419 | /* There are no images with invalid names */ |
420 | if (!image_name_is_valid(name)) | |
3a6ce860 | 421 | return -ENOENT; |
cd61c3bf | 422 | |
5ef46e5f | 423 | NULSTR_FOREACH(path, image_search_path[class]) { |
c2ce6a3d | 424 | _cleanup_closedir_ DIR *d = NULL; |
3a6ce860 | 425 | struct stat st; |
cd61c3bf | 426 | |
c2ce6a3d LP |
427 | d = opendir(path); |
428 | if (!d) { | |
429 | if (errno == ENOENT) | |
430 | continue; | |
cd61c3bf | 431 | |
c2ce6a3d LP |
432 | return -errno; |
433 | } | |
cd61c3bf | 434 | |
3a6ce860 LP |
435 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to |
436 | * symlink block devices into the search path */ | |
437 | if (fstatat(dirfd(d), name, &st, 0) < 0) { | |
aceac2f0 | 438 | _cleanup_free_ char *raw = NULL; |
5fc7f358 | 439 | |
3a6ce860 LP |
440 | if (errno != ENOENT) |
441 | return -errno; | |
442 | ||
b910cc72 | 443 | raw = strjoin(name, ".raw"); |
aceac2f0 | 444 | if (!raw) |
5fc7f358 LP |
445 | return -ENOMEM; |
446 | ||
3a6ce860 LP |
447 | if (fstatat(dirfd(d), raw, &st, 0) < 0) { |
448 | ||
449 | if (errno == ENOENT) | |
450 | continue; | |
451 | ||
452 | return -errno; | |
453 | } | |
454 | ||
455 | if (!S_ISREG(st.st_mode)) | |
5fc7f358 | 456 | continue; |
3a6ce860 LP |
457 | |
458 | r = image_make(name, dirfd(d), path, raw, &st, ret); | |
459 | ||
460 | } else { | |
461 | if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode)) | |
462 | continue; | |
463 | ||
464 | r = image_make(name, dirfd(d), path, name, &st, ret); | |
5fc7f358 | 465 | } |
3a6ce860 LP |
466 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) |
467 | continue; | |
c2ce6a3d LP |
468 | if (r < 0) |
469 | return r; | |
cd61c3bf | 470 | |
cf604fd4 LP |
471 | if (ret) |
472 | (*ret)->discoverable = true; | |
473 | ||
c2ce6a3d LP |
474 | return 1; |
475 | } | |
476 | ||
cf604fd4 LP |
477 | if (class == IMAGE_MACHINE && streq(name, ".host")) { |
478 | r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret); | |
479 | if (r < 0) | |
480 | return r; | |
481 | ||
482 | if (ret) | |
483 | (*ret)->discoverable = true; | |
484 | ||
485 | return r; | |
486 | } | |
5fc7f358 | 487 | |
3a6ce860 | 488 | return -ENOENT; |
c2ce6a3d LP |
489 | }; |
490 | ||
2ddf182b | 491 | int image_from_path(const char *path, Image **ret) { |
cf604fd4 LP |
492 | |
493 | /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether | |
494 | * the image is in the image search path. And if it is we don't know if the path we used is actually not | |
3fe91079 | 495 | * overridden by another, different image earlier in the search path */ |
cf604fd4 | 496 | |
2ddf182b LP |
497 | if (path_equal(path, "/")) |
498 | return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret); | |
499 | ||
500 | return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret); | |
501 | } | |
502 | ||
503 | int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) { | |
504 | if (image_name_is_valid(name_or_path)) | |
505 | return image_find(class, name_or_path, ret); | |
506 | ||
507 | return image_from_path(name_or_path, ret); | |
508 | } | |
509 | ||
5ef46e5f | 510 | int image_discover(ImageClass class, Hashmap *h) { |
c2ce6a3d LP |
511 | const char *path; |
512 | int r; | |
513 | ||
5ef46e5f LP |
514 | assert(class >= 0); |
515 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d LP |
516 | assert(h); |
517 | ||
5ef46e5f | 518 | NULSTR_FOREACH(path, image_search_path[class]) { |
c2ce6a3d LP |
519 | _cleanup_closedir_ DIR *d = NULL; |
520 | struct dirent *de; | |
521 | ||
522 | d = opendir(path); | |
523 | if (!d) { | |
524 | if (errno == ENOENT) | |
a67a4c8c | 525 | continue; |
c2ce6a3d LP |
526 | |
527 | return -errno; | |
528 | } | |
529 | ||
530 | FOREACH_DIRENT_ALL(de, d, return -errno) { | |
