]>
Commit | Line | Data |
---|---|---|
31db9f7c AB |
1 | /* |
2 | * Copyright (C) 2012 Alexander Block. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/bsearch.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/file.h> | |
22 | #include <linux/sort.h> | |
23 | #include <linux/mount.h> | |
24 | #include <linux/xattr.h> | |
25 | #include <linux/posix_acl_xattr.h> | |
26 | #include <linux/radix-tree.h> | |
a1857ebe | 27 | #include <linux/vmalloc.h> |
ed84885d | 28 | #include <linux/string.h> |
2351f431 | 29 | #include <linux/compat.h> |
31db9f7c AB |
30 | |
31 | #include "send.h" | |
32 | #include "backref.h" | |
0b947aff | 33 | #include "hash.h" |
31db9f7c AB |
34 | #include "locking.h" |
35 | #include "disk-io.h" | |
36 | #include "btrfs_inode.h" | |
37 | #include "transaction.h" | |
ebb8765b | 38 | #include "compression.h" |
31db9f7c | 39 | |
31db9f7c AB |
40 | /* |
41 | * A fs_path is a helper to dynamically build path names with unknown size. | |
42 | * It reallocates the internal buffer on demand. | |
43 | * It allows fast adding of path elements on the right side (normal path) and | |
44 | * fast adding to the left side (reversed path). A reversed path can also be | |
45 | * unreversed if needed. | |
46 | */ | |
47 | struct fs_path { | |
48 | union { | |
49 | struct { | |
50 | char *start; | |
51 | char *end; | |
31db9f7c AB |
52 | |
53 | char *buf; | |
1f5a7ff9 DS |
54 | unsigned short buf_len:15; |
55 | unsigned short reversed:1; | |
31db9f7c AB |
56 | char inline_buf[]; |
57 | }; | |
ace01050 DS |
58 | /* |
59 | * Average path length does not exceed 200 bytes, we'll have | |
60 | * better packing in the slab and higher chance to satisfy | |
61 | * a allocation later during send. | |
62 | */ | |
63 | char pad[256]; | |
31db9f7c AB |
64 | }; |
65 | }; | |
66 | #define FS_PATH_INLINE_SIZE \ | |
67 | (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) | |
68 | ||
69 | ||
70 | /* reused for each extent */ | |
71 | struct clone_root { | |
72 | struct btrfs_root *root; | |
73 | u64 ino; | |
74 | u64 offset; | |
75 | ||
76 | u64 found_refs; | |
77 | }; | |
78 | ||
79 | #define SEND_CTX_MAX_NAME_CACHE_SIZE 128 | |
80 | #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2) | |
81 | ||
82 | struct send_ctx { | |
83 | struct file *send_filp; | |
84 | loff_t send_off; | |
85 | char *send_buf; | |
86 | u32 send_size; | |
87 | u32 send_max_size; | |
88 | u64 total_send_size; | |
89 | u64 cmd_send_size[BTRFS_SEND_C_MAX + 1]; | |
cb95e7bf | 90 | u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ |
31db9f7c | 91 | |
31db9f7c AB |
92 | struct btrfs_root *send_root; |
93 | struct btrfs_root *parent_root; | |
94 | struct clone_root *clone_roots; | |
95 | int clone_roots_cnt; | |
96 | ||
97 | /* current state of the compare_tree call */ | |
98 | struct btrfs_path *left_path; | |
99 | struct btrfs_path *right_path; | |
100 | struct btrfs_key *cmp_key; | |
101 | ||
102 | /* | |
103 | * infos of the currently processed inode. In case of deleted inodes, | |
104 | * these are the values from the deleted inode. | |
105 | */ | |
106 | u64 cur_ino; | |
107 | u64 cur_inode_gen; | |
108 | int cur_inode_new; | |
109 | int cur_inode_new_gen; | |
110 | int cur_inode_deleted; | |
31db9f7c AB |
111 | u64 cur_inode_size; |
112 | u64 cur_inode_mode; | |
644d1940 | 113 | u64 cur_inode_rdev; |
16e7549f | 114 | u64 cur_inode_last_extent; |
31db9f7c AB |
115 | |
116 | u64 send_progress; | |
117 | ||
118 | struct list_head new_refs; | |
119 | struct list_head deleted_refs; | |
120 | ||
121 | struct radix_tree_root name_cache; | |
122 | struct list_head name_cache_list; | |
123 | int name_cache_size; | |
124 | ||
2131bcd3 LB |
125 | struct file_ra_state ra; |
126 | ||
31db9f7c | 127 | char *read_buf; |
9f03740a FDBM |
128 | |
129 | /* | |
130 | * We process inodes by their increasing order, so if before an | |
131 | * incremental send we reverse the parent/child relationship of | |
132 | * directories such that a directory with a lower inode number was | |
133 | * the parent of a directory with a higher inode number, and the one | |
134 | * becoming the new parent got renamed too, we can't rename/move the | |
135 | * directory with lower inode number when we finish processing it - we | |
136 | * must process the directory with higher inode number first, then | |
137 | * rename/move it and then rename/move the directory with lower inode | |
138 | * number. Example follows. | |
139 | * | |
140 | * Tree state when the first send was performed: | |
141 | * | |
142 | * . | |
143 | * |-- a (ino 257) | |
144 | * |-- b (ino 258) | |
145 | * | | |
146 | * | | |
147 | * |-- c (ino 259) | |
148 | * | |-- d (ino 260) | |
149 | * | | |
150 | * |-- c2 (ino 261) | |
151 | * | |
152 | * Tree state when the second (incremental) send is performed: | |
153 | * | |
154 | * . | |
155 | * |-- a (ino 257) | |
156 | * |-- b (ino 258) | |
157 | * |-- c2 (ino 261) | |
158 | * |-- d2 (ino 260) | |
159 | * |-- cc (ino 259) | |
160 | * | |
161 | * The sequence of steps that lead to the second state was: | |
162 | * | |
163 | * mv /a/b/c/d /a/b/c2/d2 | |
164 | * mv /a/b/c /a/b/c2/d2/cc | |
165 | * | |
166 | * "c" has lower inode number, but we can't move it (2nd mv operation) | |
167 | * before we move "d", which has higher inode number. | |
168 | * | |
169 | * So we just memorize which move/rename operations must be performed | |
170 | * later when their respective parent is processed and moved/renamed. | |
171 | */ | |
172 | ||
173 | /* Indexed by parent directory inode number. */ | |
174 | struct rb_root pending_dir_moves; | |
175 | ||
176 | /* | |
177 | * Reverse index, indexed by the inode number of a directory that | |
178 | * is waiting for the move/rename of its immediate parent before its | |
179 | * own move/rename can be performed. | |
180 | */ | |
181 | struct rb_root waiting_dir_moves; | |
9dc44214 FM |
182 | |
183 | /* | |
184 | * A directory that is going to be rm'ed might have a child directory | |
185 | * which is in the pending directory moves index above. In this case, | |
186 | * the directory can only be removed after the move/rename of its child | |
187 | * is performed. Example: | |
188 | * | |
189 | * Parent snapshot: | |
190 | * | |
191 | * . (ino 256) | |
192 | * |-- a/ (ino 257) | |
193 | * |-- b/ (ino 258) | |
194 | * |-- c/ (ino 259) | |
195 | * | |-- x/ (ino 260) | |
196 | * | | |
197 | * |-- y/ (ino 261) | |
198 | * | |
199 | * Send snapshot: | |
200 | * | |
201 | * . (ino 256) | |
202 | * |-- a/ (ino 257) | |
203 | * |-- b/ (ino 258) | |
204 | * |-- YY/ (ino 261) | |
205 | * |-- x/ (ino 260) | |
206 | * | |
207 | * Sequence of steps that lead to the send snapshot: | |
208 | * rm -f /a/b/c/foo.txt | |
209 | * mv /a/b/y /a/b/YY | |
210 | * mv /a/b/c/x /a/b/YY | |
211 | * rmdir /a/b/c | |
212 | * | |
213 | * When the child is processed, its move/rename is delayed until its | |
214 | * parent is processed (as explained above), but all other operations | |
215 | * like update utimes, chown, chgrp, etc, are performed and the paths | |
216 | * that it uses for those operations must use the orphanized name of | |
217 | * its parent (the directory we're going to rm later), so we need to | |
218 | * memorize that name. | |
219 | * | |
220 | * Indexed by the inode number of the directory to be deleted. | |
221 | */ | |
222 | struct rb_root orphan_dirs; | |
9f03740a FDBM |
223 | }; |
224 | ||
225 | struct pending_dir_move { | |
226 | struct rb_node node; | |
227 | struct list_head list; | |
228 | u64 parent_ino; | |
229 | u64 ino; | |
230 | u64 gen; | |
231 | struct list_head update_refs; | |
232 | }; | |
233 | ||
234 | struct waiting_dir_move { | |
235 | struct rb_node node; | |
236 | u64 ino; | |
9dc44214 FM |
237 | /* |
238 | * There might be some directory that could not be removed because it | |
239 | * was waiting for this directory inode to be moved first. Therefore | |
240 | * after this directory is moved, we can try to rmdir the ino rmdir_ino. | |
241 | */ | |
242 | u64 rmdir_ino; | |
8b191a68 | 243 | bool orphanized; |
9dc44214 FM |
244 | }; |
245 | ||
246 | struct orphan_dir_info { | |
247 | struct rb_node node; | |
248 | u64 ino; | |
249 | u64 gen; | |
31db9f7c AB |
250 | }; |
251 | ||
252 | struct name_cache_entry { | |
253 | struct list_head list; | |
7e0926fe AB |
254 | /* |
255 | * radix_tree has only 32bit entries but we need to handle 64bit inums. | |
256 | * We use the lower 32bit of the 64bit inum to store it in the tree. If | |
257 | * more then one inum would fall into the same entry, we use radix_list | |
258 | * to store the additional entries. radix_list is also used to store | |
259 | * entries where two entries have the same inum but different | |
260 | * generations. | |
261 | */ | |
262 | struct list_head radix_list; | |
31db9f7c AB |
263 | u64 ino; |
264 | u64 gen; | |
265 | u64 parent_ino; | |
266 | u64 parent_gen; | |
267 | int ret; | |
268 | int need_later_update; | |
269 | int name_len; | |
270 | char name[]; | |
271 | }; | |
272 | ||
95155585 FM |
273 | static void inconsistent_snapshot_error(struct send_ctx *sctx, |
274 | enum btrfs_compare_tree_result result, | |
275 | const char *what) | |
276 | { | |
277 | const char *result_string; | |
278 | ||
279 | switch (result) { | |
280 | case BTRFS_COMPARE_TREE_NEW: | |
281 | result_string = "new"; | |
282 | break; | |
283 | case BTRFS_COMPARE_TREE_DELETED: | |
284 | result_string = "deleted"; | |
285 | break; | |
286 | case BTRFS_COMPARE_TREE_CHANGED: | |
287 | result_string = "updated"; | |
288 | break; | |
289 | case BTRFS_COMPARE_TREE_SAME: | |
290 | ASSERT(0); | |
291 | result_string = "unchanged"; | |
292 | break; | |
293 | default: | |
294 | ASSERT(0); | |
295 | result_string = "unexpected"; | |
296 | } | |
297 | ||
298 | btrfs_err(sctx->send_root->fs_info, | |
299 | "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu", | |
300 | result_string, what, sctx->cmp_key->objectid, | |
301 | sctx->send_root->root_key.objectid, | |
302 | (sctx->parent_root ? | |
303 | sctx->parent_root->root_key.objectid : 0)); | |
304 | } | |
305 | ||
9f03740a FDBM |
306 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); |
307 | ||
9dc44214 FM |
308 | static struct waiting_dir_move * |
309 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino); | |
310 | ||
311 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino); | |
312 | ||
16e7549f JB |
313 | static int need_send_hole(struct send_ctx *sctx) |
314 | { | |
315 | return (sctx->parent_root && !sctx->cur_inode_new && | |
316 | !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && | |
317 | S_ISREG(sctx->cur_inode_mode)); | |
318 | } | |
319 | ||
31db9f7c AB |
320 | static void fs_path_reset(struct fs_path *p) |
321 | { | |
322 | if (p->reversed) { | |
323 | p->start = p->buf + p->buf_len - 1; | |
324 | p->end = p->start; | |
325 | *p->start = 0; | |
326 | } else { | |
327 | p->start = p->buf; | |
328 | p->end = p->start; | |
329 | *p->start = 0; | |
330 | } | |
331 | } | |
332 | ||
924794c9 | 333 | static struct fs_path *fs_path_alloc(void) |
31db9f7c AB |
334 | { |
