]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
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" |
e2054217 | 25 | #include "hostname-setup.h" |
c7664c07 | 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 | |
c2108701 | 132 | i = new(Image, 1); |
cd61c3bf LP |
133 | if (!i) |
134 | return -ENOMEM; | |
135 | ||
c2108701 LP |
136 | *i = (Image) { |
137 | .n_ref = 1, | |
138 | .type = t, | |
139 | .read_only = read_only, | |
140 | .crtime = crtime, | |
141 | .mtime = mtime, | |
142 | .usage = UINT64_MAX, | |
143 | .usage_exclusive = UINT64_MAX, | |
144 | .limit = UINT64_MAX, | |
145 | .limit_exclusive = UINT64_MAX, | |
146 | }; | |
cd61c3bf | 147 | |
5fc7f358 | 148 | i->name = strdup(pretty); |
cd61c3bf LP |
149 | if (!i->name) |
150 | return -ENOMEM; | |
151 | ||
657ee2d8 | 152 | i->path = path_join(path, filename); |
5fc7f358 LP |
153 | if (!i->path) |
154 | return -ENOMEM; | |
155 | ||
858d36c1 | 156 | path_simplify(i->path, false); |
cd61c3bf | 157 | |
1cc6c93a | 158 | *ret = TAKE_PTR(i); |
c2ce6a3d | 159 | |
cd61c3bf LP |
160 | return 0; |
161 | } | |
162 | ||
4756c94e LP |
163 | static int extract_pretty(const char *path, const char *suffix, char **ret) { |
164 | _cleanup_free_ char *name = NULL; | |
165 | const char *p; | |
166 | size_t n; | |
167 | ||
168 | assert(path); | |
169 | assert(ret); | |
170 | ||
171 | p = last_path_component(path); | |
172 | n = strcspn(p, "/"); | |
173 | ||
174 | name = strndup(p, n); | |
175 | if (!name) | |
176 | return -ENOMEM; | |
177 | ||
178 | if (suffix) { | |
179 | char *e; | |
180 | ||
181 | e = endswith(name, suffix); | |
182 | if (!e) | |
183 | return -EINVAL; | |
184 | ||
185 | *e = 0; | |
186 | } | |
187 | ||
188 | if (!image_name_is_valid(name)) | |
189 | return -EINVAL; | |
190 | ||
191 | *ret = TAKE_PTR(name); | |
192 | return 0; | |
193 | } | |
194 | ||
5fc7f358 LP |
195 | static int image_make( |
196 | const char *pretty, | |
197 | int dfd, | |
198 | const char *path, | |
199 | const char *filename, | |
3a6ce860 | 200 | const struct stat *st, |
5fc7f358 LP |
201 | Image **ret) { |
202 | ||
2570578d | 203 | _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL; |
3a6ce860 | 204 | struct stat stbuf; |
5fc7f358 | 205 | bool read_only; |
cd61c3bf LP |
206 | int r; |
207 | ||
3a6ce860 | 208 | assert(dfd >= 0 || dfd == AT_FDCWD); |
bcb846f3 | 209 | assert(path || dfd == AT_FDCWD); |
5fc7f358 | 210 | assert(filename); |
cd61c3bf | 211 | |
eb38edce | 212 | /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block |
3a6ce860 LP |
213 | * devices into /var/lib/machines/, and treat them normally. |
214 | * | |
215 | * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we | |
216 | * recognize. */ | |
cd61c3bf | 217 | |
3a6ce860 LP |
218 | if (!st) { |
219 | if (fstatat(dfd, filename, &stbuf, 0) < 0) | |
220 | return -errno; | |
221 | ||
222 | st = &stbuf; | |
223 | } | |
cd61c3bf | 224 | |
2570578d LP |
225 | if (!path) { |
226 | if (dfd == AT_FDCWD) | |
227 | (void) safe_getcwd(&parent); | |
228 | else | |
229 | (void) fd_get_path(dfd, &parent); | |
230 | } | |
bcb846f3 | 231 | |
5fc7f358 LP |
232 | read_only = |
233 | (path && path_startswith(path, "/usr")) || | |
08ff5529 | 234 | (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS); |
86e339c8 | 235 | |
3a6ce860 | 236 | if (S_ISDIR(st->st_mode)) { |
01b72568 LP |
237 | _cleanup_close_ int fd = -1; |
238 | unsigned file_attr = 0; | |
cd61c3bf | 239 | |
c2ce6a3d | 240 | if (!ret) |
3a6ce860 | 241 | return 0; |
cd61c3bf | 242 | |
4756c94e LP |
243 | if (!pretty) { |
244 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
245 | if (r < 0) | |
246 | return r; | |
247 | ||
248 | pretty = pretty_buffer; | |
249 | } | |
5fc7f358 | 250 | |
01b72568 LP |
251 | fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY); |
252 | if (fd < 0) | |
253 | return -errno; | |
254 | ||
c2ce6a3d | 255 | /* btrfs subvolumes have inode 256 */ |
3a6ce860 | 256 | if (st->st_ino == 256) { |
cd61c3bf | 257 | |
21222ea5 LP |
258 | r = btrfs_is_filesystem(fd); |
259 | if (r < 0) | |
260 | return r; | |
261 | if (r) { | |
10f9c755 | 262 | BtrfsSubvolInfo info; |
cd61c3bf | 263 | |
c2ce6a3d | 264 | /* It's a btrfs subvolume */ |
cd61c3bf | 265 | |
5bcd08db | 266 | r = btrfs_subvol_get_info_fd(fd, 0, &info); |
10f9c755 LP |
267 | if (r < 0) |
268 | return r; | |
c2ce6a3d LP |
269 | |
270 | r = image_new(IMAGE_SUBVOLUME, | |
5fc7f358 | 271 | pretty, |
c2ce6a3d | 272 | path, |
5fc7f358 LP |
273 | filename, |
274 | info.read_only || read_only, | |
10f9c755 | 275 | info.otime, |
c2ce6a3d | 276 | 0, |
c2ce6a3d LP |
277 | ret); |
278 | if (r < 0) | |
279 | return r; | |
280 | ||
5bcd08db LP |
281 | if (btrfs_quota_scan_ongoing(fd) == 0) { |
282 | BtrfsQuotaInfo quota; | |
b6b18498 | 283 | |
5bcd08db LP |
284 | r = btrfs_subvol_get_subtree_quota_fd(fd, 0, "a); |
285 | if (r >= 0) { | |
286 | (*ret)->usage = quota.referenced; | |
