]>
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; | |
02dcf215 | 239 | usec_t crtime = 0; |
cd61c3bf | 240 | |
c2ce6a3d | 241 | if (!ret) |
3a6ce860 | 242 | return 0; |
cd61c3bf | 243 | |
4756c94e LP |
244 | if (!pretty) { |
245 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
246 | if (r < 0) | |
247 | return r; | |
248 | ||
249 | pretty = pretty_buffer; | |
250 | } | |
5fc7f358 | 251 | |
01b72568 LP |
252 | fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY); |
253 | if (fd < 0) | |
254 | return -errno; | |
255 | ||
c2ce6a3d | 256 | /* btrfs subvolumes have inode 256 */ |
3a6ce860 | 257 | if (st->st_ino == 256) { |
cd61c3bf | 258 | |
21222ea5 LP |
259 | r = btrfs_is_filesystem(fd); |
260 | if (r < 0) | |
261 | return r; | |
262 | if (r) { | |
10f9c755 | 263 | BtrfsSubvolInfo info; |
cd61c3bf | 264 | |
c2ce6a3d | 265 | /* It's a btrfs subvolume */ |
cd61c3bf | 266 | |
5bcd08db | 267 | r = btrfs_subvol_get_info_fd(fd, 0, &info); |
10f9c755 LP |
268 | if (r < 0) |
269 | return r; | |
c2ce6a3d LP |
270 | |
271 | r = image_new(IMAGE_SUBVOLUME, | |
5fc7f358 | 272 | pretty, |
c2ce6a3d | 273 | path, |
5fc7f358 LP |
274 | filename, |
275 | info.read_only || read_only, | |
10f9c755 | 276 | info.otime, |
c2ce6a3d | 277 | 0, |
c2ce6a3d LP |
278 | ret); |
279 | if (r < 0) | |
280 | return r; | |
281 | ||
5bcd08db LP |
282 | if (btrfs_quota_scan_ongoing(fd) == 0) { |
283 | BtrfsQuotaInfo quota; | |
b6b18498 | 284 | |
5bcd08db LP |
285 | r = btrfs_subvol_get_subtree_quota_fd(fd, 0, "a); |
286 | if (r >= 0) { | |
287 | (*ret)->usage = quota.referenced; | |
288 | (*ret)->usage_exclusive = quota.exclusive; | |
289 | ||
290 | (*ret)->limit = quota.referenced_max; | |
291 | (*ret)->limit_exclusive = quota.exclusive_max; | |
292 | } | |
b6b18498 LP |
293 | } |
294 | ||
3a6ce860 | 295 | return 0; |
cd61c3bf | 296 | } |
c2ce6a3d | 297 | } |
cd61c3bf | 298 | |
02dcf215 LP |
299 | /* Get directory creation time (not available everywhere, but that's OK */ |
300 | (void) fd_getcrtime(dfd, &crtime); | |
301 | ||
302 | /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not | |
01b72568 LP |
303 | * supported everywhere we ignore failures. */ |
304 | (void) read_attr_fd(fd, &file_attr); | |
cd61c3bf | 305 | |
01b72568 | 306 | /* It's just a normal directory. */ |
c2ce6a3d | 307 | r = image_new(IMAGE_DIRECTORY, |
5fc7f358 | 308 | pretty, |
c2ce6a3d | 309 | path, |
5fc7f358 | 310 | filename, |
01b72568 | 311 | read_only || (file_attr & FS_IMMUTABLE_FL), |
02dcf215 LP |
312 | crtime, |
313 | 0, /* we don't use mtime of stat() here, since it's not the time of last change of the tree, but only of the top-level dir */ | |
c2ce6a3d LP |
314 | ret); |
315 | if (r < 0) | |
316 | return r; | |
cd61c3bf | 317 | |
3a6ce860 | 318 | return 0; |
cd61c3bf | 319 | |
3a6ce860 | 320 | } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) { |
10f9c755 | 321 | usec_t crtime = 0; |
cd61c3bf | 322 | |
aceac2f0 | 323 | /* It's a RAW disk image */ |
cd61c3bf | 324 | |
c2ce6a3d | 325 | if (!ret) |
3a6ce860 | 326 | return 0; |
cd61c3bf | 327 | |
4756c94e | 328 | (void) fd_getcrtime_at(dfd, filename, &crtime, 0); |
10f9c755 | 329 | |
4756c94e LP |
330 | if (!pretty) { |
331 | r = extract_pretty(filename, ".raw", &pretty_buffer); | |
332 | if (r < 0) | |
333 | return r; | |
334 | ||
335 | pretty = pretty_buffer; | |
336 | } | |
10f9c755 | 337 | |
aceac2f0 | 338 | r = image_new(IMAGE_RAW, |
5fc7f358 | 339 | pretty, |
c2ce6a3d | 340 | path, |
5fc7f358 | 341 | filename, |
3a6ce860 | 342 | !(st->st_mode & 0222) || read_only, |
10f9c755 | 343 | crtime, |
3a6ce860 | 344 | timespec_load(&st->st_mtim), |
c2ce6a3d LP |
345 | ret); |
346 | if (r < 0) | |
347 | return r; | |
cd61c3bf | 348 | |
3a6ce860 LP |
349 | (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512; |
350 | (*ret)->limit = (*ret)->limit_exclusive = st->st_size; | |
b6b18498 | 351 | |
3a6ce860 | 352 | return 0; |
eb38edce | 353 | |
3a6ce860 | 354 | } else if (S_ISBLK(st->st_mode)) { |
eb38edce LP |
355 | _cleanup_close_ int block_fd = -1; |
356 | uint64_t size = UINT64_MAX; | |
357 | ||
358 | /* A block device */ | |
359 | ||
360 | if (!ret) | |
3a6ce860 | 361 | return 0; |
eb38edce | 362 | |
4756c94e LP |
363 | if (!pretty) { |
364 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
365 | if (r < 0) | |
366 | return r; | |
367 | ||
368 | pretty = pretty_buffer; | |
369 | } | |
eb38edce LP |
370 | |
371 | block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
372 | if (block_fd < 0) | |