335 | struct fs_path *p; | |
336 | ||
e780b0d1 | 337 | p = kmalloc(sizeof(*p), GFP_KERNEL); |
31db9f7c AB |
338 | if (!p) |
339 | return NULL; | |
340 | p->reversed = 0; | |
31db9f7c AB |
341 | p->buf = p->inline_buf; |
342 | p->buf_len = FS_PATH_INLINE_SIZE; | |
343 | fs_path_reset(p); | |
344 | return p; | |
345 | } | |
346 | ||
924794c9 | 347 | static struct fs_path *fs_path_alloc_reversed(void) |
31db9f7c AB |
348 | { |
349 | struct fs_path *p; | |
350 | ||
924794c9 | 351 | p = fs_path_alloc(); |
31db9f7c AB |
352 | if (!p) |
353 | return NULL; | |
354 | p->reversed = 1; | |
355 | fs_path_reset(p); | |
356 | return p; | |
357 | } | |
358 | ||
924794c9 | 359 | static void fs_path_free(struct fs_path *p) |
31db9f7c AB |
360 | { |
361 | if (!p) | |
362 | return; | |
ace01050 DS |
363 | if (p->buf != p->inline_buf) |
364 | kfree(p->buf); | |
31db9f7c AB |
365 | kfree(p); |
366 | } | |
367 | ||
368 | static int fs_path_len(struct fs_path *p) | |
369 | { | |
370 | return p->end - p->start; | |
371 | } | |
372 | ||
373 | static int fs_path_ensure_buf(struct fs_path *p, int len) | |
374 | { | |
375 | char *tmp_buf; | |
376 | int path_len; | |
377 | int old_buf_len; | |
378 | ||
379 | len++; | |
380 | ||
381 | if (p->buf_len >= len) | |
382 | return 0; | |
383 | ||
cfd4a535 CM |
384 | if (len > PATH_MAX) { |
385 | WARN_ON(1); | |
386 | return -ENOMEM; | |
387 | } | |
388 | ||
1b2782c8 DS |
389 | path_len = p->end - p->start; |
390 | old_buf_len = p->buf_len; | |
391 | ||
ace01050 DS |
392 | /* |
393 | * First time the inline_buf does not suffice | |
394 | */ | |
01a9a8a9 | 395 | if (p->buf == p->inline_buf) { |
e780b0d1 | 396 | tmp_buf = kmalloc(len, GFP_KERNEL); |
01a9a8a9 FM |
397 | if (tmp_buf) |
398 | memcpy(tmp_buf, p->buf, old_buf_len); | |
399 | } else { | |
e780b0d1 | 400 | tmp_buf = krealloc(p->buf, len, GFP_KERNEL); |
01a9a8a9 | 401 | } |
9c9ca00b DS |
402 | if (!tmp_buf) |
403 | return -ENOMEM; | |
404 | p->buf = tmp_buf; | |
405 | /* | |
406 | * The real size of the buffer is bigger, this will let the fast path | |
407 | * happen most of the time | |
408 | */ | |
409 | p->buf_len = ksize(p->buf); | |
ace01050 | 410 | |
31db9f7c AB |
411 | if (p->reversed) { |
412 | tmp_buf = p->buf + old_buf_len - path_len - 1; | |
413 | p->end = p->buf + p->buf_len - 1; | |
414 | p->start = p->end - path_len; | |
415 | memmove(p->start, tmp_buf, path_len + 1); | |
416 | } else { | |
417 | p->start = p->buf; | |
418 | p->end = p->start + path_len; | |
419 | } | |
420 | return 0; | |
421 | } | |
422 | ||
b23ab57d DS |
423 | static int fs_path_prepare_for_add(struct fs_path *p, int name_len, |
424 | char **prepared) | |
31db9f7c AB |
425 | { |
426 | int ret; | |
427 | int new_len; | |
428 | ||
429 | new_len = p->end - p->start + name_len; | |
430 | if (p->start != p->end) | |
431 | new_len++; | |
432 | ret = fs_path_ensure_buf(p, new_len); | |
433 | if (ret < 0) | |
434 | goto out; | |
435 | ||
436 | if (p->reversed) { | |
437 | if (p->start != p->end) | |
438 | *--p->start = '/'; | |
439 | p->start -= name_len; | |
b23ab57d | 440 | *prepared = p->start; |
31db9f7c AB |
441 | } else { |
442 | if (p->start != p->end) | |
443 | *p->end++ = '/'; | |
b23ab57d | 444 | *prepared = p->end; |
31db9f7c AB |
445 | p->end += name_len; |
446 | *p->end = 0; | |
447 | } | |
448 | ||
449 | out: | |
450 | return ret; | |
451 | } | |
452 | ||
453 | static int fs_path_add(struct fs_path *p, const char *name, int name_len) | |
454 | { | |
455 | int ret; | |
b23ab57d | 456 | char *prepared; |
31db9f7c | 457 | |
b23ab57d | 458 | ret = fs_path_prepare_for_add(p, name_len, &prepared); |
31db9f7c AB |
459 | if (ret < 0) |
460 | goto out; | |
b23ab57d | 461 | memcpy(prepared, name, name_len); |
31db9f7c AB |
462 | |
463 | out: | |
464 | return ret; | |
465 | } | |
466 | ||
467 | static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) | |
468 | { | |
469 | int ret; | |
b23ab57d | 470 | char *prepared; |
31db9f7c | 471 | |
b23ab57d | 472 | ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); |
31db9f7c AB |
473 | if (ret < 0) |
474 | goto out; | |
b23ab57d | 475 | memcpy(prepared, p2->start, p2->end - p2->start); |
31db9f7c AB |
476 | |
477 | out: | |
478 | return ret; | |
479 | } | |
480 | ||
481 | static int fs_path_add_from_extent_buffer(struct fs_path *p, | |
482 | struct extent_buffer *eb, | |
483 | unsigned long off, int len) | |
484 | { | |
485 | int ret; | |
b23ab57d | 486 | char *prepared; |
31db9f7c | 487 | |
b23ab57d | 488 | ret = fs_path_prepare_for_add(p, len, &prepared); |
31db9f7c AB |
489 | if (ret < 0) |
490 | goto out; | |
491 | ||
b23ab57d | 492 | read_extent_buffer(eb, prepared, off, len); |
31db9f7c AB |
493 | |
494 | out: | |
495 | return ret; | |
496 | } | |
497 | ||
31db9f7c AB |
498 | static int fs_path_copy(struct fs_path *p, struct fs_path *from) |
499 | { | |
500 | int ret; | |
501 | ||
502 | p->reversed = from->reversed; | |
503 | fs_path_reset(p); | |
504 | ||
505 | ret = fs_path_add_path(p, from); | |
506 | ||
507 | return ret; | |
508 | } | |
509 | ||
510 | ||
511 | static void fs_path_unreverse(struct fs_path *p) | |
512 | { | |
513 | char *tmp; | |
514 | int len; | |
515 | ||
516 | if (!p->reversed) | |
517 | return; | |
518 | ||
519 | tmp = p->start; | |
520 | len = p->end - p->start; | |
521 | p->start = p->buf; | |
522 | p->end = p->start + len; | |
523 | memmove(p->start, tmp, len + 1); | |
524 | p->reversed = 0; | |
525 | } | |
526 | ||
527 | static struct btrfs_path *alloc_path_for_send(void) | |
528 | { | |
529 | struct btrfs_path *path; | |
530 | ||
531 | path = btrfs_alloc_path(); | |
532 | if (!path) | |
533 | return NULL; | |
534 | path->search_commit_root = 1; | |
535 | path->skip_locking = 1; | |
3f8a18cc | 536 | path->need_commit_sem = 1; |
31db9f7c AB |
537 | return path; |
538 | } | |
539 | ||
48a3b636 | 540 | static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) |
31db9f7c AB |
541 | { |
542 | int ret; | |
31db9f7c AB |
543 | u32 pos = 0; |
544 | ||
31db9f7c | 545 | while (pos < len) { |
8e93157b | 546 | ret = kernel_write(filp, buf + pos, len - pos, off); |
31db9f7c AB |
547 | /* TODO handle that correctly */ |
548 | /*if (ret == -ERESTARTSYS) { | |
549 | continue; | |
550 | }*/ | |
551 | if (ret < 0) | |
8e93157b | 552 | return ret; |
31db9f7c | 553 | if (ret == 0) { |
8e93157b | 554 | return -EIO; |
31db9f7c AB |
555 | } |
556 | pos += ret; | |
557 | } | |
558 | ||
8e93157b | 559 | return 0; |
31db9f7c AB |
560 | } |
561 | ||
562 | static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) | |
563 | { | |
564 | struct btrfs_tlv_header *hdr; | |
565 | int total_len = sizeof(*hdr) + len; | |
566 | int left = sctx->send_max_size - sctx->send_size; | |
567 | ||
568 | if (unlikely(left < total_len)) | |
569 | return -EOVERFLOW; | |
570 | ||
571 | hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); | |
572 | hdr->tlv_type = cpu_to_le16(attr); | |
573 | hdr->tlv_len = cpu_to_le16(len); | |
574 | memcpy(hdr + 1, data, len); | |
575 | sctx->send_size += total_len; | |
576 | ||
577 | return 0; | |
578 | } | |
579 | ||
95bc79d5 DS |
580 | #define TLV_PUT_DEFINE_INT(bits) \ |
581 | static int tlv_put_u##bits(struct send_ctx *sctx, \ | |
582 | u##bits attr, u##bits value) \ | |
583 | { \ | |
584 | __le##bits __tmp = cpu_to_le##bits(value); \ | |
585 | return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ | |
586 | } | |
31db9f7c | 587 | |
95bc79d5 | 588 | TLV_PUT_DEFINE_INT(64) |
31db9f7c AB |
589 | |
590 | static int tlv_put_string(struct send_ctx *sctx, u16 attr, | |
591 | const char *str, int len) | |
592 | { | |
593 | if (len == -1) | |
594 | len = strlen(str); | |
595 | return tlv_put(sctx, attr, str, len); | |
596 | } | |
597 | ||
598 | static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, | |
599 | const u8 *uuid) | |
600 | { | |
601 | return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); | |
602 | } | |
603 | ||
31db9f7c AB |
604 | static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, |
605 | struct extent_buffer *eb, | |
606 | struct btrfs_timespec *ts) | |
607 | { | |
608 | struct btrfs_timespec bts; | |
609 | read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); | |
610 | return tlv_put(sctx, attr, &bts, sizeof(bts)); | |
611 | } | |
612 | ||
613 | ||
614 | #define TLV_PUT(sctx, attrtype, attrlen, data) \ | |
615 | do { \ | |
616 | ret = tlv_put(sctx, attrtype, attrlen, data); \ | |
617 | if (ret < 0) \ | |
618 | goto tlv_put_failure; \ | |
619 | } while (0) | |
620 | ||
621 | #define TLV_PUT_INT(sctx, attrtype, bits, value) \ | |
622 | do { \ | |
623 | ret = tlv_put_u##bits(sctx, attrtype, value); \ | |
624 | if (ret < 0) \ | |
625 | goto tlv_put_failure; \ | |
626 | } while (0) | |
627 | ||
628 | #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) | |
629 | #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) | |
630 | #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) | |
631 | #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) | |
632 | #define TLV_PUT_STRING(sctx, attrtype, str, len) \ | |
633 | do { \ | |
634 | ret = tlv_put_string(sctx, attrtype, str, len); \ | |
635 | if (ret < 0) \ | |
636 | goto tlv_put_failure; \ | |
637 | } while (0) | |
638 | #define TLV_PUT_PATH(sctx, attrtype, p) \ | |
639 | do { \ | |
640 | ret = tlv_put_string(sctx, attrtype, p->start, \ | |
641 | p->end - p->start); \ | |
642 | if (ret < 0) \ | |
643 | goto tlv_put_failure; \ | |
644 | } while(0) | |
645 | #define TLV_PUT_UUID(sctx, attrtype, uuid) \ | |
646 | do { \ | |
647 | ret = tlv_put_uuid(sctx, attrtype, uuid); \ | |
648 | if (ret < 0) \ | |
649 | goto tlv_put_failure; \ | |
650 | } while (0) | |
31db9f7c AB |
651 | #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ |
652 | do { \ | |
653 | ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ | |
654 | if (ret < 0) \ | |
655 | goto tlv_put_failure; \ | |
656 | } while (0) | |
657 | ||
658 | static int send_header(struct send_ctx *sctx) | |
659 | { | |
660 | struct btrfs_stream_header hdr; | |
661 | ||
662 | strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); | |
663 | hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); | |
664 | ||
1bcea355 AJ |
665 | return write_buf(sctx->send_filp, &hdr, sizeof(hdr), |
666 | &sctx->send_off); | |
31db9f7c AB |
667 | } |
668 | ||
669 | /* | |
670 | * For each command/item we want to send to userspace, we call this function. | |
671 | */ | |
672 | static int begin_cmd(struct send_ctx *sctx, int cmd) | |
673 | { | |
674 | struct btrfs_cmd_header *hdr; | |
675 | ||
fae7f21c | 676 | if (WARN_ON(!sctx->send_buf)) |
31db9f7c | 677 | return -EINVAL; |
31db9f7c AB |
678 | |
679 | BUG_ON(sctx->send_size); | |
680 | ||
681 | sctx->send_size += sizeof(*hdr); | |
682 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
683 | hdr->cmd = cpu_to_le16(cmd); | |
684 | ||
685 | return 0; | |
686 | } | |
687 | ||
688 | static int send_cmd(struct send_ctx *sctx) | |
689 | { | |
690 | int ret; | |
691 | struct btrfs_cmd_header *hdr; | |
692 | u32 crc; | |
693 | ||
694 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
