]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
d7c7c334 | 2 | |
11c3a366 TA |
3 | #include <errno.h> |
4 | #include <fcntl.h> | |
5 | #include <inttypes.h> | |
01234e1f | 6 | #include <linux/btrfs_tree.h> |
17cbb288 | 7 | #include <linux/fs.h> |
11c3a366 | 8 | #include <linux/loop.h> |
01234e1f | 9 | #include <linux/magic.h> |
11c3a366 TA |
10 | #include <stddef.h> |
11 | #include <stdio.h> | |
d7c7c334 | 12 | #include <stdlib.h> |
ac8db36c | 13 | #include <sys/file.h> |
11c3a366 | 14 | #include <sys/ioctl.h> |
11c3a366 TA |
15 | #include <sys/sysmacros.h> |
16 | #include <unistd.h> | |
17 | ||
b5efdb8a | 18 | #include "alloc-util.h" |
18c528e9 | 19 | #include "blockdev-util.h" |
3ffd4af2 | 20 | #include "btrfs-util.h" |
24dbe603 | 21 | #include "chase.h" |
17cbb288 | 22 | #include "chattr-util.h" |
07630cea | 23 | #include "copy.h" |
3ffd4af2 | 24 | #include "fd-util.h" |
07630cea | 25 | #include "fileio.h" |
ef8becfa | 26 | #include "fs-util.h" |
a90fb858 | 27 | #include "io-util.h" |
07630cea | 28 | #include "macro.h" |
d7c7c334 | 29 | #include "path-util.h" |
17cbb288 | 30 | #include "rm-rf.h" |
d7b8eec7 | 31 | #include "smack-util.h" |
11c3a366 | 32 | #include "sparse-endian.h" |
872a590e | 33 | #include "stat-util.h" |
07630cea | 34 | #include "string-util.h" |
93cc7779 | 35 | #include "time-util.h" |
d7c7c334 | 36 | |
62572894 LP |
37 | /* WARNING: Be careful with file system ioctls! When we get an fd, we |
38 | * need to make sure it either refers to only a regular file or | |
39 | * directory, or that it is located on btrfs, before invoking any | |
40 | * btrfs ioctls. The ioctl numbers are reused by some device drivers | |
41 | * (such as DRM), and hence might have bad effects when invoked on | |
42 | * device nodes (that reference drivers) rather than fds to normal | |
43 | * files or directories. */ | |
44 | ||
6d2fd8df | 45 | int btrfs_is_subvol_at(int dir_fd, const char *path) { |
21222ea5 LP |
46 | struct stat st; |
47 | ||
6d2fd8df | 48 | assert(dir_fd >= 0 || dir_fd == AT_FDCWD); |
21222ea5 | 49 | |
cd61c3bf LP |
50 | /* On btrfs subvolumes always have the inode 256 */ |
51 | ||
e089efd7 | 52 | if (fstatat(dir_fd, strempty(path), &st, isempty(path) ? AT_EMPTY_PATH : 0) < 0) |
d7c7c334 LP |
53 | return -errno; |
54 | ||
674b04ff | 55 | if (!btrfs_might_be_subvol(&st)) |
d7c7c334 LP |
56 | return 0; |
57 | ||
6d2fd8df | 58 | return is_fs_type_at(dir_fd, path, BTRFS_SUPER_MAGIC); |
2904e949 LP |
59 | } |
60 | ||
77c66be3 DDM |
61 | int btrfs_subvol_set_read_only_at(int dir_fd, const char *path, bool b) { |
62 | _cleanup_close_ int fd = -EBADF; | |
d7c7c334 | 63 | uint64_t flags, nflags; |
0d6e763b | 64 | struct stat st; |
d7c7c334 | 65 | |
77c66be3 DDM |
66 | assert(dir_fd >= 0 || dir_fd == AT_FDCWD); |
67 | ||
e40b11be | 68 | fd = xopenat(dir_fd, path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY); |
77c66be3 DDM |
69 | if (fd < 0) |
70 | return fd; | |
0d6e763b LP |
71 | |
72 | if (fstat(fd, &st) < 0) | |
d7c7c334 LP |
73 | return -errno; |
74 | ||
674b04ff | 75 | if (!btrfs_might_be_subvol(&st)) |
0d6e763b LP |
76 | return -EINVAL; |
77 | ||
d7c7c334 LP |
78 | if (ioctl(fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags) < 0) |
79 | return -errno; | |
80 | ||
0da96503 | 81 | nflags = UPDATE_FLAG(flags, BTRFS_SUBVOL_RDONLY, b); |
d7c7c334 LP |
82 | if (flags == nflags) |
83 | return 0; | |
84 | ||
7c248223 | 85 | return RET_NERRNO(ioctl(fd, BTRFS_IOC_SUBVOL_SETFLAGS, &nflags)); |
d7c7c334 LP |
86 | } |
87 | ||
10f9c755 | 88 | int btrfs_subvol_get_read_only_fd(int fd) { |
cd61c3bf | 89 | uint64_t flags; |
62572894 LP |
90 | struct stat st; |
91 | ||
92 | assert(fd >= 0); | |
93 | ||
94 | if (fstat(fd, &st) < 0) | |
95 | return -errno; | |
96 | ||
674b04ff | 97 | if (!btrfs_might_be_subvol(&st)) |
62572894 | 98 | return -EINVAL; |
cd61c3bf LP |
99 | |
100 | if (ioctl(fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags) < 0) | |
101 | return -errno; | |
102 | ||
103 | return !!(flags & BTRFS_SUBVOL_RDONLY); | |
104 | } | |
105 | ||
c706b27f | 106 | int btrfs_get_block_device_at(int dir_fd, const char *path, dev_t *ret) { |
d7c7c334 | 107 | struct btrfs_ioctl_fs_info_args fsi = {}; |
c706b27f | 108 | _cleanup_close_ int fd = -EBADF; |
d7c7c334 | 109 | uint64_t id; |
62572894 | 110 | int r; |
d7c7c334 | 111 | |
c706b27f DDM |
112 | assert(dir_fd >= 0 || dir_fd == AT_FDCWD); |
113 | assert(path); | |
114 | assert(ret); | |
d7c7c334 | 115 | |
e40b11be | 116 | fd = xopenat(dir_fd, path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); |
698bc186 LP |
117 | if (fd < 0) |
118 | return fd; | |
119 | ||
65ddc2c5 | 120 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
62572894 LP |
121 | if (r < 0) |
122 | return r; | |
79de6eb1 | 123 | if (r == 0) |
62572894 LP |
124 | return -ENOTTY; |
125 | ||
d7c7c334 LP |
126 | if (ioctl(fd, BTRFS_IOC_FS_INFO, &fsi) < 0) |
127 | return -errno; | |
128 | ||
129 | /* We won't do this for btrfs RAID */ | |
66ae5130 | 130 | if (fsi.num_devices != 1) { |
c706b27f | 131 | *ret = 0; |
d7c7c334 | 132 | return 0; |
66ae5130 | 133 | } |
d7c7c334 LP |
134 | |
135 | for (id = 1; id <= fsi.max_id; id++) { | |
136 | struct btrfs_ioctl_dev_info_args di = { | |
137 | .devid = id, | |
138 | }; | |
139 | struct stat st; | |
140 | ||
141 | if (ioctl(fd, BTRFS_IOC_DEV_INFO, &di) < 0) { | |
142 | if (errno == ENODEV) | |
143 | continue; | |
144 | ||
145 | return -errno; | |
146 | } | |
147 | ||
67f0ac8c LP |
148 | /* For the root fs — when no initrd is involved — btrfs returns /dev/root on any kernels from |
149 | * the past few years. That sucks, as we have no API to determine the actual root then. let's | |
150 | * return an recognizable error for this case, so that the caller can maybe print a nice | |
151 | * message about this. | |
152 | * | |
153 | * https://bugzilla.kernel.org/show_bug.cgi?id=89721 */ | |
154 | if (path_equal((char*) di.path, "/dev/root")) | |
155 | return -EUCLEAN; | |
156 | ||
d7c7c334 LP |
157 | if (stat((char*) di.path, &st) < 0) |
158 | return -errno; | |
159 | ||
160 | if (!S_ISBLK(st.st_mode)) | |
3468e5ac | 161 | return -ENOTBLK; |
d7c7c334 LP |
162 | |
163 | if (major(st.st_rdev) == 0) | |
164 | return -ENODEV; | |
165 | ||
c706b27f | 166 | *ret = st.st_rdev; |
d7c7c334 LP |
167 | return 1; |
168 | } | |
169 | ||
170 | return -ENODEV; | |
171 | } | |
10f9c755 LP |
172 | |
173 | int btrfs_subvol_get_id_fd(int fd, uint64_t *ret) { | |
174 | struct btrfs_ioctl_ino_lookup_args args = { | |
175 | .objectid = BTRFS_FIRST_FREE_OBJECTID | |
176 | }; | |
62572894 | 177 | int r; |
10f9c755 LP |
178 | |
179 | assert(fd >= 0); | |
180 | assert(ret); | |
181 | ||
65ddc2c5 | 182 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
62572894 LP |
183 | if (r < 0) |
184 | return r; | |
79de6eb1 | 185 | if (r == 0) |
62572894 LP |
186 | return -ENOTTY; |
187 | ||
10f9c755 LP |
188 | if (ioctl(fd, BTRFS_IOC_INO_LOOKUP, &args) < 0) |
189 | return -errno; | |
190 | ||
191 | *ret = args.treeid; | |
192 | return 0; | |
193 | } | |
194 | ||
90578cbd | 195 | int btrfs_subvol_get_id(int fd, const char *subvol, uint64_t *ret) { |
254d1313 | 196 | _cleanup_close_ int subvol_fd = -EBADF; |
90578cbd LP |
197 | |
198 | assert(fd >= 0); | |
199 | assert(ret); | |
200 | ||
201 | subvol_fd = openat(fd, subvol, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
202 | if (subvol_fd < 0) | |
203 | return -errno; | |
204 | ||
205 | return btrfs_subvol_get_id_fd(subvol_fd, ret); | |
206 | } | |
207 | ||
5743a585 LP |
208 | static bool btrfs_ioctl_search_args_inc(struct btrfs_ioctl_search_args *args) { |
209 | assert(args); | |
210 | ||
211 | /* the objectid, type, offset together make up the btrfs key, | |
212 | * which is considered a single 136byte integer when | |
213 | * comparing. This call increases the counter by one, dealing | |
214 | * with the overflow between the overflows */ | |
215 | ||
f5fbe71d | 216 | if (args->key.min_offset < UINT64_MAX) { |
5743a585 LP |
217 | args->key.min_offset++; |
218 | return true; | |
219 | } | |
220 | ||
f5fbe71d | 221 | if (args->key.min_type < UINT8_MAX) { |
5743a585 LP |
222 | args->key.min_type++; |
223 | args->key.min_offset = 0; | |
224 | return true; | |
225 | } | |
226 | ||
f5fbe71d | 227 | if (args->key.min_objectid < UINT64_MAX) { |
5743a585 LP |
228 | args->key.min_objectid++; |
229 | args->key.min_offset = 0; | |
230 | args->key.min_type = 0; | |
231 | return true; | |
232 | } | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
237 | static void btrfs_ioctl_search_args_set(struct btrfs_ioctl_search_args *args, const struct btrfs_ioctl_search_header *h) { | |
238 | assert(args); | |
239 | assert(h); | |
240 | ||
241 | args->key.min_objectid = h->objectid; | |
242 | args->key.min_type = h->type; | |
243 | args->key.min_offset = h->offset; | |
244 | } | |
245 | ||
246 | static int btrfs_ioctl_search_args_compare(const struct btrfs_ioctl_search_args *args) { | |
90c88092 YW |
247 | int r; |
248 | ||
5743a585 LP |
249 | assert(args); |
250 | ||
251 | /* Compare min and max */ | |
252 | ||
90c88092 YW |
253 | r = CMP(args->key.min_objectid, args->key.max_objectid); |
254 | if (r != 0) | |
255 | return r; | |
5743a585 | 256 | |
90c88092 YW |
257 | r = CMP(args->key.min_type, args->key.max_type); |
258 | if (r != 0) | |
259 | return r; | |
5743a585 | 260 | |
6dd91b36 | 261 | return CMP(args->key.min_offset, args->key.max_offset); |
5743a585 LP |
262 | } |
263 | ||
801bf40c | 264 | typedef struct BtrfsForeachIterator { |
125cca1b YW |
265 | const struct btrfs_ioctl_search_args *args; |
266 | size_t offset; | |
267 | unsigned index; | |
268 | struct btrfs_ioctl_search_header *header; | |
269 | const void **body; | |
801bf40c LP |
270 | } BtrfsForeachIterator; |
271 | ||
125cca1b YW |
272 | static int btrfs_iterate(BtrfsForeachIterator *i) { |
273 | assert(i); | |
274 | assert(i->args); | |
275 | assert(i->header); | |
276 | assert(i->body); | |
277 | ||
278 | if (i->index >= i->args->key.nr_items) | |
279 | return 0; /* end */ | |
