]>
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> | |
65ddc2c5 | 7 | #include <linux/magic.h> |
a8fbdf54 TA |
8 | #include <stdio.h> |
9 | #include <stdlib.h> | |
e306723e | 10 | #include <sys/file.h> |
204f52e3 | 11 | #include <sys/ioctl.h> |
a8fbdf54 TA |
12 | #include <sys/stat.h> |
13 | #include <unistd.h> | |
546dbec5 | 14 | |
b5efdb8a | 15 | #include "alloc-util.h" |
cd61c3bf | 16 | #include "btrfs-util.h" |
f461a28d | 17 | #include "chase.h" |
c8b3094d | 18 | #include "chattr-util.h" |
ebd93cb6 | 19 | #include "copy.h" |
a0956174 | 20 | #include "dirent-util.h" |
57f1b61b | 21 | #include "discover-image.h" |
c7664c07 | 22 | #include "dissect-image.h" |
686d13b9 | 23 | #include "env-file.h" |
b6e953f2 | 24 | #include "env-util.h" |
6afa5d86 | 25 | #include "extension-util.h" |
3ffd4af2 | 26 | #include "fd-util.h" |
f4f15635 | 27 | #include "fs-util.h" |
a8fbdf54 | 28 | #include "hashmap.h" |
e2054217 | 29 | #include "hostname-setup.h" |
c7664c07 | 30 | #include "id128-util.h" |
73740c9f | 31 | #include "initrd-util.h" |
64e89f56 | 32 | #include "lock-util.h" |
a8fbdf54 | 33 | #include "log.h" |
c7664c07 | 34 | #include "loop-util.h" |
546dbec5 | 35 | #include "macro.h" |
30535c16 | 36 | #include "mkdir.h" |
d8b4d14d | 37 | #include "nulstr-util.h" |
d58ad743 | 38 | #include "os-util.h" |
8e0b6570 | 39 | #include "path-util.h" |
c6878637 | 40 | #include "rm-rf.h" |
65ddc2c5 | 41 | #include "stat-util.h" |
8b43440b | 42 | #include "string-table.h" |
07630cea | 43 | #include "string-util.h" |
8e0b6570 | 44 | #include "strv.h" |
a8fbdf54 | 45 | #include "time-util.h" |
8e0b6570 | 46 | #include "utf8.h" |
89a5a90c | 47 | #include "xattr-util.h" |
cd61c3bf | 48 | |
5ef46e5f | 49 | static const char* const image_search_path[_IMAGE_CLASS_MAX] = { |
9bca4ae4 LP |
50 | [IMAGE_MACHINE] = "/etc/machines\0" /* only place symlinks here */ |
51 | "/run/machines\0" /* and here too */ | |
52 | "/var/lib/machines\0" /* the main place for images */ | |
53 | "/var/lib/container\0" /* legacy */ | |
54 | "/usr/local/lib/machines\0" | |
55 | "/usr/lib/machines\0", | |
5ef46e5f | 56 | |
9bca4ae4 LP |
57 | [IMAGE_PORTABLE] = "/etc/portables\0" /* only place symlinks here */ |
58 | "/run/portables\0" /* and here too */ | |
59 | "/var/lib/portables\0" /* the main place for images */ | |
60 | "/usr/local/lib/portables\0" | |
61 | "/usr/lib/portables\0", | |
62 | ||
de862276 LB |
63 | /* Note that we don't allow storing extensions under /usr/, unlike with other image types. That's |
64 | * because extension images are supposed to extend /usr/, so you get into recursive races, especially | |
65 | * with directory-based extensions, as the kernel's OverlayFS explicitly checks for this and errors | |
66 | * out with -ELOOP if it finds that a lowerdir= is a child of another lowerdir=. */ | |
b60e0f57 | 67 | [IMAGE_SYSEXT] = "/etc/extensions\0" /* only place symlinks here */ |
68 | "/run/extensions\0" /* and here too */ | |
69 | "/var/lib/extensions\0", /* the main place for images */ | |
70 | ||
71 | [IMAGE_CONFEXT] = "/run/confexts\0" /* only place symlinks here */ | |
72 | "/var/lib/confexts\0" /* the main place for images */ | |
73 | "/usr/local/lib/confexts\0" | |
74 | "/usr/lib/confexts\0", | |
5ef46e5f | 75 | }; |
c2ce6a3d | 76 | |
73740c9f LP |
77 | /* Inside the initrd, use a slightly different set of search path (i.e. include .extra/sysext in extension |
78 | * search dir) */ | |
79 | static const char* const image_search_path_initrd[_IMAGE_CLASS_MAX] = { | |
80 | /* (entries that aren't listed here will get the same search path as for the non initrd-case) */ | |
81 | ||
b151e696 LP |
82 | [IMAGE_SYSEXT] = "/etc/extensions\0" /* only place symlinks here */ |
83 | "/run/extensions\0" /* and here too */ | |
84 | "/var/lib/extensions\0" /* the main place for images */ | |
85 | "/.extra/sysext\0" /* put sysext picked up by systemd-stub last, since not trusted */ | |
73740c9f LP |
86 | }; |
87 | ||
a747994b LP |
88 | static const char* image_class_suffix_table[_IMAGE_CLASS_MAX] = { |
89 | [IMAGE_SYSEXT] = ".sysext", | |
90 | [IMAGE_CONFEXT] = ".confext", | |
91 | }; | |
92 | ||
93 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(image_class_suffix, ImageClass); | |
94 | ||
8301aa0b YW |
95 | static Image *image_free(Image *i) { |
96 | assert(i); | |
9614bb06 | 97 | |
cd61c3bf LP |
98 | free(i->name); |
99 | free(i->path); | |
c7664c07 LP |
100 | |
101 | free(i->hostname); | |
102 | strv_free(i->machine_info); | |
103 | strv_free(i->os_release); | |
bcf94222 | 104 | strv_free(i->extension_release); |
c7664c07 | 105 | |
6b430fdb | 106 | return mfree(i); |
cd61c3bf LP |
107 | } |
108 | ||
8301aa0b | 109 | DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free); |
b07ec5a1 YW |
110 | DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func, |
111 | Image, image_unref); | |
9614bb06 | 112 | |
8e0b6570 LP |
113 | static char **image_settings_path(Image *image) { |
114 | _cleanup_strv_free_ char **l = NULL; | |
162f6477 LP |
115 | _cleanup_free_ char *fn = NULL; |
116 | size_t i = 0; | |
117 | int r; | |
8e0b6570 LP |
118 | |
119 | assert(image); | |
120 | ||
121 | l = new0(char*, 4); | |
122 | if (!l) | |
123 | return NULL; | |
124 | ||
162f6477 LP |
125 | fn = strjoin(image->name, ".nspawn"); |
126 | if (!fn) | |
127 | return NULL; | |
8e0b6570 | 128 | |
b910cc72 LP |
129 | FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") { |
130 | l[i] = path_join(s, fn); | |
8e0b6570 LP |
131 | if (!l[i]) |
132 | return NULL; | |
133 | ||
134 | i++; | |
135 | } | |
136 | ||
162f6477 LP |
137 | r = file_in_same_dir(image->path, fn, l + i); |
138 | if (r == -ENOMEM) | |
8e0b6570 | 139 | return NULL; |
162f6477 LP |
140 | if (r < 0) |
141 | log_debug_errno(r, "Failed to generate .nspawn settings path from image path, ignoring: %m"); | |
142 | ||
143 | strv_uniq(l); | |
8e0b6570 | 144 | |
ae2a15bc | 145 | return TAKE_PTR(l); |
8e0b6570 LP |
146 | } |
147 | ||
162f6477 LP |
148 | static int image_roothash_path(Image *image, char **ret) { |
149 | _cleanup_free_ char *fn = NULL; | |
bafbac4e LP |
150 | |
151 | assert(image); | |