2570578d | 373 | log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce | 374 | else { |
3a6ce860 LP |
375 | /* Refresh stat data after opening the node */ |
376 | if (fstat(block_fd, &stbuf) < 0) | |
eb38edce | 377 | return -errno; |
3a6ce860 LP |
378 | st = &stbuf; |
379 | ||
380 | if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */ | |
eb38edce LP |
381 | return -ENOTTY; |
382 | ||
383 | if (!read_only) { | |
384 | int state = 0; | |
385 | ||
386 | if (ioctl(block_fd, BLKROGET, &state) < 0) | |
2570578d | 387 | log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
388 | else if (state) |
389 | read_only = true; | |
390 | } | |
391 | ||
392 | if (ioctl(block_fd, BLKGETSIZE64, &size) < 0) | |
2570578d | 393 | log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
394 | |
395 | block_fd = safe_close(block_fd); | |
396 | } | |
397 | ||
398 | r = image_new(IMAGE_BLOCK, | |
399 | pretty, | |
400 | path, | |
401 | filename, | |
3a6ce860 | 402 | !(st->st_mode & 0222) || read_only, |
eb38edce LP |
403 | 0, |
404 | 0, | |
405 | ret); | |
406 | if (r < 0) | |
407 | return r; | |
408 | ||
ed0cb346 | 409 | if (!IN_SET(size, 0, UINT64_MAX)) |
eb38edce LP |
410 | (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size; |
411 | ||
3a6ce860 | 412 | return 0; |
c2ce6a3d | 413 | } |
cd61c3bf | 414 | |
3a6ce860 | 415 | return -EMEDIUMTYPE; |
c2ce6a3d | 416 | } |
cd61c3bf | 417 | |
5ef46e5f | 418 | int image_find(ImageClass class, const char *name, Image **ret) { |
c2ce6a3d LP |
419 | const char *path; |
420 | int r; | |
cd61c3bf | 421 | |
5ef46e5f LP |
422 | assert(class >= 0); |
423 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d | 424 | assert(name); |
cd61c3bf | 425 | |
c2ce6a3d LP |
426 | /* There are no images with invalid names */ |
427 | if (!image_name_is_valid(name)) | |
3a6ce860 | 428 | return -ENOENT; |
cd61c3bf | 429 | |
5ef46e5f | 430 | NULSTR_FOREACH(path, image_search_path[class]) { |
c2ce6a3d | 431 | _cleanup_closedir_ DIR *d = NULL; |
3a6ce860 | 432 | struct stat st; |
cd61c3bf | 433 | |
c2ce6a3d LP |
434 | d = opendir(path); |
435 | if (!d) { | |
436 | if (errno == ENOENT) | |
437 | continue; | |
cd61c3bf | 438 | |
c2ce6a3d LP |
439 | return -errno; |
440 | } | |
cd61c3bf | 441 | |
3a6ce860 LP |
442 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people to |
443 | * symlink block devices into the search path */ | |
444 | if (fstatat(dirfd(d), name, &st, 0) < 0) { | |
aceac2f0 | 445 | _cleanup_free_ char *raw = NULL; |
5fc7f358 | 446 | |
3a6ce860 LP |
447 | if (errno != ENOENT) |
448 | return -errno; | |
449 | ||
b910cc72 | 450 | raw = strjoin(name, ".raw"); |
aceac2f0 | 451 | if (!raw) |
5fc7f358 LP |
452 | return -ENOMEM; |
453 | ||
3a6ce860 LP |
454 | if (fstatat(dirfd(d), raw, &st, 0) < 0) { |
455 | ||
456 | if (errno == ENOENT) | |
457 | continue; | |
458 | ||
459 | return -errno; | |
460 | } | |
461 | ||
462 | if (!S_ISREG(st.st_mode)) | |
5fc7f358 | 463 | continue; |
3a6ce860 LP |
464 | |
465 | r = image_make(name, dirfd(d), path, raw, &st, ret); | |
466 | ||
467 | } else { | |
468 | if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode)) | |
469 | continue; | |
470 | ||
471 | r = image_make(name, dirfd(d), path, name, &st, ret); | |
5fc7f358 | 472 | } |
3a6ce860 LP |
473 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) |
474 | continue; | |
c2ce6a3d LP |
475 | if (r < 0) |
476 | return r; | |
cd61c3bf | 477 | |
cf604fd4 LP |
478 | if (ret) |
479 | (*ret)->discoverable = true; | |
480 | ||
c2ce6a3d LP |
481 | return 1; |
482 | } | |
483 | ||
cf604fd4 LP |
484 | if (class == IMAGE_MACHINE && streq(name, ".host")) { |
485 | r = image_make(".host", AT_FDCWD, NULL, "/", NULL, ret); | |
486 | if (r < 0) | |
487 | return r; | |
488 | ||
489 | if (ret) | |
490 | (*ret)->discoverable = true; | |
491 | ||
492 | return r; | |
493 | } | |
5fc7f358 | 494 | |
3a6ce860 | 495 | return -ENOENT; |
c2ce6a3d LP |
496 | }; |
497 | ||
2ddf182b | 498 | int image_from_path(const char *path, Image **ret) { |
cf604fd4 LP |
499 | |
500 | /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether | |
501 | * 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 | 502 | * overridden by another, different image earlier in the search path */ |
cf604fd4 | 503 | |
2ddf182b LP |
504 | if (path_equal(path, "/")) |
505 | return image_make(".host", AT_FDCWD, NULL, "/", NULL, ret); | |
506 | ||
507 | return image_make(NULL, AT_FDCWD, NULL, path, NULL, ret); | |
508 | } | |
509 | ||
510 | int image_find_harder(ImageClass class, const char *name_or_path, Image **ret) { | |