531 | _cleanup_(image_unrefp) Image *image = NULL; | |
1bf36bf9 | 532 | _cleanup_free_ char *truncated = NULL; |
3a6ce860 LP |
533 | const char *pretty; |
534 | struct stat st; | |
c2ce6a3d | 535 | |
1bf36bf9 | 536 | if (dot_or_dot_dot(de->d_name)) |
c2ce6a3d LP |
537 | continue; |
538 | ||
3a6ce860 LP |
539 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people |
540 | * to symlink block devices into the search path */ | |
541 | if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) { | |
542 | if (errno == ENOENT) | |
543 | continue; | |
544 | ||
545 | return -errno; | |
546 | } | |
547 | ||
548 | if (S_ISREG(st.st_mode)) { | |
549 | const char *e; | |
550 | ||
551 | e = endswith(de->d_name, ".raw"); | |
552 | if (!e) | |
553 | continue; | |
554 | ||
1bf36bf9 LP |
555 | truncated = strndup(de->d_name, e - de->d_name); |
556 | if (!truncated) | |
557 | return -ENOMEM; | |
558 | ||
559 | pretty = truncated; | |
3a6ce860 | 560 | } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode)) |
1bf36bf9 | 561 | pretty = de->d_name; |
3a6ce860 LP |
562 | else |
563 | continue; | |
1bf36bf9 LP |
564 | |
565 | if (!image_name_is_valid(pretty)) | |
566 | continue; | |
567 | ||
568 | if (hashmap_contains(h, pretty)) | |
c2ce6a3d LP |
569 | continue; |
570 | ||
3a6ce860 LP |
571 | r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image); |
572 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) | |
c2ce6a3d LP |
573 | continue; |
574 | if (r < 0) | |
575 | return r; | |
576 | ||
cf604fd4 LP |
577 | image->discoverable = true; |
578 | ||
c2ce6a3d LP |
579 | r = hashmap_put(h, image->name, image); |
580 | if (r < 0) | |
581 | return r; | |
582 | ||
583 | image = NULL; | |
cd61c3bf LP |
584 | } |
585 | } | |
586 | ||
5ef46e5f | 587 | if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) { |
5fc7f358 LP |
588 | _cleanup_(image_unrefp) Image *image = NULL; |
589 | ||
3a6ce860 | 590 | r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image); |
5fc7f358 LP |
591 | if (r < 0) |
592 | return r; | |
593 | ||
cf604fd4 LP |
594 | image->discoverable = true; |
595 | ||
5fc7f358 LP |
596 | r = hashmap_put(h, image->name, image); |
597 | if (r < 0) | |
598 | return r; | |
599 | ||
600 | image = NULL; | |
5fc7f358 LP |
601 | } |
602 | ||
cd61c3bf LP |
603 | return 0; |
604 | } | |
605 | ||
08682124 | 606 | int image_remove(Image *i) { |
8e766630 | 607 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
8e0b6570 | 608 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 609 | _cleanup_free_ char *roothash = NULL; |
8e0b6570 | 610 | char **j; |
30535c16 LP |
611 | int r; |
612 | ||
08682124 LP |
613 | assert(i); |
614 | ||
d94c2b06 | 615 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
08682124 LP |
616 | return -EROFS; |
617 | ||
8e0b6570 LP |
618 | settings = image_settings_path(i); |
619 | if (!settings) | |
620 | return -ENOMEM; | |
621 | ||
bafbac4e LP |
622 | roothash = image_roothash_path(i); |
623 | if (!roothash) | |
624 | return -ENOMEM; | |
625 | ||
30535c16 LP |
626 | /* Make sure we don't interfere with a running nspawn */ |
627 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
628 | if (r < 0) | |
629 | return r; | |
630 | ||
ebd93cb6 LP |
631 | switch (i->type) { |
632 | ||
633 | case IMAGE_SUBVOLUME: | |
9fb0b9c7 LP |
634 | |
635 | /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the | |
636 | * big guns */ | |
637 | if (unlink(i->path) < 0) { | |
638 | r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA); | |
639 | if (r < 0) | |
640 | return r; | |
641 | } | |
642 | ||
8e0b6570 | 643 | break; |
ebd93cb6 LP |
644 | |
645 | case IMAGE_DIRECTORY: | |
01b72568 | 646 | /* Allow deletion of read-only directories */ |