695 | hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr)); | |
696 | hdr->crc = 0; | |
697 | ||
0b947aff | 698 | crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); |
31db9f7c AB |
699 | hdr->crc = cpu_to_le32(crc); |
700 | ||
1bcea355 AJ |
701 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, |
702 | &sctx->send_off); | |
31db9f7c AB |
703 | |
704 | sctx->total_send_size += sctx->send_size; | |
705 | sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size; | |
706 | sctx->send_size = 0; | |
707 | ||
708 | return ret; | |
709 | } | |
710 | ||
711 | /* | |
712 | * Sends a move instruction to user space | |
713 | */ | |
714 | static int send_rename(struct send_ctx *sctx, | |
715 | struct fs_path *from, struct fs_path *to) | |
716 | { | |
04ab956e | 717 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
718 | int ret; |
719 | ||
04ab956e | 720 | btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start); |
31db9f7c AB |
721 | |
722 | ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); | |
723 | if (ret < 0) | |
724 | goto out; | |
725 | ||
726 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); | |
727 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); | |
728 | ||
729 | ret = send_cmd(sctx); | |
730 | ||
731 | tlv_put_failure: | |
732 | out: | |
733 | return ret; | |
734 | } | |
735 | ||
736 | /* | |
737 | * Sends a link instruction to user space | |
738 | */ | |
739 | static int send_link(struct send_ctx *sctx, | |
740 | struct fs_path *path, struct fs_path *lnk) | |
741 | { | |
04ab956e | 742 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
743 | int ret; |
744 | ||
04ab956e | 745 | btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start); |
31db9f7c AB |
746 | |
747 | ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); | |
748 | if (ret < 0) | |
749 | goto out; | |
750 | ||
751 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
752 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); | |
753 | ||
754 | ret = send_cmd(sctx); | |
755 | ||
756 | tlv_put_failure: | |
757 | out: | |
758 | return ret; | |
759 | } | |
760 | ||
761 | /* | |
762 | * Sends an unlink instruction to user space | |
763 | */ | |
764 | static int send_unlink(struct send_ctx *sctx, struct fs_path *path) | |
765 | { | |
04ab956e | 766 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
767 | int ret; |
768 | ||
04ab956e | 769 | btrfs_debug(fs_info, "send_unlink %s", path->start); |
31db9f7c AB |
770 | |
771 | ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); | |
772 | if (ret < 0) | |
773 | goto out; | |
774 | ||
775 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
776 | ||
777 | ret = send_cmd(sctx); | |
778 | ||
779 | tlv_put_failure: | |
780 | out: | |
781 | return ret; | |
782 | } | |
783 | ||
784 | /* | |
785 | * Sends a rmdir instruction to user space | |
786 | */ | |
787 | static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) | |
788 | { | |
04ab956e | 789 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
790 | int ret; |
791 | ||
04ab956e | 792 | btrfs_debug(fs_info, "send_rmdir %s", path->start); |
31db9f7c AB |
793 | |
794 | ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); | |
795 | if (ret < 0) | |
796 | goto out; | |
797 | ||
798 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
799 | ||
800 | ret = send_cmd(sctx); | |
801 | ||
802 | tlv_put_failure: | |
803 | out: | |
804 | return ret; | |
805 | } | |
806 | ||
807 | /* | |
808 | * Helper function to retrieve some fields from an inode item. | |
809 | */ | |
3f8a18cc JB |
810 | static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, |
811 | u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, | |
812 | u64 *gid, u64 *rdev) | |
31db9f7c AB |
813 | { |
814 | int ret; | |
815 | struct btrfs_inode_item *ii; | |
816 | struct btrfs_key key; | |
31db9f7c AB |
817 | |
818 | key.objectid = ino; | |
819 | key.type = BTRFS_INODE_ITEM_KEY; | |
820 | key.offset = 0; | |
821 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
31db9f7c | 822 | if (ret) { |
3f8a18cc JB |
823 | if (ret > 0) |
824 | ret = -ENOENT; | |
825 | return ret; | |
31db9f7c AB |
826 | } |
827 | ||
828 | ii = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
829 | struct btrfs_inode_item); | |
830 | if (size) | |
831 | *size = btrfs_inode_size(path->nodes[0], ii); | |
832 | if (gen) | |
833 | *gen = btrfs_inode_generation(path->nodes[0], ii); | |
834 | if (mode) | |
835 | *mode = btrfs_inode_mode(path->nodes[0], ii); | |
836 | if (uid) | |
837 | *uid = btrfs_inode_uid(path->nodes[0], ii); | |
838 | if (gid) | |
839 | *gid = btrfs_inode_gid(path->nodes[0], ii); | |
85a7b33b AB |
840 | if (rdev) |
841 | *rdev = btrfs_inode_rdev(path->nodes[0], ii); | |
31db9f7c | 842 | |
3f8a18cc JB |
843 | return ret; |
844 | } | |
845 | ||
846 | static int get_inode_info(struct btrfs_root *root, | |
847 | u64 ino, u64 *size, u64 *gen, | |
848 | u64 *mode, u64 *uid, u64 *gid, | |
849 | u64 *rdev) | |
850 | { | |
851 | struct btrfs_path *path; | |
852 | int ret; | |
853 | ||
854 | path = alloc_path_for_send(); | |
855 | if (!path) | |
856 | return -ENOMEM; | |
857 | ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, | |
858 | rdev); | |
31db9f7c AB |
859 | btrfs_free_path(path); |
860 | return ret; | |
861 | } | |
862 | ||
863 | typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, | |
864 | struct fs_path *p, | |
865 | void *ctx); | |
866 | ||
867 | /* | |
96b5bd77 JS |
868 | * Helper function to iterate the entries in ONE btrfs_inode_ref or |
869 | * btrfs_inode_extref. | |
31db9f7c AB |
870 | * The iterate callback may return a non zero value to stop iteration. This can |
871 | * be a negative value for error codes or 1 to simply stop it. | |
872 | * | |
96b5bd77 | 873 | * path must point to the INODE_REF or INODE_EXTREF when called. |
31db9f7c | 874 | */ |
924794c9 | 875 | static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
876 | struct btrfs_key *found_key, int resolve, |
877 | iterate_inode_ref_t iterate, void *ctx) | |
878 | { | |
96b5bd77 | 879 | struct extent_buffer *eb = path->nodes[0]; |
31db9f7c AB |
880 | struct btrfs_item *item; |
881 | struct btrfs_inode_ref *iref; | |
96b5bd77 | 882 | struct btrfs_inode_extref *extref; |
31db9f7c AB |
883 | struct btrfs_path *tmp_path; |
884 | struct fs_path *p; | |
96b5bd77 | 885 | u32 cur = 0; |
31db9f7c | 886 | u32 total; |
96b5bd77 | 887 | int slot = path->slots[0]; |
31db9f7c AB |
888 | u32 name_len; |
889 | char *start; | |
890 | int ret = 0; | |
96b5bd77 | 891 | int num = 0; |
31db9f7c | 892 | int index; |
96b5bd77 JS |
893 | u64 dir; |
894 | unsigned long name_off; | |
895 | unsigned long elem_size; | |
896 | unsigned long ptr; | |
31db9f7c | 897 | |
924794c9 | 898 | p = fs_path_alloc_reversed(); |
31db9f7c AB |
899 | if (!p) |
900 | return -ENOMEM; | |
901 | ||
902 | tmp_path = alloc_path_for_send(); | |
903 | if (!tmp_path) { | |
924794c9 | 904 | fs_path_free(p); |
31db9f7c AB |
905 | return -ENOMEM; |
906 | } | |
907 | ||
31db9f7c | 908 | |
96b5bd77 JS |
909 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
910 | ptr = (unsigned long)btrfs_item_ptr(eb, slot, | |
911 | struct btrfs_inode_ref); | |
dd3cc16b | 912 | item = btrfs_item_nr(slot); |
96b5bd77 JS |
913 | total = btrfs_item_size(eb, item); |
914 | elem_size = sizeof(*iref); | |
915 | } else { | |
916 | ptr = btrfs_item_ptr_offset(eb, slot); | |
917 | total = btrfs_item_size_nr(eb, slot); | |
918 | elem_size = sizeof(*extref); | |
919 | } | |
920 | ||
31db9f7c AB |
921 | while (cur < total) { |
922 | fs_path_reset(p); | |
923 | ||
96b5bd77 JS |
924 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
925 | iref = (struct btrfs_inode_ref *)(ptr + cur); | |
926 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
927 | name_off = (unsigned long)(iref + 1); | |
928 | index = btrfs_inode_ref_index(eb, iref); | |
929 | dir = found_key->offset; | |
930 | } else { | |
931 | extref = (struct btrfs_inode_extref *)(ptr + cur); | |
932 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
933 | name_off = (unsigned long)&extref->name; | |
934 | index = btrfs_inode_extref_index(eb, extref); | |
935 | dir = btrfs_inode_extref_parent(eb, extref); | |
936 | } | |
937 | ||
31db9f7c | 938 | if (resolve) { |
96b5bd77 JS |
939 | start = btrfs_ref_to_path(root, tmp_path, name_len, |
940 | name_off, eb, dir, | |
941 | p->buf, p->buf_len); | |
31db9f7c AB |
942 | if (IS_ERR(start)) { |
943 | ret = PTR_ERR(start); | |
944 | goto out; | |
945 | } | |
946 | if (start < p->buf) { | |
947 | /* overflow , try again with larger buffer */ | |
948 | ret = fs_path_ensure_buf(p, | |
949 | p->buf_len + p->buf - start); | |
950 | if (ret < 0) | |
951 | goto out; | |
96b5bd77 JS |
952 | start = btrfs_ref_to_path(root, tmp_path, |
953 | name_len, name_off, | |
954 | eb, dir, | |
955 | p->buf, p->buf_len); | |
31db9f7c AB |
956 | if (IS_ERR(start)) { |
957 | ret = PTR_ERR(start); | |
958 | goto out; | |
959 | } | |
960 | BUG_ON(start < p->buf); | |
961 | } | |
962 | p->start = start; | |
963 | } else { | |
96b5bd77 JS |
964 | ret = fs_path_add_from_extent_buffer(p, eb, name_off, |
965 | name_len); | |
31db9f7c AB |
966 | if (ret < 0) |
967 | goto out; | |
968 | } | |
969 | ||
96b5bd77 JS |
970 | cur += elem_size + name_len; |
971 | ret = iterate(num, dir, index, p, ctx); | |
31db9f7c AB |
972 | if (ret) |
973 | goto out; | |
31db9f7c AB |
974 | num++; |
975 | } | |
976 | ||
977 | out: | |
978 | btrfs_free_path(tmp_path); | |
924794c9 | 979 | fs_path_free(p); |
31db9f7c AB |
980 | return ret; |
981 | } | |
982 | ||
983 | typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, | |
984 | const char *name, int name_len, | |
985 | const char *data, int data_len, | |
986 | u8 type, void *ctx); | |
987 | ||
988 | /* | |
989 | * Helper function to iterate the entries in ONE btrfs_dir_item. | |
990 | * The iterate callback may return a non zero value to stop iteration. This can | |
991 | * be a negative value for error codes or 1 to simply stop it. | |
992 | * | |
993 | * path must point to the dir item when called. | |
994 | */ | |
924794c9 | 995 | static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
996 | iterate_dir_item_t iterate, void *ctx) |
997 | { | |
998 | int ret = 0; | |
999 | struct extent_buffer *eb; | |
1000 | struct btrfs_item *item; | |
1001 | struct btrfs_dir_item *di; | |
31db9f7c AB |
1002 | struct btrfs_key di_key; |
1003 | char *buf = NULL; | |
7e3ae33e | 1004 | int buf_len; |
31db9f7c AB |
1005 | u32 name_len; |
1006 | u32 data_len; | |
1007 | u32 cur; | |
1008 | u32 len; | |
1009 | u32 total; | |
1010 | int slot; | |
1011 | int num; | |
1012 | u8 type; | |
1013 | ||
4395e0c4 FM |
1014 | /* |
1015 | * Start with a small buffer (1 page). If later we end up needing more | |
1016 | * space, which can happen for xattrs on a fs with a leaf size greater | |
1017 | * then the page size, attempt to increase the buffer. Typically xattr | |
1018 | * values are small. | |
1019 | */ | |
1020 | buf_len = PATH_MAX; | |
e780b0d1 | 1021 | buf = kmalloc(buf_len, GFP_KERNEL); |
31db9f7c AB |
1022 | if (!buf) { |