280 | ||
281 | assert_cc(BTRFS_SEARCH_ARGS_BUFSIZE >= sizeof(struct btrfs_ioctl_search_header)); | |
282 | if (i->offset > BTRFS_SEARCH_ARGS_BUFSIZE - sizeof(struct btrfs_ioctl_search_header)) | |
283 | return -EBADMSG; | |
284 | ||
285 | struct btrfs_ioctl_search_header h; | |
286 | memcpy(&h, (const uint8_t*) i->args->buf + i->offset, sizeof(struct btrfs_ioctl_search_header)); | |
287 | ||
288 | if (i->offset > BTRFS_SEARCH_ARGS_BUFSIZE - sizeof(struct btrfs_ioctl_search_header) - h.len) | |
289 | return -EBADMSG; | |
290 | ||
291 | *i->body = (const uint8_t*) i->args->buf + i->offset + sizeof(struct btrfs_ioctl_search_header); | |
292 | *i->header = h; | |
293 | i->offset += sizeof(struct btrfs_ioctl_search_header) + h.len; | |
294 | i->index++; | |
295 | ||
296 | return 1; | |
297 | } | |
298 | ||
801bf40c LP |
299 | /* Iterates through a series of struct btrfs_file_extent_item elements. They are unfortunately not aligned, |
300 | * hence we copy out the header from them */ | |
125cca1b | 301 | #define FOREACH_BTRFS_IOCTL_SEARCH_HEADER(_sh, _body, _args) \ |
801bf40c | 302 | for (BtrfsForeachIterator iterator = { \ |
125cca1b YW |
303 | .args = &(_args), \ |
304 | .header = &(_sh), \ | |
305 | .body = &(_body), \ | |
801bf40c | 306 | }; \ |
125cca1b | 307 | btrfs_iterate(&iterator) > 0; ) |
5743a585 | 308 | |
5bcd08db | 309 | int btrfs_subvol_get_info_fd(int fd, uint64_t subvol_id, BtrfsSubvolInfo *ret) { |
10f9c755 LP |
310 | struct btrfs_ioctl_search_args args = { |
311 | /* Tree of tree roots */ | |
b6b18498 | 312 | .key.tree_id = BTRFS_ROOT_TREE_OBJECTID, |
10f9c755 LP |
313 | |
314 | /* Look precisely for the subvolume items */ | |
315 | .key.min_type = BTRFS_ROOT_ITEM_KEY, | |
316 | .key.max_type = BTRFS_ROOT_ITEM_KEY, | |
317 | ||
10f9c755 | 318 | .key.min_offset = 0, |
f5fbe71d | 319 | .key.max_offset = UINT64_MAX, |
5743a585 LP |
320 | |
321 | /* No restrictions on the other components */ | |
10f9c755 | 322 | .key.min_transid = 0, |
f5fbe71d | 323 | .key.max_transid = UINT64_MAX, |
10f9c755 LP |
324 | }; |
325 | ||
b6b18498 | 326 | bool found = false; |
10f9c755 LP |
327 | int r; |
328 | ||
329 | assert(fd >= 0); | |
330 | assert(ret); | |
331 | ||
5bcd08db LP |
332 | if (subvol_id == 0) { |
333 | r = btrfs_subvol_get_id_fd(fd, &subvol_id); | |
334 | if (r < 0) | |
335 | return r; | |
336 | } else { | |
65ddc2c5 | 337 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
5bcd08db LP |
338 | if (r < 0) |
339 | return r; | |
79de6eb1 | 340 | if (r == 0) |
5bcd08db LP |
341 | return -ENOTTY; |
342 | } | |
10f9c755 LP |
343 | |
344 | args.key.min_objectid = args.key.max_objectid = subvol_id; | |
10f9c755 | 345 | |
5743a585 | 346 | while (btrfs_ioctl_search_args_compare(&args) <= 0) { |
801bf40c LP |
347 | struct btrfs_ioctl_search_header sh; |
348 | const void *body; | |
b6b18498 LP |
349 | |
350 | args.key.nr_items = 256; | |
351 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) | |
352 | return -errno; | |
353 | ||
354 | if (args.key.nr_items <= 0) | |
355 | break; | |
10f9c755 | 356 | |
801bf40c | 357 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) { |
5743a585 | 358 | /* Make sure we start the next search at least from this entry */ |
801bf40c | 359 | btrfs_ioctl_search_args_set(&args, &sh); |
5743a585 | 360 | |
801bf40c | 361 | if (sh.objectid != subvol_id) |
b6b18498 | 362 | continue; |
801bf40c | 363 | if (sh.type != BTRFS_ROOT_ITEM_KEY) |
b6b18498 | 364 | continue; |
5743a585 LP |
365 | |
366 | /* Older versions of the struct lacked the otime setting */ | |
801bf40c | 367 | if (sh.len < offsetof(struct btrfs_root_item, otime) + sizeof(struct btrfs_timespec)) |
b6b18498 | 368 | continue; |
10f9c755 | 369 | |
801bf40c | 370 | const struct btrfs_root_item *ri = body; |
b6b18498 LP |
371 | ret->otime = (usec_t) le64toh(ri->otime.sec) * USEC_PER_SEC + |
372 | (usec_t) le32toh(ri->otime.nsec) / NSEC_PER_USEC; | |
10f9c755 | 373 | |
b6b18498 | 374 | ret->subvol_id = subvol_id; |
5d904a6a | 375 | ret->read_only = le64toh(ri->flags) & BTRFS_ROOT_SUBVOL_RDONLY; |
10f9c755 | 376 | |
b6b18498 LP |
377 | assert_cc(sizeof(ri->uuid) == sizeof(ret->uuid)); |
378 | memcpy(&ret->uuid, ri->uuid, sizeof(ret->uuid)); | |
379 | memcpy(&ret->parent_uuid, ri->parent_uuid, sizeof(ret->parent_uuid)); | |
380 | ||
381 | found = true; | |
382 | goto finish; | |
383 | } | |
384 | ||
5743a585 LP |
385 | /* Increase search key by one, to read the next item, if we can. */ |
386 | if (!btrfs_ioctl_search_args_inc(&args)) | |
b6b18498 LP |
387 | break; |
388 | } | |
389 | ||
390 | finish: | |
246caacb | 391 | return found ? 0 : -ENODATA; |
b6b18498 LP |
392 | } |
393 | ||
5bcd08db | 394 | int btrfs_qgroup_get_quota_fd(int fd, uint64_t qgroupid, BtrfsQuotaInfo *ret) { |
b6b18498 LP |
395 | |
396 | struct btrfs_ioctl_search_args args = { | |
397 | /* Tree of quota items */ | |
398 | .key.tree_id = BTRFS_QUOTA_TREE_OBJECTID, | |
399 | ||
5743a585 LP |
400 | /* The object ID is always 0 */ |
401 | .key.min_objectid = 0, | |
402 | .key.max_objectid = 0, | |
403 | ||
b6b18498 LP |
404 | /* Look precisely for the quota items */ |
405 | .key.min_type = BTRFS_QGROUP_STATUS_KEY, | |
406 | .key.max_type = BTRFS_QGROUP_LIMIT_KEY, | |
407 | ||
b6b18498 LP |
408 | /* No restrictions on the other components */ |
409 | .key.min_transid = 0, | |
f5fbe71d | 410 | .key.max_transid = UINT64_MAX, |
b6b18498 LP |
411 | }; |
412 | ||
b6b18498 LP |
413 | bool found_info = false, found_limit = false; |
414 | int r; | |
415 | ||
416 | assert(fd >= 0); | |
417 | assert(ret); | |
418 | ||
5bcd08db LP |
419 | if (qgroupid == 0) { |
420 | r = btrfs_subvol_get_id_fd(fd, &qgroupid); | |
421 | if (r < 0) | |
422 | return r; | |
423 | } else { | |
65ddc2c5 | 424 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
5bcd08db LP |
425 | if (r < 0) |
426 | return r; | |
79de6eb1 | 427 | if (r == 0) |
5bcd08db LP |
428 | return -ENOTTY; |
429 | } | |
b6b18498 | 430 | |
5bcd08db | 431 | args.key.min_offset = args.key.max_offset = qgroupid; |
b6b18498 | 432 | |
5743a585 | 433 | while (btrfs_ioctl_search_args_compare(&args) <= 0) { |
801bf40c LP |
434 | struct btrfs_ioctl_search_header sh; |
435 | const void *body; | |
b6b18498 LP |
436 | |
437 | args.key.nr_items = 256; | |
12ee6186 LP |
438 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) { |
439 | if (errno == ENOENT) /* quota tree is missing: quota disabled */ | |
440 | break; | |
441 | ||
b6b18498 | 442 | return -errno; |
12ee6186 | 443 | } |
b6b18498 LP |
444 | |
445 | if (args.key.nr_items <= 0) | |
446 | break; | |
447 | ||
801bf40c | 448 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) { |
b6b18498 | 449 | |
5743a585 | 450 | /* Make sure we start the next search at least from this entry */ |
801bf40c | 451 | btrfs_ioctl_search_args_set(&args, &sh); |
b6b18498 | 452 | |
801bf40c | 453 | if (sh.objectid != 0) |
b6b18498 | 454 | continue; |
801bf40c | 455 | if (sh.offset != qgroupid) |
b6b18498 LP |
456 | continue; |
457 | ||
801bf40c LP |
458 | if (sh.type == BTRFS_QGROUP_INFO_KEY) { |
459 | const struct btrfs_qgroup_info_item *qii = body; | |
b6b18498 | 460 | |
cb81cd80 | 461 | ret->referenced = le64toh(qii->rfer); |
b6b18498 LP |
462 | ret->exclusive = le64toh(qii->excl); |
463 | ||
464 | found_info = true; | |
465 | ||
801bf40c LP |
466 | } else if (sh.type == BTRFS_QGROUP_LIMIT_KEY) { |
467 | const struct btrfs_qgroup_limit_item *qli = body; | |
b6b18498 | 468 | |
5bcd08db LP |
469 | if (le64toh(qli->flags) & BTRFS_QGROUP_LIMIT_MAX_RFER) |
470 | ret->referenced_max = le64toh(qli->max_rfer); | |
471 | else | |
f5fbe71d | 472 | ret->referenced_max = UINT64_MAX; |
5bcd08db LP |
473 | |
474 | if (le64toh(qli->flags) & BTRFS_QGROUP_LIMIT_MAX_EXCL) | |
475 | ret->exclusive_max = le64toh(qli->max_excl); | |
476 | else | |
f5fbe71d | 477 | ret->exclusive_max = UINT64_MAX; |
b6b18498 LP |
478 | |
479 | found_limit = true; | |
480 | } | |
481 | ||
482 | if (found_info && found_limit) | |
483 | goto finish; | |
484 | } | |
485 | ||
5743a585 LP |
486 | /* Increase search key by one, to read the next item, if we can. */ |
487 | if (!btrfs_ioctl_search_args_inc(&args)) | |
b6b18498 LP |
488 | break; |
489 | } | |
490 | ||
491 | finish: | |
492 | if (!found_limit && !found_info) | |
493 | return -ENODATA; | |
494 | ||
495 | if (!found_info) { | |
f5fbe71d YW |
496 | ret->referenced = UINT64_MAX; |
497 | ret->exclusive = UINT64_MAX; | |
b6b18498 LP |
498 | } |
499 | ||
500 | if (!found_limit) { | |
f5fbe71d YW |
501 | ret->referenced_max = UINT64_MAX; |
502 | ret->exclusive_max = UINT64_MAX; | |
b6b18498 | 503 | } |
10f9c755 LP |
504 | |
505 | return 0; | |
506 | } | |
f27a3864 | 507 | |
5bcd08db | 508 | int btrfs_qgroup_get_quota(const char *path, uint64_t qgroupid, BtrfsQuotaInfo *ret) { |
254d1313 | 509 | _cleanup_close_ int fd = -EBADF; |
5bcd08db LP |
510 | |
511 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
512 | if (fd < 0) | |
513 | return -errno; | |
514 | ||
515 | return btrfs_qgroup_get_quota_fd(fd, qgroupid, ret); | |
516 | } | |
517 | ||
518 | int btrfs_subvol_find_subtree_qgroup(int fd, uint64_t subvol_id, uint64_t *ret) { | |
f5fbe71d | 519 | uint64_t level, lowest = UINT64_MAX, lowest_qgroupid = 0; |
5bcd08db | 520 | _cleanup_free_ uint64_t *qgroups = NULL; |
cecaba20 | 521 | int r, n; |
5bcd08db LP |
522 | |
523 | assert(fd >= 0); | |
524 | assert(ret); | |
525 | ||
526 | /* This finds the "subtree" qgroup for a specific | |
527 | * subvolume. This only works for subvolumes that have been | |
528 | * prepared with btrfs_subvol_auto_qgroup_fd() with | |
529 | * insert_intermediary_qgroup=true (or equivalent). For others | |
530 | * it will return the leaf qgroup instead. The two cases may | |
3a258d3a | 531 | * be distinguished via the return value, which is 1 in case |
5bcd08db LP |
532 | * an appropriate "subtree" qgroup was found, and 0 |
533 | * otherwise. */ | |
534 | ||
535 | if (subvol_id == 0) { | |
536 | r = btrfs_subvol_get_id_fd(fd, &subvol_id); | |
537 | if (r < 0) | |