db9a4254 | 647 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
8e0b6570 LP |
648 | r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME); |
649 | if (r < 0) | |
650 | return r; | |
651 | ||
652 | break; | |
01b72568 | 653 | |
eb38edce LP |
654 | case IMAGE_BLOCK: |
655 | ||
656 | /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node | |
657 | * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink | |
658 | * the thing (it's most likely a symlink after all). */ | |
659 | ||
660 | if (path_startswith(i->path, "/dev")) | |
661 | break; | |
662 | ||
4831981d | 663 | _fallthrough_; |
aceac2f0 | 664 | case IMAGE_RAW: |
41d1ed05 LP |
665 | if (unlink(i->path) < 0) |
666 | return -errno; | |
8e0b6570 | 667 | break; |
ebd93cb6 LP |
668 | |
669 | default: | |
15411c0c | 670 | return -EOPNOTSUPP; |
ebd93cb6 | 671 | } |
8e0b6570 LP |
672 | |
673 | STRV_FOREACH(j, settings) { | |
674 | if (unlink(*j) < 0 && errno != ENOENT) | |
675 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j); | |
676 | } | |
677 | ||
bafbac4e LP |
678 | if (unlink(roothash) < 0 && errno != ENOENT) |
679 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash); | |
680 | ||
8e0b6570 LP |
681 | return 0; |
682 | } | |
683 | ||
bafbac4e | 684 | static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
8e0b6570 LP |
685 | _cleanup_free_ char *rs = NULL; |
686 | const char *fn; | |
687 | ||
bafbac4e | 688 | fn = strjoina(new_name, suffix); |
8e0b6570 LP |
689 | |
690 | rs = file_in_same_dir(path, fn); | |
691 | if (!rs) | |
692 | return -ENOMEM; | |
693 | ||
694 | return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs); | |
ebd93cb6 LP |
695 | } |
696 | ||
697 | int image_rename(Image *i, const char *new_name) { | |
8e766630 | 698 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT; |
bafbac4e | 699 | _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL; |
8e0b6570 | 700 | _cleanup_strv_free_ char **settings = NULL; |
01b72568 | 701 | unsigned file_attr = 0; |
8e0b6570 | 702 | char **j; |
ebd93cb6 LP |
703 | int r; |
704 | ||
705 | assert(i); | |
706 | ||
707 | if (!image_name_is_valid(new_name)) | |
708 | return -EINVAL; | |
709 | ||
d94c2b06 | 710 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
711 | return -EROFS; |
712 | ||
8e0b6570 LP |
713 | settings = image_settings_path(i); |
714 | if (!settings) | |
715 | return -ENOMEM; | |
716 | ||
bafbac4e LP |
717 | roothash = image_roothash_path(i); |
718 | if (!roothash) | |
719 | return -ENOMEM; | |
720 | ||
30535c16 LP |
721 | /* Make sure we don't interfere with a running nspawn */ |
722 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
723 | if (r < 0) | |
724 | return r; | |
725 | ||
726 | /* Make sure nobody takes the new name, between the time we | |
727 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 728 | * time we take possession of it */ |
30535c16 LP |
729 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
730 | if (r < 0) | |
731 | return r; | |
732 | ||
5ef46e5f | 733 | r = image_find(IMAGE_MACHINE, new_name, NULL); |
3a6ce860 | 734 | if (r >= 0) |
ebd93cb6 | 735 | return -EEXIST; |
3a6ce860 LP |
736 | if (r != -ENOENT) |
737 | return r; | |
ebd93cb6 LP |
738 | |
739 | switch (i->type) { | |
740 | ||
ebd93cb6 | 741 | case IMAGE_DIRECTORY: |
01b72568 LP |
742 | /* Turn of the immutable bit while we rename the image, so that we can rename it */ |
743 | (void) read_attr_path(i->path, &file_attr); | |
744 | ||
745 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 746 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
01b72568 | 747 | |
4831981d | 748 | _fallthrough_; |
01b72568 | 749 | case IMAGE_SUBVOLUME: |
ebd93cb6 LP |
750 | new_path = file_in_same_dir(i->path, new_name); |
751 | break; | |