1023 | ret = -ENOMEM; | |
1024 | goto out; | |
1025 | } | |
1026 | ||
31db9f7c AB |
1027 | eb = path->nodes[0]; |
1028 | slot = path->slots[0]; | |
dd3cc16b | 1029 | item = btrfs_item_nr(slot); |
31db9f7c AB |
1030 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
1031 | cur = 0; | |
1032 | len = 0; | |
1033 | total = btrfs_item_size(eb, item); | |
1034 | ||
1035 | num = 0; | |
1036 | while (cur < total) { | |
1037 | name_len = btrfs_dir_name_len(eb, di); | |
1038 | data_len = btrfs_dir_data_len(eb, di); | |
1039 | type = btrfs_dir_type(eb, di); | |
1040 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1041 | ||
7e3ae33e FM |
1042 | if (type == BTRFS_FT_XATTR) { |
1043 | if (name_len > XATTR_NAME_MAX) { | |
1044 | ret = -ENAMETOOLONG; | |
1045 | goto out; | |
1046 | } | |
da17066c JM |
1047 | if (name_len + data_len > |
1048 | BTRFS_MAX_XATTR_SIZE(root->fs_info)) { | |
7e3ae33e FM |
1049 | ret = -E2BIG; |
1050 | goto out; | |
1051 | } | |
1052 | } else { | |
1053 | /* | |
1054 | * Path too long | |
1055 | */ | |
4395e0c4 | 1056 | if (name_len + data_len > PATH_MAX) { |
7e3ae33e FM |
1057 | ret = -ENAMETOOLONG; |
1058 | goto out; | |
1059 | } | |
31db9f7c AB |
1060 | } |
1061 | ||
59b0a7f2 SY |
1062 | ret = btrfs_is_name_len_valid(eb, path->slots[0], |
1063 | (unsigned long)(di + 1), name_len + data_len); | |
1064 | if (!ret) { | |
1065 | ret = -EIO; | |
1066 | goto out; | |
1067 | } | |
4395e0c4 FM |
1068 | if (name_len + data_len > buf_len) { |
1069 | buf_len = name_len + data_len; | |
1070 | if (is_vmalloc_addr(buf)) { | |
1071 | vfree(buf); | |
1072 | buf = NULL; | |
1073 | } else { | |
1074 | char *tmp = krealloc(buf, buf_len, | |
e780b0d1 | 1075 | GFP_KERNEL | __GFP_NOWARN); |
4395e0c4 FM |
1076 | |
1077 | if (!tmp) | |
1078 | kfree(buf); | |
1079 | buf = tmp; | |
1080 | } | |
1081 | if (!buf) { | |
f11f7441 | 1082 | buf = kvmalloc(buf_len, GFP_KERNEL); |
4395e0c4 FM |
1083 | if (!buf) { |
1084 | ret = -ENOMEM; | |
1085 | goto out; | |
1086 | } | |
1087 | } | |
1088 | } | |
1089 | ||
31db9f7c AB |
1090 | read_extent_buffer(eb, buf, (unsigned long)(di + 1), |
1091 | name_len + data_len); | |
1092 | ||
1093 | len = sizeof(*di) + name_len + data_len; | |
1094 | di = (struct btrfs_dir_item *)((char *)di + len); | |
1095 | cur += len; | |
1096 | ||
1097 | ret = iterate(num, &di_key, buf, name_len, buf + name_len, | |
1098 | data_len, type, ctx); | |
1099 | if (ret < 0) | |
1100 | goto out; | |
1101 | if (ret) { | |
1102 | ret = 0; | |
1103 | goto out; | |
1104 | } | |
1105 | ||
1106 | num++; | |
1107 | } | |
1108 | ||
1109 | out: | |
4395e0c4 | 1110 | kvfree(buf); |
31db9f7c AB |
1111 | return ret; |
1112 | } | |
1113 | ||
1114 | static int __copy_first_ref(int num, u64 dir, int index, | |
1115 | struct fs_path *p, void *ctx) | |
1116 | { | |
1117 | int ret; | |
1118 | struct fs_path *pt = ctx; | |
1119 | ||
1120 | ret = fs_path_copy(pt, p); | |
1121 | if (ret < 0) | |
1122 | return ret; | |
1123 | ||
1124 | /* we want the first only */ | |
1125 | return 1; | |
1126 | } | |
1127 | ||
1128 | /* | |
1129 | * Retrieve the first path of an inode. If an inode has more then one | |
1130 | * ref/hardlink, this is ignored. | |
1131 | */ | |
924794c9 | 1132 | static int get_inode_path(struct btrfs_root *root, |
31db9f7c AB |
1133 | u64 ino, struct fs_path *path) |
1134 | { | |
1135 | int ret; | |
1136 | struct btrfs_key key, found_key; | |
1137 | struct btrfs_path *p; | |
1138 | ||
1139 | p = alloc_path_for_send(); | |
1140 | if (!p) | |
1141 | return -ENOMEM; | |
1142 | ||
1143 | fs_path_reset(path); | |
1144 | ||
1145 | key.objectid = ino; | |
1146 | key.type = BTRFS_INODE_REF_KEY; | |
1147 | key.offset = 0; | |
1148 | ||
1149 | ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); | |
1150 | if (ret < 0) | |
1151 | goto out; | |
1152 | if (ret) { | |
1153 | ret = 1; | |
1154 | goto out; | |
1155 | } | |
1156 | btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); | |
1157 | if (found_key.objectid != ino || | |
96b5bd77 JS |
1158 | (found_key.type != BTRFS_INODE_REF_KEY && |
1159 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1160 | ret = -ENOENT; |
1161 | goto out; | |
1162 | } | |
1163 | ||
924794c9 TI |
1164 | ret = iterate_inode_ref(root, p, &found_key, 1, |
1165 | __copy_first_ref, path); | |
31db9f7c AB |
1166 | if (ret < 0) |
1167 | goto out; | |
1168 | ret = 0; | |
1169 | ||
1170 | out: | |
1171 | btrfs_free_path(p); | |
1172 | return ret; | |
1173 | } | |
1174 | ||
1175 | struct backref_ctx { | |
1176 | struct send_ctx *sctx; | |
1177 | ||
3f8a18cc | 1178 | struct btrfs_path *path; |
31db9f7c AB |
1179 | /* number of total found references */ |
1180 | u64 found; | |
1181 | ||
1182 | /* | |
1183 | * used for clones found in send_root. clones found behind cur_objectid | |
1184 | * and cur_offset are not considered as allowed clones. | |
1185 | */ | |
1186 | u64 cur_objectid; | |
1187 | u64 cur_offset; | |
1188 | ||
1189 | /* may be truncated in case it's the last extent in a file */ | |
1190 | u64 extent_len; | |
1191 | ||
619d8c4e FM |
1192 | /* data offset in the file extent item */ |
1193 | u64 data_offset; | |
1194 | ||
31db9f7c | 1195 | /* Just to check for bugs in backref resolving */ |
ee849c04 | 1196 | int found_itself; |
31db9f7c AB |
1197 | }; |
1198 | ||
1199 | static int __clone_root_cmp_bsearch(const void *key, const void *elt) | |
1200 | { | |
995e01b7 | 1201 | u64 root = (u64)(uintptr_t)key; |
31db9f7c AB |
1202 | struct clone_root *cr = (struct clone_root *)elt; |
1203 | ||
1204 | if (root < cr->root->objectid) | |
1205 | return -1; | |
1206 | if (root > cr->root->objectid) | |
1207 | return 1; | |
1208 | return 0; | |
1209 | } | |
1210 | ||
1211 | static int __clone_root_cmp_sort(const void *e1, const void *e2) | |
1212 | { | |
1213 | struct clone_root *cr1 = (struct clone_root *)e1; | |
1214 | struct clone_root *cr2 = (struct clone_root *)e2; | |
1215 | ||
1216 | if (cr1->root->objectid < cr2->root->objectid) | |
1217 | return -1; | |
1218 | if (cr1->root->objectid > cr2->root->objectid) | |
1219 | return 1; | |
1220 | return 0; | |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * Called for every backref that is found for the current extent. | |
766702ef | 1225 | * Results are collected in sctx->clone_roots->ino/offset/found_refs |
31db9f7c AB |
1226 | */ |
1227 | static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) | |
1228 | { | |
1229 | struct backref_ctx *bctx = ctx_; | |
1230 | struct clone_root *found; | |
1231 | int ret; | |
1232 | u64 i_size; | |
1233 | ||
1234 | /* First check if the root is in the list of accepted clone sources */ | |
995e01b7 | 1235 | found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, |
31db9f7c AB |
1236 | bctx->sctx->clone_roots_cnt, |
1237 | sizeof(struct clone_root), | |
1238 | __clone_root_cmp_bsearch); | |
1239 | if (!found) | |
1240 | return 0; | |
1241 | ||
1242 | if (found->root == bctx->sctx->send_root && | |
1243 | ino == bctx->cur_objectid && | |
1244 | offset == bctx->cur_offset) { | |
ee849c04 | 1245 | bctx->found_itself = 1; |
31db9f7c AB |
1246 | } |
1247 | ||
1248 | /* | |
766702ef | 1249 | * There are inodes that have extents that lie behind its i_size. Don't |
31db9f7c AB |
1250 | * accept clones from these extents. |
1251 | */ | |
3f8a18cc JB |
1252 | ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL, |
1253 | NULL, NULL, NULL); | |
1254 | btrfs_release_path(bctx->path); | |
31db9f7c AB |
1255 | if (ret < 0) |
1256 | return ret; | |
1257 | ||
619d8c4e | 1258 | if (offset + bctx->data_offset + bctx->extent_len > i_size) |
31db9f7c AB |
1259 | return 0; |
1260 | ||
1261 | /* | |
1262 | * Make sure we don't consider clones from send_root that are | |
1263 | * behind the current inode/offset. | |
1264 | */ | |
1265 | if (found->root == bctx->sctx->send_root) { | |
1266 | /* | |
1267 | * TODO for the moment we don't accept clones from the inode | |
1268 | * that is currently send. We may change this when | |
1269 | * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same | |
1270 | * file. | |
1271 | */ | |
1272 | if (ino >= bctx->cur_objectid) | |
1273 | return 0; | |
31db9f7c AB |
1274 | } |
1275 | ||
1276 | bctx->found++; | |
1277 | found->found_refs++; | |
1278 | if (ino < found->ino) { | |
1279 | found->ino = ino; | |
1280 | found->offset = offset; | |
1281 | } else if (found->ino == ino) { | |
1282 | /* | |
1283 | * same extent found more then once in the same file. | |
1284 | */ | |
1285 | if (found->offset > offset + bctx->extent_len) | |
1286 | found->offset = offset; | |
1287 | } | |
1288 | ||
1289 | return 0; | |
1290 | } | |
1291 | ||
1292 | /* | |
766702ef AB |
1293 | * Given an inode, offset and extent item, it finds a good clone for a clone |
1294 | * instruction. Returns -ENOENT when none could be found. The function makes | |
1295 | * sure that the returned clone is usable at the point where sending is at the | |
1296 | * moment. This means, that no clones are accepted which lie behind the current | |
1297 | * inode+offset. | |
1298 | * | |
31db9f7c AB |
1299 | * path must point to the extent item when called. |
1300 | */ | |
1301 | static int find_extent_clone(struct send_ctx *sctx, | |
1302 | struct btrfs_path *path, | |
1303 | u64 ino, u64 data_offset, | |
1304 | u64 ino_size, | |
1305 | struct clone_root **found) | |
1306 | { | |
04ab956e | 1307 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
1308 | int ret; |
1309 | int extent_type; | |
1310 | u64 logical; | |
74dd17fb | 1311 | u64 disk_byte; |
31db9f7c AB |
1312 | u64 num_bytes; |
1313 | u64 extent_item_pos; | |
69917e43 | 1314 | u64 flags = 0; |
31db9f7c AB |
1315 | struct btrfs_file_extent_item *fi; |
1316 | struct extent_buffer *eb = path->nodes[0]; | |
35075bb0 | 1317 | struct backref_ctx *backref_ctx = NULL; |
31db9f7c AB |
1318 | struct clone_root *cur_clone_root; |
1319 | struct btrfs_key found_key; | |
1320 | struct btrfs_path *tmp_path; | |
74dd17fb | 1321 | int compressed; |
31db9f7c AB |
1322 | u32 i; |
1323 | ||
1324 | tmp_path = alloc_path_for_send(); | |
1325 | if (!tmp_path) | |
1326 | return -ENOMEM; | |
1327 | ||
3f8a18cc JB |
1328 | /* We only use this path under the commit sem */ |
1329 | tmp_path->need_commit_sem = 0; | |
1330 | ||
e780b0d1 | 1331 | backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL); |
35075bb0 AB |
1332 | if (!backref_ctx) { |
1333 | ret = -ENOMEM; | |
1334 | goto out; | |
1335 | } | |
1336 | ||
3f8a18cc JB |
1337 | backref_ctx->path = tmp_path; |
1338 | ||
31db9f7c AB |
1339 | if (data_offset >= ino_size) { |
1340 | /* | |
1341 | * There may be extents that lie behind the file's size. | |
1342 | * I at least had this in combination with snapshotting while | |
1343 | * writing large files. | |
1344 | */ | |
1345 | ret = 0; | |
1346 | goto out; | |
1347 | } | |
1348 | ||
1349 | fi = btrfs_item_ptr(eb, path->slots[0], | |
1350 | struct btrfs_file_extent_item); | |
1351 | extent_type = btrfs_file_extent_type(eb, fi); | |
1352 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1353 | ret = -ENOENT; | |
1354 | goto out; | |
1355 | } | |