538 | return r; | |
539 | } | |
540 | ||
541 | r = btrfs_qgroupid_split(subvol_id, &level, NULL); | |
542 | if (r < 0) | |
543 | return r; | |
544 | if (level != 0) /* Input must be a leaf qgroup */ | |
545 | return -EINVAL; | |
546 | ||
547 | n = btrfs_qgroup_find_parents(fd, subvol_id, &qgroups); | |
548 | if (n < 0) | |
549 | return n; | |
550 | ||
cecaba20 | 551 | for (int i = 0; i < n; i++) { |
5bcd08db LP |
552 | uint64_t id; |
553 | ||
554 | r = btrfs_qgroupid_split(qgroups[i], &level, &id); | |
555 | if (r < 0) | |
556 | return r; | |
557 | ||
558 | if (id != subvol_id) | |
559 | continue; | |
560 | ||
f5fbe71d | 561 | if (lowest == UINT64_MAX || level < lowest) { |
5bcd08db LP |
562 | lowest_qgroupid = qgroups[i]; |
563 | lowest = level; | |
564 | } | |
565 | } | |
566 | ||
f5fbe71d | 567 | if (lowest == UINT64_MAX) { |
5bcd08db LP |
568 | /* No suitable higher-level qgroup found, let's return |
569 | * the leaf qgroup instead, and indicate that with the | |
570 | * return value. */ | |
571 | ||
572 | *ret = subvol_id; | |
573 | return 0; | |
574 | } | |
575 | ||
576 | *ret = lowest_qgroupid; | |
577 | return 1; | |
578 | } | |
579 | ||
580 | int btrfs_subvol_get_subtree_quota_fd(int fd, uint64_t subvol_id, BtrfsQuotaInfo *ret) { | |
581 | uint64_t qgroupid; | |
582 | int r; | |
583 | ||
584 | assert(fd >= 0); | |
585 | assert(ret); | |
586 | ||
587 | /* This determines the quota data of the qgroup with the | |
588 | * lowest level, that shares the id part with the specified | |
589 | * subvolume. This is useful for determining the quota data | |
590 | * for entire subvolume subtrees, as long as the subtrees have | |
591 | * been set up with btrfs_qgroup_subvol_auto_fd() or in a | |
592 | * compatible way */ | |
593 | ||
594 | r = btrfs_subvol_find_subtree_qgroup(fd, subvol_id, &qgroupid); | |
595 | if (r < 0) | |
596 | return r; | |
597 | ||
598 | return btrfs_qgroup_get_quota_fd(fd, qgroupid, ret); | |
599 | } | |
600 | ||
601 | int btrfs_subvol_get_subtree_quota(const char *path, uint64_t subvol_id, BtrfsQuotaInfo *ret) { | |
254d1313 | 602 | _cleanup_close_ int fd = -EBADF; |
5bcd08db LP |
603 | |
604 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
605 | if (fd < 0) | |
606 | return -errno; | |
607 | ||
608 | return btrfs_subvol_get_subtree_quota_fd(fd, subvol_id, ret); | |
609 | } | |
610 | ||
4117366a YW |
611 | int btrfs_defrag_fd(int fd) { |
612 | int r; | |
613 | ||
614 | assert(fd >= 0); | |
615 | ||
616 | r = fd_verify_regular(fd); | |
617 | if (r < 0) | |
618 | return r; | |
619 | ||
620 | return RET_NERRNO(ioctl(fd, BTRFS_IOC_DEFRAG, NULL)); | |
621 | } | |
622 | ||
f27a3864 | 623 | int btrfs_defrag(const char *p) { |
254d1313 | 624 | _cleanup_close_ int fd = -EBADF; |
f27a3864 LP |
625 | |
626 | fd = open(p, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
627 | if (fd < 0) | |
628 | return -errno; | |
629 | ||
630 | return btrfs_defrag_fd(fd); | |
631 | } | |
754061ce LP |
632 | |
633 | int btrfs_quota_enable_fd(int fd, bool b) { | |
634 | struct btrfs_ioctl_quota_ctl_args args = { | |
635 | .cmd = b ? BTRFS_QUOTA_CTL_ENABLE : BTRFS_QUOTA_CTL_DISABLE, | |
636 | }; | |
62572894 | 637 | int r; |
754061ce LP |
638 | |
639 | assert(fd >= 0); | |
640 | ||
65ddc2c5 | 641 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
62572894 LP |
642 | if (r < 0) |
643 | return r; | |
79de6eb1 | 644 | if (r == 0) |
62572894 LP |
645 | return -ENOTTY; |
646 | ||
7c248223 | 647 | return RET_NERRNO(ioctl(fd, BTRFS_IOC_QUOTA_CTL, &args)); |
754061ce LP |
648 | } |
649 | ||
650 | int btrfs_quota_enable(const char *path, bool b) { | |
254d1313 | 651 | _cleanup_close_ int fd = -EBADF; |
754061ce LP |
652 | |
653 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
654 | if (fd < 0) | |
655 | return -errno; | |
656 | ||
657 | return btrfs_quota_enable_fd(fd, b); | |
658 | } | |
d6ce17c7 | 659 | |
5bcd08db LP |
660 | int btrfs_qgroup_set_limit_fd(int fd, uint64_t qgroupid, uint64_t referenced_max) { |
661 | ||
d6ce17c7 | 662 | struct btrfs_ioctl_qgroup_limit_args args = { |
5bcd08db | 663 | .lim.max_rfer = referenced_max, |
d6ce17c7 LP |
664 | .lim.flags = BTRFS_QGROUP_LIMIT_MAX_RFER, |
665 | }; | |
62572894 | 666 | int r; |
d6ce17c7 LP |
667 | |
668 | assert(fd >= 0); | |
669 | ||
5bcd08db LP |
670 | if (qgroupid == 0) { |
671 | r = btrfs_subvol_get_id_fd(fd, &qgroupid); | |
672 | if (r < 0) | |
673 | return r; | |
674 | } else { | |
65ddc2c5 | 675 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
5bcd08db LP |
676 | if (r < 0) |
677 | return r; | |
79de6eb1 | 678 | if (r == 0) |
5bcd08db LP |
679 | return -ENOTTY; |
680 | } | |
62572894 | 681 | |
5bcd08db LP |
682 | args.qgroupid = qgroupid; |
683 | ||
cecaba20 | 684 | for (unsigned c = 0;; c++) { |
5bcd08db LP |
685 | if (ioctl(fd, BTRFS_IOC_QGROUP_LIMIT, &args) < 0) { |
686 | ||
687 | if (errno == EBUSY && c < 10) { | |
688 | (void) btrfs_quota_scan_wait(fd); | |
689 | continue; | |
690 | } | |
691 | ||
692 | return -errno; | |
693 | } | |
694 | ||
695 | break; | |
696 | } | |
d6ce17c7 LP |
697 | |
698 | return 0; | |
699 | } | |
700 | ||
5bcd08db | 701 | int btrfs_qgroup_set_limit(const char *path, uint64_t qgroupid, uint64_t referenced_max) { |
254d1313 | 702 | _cleanup_close_ int fd = -EBADF; |
5bcd08db LP |
703 | |
704 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
705 | if (fd < 0) | |
706 | return -errno; | |
707 | ||
708 | return btrfs_qgroup_set_limit_fd(fd, qgroupid, referenced_max); | |
709 | } | |
710 | ||
711 | int btrfs_subvol_set_subtree_quota_limit_fd(int fd, uint64_t subvol_id, uint64_t referenced_max) { | |
712 | uint64_t qgroupid; | |
713 | int r; | |
714 | ||
715 | assert(fd >= 0); | |
716 | ||
717 | r = btrfs_subvol_find_subtree_qgroup(fd, subvol_id, &qgroupid); | |
718 | if (r < 0) | |
719 | return r; | |
720 | ||
721 | return btrfs_qgroup_set_limit_fd(fd, qgroupid, referenced_max); | |
722 | } | |
723 | ||
724 | int btrfs_subvol_set_subtree_quota_limit(const char *path, uint64_t subvol_id, uint64_t referenced_max) { | |
254d1313 | 725 | _cleanup_close_ int fd = -EBADF; |
d6ce17c7 LP |
726 | |
727 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); | |
728 | if (fd < 0) | |
729 | return -errno; | |
730 | ||
5bcd08db | 731 | return btrfs_subvol_set_subtree_quota_limit_fd(fd, subvol_id, referenced_max); |
d6ce17c7 | 732 | } |
efe02862 | 733 | |
5bcd08db | 734 | int btrfs_qgroupid_make(uint64_t level, uint64_t id, uint64_t *ret) { |
3f952f92 LP |
735 | assert(ret); |
736 | ||
737 | if (level >= (UINT64_C(1) << (64 - BTRFS_QGROUP_LEVEL_SHIFT))) | |
738 | return -EINVAL; | |
739 | ||
740 | if (id >= (UINT64_C(1) << BTRFS_QGROUP_LEVEL_SHIFT)) | |
741 | return -EINVAL; | |
742 | ||
743 | *ret = (level << BTRFS_QGROUP_LEVEL_SHIFT) | id; | |
744 | return 0; | |
745 | } | |
746 | ||
5bcd08db LP |
747 | int btrfs_qgroupid_split(uint64_t qgroupid, uint64_t *level, uint64_t *id) { |
748 | assert(level || id); | |
749 | ||
750 | if (level) | |
751 | *level = qgroupid >> BTRFS_QGROUP_LEVEL_SHIFT; | |
752 | ||
753 | if (id) | |
754 | *id = qgroupid & ((UINT64_C(1) << BTRFS_QGROUP_LEVEL_SHIFT) - 1); | |
755 | ||
756 | return 0; | |
757 | } | |
758 | ||
759 | static int qgroup_create_or_destroy(int fd, bool b, uint64_t qgroupid) { | |
3f952f92 LP |
760 | |
761 | struct btrfs_ioctl_qgroup_create_args args = { | |
762 | .create = b, | |
5bcd08db | 763 | .qgroupid = qgroupid, |
3f952f92 | 764 | }; |
3f952f92 LP |
765 | int r; |
766 | ||
65ddc2c5 | 767 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
3f952f92 LP |
768 | if (r < 0) |
769 | return r; | |
5bcd08db LP |
770 | if (r == 0) |
771 | return -ENOTTY; | |
772 | ||
cecaba20 | 773 | for (unsigned c = 0;; c++) { |
5bcd08db LP |
774 | if (ioctl(fd, BTRFS_IOC_QGROUP_CREATE, &args) < 0) { |
775 | ||
4b019d2f LP |
776 | /* On old kernels if quota is not enabled, we get EINVAL. On newer kernels we get |
777 | * ENOTCONN. Let's always convert this to ENOTCONN to make this recognizable | |
778 | * everywhere the same way. */ | |
779 | ||
780 | if (IN_SET(errno, EINVAL, ENOTCONN)) | |
781 | return -ENOTCONN; | |
be6d467c | 782 | |
5bcd08db LP |
783 | if (errno == EBUSY && c < 10) { |
784 | (void) btrfs_quota_scan_wait(fd); | |
785 | continue; | |
786 | } | |
787 | ||
788 | return -errno; | |
789 | } | |
3f952f92 | 790 | |
5bcd08db LP |
791 | break; |
792 | } | |
793 | ||
794 | return 0; | |
795 | } | |
796 | ||
797 | int btrfs_qgroup_create(int fd, uint64_t qgroupid) { | |
798 | return qgroup_create_or_destroy(fd, true, qgroupid); | |
799 | } | |
800 | ||
801 | int btrfs_qgroup_destroy(int fd, uint64_t qgroupid) { | |
802 | return qgroup_create_or_destroy(fd, false, qgroupid); | |
803 | } | |
804 | ||
805 | int btrfs_qgroup_destroy_recursive(int fd, uint64_t qgroupid) { | |
806 | _cleanup_free_ uint64_t *qgroups = NULL; | |
807 | uint64_t subvol_id; | |
cecaba20 | 808 | int n, r; |
5bcd08db LP |
809 | |
810 | /* Destroys the specified qgroup, but unassigns it from all | |
811 | * its parents first. Also, it recursively destroys all | |
1b2a7d92 | 812 | * qgroups it is assigned to that have the same id part of the |
5bcd08db LP |
813 | * qgroupid as the specified group. */ |
814 | ||
815 | r = btrfs_qgroupid_split(qgroupid, NULL, &subvol_id); | |
816 | if (r < 0) | |
817 | return r; | |
818 | ||
819 | n = btrfs_qgroup_find_parents(fd, qgroupid, &qgroups); | |
820 | if (n < 0) | |
821 | return n; | |
822 | ||
cecaba20 | 823 | for (int i = 0; i < n; i++) { |
5bcd08db LP |
824 | uint64_t id; |
825 | ||
826 | r = btrfs_qgroupid_split(qgroups[i], NULL, &id); | |
827 | if (r < 0) | |
828 | return r; | |
829 | ||
830 | r = btrfs_qgroup_unassign(fd, qgroupid, qgroups[i]); | |
831 | if (r < 0) | |
832 | return r; | |
833 | ||
834 | if (id != subvol_id) | |
835 | continue; | |
836 | ||
837 | /* The parent qgroupid shares the same id part with | |
838 | * us? If so, destroy it too. */ | |
839 | ||
840 | (void) btrfs_qgroup_destroy_recursive(fd, qgroups[i]); | |
841 | } | |
842 | ||
843 | return btrfs_qgroup_destroy(fd, qgroupid); | |