287 | (*ret)->usage_exclusive = quota.exclusive; | |
288 | ||
289 | (*ret)->limit = quota.referenced_max; | |
290 | (*ret)->limit_exclusive = quota.exclusive_max; | |
291 | } | |
b6b18498 LP |
292 | } |
293 | ||
3a6ce860 | 294 | return 0; |
cd61c3bf | 295 | } |
c2ce6a3d | 296 | } |
cd61c3bf | 297 | |
01b72568 LP |
298 | /* If the IMMUTABLE bit is set, we consider the |
299 | * directory read-only. Since the ioctl is not | |
300 | * supported everywhere we ignore failures. */ | |
301 | (void) read_attr_fd(fd, &file_attr); | |
cd61c3bf | 302 | |
01b72568 | 303 | /* It's just a normal directory. */ |
c2ce6a3d | 304 | r = image_new(IMAGE_DIRECTORY, |
5fc7f358 | 305 | pretty, |
c2ce6a3d | 306 | path, |
5fc7f358 | 307 | filename, |
01b72568 | 308 | read_only || (file_attr & FS_IMMUTABLE_FL), |
c2ce6a3d LP |
309 | 0, |
310 | 0, | |
311 | ret); | |
312 | if (r < 0) | |
313 | return r; | |
cd61c3bf | 314 | |
3a6ce860 | 315 | return 0; |
cd61c3bf | 316 | |
3a6ce860 | 317 | } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) { |
10f9c755 | 318 | usec_t crtime = 0; |
cd61c3bf | 319 | |
aceac2f0 | 320 | /* It's a RAW disk image */ |
cd61c3bf | 321 | |
c2ce6a3d | 322 | if (!ret) |
3a6ce860 | 323 | return 0; |
cd61c3bf | 324 | |
4756c94e | 325 | (void) fd_getcrtime_at(dfd, filename, &crtime, 0); |
10f9c755 | 326 | |
4756c94e LP |
327 | if (!pretty) { |
328 | r = extract_pretty(filename, ".raw", &pretty_buffer); | |
329 | if (r < 0) | |
330 | return r; | |
331 | ||
332 | pretty = pretty_buffer; | |
333 | } | |
10f9c755 | 334 | |
aceac2f0 | 335 | r = image_new(IMAGE_RAW, |
5fc7f358 | 336 | pretty, |
c2ce6a3d | 337 | path, |
5fc7f358 | 338 | filename, |
3a6ce860 | 339 | !(st->st_mode & 0222) || read_only, |
10f9c755 | 340 | crtime, |
3a6ce860 | 341 | timespec_load(&st->st_mtim), |
c2ce6a3d LP |
342 | ret); |
343 | if (r < 0) | |
344 | return r; | |
cd61c3bf | 345 | |
3a6ce860 LP |
346 | (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512; |
347 | (*ret)->limit = (*ret)->limit_exclusive = st->st_size; | |
b6b18498 | 348 | |
3a6ce860 | 349 | return 0; |
eb38edce | 350 | |
3a6ce860 | 351 | } else if (S_ISBLK(st->st_mode)) { |
eb38edce LP |
352 | _cleanup_close_ int block_fd = -1; |
353 | uint64_t size = UINT64_MAX; | |
354 | ||
355 | /* A block device */ | |
356 | ||
357 | if (!ret) | |
3a6ce860 | 358 | return 0; |
eb38edce | 359 | |
4756c94e LP |
360 | if (!pretty) { |
361 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
362 | if (r < 0) | |
363 | return r; | |
364 | ||
365 | pretty = pretty_buffer; | |
366 | } | |
eb38edce LP |
367 | |
368 | block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
369 | if (block_fd < 0) | |
2570578d | 370 | log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce | 371 | else { |
3a6ce860 LP |
372 | /* Refresh stat data after opening the node */ |
373 | if (fstat(block_fd, &stbuf) < 0) | |
eb38edce | 374 | return -errno; |
3a6ce860 LP |
375 | st = &stbuf; |
376 | ||
377 | if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */ | |
eb38edce LP |
378 | return -ENOTTY; |
379 | ||
380 | if (!read_only) { | |
381 | int state = 0; | |
382 | ||
383 | if (ioctl(block_fd, BLKROGET, &state) < 0) | |
2570578d | 384 | log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
385 | else if (state) |
386 | read_only = true; | |
387 | } | |
388 | ||
389 | if (ioctl(block_fd, BLKGETSIZE64, &size) < 0) | |
2570578d | 390 | log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
391 | |
392 | block_fd = safe_close(block_fd); | |
393 | } | |
394 | ||
395 | r = image_new(IMAGE_BLOCK, | |
396 | pretty, | |
397 | path, | |
398 | filename, | |
3a6ce860 | 399 | !(st->st_mode & 0222) || read_only, |
eb38edce LP |
400 | 0, |
401 | 0, | |
402 | ret); | |
403 | if (r < 0) | |
404 | return r; | |
405 | ||
ed0cb346 | 406 | if (!IN_SET(size, 0, UINT64_MAX)) |
eb38edce LP |
407 | (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size; |
408 | ||
3a6ce860 | 409 | return 0; |
c2ce6a3d | 410 | } |
cd61c3bf | 411 | |
3a6ce860 | 412 | return -EMEDIUMTYPE; |
c2ce6a3d | 413 | } |
cd61c3bf | 414 | |
5ef46e5f | 415 | int image_find(ImageClass class, const char *name, Image **ret) { |
c2ce6a3d LP |
416 | const char *path; |
417 | int r; | |
cd61c3bf | 418 | |
5ef46e5f LP |
419 | assert(class >= 0); |
420 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d | 421 | assert(name); |
cd61c3bf | 422 | |
c2ce6a3d LP |
423 | /* There are no images with invalid names */ |
424 | if (!image_name_is_valid(name)) | |
3a6ce860 | 425 | return -ENOENT; |
cd61c3bf | 426 | |
5ef46e5f | 427 | NULSTR_FOREACH(path, image_search_path[class]) { |
c2ce6a3d | 428 | _cleanup_closedir_ DIR *d = NULL; |
3a6ce860 | 429 | struct stat st; |
cd61c3bf | 430 | |
c2ce6a3d LP |
431 | d = opendir(path); |
432 | if (!d) { | |
433 | if (errno == ENOENT) | |
434 | continue; | |
cd61c3bf | 435 | |
c2ce6a3d LP |
