]>
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 | ||
8301aa0b YW |
88 | static Image *image_free(Image *i) { |
89 | assert(i); | |
9614bb06 | 90 | |
cd61c3bf LP |
91 | free(i->name); |
92 | free(i->path); | |
c7664c07 LP |
93 | |
94 | free(i->hostname); | |
95 | strv_free(i->machine_info); | |
96 | strv_free(i->os_release); | |
bcf94222 | 97 | strv_free(i->extension_release); |
c7664c07 | 98 | |
6b430fdb | 99 | return mfree(i); |
cd61c3bf LP |
100 | } |
101 | ||
8301aa0b | 102 | DEFINE_TRIVIAL_REF_UNREF_FUNC(Image, image, image_free); |
b07ec5a1 YW |
103 | DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops, char, string_hash_func, string_compare_func, |
104 | Image, image_unref); | |
9614bb06 | 105 | |
8e0b6570 LP |
106 | static char **image_settings_path(Image *image) { |
107 | _cleanup_strv_free_ char **l = NULL; | |
162f6477 LP |
108 | _cleanup_free_ char *fn = NULL; |
109 | size_t i = 0; | |
110 | int r; | |
8e0b6570 LP |
111 | |
112 | assert(image); | |
113 | ||
114 | l = new0(char*, 4); | |
115 | if (!l) | |
116 | return NULL; | |
117 | ||
162f6477 LP |
118 | fn = strjoin(image->name, ".nspawn"); |
119 | if (!fn) | |
120 | return NULL; | |
8e0b6570 | 121 | |
b910cc72 LP |
122 | FOREACH_STRING(s, "/etc/systemd/nspawn", "/run/systemd/nspawn") { |
123 | l[i] = path_join(s, fn); | |
8e0b6570 LP |
124 | if (!l[i]) |
125 | return NULL; | |
126 | ||
127 | i++; | |
128 | } | |
129 | ||
162f6477 LP |
130 | r = file_in_same_dir(image->path, fn, l + i); |
131 | if (r == -ENOMEM) | |
8e0b6570 | 132 | return NULL; |
162f6477 LP |
133 | if (r < 0) |
134 | log_debug_errno(r, "Failed to generate .nspawn settings path from image path, ignoring: %m"); | |
135 | ||
136 | strv_uniq(l); | |
8e0b6570 | 137 | |
ae2a15bc | 138 | return TAKE_PTR(l); |
8e0b6570 LP |
139 | } |
140 | ||
162f6477 LP |
141 | static int image_roothash_path(Image *image, char **ret) { |
142 | _cleanup_free_ char *fn = NULL; | |
bafbac4e LP |
143 | |
144 | assert(image); | |
145 | ||
162f6477 LP |
146 | fn = strjoin(image->name, ".roothash"); |
147 | if (!fn) | |
148 | return -ENOMEM; | |
bafbac4e | 149 | |
162f6477 | 150 | return file_in_same_dir(image->path, fn, ret); |
bafbac4e LP |
151 | } |
152 | ||
c2ce6a3d | 153 | static int image_new( |
cd61c3bf | 154 | ImageType t, |
3775e141 | 155 | ImageClass c, |
5fc7f358 | 156 | const char *pretty, |
cd61c3bf | 157 | const char *path, |
5fc7f358 | 158 | const char *filename, |
cd61c3bf | 159 | bool read_only, |
10f9c755 | 160 | usec_t crtime, |
cd61c3bf | 161 | usec_t mtime, |
c2ce6a3d | 162 | Image **ret) { |
cd61c3bf LP |
163 | |
164 | _cleanup_(image_unrefp) Image *i = NULL; | |
cd61c3bf | 165 | |
cd61c3bf LP |
166 | assert(t >= 0); |
167 | assert(t < _IMAGE_TYPE_MAX); | |
5fc7f358 LP |
168 | assert(pretty); |
169 | assert(filename); | |
c2ce6a3d | 170 | assert(ret); |
cd61c3bf | 171 | |
c2108701 | 172 | i = new(Image, 1); |
cd61c3bf LP |
173 | if (!i) |
174 | return -ENOMEM; | |
175 | ||
c2108701 LP |
176 | *i = (Image) { |
177 | .n_ref = 1, | |
178 | .type = t, | |
3775e141 | 179 | .class = c, |
c2108701 LP |
180 | .read_only = read_only, |
181 | .crtime = crtime, | |
182 | .mtime = mtime, | |
183 | .usage = UINT64_MAX, | |
184 | .usage_exclusive = UINT64_MAX, | |
185 | .limit = UINT64_MAX, | |
186 | .limit_exclusive = UINT64_MAX, | |
187 | }; | |
cd61c3bf | 188 | |
5fc7f358 | 189 | i->name = strdup(pretty); |
cd61c3bf LP |
190 | if (!i->name) |
191 | return -ENOMEM; | |
192 | ||
657ee2d8 | 193 | i->path = path_join(path, filename); |
5fc7f358 LP |
194 | if (!i->path) |
195 | return -ENOMEM; | |
196 | ||
4ff361cc | 197 | path_simplify(i->path); |
cd61c3bf | 198 | |
1cc6c93a | 199 | *ret = TAKE_PTR(i); |
c2ce6a3d | 200 | |
cd61c3bf LP |
201 | return 0; |
202 | } | |
203 | ||
4756c94e LP |
204 | static int extract_pretty(const char *path, const char *suffix, char **ret) { |
205 | _cleanup_free_ char *name = NULL; | |
206 | const char *p; | |
4756c94e LP |
207 | |
208 | assert(path); | |
209 | assert(ret); | |
210 | ||
211 | p = last_path_component(path); | |
4756c94e | 212 | |
e8bec624 | 213 | name = strdupcspn(p, "/"); |
4756c94e LP |
214 | if (!name) |
215 | return -ENOMEM; | |
216 | ||
217 | if (suffix) { | |
218 | char *e; | |
219 | ||
220 | e = endswith(name, suffix); | |
221 | if (!e) | |
222 | return -EINVAL; | |
223 | ||
224 | *e = 0; | |
225 | } | |
226 | ||
227 | if (!image_name_is_valid(name)) | |
228 | return -EINVAL; | |
229 | ||
230 | *ret = TAKE_PTR(name); | |
231 | return 0; | |
232 | } | |
233 | ||
5fc7f358 | 234 | static int image_make( |
3775e141 | 235 | ImageClass c, |
5fc7f358 LP |
236 | const char *pretty, |
237 | int dfd, | |
238 | const char *path, | |
239 | const char *filename, | |
3a6ce860 | 240 | const struct stat *st, |
5fc7f358 LP |
241 | Image **ret) { |
242 | ||
2570578d | 243 | _cleanup_free_ char *pretty_buffer = NULL, *parent = NULL; |
3a6ce860 | 244 | struct stat stbuf; |
5fc7f358 | 245 | bool read_only; |
cd61c3bf LP |
246 | int r; |
247 | ||
3a6ce860 | 248 | assert(dfd >= 0 || dfd == AT_FDCWD); |
bcb846f3 | 249 | assert(path || dfd == AT_FDCWD); |
5fc7f358 | 250 | assert(filename); |
cd61c3bf | 251 | |
eb38edce | 252 | /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block |
3a6ce860 LP |
253 | * devices into /var/lib/machines/, and treat them normally. |
254 | * | |
255 | * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we | |
256 | * recognize. */ | |
cd61c3bf | 257 | |
3a6ce860 LP |
258 | if (!st) { |
259 | if (fstatat(dfd, filename, &stbuf, 0) < 0) | |
260 | return -errno; | |
261 | ||
262 | st = &stbuf; | |
263 | } | |
cd61c3bf | 264 | |
2570578d LP |
265 | if (!path) { |