152 | ||
162f6477 LP |
153 | fn = strjoin(image->name, ".roothash"); |
154 | if (!fn) | |
155 | return -ENOMEM; | |
bafbac4e | 156 | |
162f6477 | 157 | return file_in_same_dir(image->path, fn, ret); |
bafbac4e LP |
158 | } |
159 | ||
c2ce6a3d | 160 | static int image_new( |
cd61c3bf | 161 | ImageType t, |
3775e141 | 162 | ImageClass c, |
5fc7f358 | 163 | const char *pretty, |
cd61c3bf | 164 | const char *path, |
5fc7f358 | 165 | const char *filename, |
cd61c3bf | 166 | bool read_only, |
10f9c755 | 167 | usec_t crtime, |
cd61c3bf | 168 | usec_t mtime, |
c2ce6a3d | 169 | Image **ret) { |
cd61c3bf LP |
170 | |
171 | _cleanup_(image_unrefp) Image *i = NULL; | |
cd61c3bf | 172 | |
cd61c3bf LP |
173 | assert(t >= 0); |
174 | assert(t < _IMAGE_TYPE_MAX); | |
5fc7f358 LP |
175 | assert(pretty); |
176 | assert(filename); | |
c2ce6a3d | 177 | assert(ret); |
cd61c3bf | 178 | |
c2108701 | 179 | i = new(Image, 1); |
cd61c3bf LP |
180 | if (!i) |
181 | return -ENOMEM; | |
182 | ||
c2108701 LP |
183 | *i = (Image) { |
184 | .n_ref = 1, | |
185 | .type = t, | |
3775e141 | 186 | .class = c, |
c2108701 LP |
187 | .read_only = read_only, |
188 | .crtime = crtime, | |
189 | .mtime = mtime, | |
190 | .usage = UINT64_MAX, | |
191 | .usage_exclusive = UINT64_MAX, | |
192 | .limit = UINT64_MAX, | |
193 | .limit_exclusive = UINT64_MAX, | |
194 | }; | |
cd61c3bf | 195 | |
5fc7f358 | 196 | i->name = strdup(pretty); |
cd61c3bf LP |
197 | if (!i->name) |
198 | return -ENOMEM; | |
199 | ||
657ee2d8 | 200 | i->path = path_join(path, filename); |
5fc7f358 LP |
201 | if (!i->path) |
202 | return -ENOMEM; | |
203 | ||
4ff361cc | 204 | path_simplify(i->path); |
cd61c3bf | 205 | |
1cc6c93a | 206 | *ret = TAKE_PTR(i); |
c2ce6a3d | 207 | |
cd61c3bf LP |
208 | return 0; |
209 | } | |
210 | ||
a747994b LP |
211 | static int extract_pretty( |
212 | const char *path, | |
213 | const char *class_suffix, | |
214 | const char *format_suffix, | |
215 | char **ret) { | |
216 | ||
4756c94e | 217 | _cleanup_free_ char *name = NULL; |
a747994b | 218 | int r; |
4756c94e LP |
219 | |
220 | assert(path); | |
221 | assert(ret); | |
222 | ||
a747994b LP |
223 | r = path_extract_filename(path, &name); |
224 | if (r < 0) | |
225 | return r; | |
4756c94e | 226 | |
a747994b LP |
227 | if (format_suffix) { |
228 | char *e = endswith(name, format_suffix); | |
229 | if (!e) /* Format suffix is required */ | |
4756c94e LP |
230 | return -EINVAL; |
231 | ||
232 | *e = 0; | |
233 | } | |
234 | ||
a747994b LP |
235 | if (class_suffix) { |
236 | char *e = endswith(name, class_suffix); | |
237 | if (e) /* Class suffix is optional */ | |
238 | *e = 0; | |
239 | } | |
240 | ||
4756c94e LP |
241 | if (!image_name_is_valid(name)) |
242 | return -EINVAL; | |
243 | ||
244 | *ret = TAKE_PTR(name); | |
245 | return 0; | |
246 | } | |
247 | ||
5fc7f358 | 248 | static int image_make( |
3775e141 | 249 | ImageClass c, |
5fc7f358 LP |
250 | const char *pretty, |
251 | int dfd, | |
252 | const char *path, | |
253 | const char *filename, | |
3a6ce860 | 254 | const struct stat *st, |
5fc7f358 LP |
255 | Image **ret) { |
256 | ||
2570578d | 257 | _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL; |
3a6ce860 | 258 | struct stat stbuf; |
5fc7f358 | 259 | bool read_only; |
cd61c3bf LP |
260 | int r; |
261 | ||
3a6ce860 | 262 | assert(dfd >= 0 || dfd == AT_FDCWD); |
bcb846f3 | 263 | assert(path || dfd == AT_FDCWD); |
5fc7f358 | 264 | assert(filename); |
cd61c3bf | 265 | |
eb38edce | 266 | /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block |
3a6ce860 LP |
267 | * devices into /var/lib/machines/, and treat them normally. |
268 | * | |
269 | * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we | |
270 | * recognize. */ | |
cd61c3bf | 271 | |
3a6ce860 LP |
272 | if (!st) { |
273 | if (fstatat(dfd, filename, &stbuf, 0) < 0) | |
274 | return -errno; | |
275 | ||
276 | st = &stbuf; | |
277 | } | |
cd61c3bf | 278 | |
2570578d LP |
279 | if (!path) { |
280 | if (dfd == AT_FDCWD) | |
281 | (void) safe_getcwd(&parent); | |
282 | else | |
283 | (void) fd_get_path(dfd, &parent); | |
284 | } | |
bcb846f3 | 285 | |
5fc7f358 LP |
286 | read_only = |
287 | (path && path_startswith(path, "/usr")) || | |
08ff5529 | 288 | (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS); |
86e339c8 | 289 | |
3a6ce860 | 290 | if (S_ISDIR(st->st_mode)) { |
254d1313 | 291 | _cleanup_close_ int fd = -EBADF; |
01b72568 | 292 | unsigned file_attr = 0; |
02dcf215 | 293 | usec_t crtime = 0; |
cd61c3bf | 294 | |
c2ce6a3d | 295 | if (!ret) |
3a6ce860 | 296 | return 0; |
cd61c3bf | 297 | |
4756c94e | 298 | if (!pretty) { |
a747994b | 299 | r = extract_pretty(filename, image_class_suffix_to_string(c), NULL, &pretty_buffer); |
4756c94e LP |
300 | if (r < 0) |
301 | return r; | |
302 | ||
303 | pretty = pretty_buffer; | |
304 | } | |
5fc7f358 | 305 | |
01b72568 LP |
306 | fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY); |
307 | if (fd < 0) | |
308 | return -errno; | |
309 | ||
674b04ff | 310 | if (btrfs_might_be_subvol(st)) { |
cd61c3bf | 311 | |
65ddc2c5 | 312 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
21222ea5 LP |
313 | if (r < 0) |
314 | return r; | |
79de6eb1 | 315 | if (r > 0) { |
10f9c755 | 316 | BtrfsSubvolInfo info; |
cd61c3bf | 317 | |
c2ce6a3d | 318 | /* It's a btrfs subvolume */ |
cd61c3bf | 319 | |
5bcd08db | 320 | r = btrfs_subvol_get_info_fd(fd, 0, &info); |
10f9c755 LP |
321 | if (r < 0) |
322 | return r; | |
c2ce6a3d LP |
323 | |
324 | r = image_new(IMAGE_SUBVOLUME, | |
3775e141 | 325 | c, |
5fc7f358 | 326 | pretty, |
c2ce6a3d | 327 | path, |
5fc7f358 LP |
328 | filename, |
329 | info.read_only || read_only, | |
10f9c755 | 330 | info.otime, |
c2ce6a3d | 331 | 0, |
c2ce6a3d LP |
332 | ret); |
333 | if (r < 0) | |
334 | return r; | |
335 | ||
5bcd08db LP |
336 | if (btrfs_quota_scan_ongoing(fd) == 0) { |
337 | BtrfsQuotaInfo quota; | |
b6b18498 | 338 | |
5bcd08db LP |
339 | r = btrfs_subvol_get_subtree_quota_fd(fd, 0, "a); |
340 | if (r >= 0) { | |
341 | (*ret)->usage = quota.referenced; | |
342 | (*ret)->usage_exclusive = quota.exclusive; | |
343 | ||
344 | (*ret)->limit = quota.referenced_max; | |