436 | return -errno; |
437 | } | |
cd61c3bf | 438 | |
3a6ce860 LP |
439 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to |
440 | * symlink block devices into the search path */ | |
441 | if (fstatat(dirfd(d), name, &st, 0) < 0) { | |
aceac2f0 | 442 | _cleanup_free_ char *raw = NULL; |
5fc7f358 | 443 | |
3a6ce860 LP |
444 | if (errno != ENOENT) |
445 | return -errno; | |
446 | ||
b910cc72 | 447 | raw = strjoin(name, ".raw"); |
aceac2f0 | 448 | if (!raw) |
5fc7f358 LP |
449 | return -ENOMEM; |
450 | ||
3a6ce860 LP |
451 | if (fstatat(dirfd(d), raw, &st, 0) < 0) { |
452 | ||
453 | if (errno == ENOENT) | |
454 | continue; | |
455 | ||
456 | return -errno; | |
457 | } | |
458 | ||
459 | if (!S_ISREG(st.st_mode)) | |
5fc7f358 | 460 | continue; |
3a6ce860 LP |
461 | |
462 | r = image_make(name, dirfd(d), path, raw, &st, ret); | |
463 | ||
464 | } else { | |
465 | if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode)) | |
466 | continue; | |
467 | ||
468 | r = image_make(name, dirfd(d), path, name, &st, ret); | |
5fc7f358 | 469 | } |
3a6ce860 LP |
470 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) |
471 | continue; | |
c2ce6a3d LP |
472 | if (r < 0) |
473 | return r; | |
cd61c3bf | 474 | |
cf604fd4 LP |
475 | if (ret) |
476 | (*ret)->discoverable = true; | |
477 | ||
c2ce6a3d LP |
478 | return 1; |
479 | } | |
480 | ||
cf604fd4 LP |
481 | if (class == IMAGE_MACHINE && streq(name, ".host")) { |
482 | r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret); | |
483 | if (r < 0) | |
484 | return r; | |
485 | ||
486 | if (ret) | |
487 | (*ret)->discoverable = true; | |
488 | ||
489 | return r; | |
490 | } | |
5fc7f358 | 491 | |
3a6ce860 | 492 | return -ENOENT; |
c2ce6a3d LP |
493 | }; |
494 | ||
2ddf182b | 495 | int image_from_path(const char *path, Image **ret) { |
cf604fd4 LP |
496 | |
497 | /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether | |
498 | * 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 | 499 | * overridden by another, different image earlier in the search path */ |
cf604fd4 | 500 | |
2ddf182b LP |
501 | if (path_equal(path, "/")) |
502 | return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret); | |
503 | ||
504 | return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret); | |
505 | } | |
506 | ||
507 | int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) { | |
508 | if (image_name_is_valid(name_or_path)) | |
509 | return image_find(class, name_or_path, ret); | |
510 | ||
511 | return image_from_path(name_or_path, ret); | |
512 | } | |
513 | ||
5ef46e5f | 514 | int image_discover(ImageClass class, Hashmap *h) { |
c2ce6a3d LP |
515 | const char *path; |
516 | int r; | |
517 | ||
5ef46e5f LP |
518 | assert(class >= 0); |
519 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d LP |
520 | assert(h); |
521 | ||
5ef46e5f | 522 | NULSTR_FOREACH(path, image_search_path[class]) { |
c2ce6a3d LP |
523 | _cleanup_closedir_ DIR *d = NULL; |
524 | struct dirent *de; | |
525 | ||
526 | d = opendir(path); | |
527 | if (!d) { | |
528 | if (errno == ENOENT) | |
a67a4c8c | 529 | continue; |
c2ce6a3d LP |
530 | |
531 | return -errno; | |
532 | } | |
533 | ||
534 | FOREACH_DIRENT_ALL(de, d, return -errno) { | |
535 | _cleanup_(image_unrefp) Image *image = NULL; | |
1bf36bf9 | 536 | _cleanup_free_ char *truncated = NULL; |
3a6ce860 LP |
537 | const char *pretty; |
538 | struct stat st; | |
c2ce6a3d | 539 | |
1bf36bf9 | 540 | if (dot_or_dot_dot(de->d_name)) |
c2ce6a3d LP |
541 | continue; |
542 | ||
3a6ce860 LP |
543 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people |
544 | * to symlink block devices into the search path */ | |
545 | if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) { | |
546 | if (errno == ENOENT) | |
547 | continue; | |
548 | ||
549 | return -errno; | |
550 | } | |
551 | ||
552 | if (S_ISREG(st.st_mode)) { | |
553 | const char *e; | |
554 | ||
555 | e = endswith(de->d_name, ".raw"); | |
556 | if (!e) | |
557 | continue; | |
558 | ||
1bf36bf9 LP |
559 | truncated = strndup(de->d_name, e - de->d_name); |
560 | if (!truncated) | |
561 | return -ENOMEM; | |
562 | ||
563 | pretty = truncated; | |
3a6ce860 | 564 | } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode)) |
1bf36bf9 | 565 | pretty = de->d_name; |
3a6ce860 LP |
566 | else |
567 | continue; | |
1bf36bf9 LP |
568 | |
569 | if (!image_name_is_valid(pretty)) | |
570 | continue; | |
571 | ||
572 | if (hashmap_contains(h, pretty)) | |
c2ce6a3d LP |
573 | continue; |
574 | ||
3a6ce860 LP |
575 | r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image); |
576 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) | |
c2ce6a3d LP |
577 | continue; |
578 | if (r < 0) | |
579 | return r; | |
580 | ||
cf604fd4 LP |
581 | image->discoverable = true; |
582 | ||
c2ce6a3d LP |
583 | r = hashmap_put(h, image->name, image); |
584 | if (r < 0) | |
585 | return r; | |
586 | ||