266 | if (dfd == AT_FDCWD) | |
267 | (void) safe_getcwd(&parent); | |
268 | else | |
269 | (void) fd_get_path(dfd, &parent); | |
270 | } | |
bcb846f3 | 271 | |
5fc7f358 LP |
272 | read_only = |
273 | (path && path_startswith(path, "/usr")) || | |
08ff5529 | 274 | (faccessat(dfd, filename, W_OK, AT_EACCESS) < 0 && errno == EROFS); |
86e339c8 | 275 | |
3a6ce860 | 276 | if (S_ISDIR(st->st_mode)) { |
254d1313 | 277 | _cleanup_close_ int fd = -EBADF; |
01b72568 | 278 | unsigned file_attr = 0; |
02dcf215 | 279 | usec_t crtime = 0; |
cd61c3bf | 280 | |
c2ce6a3d | 281 | if (!ret) |
3a6ce860 | 282 | return 0; |
cd61c3bf | 283 | |
4756c94e LP |
284 | if (!pretty) { |
285 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
286 | if (r < 0) | |
287 | return r; | |
288 | ||
289 | pretty = pretty_buffer; | |
290 | } | |
5fc7f358 | 291 | |
01b72568 LP |
292 | fd = openat(dfd, filename, O_CLOEXEC|O_NOCTTY|O_DIRECTORY); |
293 | if (fd < 0) | |
294 | return -errno; | |
295 | ||
674b04ff | 296 | if (btrfs_might_be_subvol(st)) { |
cd61c3bf | 297 | |
65ddc2c5 | 298 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
21222ea5 LP |
299 | if (r < 0) |
300 | return r; | |
301 | if (r) { | |
10f9c755 | 302 | BtrfsSubvolInfo info; |
cd61c3bf | 303 | |
c2ce6a3d | 304 | /* It's a btrfs subvolume */ |
cd61c3bf | 305 | |
5bcd08db | 306 | r = btrfs_subvol_get_info_fd(fd, 0, &info); |
10f9c755 LP |
307 | if (r < 0) |
308 | return r; | |
c2ce6a3d LP |
309 | |
310 | r = image_new(IMAGE_SUBVOLUME, | |
3775e141 | 311 | c, |
5fc7f358 | 312 | pretty, |
c2ce6a3d | 313 | path, |
5fc7f358 LP |
314 | filename, |
315 | info.read_only || read_only, | |
10f9c755 | 316 | info.otime, |
c2ce6a3d | 317 | 0, |
c2ce6a3d LP |
318 | ret); |
319 | if (r < 0) | |
320 | return r; | |
321 | ||
5bcd08db LP |
322 | if (btrfs_quota_scan_ongoing(fd) == 0) { |
323 | BtrfsQuotaInfo quota; | |
b6b18498 | 324 | |
5bcd08db LP |
325 | r = btrfs_subvol_get_subtree_quota_fd(fd, 0, "a); |
326 | if (r >= 0) { | |
327 | (*ret)->usage = quota.referenced; | |
328 | (*ret)->usage_exclusive = quota.exclusive; | |
329 | ||
330 | (*ret)->limit = quota.referenced_max; | |
331 | (*ret)->limit_exclusive = quota.exclusive_max; | |
332 | } | |
b6b18498 LP |
333 | } |
334 | ||
3a6ce860 | 335 | return 0; |
cd61c3bf | 336 | } |
c2ce6a3d | 337 | } |
cd61c3bf | 338 | |
02dcf215 | 339 | /* Get directory creation time (not available everywhere, but that's OK */ |
12a7f04a | 340 | (void) fd_getcrtime(fd, &crtime); |
02dcf215 LP |
341 | |
342 | /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not | |
01b72568 LP |
343 | * supported everywhere we ignore failures. */ |
344 | (void) read_attr_fd(fd, &file_attr); | |
cd61c3bf | 345 | |
01b72568 | 346 | /* It's just a normal directory. */ |
c2ce6a3d | 347 | r = image_new(IMAGE_DIRECTORY, |
3775e141 | 348 | c, |
5fc7f358 | 349 | pretty, |
c2ce6a3d | 350 | path, |
5fc7f358 | 351 | filename, |
01b72568 | 352 | read_only || (file_attr & FS_IMMUTABLE_FL), |
02dcf215 LP |
353 | crtime, |
354 | 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 |
355 | ret); |
356 | if (r < 0) | |
357 | return r; | |
cd61c3bf | 358 | |
3a6ce860 | 359 | return 0; |
cd61c3bf | 360 | |
3a6ce860 | 361 | } else if (S_ISREG(st->st_mode) && endswith(filename, ".raw")) { |
10f9c755 | 362 | usec_t crtime = 0; |
cd61c3bf | 363 | |
aceac2f0 | 364 | /* It's a RAW disk image */ |
cd61c3bf | 365 | |
c2ce6a3d | 366 | if (!ret) |
3a6ce860 | 367 | return 0; |
cd61c3bf | 368 | |
c53e07e2 | 369 | (void) fd_getcrtime_at(dfd, filename, AT_SYMLINK_FOLLOW, &crtime); |
10f9c755 | 370 | |
4756c94e LP |
371 | if (!pretty) { |
372 | r = extract_pretty(filename, ".raw", &pretty_buffer); | |
373 | if (r < 0) | |
374 | return r; | |
375 | ||
376 | pretty = pretty_buffer; | |
377 | } | |
10f9c755 | 378 | |
aceac2f0 | 379 | r = image_new(IMAGE_RAW, |
3775e141 | 380 | c, |
5fc7f358 | 381 | pretty, |
c2ce6a3d | 382 | path, |
5fc7f358 | 383 | filename, |
3a6ce860 | 384 | !(st->st_mode & 0222) || read_only, |
10f9c755 | 385 | crtime, |
3a6ce860 | 386 | timespec_load(&st->st_mtim), |
c2ce6a3d LP |
387 | ret); |
388 | if (r < 0) | |
389 | return r; | |
cd61c3bf | 390 | |
3a6ce860 LP |
391 | (*ret)->usage = (*ret)->usage_exclusive = st->st_blocks * 512; |
392 | (*ret)->limit = (*ret)->limit_exclusive = st->st_size; | |
b6b18498 | 393 | |
3a6ce860 | 394 | return 0; |
eb38edce | 395 | |
3a6ce860 | 396 | } else if (S_ISBLK(st->st_mode)) { |
254d1313 | 397 | _cleanup_close_ int block_fd = -EBADF; |
eb38edce LP |
398 | uint64_t size = UINT64_MAX; |
399 | ||
400 | /* A block device */ | |
401 | ||
402 | if (!ret) | |
3a6ce860 | 403 | return 0; |
eb38edce | 404 | |
4756c94e LP |
405 | if (!pretty) { |
406 | r = extract_pretty(filename, NULL, &pretty_buffer); | |
407 | if (r < 0) | |
408 | return r; | |
409 | ||
410 | pretty = pretty_buffer; | |
411 | } | |
eb38edce LP |
412 | |
413 | block_fd = openat(dfd, filename, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
414 | if (block_fd < 0) | |
2570578d | 415 | log_debug_errno(errno, "Failed to open block device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce | 416 | else { |
3a6ce860 LP |
417 | /* Refresh stat data after opening the node */ |
418 | if (fstat(block_fd, &stbuf) < 0) | |
eb38edce | 419 | return -errno; |
3a6ce860 LP |
420 | st = &stbuf; |
421 | ||
422 | if (!S_ISBLK(st->st_mode)) /* Verify that what we opened is actually what we think it is */ | |
eb38edce LP |
423 | return -ENOTTY; |
424 | ||
425 | if (!read_only) { | |
426 | int state = 0; | |
427 | ||
428 | if (ioctl(block_fd, BLKROGET, &state) < 0) | |
2570578d | 429 | log_debug_errno(errno, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