844 | } | |
845 | ||
846 | int btrfs_quota_scan_start(int fd) { | |
847 | struct btrfs_ioctl_quota_rescan_args args = {}; | |
848 | ||
849 | assert(fd >= 0); | |
850 | ||
7c248223 | 851 | return RET_NERRNO(ioctl(fd, BTRFS_IOC_QUOTA_RESCAN, &args)); |
3f952f92 LP |
852 | } |
853 | ||
5bcd08db LP |
854 | int btrfs_quota_scan_wait(int fd) { |
855 | assert(fd >= 0); | |
856 | ||
7c248223 | 857 | return RET_NERRNO(ioctl(fd, BTRFS_IOC_QUOTA_RESCAN_WAIT)); |
3f952f92 LP |
858 | } |
859 | ||
5bcd08db LP |
860 | int btrfs_quota_scan_ongoing(int fd) { |
861 | struct btrfs_ioctl_quota_rescan_args args = {}; | |
862 | ||
863 | assert(fd >= 0); | |
864 | ||
865 | if (ioctl(fd, BTRFS_IOC_QUOTA_RESCAN_STATUS, &args) < 0) | |
866 | return -errno; | |
867 | ||
868 | return !!args.flags; | |
3f952f92 LP |
869 | } |
870 | ||
5bcd08db LP |
871 | static int qgroup_assign_or_unassign(int fd, bool b, uint64_t child, uint64_t parent) { |
872 | struct btrfs_ioctl_qgroup_assign_args args = { | |
873 | .assign = b, | |
874 | .src = child, | |
875 | .dst = parent, | |
876 | }; | |
5bcd08db LP |
877 | int r; |
878 | ||
65ddc2c5 | 879 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
5bcd08db LP |
880 | if (r < 0) |
881 | return r; | |
882 | if (r == 0) | |
883 | return -ENOTTY; | |
884 | ||
cecaba20 | 885 | for (unsigned c = 0;; c++) { |
5bcd08db LP |
886 | r = ioctl(fd, BTRFS_IOC_QGROUP_ASSIGN, &args); |
887 | if (r < 0) { | |
888 | if (errno == EBUSY && c < 10) { | |
889 | (void) btrfs_quota_scan_wait(fd); | |
890 | continue; | |
891 | } | |
892 | ||
893 | return -errno; | |
894 | } | |
895 | ||
896 | if (r == 0) | |
897 | return 0; | |
898 | ||
899 | /* If the return value is > 0, we need to request a rescan */ | |
900 | ||
901 | (void) btrfs_quota_scan_start(fd); | |
902 | return 1; | |
903 | } | |
904 | } | |
905 | ||
906 | int btrfs_qgroup_assign(int fd, uint64_t child, uint64_t parent) { | |
907 | return qgroup_assign_or_unassign(fd, true, child, parent); | |
908 | } | |
909 | ||
910 | int btrfs_qgroup_unassign(int fd, uint64_t child, uint64_t parent) { | |
911 | return qgroup_assign_or_unassign(fd, false, child, parent); | |
912 | } | |
913 | ||
914 | static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol_id, BtrfsRemoveFlags flags) { | |
d9e2daaf LP |
915 | struct btrfs_ioctl_search_args args = { |
916 | .key.tree_id = BTRFS_ROOT_TREE_OBJECTID, | |
917 | ||
918 | .key.min_objectid = BTRFS_FIRST_FREE_OBJECTID, | |
919 | .key.max_objectid = BTRFS_LAST_FREE_OBJECTID, | |
920 | ||
921 | .key.min_type = BTRFS_ROOT_BACKREF_KEY, | |
922 | .key.max_type = BTRFS_ROOT_BACKREF_KEY, | |
923 | ||
924 | .key.min_transid = 0, | |
f5fbe71d | 925 | .key.max_transid = UINT64_MAX, |
d9e2daaf LP |
926 | }; |
927 | ||
928 | struct btrfs_ioctl_vol_args vol_args = {}; | |
254d1313 | 929 | _cleanup_close_ int subvol_fd = -EBADF; |
62572894 | 930 | struct stat st; |
3986b258 | 931 | bool made_writable = false; |
d9e2daaf LP |
932 | int r; |
933 | ||
934 | assert(fd >= 0); | |
935 | assert(subvolume); | |
936 | ||
62572894 LP |
937 | if (fstat(fd, &st) < 0) |
938 | return -errno; | |
939 | ||
940 | if (!S_ISDIR(st.st_mode)) | |
941 | return -EINVAL; | |
942 | ||
f7c9f4a2 | 943 | subvol_fd = openat(fd, subvolume, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW); |
d9e2daaf LP |
944 | if (subvol_fd < 0) |
945 | return -errno; | |
946 | ||
ae1940d2 LP |
947 | /* Let's check if this is actually a subvolume. Note that this is mostly redundant, as BTRFS_IOC_SNAP_DESTROY |
948 | * would fail anyway if it is not. However, it's a good thing to check this ahead of time so that we can return | |
949 | * ENOTTY unconditionally in this case. This is different from the ioctl() which will return EPERM/EACCES if we | |
950 | * don't have the privileges to remove subvolumes, regardless if the specified directory is actually a | |
951 | * subvolume or not. In order to make it easy for callers to cover the "this is not a btrfs subvolume" case | |
952 | * let's prefer ENOTTY over EPERM/EACCES though. */ | |
953 | r = btrfs_is_subvol_fd(subvol_fd); | |
954 | if (r < 0) | |
955 | return r; | |
956 | if (r == 0) /* Not a btrfs subvolume */ | |
957 | return -ENOTTY; | |
958 | ||
d9e2daaf LP |
959 | if (subvol_id == 0) { |
960 | r = btrfs_subvol_get_id_fd(subvol_fd, &subvol_id); | |
961 | if (r < 0) | |
962 | return r; | |
963 | } | |
964 | ||
3f952f92 LP |
965 | /* First, try to remove the subvolume. If it happens to be |
966 | * already empty, this will just work. */ | |
967 | strncpy(vol_args.name, subvolume, sizeof(vol_args.name)-1); | |
968 | if (ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &vol_args) >= 0) { | |
5bcd08db | 969 | (void) btrfs_qgroup_destroy_recursive(fd, subvol_id); /* for the leaf subvolumes, the qgroup id is identical to the subvol id */ |
3f952f92 LP |
970 | return 0; |
971 | } | |
5bcd08db | 972 | if (!(flags & BTRFS_REMOVE_RECURSIVE) || errno != ENOTEMPTY) |
3f952f92 LP |
973 | return -errno; |
974 | ||
975 | /* OK, the subvolume is not empty, let's look for child | |
976 | * subvolumes, and remove them, first */ | |
977 | ||
d9e2daaf LP |
978 | args.key.min_offset = args.key.max_offset = subvol_id; |
979 | ||
980 | while (btrfs_ioctl_search_args_compare(&args) <= 0) { | |
801bf40c LP |
981 | struct btrfs_ioctl_search_header sh; |
982 | const void *body; | |
d9e2daaf LP |
983 | |
984 | args.key.nr_items = 256; | |
985 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) | |
986 | return -errno; | |
987 | ||
988 | if (args.key.nr_items <= 0) | |
989 | break; | |
990 | ||
801bf40c | 991 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) { |
d9e2daaf | 992 | _cleanup_free_ char *p = NULL; |
d9e2daaf | 993 | |
801bf40c | 994 | btrfs_ioctl_search_args_set(&args, &sh); |
d9e2daaf | 995 | |
801bf40c | 996 | if (sh.type != BTRFS_ROOT_BACKREF_KEY) |
d9e2daaf | 997 | continue; |
801bf40c | 998 | if (sh.offset != subvol_id) |
d9e2daaf LP |
999 | continue; |
1000 | ||
801bf40c | 1001 | const struct btrfs_root_ref *ref = body; |
e5c41c61 | 1002 | p = memdup_suffix0((char*) ref + sizeof(struct btrfs_root_ref), le64toh(ref->name_len)); |
d9e2daaf LP |
1003 | if (!p) |
1004 | return -ENOMEM; | |
1005 | ||
41ab8c67 LP |
1006 | struct btrfs_ioctl_ino_lookup_args ino_args = { |
1007 | .treeid = subvol_id, | |
1008 | .objectid = htole64(ref->dirid), | |
1009 | }; | |
d9e2daaf LP |
1010 | |
1011 | if (ioctl(fd, BTRFS_IOC_INO_LOOKUP, &ino_args) < 0) | |
1012 | return -errno; | |
1013 | ||
3986b258 LP |
1014 | if (!made_writable) { |
1015 | r = btrfs_subvol_set_read_only_fd(subvol_fd, false); | |
1016 | if (r < 0) | |
1017 | return r; | |
1018 | ||
1019 | made_writable = true; | |
1020 | } | |
1021 | ||
d9e2daaf LP |
1022 | if (isempty(ino_args.name)) |
1023 | /* Subvolume is in the top-level | |
1024 | * directory of the subvolume. */ | |
801bf40c | 1025 | r = subvol_remove_children(subvol_fd, p, sh.objectid, flags); |
d9e2daaf | 1026 | else { |
254d1313 | 1027 | _cleanup_close_ int child_fd = -EBADF; |
d9e2daaf LP |
1028 | |
1029 | /* Subvolume is somewhere further down, | |
1030 | * hence we need to open the | |
1031 | * containing directory first */ | |
1032 | ||
f7c9f4a2 | 1033 | child_fd = openat(subvol_fd, ino_args.name, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW); |
d9e2daaf LP |
1034 | if (child_fd < 0) |
1035 | return -errno; | |
1036 | ||
801bf40c | 1037 | r = subvol_remove_children(child_fd, p, sh.objectid, flags); |
d9e2daaf LP |
1038 | } |
1039 | if (r < 0) | |
1040 | return r; | |
1041 | } | |
1042 | ||
1043 | /* Increase search key by one, to read the next item, if we can. */ | |
1044 | if (!btrfs_ioctl_search_args_inc(&args)) | |
1045 | break; | |
1046 | } | |
1047 | ||
1048 | /* OK, the child subvolumes should all be gone now, let's try | |
1049 | * again to remove the subvolume */ | |
1050 | if (ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &vol_args) < 0) | |
1051 | return -errno; | |
1052 | ||
5bcd08db | 1053 | (void) btrfs_qgroup_destroy_recursive(fd, subvol_id); |
d9e2daaf LP |
1054 | return 0; |
1055 | } | |
1056 | ||
24dbe603 | 1057 | int btrfs_subvol_remove_at(int dir_fd, const char *path, BtrfsRemoveFlags flags) { |
03469b77 | 1058 | _cleanup_free_ char *subvolume = NULL; |
254d1313 | 1059 | _cleanup_close_ int fd = -EBADF; |
d9e2daaf LP |
1060 | int r; |
1061 | ||
1062 | assert(path); | |
1063 | ||
24dbe603 | 1064 | fd = chase_and_openat(dir_fd, path, CHASE_PARENT|CHASE_EXTRACT_FILENAME, O_CLOEXEC, &subvolume); |
d9e2daaf LP |
1065 | if (fd < 0) |
1066 | return fd; | |
1067 | ||
e54c79cc | 1068 | r = btrfs_validate_subvolume_name(subvolume); |
24dbe603 DDM |
1069 | if (r < 0) |
1070 | return r; | |
5bcd08db | 1071 | |
5bcd08db LP |
1072 | return subvol_remove_children(fd, subvolume, 0, flags); |
1073 | } | |
1074 | ||
1075 | int btrfs_qgroup_copy_limits(int fd, uint64_t old_qgroupid, uint64_t new_qgroupid) { | |
1076 | ||
1077 | struct btrfs_ioctl_search_args args = { | |
1078 | /* Tree of quota items */ | |
1079 | .key.tree_id = BTRFS_QUOTA_TREE_OBJECTID, | |
1080 | ||
1081 | /* The object ID is always 0 */ | |
1082 | .key.min_objectid = 0, | |
1083 | .key.max_objectid = 0, | |
1084 | ||
1085 | /* Look precisely for the quota items */ | |
1086 | .key.min_type = BTRFS_QGROUP_LIMIT_KEY, | |
1087 | .key.max_type = BTRFS_QGROUP_LIMIT_KEY, | |
1088 | ||
1089 | /* For our qgroup */ | |
1090 | .key.min_offset = old_qgroupid, | |
1091 | .key.max_offset = old_qgroupid, | |
1092 | ||
1093 | /* No restrictions on the other components */ | |
1094 | .key.min_transid = 0, | |
f5fbe71d | 1095 | .key.max_transid = UINT64_MAX, |
5bcd08db LP |
1096 | }; |
1097 | ||
1098 | int r; | |
1099 | ||
65ddc2c5 | 1100 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
5bcd08db LP |
1101 | if (r < 0) |
1102 | return r; | |
79de6eb1 | 1103 | if (r == 0) |
5bcd08db LP |
1104 | return -ENOTTY; |
1105 | ||
1106 | while (btrfs_ioctl_search_args_compare(&args) <= 0) { | |
801bf40c LP |
1107 | struct btrfs_ioctl_search_header sh; |