511 | if (image_name_is_valid(name_or_path)) | |
512 | return image_find(class, name_or_path, ret); | |
513 | ||
514 | return image_from_path(name_or_path, ret); | |
515 | } | |
516 | ||
5ef46e5f | 517 | int image_discover(ImageClass class, Hashmap *h) { |
c2ce6a3d LP |
518 | const char *path; |
519 | int r; | |
520 | ||
5ef46e5f LP |
521 | assert(class >= 0); |
522 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d LP |
523 | assert(h); |
524 | ||
5ef46e5f | 525 | NULSTR_FOREACH(path, image_search_path[class]) { |
c2ce6a3d LP |
526 | _cleanup_closedir_ DIR *d = NULL; |
527 | struct dirent *de; | |
528 | ||
529 | d = opendir(path); | |
530 | if (!d) { | |
531 | if (errno == ENOENT) | |
a67a4c8c | 532 | continue; |
c2ce6a3d LP |
533 | |
534 | return -errno; | |
535 | } | |
536 | ||
537 | FOREACH_DIRENT_ALL(de, d, return -errno) { | |
538 | _cleanup_(image_unrefp) Image *image = NULL; | |
1bf36bf9 | 539 | _cleanup_free_ char *truncated = NULL; |
3a6ce860 LP |
540 | const char *pretty; |
541 | struct stat st; | |
c2ce6a3d | 542 | |
1bf36bf9 | 543 | if (dot_or_dot_dot(de->d_name)) |
c2ce6a3d LP |
544 | continue; |
545 | ||
3a6ce860 LP |
546 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people |
547 | * to symlink block devices into the search path */ | |
548 | if (fstatat(dirfd(d), de->d_name, &st, 0) < 0) { | |
549 | if (errno == ENOENT) | |
550 | continue; | |
551 | ||
552 | return -errno; | |
553 | } | |
554 | ||
555 | if (S_ISREG(st.st_mode)) { | |
556 | const char *e; | |
557 | ||
558 | e = endswith(de->d_name, ".raw"); | |
559 | if (!e) | |
560 | continue; | |
561 | ||
1bf36bf9 LP |
562 | truncated = strndup(de->d_name, e - de->d_name); |
563 | if (!truncated) | |
564 | return -ENOMEM; | |
565 | ||
566 | pretty = truncated; | |
3a6ce860 | 567 | } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode)) |
1bf36bf9 | 568 | pretty = de->d_name; |
3a6ce860 LP |
569 | else |
570 | continue; | |
1bf36bf9 LP |
571 | |
572 | if (!image_name_is_valid(pretty)) | |
573 | continue; | |
574 | ||
575 | if (hashmap_contains(h, pretty)) | |
c2ce6a3d LP |
576 | continue; |
577 | ||
3a6ce860 LP |
578 | r = image_make(pretty, dirfd(d), path, de->d_name, &st, &image); |
579 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) | |
c2ce6a3d LP |
580 | continue; |
581 | if (r < 0) | |
582 | return r; | |
583 | ||
cf604fd4 LP |
584 | image->discoverable = true; |
585 | ||
c2ce6a3d LP |
586 | r = hashmap_put(h, image->name, image); |
587 | if (r < 0) | |
588 | return r; | |
589 | ||
590 | image = NULL; | |
cd61c3bf LP |
591 | } |
592 | } | |
593 | ||
5ef46e5f | 594 | if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) { |
5fc7f358 LP |
595 | _cleanup_(image_unrefp) Image *image = NULL; |
596 | ||
3a6ce860 | 597 | r = image_make(".host", AT_FDCWD, NULL, "/", NULL, &image); |
5fc7f358 LP |
598 | if (r < 0) |
599 | return r; | |
600 | ||
cf604fd4 LP |
601 | image->discoverable = true; |
602 | ||
5fc7f358 LP |
603 | r = hashmap_put(h, image->name, image); |
604 | if (r < 0) | |
605 | return r; | |
606 | ||
607 | image = NULL; | |
5fc7f358 LP |
608 | } |
609 | ||
cd61c3bf LP |
610 | return 0; |
611 | } | |
612 | ||
08682124 | 613 | int image_remove(Image *i) { |
8e766630 | 614 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
8e0b6570 | 615 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 616 | _cleanup_free_ char *roothash = NULL; |
8e0b6570 | 617 | char **j; |
30535c16 LP |
618 | int r; |
619 | ||
08682124 LP |
620 | assert(i); |
621 | ||
d94c2b06 | 622 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
08682124 LP |
623 | return -EROFS; |
624 | ||
8e0b6570 LP |
625 | settings = image_settings_path(i); |
626 | if (!settings) | |
627 | return -ENOMEM; | |
628 | ||
bafbac4e LP |
629 | roothash = image_roothash_path(i); |
630 | if (!roothash) | |
631 | return -ENOMEM; | |
632 | ||
30535c16 LP |
633 | /* Make sure we don't interfere with a running nspawn */ |
634 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
635 | if (r < 0) | |
636 | return r; | |
637 | ||
ebd93cb6 LP |
638 | switch (i->type) { |
639 | ||
640 | case IMAGE_SUBVOLUME: | |
9fb0b9c7 LP |
641 | |
642 | /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the | |
643 | * big guns */ | |
644 | if (unlink(i->path) < 0) { | |
645 | r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA); | |
646 | if (r < 0) | |
647 | return r; | |
648 | } | |
649 | ||
8e0b6570 | 650 | break; |
ebd93cb6 LP |
651 | |
652 | case IMAGE_DIRECTORY: | |