74dd17fb | 1356 | compressed = btrfs_file_extent_compression(eb, fi); |
31db9f7c AB |
1357 | |
1358 | num_bytes = btrfs_file_extent_num_bytes(eb, fi); | |
74dd17fb CM |
1359 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
1360 | if (disk_byte == 0) { | |
31db9f7c AB |
1361 | ret = -ENOENT; |
1362 | goto out; | |
1363 | } | |
74dd17fb | 1364 | logical = disk_byte + btrfs_file_extent_offset(eb, fi); |
31db9f7c | 1365 | |
04ab956e JM |
1366 | down_read(&fs_info->commit_root_sem); |
1367 | ret = extent_from_logical(fs_info, disk_byte, tmp_path, | |
69917e43 | 1368 | &found_key, &flags); |
04ab956e | 1369 | up_read(&fs_info->commit_root_sem); |
31db9f7c AB |
1370 | btrfs_release_path(tmp_path); |
1371 | ||
1372 | if (ret < 0) | |
1373 | goto out; | |
69917e43 | 1374 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
31db9f7c AB |
1375 | ret = -EIO; |
1376 | goto out; | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * Setup the clone roots. | |
1381 | */ | |
1382 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1383 | cur_clone_root = sctx->clone_roots + i; | |
1384 | cur_clone_root->ino = (u64)-1; | |
1385 | cur_clone_root->offset = 0; | |
1386 | cur_clone_root->found_refs = 0; | |
1387 | } | |
1388 | ||
35075bb0 AB |
1389 | backref_ctx->sctx = sctx; |
1390 | backref_ctx->found = 0; | |
1391 | backref_ctx->cur_objectid = ino; | |
1392 | backref_ctx->cur_offset = data_offset; | |
1393 | backref_ctx->found_itself = 0; | |
1394 | backref_ctx->extent_len = num_bytes; | |
619d8c4e FM |
1395 | /* |
1396 | * For non-compressed extents iterate_extent_inodes() gives us extent | |
1397 | * offsets that already take into account the data offset, but not for | |
1398 | * compressed extents, since the offset is logical and not relative to | |
1399 | * the physical extent locations. We must take this into account to | |
1400 | * avoid sending clone offsets that go beyond the source file's size, | |
1401 | * which would result in the clone ioctl failing with -EINVAL on the | |
1402 | * receiving end. | |
1403 | */ | |
1404 | if (compressed == BTRFS_COMPRESS_NONE) | |
1405 | backref_ctx->data_offset = 0; | |
1406 | else | |
1407 | backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi); | |
31db9f7c AB |
1408 | |
1409 | /* | |
1410 | * The last extent of a file may be too large due to page alignment. | |
1411 | * We need to adjust extent_len in this case so that the checks in | |
1412 | * __iterate_backrefs work. | |
1413 | */ | |
1414 | if (data_offset + num_bytes >= ino_size) | |
35075bb0 | 1415 | backref_ctx->extent_len = ino_size - data_offset; |
31db9f7c AB |
1416 | |
1417 | /* | |
1418 | * Now collect all backrefs. | |
1419 | */ | |
74dd17fb CM |
1420 | if (compressed == BTRFS_COMPRESS_NONE) |
1421 | extent_item_pos = logical - found_key.objectid; | |
1422 | else | |
1423 | extent_item_pos = 0; | |
0b246afa JM |
1424 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
1425 | extent_item_pos, 1, __iterate_backrefs, | |
c995ab3c | 1426 | backref_ctx, false); |
74dd17fb | 1427 | |
31db9f7c AB |
1428 | if (ret < 0) |
1429 | goto out; | |
1430 | ||
35075bb0 | 1431 | if (!backref_ctx->found_itself) { |
31db9f7c AB |
1432 | /* found a bug in backref code? */ |
1433 | ret = -EIO; | |
04ab956e | 1434 | btrfs_err(fs_info, |
5d163e0e | 1435 | "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu", |
04ab956e | 1436 | ino, data_offset, disk_byte, found_key.objectid); |
31db9f7c AB |
1437 | goto out; |
1438 | } | |
1439 | ||
04ab956e JM |
1440 | btrfs_debug(fs_info, |
1441 | "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu", | |
1442 | data_offset, ino, num_bytes, logical); | |
31db9f7c | 1443 | |
35075bb0 | 1444 | if (!backref_ctx->found) |
04ab956e | 1445 | btrfs_debug(fs_info, "no clones found"); |
31db9f7c AB |
1446 | |
1447 | cur_clone_root = NULL; | |
1448 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1449 | if (sctx->clone_roots[i].found_refs) { | |
1450 | if (!cur_clone_root) | |
1451 | cur_clone_root = sctx->clone_roots + i; | |
1452 | else if (sctx->clone_roots[i].root == sctx->send_root) | |
1453 | /* prefer clones from send_root over others */ | |
1454 | cur_clone_root = sctx->clone_roots + i; | |
31db9f7c AB |
1455 | } |
1456 | ||
1457 | } | |
1458 | ||
1459 | if (cur_clone_root) { | |
1460 | *found = cur_clone_root; | |
1461 | ret = 0; | |
1462 | } else { | |
1463 | ret = -ENOENT; | |
1464 | } | |
1465 | ||
1466 | out: | |
1467 | btrfs_free_path(tmp_path); | |
35075bb0 | 1468 | kfree(backref_ctx); |
31db9f7c AB |
1469 | return ret; |
1470 | } | |
1471 | ||
924794c9 | 1472 | static int read_symlink(struct btrfs_root *root, |
31db9f7c AB |
1473 | u64 ino, |
1474 | struct fs_path *dest) | |
1475 | { | |
1476 | int ret; | |
1477 | struct btrfs_path *path; | |
1478 | struct btrfs_key key; | |
1479 | struct btrfs_file_extent_item *ei; | |
1480 | u8 type; | |
1481 | u8 compression; | |
1482 | unsigned long off; | |
1483 | int len; | |
1484 | ||
1485 | path = alloc_path_for_send(); | |
1486 | if (!path) | |
1487 | return -ENOMEM; | |
1488 | ||
1489 | key.objectid = ino; | |
1490 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1491 | key.offset = 0; | |
1492 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1493 | if (ret < 0) | |
1494 | goto out; | |
a879719b FM |
1495 | if (ret) { |
1496 | /* | |
1497 | * An empty symlink inode. Can happen in rare error paths when | |
1498 | * creating a symlink (transaction committed before the inode | |
1499 | * eviction handler removed the symlink inode items and a crash | |
1500 | * happened in between or the subvol was snapshoted in between). | |
1501 | * Print an informative message to dmesg/syslog so that the user | |
1502 | * can delete the symlink. | |
1503 | */ | |
1504 | btrfs_err(root->fs_info, | |
1505 | "Found empty symlink inode %llu at root %llu", | |
1506 | ino, root->root_key.objectid); | |
1507 | ret = -EIO; | |
1508 | goto out; | |
1509 | } | |
31db9f7c AB |
1510 | |
1511 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1512 | struct btrfs_file_extent_item); | |
1513 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
1514 | compression = btrfs_file_extent_compression(path->nodes[0], ei); | |
1515 | BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); | |
1516 | BUG_ON(compression); | |
1517 | ||
1518 | off = btrfs_file_extent_inline_start(ei); | |
514ac8ad | 1519 | len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei); |
31db9f7c AB |
1520 | |
1521 | ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); | |
31db9f7c AB |
1522 | |
1523 | out: | |
1524 | btrfs_free_path(path); | |
1525 | return ret; | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * Helper function to generate a file name that is unique in the root of | |
1530 | * send_root and parent_root. This is used to generate names for orphan inodes. | |
1531 | */ | |
1532 | static int gen_unique_name(struct send_ctx *sctx, | |
1533 | u64 ino, u64 gen, | |
1534 | struct fs_path *dest) | |
1535 | { | |
1536 | int ret = 0; | |
1537 | struct btrfs_path *path; | |
1538 | struct btrfs_dir_item *di; | |
1539 | char tmp[64]; | |
1540 | int len; | |
1541 | u64 idx = 0; | |
1542 | ||
1543 | path = alloc_path_for_send(); | |
1544 | if (!path) | |
1545 | return -ENOMEM; | |
1546 | ||
1547 | while (1) { | |
f74b86d8 | 1548 | len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", |
31db9f7c | 1549 | ino, gen, idx); |
64792f25 | 1550 | ASSERT(len < sizeof(tmp)); |
31db9f7c AB |
1551 | |
1552 | di = btrfs_lookup_dir_item(NULL, sctx->send_root, | |
1553 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1554 | tmp, strlen(tmp), 0); | |
1555 | btrfs_release_path(path); | |
1556 | if (IS_ERR(di)) { | |
1557 | ret = PTR_ERR(di); | |
1558 | goto out; | |
1559 | } | |
1560 | if (di) { | |
1561 | /* not unique, try again */ | |
1562 | idx++; | |
1563 | continue; | |
1564 | } | |
1565 | ||
1566 | if (!sctx->parent_root) { | |
1567 | /* unique */ | |
1568 | ret = 0; | |
1569 | break; | |
1570 | } | |
1571 | ||
1572 | di = btrfs_lookup_dir_item(NULL, sctx->parent_root, | |
1573 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1574 | tmp, strlen(tmp), 0); | |
1575 | btrfs_release_path(path); | |
1576 | if (IS_ERR(di)) { | |
1577 | ret = PTR_ERR(di); | |
1578 | goto out; | |
1579 | } | |
1580 | if (di) { | |
1581 | /* not unique, try again */ | |
1582 | idx++; | |
1583 | continue; | |
1584 | } | |
1585 | /* unique */ | |
1586 | break; | |
1587 | } | |
1588 | ||
1589 | ret = fs_path_add(dest, tmp, strlen(tmp)); | |
1590 | ||
1591 | out: | |
1592 | btrfs_free_path(path); | |
1593 | return ret; | |
1594 | } | |
1595 | ||
1596 | enum inode_state { | |
1597 | inode_state_no_change, | |
1598 | inode_state_will_create, | |
1599 | inode_state_did_create, | |
1600 | inode_state_will_delete, | |
1601 | inode_state_did_delete, | |
1602 | }; | |
1603 | ||
1604 | static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) | |
1605 | { | |
1606 | int ret; | |
1607 | int left_ret; | |
1608 | int right_ret; | |
1609 | u64 left_gen; | |
1610 | u64 right_gen; | |
1611 | ||
1612 | ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, | |
85a7b33b | 1613 | NULL, NULL); |
31db9f7c AB |
1614 | if (ret < 0 && ret != -ENOENT) |
1615 | goto out; | |
1616 | left_ret = ret; | |
1617 | ||
1618 | if (!sctx->parent_root) { | |
1619 | right_ret = -ENOENT; | |
1620 | } else { | |
1621 | ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, | |
85a7b33b | 1622 | NULL, NULL, NULL, NULL); |
31db9f7c AB |
1623 | if (ret < 0 && ret != -ENOENT) |
1624 | goto out; | |
1625 | right_ret = ret; | |
1626 | } | |
1627 | ||
1628 | if (!left_ret && !right_ret) { | |
e938c8ad | 1629 | if (left_gen == gen && right_gen == gen) { |
31db9f7c | 1630 | ret = inode_state_no_change; |
e938c8ad | 1631 | } else if (left_gen == gen) { |
31db9f7c AB |
1632 | if (ino < sctx->send_progress) |
1633 | ret = inode_state_did_create; | |
1634 | else | |
1635 | ret = inode_state_will_create; | |
1636 | } else if (right_gen == gen) { | |
1637 | if (ino < sctx->send_progress) | |
1638 | ret = inode_state_did_delete; | |
1639 | else | |
1640 | ret = inode_state_will_delete; | |
1641 | } else { | |
1642 | ret = -ENOENT; | |
1643 | } | |
1644 | } else if (!left_ret) { | |
1645 | if (left_gen == gen) { | |
1646 | if (ino < sctx->send_progress) | |
1647 | ret = inode_state_did_create; | |
1648 | else | |
1649 | ret = inode_state_will_create; | |
1650 | } else { | |
1651 | ret = -ENOENT; | |
1652 | } | |
1653 | } else if (!right_ret) { | |
1654 | if (right_gen == gen) { | |
1655 | if (ino < sctx->send_progress) | |
1656 | ret = inode_state_did_delete; | |
1657 | else | |
1658 | ret = inode_state_will_delete; | |
1659 | } else { | |
1660 | ret = -ENOENT; | |
1661 | } | |
1662 | } else { | |
1663 | ret = -ENOENT; | |
1664 | } | |
1665 | ||
1666 | out: | |
1667 | return ret; | |
1668 | } | |
1669 | ||
1670 | static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) | |
1671 | { | |
1672 | int ret; | |
1673 | ||
4dd9920d RK |
1674 | if (ino == BTRFS_FIRST_FREE_OBJECTID) |
1675 | return 1; | |
1676 | ||
31db9f7c AB |