345 | (*ret)->limit_exclusive = quota.exclusive_max; | |
346 | } | |
b6b18498 LP |
347 | } |
348 | ||
3a6ce860 | 349 | return 0; |
cd61c3bf | 350 | } |
c2ce6a3d | 351 | } |
cd61c3bf | 352 | |
02dcf215 | 353 | /* Get directory creation time (not available everywhere, but that's OK */ |
12a7f04a | 354 | (void) fd_getcrtime(fd, &crtime); |
02dcf215 LP |
355 | |
356 | /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not | |
01b72568 LP |
357 | * supported everywhere we ignore failures. */ |
358 | (void) read_attr_fd(fd, &file_attr); | |
cd61c3bf | 359 | |
01b72568 | 360 | /* It's just a normal directory. */ |
c2ce6a3d | 361 | r = image_new(IMAGE_DIRECTORY, |
3775e141 | 362 | c, |
5fc7f358 | 363 | pretty, |
c2ce6a3d | 364 | path, |
5fc7f358 | 365 | filename, |
01b72568 | 366 | read_only || (file_attr & FS_IMMUTABLE_FL), |
02dcf215 LP |
367 | crtime, |
368 | 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 |
369 | ret); |
370 | if (r < 0) | |
371 | return r; | |
cd61c3bf | 372 | |
3a6ce860 | 373 | return 0; |
cd61c3bf | 374 | |
3a6ce860 | 375 | } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) { |
10f9c755 | 376 | usec_t crtime = 0; |
cd61c3bf | 377 | |
aceac2f0 | 378 | /* It's a RAW disk image */ |
cd61c3bf | 379 | |
c2ce6a3d | 380 | if (!ret) |
3a6ce860 | 381 | return 0; |
cd61c3bf | 382 | |
c53e07e2 | 383 | (void) fd_getcrtime_at(dfd, filename, AT_SYMLINK_FOLLOW, &crtime); |
10f9c755 | 384 | |
4756c94e | 385 | if (!pretty) { |
a747994b | 386 | r = extract_pretty(filename, image_class_suffix_to_string(c), ".raw", &pretty_buffer); |
4756c94e LP |
387 | if (r < 0) |
388 | return r; | |
389 | ||
390 | pretty = pretty_buffer; | |
391 | } | |
10f9c755 | 392 | |
aceac2f0 | 393 | r = image_new(IMAGE_RAW, |
3775e141 | 394 | c, |
5fc7f358 | 395 | pretty, |
c2ce6a3d | 396 | path, |
5fc7f358 | 397 | filename, |
3a6ce860 | 398 | !(st->st_mode & 0222) || read_only, |
10f9c755 | 399 | crtime, |
3a6ce860 | 400 | timespec_load(&st->st_mtim), |
c2ce6a3d LP |
401 | ret); |
402 | if (r < 0) | |
403 | return r; | |
cd61c3bf | 404 | |
3a6ce860 LP |
405 | (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512; |
406 | (*ret)->limit = (*ret)->limit_exclusive = st->st_size; | |
b6b18498 | 407 | |
3a6ce860 | 408 | return 0; |
eb38edce | 409 | |
3a6ce860 | 410 | } else if (S_ISBLK(st->st_mode)) { |
254d1313 | 411 | _cleanup_close_ int block_fd = -EBADF; |
eb38edce LP |
412 | uint64_t size = UINT64_MAX; |
413 | ||
414 | /* A block device */ | |
415 | ||
416 | if (!ret) | |
3a6ce860 | 417 | return 0; |
eb38edce | 418 | |
4756c94e | 419 | if (!pretty) { |
a747994b | 420 | r = extract_pretty(filename, NULL, NULL, &pretty_buffer); |
4756c94e LP |
421 | if (r < 0) |
422 | return r; | |
423 | ||
424 | pretty = pretty_buffer; | |
425 | } | |
eb38edce LP |
426 | |
427 | block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
428 | if (block_fd < 0) | |
2570578d | 429 | log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce | 430 | else { |
3a6ce860 LP |
431 | /* Refresh stat data after opening the node */ |
432 | if (fstat(block_fd, &stbuf) < 0) | |
eb38edce | 433 | return -errno; |
3a6ce860 LP |
434 | st = &stbuf; |
435 | ||
436 | if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */ | |
eb38edce LP |
437 | return -ENOTTY; |
438 | ||
439 | if (!read_only) { | |
440 | int state = 0; | |
441 | ||
442 | if (ioctl(block_fd, BLKROGET, &state) < 0) | |
2570578d | 443 | log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
444 | else if (state) |
445 | read_only = true; | |
446 | } | |
447 | ||
448 | if (ioctl(block_fd, BLKGETSIZE64, &size) < 0) | |
2570578d | 449 | log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
450 | |
451 | block_fd = safe_close(block_fd); | |
452 | } | |
453 | ||
454 | r = image_new(IMAGE_BLOCK, | |
3775e141 | 455 | c, |
eb38edce LP |
456 | pretty, |
457 | path, | |
458 | filename, | |
3a6ce860 | 459 | !(st->st_mode & 0222) || read_only, |
eb38edce LP |
460 | 0, |
461 | 0, | |
462 | ret); | |
463 | if (r < 0) | |
464 | return r; | |
465 | ||
ed0cb346 | 466 | if (!IN_SET(size, 0, UINT64_MAX)) |
eb38edce LP |
467 | (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size; |
468 | ||
3a6ce860 | 469 | return 0; |
c2ce6a3d | 470 | } |
cd61c3bf | 471 | |
3a6ce860 | 472 | return -EMEDIUMTYPE; |
c2ce6a3d | 473 | } |
cd61c3bf | 474 | |
73740c9f LP |
475 | static const char *pick_image_search_path(ImageClass class) { |
476 | if (class < 0 || class >= _IMAGE_CLASS_MAX) | |
477 | return NULL; | |
478 | ||
479 | /* Use the initrd search path if there is one, otherwise use the common one */ | |
480 | return in_initrd() && image_search_path_initrd[class] ? image_search_path_initrd[class] : image_search_path[class]; | |
481 | } | |
482 | ||
d577d4a4 LP |
483 | int image_find(ImageClass class, |
484 | const char *name, | |
485 | const char *root, | |
486 | Image **ret) { | |
487 | ||
c2ce6a3d | 488 | int r; |
cd61c3bf | 489 | |
5ef46e5f LP |
490 | assert(class >= 0); |
491 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d | 492 | assert(name); |
cd61c3bf | 493 | |
c2ce6a3d LP |
494 | /* There are no images with invalid names */ |
495 | if (!image_name_is_valid(name)) | |
3a6ce860 | 496 | return -ENOENT; |
cd61c3bf | 497 | |
73740c9f | 498 | NULSTR_FOREACH(path, pick_image_search_path(class)) { |
d577d4a4 | 499 | _cleanup_free_ char *resolved = NULL; |
c2ce6a3d | 500 | _cleanup_closedir_ DIR *d = NULL; |
3a6ce860 | 501 | struct stat st; |
d577d4a4 | 502 | int flags; |
cd61c3bf | 503 | |
f461a28d | 504 | r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d); |
d577d4a4 LP |
505 | if (r == -ENOENT) |
506 | continue; | |
507 | if (r < 0) | |
508 | return r; | |
cd61c3bf | 509 | |
d577d4a4 LP |
510 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people |
511 | * to symlink block devices into the search path. (For now, we disable that when operating | |
512 | * relative to some root directory.) */ | |
513 | flags = root ? AT_SYMLINK_NOFOLLOW : 0; | |
514 | if (fstatat(dirfd(d), name, &st, flags) < 0) { | |
aceac2f0 | 515 | _cleanup_free_ char *raw = NULL; |
5fc7f358 | 516 | |
3a6ce860 LP |