587 | image = NULL; | |
cd61c3bf LP |
588 | } |
589 | } | |
590 | ||
5ef46e5f | 591 | if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) { |
5fc7f358 LP |
592 | _cleanup_(image_unrefp) Image *image = NULL; |
593 | ||
3a6ce860 | 594 | r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image); |
5fc7f358 LP |
595 | if (r < 0) |
596 | return r; | |
597 | ||
cf604fd4 LP |
598 | image->discoverable = true; |
599 | ||
5fc7f358 LP |
600 | r = hashmap_put(h, image->name, image); |
601 | if (r < 0) | |
602 | return r; | |
603 | ||
604 | image = NULL; | |
5fc7f358 LP |
605 | } |
606 | ||
cd61c3bf LP |
607 | return 0; |
608 | } | |
609 | ||
08682124 | 610 | int image_remove(Image *i) { |
8e766630 | 611 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
8e0b6570 | 612 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 613 | _cleanup_free_ char *roothash = NULL; |
8e0b6570 | 614 | char **j; |
30535c16 LP |
615 | int r; |
616 | ||
08682124 LP |
617 | assert(i); |
618 | ||
d94c2b06 | 619 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
08682124 LP |
620 | return -EROFS; |
621 | ||
8e0b6570 LP |
622 | settings = image_settings_path(i); |
623 | if (!settings) | |
624 | return -ENOMEM; | |
625 | ||
bafbac4e LP |
626 | roothash = image_roothash_path(i); |
627 | if (!roothash) | |
628 | return -ENOMEM; | |
629 | ||
30535c16 LP |
630 | /* Make sure we don't interfere with a running nspawn */ |
631 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
632 | if (r < 0) | |
633 | return r; | |
634 | ||
ebd93cb6 LP |
635 | switch (i->type) { |
636 | ||
637 | case IMAGE_SUBVOLUME: | |
9fb0b9c7 LP |
638 | |
639 | /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the | |
640 | * big guns */ | |
641 | if (unlink(i->path) < 0) { | |
642 | r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA); | |
643 | if (r < 0) | |
644 | return r; | |
645 | } | |
646 | ||
8e0b6570 | 647 | break; |
ebd93cb6 LP |
648 | |
649 | case IMAGE_DIRECTORY: | |
01b72568 | 650 | /* Allow deletion of read-only directories */ |
db9a4254 | 651 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
8e0b6570 LP |
652 | r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME); |
653 | if (r < 0) | |
654 | return r; | |
655 | ||
656 | break; | |
01b72568 | 657 | |
eb38edce LP |
658 | case IMAGE_BLOCK: |
659 | ||
660 | /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node | |
661 | * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink | |
662 | * the thing (it's most likely a symlink after all). */ | |
663 | ||
664 | if (path_startswith(i->path, "/dev")) | |
665 | break; | |
666 | ||
4831981d | 667 | _fallthrough_; |
aceac2f0 | 668 | case IMAGE_RAW: |
41d1ed05 LP |
669 | if (unlink(i->path) < 0) |
670 | return -errno; | |
8e0b6570 | 671 | break; |
ebd93cb6 LP |
672 | |
673 | default: | |
15411c0c | 674 | return -EOPNOTSUPP; |
ebd93cb6 | 675 | } |
8e0b6570 LP |
676 | |
677 | STRV_FOREACH(j, settings) { | |
678 | if (unlink(*j) < 0 && errno != ENOENT) | |
679 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j); | |
680 | } | |
681 | ||
bafbac4e LP |
682 | if (unlink(roothash) < 0 && errno != ENOENT) |
683 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash); | |
684 | ||
8e0b6570 LP |
685 | return 0; |
686 | } | |
687 | ||
bafbac4e | 688 | static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
8e0b6570 LP |
689 | _cleanup_free_ char *rs = NULL; |
690 | const char *fn; | |
691 | ||
bafbac4e | 692 | fn = strjoina(new_name, suffix); |
8e0b6570 LP |
693 | |
694 | rs = file_in_same_dir(path, fn); | |
695 | if (!rs) | |
696 | return -ENOMEM; | |
697 | ||
698 | return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs); | |
ebd93cb6 LP |
699 | } |
700 | ||
701 | int image_rename(Image *i, const char *new_name) { | |
8e766630 | 702 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT; |
bafbac4e | 703 | _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL; |
8e0b6570 | 704 | _cleanup_strv_free_ char **settings = NULL; |
01b72568 | 705 | unsigned file_attr = 0; |
8e0b6570 | 706 | char **j; |
ebd93cb6 LP |
707 | int r; |
708 | ||
709 | assert(i); | |
710 | ||
711 | if (!image_name_is_valid(new_name)) | |
712 | return -EINVAL; | |
713 | ||
d94c2b06 | 714 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
715 | return -EROFS; |
716 | ||
8e0b6570 LP |
717 | settings = image_settings_path(i); |
718 | if (!settings) | |
719 | return -ENOMEM; | |
720 | ||
bafbac4e LP |
721 | roothash = image_roothash_path(i); |
722 | if (!roothash) | |
723 | return -ENOMEM; | |
724 | ||
30535c16 LP |
725 | /* Make sure we don't interfere with a running nspawn */ |
726 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
727 | if (r < 0) | |
728 | return r; | |
729 | ||
730 | /* Make sure nobody takes the new name, between the time we | |
731 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 732 | * time we take possession of it */ |