1108 | const void *body; | |
5bcd08db LP |
1109 | |
1110 | args.key.nr_items = 256; | |
12ee6186 LP |
1111 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) { |
1112 | if (errno == ENOENT) /* quota tree missing: quota is not enabled, hence nothing to copy */ | |
1113 | break; | |
1114 | ||
5bcd08db | 1115 | return -errno; |
12ee6186 | 1116 | } |
5bcd08db LP |
1117 | |
1118 | if (args.key.nr_items <= 0) | |
1119 | break; | |
1120 | ||
801bf40c | 1121 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) { |
5bcd08db LP |
1122 | struct btrfs_ioctl_qgroup_limit_args qargs; |
1123 | unsigned c; | |
1124 | ||
1125 | /* Make sure we start the next search at least from this entry */ | |
801bf40c | 1126 | btrfs_ioctl_search_args_set(&args, &sh); |
5bcd08db | 1127 | |
801bf40c | 1128 | if (sh.objectid != 0) |
5bcd08db | 1129 | continue; |
801bf40c | 1130 | if (sh.type != BTRFS_QGROUP_LIMIT_KEY) |
5bcd08db | 1131 | continue; |
801bf40c | 1132 | if (sh.offset != old_qgroupid) |
5bcd08db LP |
1133 | continue; |
1134 | ||
1135 | /* We found the entry, now copy things over. */ | |
1136 | ||
801bf40c | 1137 | const struct btrfs_qgroup_limit_item *qli = body; |
5bcd08db LP |
1138 | qargs = (struct btrfs_ioctl_qgroup_limit_args) { |
1139 | .qgroupid = new_qgroupid, | |
1140 | ||
1141 | .lim.max_rfer = le64toh(qli->max_rfer), | |
1142 | .lim.max_excl = le64toh(qli->max_excl), | |
1143 | .lim.rsv_rfer = le64toh(qli->rsv_rfer), | |
1144 | .lim.rsv_excl = le64toh(qli->rsv_excl), | |
1145 | ||
1146 | .lim.flags = le64toh(qli->flags) & (BTRFS_QGROUP_LIMIT_MAX_RFER| | |
1147 | BTRFS_QGROUP_LIMIT_MAX_EXCL| | |
1148 | BTRFS_QGROUP_LIMIT_RSV_RFER| | |
1149 | BTRFS_QGROUP_LIMIT_RSV_EXCL), | |
1150 | }; | |
1151 | ||
1152 | for (c = 0;; c++) { | |
1153 | if (ioctl(fd, BTRFS_IOC_QGROUP_LIMIT, &qargs) < 0) { | |
1154 | if (errno == EBUSY && c < 10) { | |
1155 | (void) btrfs_quota_scan_wait(fd); | |
1156 | continue; | |
1157 | } | |
1158 | return -errno; | |
1159 | } | |
1160 | ||
1161 | break; | |
1162 | } | |
1163 | ||
1164 | return 1; | |
1165 | } | |
1166 | ||
1167 | /* Increase search key by one, to read the next item, if we can. */ | |
1168 | if (!btrfs_ioctl_search_args_inc(&args)) | |
1169 | break; | |
1170 | } | |
1171 | ||
1172 | return 0; | |
d9e2daaf LP |
1173 | } |
1174 | ||
5bcd08db LP |
1175 | static int copy_quota_hierarchy(int fd, uint64_t old_subvol_id, uint64_t new_subvol_id) { |
1176 | _cleanup_free_ uint64_t *old_qgroups = NULL, *old_parent_qgroups = NULL; | |
1177 | bool copy_from_parent = false, insert_intermediary_qgroup = false; | |
cecaba20 | 1178 | int n_old_qgroups, n_old_parent_qgroups, r; |
5bcd08db LP |
1179 | uint64_t old_parent_id; |
1180 | ||
1181 | assert(fd >= 0); | |
1182 | ||
1183 | /* Copies a reduced form of quota information from the old to | |
1184 | * the new subvolume. */ | |
1185 | ||
1186 | n_old_qgroups = btrfs_qgroup_find_parents(fd, old_subvol_id, &old_qgroups); | |
1187 | if (n_old_qgroups <= 0) /* Nothing to copy */ | |
1188 | return n_old_qgroups; | |
1189 | ||
1190 | r = btrfs_subvol_get_parent(fd, old_subvol_id, &old_parent_id); | |
08c77cf3 LP |
1191 | if (r == -ENXIO) |
1192 | /* We have no parent, hence nothing to copy. */ | |
1193 | n_old_parent_qgroups = 0; | |
1194 | else if (r < 0) | |
5bcd08db | 1195 | return r; |
08c77cf3 LP |
1196 | else { |
1197 | n_old_parent_qgroups = btrfs_qgroup_find_parents(fd, old_parent_id, &old_parent_qgroups); | |
1198 | if (n_old_parent_qgroups < 0) | |
1199 | return n_old_parent_qgroups; | |
1200 | } | |
5bcd08db | 1201 | |
cecaba20 | 1202 | for (int i = 0; i < n_old_qgroups; i++) { |
5bcd08db | 1203 | uint64_t id; |
5bcd08db LP |
1204 | |
1205 | r = btrfs_qgroupid_split(old_qgroups[i], NULL, &id); | |
1206 | if (r < 0) | |
1207 | return r; | |
1208 | ||
1209 | if (id == old_subvol_id) { | |
1210 | /* The old subvolume was member of a qgroup | |
1211 | * that had the same id, but a different level | |
1212 | * as it self. Let's set up something similar | |
1213 | * in the destination. */ | |
1214 | insert_intermediary_qgroup = true; | |
1215 | break; | |
1216 | } | |
1217 | ||
cecaba20 | 1218 | for (int j = 0; j < n_old_parent_qgroups; j++) |
d46b79bb | 1219 | if (old_parent_qgroups[j] == old_qgroups[i]) |
5bcd08db LP |
1220 | /* The old subvolume shared a common |
1221 | * parent qgroup with its parent | |
1222 | * subvolume. Let's set up something | |
1223 | * similar in the destination. */ | |
1224 | copy_from_parent = true; | |
5bcd08db LP |
1225 | } |
1226 | ||
1227 | if (!insert_intermediary_qgroup && !copy_from_parent) | |
1228 | return 0; | |
1229 | ||
1230 | return btrfs_subvol_auto_qgroup_fd(fd, new_subvol_id, insert_intermediary_qgroup); | |
1231 | } | |
1232 | ||
1233 | static int copy_subtree_quota_limits(int fd, uint64_t old_subvol, uint64_t new_subvol) { | |
1234 | uint64_t old_subtree_qgroup, new_subtree_qgroup; | |
1235 | bool changed; | |
1236 | int r; | |
1237 | ||
1238 | /* First copy the leaf limits */ | |
1239 | r = btrfs_qgroup_copy_limits(fd, old_subvol, new_subvol); | |
1240 | if (r < 0) | |
1241 | return r; | |
1242 | changed = r > 0; | |
1243 | ||
1244 | /* Then, try to copy the subtree limits, if there are any. */ | |
1245 | r = btrfs_subvol_find_subtree_qgroup(fd, old_subvol, &old_subtree_qgroup); | |
1246 | if (r < 0) | |
1247 | return r; | |
1248 | if (r == 0) | |
1249 | return changed; | |
1250 | ||
1251 | r = btrfs_subvol_find_subtree_qgroup(fd, new_subvol, &new_subtree_qgroup); | |
1252 | if (r < 0) | |
1253 | return r; | |
1254 | if (r == 0) | |
1255 | return changed; | |
1256 | ||
1257 | r = btrfs_qgroup_copy_limits(fd, old_subtree_qgroup, new_subtree_qgroup); | |
1258 | if (r != 0) | |
1259 | return r; | |
1260 | ||
1261 | return changed; | |
d9e2daaf | 1262 | } |
f70a17f8 | 1263 | |
b3cade0c LP |
1264 | static int subvol_snapshot_children( |
1265 | int old_fd, | |
1266 | int new_fd, | |
1267 | const char *subvolume, | |
1268 | uint64_t old_subvol_id, | |
1269 | BtrfsSnapshotFlags flags) { | |
f70a17f8 LP |
1270 | |
1271 | struct btrfs_ioctl_search_args args = { | |
1272 | .key.tree_id = BTRFS_ROOT_TREE_OBJECTID, | |
1273 | ||
1274 | .key.min_objectid = BTRFS_FIRST_FREE_OBJECTID, | |
1275 | .key.max_objectid = BTRFS_LAST_FREE_OBJECTID, | |
1276 | ||
1277 | .key.min_type = BTRFS_ROOT_BACKREF_KEY, | |
1278 | .key.max_type = BTRFS_ROOT_BACKREF_KEY, | |
1279 | ||
1280 | .key.min_transid = 0, | |
f5fbe71d | 1281 | .key.max_transid = UINT64_MAX, |
f70a17f8 LP |
1282 | }; |
1283 | ||
1284 | struct btrfs_ioctl_vol_args_v2 vol_args = { | |
1285 | .flags = flags & BTRFS_SNAPSHOT_READ_ONLY ? BTRFS_SUBVOL_RDONLY : 0, | |
1286 | .fd = old_fd, | |
1287 | }; | |
254d1313 | 1288 | _cleanup_close_ int subvolume_fd = -EBADF; |
90578cbd LP |
1289 | uint64_t new_subvol_id; |
1290 | int r; | |
f70a17f8 LP |
1291 | |
1292 | assert(old_fd >= 0); | |
1293 | assert(new_fd >= 0); | |
1294 | assert(subvolume); | |
1295 | ||
1296 | strncpy(vol_args.name, subvolume, sizeof(vol_args.name)-1); | |
f70a17f8 LP |
1297 | |
1298 | if (ioctl(new_fd, BTRFS_IOC_SNAP_CREATE_V2, &vol_args) < 0) | |
1299 | return -errno; | |
1300 | ||
d54f60c2 DDM |
1301 | if (FLAGS_SET(flags, BTRFS_SNAPSHOT_LOCK_BSD)) { |
1302 | subvolume_fd = xopenat_lock(new_fd, subvolume, | |
1303 | O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW, | |
d54f60c2 DDM |
1304 | LOCK_BSD, |
1305 | LOCK_EX); | |
1306 | if (subvolume_fd < 0) | |
1307 | return subvolume_fd; | |
1308 | ||
1309 | r = btrfs_is_subvol_fd(subvolume_fd); | |
1310 | if (r < 0) | |
1311 | return r; | |
1312 | if (r == 0) | |
1313 | return -EEXIST; | |
1314 | } | |
1315 | ||
5bcd08db LP |
1316 | if (!(flags & BTRFS_SNAPSHOT_RECURSIVE) && |
1317 | !(flags & BTRFS_SNAPSHOT_QUOTA)) | |
d54f60c2 | 1318 | return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0; |
f70a17f8 | 1319 | |
90578cbd LP |
1320 | if (old_subvol_id == 0) { |
1321 | r = btrfs_subvol_get_id_fd(old_fd, &old_subvol_id); | |
f70a17f8 LP |
1322 | if (r < 0) |
1323 | return r; | |
1324 | } | |
1325 | ||
90578cbd LP |
1326 | r = btrfs_subvol_get_id(new_fd, vol_args.name, &new_subvol_id); |
1327 | if (r < 0) | |
1328 | return r; | |
1329 | ||
5bcd08db LP |
1330 | if (flags & BTRFS_SNAPSHOT_QUOTA) |
1331 | (void) copy_quota_hierarchy(new_fd, old_subvol_id, new_subvol_id); | |
1332 | ||
1333 | if (!(flags & BTRFS_SNAPSHOT_RECURSIVE)) { | |
1334 | ||
1335 | if (flags & BTRFS_SNAPSHOT_QUOTA) | |
1336 | (void) copy_subtree_quota_limits(new_fd, old_subvol_id, new_subvol_id); | |
1337 | ||
d54f60c2 | 1338 | return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0; |
5bcd08db LP |
1339 | } |
1340 | ||
90578cbd | 1341 | args.key.min_offset = args.key.max_offset = old_subvol_id; |
f70a17f8 LP |
1342 | |
1343 | while (btrfs_ioctl_search_args_compare(&args) <= 0) { | |
801bf40c LP |
1344 | struct btrfs_ioctl_search_header sh; |
1345 | const void *body; | |
f70a17f8 LP |
1346 | |
1347 | args.key.nr_items = 256; | |
1348 | if (ioctl(old_fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) | |
1349 | return -errno; | |
1350 | ||
1351 | if (args.key.nr_items <= 0) | |
1352 | break; | |
1353 | ||
801bf40c | 1354 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) { |
f70a17f8 | 1355 | _cleanup_free_ char *p = NULL, *c = NULL, *np = NULL; |
254d1313 | 1356 | _cleanup_close_ int old_child_fd = -EBADF, new_child_fd = -EBADF; |
f70a17f8 | 1357 | |
801bf40c | 1358 | btrfs_ioctl_search_args_set(&args, &sh); |
f70a17f8 | 1359 | |
801bf40c | 1360 | if (sh.type != BTRFS_ROOT_BACKREF_KEY) |
f70a17f8 | 1361 | continue; |
90578cbd | 1362 | |
801bf40c LP |
1363 | /* Avoid finding the source subvolume a second time */ |
1364 | if (sh.offset != old_subvol_id) | |
f70a17f8 LP |
1365 | continue; |
1366 | ||
801bf40c LP |
1367 | /* Avoid running into loops if the new subvolume is below the old one. */ |
1368 | if (sh.objectid == new_subvol_id) | |
90578cbd | 1369 | continue; |
f70a17f8 | 1370 | |
801bf40c | 1371 | const struct btrfs_root_ref *ref = body; |