517 | if (errno != ENOENT) |
518 | return -errno; | |
519 | ||
b910cc72 | 520 | raw = strjoin(name, ".raw"); |
aceac2f0 | 521 | if (!raw) |
5fc7f358 LP |
522 | return -ENOMEM; |
523 | ||
d577d4a4 | 524 | if (fstatat(dirfd(d), raw, &st, flags) < 0) { |
3a6ce860 LP |
525 | if (errno == ENOENT) |
526 | continue; | |
527 | ||
528 | return -errno; | |
529 | } | |
530 | ||
531 | if (!S_ISREG(st.st_mode)) | |
5fc7f358 | 532 | continue; |
3a6ce860 | 533 | |
3775e141 | 534 | r = image_make(class, name, dirfd(d), resolved, raw, &st, ret); |
3a6ce860 LP |
535 | |
536 | } else { | |
537 | if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode)) | |
538 | continue; | |
539 | ||
3775e141 | 540 | r = image_make(class, name, dirfd(d), resolved, name, &st, ret); |
5fc7f358 | 541 | } |
3a6ce860 LP |
542 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) |
543 | continue; | |
c2ce6a3d LP |
544 | if (r < 0) |
545 | return r; | |
cd61c3bf | 546 | |
cf604fd4 LP |
547 | if (ret) |
548 | (*ret)->discoverable = true; | |
549 | ||
c2ce6a3d LP |
550 | return 1; |
551 | } | |
552 | ||
cf604fd4 | 553 | if (class == IMAGE_MACHINE && streq(name, ".host")) { |
3775e141 | 554 | r = image_make(class, ".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret); |
cf604fd4 LP |
555 | if (r < 0) |
556 | return r; | |
557 | ||
558 | if (ret) | |
559 | (*ret)->discoverable = true; | |
560 | ||
561 | return r; | |
562 | } | |
5fc7f358 | 563 | |
3a6ce860 | 564 | return -ENOENT; |
c2ce6a3d LP |
565 | }; |
566 | ||
2ddf182b | 567 | int image_from_path(const char *path, Image **ret) { |
cf604fd4 LP |
568 | |
569 | /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether | |
570 | * 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 | 571 | * overridden by another, different image earlier in the search path */ |
cf604fd4 | 572 | |
2ddf182b | 573 | if (path_equal(path, "/")) |
3775e141 | 574 | return image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, "/", NULL, ret); |
2ddf182b | 575 | |
3775e141 | 576 | return image_make(_IMAGE_CLASS_INVALID, NULL, AT_FDCWD, NULL, path, NULL, ret); |
2ddf182b LP |
577 | } |
578 | ||
d577d4a4 | 579 | int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) { |
2ddf182b | 580 | if (image_name_is_valid(name_or_path)) |
d577d4a4 | 581 | return image_find(class, name_or_path, root, ret); |
2ddf182b LP |
582 | |
583 | return image_from_path(name_or_path, ret); | |
584 | } | |
585 | ||
d577d4a4 LP |
586 | int image_discover( |
587 | ImageClass class, | |
588 | const char *root, | |
589 | Hashmap *h) { | |
590 | ||
c2ce6a3d LP |
591 | int r; |
592 | ||
5ef46e5f LP |
593 | assert(class >= 0); |
594 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d LP |
595 | assert(h); |
596 | ||
73740c9f | 597 | NULSTR_FOREACH(path, pick_image_search_path(class)) { |
d577d4a4 | 598 | _cleanup_free_ char *resolved = NULL; |
c2ce6a3d | 599 | _cleanup_closedir_ DIR *d = NULL; |
c2ce6a3d | 600 | |
f461a28d | 601 | r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d); |
d577d4a4 LP |
602 | if (r == -ENOENT) |
603 | continue; | |
604 | if (r < 0) | |
605 | return r; | |
c2ce6a3d LP |
606 | |
607 | FOREACH_DIRENT_ALL(de, d, return -errno) { | |
608 | _cleanup_(image_unrefp) Image *image = NULL; | |
a747994b | 609 | _cleanup_free_ char *pretty = NULL; |
3a6ce860 | 610 | struct stat st; |
d577d4a4 | 611 | int flags; |
c2ce6a3d | 612 | |
1bf36bf9 | 613 | if (dot_or_dot_dot(de->d_name)) |
c2ce6a3d LP |
614 | continue; |
615 | ||
d577d4a4 LP |
616 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to |
617 | * permit people to symlink block devices into the search path. */ | |
618 | flags = root ? AT_SYMLINK_NOFOLLOW : 0; | |
619 | if (fstatat(dirfd(d), de->d_name, &st, flags) < 0) { | |
3a6ce860 LP |
620 | if (errno == ENOENT) |
621 | continue; | |
622 | ||
623 | return -errno; | |
624 | } | |
625 | ||
a747994b LP |
626 | if (S_ISREG(st.st_mode)) |
627 | r = extract_pretty(de->d_name, image_class_suffix_to_string(class), ".raw", &pretty); | |
628 | else if (S_ISDIR(st.st_mode)) | |
629 | r = extract_pretty(de->d_name, image_class_suffix_to_string(class), NULL, &pretty); | |
630 | else if (S_ISBLK(st.st_mode)) | |
631 | r = extract_pretty(de->d_name, NULL, NULL, &pretty); | |
3a6ce860 LP |
632 | else |
633 | continue; | |
a747994b LP |
634 | if (r < 0) |
635 | return r; | |
1bf36bf9 LP |
636 | |
637 | if (!image_name_is_valid(pretty)) | |
638 | continue; | |
639 | ||
640 | if (hashmap_contains(h, pretty)) | |
c2ce6a3d LP |
641 | continue; |
642 | ||
3775e141 | 643 | r = image_make(class, pretty, dirfd(d), resolved, de->d_name, &st, &image); |
3a6ce860 | 644 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) |
c2ce6a3d LP |
645 | continue; |
646 | if (r < 0) | |
647 | return r; | |
648 | ||
cf604fd4 LP |
649 | image->discoverable = true; |
650 | ||
c2ce6a3d LP |
651 | r = hashmap_put(h, image->name, image); |
652 | if (r < 0) | |
653 | return r; | |
654 | ||
655 | image = NULL; | |
cd61c3bf LP |
656 | } |
657 | } | |
658 | ||
5ef46e5f | 659 | if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) { |
5fc7f358 LP |
660 | _cleanup_(image_unrefp) Image *image = NULL; |
661 | ||
3775e141 | 662 | r = image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image); |
5fc7f358 LP |
663 | if (r < 0) |
664 | return r; | |
665 | ||
cf604fd4 LP |
666 | image->discoverable = true; |
667 | ||
5fc7f358 LP |
668 | r = hashmap_put(h, image->name, image); |
669 | if (r < 0) | |
670 | return r; | |
671 | ||
672 | image = NULL; | |
5fc7f358 LP |
673 | } |
674 | ||
cd61c3bf LP |
675 | return 0; |
676 | } | |
677 | ||
08682124 | 678 | int image_remove(Image *i) { |
8e766630 | 679 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
8e0b6570 | 680 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 681 | _cleanup_free_ char *roothash = NULL; |
30535c16 LP |
682 | int r; |
683 | ||
08682124 LP |
684 | assert(i); |
685 | ||
d94c2b06 | 686 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
08682124 LP |
687 | return -EROFS; |
688 | ||
8e0b6570 LP |
689 | settings = image_settings_path(i); |
690 | if (!settings) | |
691 | return -ENOMEM; | |