752 | ||
eb38edce LP |
753 | case IMAGE_BLOCK: |
754 | ||
755 | /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */ | |
756 | if (path_startswith(i->path, "/dev")) | |
757 | return -EROFS; | |
758 | ||
759 | new_path = file_in_same_dir(i->path, new_name); | |
760 | break; | |
761 | ||
aceac2f0 | 762 | case IMAGE_RAW: { |
ebd93cb6 LP |
763 | const char *fn; |
764 | ||
63c372cb | 765 | fn = strjoina(new_name, ".raw"); |
ebd93cb6 LP |
766 | new_path = file_in_same_dir(i->path, fn); |
767 | break; | |
768 | } | |
769 | ||
770 | default: | |
15411c0c | 771 | return -EOPNOTSUPP; |
ebd93cb6 LP |
772 | } |
773 | ||
774 | if (!new_path) | |
775 | return -ENOMEM; | |
776 | ||
777 | nn = strdup(new_name); | |
778 | if (!nn) | |
779 | return -ENOMEM; | |
780 | ||
f85ef957 AC |
781 | r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path); |
782 | if (r < 0) | |
783 | return r; | |
ebd93cb6 | 784 | |
01b72568 LP |
785 | /* Restore the immutable bit, if it was set before */ |
786 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 787 | (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL); |
01b72568 | 788 | |
f9ecfd3b DL |
789 | free_and_replace(i->path, new_path); |
790 | free_and_replace(i->name, nn); | |
ebd93cb6 | 791 | |
8e0b6570 | 792 | STRV_FOREACH(j, settings) { |
bafbac4e | 793 | r = rename_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
794 | if (r < 0 && r != -ENOENT) |
795 | log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j); | |
796 | } | |
797 | ||
bafbac4e LP |
798 | r = rename_auxiliary_file(roothash, new_name, ".roothash"); |
799 | if (r < 0 && r != -ENOENT) | |
800 | log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash); | |
801 | ||
ebd93cb6 LP |
802 | return 0; |
803 | } | |
804 | ||
bafbac4e | 805 | static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
8e0b6570 LP |
806 | _cleanup_free_ char *rs = NULL; |
807 | const char *fn; | |
808 | ||
bafbac4e | 809 | fn = strjoina(new_name, suffix); |
8e0b6570 LP |
810 | |
811 | rs = file_in_same_dir(path, fn); | |
812 | if (!rs) | |
813 | return -ENOMEM; | |
814 | ||
8a016c74 | 815 | return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK); |
8e0b6570 LP |
816 | } |
817 | ||
ebd93cb6 | 818 | int image_clone(Image *i, const char *new_name, bool read_only) { |
8e766630 | 819 | _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT; |
8e0b6570 | 820 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 821 | _cleanup_free_ char *roothash = NULL; |
ebd93cb6 | 822 | const char *new_path; |
8e0b6570 | 823 | char **j; |
ebd93cb6 LP |
824 | int r; |
825 | ||
826 | assert(i); | |
827 | ||
828 | if (!image_name_is_valid(new_name)) | |
829 | return -EINVAL; | |
830 | ||
8e0b6570 LP |
831 | settings = image_settings_path(i); |
832 | if (!settings) | |
833 | return -ENOMEM; | |
834 | ||
bafbac4e LP |
835 | roothash = image_roothash_path(i); |
836 | if (!roothash) | |
837 | return -ENOMEM; | |
838 | ||
30535c16 LP |
839 | /* Make sure nobody takes the new name, between the time we |
840 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 841 | * time we take possession of it */ |
30535c16 LP |
842 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
843 | if (r < 0) | |
844 | return r; | |
845 | ||
5ef46e5f | 846 | r = image_find(IMAGE_MACHINE, new_name, NULL); |
3a6ce860 | 847 | if (r >= 0) |
ebd93cb6 | 848 | return -EEXIST; |
3a6ce860 LP |
849 | if (r != -ENOENT) |
850 | return r; | |
ebd93cb6 LP |
851 | |
852 | switch (i->type) { | |
853 | ||
854 | case IMAGE_SUBVOLUME: | |
855 | case IMAGE_DIRECTORY: | |
9a50e3ca | 856 | /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain |
13e785f7 | 857 | * directory. */ |
9a50e3ca | 858 | |
63c372cb | 859 | new_path = strjoina("/var/lib/machines/", new_name); |