30535c16 LP |
733 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
734 | if (r < 0) | |
735 | return r; | |
736 | ||
5ef46e5f | 737 | r = image_find(IMAGE_MACHINE, new_name, NULL); |
3a6ce860 | 738 | if (r >= 0) |
ebd93cb6 | 739 | return -EEXIST; |
3a6ce860 LP |
740 | if (r != -ENOENT) |
741 | return r; | |
ebd93cb6 LP |
742 | |
743 | switch (i->type) { | |
744 | ||
ebd93cb6 | 745 | case IMAGE_DIRECTORY: |
01b72568 LP |
746 | /* Turn of the immutable bit while we rename the image, so that we can rename it */ |
747 | (void) read_attr_path(i->path, &file_attr); | |
748 | ||
749 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 750 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
01b72568 | 751 | |
4831981d | 752 | _fallthrough_; |
01b72568 | 753 | case IMAGE_SUBVOLUME: |
ebd93cb6 LP |
754 | new_path = file_in_same_dir(i->path, new_name); |
755 | break; | |
756 | ||
eb38edce LP |
757 | case IMAGE_BLOCK: |
758 | ||
759 | /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */ | |
760 | if (path_startswith(i->path, "/dev")) | |
761 | return -EROFS; | |
762 | ||
763 | new_path = file_in_same_dir(i->path, new_name); | |
764 | break; | |
765 | ||
aceac2f0 | 766 | case IMAGE_RAW: { |
ebd93cb6 LP |
767 | const char *fn; |
768 | ||
63c372cb | 769 | fn = strjoina(new_name, ".raw"); |
ebd93cb6 LP |
770 | new_path = file_in_same_dir(i->path, fn); |
771 | break; | |
772 | } | |
773 | ||
774 | default: | |
15411c0c | 775 | return -EOPNOTSUPP; |
ebd93cb6 LP |
776 | } |
777 | ||
778 | if (!new_path) | |
779 | return -ENOMEM; | |
780 | ||
781 | nn = strdup(new_name); | |
782 | if (!nn) | |
783 | return -ENOMEM; | |
784 | ||
f85ef957 AC |
785 | r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path); |
786 | if (r < 0) | |
787 | return r; | |
ebd93cb6 | 788 | |
01b72568 LP |
789 | /* Restore the immutable bit, if it was set before */ |
790 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 791 | (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL); |
01b72568 | 792 | |
f9ecfd3b DL |
793 | free_and_replace(i->path, new_path); |
794 | free_and_replace(i->name, nn); | |
ebd93cb6 | 795 | |
8e0b6570 | 796 | STRV_FOREACH(j, settings) { |
bafbac4e | 797 | r = rename_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
798 | if (r < 0 && r != -ENOENT) |
799 | log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j); | |
800 | } | |
801 | ||
bafbac4e LP |
802 | r = rename_auxiliary_file(roothash, new_name, ".roothash"); |
803 | if (r < 0 && r != -ENOENT) | |
804 | log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash); | |
805 | ||
ebd93cb6 LP |
806 | return 0; |
807 | } | |
808 | ||
bafbac4e | 809 | static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
8e0b6570 LP |
810 | _cleanup_free_ char *rs = NULL; |
811 | const char *fn; | |
812 | ||
bafbac4e | 813 | fn = strjoina(new_name, suffix); |
8e0b6570 LP |
814 | |
815 | rs = file_in_same_dir(path, fn); | |
816 | if (!rs) | |
817 | return -ENOMEM; | |
818 | ||
8a016c74 | 819 | return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK); |
8e0b6570 LP |
820 | } |
821 | ||
ebd93cb6 | 822 | int image_clone(Image *i, const char *new_name, bool read_only) { |
8e766630 | 823 | _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT; |
8e0b6570 | 824 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 825 | _cleanup_free_ char *roothash = NULL; |
ebd93cb6 | 826 | const char *new_path; |
8e0b6570 | 827 | char **j; |
ebd93cb6 LP |
828 | int r; |
829 | ||
830 | assert(i); | |
831 | ||
832 | if (!image_name_is_valid(new_name)) | |
833 | return -EINVAL; | |
834 | ||
8e0b6570 LP |
835 | settings = image_settings_path(i); |
836 | if (!settings) | |
837 | return -ENOMEM; | |
838 | ||
bafbac4e LP |
839 | roothash = image_roothash_path(i); |
840 | if (!roothash) | |
841 | return -ENOMEM; | |
842 | ||
30535c16 LP |
843 | /* Make sure nobody takes the new name, between the time we |
844 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 845 | * time we take possession of it */ |
30535c16 LP |
846 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
847 | if (r < 0) | |
848 | return r; | |
849 | ||
5ef46e5f | 850 | r = image_find(IMAGE_MACHINE, new_name, NULL); |
3a6ce860 | 851 | if (r >= 0) |
ebd93cb6 | 852 | return -EEXIST; |
3a6ce860 LP |
853 | if (r != -ENOENT) |
854 | return r; | |
ebd93cb6 LP |
855 | |
856 | switch (i->type) { | |
857 | ||
858 | case IMAGE_SUBVOLUME: | |
859 | case IMAGE_DIRECTORY: | |
9a50e3ca | 860 | /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain |
13e785f7 | 861 | * directory. */ |
9a50e3ca | 862 | |
63c372cb | 863 | new_path = strjoina("/var/lib/machines/", new_name); |
ebd93cb6 | 864 | |
17cbb288 LP |
865 | r = btrfs_subvol_snapshot(i->path, new_path, |
866 | (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | | |
867 | BTRFS_SNAPSHOT_FALLBACK_COPY | | |
868 | BTRFS_SNAPSHOT_FALLBACK_DIRECTORY | | |