430 | else if (state) |
431 | read_only = true; | |
432 | } | |
433 | ||
434 | if (ioctl(block_fd, BLKGETSIZE64, &size) < 0) | |
2570578d | 435 | log_debug_errno(errno, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path ?: strnull(parent), filename); |
eb38edce LP |
436 | |
437 | block_fd = safe_close(block_fd); | |
438 | } | |
439 | ||
440 | r = image_new(IMAGE_BLOCK, | |
3775e141 | 441 | c, |
eb38edce LP |
442 | pretty, |
443 | path, | |
444 | filename, | |
3a6ce860 | 445 | !(st->st_mode & 0222) || read_only, |
eb38edce LP |
446 | 0, |
447 | 0, | |
448 | ret); | |
449 | if (r < 0) | |
450 | return r; | |
451 | ||
ed0cb346 | 452 | if (!IN_SET(size, 0, UINT64_MAX)) |
eb38edce LP |
453 | (*ret)->usage = (*ret)->usage_exclusive = (*ret)->limit = (*ret)->limit_exclusive = size; |
454 | ||
3a6ce860 | 455 | return 0; |
c2ce6a3d | 456 | } |
cd61c3bf | 457 | |
3a6ce860 | 458 | return -EMEDIUMTYPE; |
c2ce6a3d | 459 | } |
cd61c3bf | 460 | |
73740c9f LP |
461 | static const char *pick_image_search_path(ImageClass class) { |
462 | if (class < 0 || class >= _IMAGE_CLASS_MAX) | |
463 | return NULL; | |
464 | ||
465 | /* Use the initrd search path if there is one, otherwise use the common one */ | |
466 | return in_initrd() && image_search_path_initrd[class] ? image_search_path_initrd[class] : image_search_path[class]; | |
467 | } | |
468 | ||
d577d4a4 LP |
469 | int image_find(ImageClass class, |
470 | const char *name, | |
471 | const char *root, | |
472 | Image **ret) { | |
473 | ||
c2ce6a3d | 474 | int r; |
cd61c3bf | 475 | |
5ef46e5f LP |
476 | assert(class >= 0); |
477 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d | 478 | assert(name); |
cd61c3bf | 479 | |
c2ce6a3d LP |
480 | /* There are no images with invalid names */ |
481 | if (!image_name_is_valid(name)) | |
3a6ce860 | 482 | return -ENOENT; |
cd61c3bf | 483 | |
73740c9f | 484 | NULSTR_FOREACH(path, pick_image_search_path(class)) { |
d577d4a4 | 485 | _cleanup_free_ char *resolved = NULL; |
c2ce6a3d | 486 | _cleanup_closedir_ DIR *d = NULL; |
3a6ce860 | 487 | struct stat st; |
d577d4a4 | 488 | int flags; |
cd61c3bf | 489 | |
f461a28d | 490 | r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d); |
d577d4a4 LP |
491 | if (r == -ENOENT) |
492 | continue; | |
493 | if (r < 0) | |
494 | return r; | |
cd61c3bf | 495 | |
d577d4a4 LP |
496 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people |
497 | * to symlink block devices into the search path. (For now, we disable that when operating | |
498 | * relative to some root directory.) */ | |
499 | flags = root ? AT_SYMLINK_NOFOLLOW : 0; | |
500 | if (fstatat(dirfd(d), name, &st, flags) < 0) { | |
aceac2f0 | 501 | _cleanup_free_ char *raw = NULL; |
5fc7f358 | 502 | |
3a6ce860 LP |
503 | if (errno != ENOENT) |
504 | return -errno; | |
505 | ||
b910cc72 | 506 | raw = strjoin(name, ".raw"); |
aceac2f0 | 507 | if (!raw) |
5fc7f358 LP |
508 | return -ENOMEM; |
509 | ||
d577d4a4 | 510 | if (fstatat(dirfd(d), raw, &st, flags) < 0) { |
3a6ce860 LP |
511 | if (errno == ENOENT) |
512 | continue; | |
513 | ||
514 | return -errno; | |
515 | } | |
516 | ||
517 | if (!S_ISREG(st.st_mode)) | |
5fc7f358 | 518 | continue; |
3a6ce860 | 519 | |
3775e141 | 520 | r = image_make(class, name, dirfd(d), resolved, raw, &st, ret); |
3a6ce860 LP |
521 | |
522 | } else { | |
523 | if (!S_ISDIR(st.st_mode) && !S_ISBLK(st.st_mode)) | |
524 | continue; | |
525 | ||
3775e141 | 526 | r = image_make(class, name, dirfd(d), resolved, name, &st, ret); |
5fc7f358 | 527 | } |
3a6ce860 LP |
528 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) |
529 | continue; | |
c2ce6a3d LP |
530 | if (r < 0) |
531 | return r; | |
cd61c3bf | 532 | |
cf604fd4 LP |
533 | if (ret) |
534 | (*ret)->discoverable = true; | |
535 | ||
c2ce6a3d LP |
536 | return 1; |
537 | } | |
538 | ||
cf604fd4 | 539 | if (class == IMAGE_MACHINE && streq(name, ".host")) { |
3775e141 | 540 | r = image_make(class, ".host", AT_FDCWD, NULL, empty_to_root(root), NULL, ret); |
cf604fd4 LP |
541 | if (r < 0) |
542 | return r; | |
543 | ||
544 | if (ret) | |
545 | (*ret)->discoverable = true; | |
546 | ||
547 | return r; | |
548 | } | |
5fc7f358 | 549 | |
3a6ce860 | 550 | return -ENOENT; |
c2ce6a3d LP |
551 | }; |
552 | ||
2ddf182b | 553 | int image_from_path(const char *path, Image **ret) { |
cf604fd4 LP |
554 | |
555 | /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether | |
556 | * 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 | 557 | * overridden by another, different image earlier in the search path */ |
cf604fd4 | 558 | |
2ddf182b | 559 | if (path_equal(path, "/")) |
3775e141 | 560 | return image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, "/", NULL, ret); |
2ddf182b | 561 | |
3775e141 | 562 | return image_make(_IMAGE_CLASS_INVALID, NULL, AT_FDCWD, NULL, path, NULL, ret); |
2ddf182b LP |
563 | } |
564 | ||
d577d4a4 | 565 | int image_find_harder(ImageClass class, const char *name_or_path, const char *root, Image **ret) { |
2ddf182b | 566 | if (image_name_is_valid(name_or_path)) |
d577d4a4 | 567 | return image_find(class, name_or_path, root, ret); |
2ddf182b LP |
568 | |
569 | return image_from_path(name_or_path, ret); | |
570 | } | |
571 | ||
d577d4a4 LP |
572 | int image_discover( |
573 | ImageClass class, | |
574 | const char *root, | |
575 | Hashmap *h) { | |
576 | ||
c2ce6a3d LP |
577 | int r; |
578 | ||
5ef46e5f LP |
579 | assert(class >= 0); |
580 | assert(class < _IMAGE_CLASS_MAX); | |
c2ce6a3d LP |
581 | assert(h); |
582 | ||
73740c9f | 583 | NULSTR_FOREACH(path, pick_image_search_path(class)) { |
d577d4a4 | 584 | _cleanup_free_ char *resolved = NULL; |
c2ce6a3d | 585 | _cleanup_closedir_ DIR *d = NULL; |