692 | ||
162f6477 LP |
693 | r = image_roothash_path(i, &roothash); |
694 | if (r < 0) | |
695 | return r; | |
bafbac4e | 696 | |
30535c16 LP |
697 | /* Make sure we don't interfere with a running nspawn */ |
698 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
699 | if (r < 0) | |
700 | return r; | |
701 | ||
ebd93cb6 LP |
702 | switch (i->type) { |
703 | ||
704 | case IMAGE_SUBVOLUME: | |
9fb0b9c7 LP |
705 | |
706 | /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the | |
707 | * big guns */ | |
708 | if (unlink(i->path) < 0) { | |
709 | r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA); | |
710 | if (r < 0) | |
711 | return r; | |
712 | } | |
713 | ||
8e0b6570 | 714 | break; |
ebd93cb6 LP |
715 | |
716 | case IMAGE_DIRECTORY: | |
01b72568 | 717 | /* Allow deletion of read-only directories */ |
db9a4254 | 718 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
8e0b6570 LP |
719 | r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME); |
720 | if (r < 0) | |
721 | return r; | |
722 | ||
723 | break; | |
01b72568 | 724 | |
eb38edce LP |
725 | case IMAGE_BLOCK: |
726 | ||
727 | /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node | |
728 | * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink | |
729 | * the thing (it's most likely a symlink after all). */ | |
730 | ||
731 | if (path_startswith(i->path, "/dev")) | |
732 | break; | |
733 | ||
4831981d | 734 | _fallthrough_; |
aceac2f0 | 735 | case IMAGE_RAW: |
41d1ed05 LP |
736 | if (unlink(i->path) < 0) |
737 | return -errno; | |
8e0b6570 | 738 | break; |
ebd93cb6 LP |
739 | |
740 | default: | |
15411c0c | 741 | return -EOPNOTSUPP; |
ebd93cb6 | 742 | } |
8e0b6570 | 743 | |
de010b0b | 744 | STRV_FOREACH(j, settings) |
8e0b6570 LP |
745 | if (unlink(*j) < 0 && errno != ENOENT) |
746 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j); | |
8e0b6570 | 747 | |
bafbac4e LP |
748 | if (unlink(roothash) < 0 && errno != ENOENT) |
749 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash); | |
750 | ||
8e0b6570 LP |
751 | return 0; |
752 | } | |
753 | ||
bafbac4e | 754 | static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
162f6477 LP |
755 | _cleanup_free_ char *fn = NULL, *rs = NULL; |
756 | int r; | |
8e0b6570 | 757 | |
162f6477 LP |
758 | fn = strjoin(new_name, suffix); |
759 | if (!fn) | |
8e0b6570 LP |
760 | return -ENOMEM; |
761 | ||
162f6477 LP |
762 | r = file_in_same_dir(path, fn, &rs); |
763 | if (r < 0) | |
764 | return r; | |
765 | ||
8e0b6570 | 766 | return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs); |
ebd93cb6 LP |
767 | } |
768 | ||
769 | int image_rename(Image *i, const char *new_name) { | |
8e766630 | 770 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT; |
bafbac4e | 771 | _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL; |
8e0b6570 | 772 | _cleanup_strv_free_ char **settings = NULL; |
01b72568 | 773 | unsigned file_attr = 0; |
ebd93cb6 LP |
774 | int r; |
775 | ||
776 | assert(i); | |
777 | ||
778 | if (!image_name_is_valid(new_name)) | |
779 | return -EINVAL; | |
780 | ||
d94c2b06 | 781 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
782 | return -EROFS; |
783 | ||
8e0b6570 LP |
784 | settings = image_settings_path(i); |
785 | if (!settings) | |
786 | return -ENOMEM; | |
787 | ||
162f6477 LP |
788 | r = image_roothash_path(i, &roothash); |
789 | if (r < 0) | |
790 | return r; | |
bafbac4e | 791 | |
30535c16 LP |
792 | /* Make sure we don't interfere with a running nspawn */ |
793 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
794 | if (r < 0) | |
795 | return r; | |
796 | ||
797 | /* Make sure nobody takes the new name, between the time we | |
798 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 799 | * time we take possession of it */ |
30535c16 LP |
800 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
801 | if (r < 0) | |
802 | return r; | |
803 | ||
d577d4a4 | 804 | r = image_find(IMAGE_MACHINE, new_name, NULL, NULL); |
3a6ce860 | 805 | if (r >= 0) |
ebd93cb6 | 806 | return -EEXIST; |
3a6ce860 LP |
807 | if (r != -ENOENT) |
808 | return r; | |
ebd93cb6 LP |
809 | |
810 | switch (i->type) { | |
811 | ||
ebd93cb6 | 812 | case IMAGE_DIRECTORY: |
01b72568 LP |
813 | /* Turn of the immutable bit while we rename the image, so that we can rename it */ |
814 | (void) read_attr_path(i->path, &file_attr); | |
815 | ||
816 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 817 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
01b72568 | 818 | |
4831981d | 819 | _fallthrough_; |
01b72568 | 820 | case IMAGE_SUBVOLUME: |
162f6477 | 821 | r = file_in_same_dir(i->path, new_name, &new_path); |
ebd93cb6 LP |
822 | break; |
823 | ||
eb38edce LP |
824 | case IMAGE_BLOCK: |
825 | ||
826 | /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */ | |
827 | if (path_startswith(i->path, "/dev")) | |
828 | return -EROFS; | |
829 | ||
162f6477 | 830 | r = file_in_same_dir(i->path, new_name, &new_path); |
eb38edce LP |
831 | break; |
832 | ||
aceac2f0 | 833 | case IMAGE_RAW: { |
ebd93cb6 LP |
834 | const char *fn; |
835 | ||
63c372cb | 836 | fn = strjoina(new_name, ".raw"); |
162f6477 LP |
837 | |
838 | r = file_in_same_dir(i->path, fn, &new_path); | |
ebd93cb6 LP |
839 | break; |
840 | } | |
841 | ||
842 | default: | |
15411c0c | 843 | return -EOPNOTSUPP; |
ebd93cb6 | 844 | } |
162f6477 LP |
845 | if (r < 0) |
846 | return r; | |
ebd93cb6 LP |
847 | |
848 | nn = strdup(new_name); | |
849 | if (!nn) | |
850 | return -ENOMEM; | |
851 | ||
f85ef957 AC |
852 | r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path); |
853 | if (r < 0) | |
854 | return r; | |
ebd93cb6 | 855 | |
01b72568 LP |
856 | /* Restore the immutable bit, if it was set before */ |
857 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 858 | (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL); |
01b72568 | 859 | |
f9ecfd3b DL |
860 | free_and_replace(i->path, new_path); |
861 | free_and_replace(i->name, nn); | |
ebd93cb6 | 862 | |
8e0b6570 | 863 | STRV_FOREACH(j, settings) { |
bafbac4e | 864 | r = rename_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
865 | if (r < 0 && r != -ENOENT) |