1677 | ret = get_cur_inode_state(sctx, ino, gen); |
1678 | if (ret < 0) | |
1679 | goto out; | |
1680 | ||
1681 | if (ret == inode_state_no_change || | |
1682 | ret == inode_state_did_create || | |
1683 | ret == inode_state_will_delete) | |
1684 | ret = 1; | |
1685 | else | |
1686 | ret = 0; | |
1687 | ||
1688 | out: | |
1689 | return ret; | |
1690 | } | |
1691 | ||
1692 | /* | |
1693 | * Helper function to lookup a dir item in a dir. | |
1694 | */ | |
1695 | static int lookup_dir_item_inode(struct btrfs_root *root, | |
1696 | u64 dir, const char *name, int name_len, | |
1697 | u64 *found_inode, | |
1698 | u8 *found_type) | |
1699 | { | |
1700 | int ret = 0; | |
1701 | struct btrfs_dir_item *di; | |
1702 | struct btrfs_key key; | |
1703 | struct btrfs_path *path; | |
1704 | ||
1705 | path = alloc_path_for_send(); | |
1706 | if (!path) | |
1707 | return -ENOMEM; | |
1708 | ||
1709 | di = btrfs_lookup_dir_item(NULL, root, path, | |
1710 | dir, name, name_len, 0); | |
1711 | if (!di) { | |
1712 | ret = -ENOENT; | |
1713 | goto out; | |
1714 | } | |
1715 | if (IS_ERR(di)) { | |
1716 | ret = PTR_ERR(di); | |
1717 | goto out; | |
1718 | } | |
1719 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
1af56070 FM |
1720 | if (key.type == BTRFS_ROOT_ITEM_KEY) { |
1721 | ret = -ENOENT; | |
1722 | goto out; | |
1723 | } | |
31db9f7c AB |
1724 | *found_inode = key.objectid; |
1725 | *found_type = btrfs_dir_type(path->nodes[0], di); | |
1726 | ||
1727 | out: | |
1728 | btrfs_free_path(path); | |
1729 | return ret; | |
1730 | } | |
1731 | ||
766702ef AB |
1732 | /* |
1733 | * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, | |
1734 | * generation of the parent dir and the name of the dir entry. | |
1735 | */ | |
924794c9 | 1736 | static int get_first_ref(struct btrfs_root *root, u64 ino, |
31db9f7c AB |
1737 | u64 *dir, u64 *dir_gen, struct fs_path *name) |
1738 | { | |
1739 | int ret; | |
1740 | struct btrfs_key key; | |
1741 | struct btrfs_key found_key; | |
1742 | struct btrfs_path *path; | |
31db9f7c | 1743 | int len; |
96b5bd77 | 1744 | u64 parent_dir; |
31db9f7c AB |
1745 | |
1746 | path = alloc_path_for_send(); | |
1747 | if (!path) | |
1748 | return -ENOMEM; | |
1749 | ||
1750 | key.objectid = ino; | |
1751 | key.type = BTRFS_INODE_REF_KEY; | |
1752 | key.offset = 0; | |
1753 | ||
1754 | ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); | |
1755 | if (ret < 0) | |
1756 | goto out; | |
1757 | if (!ret) | |
1758 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1759 | path->slots[0]); | |
96b5bd77 JS |
1760 | if (ret || found_key.objectid != ino || |
1761 | (found_key.type != BTRFS_INODE_REF_KEY && | |
1762 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1763 | ret = -ENOENT; |
1764 | goto out; | |
1765 | } | |
1766 | ||
51a60253 | 1767 | if (found_key.type == BTRFS_INODE_REF_KEY) { |
96b5bd77 JS |
1768 | struct btrfs_inode_ref *iref; |
1769 | iref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1770 | struct btrfs_inode_ref); | |
1771 | len = btrfs_inode_ref_name_len(path->nodes[0], iref); | |
1772 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1773 | (unsigned long)(iref + 1), | |
1774 | len); | |
1775 | parent_dir = found_key.offset; | |
1776 | } else { | |
1777 | struct btrfs_inode_extref *extref; | |
1778 | extref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1779 | struct btrfs_inode_extref); | |
1780 | len = btrfs_inode_extref_name_len(path->nodes[0], extref); | |
1781 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1782 | (unsigned long)&extref->name, len); | |
1783 | parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); | |
1784 | } | |
31db9f7c AB |
1785 | if (ret < 0) |
1786 | goto out; | |
1787 | btrfs_release_path(path); | |
1788 | ||
b46ab97b FM |
1789 | if (dir_gen) { |
1790 | ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, | |
1791 | NULL, NULL, NULL); | |
1792 | if (ret < 0) | |
1793 | goto out; | |
1794 | } | |
31db9f7c | 1795 | |
96b5bd77 | 1796 | *dir = parent_dir; |
31db9f7c AB |
1797 | |
1798 | out: | |
1799 | btrfs_free_path(path); | |
1800 | return ret; | |
1801 | } | |
1802 | ||
924794c9 | 1803 | static int is_first_ref(struct btrfs_root *root, |
31db9f7c AB |
1804 | u64 ino, u64 dir, |
1805 | const char *name, int name_len) | |
1806 | { | |
1807 | int ret; | |
1808 | struct fs_path *tmp_name; | |
1809 | u64 tmp_dir; | |
31db9f7c | 1810 | |
924794c9 | 1811 | tmp_name = fs_path_alloc(); |
31db9f7c AB |
1812 | if (!tmp_name) |
1813 | return -ENOMEM; | |
1814 | ||
b46ab97b | 1815 | ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name); |
31db9f7c AB |
1816 | if (ret < 0) |
1817 | goto out; | |
1818 | ||
b9291aff | 1819 | if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { |
31db9f7c AB |
1820 | ret = 0; |
1821 | goto out; | |
1822 | } | |
1823 | ||
e938c8ad | 1824 | ret = !memcmp(tmp_name->start, name, name_len); |
31db9f7c AB |
1825 | |
1826 | out: | |
924794c9 | 1827 | fs_path_free(tmp_name); |
31db9f7c AB |
1828 | return ret; |
1829 | } | |
1830 | ||
766702ef AB |
1831 | /* |
1832 | * Used by process_recorded_refs to determine if a new ref would overwrite an | |
1833 | * already existing ref. In case it detects an overwrite, it returns the | |
1834 | * inode/gen in who_ino/who_gen. | |
1835 | * When an overwrite is detected, process_recorded_refs does proper orphanizing | |
1836 | * to make sure later references to the overwritten inode are possible. | |
1837 | * Orphanizing is however only required for the first ref of an inode. | |
1838 | * process_recorded_refs does an additional is_first_ref check to see if | |
1839 | * orphanizing is really required. | |
1840 | */ | |
31db9f7c AB |
1841 | static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
1842 | const char *name, int name_len, | |
f5962781 | 1843 | u64 *who_ino, u64 *who_gen, u64 *who_mode) |
31db9f7c AB |
1844 | { |
1845 | int ret = 0; | |
ebdad913 | 1846 | u64 gen; |
31db9f7c AB |
1847 | u64 other_inode = 0; |
1848 | u8 other_type = 0; | |
1849 | ||
1850 | if (!sctx->parent_root) | |
1851 | goto out; | |
1852 | ||
1853 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1854 | if (ret <= 0) | |
1855 | goto out; | |
1856 | ||
ebdad913 JB |
1857 | /* |
1858 | * If we have a parent root we need to verify that the parent dir was | |
01327610 | 1859 | * not deleted and then re-created, if it was then we have no overwrite |
ebdad913 JB |
1860 | * and we can just unlink this entry. |
1861 | */ | |
4dd9920d | 1862 | if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) { |
ebdad913 JB |
1863 | ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, |
1864 | NULL, NULL, NULL); | |
1865 | if (ret < 0 && ret != -ENOENT) | |
1866 | goto out; | |
1867 | if (ret) { | |
1868 | ret = 0; | |
1869 | goto out; | |
1870 | } | |
1871 | if (gen != dir_gen) | |
1872 | goto out; | |
1873 | } | |
1874 | ||
31db9f7c AB |
1875 | ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, |
1876 | &other_inode, &other_type); | |
1877 | if (ret < 0 && ret != -ENOENT) | |
1878 | goto out; | |
1879 | if (ret) { | |
1880 | ret = 0; | |
1881 | goto out; | |
1882 | } | |
1883 | ||
766702ef AB |
1884 | /* |
1885 | * Check if the overwritten ref was already processed. If yes, the ref | |
1886 | * was already unlinked/moved, so we can safely assume that we will not | |
1887 | * overwrite anything at this point in time. | |
1888 | */ | |
801bec36 RK |
1889 | if (other_inode > sctx->send_progress || |
1890 | is_waiting_for_move(sctx, other_inode)) { | |
31db9f7c | 1891 | ret = get_inode_info(sctx->parent_root, other_inode, NULL, |
f5962781 | 1892 | who_gen, who_mode, NULL, NULL, NULL); |
31db9f7c AB |
1893 | if (ret < 0) |
1894 | goto out; | |
1895 | ||
1896 | ret = 1; | |
1897 | *who_ino = other_inode; | |
1898 | } else { | |
1899 | ret = 0; | |
1900 | } | |
1901 | ||
1902 | out: | |
1903 | return ret; | |
1904 | } | |
1905 | ||
766702ef AB |
1906 | /* |
1907 | * Checks if the ref was overwritten by an already processed inode. This is | |
1908 | * used by __get_cur_name_and_parent to find out if the ref was orphanized and | |
1909 | * thus the orphan name needs be used. | |
1910 | * process_recorded_refs also uses it to avoid unlinking of refs that were | |
1911 | * overwritten. | |
1912 | */ | |
31db9f7c AB |
1913 | static int did_overwrite_ref(struct send_ctx *sctx, |
1914 | u64 dir, u64 dir_gen, | |
1915 | u64 ino, u64 ino_gen, | |
1916 | const char *name, int name_len) | |
1917 | { | |
1918 | int ret = 0; | |
1919 | u64 gen; | |
1920 | u64 ow_inode; | |
1921 | u8 other_type; | |
1922 | ||
1923 | if (!sctx->parent_root) | |
1924 | goto out; | |
1925 | ||
1926 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1927 | if (ret <= 0) | |
1928 | goto out; | |
1929 | ||
01914101 RK |
1930 | if (dir != BTRFS_FIRST_FREE_OBJECTID) { |
1931 | ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, | |
1932 | NULL, NULL, NULL); | |
1933 | if (ret < 0 && ret != -ENOENT) | |
1934 | goto out; | |
1935 | if (ret) { | |
1936 | ret = 0; | |
1937 | goto out; | |
1938 | } | |
1939 | if (gen != dir_gen) | |
1940 | goto out; | |
1941 | } | |
1942 | ||
31db9f7c AB |
1943 | /* check if the ref was overwritten by another ref */ |
1944 | ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, | |
1945 | &ow_inode, &other_type); | |
1946 | if (ret < 0 && ret != -ENOENT) | |
1947 | goto out; | |
1948 | if (ret) { | |
1949 | /* was never and will never be overwritten */ | |
1950 | ret = 0; | |
1951 | goto out; | |
1952 | } | |
1953 | ||
1954 | ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, | |
85a7b33b | 1955 | NULL, NULL); |
31db9f7c AB |
1956 | if (ret < 0) |
1957 | goto out; | |
1958 | ||
1959 | if (ow_inode == ino && gen == ino_gen) { | |
1960 | ret = 0; | |
1961 | goto out; | |
1962 | } | |
1963 | ||
8b191a68 FM |
1964 | /* |
1965 | * We know that it is or will be overwritten. Check this now. | |
1966 | * The current inode being processed might have been the one that caused | |
b786f16a FM |
1967 | * inode 'ino' to be orphanized, therefore check if ow_inode matches |
1968 | * the current inode being processed. | |
8b191a68 | 1969 | */ |
b786f16a FM |
1970 | if ((ow_inode < sctx->send_progress) || |
1971 | (ino != sctx->cur_ino && ow_inode == sctx->cur_ino && | |
1972 | gen == sctx->cur_inode_gen)) | |
31db9f7c AB |
1973 | ret = 1; |
1974 | else | |
1975 | ret = 0; | |
1976 | ||
1977 | out: | |
1978 | return ret; | |
1979 | } | |
1980 | ||
766702ef AB |
1981 | /* |
1982 | * Same as did_overwrite_ref, but also checks if it is the first ref of an inode | |
1983 | * that got overwritten. This is used by process_recorded_refs to determine | |
1984 | * if it has to use the path as returned by get_cur_path or the orphan name. | |
1985 | */ | |
31db9f7c AB |
1986 | static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) |
1987 | { | |
1988 | int ret = 0; | |
1989 | struct fs_path *name = NULL; | |
1990 | u64 dir; | |
1991 | u64 dir_gen; | |
1992 | ||
1993 | if (!sctx->parent_root) | |
1994 | goto out; | |
1995 | ||
924794c9 | 1996 | name = fs_path_alloc(); |
31db9f7c AB |