01b72568 | 653 | /* Allow deletion of read-only directories */ |
db9a4254 | 654 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
8e0b6570 LP |
655 | r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME); |
656 | if (r < 0) | |
657 | return r; | |
658 | ||
659 | break; | |
01b72568 | 660 | |
eb38edce LP |
661 | case IMAGE_BLOCK: |
662 | ||
663 | /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node | |
664 | * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink | |
665 | * the thing (it's most likely a symlink after all). */ | |
666 | ||
667 | if (path_startswith(i->path, "/dev")) | |
668 | break; | |
669 | ||
4831981d | 670 | _fallthrough_; |
aceac2f0 | 671 | case IMAGE_RAW: |
41d1ed05 LP |
672 | if (unlink(i->path) < 0) |
673 | return -errno; | |
8e0b6570 | 674 | break; |
ebd93cb6 LP |
675 | |
676 | default: | |
15411c0c | 677 | return -EOPNOTSUPP; |
ebd93cb6 | 678 | } |
8e0b6570 LP |
679 | |
680 | STRV_FOREACH(j, settings) { | |
681 | if (unlink(*j) < 0 && errno != ENOENT) | |
682 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j); | |
683 | } | |
684 | ||
bafbac4e LP |
685 | if (unlink(roothash) < 0 && errno != ENOENT) |
686 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash); | |
687 | ||
8e0b6570 LP |
688 | return 0; |
689 | } | |
690 | ||
bafbac4e | 691 | static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
8e0b6570 LP |
692 | _cleanup_free_ char *rs = NULL; |
693 | const char *fn; | |
694 | ||
bafbac4e | 695 | fn = strjoina(new_name, suffix); |
8e0b6570 LP |
696 | |
697 | rs = file_in_same_dir(path, fn); | |
698 | if (!rs) | |
699 | return -ENOMEM; | |
700 | ||
701 | return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs); | |
ebd93cb6 LP |
702 | } |
703 | ||
704 | int image_rename(Image *i, const char *new_name) { | |
8e766630 | 705 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT; |
bafbac4e | 706 | _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL; |
8e0b6570 | 707 | _cleanup_strv_free_ char **settings = NULL; |
01b72568 | 708 | unsigned file_attr = 0; |
8e0b6570 | 709 | char **j; |
ebd93cb6 LP |
710 | int r; |
711 | ||
712 | assert(i); | |
713 | ||
714 | if (!image_name_is_valid(new_name)) | |
715 | return -EINVAL; | |
716 | ||
d94c2b06 | 717 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
718 | return -EROFS; |
719 | ||
8e0b6570 LP |
720 | settings = image_settings_path(i); |
721 | if (!settings) | |
722 | return -ENOMEM; | |
723 | ||
bafbac4e LP |
724 | roothash = image_roothash_path(i); |
725 | if (!roothash) | |
726 | return -ENOMEM; | |
727 | ||
30535c16 LP |
728 | /* Make sure we don't interfere with a running nspawn */ |
729 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
730 | if (r < 0) | |
731 | return r; | |
732 | ||
733 | /* Make sure nobody takes the new name, between the time we | |
734 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 735 | * time we take possession of it */ |
30535c16 LP |
736 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
737 | if (r < 0) | |
738 | return r; | |
739 | ||
5ef46e5f | 740 | r = image_find(IMAGE_MACHINE, new_name, NULL); |
3a6ce860 | 741 | if (r >= 0) |
ebd93cb6 | 742 | return -EEXIST; |
3a6ce860 LP |
743 | if (r != -ENOENT) |
744 | return r; | |
ebd93cb6 LP |
745 | |
746 | switch (i->type) { | |
747 | ||
ebd93cb6 | 748 | case IMAGE_DIRECTORY: |
01b72568 LP |
749 | /* Turn of the immutable bit while we rename the image, so that we can rename it */ |
750 | (void) read_attr_path(i->path, &file_attr); | |
751 | ||
752 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 753 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
01b72568 | 754 | |
4831981d | 755 | _fallthrough_; |
01b72568 | 756 | case IMAGE_SUBVOLUME: |
ebd93cb6 LP |
757 | new_path = file_in_same_dir(i->path, new_name); |
758 | break; | |
759 | ||
eb38edce LP |
760 | case IMAGE_BLOCK: |
761 | ||
762 | /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */ | |
763 | if (path_startswith(i->path, "/dev")) | |
764 | return -EROFS; | |
765 | ||
766 | new_path = file_in_same_dir(i->path, new_name); | |
767 | break; | |
768 | ||
aceac2f0 | 769 | case IMAGE_RAW: { |
ebd93cb6 LP |
770 | const char *fn; |
771 | ||
63c372cb | 772 | fn = strjoina(new_name, ".raw"); |
ebd93cb6 LP |
773 | new_path = file_in_same_dir(i->path, fn); |
774 | break; | |
775 | } | |
776 | ||
777 | default: | |
15411c0c | 778 | return -EOPNOTSUPP; |
ebd93cb6 LP |
779 | } |
780 | ||
781 | if (!new_path) | |
782 | return -ENOMEM; | |
783 | ||