c2ce6a3d | 586 | |
f461a28d | 587 | r = chase_and_opendir(path, root, CHASE_PREFIX_ROOT, &resolved, &d); |
d577d4a4 LP |
588 | if (r == -ENOENT) |
589 | continue; | |
590 | if (r < 0) | |
591 | return r; | |
c2ce6a3d LP |
592 | |
593 | FOREACH_DIRENT_ALL(de, d, return -errno) { | |
594 | _cleanup_(image_unrefp) Image *image = NULL; | |
1bf36bf9 | 595 | _cleanup_free_ char *truncated = NULL; |
3a6ce860 LP |
596 | const char *pretty; |
597 | struct stat st; | |
d577d4a4 | 598 | int flags; |
c2ce6a3d | 599 | |
1bf36bf9 | 600 | if (dot_or_dot_dot(de->d_name)) |
c2ce6a3d LP |
601 | continue; |
602 | ||
d577d4a4 LP |
603 | /* As mentioned above, we follow symlinks on this fstatat(), because we want to |
604 | * permit people to symlink block devices into the search path. */ | |
605 | flags = root ? AT_SYMLINK_NOFOLLOW : 0; | |
606 | if (fstatat(dirfd(d), de->d_name, &st, flags) < 0) { | |
3a6ce860 LP |
607 | if (errno == ENOENT) |
608 | continue; | |
609 | ||
610 | return -errno; | |
611 | } | |
612 | ||
613 | if (S_ISREG(st.st_mode)) { | |
614 | const char *e; | |
615 | ||
616 | e = endswith(de->d_name, ".raw"); | |
617 | if (!e) | |
618 | continue; | |
619 | ||
1bf36bf9 LP |
620 | truncated = strndup(de->d_name, e - de->d_name); |
621 | if (!truncated) | |
622 | return -ENOMEM; | |
623 | ||
624 | pretty = truncated; | |
3a6ce860 | 625 | } else if (S_ISDIR(st.st_mode) || S_ISBLK(st.st_mode)) |
1bf36bf9 | 626 | pretty = de->d_name; |
3a6ce860 LP |
627 | else |
628 | continue; | |
1bf36bf9 LP |
629 | |
630 | if (!image_name_is_valid(pretty)) | |
631 | continue; | |
632 | ||
633 | if (hashmap_contains(h, pretty)) | |
c2ce6a3d LP |
634 | continue; |
635 | ||
3775e141 | 636 | r = image_make(class, pretty, dirfd(d), resolved, de->d_name, &st, &image); |
3a6ce860 | 637 | if (IN_SET(r, -ENOENT, -EMEDIUMTYPE)) |
c2ce6a3d LP |
638 | continue; |
639 | if (r < 0) | |
640 | return r; | |
641 | ||
cf604fd4 LP |
642 | image->discoverable = true; |
643 | ||
c2ce6a3d LP |
644 | r = hashmap_put(h, image->name, image); |
645 | if (r < 0) | |
646 | return r; | |
647 | ||
648 | image = NULL; | |
cd61c3bf LP |
649 | } |
650 | } | |
651 | ||
5ef46e5f | 652 | if (class == IMAGE_MACHINE && !hashmap_contains(h, ".host")) { |
5fc7f358 LP |
653 | _cleanup_(image_unrefp) Image *image = NULL; |
654 | ||
3775e141 | 655 | r = image_make(IMAGE_MACHINE, ".host", AT_FDCWD, NULL, empty_to_root("/"), NULL, &image); |
5fc7f358 LP |
656 | if (r < 0) |
657 | return r; | |
658 | ||
cf604fd4 LP |
659 | image->discoverable = true; |
660 | ||
5fc7f358 LP |
661 | r = hashmap_put(h, image->name, image); |
662 | if (r < 0) | |
663 | return r; | |
664 | ||
665 | image = NULL; | |
5fc7f358 LP |
666 | } |
667 | ||
cd61c3bf LP |
668 | return 0; |
669 | } | |
670 | ||
08682124 | 671 | int image_remove(Image *i) { |
8e766630 | 672 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
8e0b6570 | 673 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 674 | _cleanup_free_ char *roothash = NULL; |
30535c16 LP |
675 | int r; |
676 | ||
08682124 LP |
677 | assert(i); |
678 | ||
d94c2b06 | 679 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
08682124 LP |
680 | return -EROFS; |
681 | ||
8e0b6570 LP |
682 | settings = image_settings_path(i); |
683 | if (!settings) | |
684 | return -ENOMEM; | |
685 | ||
162f6477 LP |
686 | r = image_roothash_path(i, &roothash); |
687 | if (r < 0) | |
688 | return r; | |
bafbac4e | 689 | |
30535c16 LP |
690 | /* Make sure we don't interfere with a running nspawn */ |
691 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
692 | if (r < 0) | |
693 | return r; | |
694 | ||
ebd93cb6 LP |
695 | switch (i->type) { |
696 | ||
697 | case IMAGE_SUBVOLUME: | |
9fb0b9c7 LP |
698 | |
699 | /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the | |
700 | * big guns */ | |
701 | if (unlink(i->path) < 0) { | |
702 | r = btrfs_subvol_remove(i->path, BTRFS_REMOVE_RECURSIVE|BTRFS_REMOVE_QUOTA); | |
703 | if (r < 0) | |
704 | return r; | |
705 | } | |
706 | ||
8e0b6570 | 707 | break; |
ebd93cb6 LP |
708 | |
709 | case IMAGE_DIRECTORY: | |
01b72568 | 710 | /* Allow deletion of read-only directories */ |
db9a4254 | 711 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
8e0b6570 LP |
712 | r = rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME); |
713 | if (r < 0) | |
714 | return r; | |
715 | ||
716 | break; | |
01b72568 | 717 | |
eb38edce LP |
718 | case IMAGE_BLOCK: |
719 | ||
720 | /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node | |
721 | * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink | |
722 | * the thing (it's most likely a symlink after all). */ | |
723 | ||
724 | if (path_startswith(i->path, "/dev")) | |
725 | break; | |
726 | ||
4831981d | 727 | _fallthrough_; |
aceac2f0 | 728 | case IMAGE_RAW: |
41d1ed05 LP |
729 | if (unlink(i->path) < 0) |
730 | return -errno; | |
8e0b6570 | 731 | break; |
ebd93cb6 LP |
732 | |
733 | default: | |
15411c0c | 734 | return -EOPNOTSUPP; |
ebd93cb6 | 735 | } |
8e0b6570 | 736 | |
de010b0b | 737 | STRV_FOREACH(j, settings) |
8e0b6570 LP |
738 | if (unlink(*j) < 0 && errno != ENOENT) |
739 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", *j); | |
8e0b6570 | 740 | |
bafbac4e LP |
741 | if (unlink(roothash) < 0 && errno != ENOENT) |
742 | log_debug_errno(errno, "Failed to unlink %s, ignoring: %m", roothash); | |
743 | ||
8e0b6570 LP |
744 | return 0; |
745 | } | |
746 | ||
bafbac4e | 747 | static int rename_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
162f6477 LP |
748 | _cleanup_free_ char *fn = NULL, *rs = NULL; |
749 | int r; | |
8e0b6570 | 750 | |
162f6477 LP |
751 | fn = strjoin(new_name, suffix); |