1997 | if (!name) |
1998 | return -ENOMEM; | |
1999 | ||
924794c9 | 2000 | ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); |
31db9f7c AB |
2001 | if (ret < 0) |
2002 | goto out; | |
2003 | ||
2004 | ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, | |
2005 | name->start, fs_path_len(name)); | |
31db9f7c AB |
2006 | |
2007 | out: | |
924794c9 | 2008 | fs_path_free(name); |
31db9f7c AB |
2009 | return ret; |
2010 | } | |
2011 | ||
766702ef AB |
2012 | /* |
2013 | * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, | |
2014 | * so we need to do some special handling in case we have clashes. This function | |
2015 | * takes care of this with the help of name_cache_entry::radix_list. | |
5dc67d0b | 2016 | * In case of error, nce is kfreed. |
766702ef | 2017 | */ |
31db9f7c AB |
2018 | static int name_cache_insert(struct send_ctx *sctx, |
2019 | struct name_cache_entry *nce) | |
2020 | { | |
2021 | int ret = 0; | |
7e0926fe AB |
2022 | struct list_head *nce_head; |
2023 | ||
2024 | nce_head = radix_tree_lookup(&sctx->name_cache, | |
2025 | (unsigned long)nce->ino); | |
2026 | if (!nce_head) { | |
e780b0d1 | 2027 | nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL); |
cfa7a9cc TI |
2028 | if (!nce_head) { |
2029 | kfree(nce); | |
31db9f7c | 2030 | return -ENOMEM; |
cfa7a9cc | 2031 | } |
7e0926fe | 2032 | INIT_LIST_HEAD(nce_head); |
31db9f7c | 2033 | |
7e0926fe | 2034 | ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); |
5dc67d0b AB |
2035 | if (ret < 0) { |
2036 | kfree(nce_head); | |
2037 | kfree(nce); | |
31db9f7c | 2038 | return ret; |
5dc67d0b | 2039 | } |
31db9f7c | 2040 | } |
7e0926fe | 2041 | list_add_tail(&nce->radix_list, nce_head); |
31db9f7c AB |
2042 | list_add_tail(&nce->list, &sctx->name_cache_list); |
2043 | sctx->name_cache_size++; | |
2044 | ||
2045 | return ret; | |
2046 | } | |
2047 | ||
2048 | static void name_cache_delete(struct send_ctx *sctx, | |
2049 | struct name_cache_entry *nce) | |
2050 | { | |
7e0926fe | 2051 | struct list_head *nce_head; |
31db9f7c | 2052 | |
7e0926fe AB |
2053 | nce_head = radix_tree_lookup(&sctx->name_cache, |
2054 | (unsigned long)nce->ino); | |
57fb8910 DS |
2055 | if (!nce_head) { |
2056 | btrfs_err(sctx->send_root->fs_info, | |
2057 | "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", | |
2058 | nce->ino, sctx->name_cache_size); | |
2059 | } | |
31db9f7c | 2060 | |
7e0926fe | 2061 | list_del(&nce->radix_list); |
31db9f7c | 2062 | list_del(&nce->list); |
31db9f7c | 2063 | sctx->name_cache_size--; |
7e0926fe | 2064 | |
57fb8910 DS |
2065 | /* |
2066 | * We may not get to the final release of nce_head if the lookup fails | |
2067 | */ | |
2068 | if (nce_head && list_empty(nce_head)) { | |
7e0926fe AB |
2069 | radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); |
2070 | kfree(nce_head); | |
2071 | } | |
31db9f7c AB |
2072 | } |
2073 | ||
2074 | static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, | |
2075 | u64 ino, u64 gen) | |
2076 | { | |
7e0926fe AB |
2077 | struct list_head *nce_head; |
2078 | struct name_cache_entry *cur; | |
31db9f7c | 2079 | |
7e0926fe AB |
2080 | nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); |
2081 | if (!nce_head) | |
31db9f7c AB |
2082 | return NULL; |
2083 | ||
7e0926fe AB |
2084 | list_for_each_entry(cur, nce_head, radix_list) { |
2085 | if (cur->ino == ino && cur->gen == gen) | |
2086 | return cur; | |
2087 | } | |
31db9f7c AB |
2088 | return NULL; |
2089 | } | |
2090 | ||
766702ef AB |
2091 | /* |
2092 | * Removes the entry from the list and adds it back to the end. This marks the | |
2093 | * entry as recently used so that name_cache_clean_unused does not remove it. | |
2094 | */ | |
31db9f7c AB |
2095 | static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) |
2096 | { | |
2097 | list_del(&nce->list); | |
2098 | list_add_tail(&nce->list, &sctx->name_cache_list); | |
2099 | } | |
2100 | ||
766702ef AB |
2101 | /* |
2102 | * Remove some entries from the beginning of name_cache_list. | |
2103 | */ | |
31db9f7c AB |
2104 | static void name_cache_clean_unused(struct send_ctx *sctx) |
2105 | { | |
2106 | struct name_cache_entry *nce; | |
2107 | ||
2108 | if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) | |
2109 | return; | |
2110 | ||
2111 | while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { | |
2112 | nce = list_entry(sctx->name_cache_list.next, | |
2113 | struct name_cache_entry, list); | |
2114 | name_cache_delete(sctx, nce); | |
2115 | kfree(nce); | |
2116 | } | |
2117 | } | |
2118 | ||
2119 | static void name_cache_free(struct send_ctx *sctx) | |
2120 | { | |
2121 | struct name_cache_entry *nce; | |
31db9f7c | 2122 | |
e938c8ad AB |
2123 | while (!list_empty(&sctx->name_cache_list)) { |
2124 | nce = list_entry(sctx->name_cache_list.next, | |
2125 | struct name_cache_entry, list); | |
31db9f7c | 2126 | name_cache_delete(sctx, nce); |
17589bd9 | 2127 | kfree(nce); |
31db9f7c AB |
2128 | } |
2129 | } | |
2130 | ||
766702ef AB |
2131 | /* |
2132 | * Used by get_cur_path for each ref up to the root. | |
2133 | * Returns 0 if it succeeded. | |
2134 | * Returns 1 if the inode is not existent or got overwritten. In that case, the | |
2135 | * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 | |
2136 | * is returned, parent_ino/parent_gen are not guaranteed to be valid. | |
2137 | * Returns <0 in case of error. | |
2138 | */ | |
31db9f7c AB |
2139 | static int __get_cur_name_and_parent(struct send_ctx *sctx, |
2140 | u64 ino, u64 gen, | |
2141 | u64 *parent_ino, | |
2142 | u64 *parent_gen, | |
2143 | struct fs_path *dest) | |
2144 | { | |
2145 | int ret; | |
2146 | int nce_ret; | |
31db9f7c AB |
2147 | struct name_cache_entry *nce = NULL; |
2148 | ||
766702ef AB |
2149 | /* |
2150 | * First check if we already did a call to this function with the same | |
2151 | * ino/gen. If yes, check if the cache entry is still up-to-date. If yes | |
2152 | * return the cached result. | |
2153 | */ | |
31db9f7c AB |
2154 | nce = name_cache_search(sctx, ino, gen); |
2155 | if (nce) { | |
2156 | if (ino < sctx->send_progress && nce->need_later_update) { | |
2157 | name_cache_delete(sctx, nce); | |
2158 | kfree(nce); | |
2159 | nce = NULL; | |
2160 | } else { | |
2161 | name_cache_used(sctx, nce); | |
2162 | *parent_ino = nce->parent_ino; | |
2163 | *parent_gen = nce->parent_gen; | |
2164 | ret = fs_path_add(dest, nce->name, nce->name_len); | |
2165 | if (ret < 0) | |
2166 | goto out; | |
2167 | ret = nce->ret; | |
2168 | goto out; | |
2169 | } | |
2170 | } | |
2171 | ||
766702ef AB |
2172 | /* |
2173 | * If the inode is not existent yet, add the orphan name and return 1. | |
2174 | * This should only happen for the parent dir that we determine in | |
2175 | * __record_new_ref | |
2176 | */ | |
31db9f7c AB |
2177 | ret = is_inode_existent(sctx, ino, gen); |
2178 | if (ret < 0) | |
2179 | goto out; | |
2180 | ||
2181 | if (!ret) { | |
2182 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2183 | if (ret < 0) | |
2184 | goto out; | |
2185 | ret = 1; | |
2186 | goto out_cache; | |
2187 | } | |
2188 | ||
766702ef AB |
2189 | /* |
2190 | * Depending on whether the inode was already processed or not, use | |
2191 | * send_root or parent_root for ref lookup. | |
2192 | */ | |
bf0d1f44 | 2193 | if (ino < sctx->send_progress) |
924794c9 TI |
2194 | ret = get_first_ref(sctx->send_root, ino, |
2195 | parent_ino, parent_gen, dest); | |
31db9f7c | 2196 | else |
924794c9 TI |
2197 | ret = get_first_ref(sctx->parent_root, ino, |
2198 | parent_ino, parent_gen, dest); | |
31db9f7c AB |
2199 | if (ret < 0) |
2200 | goto out; | |
2201 | ||
766702ef AB |
2202 | /* |
2203 | * Check if the ref was overwritten by an inode's ref that was processed | |
2204 | * earlier. If yes, treat as orphan and return 1. | |
2205 | */ | |
31db9f7c AB |
2206 | ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, |
2207 | dest->start, dest->end - dest->start); | |
2208 | if (ret < 0) | |
2209 | goto out; | |
2210 | if (ret) { | |
2211 | fs_path_reset(dest); | |
2212 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2213 | if (ret < 0) | |
2214 | goto out; | |
2215 | ret = 1; | |
2216 | } | |
2217 | ||
2218 | out_cache: | |
766702ef AB |
2219 | /* |
2220 | * Store the result of the lookup in the name cache. | |
2221 | */ | |
e780b0d1 | 2222 | nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); |
31db9f7c AB |
2223 | if (!nce) { |
2224 | ret = -ENOMEM; | |
2225 | goto out; | |
2226 | } | |
2227 | ||
2228 | nce->ino = ino; | |
2229 | nce->gen = gen; | |
2230 | nce->parent_ino = *parent_ino; | |
2231 | nce->parent_gen = *parent_gen; | |
2232 | nce->name_len = fs_path_len(dest); | |
2233 | nce->ret = ret; | |
2234 | strcpy(nce->name, dest->start); | |
31db9f7c AB |
2235 | |
2236 | if (ino < sctx->send_progress) | |
2237 | nce->need_later_update = 0; | |
2238 | else | |
2239 | nce->need_later_update = 1; | |
2240 | ||
2241 | nce_ret = name_cache_insert(sctx, nce); | |
2242 | if (nce_ret < 0) | |
2243 | ret = nce_ret; | |
2244 | name_cache_clean_unused(sctx); | |
2245 | ||
2246 | out: | |
31db9f7c AB |
2247 | return ret; |
2248 | } | |
2249 | ||
2250 | /* | |
2251 | * Magic happens here. This function returns the first ref to an inode as it | |
2252 | * would look like while receiving the stream at this point in time. | |
2253 | * We walk the path up to the root. For every inode in between, we check if it | |
2254 | * was already processed/sent. If yes, we continue with the parent as found | |
2255 | * in send_root. If not, we continue with the parent as found in parent_root. | |
2256 | * If we encounter an inode that was deleted at this point in time, we use the | |
2257 | * inodes "orphan" name instead of the real name and stop. Same with new inodes | |
2258 | * that were not created yet and overwritten inodes/refs. | |
2259 | * | |
2260 | * When do we have have orphan inodes: | |
2261 | * 1. When an inode is freshly created and thus no valid refs are available yet | |
2262 | * 2. When a directory lost all it's refs (deleted) but still has dir items | |
2263 | * inside which were not processed yet (pending for move/delete). If anyone | |
2264 | * tried to get the path to the dir items, it would get a path inside that | |
2265 | * orphan directory. | |
2266 | * 3. When an inode is moved around or gets new links, it may overwrite the ref | |
2267 | * of an unprocessed inode. If in that case the first ref would be | |
2268 | * overwritten, the overwritten inode gets "orphanized". Later when we | |
2269 | * process this overwritten inode, it is restored at a new place by moving | |
2270 | * the orphan inode. | |
2271 | * | |
2272 | * sctx->send_progress tells this function at which point in time receiving | |
2273 | * would be. | |
2274 | */ | |
2275 | static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, | |
2276 | struct fs_path *dest) | |
2277 | { | |
2278 | int ret = 0; | |
2279 | struct fs_path *name = NULL; | |
2280 | u64 parent_inode = 0; | |
2281 | u64 parent_gen = 0; | |
2282 | int stop = 0; | |
2283 | ||
924794c9 | 2284 | name = fs_path_alloc(); |
31db9f7c AB |