784 | nn = strdup(new_name); | |
785 | if (!nn) | |
786 | return -ENOMEM; | |
787 | ||
f85ef957 AC |
788 | r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path); |
789 | if (r < 0) | |
790 | return r; | |
ebd93cb6 | 791 | |
01b72568 LP |
792 | /* Restore the immutable bit, if it was set before */ |
793 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 794 | (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL); |
01b72568 | 795 | |
f9ecfd3b DL |
796 | free_and_replace(i->path, new_path); |
797 | free_and_replace(i->name, nn); | |
ebd93cb6 | 798 | |
8e0b6570 | 799 | STRV_FOREACH(j, settings) { |
bafbac4e | 800 | r = rename_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
801 | if (r < 0 && r != -ENOENT) |
802 | log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j); | |
803 | } | |
804 | ||
bafbac4e LP |
805 | r = rename_auxiliary_file(roothash, new_name, ".roothash"); |
806 | if (r < 0 && r != -ENOENT) | |
807 | log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash); | |
808 | ||
ebd93cb6 LP |
809 | return 0; |
810 | } | |
811 | ||
bafbac4e | 812 | static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
8e0b6570 LP |
813 | _cleanup_free_ char *rs = NULL; |
814 | const char *fn; | |
815 | ||
bafbac4e | 816 | fn = strjoina(new_name, suffix); |
8e0b6570 LP |
817 | |
818 | rs = file_in_same_dir(path, fn); | |
819 | if (!rs) | |
820 | return -ENOMEM; | |
821 | ||
8a016c74 | 822 | return copy_file_atomic(path, rs, 0664, 0, 0, COPY_REFLINK); |
8e0b6570 LP |
823 | } |
824 | ||
ebd93cb6 | 825 | int image_clone(Image *i, const char *new_name, bool read_only) { |
8e766630 | 826 | _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT; |
8e0b6570 | 827 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 828 | _cleanup_free_ char *roothash = NULL; |
ebd93cb6 | 829 | const char *new_path; |
8e0b6570 | 830 | char **j; |
ebd93cb6 LP |
831 | int r; |
832 | ||
833 | assert(i); | |
834 | ||
835 | if (!image_name_is_valid(new_name)) | |
836 | return -EINVAL; | |
837 | ||
8e0b6570 LP |
838 | settings = image_settings_path(i); |
839 | if (!settings) | |
840 | return -ENOMEM; | |
841 | ||
bafbac4e LP |
842 | roothash = image_roothash_path(i); |
843 | if (!roothash) | |
844 | return -ENOMEM; | |
845 | ||
30535c16 LP |
846 | /* Make sure nobody takes the new name, between the time we |
847 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 848 | * time we take possession of it */ |
30535c16 LP |
849 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
850 | if (r < 0) | |
851 | return r; | |
852 | ||
5ef46e5f | 853 | r = image_find(IMAGE_MACHINE, new_name, NULL); |
3a6ce860 | 854 | if (r >= 0) |
ebd93cb6 | 855 | return -EEXIST; |
3a6ce860 LP |
856 | if (r != -ENOENT) |
857 | return r; | |
ebd93cb6 LP |
858 | |
859 | switch (i->type) { | |
860 | ||
861 | case IMAGE_SUBVOLUME: | |
862 | case IMAGE_DIRECTORY: | |
9a50e3ca | 863 | /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain |
13e785f7 | 864 | * directory. */ |
9a50e3ca | 865 | |
63c372cb | 866 | new_path = strjoina("/var/lib/machines/", new_name); |
ebd93cb6 | 867 | |
17cbb288 LP |
868 | r = btrfs_subvol_snapshot(i->path, new_path, |
869 | (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | | |
870 | BTRFS_SNAPSHOT_FALLBACK_COPY | | |
871 | BTRFS_SNAPSHOT_FALLBACK_DIRECTORY | | |
872 | BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE | | |
873 | BTRFS_SNAPSHOT_RECURSIVE | | |
874 | BTRFS_SNAPSHOT_QUOTA); | |
875 | if (r >= 0) | |
9a50e3ca | 876 | /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */ |
8120ee28 | 877 | (void) btrfs_subvol_auto_qgroup(new_path, 0, true); |
5bcd08db | 878 | |
ebd93cb6 LP |
879 | break; |
880 | ||
aceac2f0 | 881 | case IMAGE_RAW: |
63c372cb | 882 | new_path = strjoina("/var/lib/machines/", new_name, ".raw"); |
ebd93cb6 | 883 | |
8a016c74 | 884 | r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME); |
ebd93cb6 LP |
885 | break; |
886 | ||
eb38edce | 887 | case IMAGE_BLOCK: |
ebd93cb6 | 888 | default: |
15411c0c | 889 | return -EOPNOTSUPP; |
ebd93cb6 LP |
890 | } |
891 | ||
892 | if (r < 0) | |
893 | return r; | |
894 | ||
8e0b6570 | 895 | STRV_FOREACH(j, settings) { |
bafbac4e | 896 | r = clone_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
897 | if (r < 0 && r != -ENOENT) |
898 | log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j); | |
899 | } | |
900 | ||
bafbac4e LP |
901 | r = clone_auxiliary_file(roothash, new_name, ".roothash"); |
902 | if (r < 0 && r != -ENOENT) | |