752 | if (!fn) | |
8e0b6570 LP |
753 | return -ENOMEM; |
754 | ||
162f6477 LP |
755 | r = file_in_same_dir(path, fn, &rs); |
756 | if (r < 0) | |
757 | return r; | |
758 | ||
8e0b6570 | 759 | return rename_noreplace(AT_FDCWD, path, AT_FDCWD, rs); |
ebd93cb6 LP |
760 | } |
761 | ||
762 | int image_rename(Image *i, const char *new_name) { | |
8e766630 | 763 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT; |
bafbac4e | 764 | _cleanup_free_ char *new_path = NULL, *nn = NULL, *roothash = NULL; |
8e0b6570 | 765 | _cleanup_strv_free_ char **settings = NULL; |
01b72568 | 766 | unsigned file_attr = 0; |
ebd93cb6 LP |
767 | int r; |
768 | ||
769 | assert(i); | |
770 | ||
771 | if (!image_name_is_valid(new_name)) | |
772 | return -EINVAL; | |
773 | ||
d94c2b06 | 774 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
775 | return -EROFS; |
776 | ||
8e0b6570 LP |
777 | settings = image_settings_path(i); |
778 | if (!settings) | |
779 | return -ENOMEM; | |
780 | ||
162f6477 LP |
781 | r = image_roothash_path(i, &roothash); |
782 | if (r < 0) | |
783 | return r; | |
bafbac4e | 784 | |
30535c16 LP |
785 | /* Make sure we don't interfere with a running nspawn */ |
786 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
787 | if (r < 0) | |
788 | return r; | |
789 | ||
790 | /* Make sure nobody takes the new name, between the time we | |
791 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 792 | * time we take possession of it */ |
30535c16 LP |
793 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
794 | if (r < 0) | |
795 | return r; | |
796 | ||
d577d4a4 | 797 | r = image_find(IMAGE_MACHINE, new_name, NULL, NULL); |
3a6ce860 | 798 | if (r >= 0) |
ebd93cb6 | 799 | return -EEXIST; |
3a6ce860 LP |
800 | if (r != -ENOENT) |
801 | return r; | |
ebd93cb6 LP |
802 | |
803 | switch (i->type) { | |
804 | ||
ebd93cb6 | 805 | case IMAGE_DIRECTORY: |
01b72568 LP |
806 | /* Turn of the immutable bit while we rename the image, so that we can rename it */ |
807 | (void) read_attr_path(i->path, &file_attr); | |
808 | ||
809 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 810 | (void) chattr_path(i->path, 0, FS_IMMUTABLE_FL, NULL); |
01b72568 | 811 | |
4831981d | 812 | _fallthrough_; |
01b72568 | 813 | case IMAGE_SUBVOLUME: |
162f6477 | 814 | r = file_in_same_dir(i->path, new_name, &new_path); |
ebd93cb6 LP |
815 | break; |
816 | ||
eb38edce LP |
817 | case IMAGE_BLOCK: |
818 | ||
819 | /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */ | |
820 | if (path_startswith(i->path, "/dev")) | |
821 | return -EROFS; | |
822 | ||
162f6477 | 823 | r = file_in_same_dir(i->path, new_name, &new_path); |
eb38edce LP |
824 | break; |
825 | ||
aceac2f0 | 826 | case IMAGE_RAW: { |
ebd93cb6 LP |
827 | const char *fn; |
828 | ||
63c372cb | 829 | fn = strjoina(new_name, ".raw"); |
162f6477 LP |
830 | |
831 | r = file_in_same_dir(i->path, fn, &new_path); | |
ebd93cb6 LP |
832 | break; |
833 | } | |
834 | ||
835 | default: | |
15411c0c | 836 | return -EOPNOTSUPP; |
ebd93cb6 | 837 | } |
162f6477 LP |
838 | if (r < 0) |
839 | return r; | |
ebd93cb6 LP |
840 | |
841 | nn = strdup(new_name); | |
842 | if (!nn) | |
843 | return -ENOMEM; | |
844 | ||
f85ef957 AC |
845 | r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path); |
846 | if (r < 0) | |
847 | return r; | |
ebd93cb6 | 848 | |
01b72568 LP |
849 | /* Restore the immutable bit, if it was set before */ |
850 | if (file_attr & FS_IMMUTABLE_FL) | |
db9a4254 | 851 | (void) chattr_path(new_path, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL); |
01b72568 | 852 | |
f9ecfd3b DL |
853 | free_and_replace(i->path, new_path); |
854 | free_and_replace(i->name, nn); | |
ebd93cb6 | 855 | |
8e0b6570 | 856 | STRV_FOREACH(j, settings) { |
bafbac4e | 857 | r = rename_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
858 | if (r < 0 && r != -ENOENT) |
859 | log_debug_errno(r, "Failed to rename settings file %s, ignoring: %m", *j); | |
860 | } | |
861 | ||
bafbac4e LP |
862 | r = rename_auxiliary_file(roothash, new_name, ".roothash"); |
863 | if (r < 0 && r != -ENOENT) | |
864 | log_debug_errno(r, "Failed to rename roothash file %s, ignoring: %m", roothash); | |
865 | ||
ebd93cb6 LP |
866 | return 0; |
867 | } | |
868 | ||
bafbac4e | 869 | static int clone_auxiliary_file(const char *path, const char *new_name, const char *suffix) { |
162f6477 LP |
870 | _cleanup_free_ char *fn = NULL, *rs = NULL; |
871 | int r; | |
8e0b6570 | 872 | |
162f6477 LP |
873 | fn = strjoin(new_name, suffix); |
874 | if (!fn) | |
8e0b6570 LP |
875 | return -ENOMEM; |
876 | ||
162f6477 LP |
877 | r = file_in_same_dir(path, fn, &rs); |
878 | if (r < 0) | |
879 | return r; | |
880 | ||
7c2f5495 | 881 | return copy_file_atomic(path, rs, 0664, COPY_REFLINK); |
8e0b6570 LP |
882 | } |
883 | ||
ebd93cb6 | 884 | int image_clone(Image *i, const char *new_name, bool read_only) { |
8e766630 | 885 | _cleanup_(release_lock_file) LockFile name_lock = LOCK_FILE_INIT; |
8e0b6570 | 886 | _cleanup_strv_free_ char **settings = NULL; |
bafbac4e | 887 | _cleanup_free_ char *roothash = NULL; |
ebd93cb6 LP |
888 | const char *new_path; |
889 | int r; | |
890 | ||
891 | assert(i); | |
892 | ||
893 | if (!image_name_is_valid(new_name)) | |
894 | return -EINVAL; | |
895 | ||
8e0b6570 LP |
896 | settings = image_settings_path(i); |
897 | if (!settings) | |
898 | return -ENOMEM; | |
899 | ||
162f6477 LP |
900 | r = image_roothash_path(i, &roothash); |
901 | if (r < 0) | |
902 | return r; | |
bafbac4e | 903 | |
30535c16 LP |
904 | /* Make sure nobody takes the new name, between the time we |
905 | * checked it is currently unused in all search paths, and the | |