ebd93cb6 | 860 | |
17cbb288 LP |
861 | r = btrfs_subvol_snapshot(i->path, new_path, |
862 | (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | | |
863 | BTRFS_SNAPSHOT_FALLBACK_COPY | | |
864 | BTRFS_SNAPSHOT_FALLBACK_DIRECTORY | | |
865 | BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE | | |
866 | BTRFS_SNAPSHOT_RECURSIVE | | |
867 | BTRFS_SNAPSHOT_QUOTA); | |
868 | if (r >= 0) | |
9a50e3ca | 869 | /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */ |
8120ee28 | 870 | (void) btrfs_subvol_auto_qgroup(new_path, 0, true); |
5bcd08db | 871 | |
ebd93cb6 LP |
872 | break; |
873 | ||
aceac2f0 | 874 | case IMAGE_RAW: |
63c372cb | 875 | new_path = strjoina("/var/lib/machines/", new_name, ".raw"); |
ebd93cb6 | 876 | |
8a016c74 | 877 | r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME); |
ebd93cb6 LP |
878 | break; |
879 | ||
eb38edce | 880 | case IMAGE_BLOCK: |
ebd93cb6 | 881 | default: |
15411c0c | 882 | return -EOPNOTSUPP; |
ebd93cb6 LP |
883 | } |
884 | ||
885 | if (r < 0) | |
886 | return r; | |
887 | ||
8e0b6570 | 888 | STRV_FOREACH(j, settings) { |
bafbac4e | 889 | r = clone_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
890 | if (r < 0 && r != -ENOENT) |
891 | log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j); | |
892 | } | |
893 | ||
bafbac4e LP |
894 | r = clone_auxiliary_file(roothash, new_name, ".roothash"); |
895 | if (r < 0 && r != -ENOENT) | |
896 | log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash); | |
897 | ||
ebd93cb6 LP |
898 | return 0; |
899 | } | |
900 | ||
901 | int image_read_only(Image *i, bool b) { | |
8e766630 | 902 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
ebd93cb6 | 903 | int r; |
c7664c07 | 904 | |
ebd93cb6 LP |
905 | assert(i); |
906 | ||
d94c2b06 | 907 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
908 | return -EROFS; |
909 | ||
30535c16 LP |
910 | /* Make sure we don't interfere with a running nspawn */ |
911 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
912 | if (r < 0) | |
913 | return r; | |
914 | ||
ebd93cb6 LP |
915 | switch (i->type) { |
916 | ||
917 | case IMAGE_SUBVOLUME: | |
5bcd08db LP |
918 | |
919 | /* Note that we set the flag only on the top-level | |
920 | * subvolume of the image. */ | |
921 | ||
ebd93cb6 LP |
922 | r = btrfs_subvol_set_read_only(i->path, b); |
923 | if (r < 0) | |
924 | return r; | |
01b72568 LP |
925 | |
926 | break; | |
927 | ||
928 | case IMAGE_DIRECTORY: | |
929 | /* For simple directory trees we cannot use the access | |
930 | mode of the top-level directory, since it has an | |
931 | effect on the container itself. However, we can | |
932 | use the "immutable" flag, to at least make the | |
933 | top-level directory read-only. It's not as good as | |
934 | a read-only subvolume, but at least something, and | |
13e785f7 | 935 | we can read the value back. */ |
01b72568 | 936 | |
db9a4254 | 937 | r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL); |
01b72568 LP |
938 | if (r < 0) |
939 | return r; | |
940 | ||
ebd93cb6 LP |
941 | break; |
942 | ||
aceac2f0 | 943 | case IMAGE_RAW: { |
ebd93cb6 LP |
944 | struct stat st; |
945 | ||
946 | if (stat(i->path, &st) < 0) | |
947 | return -errno; | |
948 | ||
949 | if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0) | |
950 | return -errno; | |
f2068bcc LP |
951 | |
952 | /* If the images is now read-only, it's a good time to | |
953 | * defrag it, given that no write patterns will | |
954 | * fragment it again. */ | |
955 | if (b) | |
956 | (void) btrfs_defrag(i->path); | |
ebd93cb6 LP |
957 | break; |
958 | } | |
959 | ||
eb38edce LP |
960 | case IMAGE_BLOCK: { |
961 | _cleanup_close_ int fd = -1; | |
962 | struct stat st; | |