e5c41c61 | 1372 | p = memdup_suffix0((char*) ref + sizeof(struct btrfs_root_ref), le64toh(ref->name_len)); |
f70a17f8 LP |
1373 | if (!p) |
1374 | return -ENOMEM; | |
1375 | ||
41ab8c67 LP |
1376 | struct btrfs_ioctl_ino_lookup_args ino_args = { |
1377 | .treeid = old_subvol_id, | |
1378 | .objectid = htole64(ref->dirid), | |
1379 | }; | |
f70a17f8 LP |
1380 | |
1381 | if (ioctl(old_fd, BTRFS_IOC_INO_LOOKUP, &ino_args) < 0) | |
1382 | return -errno; | |
1383 | ||
b910cc72 | 1384 | c = path_join(ino_args.name, p); |
f70a17f8 LP |
1385 | if (!c) |
1386 | return -ENOMEM; | |
1387 | ||
f7c9f4a2 | 1388 | old_child_fd = openat(old_fd, c, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW); |
f70a17f8 LP |
1389 | if (old_child_fd < 0) |
1390 | return -errno; | |
1391 | ||
657ee2d8 | 1392 | np = path_join(subvolume, ino_args.name); |
f70a17f8 LP |
1393 | if (!np) |
1394 | return -ENOMEM; | |
1395 | ||
f7c9f4a2 | 1396 | new_child_fd = openat(new_fd, np, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW); |
f70a17f8 LP |
1397 | if (new_child_fd < 0) |
1398 | return -errno; | |
1399 | ||
ffb296b2 | 1400 | if (flags & BTRFS_SNAPSHOT_READ_ONLY) { |
801bf40c LP |
1401 | /* If the snapshot is read-only we need to mark it writable temporarily, to |
1402 | * put the subsnapshot into place. */ | |
ffb296b2 LP |
1403 | |
1404 | if (subvolume_fd < 0) { | |
f7c9f4a2 | 1405 | subvolume_fd = openat(new_fd, subvolume, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW); |
ffb296b2 LP |
1406 | if (subvolume_fd < 0) |
1407 | return -errno; | |
1408 | } | |
1409 | ||
1410 | r = btrfs_subvol_set_read_only_fd(subvolume_fd, false); | |
1411 | if (r < 0) | |
1412 | return r; | |
1413 | } | |
1414 | ||
801bf40c LP |
1415 | /* When btrfs clones the subvolumes, child subvolumes appear as empty |
1416 | * directories. Remove them, so that we can create a new snapshot in their place */ | |
ffb296b2 LP |
1417 | if (unlinkat(new_child_fd, p, AT_REMOVEDIR) < 0) { |
1418 | int k = -errno; | |
1419 | ||
1420 | if (flags & BTRFS_SNAPSHOT_READ_ONLY) | |
1421 | (void) btrfs_subvol_set_read_only_fd(subvolume_fd, true); | |
1422 | ||
1423 | return k; | |
1424 | } | |
f70a17f8 | 1425 | |
801bf40c | 1426 | r = subvol_snapshot_children(old_child_fd, new_child_fd, p, sh.objectid, |
d54f60c2 | 1427 | flags & ~(BTRFS_SNAPSHOT_FALLBACK_COPY|BTRFS_SNAPSHOT_LOCK_BSD)); |
ffb296b2 LP |
1428 | |
1429 | /* Restore the readonly flag */ | |
1430 | if (flags & BTRFS_SNAPSHOT_READ_ONLY) { | |
1431 | int k; | |
1432 | ||
1433 | k = btrfs_subvol_set_read_only_fd(subvolume_fd, true); | |
1434 | if (r >= 0 && k < 0) | |
1435 | return k; | |
1436 | } | |
1437 | ||
f70a17f8 LP |
1438 | if (r < 0) |
1439 | return r; | |
1440 | } | |
1441 | ||
1442 | /* Increase search key by one, to read the next item, if we can. */ | |
1443 | if (!btrfs_ioctl_search_args_inc(&args)) | |
1444 | break; | |
1445 | } | |
1446 | ||
5bcd08db LP |
1447 | if (flags & BTRFS_SNAPSHOT_QUOTA) |
1448 | (void) copy_subtree_quota_limits(new_fd, old_subvol_id, new_subvol_id); | |
1449 | ||
d54f60c2 | 1450 | return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0; |
f70a17f8 LP |
1451 | } |
1452 | ||
fab4ef72 DDM |
1453 | int btrfs_subvol_snapshot_at_full( |
1454 | int dir_fdf, | |
1455 | const char *from, | |
1456 | int dir_fdt, | |
1457 | const char *to, | |
b3cade0c LP |
1458 | BtrfsSnapshotFlags flags, |
1459 | copy_progress_path_t progress_path, | |
1460 | copy_progress_bytes_t progress_bytes, | |
1461 | void *userdata) { | |
1462 | ||
03469b77 | 1463 | _cleanup_free_ char *subvolume = NULL; |
d54f60c2 | 1464 | _cleanup_close_ int old_fd = -EBADF, new_fd = -EBADF, subvolume_fd = -EBADF; |
f70a17f8 LP |
1465 | int r; |
1466 | ||
fab4ef72 DDM |
1467 | assert(dir_fdf >= 0 || dir_fdf == AT_FDCWD); |
1468 | assert(dir_fdt >= 0 || dir_fdt == AT_FDCWD); | |
1469 | assert(to); | |
1470 | ||
e40b11be | 1471 | old_fd = xopenat(dir_fdf, from, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY); |
fab4ef72 DDM |
1472 | if (old_fd < 0) |
1473 | return old_fd; | |
f70a17f8 | 1474 | |
fab4ef72 DDM |
1475 | new_fd = chase_and_openat(dir_fdt, to, CHASE_PARENT|CHASE_EXTRACT_FILENAME, O_CLOEXEC, &subvolume); |
1476 | if (new_fd < 0) | |
1477 | return new_fd; | |
1478 | ||
e54c79cc | 1479 | r = btrfs_validate_subvolume_name(subvolume); |
fab4ef72 DDM |
1480 | if (r < 0) |
1481 | return r; | |
1482 | ||
1483 | r = btrfs_is_subvol_at(dir_fdf, from); | |
f70a17f8 LP |
1484 | if (r < 0) |
1485 | return r; | |
1486 | if (r == 0) { | |
17cbb288 LP |
1487 | bool plain_directory = false; |
1488 | ||
1489 | /* If the source isn't a proper subvolume, fail unless fallback is requested */ | |
f70a17f8 LP |
1490 | if (!(flags & BTRFS_SNAPSHOT_FALLBACK_COPY)) |
1491 | return -EISDIR; | |
1492 | ||
e54c79cc | 1493 | r = btrfs_subvol_make(new_fd, subvolume); |
08b8e913 DL |
1494 | if (r < 0) { |
1495 | if (ERRNO_IS_NOT_SUPPORTED(r) && (flags & BTRFS_SNAPSHOT_FALLBACK_DIRECTORY)) { | |
1496 | /* If the destination doesn't support subvolumes, then use a plain directory, if that's requested. */ | |
1497 | if (mkdirat(new_fd, subvolume, 0755) < 0) | |
1498 | return -errno; | |
17cbb288 | 1499 | |
08b8e913 DL |
1500 | plain_directory = true; |
1501 | } else | |
1502 | return r; | |
1503 | } | |
f70a17f8 | 1504 | |
d54f60c2 DDM |
1505 | if (FLAGS_SET(flags, BTRFS_SNAPSHOT_LOCK_BSD)) { |
1506 | subvolume_fd = xopenat_lock(new_fd, subvolume, | |
1507 | O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW, | |
d54f60c2 DDM |
1508 | LOCK_BSD, |
1509 | LOCK_EX); | |
1510 | if (subvolume_fd < 0) | |
1511 | return subvolume_fd; | |
1512 | ||
1513 | if (!plain_directory) { | |
1514 | r = btrfs_is_subvol_fd(subvolume_fd); | |
1515 | if (r < 0) | |
1516 | return r; | |
1517 | if (r == 0) | |
1518 | return -EEXIST; | |
1519 | } | |
1520 | } | |
1521 | ||
f9f70e06 | 1522 | r = copy_directory_at_full( |
fab4ef72 DDM |
1523 | dir_fdf, from, |
1524 | new_fd, subvolume, | |
28ba7e36 LP |
1525 | COPY_MERGE_EMPTY| |
1526 | COPY_REFLINK| | |
1527 | COPY_SAME_MOUNT| | |
1528 | COPY_HARDLINKS| | |
23e026de | 1529 | COPY_ALL_XATTRS| |
28ba7e36 LP |
1530 | (FLAGS_SET(flags, BTRFS_SNAPSHOT_SIGINT) ? COPY_SIGINT : 0)| |
1531 | (FLAGS_SET(flags, BTRFS_SNAPSHOT_SIGTERM) ? COPY_SIGTERM : 0), | |
1532 | progress_path, | |
1533 | progress_bytes, | |
1534 | userdata); | |
17cbb288 LP |
1535 | if (r < 0) |
1536 | goto fallback_fail; | |
f70a17f8 LP |
1537 | |
1538 | if (flags & BTRFS_SNAPSHOT_READ_ONLY) { | |
17cbb288 LP |
1539 | |
1540 | if (plain_directory) { | |
1541 | /* Plain directories have no recursive read-only flag, but something pretty close to | |
1542 | * it: the IMMUTABLE bit. Let's use this here, if this is requested. */ | |
1543 | ||
1544 | if (flags & BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE) | |
fab4ef72 | 1545 | (void) chattr_at(new_fd, subvolume, FS_IMMUTABLE_FL, FS_IMMUTABLE_FL, NULL); |
17cbb288 | 1546 | } else { |
fab4ef72 | 1547 | r = btrfs_subvol_set_read_only_at(new_fd, subvolume, true); |
17cbb288 LP |
1548 | if (r < 0) |
1549 | goto fallback_fail; | |
f70a17f8 LP |
1550 | } |
1551 | } | |
1552 | ||
d54f60c2 | 1553 | return flags & BTRFS_SNAPSHOT_LOCK_BSD ? TAKE_FD(subvolume_fd) : 0; |
17cbb288 LP |
1554 | |
1555 | fallback_fail: | |
fab4ef72 | 1556 | (void) rm_rf_at(new_fd, subvolume, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME); |
17cbb288 | 1557 | return r; |
f70a17f8 LP |
1558 | } |
1559 | ||
f70a17f8 LP |
1560 | return subvol_snapshot_children(old_fd, new_fd, subvolume, 0, flags); |
1561 | } | |
1562 | ||
5bcd08db LP |
1563 | int btrfs_qgroup_find_parents(int fd, uint64_t qgroupid, uint64_t **ret) { |
1564 | ||
1565 | struct btrfs_ioctl_search_args args = { | |
1566 | /* Tree of quota items */ | |
1567 | .key.tree_id = BTRFS_QUOTA_TREE_OBJECTID, | |
1568 | ||
1569 | /* Look precisely for the quota relation items */ | |
1570 | .key.min_type = BTRFS_QGROUP_RELATION_KEY, | |
1571 | .key.max_type = BTRFS_QGROUP_RELATION_KEY, | |
1572 | ||
1573 | /* No restrictions on the other components */ | |
1574 | .key.min_offset = 0, | |
f5fbe71d | 1575 | .key.max_offset = UINT64_MAX, |
5bcd08db LP |
1576 | |
1577 | .key.min_transid = 0, | |
f5fbe71d | 1578 | .key.max_transid = UINT64_MAX, |
5bcd08db LP |
1579 | }; |
1580 | ||
1581 | _cleanup_free_ uint64_t *items = NULL; | |
319a4f4b | 1582 | size_t n_items = 0; |
5bcd08db LP |
1583 | int r; |
1584 | ||
1585 | assert(fd >= 0); | |
1586 | assert(ret); | |
1587 | ||
1588 | if (qgroupid == 0) { | |
1589 | r = btrfs_subvol_get_id_fd(fd, &qgroupid); | |
1590 | if (r < 0) | |
1591 | return r; | |
1592 | } else { | |
65ddc2c5 | 1593 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
5bcd08db LP |
1594 | if (r < 0) |
1595 | return r; | |
79de6eb1 | 1596 | if (r == 0) |
5bcd08db LP |
1597 | return -ENOTTY; |
1598 | } | |
1599 | ||
1600 | args.key.min_objectid = args.key.max_objectid = qgroupid; | |
1601 | ||
1602 | while (btrfs_ioctl_search_args_compare(&args) <= 0) { | |
801bf40c LP |
1603 | struct btrfs_ioctl_search_header sh; |
1604 | _unused_ const void *body; | |
5bcd08db LP |
1605 | |
1606 | args.key.nr_items = 256; | |
12ee6186 LP |
1607 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) { |
1608 | if (errno == ENOENT) /* quota tree missing: quota is disabled */ | |
1609 | break; | |
1610 | ||
5bcd08db | 1611 | return -errno; |
12ee6186 | 1612 | } |
5bcd08db LP |
1613 | |
1614 | if (args.key.nr_items <= 0) | |
1615 | break; | |
1616 | ||
801bf40c | 1617 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) { |
5bcd08db LP |
1618 | |
1619 | /* Make sure we start the next search at least from this entry */ | |
801bf40c | 1620 | btrfs_ioctl_search_args_set(&args, &sh); |
5bcd08db | 1621 | |
801bf40c | 1622 | if (sh.type != BTRFS_QGROUP_RELATION_KEY) |
5bcd08db | 1623 | continue; |
801bf40c | 1624 | if (sh.offset < sh.objectid) |
5bcd08db | 1625 | continue; |
801bf40c | 1626 | if (sh.objectid != qgroupid) |
5bcd08db LP |
1627 | continue; |
1628 | ||
319a4f4b | 1629 | if (!GREEDY_REALLOC(items, n_items+1)) |
5bcd08db LP |