f8e2f4d6 | 906 | * time we take possession of it */ |
30535c16 LP |
907 | r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock); |
908 | if (r < 0) | |
909 | return r; | |
910 | ||
d577d4a4 | 911 | r = image_find(IMAGE_MACHINE, new_name, NULL, NULL); |
3a6ce860 | 912 | if (r >= 0) |
ebd93cb6 | 913 | return -EEXIST; |
3a6ce860 LP |
914 | if (r != -ENOENT) |
915 | return r; | |
ebd93cb6 LP |
916 | |
917 | switch (i->type) { | |
918 | ||
919 | case IMAGE_SUBVOLUME: | |
920 | case IMAGE_DIRECTORY: | |
9a50e3ca | 921 | /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain |
13e785f7 | 922 | * directory. */ |
9a50e3ca | 923 | |
63c372cb | 924 | new_path = strjoina("/var/lib/machines/", new_name); |
ebd93cb6 | 925 | |
17cbb288 LP |
926 | r = btrfs_subvol_snapshot(i->path, new_path, |
927 | (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | | |
928 | BTRFS_SNAPSHOT_FALLBACK_COPY | | |
929 | BTRFS_SNAPSHOT_FALLBACK_DIRECTORY | | |
930 | BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE | | |
931 | BTRFS_SNAPSHOT_RECURSIVE | | |
932 | BTRFS_SNAPSHOT_QUOTA); | |
933 | if (r >= 0) | |
9a50e3ca | 934 | /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */ |
8120ee28 | 935 | (void) btrfs_subvol_auto_qgroup(new_path, 0, true); |
5bcd08db | 936 | |
ebd93cb6 LP |
937 | break; |
938 | ||
aceac2f0 | 939 | case IMAGE_RAW: |
63c372cb | 940 | new_path = strjoina("/var/lib/machines/", new_name, ".raw"); |
ebd93cb6 | 941 | |
7c2f5495 DDM |
942 | r = copy_file_atomic_full(i->path, new_path, read_only ? 0444 : 0644, FS_NOCOW_FL, FS_NOCOW_FL, |
943 | COPY_REFLINK|COPY_CRTIME, NULL, NULL); | |
ebd93cb6 LP |
944 | break; |
945 | ||
eb38edce | 946 | case IMAGE_BLOCK: |
ebd93cb6 | 947 | default: |
15411c0c | 948 | return -EOPNOTSUPP; |
ebd93cb6 LP |
949 | } |
950 | ||
951 | if (r < 0) | |
952 | return r; | |
953 | ||
8e0b6570 | 954 | STRV_FOREACH(j, settings) { |
bafbac4e | 955 | r = clone_auxiliary_file(*j, new_name, ".nspawn"); |
8e0b6570 LP |
956 | if (r < 0 && r != -ENOENT) |
957 | log_debug_errno(r, "Failed to clone settings %s, ignoring: %m", *j); | |
958 | } | |
959 | ||
bafbac4e LP |
960 | r = clone_auxiliary_file(roothash, new_name, ".roothash"); |
961 | if (r < 0 && r != -ENOENT) | |
962 | log_debug_errno(r, "Failed to clone root hash file %s, ignoring: %m", roothash); | |
963 | ||
ebd93cb6 LP |
964 | return 0; |
965 | } | |
966 | ||
967 | int image_read_only(Image *i, bool b) { | |
8e766630 | 968 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
ebd93cb6 | 969 | int r; |
c7664c07 | 970 | |
ebd93cb6 LP |
971 | assert(i); |
972 | ||
d94c2b06 | 973 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
ebd93cb6 LP |
974 | return -EROFS; |
975 | ||
30535c16 LP |
976 | /* Make sure we don't interfere with a running nspawn */ |
977 | r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock); | |
978 | if (r < 0) | |
979 | return r; | |
980 | ||
ebd93cb6 LP |
981 | switch (i->type) { |
982 | ||
983 | case IMAGE_SUBVOLUME: | |
5bcd08db LP |
984 | |
985 | /* Note that we set the flag only on the top-level | |
986 | * subvolume of the image. */ | |
987 | ||
ebd93cb6 LP |
988 | r = btrfs_subvol_set_read_only(i->path, b); |
989 | if (r < 0) | |
990 | return r; | |
01b72568 LP |
991 | |
992 | break; | |
993 | ||
994 | case IMAGE_DIRECTORY: | |
995 | /* For simple directory trees we cannot use the access | |
996 | mode of the top-level directory, since it has an | |
997 | effect on the container itself. However, we can | |
998 | use the "immutable" flag, to at least make the | |
999 | top-level directory read-only. It's not as good as | |
1000 | a read-only subvolume, but at least something, and | |
13e785f7 | 1001 | we can read the value back. */ |
01b72568 | 1002 | |
db9a4254 | 1003 | r = chattr_path(i->path, b ? FS_IMMUTABLE_FL : 0, FS_IMMUTABLE_FL, NULL); |
01b72568 LP |
1004 | if (r < 0) |
1005 | return r; | |
1006 | ||
ebd93cb6 LP |
1007 | break; |
1008 | ||
aceac2f0 | 1009 | case IMAGE_RAW: { |
ebd93cb6 LP |
1010 | struct stat st; |
1011 | ||
1012 | if (stat(i->path, &st) < 0) | |
1013 | return -errno; | |
1014 | ||
1015 | if (chmod(i->path, (st.st_mode & 0444) | (b ? 0000 : 0200)) < 0) | |
1016 | return -errno; | |
f2068bcc LP |
1017 | |
1018 | /* If the images is now read-only, it's a good time to | |
1019 | * defrag it, given that no write patterns will | |
1020 | * fragment it again. */ | |
1021 | if (b) | |
1022 | (void) btrfs_defrag(i->path); | |
ebd93cb6 LP |
1023 | break; |
1024 | } | |
1025 | ||
eb38edce | 1026 | case IMAGE_BLOCK: { |
254d1313 | 1027 | _cleanup_close_ int fd = -EBADF; |
eb38edce LP |
1028 | struct stat st; |
1029 | int state = b; | |
1030 | ||
1031 | fd = open(i->path, O_CLOEXEC|O_RDONLY|O_NONBLOCK|O_NOCTTY); | |
1032 | if (fd < 0) | |
1033 | return -errno; | |
1034 | ||
1035 | if (fstat(fd, &st) < 0) | |
1036 | return -errno; | |
1037 | if (!S_ISBLK(st.st_mode)) | |
1038 | return -ENOTTY; | |
1039 | ||
1040 | if (ioctl(fd, BLKROSET, &state) < 0) | |
1041 | return -errno; | |
1042 | ||
1043 | break; | |
1044 | } | |
1045 | ||
ebd93cb6 | 1046 | default: |
15411c0c | 1047 | return -EOPNOTSUPP; |
ebd93cb6 LP |
1048 | } |
1049 | ||
1050 | return 0; | |
08682124 LP |
1051 | } |
1052 | ||
30535c16 LP |
1053 | int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) { |
1054 | _cleanup_free_ char *p = NULL; | |
1055 | LockFile t = LOCK_FILE_INIT; | |
1056 | struct stat st; | |
f25bed67 | 1057 | bool exclusive; |
30535c16 LP |
1058 | int r; |
1059 | ||
1060 | assert(path); | |
1061 | assert(global); | |
1062 | assert(local); | |
1063 | ||
f25bed67 LP |
1064 | /* Locks an image path. This actually creates two locks: one "local" one, next to the image path |