866 | log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j); | |
867 | } | |
868 | ||
bafbac4e LP |
869 | r = rename_auxiliary_file(roothash, new_name, ".roothash"); |
870 | if (r < 0 && r != -ENOENT) | |
871 | log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash); | |
872 | ||
ebd93cb6 LP |
873 | return 0; |
874 | } | |
875 | ||
bafbac4e | 876 | static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
162f6477 LP |
877 | _cleanup_free_ char *fn = NULL, *rs = NULL; |
878 | int r; | |
8e0b6570 | 879 | |
162f6477 LP |
880 | fn = strjoin(new_name, suffix); |
881 | if (!fn) | |
8e0b6570 LP |
882 | return -ENOMEM; |
883 | ||
162f6477 LP |
884 | r = file_in_same_dir(path, fn, &rs); |
885 | if (r < 0) | |
886 | return r; | |
887 | ||
7c2f5495 | 888 | return copy_file_atomic(path, rs, 0664, COPY_REFLINK); |
8e0b6570 LP |
889 | } |
890 | ||
ebd93cb6 | 891 | int image_clone(Image *i, const char *new_name, bool read_only) { |
8e766630 | 892 | _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT; |
8e0b6570 | 893 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 894 | _cleanup_free_ char *roothash = NULL; |
ebd93cb6 LP |
895 | const char *new_path; |
896 | int r; | |
897 | ||
898 | assert(i); | |
899 | ||
900 | if (!image_name_is_valid(new_name)) | |
901 | return -EINVAL; | |
902 | ||
8e0b6570 LP |
903 | settings = image_settings_path(i); |
904 | if (!settings) | |
905 | return -ENOMEM; | |
906 | ||
162f6477 LP |
907 | r = image_roothash_path(i, &roothash); |
908 | if (r < 0) | |
909 | return r; | |
bafbac4e | 910 | |
30535c16 LP |
911 | /* Make sure nobody takes the new name, between the time we |
912 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 913 | * time we take possession of it */ |
30535c16 LP |
914 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
915 | if (r < 0) | |
916 | return r; | |
917 | ||
d577d4a4 | 918 | r = image_find(IMAGE_MACHINE, new_name, NULL, NULL); |
3a6ce860 | 919 | if (r >= 0) |
ebd93cb6 | 920 | return -EEXIST; |
3a6ce860 LP |
921 | if (r != -ENOENT) |
922 | return r; | |
ebd93cb6 LP |
923 | |
924 | switch (i->type) { | |
925 | ||
926 | case IMAGE_SUBVOLUME: | |
927 | case IMAGE_DIRECTORY: | |
9a50e3ca | 928 | /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain |
13e785f7 | 929 | * directory. */ |
9a50e3ca | 930 | |
63c372cb | 931 | new_path = strjoina("/var/lib/machines/", new_name); |
ebd93cb6 | 932 | |
fab4ef72 DDM |
933 | r = btrfs_subvol_snapshot_at(AT_FDCWD, i->path, AT_FDCWD, new_path, |
934 | (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | | |
935 | BTRFS_SNAPSHOT_FALLBACK_COPY | | |
936 | BTRFS_SNAPSHOT_FALLBACK_DIRECTORY | | |
937 | BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE | | |
938 | BTRFS_SNAPSHOT_RECURSIVE | | |
939 | BTRFS_SNAPSHOT_QUOTA); | |
17cbb288 | 940 | if (r >= 0) |
9a50e3ca | 941 | /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */ |
8120ee28 | 942 | (void) btrfs_subvol_auto_qgroup(new_path, 0, true); |
5bcd08db | 943 | |
ebd93cb6 LP |
944 | break; |
945 | ||
aceac2f0 | 946 | case IMAGE_RAW: |
63c372cb | 947 | new_path = strjoina("/var/lib/machines/", new_name, ".raw"); |
ebd93cb6 | 948 | |
7c2f5495 DDM |
949 | r = copy_file_atomic_full(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, |
950 | COPY_REFLINK|COPY_CRTIME, NULL, NULL); | |
ebd93cb6 LP |
951 | break; |
952 | ||
eb38edce | 953 | case IMAGE_BLOCK: |
ebd93cb6 | 954 | default: |
15411c0c | 955 | return -EOPNOTSUPP; |
ebd93cb6 LP |
956 | } |
957 | ||
958 | if (r < 0) | |
959 | return r; | |
960 | ||
8e0b6570 | 961 | STRV_FOREACH(j, settings) { |
bafbac4e | 962 | r = clone_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
963 | if (r < 0 && r != -ENOENT) |
964 | log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j); | |
965 | } | |
966 | ||
bafbac4e LP |
967 | r = clone_auxiliary_file(roothash, new_name, ".roothash"); |
968 | if (r < 0 && r != -ENOENT) | |
969 | log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash); | |
970 | ||
ebd93cb6 LP |
971 | return 0; |
972 | } | |
973 | ||
974 | int image_read_only(Image *i, bool b) { | |
8e766630 | 975 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
ebd93cb6 | 976 | int r; |
c7664c07 | 977 | |
ebd93cb6 LP |
978 | assert(i); |
979 | ||
d94c2b06 | 980 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
981 | return -EROFS; |
982 | ||
30535c16 LP |
983 | /* Make sure we don't interfere with a running nspawn */ |
984 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
985 | if (r < 0) | |
986 | return r; | |
987 | ||
ebd93cb6 LP |
988 | switch (i->type) { |
989 | ||
990 | case IMAGE_SUBVOLUME: | |
5bcd08db LP |
991 | |
992 | /* Note that we set the flag only on the top-level | |
993 | * subvolume of the image. */ | |
994 | ||
ebd93cb6 LP |
995 | r = btrfs_subvol_set_read_only(i->path, b); |
996 | if (r < 0) | |
997 | return r; | |
01b72568 LP |
998 | |
999 | break; | |
1000 | ||
1001 | case IMAGE_DIRECTORY: | |
1002 | /* For simple directory trees we cannot use the access | |
1003 | mode of the top-level directory, since it has an | |
1004 | effect on the container itself. However, we can | |
1005 | use the "immutable" flag, to at least make the | |
1006 | top-level directory read-only. It's not as good as | |
1007 | a read-only subvolume, but at least something, and | |
13e785f7 | 1008 | we can read the value back. */ |
01b72568 | 1009 | |
db9a4254 | 1010 | r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL); |
01b72568 LP |
1011 | if (r < 0) |
1012 | return r; | |
1013 | ||
ebd93cb6 LP |
1014 | break; |
1015 | ||
aceac2f0 | 1016 | case IMAGE_RAW: { |
ebd93cb6 LP |
1017 | struct stat st; |
1018 | ||
1019 | if (stat(i->path, &st) < 0) | |
1020 | return -errno; | |
1021 | ||
1022 | if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0) | |
1023 | return -errno; | |
f2068bcc LP |
1024 | |
1025 | /* If the images is now read-only, it's a good time to | |
1026 | * defrag it, given that no write patterns will | |
1027 | * fragment it again. */ | |
1028 | if (b) | |
1029 | (void) btrfs_defrag(i->path); | |
ebd93cb6 LP |
1030 | break; |
1031 | } | |
1032 | ||