963 | int state = b; | |
964 | ||
965 | fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY); | |
966 | if (fd < 0) | |
967 | return -errno; | |
968 | ||
969 | if (fstat(fd, &st) < 0) | |
970 | return -errno; | |
971 | if (!S_ISBLK(st.st_mode)) | |
972 | return -ENOTTY; | |
973 | ||
974 | if (ioctl(fd, BLKROSET, &state) < 0) | |
975 | return -errno; | |
976 | ||
977 | break; | |
978 | } | |
979 | ||
ebd93cb6 | 980 | default: |
15411c0c | 981 | return -EOPNOTSUPP; |
ebd93cb6 LP |
982 | } |
983 | ||
984 | return 0; | |
08682124 LP |
985 | } |
986 | ||
30535c16 LP |
987 | int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) { |
988 | _cleanup_free_ char *p = NULL; | |
989 | LockFile t = LOCK_FILE_INIT; | |
990 | struct stat st; | |
f25bed67 | 991 | bool exclusive; |
30535c16 LP |
992 | int r; |
993 | ||
994 | assert(path); | |
995 | assert(global); | |
996 | assert(local); | |
997 | ||
f25bed67 LP |
998 | /* Locks an image path. This actually creates two locks: one "local" one, next to the image path |
999 | * itself, which might be shared via NFS. And another "global" one, in /run, that uses the | |
1000 | * device/inode number. This has the benefit that we can even lock a tree that is a mount point, | |
1001 | * correctly. */ | |
30535c16 | 1002 | |
30535c16 LP |
1003 | if (!path_is_absolute(path)) |
1004 | return -EINVAL; | |
1005 | ||
f25bed67 LP |
1006 | switch (operation & (LOCK_SH|LOCK_EX)) { |
1007 | case LOCK_SH: | |
1008 | exclusive = false; | |
1009 | break; | |
1010 | case LOCK_EX: | |
1011 | exclusive = true; | |
1012 | break; | |
1013 | default: | |
1014 | return -EINVAL; | |
1015 | } | |
1016 | ||
b6e953f2 LP |
1017 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1018 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1019 | return 0; | |
1020 | } | |
1021 | ||
f25bed67 LP |
1022 | /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are |
1023 | * running off it after all, and we don't want any images to manipulate the host image. We make an | |
1024 | * exception for shared locks however: we allow those (and make them NOPs since there's no point in | |
1025 | * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well, | |
1026 | * since it means changes made to the host might propagate to the container as they happen (and a | |
1027 | * shared lock kinda suggests that no changes happen at all while it is in place), but it's too | |
1028 | * useful not to allow read-only containers off the host root, hence let's support this, and trust | |
1029 | * the user to do the right thing with this. */ | |
1030 | if (path_equal(path, "/")) { | |
1031 | if (exclusive) | |
1032 | return -EBUSY; | |
1033 | ||
1034 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1035 | return 0; | |
1036 | } | |
b6e953f2 | 1037 | |
30535c16 | 1038 | if (stat(path, &st) >= 0) { |
eb38edce LP |
1039 | if (S_ISBLK(st.st_mode)) |
1040 | r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev)); | |
1041 | else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode)) | |
1042 | r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino); | |
1043 | else | |
1044 | return -ENOTTY; | |
1045 | ||
1046 | if (r < 0) | |
30535c16 LP |
1047 | return -ENOMEM; |
1048 | } | |
1049 | ||
f25bed67 LP |
1050 | /* For block devices we don't need the "local" lock, as the major/minor lock above should be |
1051 | * sufficient, since block devices are host local anyway. */ | |
1052 | if (!path_startswith(path, "/dev/")) { | |
eb38edce | 1053 | r = make_lock_file_for(path, operation, &t); |
8be17c9b | 1054 | if (r < 0) { |
f25bed67 | 1055 | if (!exclusive && r == -EROFS) |
771b7ead | 1056 | log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path); |
8be17c9b LT |
1057 | else |