1065 | * itself, which might be shared via NFS. And another "global" one, in /run, that uses the | |
1066 | * device/inode number. This has the benefit that we can even lock a tree that is a mount point, | |
1067 | * correctly. */ | |
30535c16 | 1068 | |
30535c16 LP |
1069 | if (!path_is_absolute(path)) |
1070 | return -EINVAL; | |
1071 | ||
f25bed67 LP |
1072 | switch (operation & (LOCK_SH|LOCK_EX)) { |
1073 | case LOCK_SH: | |
1074 | exclusive = false; | |
1075 | break; | |
1076 | case LOCK_EX: | |
1077 | exclusive = true; | |
1078 | break; | |
1079 | default: | |
1080 | return -EINVAL; | |
1081 | } | |
1082 | ||
b6e953f2 LP |
1083 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1084 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1085 | return 0; | |
1086 | } | |
1087 | ||
f25bed67 LP |
1088 | /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are |
1089 | * running off it after all, and we don't want any images to manipulate the host image. We make an | |
1090 | * exception for shared locks however: we allow those (and make them NOPs since there's no point in | |
1091 | * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well, | |
1092 | * since it means changes made to the host might propagate to the container as they happen (and a | |
1093 | * shared lock kinda suggests that no changes happen at all while it is in place), but it's too | |
1094 | * useful not to allow read-only containers off the host root, hence let's support this, and trust | |
1095 | * the user to do the right thing with this. */ | |
1096 | if (path_equal(path, "/")) { | |
1097 | if (exclusive) | |
1098 | return -EBUSY; | |
1099 | ||
1100 | *local = *global = (LockFile) LOCK_FILE_INIT; | |
1101 | return 0; | |
1102 | } | |
b6e953f2 | 1103 | |
30535c16 | 1104 | if (stat(path, &st) >= 0) { |
eb38edce LP |
1105 | if (S_ISBLK(st.st_mode)) |
1106 | r = asprintf(&p, "/run/systemd/nspawn/locks/block-%u:%u", major(st.st_rdev), minor(st.st_rdev)); | |
1107 | else if (S_ISDIR(st.st_mode) || S_ISREG(st.st_mode)) | |
1108 | r = asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino); | |
1109 | else | |
1110 | return -ENOTTY; | |
eb38edce | 1111 | if (r < 0) |
30535c16 LP |
1112 | return -ENOMEM; |
1113 | } | |
1114 | ||
f25bed67 LP |
1115 | /* For block devices we don't need the "local" lock, as the major/minor lock above should be |
1116 | * sufficient, since block devices are host local anyway. */ | |
1117 | if (!path_startswith(path, "/dev/")) { | |
eb38edce | 1118 | r = make_lock_file_for(path, operation, &t); |
8be17c9b | 1119 | if (r < 0) { |
f25bed67 | 1120 | if (!exclusive && r == -EROFS) |
771b7ead | 1121 | log_debug_errno(r, "Failed to create shared lock for '%s', ignoring: %m", path); |
8be17c9b LT |
1122 | else |
1123 | return r; | |
1124 | } | |
eb38edce | 1125 | } |
30535c16 LP |
1126 | |
1127 | if (p) { | |
6e5dcce4 | 1128 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); |
30535c16 LP |
1129 | |
1130 | r = make_lock_file(p, operation, global); | |
1131 | if (r < 0) { | |
1132 | release_lock_file(&t); | |
1133 | return r; | |
1134 | } | |
546dbec5 LP |
1135 | } else |
1136 | *global = (LockFile) LOCK_FILE_INIT; | |
30535c16 LP |
1137 | |
1138 | *local = t; | |
1139 | return 0; | |
1140 | } | |
1141 | ||
cb81cd80 | 1142 | int image_set_limit(Image *i, uint64_t referenced_max) { |
d6ce17c7 LP |
1143 | assert(i); |
1144 | ||
d94c2b06 | 1145 | if (IMAGE_IS_VENDOR(i) || IMAGE_IS_HOST(i)) |
d6ce17c7 LP |
1146 | return -EROFS; |
1147 | ||
1148 | if (i->type != IMAGE_SUBVOLUME) | |
15411c0c | 1149 | return -EOPNOTSUPP; |
d6ce17c7 | 1150 | |
5bcd08db LP |
1151 | /* We set the quota both for the subvolume as well as for the |
1152 | * subtree. The latter is mostly for historical reasons, since | |
1153 | * we didn't use to have a concept of subtree quota, and hence | |
1154 | * only modified the subvolume quota. */ | |
1155 | ||
1156 | (void) btrfs_qgroup_set_limit(i->path, 0, referenced_max); | |
1157 | (void) btrfs_subvol_auto_qgroup(i->path, 0, true); | |
1158 | return btrfs_subvol_set_subtree_quota_limit(i->path, 0, referenced_max); | |
d6ce17c7 LP |
1159 | } |
1160 | ||
84be0c71 | 1161 | int image_read_metadata(Image *i, const ImagePolicy *image_policy) { |
8e766630 | 1162 | _cleanup_(release_lock_file) LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT; |
c7664c07 LP |
1163 | int r; |
1164 | ||
1165 | assert(i); | |
1166 | ||
1167 | r = image_path_lock(i->path, LOCK_SH|LOCK_NB, &global_lock, &local_lock); | |
1168 | if (r < 0) | |
1169 | return r; | |
1170 | ||
1171 | switch (i->type) { | |
1172 | ||
1173 | case IMAGE_SUBVOLUME: | |
1174 | case IMAGE_DIRECTORY: { | |
bcf94222 | 1175 | _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL; |
c7664c07 LP |
1176 | sd_id128_t machine_id = SD_ID128_NULL; |
1177 | _cleanup_free_ char *hostname = NULL; | |
1178 | _cleanup_free_ char *path = NULL; | |
1179 | ||
b60e0f57 | 1180 | if (i->class == IMAGE_SYSEXT) { |
6afa5d86 LB |
1181 | r = extension_has_forbidden_content(i->path); |
1182 | if (r < 0) | |
1183 | return r; | |
1184 | if (r > 0) | |
1185 | return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM), | |
1186 | "Conflicting content found in image %s, refusing.", | |
1187 | i->name); | |
1188 | } | |
1189 | ||
f461a28d | 1190 | r = chase("/etc/hostname", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1191 | if (r < 0 && r != -ENOENT) |
1192 | log_debug_errno(r, "Failed to chase /etc/hostname in image %s: %m", i->name); | |
1193 | else if (r >= 0) { | |
1194 | r = read_etc_hostname(path, &hostname); | |
1195 | if (r < 0) | |
1196 | log_debug_errno(errno, "Failed to read /etc/hostname of image %s: %m", i->name); | |