903 | log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash); | |
904 | ||
ebd93cb6 LP |
905 | return 0; |
906 | } | |
907 | ||
908 | int image_read_only(Image *i, bool b) { | |
8e766630 | 909 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
ebd93cb6 | 910 | int r; |
c7664c07 | 911 | |
ebd93cb6 LP |
912 | assert(i); |
913 | ||
d94c2b06 | 914 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
915 | return -EROFS; |
916 | ||
30535c16 LP |
917 | /* Make sure we don't interfere with a running nspawn */ |
918 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
919 | if (r < 0) | |
920 | return r; | |
921 | ||
ebd93cb6 LP |
922 | switch (i->type) { |
923 | ||
924 | case IMAGE_SUBVOLUME: | |
5bcd08db LP |
925 | |
926 | /* Note that we set the flag only on the top-level | |
927 | * subvolume of the image. */ | |
928 | ||
ebd93cb6 LP |
929 | r = btrfs_subvol_set_read_only(i->path, b); |
930 | if (r < 0) | |
931 | return r; | |
01b72568 LP |
932 | |
933 | break; | |
934 | ||
935 | case IMAGE_DIRECTORY: | |
936 | /* For simple directory trees we cannot use the access | |
937 | mode of the top-level directory, since it has an | |
938 | effect on the container itself. However, we can | |
939 | use the "immutable" flag, to at least make the | |
940 | top-level directory read-only. It's not as good as | |
941 | a read-only subvolume, but at least something, and | |
13e785f7 | 942 | we can read the value back. */ |
01b72568 | 943 | |
db9a4254 | 944 | r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL); |
01b72568 LP |
945 | if (r < 0) |
946 | return r; | |
947 | ||
ebd93cb6 LP |
948 | break; |
949 | ||
aceac2f0 | 950 | case IMAGE_RAW: { |
ebd93cb6 LP |
951 | struct stat st; |
952 | ||
953 | if (stat(i->path, &st) < 0) | |
954 | return -errno; | |
955 | ||
956 | if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0) | |
957 | return -errno; | |
f2068bcc LP |
958 | |
959 | /* If the images is now read-only, it's a good time to | |
960 | * defrag it, given that no write patterns will | |
961 | * fragment it again. */ | |
962 | if (b) | |
963 | (void) btrfs_defrag(i->path); | |
ebd93cb6 LP |
964 | break; |
965 | } | |
966 | ||
eb38edce LP |
967 | case IMAGE_BLOCK: { |
968 | _cleanup_close_ int fd = -1; | |
969 | struct stat st; | |
970 | int state = b; | |
971 | ||
972 | fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY); | |
973 | if (fd < 0) | |
974 | return -errno; | |
975 | ||
976 | if (fstat(fd, &st) < 0) | |
977 | return -errno; | |
978 | if (!S_ISBLK(st.st_mode)) | |
979 | return -ENOTTY; | |
980 | ||
981 | if (ioctl(fd, BLKROSET, &state) < 0) | |
982 | return -errno; | |
983 | ||
984 | break; | |
985 | } | |
986 | ||
ebd93cb6 | 987 | default: |
15411c0c | 988 | return -EOPNOTSUPP; |
ebd93cb6 LP |
989 | } |
990 | ||
991 | return 0; | |
08682124 LP |
992 | } |
993 | ||
30535c16 LP |
994 | int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) { |
995 | _cleanup_free_ char *p = NULL; | |
996 | LockFile t = LOCK_FILE_INIT; | |
997 | struct stat st; | |
f25bed67 | 998 | bool exclusive; |
30535c16 LP |
999 | int r; |
1000 | ||
1001 | assert(path); | |
1002 | assert(global); | |
1003 | assert(local); | |
1004 | ||
f25bed67 LP |
1005 | /* Locks an image path. This actually creates two locks: one "local" one, next to the image path |
1006 | * itself, which might be shared via NFS. And another "global" one, in /run, that uses the | |
1007 | * device/inode number. This has the benefit that we can even lock a tree that is a mount point, | |
1008 | * correctly. */ | |
30535c16 | 1009 | |
30535c16 LP |
1010 | if (!path_is_absolute(path)) |
1011 | return -EINVAL; | |
1012 | ||
f25bed67 LP |
1013 | switch (operation & (LOCK_SH|LOCK_EX)) { |
1014 | case LOCK_SH: | |
1015 | exclusive = false; | |
1016 | break; | |
1017 | case LOCK_EX: | |
1018 | exclusive = true; | |
1019 | break; | |
1020 | default: | |
1021 | return -EINVAL; | |
1022 | } | |
1023 | ||
b6e953f2 LP |
1024 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1025 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1026 | return 0; | |
1027 | } | |
1028 | ||
f25bed67 LP |
1029 | /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are |
1030 | * running off it after all, and we don't want any images to manipulate the host image. We make an | |
1031 | * exception for shared locks however: we allow those (and make them NOPs since there's no point in | |
1032 | * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well, | |
1033 | * since it means changes made to the host might propagate to the container as they happen (and a | |