869 | BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE | | |
870 | BTRFS_SNAPSHOT_RECURSIVE | | |
871 | BTRFS_SNAPSHOT_QUOTA); | |
872 | if (r >= 0) | |
9a50e3ca | 873 | /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */ |
8120ee28 | 874 | (void) btrfs_subvol_auto_qgroup(new_path, 0, true); |
5bcd08db | 875 | |
ebd93cb6 LP |
876 | break; |
877 | ||
aceac2f0 | 878 | case IMAGE_RAW: |
63c372cb | 879 | new_path = strjoina("/var/lib/machines/", new_name, ".raw"); |
ebd93cb6 | 880 | |
8a016c74 | 881 | r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME); |
ebd93cb6 LP |
882 | break; |
883 | ||
eb38edce | 884 | case IMAGE_BLOCK: |
ebd93cb6 | 885 | default: |
15411c0c | 886 | return -EOPNOTSUPP; |
ebd93cb6 LP |
887 | } |
888 | ||
889 | if (r < 0) | |
890 | return r; | |
891 | ||
8e0b6570 | 892 | STRV_FOREACH(j, settings) { |
bafbac4e | 893 | r = clone_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
894 | if (r < 0 && r != -ENOENT) |
895 | log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j); | |
896 | } | |
897 | ||
bafbac4e LP |
898 | r = clone_auxiliary_file(roothash, new_name, ".roothash"); |
899 | if (r < 0 && r != -ENOENT) | |
900 | log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash); | |
901 | ||
ebd93cb6 LP |
902 | return 0; |
903 | } | |
904 | ||
905 | int image_read_only(Image *i, bool b) { | |
8e766630 | 906 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
ebd93cb6 | 907 | int r; |
c7664c07 | 908 | |
ebd93cb6 LP |
909 | assert(i); |
910 | ||
d94c2b06 | 911 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
912 | return -EROFS; |
913 | ||
30535c16 LP |
914 | /* Make sure we don't interfere with a running nspawn */ |
915 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
916 | if (r < 0) | |
917 | return r; | |
918 | ||
ebd93cb6 LP |
919 | switch (i->type) { |
920 | ||
921 | case IMAGE_SUBVOLUME: | |
5bcd08db LP |
922 | |
923 | /* Note that we set the flag only on the top-level | |
924 | * subvolume of the image. */ | |
925 | ||
ebd93cb6 LP |
926 | r = btrfs_subvol_set_read_only(i->path, b); |
927 | if (r < 0) | |
928 | return r; | |
01b72568 LP |
929 | |
930 | break; | |
931 | ||
932 | case IMAGE_DIRECTORY: | |
933 | /* For simple directory trees we cannot use the access | |
934 | mode of the top-level directory, since it has an | |
935 | effect on the container itself. However, we can | |
936 | use the "immutable" flag, to at least make the | |
937 | top-level directory read-only. It's not as good as | |
938 | a read-only subvolume, but at least something, and | |
13e785f7 | 939 | we can read the value back. */ |
01b72568 | 940 | |
db9a4254 | 941 | r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL); |
01b72568 LP |
942 | if (r < 0) |
943 | return r; | |
944 | ||
ebd93cb6 LP |
945 | break; |
946 | ||
aceac2f0 | 947 | case IMAGE_RAW: { |
ebd93cb6 LP |
948 | struct stat st; |
949 | ||
950 | if (stat(i->path, &st) < 0) | |
951 | return -errno; | |
952 | ||
953 | if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0) | |
954 | return -errno; | |
f2068bcc LP |
955 | |
956 | /* If the images is now read-only, it's a good time to | |
957 | * defrag it, given that no write patterns will | |
958 | * fragment it again. */ | |
959 | if (b) | |
960 | (void) btrfs_defrag(i->path); | |
ebd93cb6 LP |
961 | break; |
962 | } | |
963 | ||
eb38edce LP |
964 | case IMAGE_BLOCK: { |
965 | _cleanup_close_ int fd = -1; | |
966 | struct stat st; | |
967 | int state = b; | |
968 | ||
969 | fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY); | |
970 | if (fd < 0) | |
971 | return -errno; | |
972 | ||
973 | if (fstat(fd, &st) < 0) | |
974 | return -errno; | |
975 | if (!S_ISBLK(st.st_mode)) | |
976 | return -ENOTTY; | |
977 | ||
978 | if (ioctl(fd, BLKROSET, &state) < 0) | |
979 | return -errno; | |
980 | ||
981 | break; | |
982 | } | |
983 | ||
ebd93cb6 | 984 | default: |
15411c0c | 985 | return -EOPNOTSUPP; |
ebd93cb6 LP |
986 | } |
987 | ||
988 | return 0; | |
08682124 LP |
989 | } |
990 | ||
30535c16 LP |
991 | int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) { |
992 | _cleanup_free_ char *p = NULL; | |
993 | LockFile t = LOCK_FILE_INIT; | |
994 | struct stat st; | |
f25bed67 | 995 | bool exclusive; |
30535c16 LP |
996 | int r; |
997 | ||
998 | assert(path); | |
999 | assert(global); | |
1000 | assert(local); | |
1001 | ||
f25bed67 LP |
1002 | /* Locks an image path. This actually creates two locks: one "local" one, next to the image path |
1003 | * itself, which might be shared via NFS. And another "global" one, in /run, that uses the | |
1004 | * device/inode number. This has the benefit that we can even lock a tree that is a mount point, | |
1005 | * correctly. */ | |
30535c16 | 1006 | |
30535c16 LP |
1007 | if (!path_is_absolute(path)) |
1008 | return -EINVAL; | |
1009 | ||
f25bed67 LP |
1010 | switch (operation & (LOCK_SH|LOCK_EX)) { |