1197 | } | |
1198 | ||
1199 | path = mfree(path); | |
1200 | ||
f7b5f399 YW |
1201 | r = id128_get_machine(i->path, &machine_id); |
1202 | if (r < 0) | |
1203 | log_debug_errno(r, "Failed to read machine ID in image %s, ignoring: %m", i->name); | |
c7664c07 | 1204 | |
f461a28d | 1205 | r = chase("/etc/machine-info", i->path, CHASE_PREFIX_ROOT|CHASE_TRAIL_SLASH, &path, NULL); |
c7664c07 LP |
1206 | if (r < 0 && r != -ENOENT) |
1207 | log_debug_errno(r, "Failed to chase /etc/machine-info in image %s: %m", i->name); | |
1208 | else if (r >= 0) { | |
aa8fbc74 | 1209 | r = load_env_file_pairs(NULL, path, &machine_info); |
c7664c07 LP |
1210 | if (r < 0) |
1211 | log_debug_errno(r, "Failed to parse machine-info data of %s: %m", i->name); | |
1212 | } | |
1213 | ||
d58ad743 LP |
1214 | r = load_os_release_pairs(i->path, &os_release); |
1215 | if (r < 0) | |
1216 | log_debug_errno(r, "Failed to read os-release in image, ignoring: %m"); | |
c7664c07 | 1217 | |
b60e0f57 | 1218 | r = load_extension_release_pairs(i->path, i->class, i->name, /* relax_extension_release_check= */ false, &extension_release); |
bcf94222 LB |
1219 | if (r < 0) |
1220 | log_debug_errno(r, "Failed to read extension-release in image, ignoring: %m"); | |
1221 | ||
c7664c07 LP |
1222 | free_and_replace(i->hostname, hostname); |
1223 | i->machine_id = machine_id; | |
1224 | strv_free_and_replace(i->machine_info, machine_info); | |
1225 | strv_free_and_replace(i->os_release, os_release); | |
bcf94222 | 1226 | strv_free_and_replace(i->extension_release, extension_release); |
c7664c07 LP |
1227 | |
1228 | break; | |
1229 | } | |
1230 | ||
1231 | case IMAGE_RAW: | |
1232 | case IMAGE_BLOCK: { | |
1233 | _cleanup_(loop_device_unrefp) LoopDevice *d = NULL; | |
1234 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
1235 | ||
22ee78a8 | 1236 | r = loop_device_make_by_path(i->path, O_RDONLY, /* sector_size= */ UINT32_MAX, LO_FLAGS_PARTSCAN, LOCK_SH, &d); |
41bc4849 LP |
1237 | if (r < 0) |
1238 | return r; | |
1239 | ||
bad31660 YW |
1240 | r = dissect_loop_device( |
1241 | d, | |
84be0c71 LP |
1242 | /* verity= */ NULL, |
1243 | /* mount_options= */ NULL, | |
1244 | image_policy, | |
4b5de5dd LP |
1245 | DISSECT_IMAGE_GENERIC_ROOT | |
1246 | DISSECT_IMAGE_REQUIRE_ROOT | | |
1247 | DISSECT_IMAGE_RELAX_VAR_CHECK | | |
f6f4ec79 | 1248 | DISSECT_IMAGE_READ_ONLY | |
73d88b80 LP |
1249 | DISSECT_IMAGE_USR_NO_ROOT | |
1250 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | | |
1251 | DISSECT_IMAGE_PIN_PARTITION_DEVICES, | |
75dc190d | 1252 | &m); |
c7664c07 LP |
1253 | if (r < 0) |
1254 | return r; | |
1255 | ||
22847508 ZJS |
1256 | r = dissected_image_acquire_metadata(m, |
1257 | DISSECT_IMAGE_VALIDATE_OS | | |
1258 | DISSECT_IMAGE_VALIDATE_OS_EXT); | |
c7664c07 LP |
1259 | if (r < 0) |
1260 | return r; | |
1261 | ||
1262 | free_and_replace(i->hostname, m->hostname); | |
1263 | i->machine_id = m->machine_id; | |
1264 | strv_free_and_replace(i->machine_info, m->machine_info); | |
1265 | strv_free_and_replace(i->os_release, m->os_release); | |
bcf94222 | 1266 | strv_free_and_replace(i->extension_release, m->extension_release); |
c7664c07 LP |
1267 | |
1268 | break; | |
1269 | } | |
1270 | ||
1271 | default: | |
1272 | return -EOPNOTSUPP; | |
1273 | } | |
1274 | ||
1275 | i->metadata_valid = true; | |
1276 | ||
1277 | return 0; | |
1278 | } | |
1279 | ||
30535c16 | 1280 | int image_name_lock(const char *name, int operation, LockFile *ret) { |
99d97afc LP |
1281 | const char *p; |
1282 | ||
30535c16 LP |
1283 | assert(name); |
1284 | assert(ret); | |
1285 | ||
1286 | /* Locks an image name, regardless of the precise path used. */ | |
1287 | ||
99d97afc LP |
1288 | if (streq(name, ".host")) |
1289 | return -EBUSY; | |
1290 | ||
30535c16 LP |
1291 | if (!image_name_is_valid(name)) |
1292 | return -EINVAL; | |
1293 | ||
b6e953f2 LP |
1294 | if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) { |
1295 | *ret = (LockFile) LOCK_FILE_INIT; | |
1296 | return 0; | |
1297 | } | |
1298 | ||
6e5dcce4 | 1299 | (void) mkdir_p("/run/systemd/nspawn/locks", 0700); |
99d97afc LP |
1300 | |
1301 | p = strjoina("/run/systemd/nspawn/locks/name-", name); | |
30535c16 LP |
1302 | return make_lock_file(p, operation, ret); |
1303 | } | |
1304 | ||
d577d4a4 LP |
1305 | bool image_in_search_path( |
1306 | ImageClass class, | |
1307 | const char *root, | |
1308 | const char *image) { | |
1309 | ||
ace9ab19 LP |
1310 | assert(image); |
1311 | ||
73740c9f | 1312 | NULSTR_FOREACH(path, pick_image_search_path(class)) { |
d577d4a4 | 1313 | const char *p, *q; |
ace9ab19 LP |
1314 | size_t k; |
1315 | ||
d577d4a4 LP |
1316 | if (!empty_or_root(root)) { |
1317 | q = path_startswith(path, root); | |
1318 | if (!q) | |
1319 | continue; | |
1320 | } else | |
1321 | q = path; | |
1322 | ||
1323 | p = path_startswith(q, path); | |
ace9ab19 LP |
1324 | if (!p) |
1325 | continue; | |
1326 | ||
1327 | /* Make sure there's a filename following */ | |
1328 | k = strcspn(p, "/"); | |
1329 | if (k == 0) | |
1330 | continue; | |
1331 | ||
1332 | p += k; | |
1333 | ||
1334 | /* Accept trailing slashes */ | |
1335 | if (p[strspn(p, "/")] == 0) | |
1336 | return true; | |
1337 | ||
1338 | } | |
1339 | ||
1340 | return false; | |
1341 | } | |
1342 | ||
cd61c3bf LP |
1343 | static const char* const image_type_table[_IMAGE_TYPE_MAX] = { |
1344 | [IMAGE_DIRECTORY] = "directory", | |
1345 | [IMAGE_SUBVOLUME] = "subvolume", | |
aceac2f0 | 1346 | [IMAGE_RAW] = "raw", |
eb38edce | 1347 | [IMAGE_BLOCK] = "block", |
cd61c3bf LP |
1348 | }; |
1349 | ||
1350 | DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType); |