1034 | * shared lock kinda suggests that no changes happen at all while it is in place), but it's too | |
1035 | * useful not to allow read-only containers off the host root, hence let's support this, and trust | |
1036 | * the user to do the right thing with this. */ | |
1037 | if (path_equal(path, "/")) { | |
1038 | if (exclusive) | |
1039 | return -EBUSY; | |
1040 | ||
1041 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1042 | return 0; | |
1043 | } | |
b6e953f2 | 1044 | |
30535c16 | 1045 | if (stat(path, &st) >= 0) { |
eb38edce LP |
1046 | if (S_ISBLK(st.st_mode)) |
1047 | r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev)); | |
1048 | else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode)) | |
1049 | r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino); | |
1050 | else | |
1051 | return -ENOTTY; | |
1052 | ||
1053 | if (r < 0) | |
30535c16 LP |
1054 | return -ENOMEM; |
1055 | } | |
1056 | ||
f25bed67 LP |
1057 | /* For block devices we don't need the "local" lock, as the major/minor lock above should be |
1058 | * sufficient, since block devices are host local anyway. */ | |
1059 | if (!path_startswith(path, "/dev/")) { | |
eb38edce | 1060 | r = make_lock_file_for(path, operation, &t); |
8be17c9b | 1061 | if (r < 0) { |
f25bed67 | 1062 | if (!exclusive && r == -EROFS) |
771b7ead | 1063 | log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path); |
8be17c9b LT |
1064 | else |
1065 | return r; | |
1066 | } | |
eb38edce | 1067 | } |
30535c16 LP |
1068 | |
1069 | if (p) { | |
6e5dcce4 | 1070 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); |
30535c16 LP |
1071 | |
1072 | r = make_lock_file(p, operation, global); | |
1073 | if (r < 0) { | |
1074 | release_lock_file(&t); | |
1075 | return r; | |
1076 | } | |
546dbec5 LP |
1077 | } else |
1078 | *global = (LockFile) LOCK_FILE_INIT; | |
30535c16 LP |
1079 | |
1080 | *local = t; | |
1081 | return 0; | |
1082 | } | |
1083 | ||
cb81cd80 | 1084 | int image_set_limit(Image *i, uint64_t referenced_max) { |
d6ce17c7 LP |
1085 | assert(i); |
1086 | ||
d94c2b06 | 1087 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
d6ce17c7 LP |
1088 | return -EROFS; |
1089 | ||
1090 | if (i->type != IMAGE_SUBVOLUME) | |
15411c0c | 1091 | return -EOPNOTSUPP; |
d6ce17c7 | 1092 | |
5bcd08db LP |
1093 | /* We set the quota both for the subvolume as well as for the |
1094 | * subtree. The latter is mostly for historical reasons, since | |
1095 | * we didn't use to have a concept of subtree quota, and hence | |
1096 | * only modified the subvolume quota. */ | |
1097 | ||
1098 | (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max); | |
1099 | (void) btrfs_subvol_auto_qgroup(i->path, 0, true); | |
1100 | return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max); | |
d6ce17c7 LP |
1101 | } |
1102 | ||
c7664c07 | 1103 | int image_read_metadata(Image *i) { |
8e766630 | 1104 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
c7664c07 LP |
1105 | int r; |
1106 | ||
1107 | assert(i); | |
1108 | ||
1109 | r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock); | |
1110 | if (r < 0) | |
1111 | return r; | |
1112 | ||
1113 | switch (i->type) { | |
1114 | ||
1115 | case IMAGE_SUBVOLUME: | |
1116 | case IMAGE_DIRECTORY: { | |
1117 | _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL; | |
1118 | sd_id128_t machine_id = SD_ID128_NULL; | |
1119 | _cleanup_free_ char *hostname = NULL; | |
1120 | _cleanup_free_ char *path = NULL; | |
1121 | ||
a5648b80 | 1122 | r = chase_symlinks("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1123 | if (r < 0 && r != -ENOENT) |
1124 | log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name); | |
1125 | else if (r >= 0) { | |
1126 | r = read_etc_hostname(path, &hostname); | |
1127 | if (r < 0) | |
1128 | log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name); | |
1129 | } | |
1130 | ||
1131 | path = mfree(path); | |
1132 | ||
a5648b80 | 1133 | r = chase_symlinks("/etc/machine-id", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1134 | if (r < 0 && r != -ENOENT) |
1135 | log_debug_errno(r, "Failed to chase /etc/machine-id in image %s: %m", i->name); | |
1136 | else if (r >= 0) { | |
1137 | _cleanup_close_ int fd = -1; | |
1138 | ||
1139 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY); | |
1140 | if (fd < 0) | |
1141 | log_debug_errno(errno, "Failed to open %s: %m", path); | |
1142 | else { | |
1143 | r = id128_read_fd(fd, ID128_PLAIN, &machine_id); | |
1144 | if (r < 0) | |