1011 | case LOCK_SH: | |
1012 | exclusive = false; | |
1013 | break; | |
1014 | case LOCK_EX: | |
1015 | exclusive = true; | |
1016 | break; | |
1017 | default: | |
1018 | return -EINVAL; | |
1019 | } | |
1020 | ||
b6e953f2 LP |
1021 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1022 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1023 | return 0; | |
1024 | } | |
1025 | ||
f25bed67 LP |
1026 | /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are |
1027 | * running off it after all, and we don't want any images to manipulate the host image. We make an | |
1028 | * exception for shared locks however: we allow those (and make them NOPs since there's no point in | |
1029 | * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well, | |
1030 | * since it means changes made to the host might propagate to the container as they happen (and a | |
1031 | * shared lock kinda suggests that no changes happen at all while it is in place), but it's too | |
1032 | * useful not to allow read-only containers off the host root, hence let's support this, and trust | |
1033 | * the user to do the right thing with this. */ | |
1034 | if (path_equal(path, "/")) { | |
1035 | if (exclusive) | |
1036 | return -EBUSY; | |
1037 | ||
1038 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1039 | return 0; | |
1040 | } | |
b6e953f2 | 1041 | |
30535c16 | 1042 | if (stat(path, &st) >= 0) { |
eb38edce LP |
1043 | if (S_ISBLK(st.st_mode)) |
1044 | r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev)); | |
1045 | else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode)) | |
1046 | r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino); | |
1047 | else | |
1048 | return -ENOTTY; | |
1049 | ||
1050 | if (r < 0) | |
30535c16 LP |
1051 | return -ENOMEM; |
1052 | } | |
1053 | ||
f25bed67 LP |
1054 | /* For block devices we don't need the "local" lock, as the major/minor lock above should be |
1055 | * sufficient, since block devices are host local anyway. */ | |
1056 | if (!path_startswith(path, "/dev/")) { | |
eb38edce | 1057 | r = make_lock_file_for(path, operation, &t); |
8be17c9b | 1058 | if (r < 0) { |
f25bed67 | 1059 | if (!exclusive && r == -EROFS) |
771b7ead | 1060 | log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path); |
8be17c9b LT |
1061 | else |
1062 | return r; | |
1063 | } | |
eb38edce | 1064 | } |
30535c16 LP |
1065 | |
1066 | if (p) { | |
6e5dcce4 | 1067 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); |
30535c16 LP |
1068 | |
1069 | r = make_lock_file(p, operation, global); | |
1070 | if (r < 0) { | |
1071 | release_lock_file(&t); | |
1072 | return r; | |
1073 | } | |
546dbec5 LP |
1074 | } else |
1075 | *global = (LockFile) LOCK_FILE_INIT; | |
30535c16 LP |
1076 | |
1077 | *local = t; | |
1078 | return 0; | |
1079 | } | |
1080 | ||
cb81cd80 | 1081 | int image_set_limit(Image *i, uint64_t referenced_max) { |
d6ce17c7 LP |
1082 | assert(i); |
1083 | ||
d94c2b06 | 1084 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
d6ce17c7 LP |
1085 | return -EROFS; |
1086 | ||
1087 | if (i->type != IMAGE_SUBVOLUME) | |
15411c0c | 1088 | return -EOPNOTSUPP; |
d6ce17c7 | 1089 | |
5bcd08db LP |
1090 | /* We set the quota both for the subvolume as well as for the |
1091 | * subtree. The latter is mostly for historical reasons, since | |
1092 | * we didn't use to have a concept of subtree quota, and hence | |
1093 | * only modified the subvolume quota. */ | |
1094 | ||
1095 | (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max); | |
1096 | (void) btrfs_subvol_auto_qgroup(i->path, 0, true); | |
1097 | return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max); | |
d6ce17c7 LP |
1098 | } |
1099 | ||
c7664c07 | 1100 | int image_read_metadata(Image *i) { |
8e766630 | 1101 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
c7664c07 LP |
1102 | int r; |
1103 | ||
1104 | assert(i); | |
1105 | ||
1106 | r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock); | |
1107 | if (r < 0) | |
1108 | return r; | |
1109 | ||
1110 | switch (i->type) { | |
1111 | ||
1112 | case IMAGE_SUBVOLUME: | |
1113 | case IMAGE_DIRECTORY: { | |
1114 | _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL; | |
1115 | sd_id128_t machine_id = SD_ID128_NULL; | |
1116 | _cleanup_free_ char *hostname = NULL; | |
1117 | _cleanup_free_ char *path = NULL; | |
1118 | ||
a5648b80 | 1119 | r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1120 | if (r < 0 && r != -ENOENT) |
1121 | log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name); | |
1122 | else if (r >= 0) { | |
1123 | r = read_etc_hostname(path, &hostname); | |
1124 | if (r < 0) | |
1125 | log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name); | |
1126 | } | |
1127 | ||
1128 | path = mfree(path); | |
1129 | ||
a5648b80 | 1130 | r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1131 | if (r < 0 && r != -ENOENT) |