eb38edce | 1033 | case IMAGE_BLOCK: { |
254d1313 | 1034 | _cleanup_close_ int fd = -EBADF; |
eb38edce LP |
1035 | struct stat st; |
1036 | int state = b; | |
1037 | ||
1038 | fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY); | |
1039 | if (fd < 0) | |
1040 | return -errno; | |
1041 | ||
1042 | if (fstat(fd, &st) < 0) | |
1043 | return -errno; | |
1044 | if (!S_ISBLK(st.st_mode)) | |
1045 | return -ENOTTY; | |
1046 | ||
1047 | if (ioctl(fd, BLKROSET, &state) < 0) | |
1048 | return -errno; | |
1049 | ||
1050 | break; | |
1051 | } | |
1052 | ||
ebd93cb6 | 1053 | default: |
15411c0c | 1054 | return -EOPNOTSUPP; |
ebd93cb6 LP |
1055 | } |
1056 | ||
1057 | return 0; | |
08682124 LP |
1058 | } |
1059 | ||
30535c16 LP |
1060 | int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) { |
1061 | _cleanup_free_ char *p = NULL; | |
1062 | LockFile t = LOCK_FILE_INIT; | |
1063 | struct stat st; | |
f25bed67 | 1064 | bool exclusive; |
30535c16 LP |
1065 | int r; |
1066 | ||
1067 | assert(path); | |
1068 | assert(global); | |
1069 | assert(local); | |
1070 | ||
f25bed67 LP |
1071 | /* Locks an image path. This actually creates two locks: one "local" one, next to the image path |
1072 | * itself, which might be shared via NFS. And another "global" one, in /run, that uses the | |
1073 | * device/inode number. This has the benefit that we can even lock a tree that is a mount point, | |
1074 | * correctly. */ | |
30535c16 | 1075 | |
30535c16 LP |
1076 | if (!path_is_absolute(path)) |
1077 | return -EINVAL; | |
1078 | ||
f25bed67 LP |
1079 | switch (operation & (LOCK_SH|LOCK_EX)) { |
1080 | case LOCK_SH: | |
1081 | exclusive = false; | |
1082 | break; | |
1083 | case LOCK_EX: | |
1084 | exclusive = true; | |
1085 | break; | |
1086 | default: | |
1087 | return -EINVAL; | |
1088 | } | |
1089 | ||
b6e953f2 LP |
1090 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1091 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1092 | return 0; | |
1093 | } | |
1094 | ||
f25bed67 LP |
1095 | /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are |
1096 | * running off it after all, and we don't want any images to manipulate the host image. We make an | |
1097 | * exception for shared locks however: we allow those (and make them NOPs since there's no point in | |
1098 | * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well, | |
1099 | * since it means changes made to the host might propagate to the container as they happen (and a | |
1100 | * shared lock kinda suggests that no changes happen at all while it is in place), but it's too | |
1101 | * useful not to allow read-only containers off the host root, hence let's support this, and trust | |
1102 | * the user to do the right thing with this. */ | |
1103 | if (path_equal(path, "/")) { | |
1104 | if (exclusive) | |
1105 | return -EBUSY; | |
1106 | ||
1107 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1108 | return 0; | |
1109 | } | |
b6e953f2 | 1110 | |
30535c16 | 1111 | if (stat(path, &st) >= 0) { |
eb38edce LP |
1112 | if (S_ISBLK(st.st_mode)) |
1113 | r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev)); | |
1114 | else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode)) | |
1115 | r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino); | |
1116 | else | |
1117 | return -ENOTTY; | |
eb38edce | 1118 | if (r < 0) |
30535c16 LP |
1119 | return -ENOMEM; |
1120 | } | |
1121 | ||
f25bed67 LP |
1122 | /* For block devices we don't need the "local" lock, as the major/minor lock above should be |
1123 | * sufficient, since block devices are host local anyway. */ | |
1124 | if (!path_startswith(path, "/dev/")) { | |
eb38edce | 1125 | r = make_lock_file_for(path, operation, &t); |
8be17c9b | 1126 | if (r < 0) { |
f25bed67 | 1127 | if (!exclusive && r == -EROFS) |
771b7ead | 1128 | log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path); |
8be17c9b LT |
1129 | else |
1130 | return r; | |
1131 | } | |
eb38edce | 1132 | } |
30535c16 LP |
1133 | |
1134 | if (p) { | |
6e5dcce4 | 1135 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); |
30535c16 LP |
1136 | |
1137 | r = make_lock_file(p, operation, global); | |
1138 | if (r < 0) { | |
1139 | release_lock_file(&t); | |
1140 | return r; | |
1141 | } | |
546dbec5 LP |
1142 | } else |
1143 | *global = (LockFile) LOCK_FILE_INIT; | |
30535c16 LP |
1144 | |
1145 | *local = t; | |
1146 | return 0; | |
1147 | } | |
1148 | ||
cb81cd80 | 1149 | int image_set_limit(Image *i, uint64_t referenced_max) { |
d6ce17c7 LP |
1150 | assert(i); |
1151 | ||
d94c2b06 | 1152 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
d6ce17c7 LP |
1153 | return -EROFS; |
1154 | ||
1155 | if (i->type != IMAGE_SUBVOLUME) | |
15411c0c | 1156 | return -EOPNOTSUPP; |
d6ce17c7 | 1157 | |
5bcd08db LP |
1158 | /* We set the quota both for the subvolume as well as for the |
1159 | * subtree. The latter is mostly for historical reasons, since | |
1160 | * we didn't use to have a concept of subtree quota, and hence | |
1161 | * only modified the subvolume quota. */ | |
1162 | ||
1163 | (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max); | |
1164 | (void) btrfs_subvol_auto_qgroup(i->path, 0, true); | |
1165 | return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max); | |
d6ce17c7 LP |
1166 | } |
1167 | ||
84be0c71 | 1168 | int image_read_metadata(Image *i, const ImagePolicy *image_policy) { |
8e766630 | 1169 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
c7664c07 LP |
1170 | int r; |
1171 | ||
1172 | assert(i); | |
1173 | ||
1174 | r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock); | |
1175 | if (r < 0) | |
1176 | return r; | |
1177 | ||
1178 | switch (i->type) { | |
1179 | ||
1180 | case IMAGE_SUBVOLUME: | |
1181 | case IMAGE_DIRECTORY: { | |
bcf94222 | 1182 | _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL; |
c7664c07 LP |
1183 | sd_id128_t machine_id = SD_ID128_NULL; |
1184 | _cleanup_free_ char *hostname = NULL; | |
1185 | _cleanup_free_ char *path = NULL; | |
1186 | ||
b60e0f57 | 1187 | if (i->class == IMAGE_SYSEXT) { |
6afa5d86 LB |
1188 | r = extension_has_forbidden_content(i->path); |
1189 | if (r < 0) | |