1145 | log_debug_errno(r, "Image %s contains invalid machine ID.", i->name); | |
1146 | } | |
1147 | } | |
1148 | ||
1149 | path = mfree(path); | |
1150 | ||
a5648b80 | 1151 | r = chase_symlinks("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1152 | if (r < 0 && r != -ENOENT) |
1153 | log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name); | |
1154 | else if (r >= 0) { | |
aa8fbc74 | 1155 | r = load_env_file_pairs(NULL, path, &machine_info); |
c7664c07 LP |
1156 | if (r < 0) |
1157 | log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name); | |
1158 | } | |
1159 | ||
d58ad743 LP |
1160 | r = load_os_release_pairs(i->path, &os_release); |
1161 | if (r < 0) | |
1162 | log_debug_errno(r, "Failed to read os-release in image, ignoring: %m"); | |
c7664c07 LP |
1163 | |
1164 | free_and_replace(i->hostname, hostname); | |
1165 | i->machine_id = machine_id; | |
1166 | strv_free_and_replace(i->machine_info, machine_info); | |
1167 | strv_free_and_replace(i->os_release, os_release); | |
1168 | ||
1169 | break; | |
1170 | } | |
1171 | ||
1172 | case IMAGE_RAW: | |
1173 | case IMAGE_BLOCK: { | |
1174 | _cleanup_(loop_device_unrefp) LoopDevice *d = NULL; | |
1175 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
1176 | ||
e08f94ac | 1177 | r = loop_device_make_by_path(i->path, O_RDONLY, LO_FLAGS_PARTSCAN, &d); |
c7664c07 LP |
1178 | if (r < 0) |
1179 | return r; | |
1180 | ||
89e62e0b | 1181 | r = dissect_image(d->fd, NULL, NULL, DISSECT_IMAGE_REQUIRE_ROOT|DISSECT_IMAGE_RELAX_VAR_CHECK, &m); |
c7664c07 LP |
1182 | if (r < 0) |
1183 | return r; | |
1184 | ||
1185 | r = dissected_image_acquire_metadata(m); | |
1186 | if (r < 0) | |
1187 | return r; | |
1188 | ||
1189 | free_and_replace(i->hostname, m->hostname); | |
1190 | i->machine_id = m->machine_id; | |
1191 | strv_free_and_replace(i->machine_info, m->machine_info); | |
1192 | strv_free_and_replace(i->os_release, m->os_release); | |
1193 | ||
1194 | break; | |
1195 | } | |
1196 | ||
1197 | default: | |
1198 | return -EOPNOTSUPP; | |
1199 | } | |
1200 | ||
1201 | i->metadata_valid = true; | |
1202 | ||
1203 | return 0; | |
1204 | } | |
1205 | ||
30535c16 | 1206 | int image_name_lock(const char *name, int operation, LockFile *ret) { |
30535c16 LP |
1207 | assert(name); |
1208 | assert(ret); | |
1209 | ||
1210 | /* Locks an image name, regardless of the precise path used. */ | |
1211 | ||
1212 | if (!image_name_is_valid(name)) | |
1213 | return -EINVAL; | |
1214 | ||
b6e953f2 LP |
1215 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1216 | *ret = (LockFile) LOCK_FILE_INIT; | |
1217 | return 0; | |
1218 | } | |
1219 | ||
30535c16 LP |
1220 | if (streq(name, ".host")) |
1221 | return -EBUSY; | |
1222 | ||
6e5dcce4 ZJS |
1223 | const char *p = strjoina("/run/systemd/nspawn/locks/name-", name); |
1224 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); | |
30535c16 LP |
1225 | return make_lock_file(p, operation, ret); |
1226 | } | |
1227 | ||
1228 | bool image_name_is_valid(const char *s) { | |
1229 | if (!filename_is_valid(s)) | |
1230 | return false; | |
1231 | ||
1232 | if (string_has_cc(s, NULL)) | |
1233 | return false; | |
1234 | ||
1235 | if (!utf8_is_valid(s)) | |
1236 | return false; | |
1237 | ||
1238 | /* Temporary files for atomically creating new files */ | |
1239 | if (startswith(s, ".#")) | |
1240 | return false; | |
1241 | ||
1242 | return true; | |
1243 | } | |
1244 | ||
ace9ab19 LP |
1245 | bool image_in_search_path(ImageClass class, const char *image) { |
1246 | const char *path; | |
1247 | ||
1248 | assert(image); | |
1249 | ||
1250 | NULSTR_FOREACH(path, image_search_path[class]) { | |
1251 | const char *p; | |
1252 | size_t k; | |
1253 | ||
1254 | p = path_startswith(image, path); | |
1255 | if (!p) | |
1256 | continue; | |
1257 | ||
1258 | /* Make sure there's a filename following */ | |
1259 | k = strcspn(p, "/"); | |
1260 | if (k == 0) | |
1261 | continue; | |
1262 | ||
1263 | p += k; | |
1264 | ||
1265 | /* Accept trailing slashes */ | |
1266 | if (p[strspn(p, "/")] == 0) | |
1267 | return true; | |
1268 | ||
1269 | } | |
1270 | ||
1271 | return false; | |
1272 | } | |
1273 | ||
cd61c3bf LP |
1274 | static const char* const image_type_table[_IMAGE_TYPE_MAX] = { |
1275 | [IMAGE_DIRECTORY] = "directory", | |
1276 | [IMAGE_SUBVOLUME] = "subvolume", | |
aceac2f0 | 1277 | [IMAGE_RAW] = "raw", |
eb38edce | 1278 | [IMAGE_BLOCK] = "block", |
cd61c3bf LP |
1279 | }; |
1280 | ||
1281 | DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType); |