1630 | return -ENOMEM; |
1631 | ||
801bf40c | 1632 | items[n_items++] = sh.offset; |
5bcd08db LP |
1633 | } |
1634 | ||
1635 | /* Increase search key by one, to read the next item, if we can. */ | |
1636 | if (!btrfs_ioctl_search_args_inc(&args)) | |
1637 | break; | |
1638 | } | |
1639 | ||
1640 | if (n_items <= 0) { | |
1641 | *ret = NULL; | |
1642 | return 0; | |
1643 | } | |
1644 | ||
1cc6c93a | 1645 | *ret = TAKE_PTR(items); |
5bcd08db LP |
1646 | |
1647 | return (int) n_items; | |
1648 | } | |
1649 | ||
1650 | int btrfs_subvol_auto_qgroup_fd(int fd, uint64_t subvol_id, bool insert_intermediary_qgroup) { | |
1651 | _cleanup_free_ uint64_t *qgroups = NULL; | |
254d1313 | 1652 | _cleanup_close_ int real_fd = -EBADF; |
5bcd08db LP |
1653 | uint64_t parent_subvol; |
1654 | bool changed = false; | |
1655 | int n = 0, r; | |
1656 | ||
1657 | assert(fd >= 0); | |
1658 | ||
1659 | /* | |
1660 | * Sets up the specified subvolume's qgroup automatically in | |
1661 | * one of two ways: | |
1662 | * | |
1663 | * If insert_intermediary_qgroup is false, the subvolume's | |
1664 | * leaf qgroup will be assigned to the same parent qgroups as | |
1665 | * the subvolume's parent subvolume. | |
1666 | * | |
1667 | * If insert_intermediary_qgroup is true a new intermediary | |
1668 | * higher-level qgroup is created, with a higher level number, | |
1669 | * but reusing the id of the subvolume. The level number is | |
1670 | * picked as one smaller than the lowest level qgroup the | |
1671 | * parent subvolume is a member of. If the parent subvolume's | |
1672 | * leaf qgroup is assigned to no higher-level qgroup a new | |
1673 | * qgroup of level 255 is created instead. Either way, the new | |
1674 | * qgroup is then assigned to the parent's higher-level | |
1675 | * qgroup, and the subvolume itself is assigned to it. | |
1676 | * | |
1677 | * If the subvolume is already assigned to a higher level | |
1678 | * qgroup, no operation is executed. | |
1679 | * | |
1680 | * Effectively this means: regardless if | |
1681 | * insert_intermediary_qgroup is true or not, after this | |
1682 | * function is invoked the subvolume will be accounted within | |
1683 | * the same qgroups as the parent. However, if it is true, it | |
1684 | * will also get its own higher-level qgroup, which may in | |
1685 | * turn be used by subvolumes created beneath this subvolume | |
1686 | * later on. | |
1687 | * | |
1688 | * This hence defines a simple default qgroup setup for | |
1689 | * subvolumes, as long as this function is invoked on each | |
1690 | * created subvolume: each subvolume is always accounting | |
1691 | * together with its immediate parents. Optionally, if | |
1692 | * insert_intermediary_qgroup is true, it will also get a | |
1693 | * qgroup that then includes all its own child subvolumes. | |
1694 | */ | |
1695 | ||
e6d1d4c0 LP |
1696 | /* Turn this into a proper fd, if it is currently O_PATH */ |
1697 | fd = fd_reopen_condition(fd, O_RDONLY|O_CLOEXEC, O_PATH, &real_fd); | |
1698 | if (fd < 0) | |
1699 | return fd; | |
1700 | ||
5bcd08db | 1701 | if (subvol_id == 0) { |
2904e949 | 1702 | r = btrfs_is_subvol_fd(fd); |
5bcd08db LP |
1703 | if (r < 0) |
1704 | return r; | |
1705 | if (!r) | |
1706 | return -ENOTTY; | |
1707 | ||
1708 | r = btrfs_subvol_get_id_fd(fd, &subvol_id); | |
1709 | if (r < 0) | |
1710 | return r; | |
1711 | } | |
1712 | ||
1713 | n = btrfs_qgroup_find_parents(fd, subvol_id, &qgroups); | |
1714 | if (n < 0) | |
1715 | return n; | |
1716 | if (n > 0) /* already parent qgroups set up, let's bail */ | |
1717 | return 0; | |
1718 | ||
08c77cf3 LP |
1719 | qgroups = mfree(qgroups); |
1720 | ||
5bcd08db | 1721 | r = btrfs_subvol_get_parent(fd, subvol_id, &parent_subvol); |
08c77cf3 LP |
1722 | if (r == -ENXIO) |
1723 | /* No parent, hence no qgroup memberships */ | |
1724 | n = 0; | |
1725 | else if (r < 0) | |
5bcd08db | 1726 | return r; |
08c77cf3 LP |
1727 | else { |
1728 | n = btrfs_qgroup_find_parents(fd, parent_subvol, &qgroups); | |
1729 | if (n < 0) | |
1730 | return n; | |
1731 | } | |
5bcd08db LP |
1732 | |
1733 | if (insert_intermediary_qgroup) { | |
1734 | uint64_t lowest = 256, new_qgroupid; | |
1735 | bool created = false; | |
5bcd08db LP |
1736 | |
1737 | /* Determine the lowest qgroup that the parent | |
1738 | * subvolume is assigned to. */ | |
1739 | ||
cecaba20 | 1740 | for (int i = 0; i < n; i++) { |
5bcd08db LP |
1741 | uint64_t level; |
1742 | ||
1743 | r = btrfs_qgroupid_split(qgroups[i], &level, NULL); | |
1744 | if (r < 0) | |
1745 | return r; | |
1746 | ||
1747 | if (level < lowest) | |
1748 | lowest = level; | |
1749 | } | |
1750 | ||
1751 | if (lowest <= 1) /* There are no levels left we could use insert an intermediary qgroup at */ | |
1752 | return -EBUSY; | |
1753 | ||
1754 | r = btrfs_qgroupid_make(lowest - 1, subvol_id, &new_qgroupid); | |
1755 | if (r < 0) | |
1756 | return r; | |
1757 | ||
1758 | /* Create the new intermediary group, unless it already exists */ | |
1759 | r = btrfs_qgroup_create(fd, new_qgroupid); | |
1760 | if (r < 0 && r != -EEXIST) | |
1761 | return r; | |
1762 | if (r >= 0) | |
1763 | changed = created = true; | |
1764 | ||
cecaba20 | 1765 | for (int i = 0; i < n; i++) { |
5bcd08db LP |
1766 | r = btrfs_qgroup_assign(fd, new_qgroupid, qgroups[i]); |
1767 | if (r < 0 && r != -EEXIST) { | |
1768 | if (created) | |
1769 | (void) btrfs_qgroup_destroy_recursive(fd, new_qgroupid); | |
1770 | ||
1771 | return r; | |
1772 | } | |
1773 | if (r >= 0) | |
1774 | changed = true; | |
1775 | } | |
1776 | ||
1777 | r = btrfs_qgroup_assign(fd, subvol_id, new_qgroupid); | |
1778 | if (r < 0 && r != -EEXIST) { | |
1779 | if (created) | |
1780 | (void) btrfs_qgroup_destroy_recursive(fd, new_qgroupid); | |
1781 | return r; | |
1782 | } | |
1783 | if (r >= 0) | |
1784 | changed = true; | |
1785 | ||
1786 | } else { | |
1787 | int i; | |
1788 | ||
1789 | /* Assign our subvolume to all the same qgroups as the parent */ | |
1790 | ||
1791 | for (i = 0; i < n; i++) { | |
1792 | r = btrfs_qgroup_assign(fd, subvol_id, qgroups[i]); | |
1793 | if (r < 0 && r != -EEXIST) | |
1794 | return r; | |
1795 | if (r >= 0) | |
1796 | changed = true; | |
1797 | } | |
1798 | } | |
1799 | ||
1800 | return changed; | |
1801 | } | |
1802 | ||
1803 | int btrfs_subvol_auto_qgroup(const char *path, uint64_t subvol_id, bool create_intermediary_qgroup) { | |
254d1313 | 1804 | _cleanup_close_ int fd = -EBADF; |
5bcd08db LP |
1805 | |
1806 | fd = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY); | |
1807 | if (fd < 0) | |
1808 | return -errno; | |
1809 | ||
1810 | return btrfs_subvol_auto_qgroup_fd(fd, subvol_id, create_intermediary_qgroup); | |
1811 | } | |
1812 | ||
3799fa80 DDM |
1813 | int btrfs_subvol_make_default(const char *path) { |
1814 | _cleanup_close_ int fd = -EBADF; | |
1815 | uint64_t id; | |
1816 | int r; | |
1817 | ||
1818 | assert(path); | |
1819 | ||
1820 | fd = open(path, O_NOCTTY|O_CLOEXEC|O_DIRECTORY); | |
1821 | if (fd < 0) | |
1822 | return -errno; | |
1823 | ||
1824 | r = btrfs_subvol_get_id_fd(fd, &id); | |
1825 | if (r < 0) | |
1826 | return r; | |
1827 | ||
1828 | return RET_NERRNO(ioctl(fd, BTRFS_IOC_DEFAULT_SUBVOL, &id)); | |
1829 | } | |
1830 | ||
5bcd08db LP |
1831 | int btrfs_subvol_get_parent(int fd, uint64_t subvol_id, uint64_t *ret) { |
1832 | ||
1833 | struct btrfs_ioctl_search_args args = { | |
1834 | /* Tree of tree roots */ | |
1835 | .key.tree_id = BTRFS_ROOT_TREE_OBJECTID, | |
1836 | ||
1837 | /* Look precisely for the subvolume items */ | |
1838 | .key.min_type = BTRFS_ROOT_BACKREF_KEY, | |
1839 | .key.max_type = BTRFS_ROOT_BACKREF_KEY, | |
1840 | ||
1841 | /* No restrictions on the other components */ | |
1842 | .key.min_offset = 0, | |
f5fbe71d | 1843 | .key.max_offset = UINT64_MAX, |
5bcd08db LP |
1844 | |
1845 | .key.min_transid = 0, | |
f5fbe71d | 1846 | .key.max_transid = UINT64_MAX, |
5bcd08db LP |
1847 | }; |
1848 | int r; | |
1849 | ||
1850 | assert(fd >= 0); | |
1851 | assert(ret); | |
1852 | ||
1853 | if (subvol_id == 0) { | |
1854 | r = btrfs_subvol_get_id_fd(fd, &subvol_id); | |
1855 | if (r < 0) | |
1856 | return r; | |
1857 | } else { | |
65ddc2c5 | 1858 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
5bcd08db LP |
1859 | if (r < 0) |
1860 | return r; | |
79de6eb1 | 1861 | if (r == 0) |
5bcd08db LP |
1862 | return -ENOTTY; |
1863 | } | |
1864 | ||
1865 | args.key.min_objectid = args.key.max_objectid = subvol_id; | |
1866 | ||
1867 | while (btrfs_ioctl_search_args_compare(&args) <= 0) { | |
801bf40c LP |
1868 | struct btrfs_ioctl_search_header sh; |
1869 | _unused_ const void *body = NULL; | |
5bcd08db LP |
1870 | |
1871 | args.key.nr_items = 256; | |
1872 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0) | |
9c4615fb | 1873 | return negative_errno(); |
5bcd08db LP |
1874 | |
1875 | if (args.key.nr_items <= 0) | |
1876 | break; | |
1877 | ||
801bf40c | 1878 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, args) { |
5bcd08db | 1879 | |
801bf40c | 1880 | if (sh.type != BTRFS_ROOT_BACKREF_KEY) |
5bcd08db | 1881 | continue; |
801bf40c | 1882 | if (sh.objectid != subvol_id) |
5bcd08db LP |
1883 | continue; |
1884 | ||
801bf40c | 1885 | *ret = sh.offset; |
5bcd08db LP |
1886 | return 0; |
1887 | } | |
1888 | } | |
1889 | ||
1890 | return -ENXIO; | |
1891 | } | |
5228c58e DDM |
1892 | |
1893 | int btrfs_forget_device(const char *path) { | |
1894 | _cleanup_close_ int control_fd = -EBADF; | |
1895 | struct btrfs_ioctl_vol_args args = {}; | |
1896 | ||
1897 | assert(path); | |
1898 | ||
1899 | if (strlen(path) > BTRFS_PATH_NAME_MAX) | |
1900 | return -E2BIG; | |
1901 | ||
1902 | strcpy(args.name, path); | |
1903 | ||
1904 | control_fd = open("/dev/btrfs-control", O_RDWR|O_CLOEXEC); | |
1905 | if (control_fd < 0) | |
1906 | return -errno; | |
1907 | ||
1908 | return RET_NERRNO(ioctl(control_fd, BTRFS_IOC_FORGET_DEV, &args)); | |
1909 | } | |
efb6a76a MY |
1910 | |
1911 | typedef struct BtrfsStripe { | |
1912 | uint64_t devid; | |
1913 | uint64_t offset; | |