1132 | log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name); | |
1133 | else if (r >= 0) { | |
1134 | _cleanup_close_ int fd = -1; | |
1135 | ||
1136 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
1137 | if (fd < 0) | |
1138 | log_debug_errno(errno, "Failed to open %s: %m", path); | |
1139 | else { | |
1140 | r = id128_read_fd(fd, ID128_PLAIN, &machine_id); | |
1141 | if (r < 0) | |
1142 | log_debug_errno(r, "Image %s contains invalid machine ID.", i->name); | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | path = mfree(path); | |
1147 | ||
a5648b80 | 1148 | r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1149 | if (r < 0 && r != -ENOENT) |
1150 | log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name); | |
1151 | else if (r >= 0) { | |
aa8fbc74 | 1152 | r = load_env_file_pairs(NULL, path, &machine_info); |
c7664c07 LP |
1153 | if (r < 0) |
1154 | log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name); | |
1155 | } | |
1156 | ||
d58ad743 LP |
1157 | r = load_os_release_pairs(i->path, &os_release); |
1158 | if (r < 0) | |
1159 | log_debug_errno(r, "Failed to read os-release in image, ignoring: %m"); | |
c7664c07 LP |
1160 | |
1161 | free_and_replace(i->hostname, hostname); | |
1162 | i->machine_id = machine_id; | |
1163 | strv_free_and_replace(i->machine_info, machine_info); | |
1164 | strv_free_and_replace(i->os_release, os_release); | |
1165 | ||
1166 | break; | |
1167 | } | |
1168 | ||
1169 | case IMAGE_RAW: | |
1170 | case IMAGE_BLOCK: { | |
1171 | _cleanup_(loop_device_unrefp) LoopDevice *d = NULL; | |
1172 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
1173 | ||
e08f94ac | 1174 | r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, &d); |
c7664c07 LP |
1175 | if (r < 0) |
1176 | return r; | |
1177 | ||
89e62e0b | 1178 | r = dissect_image(d->fd, NULL, NULL, DISSECT_IMAGE_REQUIRE_ROOT|DISSECT_IMAGE_RELAX_VAR_CHECK, &m); |
c7664c07 LP |
1179 | if (r < 0) |
1180 | return r; | |
1181 | ||
1182 | r = dissected_image_acquire_metadata(m); | |
1183 | if (r < 0) | |
1184 | return r; | |
1185 | ||
1186 | free_and_replace(i->hostname, m->hostname); | |
1187 | i->machine_id = m->machine_id; | |
1188 | strv_free_and_replace(i->machine_info, m->machine_info); | |
1189 | strv_free_and_replace(i->os_release, m->os_release); | |
1190 | ||
1191 | break; | |
1192 | } | |
1193 | ||
1194 | default: | |
1195 | return -EOPNOTSUPP; | |
1196 | } | |
1197 | ||
1198 | i->metadata_valid = true; | |
1199 | ||
1200 | return 0; | |
1201 | } | |
1202 | ||
30535c16 | 1203 | int image_name_lock(const char *name, int operation, LockFile *ret) { |
30535c16 LP |
1204 | assert(name); |
1205 | assert(ret); | |
1206 | ||
1207 | /* Locks an image name, regardless of the precise path used. */ | |
1208 | ||
1209 | if (!image_name_is_valid(name)) | |
1210 | return -EINVAL; | |
1211 | ||
b6e953f2 LP |
1212 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1213 | *ret = (LockFile) LOCK_FILE_INIT; | |
1214 | return 0; | |
1215 | } | |
1216 | ||
30535c16 LP |
1217 | if (streq(name, ".host")) |
1218 | return -EBUSY; | |
1219 | ||
6e5dcce4 ZJS |
1220 | const char *p = strjoina("/run/systemd/nspawn/locks/name-", name); |
1221 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); | |
30535c16 LP |
1222 | return make_lock_file(p, operation, ret); |
1223 | } | |
1224 | ||
1225 | bool image_name_is_valid(const char *s) { | |
1226 | if (!filename_is_valid(s)) | |
1227 | return false; | |
1228 | ||
1229 | if (string_has_cc(s, NULL)) | |
1230 | return false; | |
1231 | ||
1232 | if (!utf8_is_valid(s)) | |
1233 | return false; | |
1234 | ||
1235 | /* Temporary files for atomically creating new files */ | |
1236 | if (startswith(s, ".#")) | |
1237 | return false; | |
1238 | ||
1239 | return true; | |
1240 | } | |
1241 | ||
ace9ab19 LP |
1242 | bool image_in_search_path(ImageClass class, const char *image) { |
1243 | const char *path; | |
1244 | ||
1245 | assert(image); | |
1246 | ||
1247 | NULSTR_FOREACH(path, image_search_path[class]) { | |
1248 | const char *p; | |
1249 | size_t k; | |
1250 | ||
1251 | p = path_startswith(image, path); | |
1252 | if (!p) | |
1253 | continue; | |
1254 | ||
1255 | /* Make sure there's a filename following */ | |
1256 | k = strcspn(p, "/"); | |
1257 | if (k == 0) | |
1258 | continue; | |
1259 | ||
1260 | p += k; | |
1261 | ||
1262 | /* Accept trailing slashes */ | |
1263 | if (p[strspn(p, "/")] == 0) | |
1264 | return true; | |
1265 | ||
1266 | } | |
1267 | ||
1268 | return false; | |
1269 | } | |
1270 | ||
cd61c3bf LP |
1271 | static const char* const image_type_table[_IMAGE_TYPE_MAX] = { |
1272 | [IMAGE_DIRECTORY] = "directory", | |
1273 | [IMAGE_SUBVOLUME] = "subvolume", | |
aceac2f0 | 1274 | [IMAGE_RAW] = "raw", |
eb38edce | 1275 | [IMAGE_BLOCK] = "block", |
cd61c3bf LP |
1276 | }; |
1277 | ||
1278 | DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType); |