1190 | return r; | |
1191 | if (r > 0) | |
1192 | return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM), | |
1193 | "Conflicting content found in image %s, refusing.", | |
1194 | i->name); | |
1195 | } | |
1196 | ||
f461a28d | 1197 | r = chase("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1198 | if (r < 0 && r != -ENOENT) |
1199 | log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name); | |
1200 | else if (r >= 0) { | |
1201 | r = read_etc_hostname(path, &hostname); | |
1202 | if (r < 0) | |
1203 | log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name); | |
1204 | } | |
1205 | ||
1206 | path = mfree(path); | |
1207 | ||
f7b5f399 YW |
1208 | r = id128_get_machine(i->path, &machine_id); |
1209 | if (r < 0) | |
1210 | log_debug_errno(r, "Failed to read machine ID in image %s, ignoring: %m", i->name); | |
c7664c07 | 1211 | |
f461a28d | 1212 | r = chase("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1213 | if (r < 0 && r != -ENOENT) |
1214 | log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name); | |
1215 | else if (r >= 0) { | |
aa8fbc74 | 1216 | r = load_env_file_pairs(NULL, path, &machine_info); |
c7664c07 LP |
1217 | if (r < 0) |
1218 | log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name); | |
1219 | } | |
1220 | ||
d58ad743 LP |
1221 | r = load_os_release_pairs(i->path, &os_release); |
1222 | if (r < 0) | |
1223 | log_debug_errno(r, "Failed to read os-release in image, ignoring: %m"); | |
c7664c07 | 1224 | |
b60e0f57 | 1225 | r = load_extension_release_pairs(i->path, i->class, i->name, /* relax_extension_release_check= */ false, &extension_release); |
bcf94222 LB |
1226 | if (r < 0) |
1227 | log_debug_errno(r, "Failed to read extension-release in image, ignoring: %m"); | |
1228 | ||
c7664c07 LP |
1229 | free_and_replace(i->hostname, hostname); |
1230 | i->machine_id = machine_id; | |
1231 | strv_free_and_replace(i->machine_info, machine_info); | |
1232 | strv_free_and_replace(i->os_release, os_release); | |
bcf94222 | 1233 | strv_free_and_replace(i->extension_release, extension_release); |
c7664c07 LP |
1234 | |
1235 | break; | |
1236 | } | |
1237 | ||
1238 | case IMAGE_RAW: | |
1239 | case IMAGE_BLOCK: { | |
1240 | _cleanup_(loop_device_unrefp) LoopDevice *d = NULL; | |
1241 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
1242 | ||
22ee78a8 | 1243 | r = loop_device_make_by_path(i->path, O_RDONLY, /* sector_size= */ UINT32_MAX, LO_FLAGS_PARTSCAN, LOCK_SH, &d); |
41bc4849 LP |
1244 | if (r < 0) |
1245 | return r; | |
1246 | ||
bad31660 YW |
1247 | r = dissect_loop_device( |
1248 | d, | |
84be0c71 LP |
1249 | /* verity= */ NULL, |
1250 | /* mount_options= */ NULL, | |
1251 | image_policy, | |
4b5de5dd LP |
1252 | DISSECT_IMAGE_GENERIC_ROOT | |
1253 | DISSECT_IMAGE_REQUIRE_ROOT | | |
1254 | DISSECT_IMAGE_RELAX_VAR_CHECK | | |
f6f4ec79 | 1255 | DISSECT_IMAGE_READ_ONLY | |
73d88b80 LP |
1256 | DISSECT_IMAGE_USR_NO_ROOT | |
1257 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | | |
1258 | DISSECT_IMAGE_PIN_PARTITION_DEVICES, | |
75dc190d | 1259 | &m); |
c7664c07 LP |
1260 | if (r < 0) |
1261 | return r; | |
1262 | ||
22847508 ZJS |
1263 | r = dissected_image_acquire_metadata(m, |
1264 | DISSECT_IMAGE_VALIDATE_OS | | |
1265 | DISSECT_IMAGE_VALIDATE_OS_EXT); | |
c7664c07 LP |
1266 | if (r < 0) |
1267 | return r; | |
1268 | ||
1269 | free_and_replace(i->hostname, m->hostname); | |
1270 | i->machine_id = m->machine_id; | |
1271 | strv_free_and_replace(i->machine_info, m->machine_info); | |
1272 | strv_free_and_replace(i->os_release, m->os_release); | |
bcf94222 | 1273 | strv_free_and_replace(i->extension_release, m->extension_release); |
c7664c07 LP |
1274 | |
1275 | break; | |
1276 | } | |
1277 | ||
1278 | default: | |
1279 | return -EOPNOTSUPP; | |
1280 | } | |
1281 | ||
1282 | i->metadata_valid = true; | |
1283 | ||
1284 | return 0; | |
1285 | } | |
1286 | ||
30535c16 | 1287 | int image_name_lock(const char *name, int operation, LockFile *ret) { |
99d97afc LP |
1288 | const char *p; |
1289 | ||
30535c16 LP |
1290 | assert(name); |
1291 | assert(ret); | |
1292 | ||
1293 | /* Locks an image name, regardless of the precise path used. */ | |
1294 | ||
99d97afc LP |
1295 | if (streq(name, ".host")) |
1296 | return -EBUSY; | |
1297 | ||
30535c16 LP |
1298 | if (!image_name_is_valid(name)) |
1299 | return -EINVAL; | |
1300 | ||
b6e953f2 LP |
1301 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1302 | *ret = (LockFile) LOCK_FILE_INIT; | |
1303 | return 0; | |
1304 | } | |
1305 | ||
6e5dcce4 | 1306 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); |
99d97afc LP |
1307 | |
1308 | p = strjoina("/run/systemd/nspawn/locks/name-", name); | |
30535c16 LP |
1309 | return make_lock_file(p, operation, ret); |
1310 | } | |
1311 | ||
d577d4a4 LP |
1312 | bool image_in_search_path( |
1313 | ImageClass class, | |
1314 | const char *root, | |
1315 | const char *image) { | |
1316 | ||
ace9ab19 LP |
1317 | assert(image); |
1318 | ||
73740c9f | 1319 | NULSTR_FOREACH(path, pick_image_search_path(class)) { |
d577d4a4 | 1320 | const char *p, *q; |
ace9ab19 LP |
1321 | size_t k; |
1322 | ||
d577d4a4 LP |
1323 | if (!empty_or_root(root)) { |
1324 | q = path_startswith(path, root); | |
1325 | if (!q) | |
1326 | continue; | |
1327 | } else | |
1328 | q = path; | |
1329 | ||
1330 | p = path_startswith(q, path); | |
ace9ab19 LP |
1331 | if (!p) |
1332 | continue; | |
1333 | ||
1334 | /* Make sure there's a filename following */ | |
1335 | k = strcspn(p, "/"); | |
1336 | if (k == 0) | |
1337 | continue; | |
1338 | ||
1339 | p += k; | |
1340 | ||
1341 | /* Accept trailing slashes */ | |
1342 | if (p[strspn(p, "/")] == 0) | |
1343 | return true; | |
1344 | ||
1345 | } | |
1346 | ||
1347 | return false; | |
1348 | } | |
1349 | ||
cd61c3bf LP |
1350 | static const char* const image_type_table[_IMAGE_TYPE_MAX] = { |
1351 | [IMAGE_DIRECTORY] = "directory", | |
1352 | [IMAGE_SUBVOLUME] = "subvolume", | |
aceac2f0 | 1353 | [IMAGE_RAW] = "raw", |
eb38edce | 1354 | [IMAGE_BLOCK] = "block", |
cd61c3bf LP |
1355 | }; |
1356 | ||
1357 | DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType); |