1914 | } BtrfsStripe; | |
1915 | ||
1916 | typedef struct BtrfsChunk { | |
1917 | uint64_t offset; | |
1918 | uint64_t length; | |
1919 | uint64_t type; | |
1920 | ||
1921 | BtrfsStripe *stripes; | |
1922 | uint16_t n_stripes; | |
1923 | uint64_t stripe_len; | |
1924 | } BtrfsChunk; | |
1925 | ||
1926 | typedef struct BtrfsChunkTree { | |
1927 | BtrfsChunk **chunks; | |
1928 | size_t n_chunks; | |
1929 | } BtrfsChunkTree; | |
1930 | ||
1931 | static BtrfsChunk* btrfs_chunk_free(BtrfsChunk *chunk) { | |
1932 | if (!chunk) | |
1933 | return NULL; | |
1934 | ||
1935 | free(chunk->stripes); | |
1936 | ||
1937 | return mfree(chunk); | |
1938 | } | |
1939 | ||
1940 | DEFINE_TRIVIAL_CLEANUP_FUNC(BtrfsChunk*, btrfs_chunk_free); | |
1941 | ||
1942 | static void btrfs_chunk_tree_done(BtrfsChunkTree *tree) { | |
1943 | assert(tree); | |
1944 | ||
1945 | FOREACH_ARRAY(i, tree->chunks, tree->n_chunks) | |
1946 | btrfs_chunk_free(*i); | |
e504e465 MY |
1947 | |
1948 | free(tree->chunks); | |
efb6a76a MY |
1949 | } |
1950 | ||
1951 | static int btrfs_read_chunk_tree_fd(int fd, BtrfsChunkTree *ret) { | |
1952 | ||
1953 | struct btrfs_ioctl_search_args search_args = { | |
1954 | .key.tree_id = BTRFS_CHUNK_TREE_OBJECTID, | |
1955 | ||
1956 | .key.min_type = BTRFS_CHUNK_ITEM_KEY, | |
1957 | .key.max_type = BTRFS_CHUNK_ITEM_KEY, | |
1958 | ||
1959 | .key.min_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
1960 | .key.max_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
1961 | ||
1962 | .key.min_offset = 0, | |
1963 | .key.max_offset = UINT64_MAX, | |
1964 | ||
1965 | .key.min_transid = 0, | |
1966 | .key.max_transid = UINT64_MAX, | |
1967 | }; | |
1968 | ||
1969 | _cleanup_(btrfs_chunk_tree_done) BtrfsChunkTree tree = {}; | |
1970 | ||
1971 | assert(fd >= 0); | |
1972 | assert(ret); | |
1973 | ||
1974 | while (btrfs_ioctl_search_args_compare(&search_args) <= 0) { | |
801bf40c LP |
1975 | struct btrfs_ioctl_search_header sh; |
1976 | const void *body; | |
efb6a76a MY |
1977 | |
1978 | search_args.key.nr_items = 256; | |
1979 | ||
1980 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &search_args) < 0) | |
1981 | return -errno; | |
1982 | ||
1983 | if (search_args.key.nr_items == 0) | |
1984 | break; | |
1985 | ||
801bf40c | 1986 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, search_args) { |
efb6a76a | 1987 | _cleanup_(btrfs_chunk_freep) BtrfsChunk *chunk = NULL; |
efb6a76a | 1988 | |
801bf40c | 1989 | btrfs_ioctl_search_args_set(&search_args, &sh); |
efb6a76a | 1990 | |
801bf40c | 1991 | if (sh.objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID) |
efb6a76a | 1992 | continue; |
801bf40c | 1993 | if (sh.type != BTRFS_CHUNK_ITEM_KEY) |
efb6a76a MY |
1994 | continue; |
1995 | ||
1996 | chunk = new(BtrfsChunk, 1); | |
1997 | if (!chunk) | |
1998 | return -ENOMEM; | |
1999 | ||
801bf40c | 2000 | const struct btrfs_chunk *item = body; |
efb6a76a | 2001 | *chunk = (BtrfsChunk) { |
801bf40c | 2002 | .offset = sh.offset, |
efb6a76a MY |
2003 | .length = le64toh(item->length), |
2004 | .type = le64toh(item->type), | |
2005 | .n_stripes = le16toh(item->num_stripes), | |
2006 | .stripe_len = le64toh(item->stripe_len), | |
2007 | }; | |
2008 | ||
2009 | chunk->stripes = new(BtrfsStripe, chunk->n_stripes); | |
2010 | if (!chunk->stripes) | |
2011 | return -ENOMEM; | |
2012 | ||
2013 | for (size_t j = 0; j < chunk->n_stripes; j++) { | |
2014 | const struct btrfs_stripe *stripe = &item->stripe + j; | |
2015 | ||
2016 | chunk->stripes[j] = (BtrfsStripe) { | |
2017 | .devid = le64toh(stripe->devid), | |
2018 | .offset = le64toh(stripe->offset), | |
2019 | }; | |
2020 | } | |
2021 | ||
2022 | if (!GREEDY_REALLOC(tree.chunks, tree.n_chunks + 1)) | |
2023 | return -ENOMEM; | |
2024 | ||
2025 | tree.chunks[tree.n_chunks++] = TAKE_PTR(chunk); | |
2026 | } | |
2027 | ||
2028 | if (!btrfs_ioctl_search_args_inc(&search_args)) | |
2029 | break; | |
2030 | } | |
2031 | ||
2032 | *ret = TAKE_STRUCT(tree); | |
2033 | return 0; | |
2034 | } | |
2035 | ||
2036 | static BtrfsChunk* btrfs_find_chunk_from_logical_address(const BtrfsChunkTree *tree, uint64_t logical) { | |
2037 | size_t min_index, max_index; | |
2038 | ||
2039 | assert(tree); | |
2040 | assert(tree->chunks || tree->n_chunks == 0); | |
2041 | ||
2042 | if (tree->n_chunks == 0) | |
2043 | return NULL; | |
2044 | ||
2045 | /* bisection */ | |
2046 | min_index = 0; | |
2047 | max_index = tree->n_chunks - 1; | |
2048 | ||
2049 | while (min_index <= max_index) { | |
2050 | size_t mid = (min_index + max_index) / 2; | |
2051 | ||
2052 | if (logical < tree->chunks[mid]->offset) { | |
2053 | if (mid < 1) | |
2054 | return NULL; | |
2055 | ||
2056 | max_index = mid - 1; | |
2057 | } else if (logical >= tree->chunks[mid]->offset + tree->chunks[mid]->length) | |
2058 | min_index = mid + 1; | |
2059 | else | |
2060 | return tree->chunks[mid]; | |
2061 | } | |
2062 | ||
2063 | return NULL; | |
2064 | } | |
2065 | ||
2066 | static int btrfs_is_nocow_fd(int fd) { | |
efb6a76a | 2067 | unsigned flags; |
05f38c89 | 2068 | int r; |
efb6a76a MY |
2069 | |
2070 | assert(fd >= 0); | |
2071 | ||
05f38c89 LP |
2072 | r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC); |
2073 | if (r < 0) | |
2074 | return r; | |
2075 | if (r == 0) | |
efb6a76a MY |
2076 | return -ENOTTY; |
2077 | ||
05f38c89 LP |
2078 | r = read_attr_fd(fd, &flags); |
2079 | if (r < 0) | |
2080 | return r; | |
efb6a76a MY |
2081 | |
2082 | return FLAGS_SET(flags, FS_NOCOW_FL) && !FLAGS_SET(flags, FS_COMPR_FL); | |
2083 | } | |
2084 | ||
2085 | int btrfs_get_file_physical_offset_fd(int fd, uint64_t *ret) { | |
2086 | ||
2087 | struct btrfs_ioctl_search_args search_args = { | |
2088 | .key.min_type = BTRFS_EXTENT_DATA_KEY, | |
2089 | .key.max_type = BTRFS_EXTENT_DATA_KEY, | |
2090 | ||
2091 | .key.min_offset = 0, | |
2092 | .key.max_offset = UINT64_MAX, | |
2093 | ||
2094 | .key.min_transid = 0, | |
2095 | .key.max_transid = UINT64_MAX, | |
2096 | }; | |
2097 | ||
2098 | _cleanup_(btrfs_chunk_tree_done) BtrfsChunkTree tree = {}; | |
2099 | uint64_t subvol_id; | |
2100 | struct stat st; | |
2101 | int r; | |
2102 | ||
2103 | assert(fd >= 0); | |
2104 | assert(ret); | |
2105 | ||
2106 | if (fstat(fd, &st) < 0) | |
2107 | return -errno; | |
2108 | ||
2109 | r = stat_verify_regular(&st); | |
2110 | if (r < 0) | |
2111 | return r; | |
2112 | ||
2113 | r = btrfs_is_nocow_fd(fd); | |
2114 | if (r < 0) | |
2115 | return r; | |
2116 | if (r == 0) | |
2117 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
2118 | "Cannot get physical address for btrfs extent: CoW enabled"); | |
2119 | ||
2120 | r = btrfs_subvol_get_id_fd(fd, &subvol_id); | |
2121 | if (r < 0) | |
2122 | return r; | |
2123 | ||
2124 | r = btrfs_read_chunk_tree_fd(fd, &tree); | |
2125 | if (r < 0) | |
2126 | return r; | |
2127 | ||
2128 | search_args.key.tree_id = subvol_id; | |
2129 | search_args.key.min_objectid = search_args.key.max_objectid = st.st_ino; | |
2130 | ||
2131 | while (btrfs_ioctl_search_args_compare(&search_args) <= 0) { | |
801bf40c LP |
2132 | struct btrfs_ioctl_search_header sh; |
2133 | const void *body; | |
efb6a76a MY |
2134 | |
2135 | search_args.key.nr_items = 256; | |
2136 | ||
2137 | if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &search_args) < 0) | |
2138 | return -errno; | |
2139 | ||
2140 | if (search_args.key.nr_items == 0) | |
2141 | break; | |
2142 | ||
801bf40c | 2143 | FOREACH_BTRFS_IOCTL_SEARCH_HEADER(sh, body, search_args) { |
efb6a76a MY |
2144 | uint64_t logical_offset; |
2145 | BtrfsChunk *chunk; | |
2146 | ||
801bf40c | 2147 | btrfs_ioctl_search_args_set(&search_args, &sh); |
efb6a76a | 2148 | |
801bf40c | 2149 | if (sh.type != BTRFS_EXTENT_DATA_KEY) |
efb6a76a MY |
2150 | continue; |
2151 | ||
801bf40c | 2152 | if (sh.objectid != st.st_ino) |
efb6a76a MY |
2153 | continue; |
2154 | ||
801bf40c | 2155 | const struct btrfs_file_extent_item *item = body; |
efb6a76a MY |
2156 | if (!IN_SET(item->type, BTRFS_FILE_EXTENT_REG, BTRFS_FILE_EXTENT_PREALLOC)) |
2157 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
2158 | "Cannot get physical address for btrfs extent: invalid type %" PRIu8, | |
2159 | item->type); | |
2160 | ||
2161 | if (item->compression != 0 || item->encryption != 0 || item->other_encoding != 0) | |
2162 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
2163 | "Cannot get physical address for btrfs extent: has incompatible property"); | |
2164 | ||
2165 | logical_offset = le64toh(item->disk_bytenr); | |
2166 | if (logical_offset == 0) | |
2167 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
2168 | "Cannot get physical address for btrfs extent: failed to get logical offset"); | |
2169 | ||
2170 | chunk = btrfs_find_chunk_from_logical_address(&tree, logical_offset); | |
2171 | if (!chunk) | |
2172 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
2173 | "Cannot get physical address for btrfs extent: no matching chunk found"); | |
2174 | ||
2175 | if ((chunk->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0) | |
2176 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), | |
2177 | "Cannot get physical address for btrfs extent: unsupported profile"); | |
2178 | ||
2179 | uint64_t relative_chunk, relative_stripe, stripe_nr; | |
2180 | uint16_t stripe_index; | |
2181 | ||
2182 | assert(logical_offset >= chunk->offset); | |
2183 | assert(chunk->n_stripes > 0); | |
2184 | assert(chunk->stripe_len > 0); | |
2185 | ||
2186 | relative_chunk = logical_offset - chunk->offset; | |
2187 | stripe_nr = relative_chunk / chunk->stripe_len; | |
2188 | relative_stripe = relative_chunk - stripe_nr * chunk->stripe_len; | |
2189 | stripe_index = stripe_nr % chunk->n_stripes; | |
2190 | ||
2191 | *ret = chunk->stripes[stripe_index].offset + | |
2192 | stripe_nr / chunk->n_stripes * chunk->stripe_len + | |
2193 | relative_stripe; | |
2194 | ||
2195 | return 0; | |
2196 | } | |
2197 | ||
2198 | if (!btrfs_ioctl_search_args_inc(&search_args)) | |
2199 | break; | |
2200 | } | |
2201 | ||
2202 | return -ENODATA; | |
2203 | } |