1058 | return r; | |
1059 | } | |
eb38edce | 1060 | } |
30535c16 LP |
1061 | |
1062 | if (p) { | |
6e5dcce4 | 1063 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); |
30535c16 LP |
1064 | |
1065 | r = make_lock_file(p, operation, global); | |
1066 | if (r < 0) { | |
1067 | release_lock_file(&t); | |
1068 | return r; | |
1069 | } | |
546dbec5 LP |
1070 | } else |
1071 | *global = (LockFile) LOCK_FILE_INIT; | |
30535c16 LP |
1072 | |
1073 | *local = t; | |
1074 | return 0; | |
1075 | } | |
1076 | ||
cb81cd80 | 1077 | int image_set_limit(Image *i, uint64_t referenced_max) { |
d6ce17c7 LP |
1078 | assert(i); |
1079 | ||
d94c2b06 | 1080 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
d6ce17c7 LP |
1081 | return -EROFS; |
1082 | ||
1083 | if (i->type != IMAGE_SUBVOLUME) | |
15411c0c | 1084 | return -EOPNOTSUPP; |
d6ce17c7 | 1085 | |
5bcd08db LP |
1086 | /* We set the quota both for the subvolume as well as for the |
1087 | * subtree. The latter is mostly for historical reasons, since | |
1088 | * we didn't use to have a concept of subtree quota, and hence | |
1089 | * only modified the subvolume quota. */ | |
1090 | ||
1091 | (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max); | |
1092 | (void) btrfs_subvol_auto_qgroup(i->path, 0, true); | |
1093 | return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max); | |
d6ce17c7 LP |
1094 | } |
1095 | ||
c7664c07 | 1096 | int image_read_metadata(Image *i) { |
8e766630 | 1097 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
c7664c07 LP |
1098 | int r; |
1099 | ||
1100 | assert(i); | |
1101 | ||
1102 | r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock); | |
1103 | if (r < 0) | |
1104 | return r; | |
1105 | ||
1106 | switch (i->type) { | |
1107 | ||
1108 | case IMAGE_SUBVOLUME: | |
1109 | case IMAGE_DIRECTORY: { | |
1110 | _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL; | |
1111 | sd_id128_t machine_id = SD_ID128_NULL; | |
1112 | _cleanup_free_ char *hostname = NULL; | |
1113 | _cleanup_free_ char *path = NULL; | |
1114 | ||
a5648b80 | 1115 | r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1116 | if (r < 0 && r != -ENOENT) |
1117 | log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name); | |
1118 | else if (r >= 0) { | |
1119 | r = read_etc_hostname(path, &hostname); | |
1120 | if (r < 0) | |
1121 | log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name); | |
1122 | } | |
1123 | ||
1124 | path = mfree(path); | |
1125 | ||
a5648b80 | 1126 | r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1127 | if (r < 0 && r != -ENOENT) |
1128 | log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name); | |
1129 | else if (r >= 0) { | |
1130 | _cleanup_close_ int fd = -1; | |
1131 | ||
1132 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
1133 | if (fd < 0) | |
1134 | log_debug_errno(errno, "Failed to open %s: %m", path); | |
1135 | else { | |
1136 | r = id128_read_fd(fd, ID128_PLAIN, &machine_id); | |
1137 | if (r < 0) | |
1138 | log_debug_errno(r, "Image %s contains invalid machine ID.", i->name); | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | path = mfree(path); | |
1143 | ||
a5648b80 | 1144 | r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1145 | if (r < 0 && r != -ENOENT) |
1146 | log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name); | |
1147 | else if (r >= 0) { | |
aa8fbc74 | 1148 | r = load_env_file_pairs(NULL, path, &machine_info); |
c7664c07 LP |
1149 | if (r < 0) |
1150 | log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name); | |
1151 | } | |
1152 | ||
d58ad743 LP |
1153 | r = load_os_release_pairs(i->path, &os_release); |
1154 | if (r < 0) | |
1155 | log_debug_errno(r, "Failed to read os-release in image, ignoring: %m"); | |
c7664c07 LP |
1156 | |
1157 | free_and_replace(i->hostname, hostname); | |
1158 | i->machine_id = machine_id; | |
1159 | strv_free_and_replace(i->machine_info, machine_info); | |
1160 | strv_free_and_replace(i->os_release, os_release); | |
1161 | ||
1162 | break; | |
1163 | } | |
1164 | ||
1165 | case IMAGE_RAW: | |
1166 | case IMAGE_BLOCK: { | |
1167 | _cleanup_(loop_device_unrefp) LoopDevice *d = NULL; | |
1168 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
1169 | ||
e08f94ac | 1170 | r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, &d); |
c7664c07 LP |
1171 | if (r < 0) |
1172 | return r; | |
1173 | ||
e7cbe5cb | 1174 | r = dissect_image(d->fd, NULL, 0, NULL, DISSECT_IMAGE_REQUIRE_ROOT|DISSECT_IMAGE_RELAX_VAR_CHECK, &m); |
c7664c07 LP |
1175 | if (r < 0) |
1176 | return r; | |
1177 | ||
1178 | r = dissected_image_acquire_metadata(m); | |
1179 | if (r < 0) | |
1180 | return r; | |
1181 | ||
1182 | free_and_replace(i->hostname, m->hostname); | |
1183 | i->machine_id = m->machine_id; | |
1184 | strv_free_and_replace(i->machine_info, m->machine_info); | |
1185 | strv_free_and_replace(i->os_release, m->os_release); | |
1186 | ||
1187 | break; | |
1188 | } | |
1189 | ||
1190 | default: | |
1191 | return -EOPNOTSUPP; | |
1192 | } | |
1193 | ||
1194 | i->metadata_valid = true; | |
1195 | ||
1196 | return 0; | |
1197 | } | |
1198 | ||
30535c16 | 1199 | int image_name_lock(const char *name, int operation, LockFile *ret) { |
30535c16 LP |
1200 | assert(name); |
1201 | assert(ret); | |
1202 | ||
1203 | /* Locks an image name, regardless of the precise path used. */ | |
1204 | ||
1205 | if (!image_name_is_valid(name)) | |
1206 | return -EINVAL; | |
1207 | ||
b6e953f2 LP |
1208 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1209 | *ret = (LockFile) LOCK_FILE_INIT; | |
1210 | return 0; | |
1211 | } | |
1212 | ||
30535c16 LP |
1213 | if (streq(name, ".host")) |
1214 | return -EBUSY; | |
1215 | ||
6e5dcce4 ZJS |
1216 | const char *p = strjoina("/run/systemd/nspawn/locks/name-", name); |
1217 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); | |
30535c16 LP |
1218 | return make_lock_file(p, operation, ret); |
1219 | } | |
1220 | ||
1221 | bool image_name_is_valid(const char *s) { | |
1222 | if (!filename_is_valid(s)) | |
1223 | return false; | |
1224 | ||
1225 | if (string_has_cc(s, NULL)) | |
1226 | return false; | |
1227 | ||
1228 | if (!utf8_is_valid(s)) | |
1229 | return false; | |
1230 | ||
1231 | /* Temporary files for atomically creating new files */ | |
1232 | if (startswith(s, ".#")) | |
1233 | return false; | |
1234 | ||
1235 | return true; | |
1236 | } | |
1237 | ||
ace9ab19 LP |
1238 | bool image_in_search_path(ImageClass class, const char *image) { |
1239 | const char *path; | |
1240 | ||
1241 | assert(image); | |
1242 | ||
1243 | NULSTR_FOREACH(path, image_search_path[class]) { | |
1244 | const char *p; | |
1245 | size_t k; | |
1246 | ||
1247 | p = path_startswith(image, path); | |
1248 | if (!p) | |
1249 | continue; | |
1250 | ||
1251 | /* Make sure there's a filename following */ | |
1252 | k = strcspn(p, "/"); | |
1253 | if (k == 0) | |
1254 | continue; | |
1255 | ||
1256 | p += k; | |
1257 | ||
1258 | /* Accept trailing slashes */ | |
1259 | if (p[strspn(p, "/")] == 0) | |
1260 | return true; | |
1261 | ||
1262 | } | |
1263 | ||
1264 | return false; | |
1265 | } | |
1266 | ||
cd61c3bf LP |
1267 | static const char* const image_type_table[_IMAGE_TYPE_MAX] = { |
1268 | [IMAGE_DIRECTORY] = "directory", | |
1269 | [IMAGE_SUBVOLUME] = "subvolume", | |
aceac2f0 | 1270 | [IMAGE_RAW] = "raw", |
eb38edce | 1271 | [IMAGE_BLOCK] = "block", |
cd61c3bf LP |
1272 | }; |
1273 | ||
1274 | DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType); |