2285 | if (!name) { |
2286 | ret = -ENOMEM; | |
2287 | goto out; | |
2288 | } | |
2289 | ||
2290 | dest->reversed = 1; | |
2291 | fs_path_reset(dest); | |
2292 | ||
2293 | while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { | |
8b191a68 FM |
2294 | struct waiting_dir_move *wdm; |
2295 | ||
31db9f7c AB |
2296 | fs_path_reset(name); |
2297 | ||
9dc44214 FM |
2298 | if (is_waiting_for_rm(sctx, ino)) { |
2299 | ret = gen_unique_name(sctx, ino, gen, name); | |
2300 | if (ret < 0) | |
2301 | goto out; | |
2302 | ret = fs_path_add_path(dest, name); | |
2303 | break; | |
2304 | } | |
2305 | ||
8b191a68 FM |
2306 | wdm = get_waiting_dir_move(sctx, ino); |
2307 | if (wdm && wdm->orphanized) { | |
2308 | ret = gen_unique_name(sctx, ino, gen, name); | |
2309 | stop = 1; | |
2310 | } else if (wdm) { | |
bf0d1f44 FM |
2311 | ret = get_first_ref(sctx->parent_root, ino, |
2312 | &parent_inode, &parent_gen, name); | |
2313 | } else { | |
2314 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
2315 | &parent_inode, | |
2316 | &parent_gen, name); | |
2317 | if (ret) | |
2318 | stop = 1; | |
2319 | } | |
2320 | ||
31db9f7c AB |
2321 | if (ret < 0) |
2322 | goto out; | |
9f03740a | 2323 | |
31db9f7c AB |
2324 | ret = fs_path_add_path(dest, name); |
2325 | if (ret < 0) | |
2326 | goto out; | |
2327 | ||
2328 | ino = parent_inode; | |
2329 | gen = parent_gen; | |
2330 | } | |
2331 | ||
2332 | out: | |
924794c9 | 2333 | fs_path_free(name); |
31db9f7c AB |
2334 | if (!ret) |
2335 | fs_path_unreverse(dest); | |
2336 | return ret; | |
2337 | } | |
2338 | ||
31db9f7c AB |
2339 | /* |
2340 | * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace | |
2341 | */ | |
2342 | static int send_subvol_begin(struct send_ctx *sctx) | |
2343 | { | |
2344 | int ret; | |
2345 | struct btrfs_root *send_root = sctx->send_root; | |
2346 | struct btrfs_root *parent_root = sctx->parent_root; | |
2347 | struct btrfs_path *path; | |
2348 | struct btrfs_key key; | |
2349 | struct btrfs_root_ref *ref; | |
2350 | struct extent_buffer *leaf; | |
2351 | char *name = NULL; | |
2352 | int namelen; | |
2353 | ||
ffcfaf81 | 2354 | path = btrfs_alloc_path(); |
31db9f7c AB |
2355 | if (!path) |
2356 | return -ENOMEM; | |
2357 | ||
e780b0d1 | 2358 | name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); |
31db9f7c AB |
2359 | if (!name) { |
2360 | btrfs_free_path(path); | |
2361 | return -ENOMEM; | |
2362 | } | |
2363 | ||
2364 | key.objectid = send_root->objectid; | |
2365 | key.type = BTRFS_ROOT_BACKREF_KEY; | |
2366 | key.offset = 0; | |
2367 | ||
2368 | ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, | |
2369 | &key, path, 1, 0); | |
2370 | if (ret < 0) | |
2371 | goto out; | |
2372 | if (ret) { | |
2373 | ret = -ENOENT; | |
2374 | goto out; | |
2375 | } | |
2376 | ||
2377 | leaf = path->nodes[0]; | |
2378 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2379 | if (key.type != BTRFS_ROOT_BACKREF_KEY || | |
2380 | key.objectid != send_root->objectid) { | |
2381 | ret = -ENOENT; | |
2382 | goto out; | |
2383 | } | |
2384 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
2385 | namelen = btrfs_root_ref_name_len(leaf, ref); | |
2386 | read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); | |
2387 | btrfs_release_path(path); | |
2388 | ||
31db9f7c AB |
2389 | if (parent_root) { |
2390 | ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); | |
2391 | if (ret < 0) | |
2392 | goto out; | |
2393 | } else { | |
2394 | ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); | |
2395 | if (ret < 0) | |
2396 | goto out; | |
2397 | } | |
2398 | ||
2399 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); | |
b96b1db0 RR |
2400 | |
2401 | if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid)) | |
2402 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2403 | sctx->send_root->root_item.received_uuid); | |
2404 | else | |
2405 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2406 | sctx->send_root->root_item.uuid); | |
2407 | ||
31db9f7c | 2408 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, |
5a0f4e2c | 2409 | le64_to_cpu(sctx->send_root->root_item.ctransid)); |
31db9f7c | 2410 | if (parent_root) { |
37b8d27d JB |
2411 | if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) |
2412 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2413 | parent_root->root_item.received_uuid); | |
2414 | else | |
2415 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2416 | parent_root->root_item.uuid); | |
31db9f7c | 2417 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 2418 | le64_to_cpu(sctx->parent_root->root_item.ctransid)); |
31db9f7c AB |
2419 | } |
2420 | ||
2421 | ret = send_cmd(sctx); | |
2422 | ||
2423 | tlv_put_failure: | |
2424 | out: | |
2425 | btrfs_free_path(path); | |
2426 | kfree(name); | |
2427 | return ret; | |
2428 | } | |
2429 | ||
2430 | static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) | |
2431 | { | |
04ab956e | 2432 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2433 | int ret = 0; |
2434 | struct fs_path *p; | |
2435 | ||
04ab956e | 2436 | btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size); |
31db9f7c | 2437 | |
924794c9 | 2438 | p = fs_path_alloc(); |
31db9f7c AB |
2439 | if (!p) |
2440 | return -ENOMEM; | |
2441 | ||
2442 | ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); | |
2443 | if (ret < 0) | |
2444 | goto out; | |
2445 | ||
2446 | ret = get_cur_path(sctx, ino, gen, p); | |
2447 | if (ret < 0) | |
2448 | goto out; | |
2449 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2450 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); | |
2451 | ||
2452 | ret = send_cmd(sctx); | |
2453 | ||
2454 | tlv_put_failure: | |
2455 | out: | |
924794c9 | 2456 | fs_path_free(p); |
31db9f7c AB |
2457 | return ret; |
2458 | } | |
2459 | ||
2460 | static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) | |
2461 | { | |
04ab956e | 2462 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2463 | int ret = 0; |
2464 | struct fs_path *p; | |
2465 | ||
04ab956e | 2466 | btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode); |
31db9f7c | 2467 | |
924794c9 | 2468 | p = fs_path_alloc(); |
31db9f7c AB |
2469 | if (!p) |
2470 | return -ENOMEM; | |
2471 | ||
2472 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); | |
2473 | if (ret < 0) | |
2474 | goto out; | |
2475 | ||
2476 | ret = get_cur_path(sctx, ino, gen, p); | |
2477 | if (ret < 0) | |
2478 | goto out; | |
2479 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2480 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); | |
2481 | ||
2482 | ret = send_cmd(sctx); | |
2483 | ||
2484 | tlv_put_failure: | |
2485 | out: | |
924794c9 | 2486 | fs_path_free(p); |
31db9f7c AB |
2487 | return ret; |
2488 | } | |
2489 | ||
2490 | static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) | |
2491 | { | |
04ab956e | 2492 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2493 | int ret = 0; |
2494 | struct fs_path *p; | |
2495 | ||
04ab956e JM |
2496 | btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu", |
2497 | ino, uid, gid); | |
31db9f7c | 2498 | |
924794c9 | 2499 | p = fs_path_alloc(); |
31db9f7c AB |
2500 | if (!p) |
2501 | return -ENOMEM; | |
2502 | ||
2503 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); | |
2504 | if (ret < 0) | |
2505 | goto out; | |
2506 | ||
2507 | ret = get_cur_path(sctx, ino, gen, p); | |
2508 | if (ret < 0) | |
2509 | goto out; | |
2510 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2511 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); | |
2512 | TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); | |
2513 | ||
2514 | ret = send_cmd(sctx); | |
2515 | ||
2516 | tlv_put_failure: | |
2517 | out: | |
924794c9 | 2518 | fs_path_free(p); |
31db9f7c AB |
2519 | return ret; |
2520 | } | |
2521 | ||
2522 | static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) | |
2523 | { | |
04ab956e | 2524 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2525 | int ret = 0; |
2526 | struct fs_path *p = NULL; | |
2527 | struct btrfs_inode_item *ii; | |
2528 | struct btrfs_path *path = NULL; | |
2529 | struct extent_buffer *eb; | |
2530 | struct btrfs_key key; | |
2531 | int slot; | |
2532 | ||
04ab956e | 2533 | btrfs_debug(fs_info, "send_utimes %llu", ino); |
31db9f7c | 2534 | |
924794c9 | 2535 | p = fs_path_alloc(); |
31db9f7c AB |
2536 | if (!p) |
2537 | return -ENOMEM; | |
2538 | ||
2539 | path = alloc_path_for_send(); | |
2540 | if (!path) { | |
2541 | ret = -ENOMEM; | |
2542 | goto out; | |
2543 | } | |
2544 | ||
2545 | key.objectid = ino; | |
2546 | key.type = BTRFS_INODE_ITEM_KEY; | |
2547 | key.offset = 0; | |
2548 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); | |
15b253ea FM |
2549 | if (ret > 0) |
2550 | ret = -ENOENT; | |
31db9f7c AB |
2551 | if (ret < 0) |
2552 | goto out; | |
2553 | ||
2554 | eb = path->nodes[0]; | |
2555 | slot = path->slots[0]; | |
2556 | ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
2557 | ||
2558 | ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); | |
2559 | if (ret < 0) | |
2560 | goto out; | |
2561 | ||
2562 | ret = get_cur_path(sctx, ino, gen, p); | |
2563 | if (ret < 0) | |
2564 | goto out; | |
2565 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
a937b979 DS |
2566 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); |
2567 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); | |
2568 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); | |
766702ef | 2569 | /* TODO Add otime support when the otime patches get into upstream */ |
31db9f7c AB |
2570 | |
2571 | ret = send_cmd(sctx); | |
2572 | ||
2573 | tlv_put_failure: | |
2574 | out: | |
924794c9 | 2575 | fs_path_free(p); |
31db9f7c AB |
2576 | btrfs_free_path(path); |
2577 | return ret; | |
2578 | } | |
2579 | ||
2580 | /* | |
2581 | * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have | |
2582 | * a valid path yet because we did not process the refs yet. So, the inode | |
2583 | * is created as orphan. | |
2584 | */ | |
1f4692da | 2585 | static int send_create_inode(struct send_ctx *sctx, u64 ino) |
31db9f7c | 2586 | { |
04ab956e | 2587 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c | 2588 | int ret = 0; |
31db9f7c | 2589 | struct fs_path *p; |
31db9f7c | 2590 | int cmd; |
1f4692da | 2591 | u64 gen; |
31db9f7c | 2592 | u64 mode; |
1f4692da | 2593 | u64 rdev; |
31db9f7c | 2594 | |
04ab956e | 2595 | btrfs_debug(fs_info, "send_create_inode %llu", ino); |
31db9f7c | 2596 | |
924794c9 | 2597 | p = fs_path_alloc(); |
31db9f7c AB |
2598 | if (!p) |
2599 | return -ENOMEM; | |
2600 | ||
644d1940 LB |
2601 | if (ino != sctx->cur_ino) { |
2602 | ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, | |
2603 | NULL, NULL, &rdev); | |
2604 | if (ret < 0) | |
2605 | goto out; | |
2606 | } else { | |
2607 | gen = sctx->cur_inode_gen; | |
2608 | mode = sctx->cur_inode_mode; | |
2609 | rdev = sctx->cur_inode_rdev; | |
2610 | } | |
31db9f7c | 2611 | |
e938c8ad | 2612 | if (S_ISREG(mode)) { |
31db9f7c | 2613 | cmd = BTRFS_SEND_C_MKFILE; |
e938c8ad | 2614 | } else if (S_ISDIR(mode)) { |
31db9f7c | 2615 | cmd = BTRFS_SEND_C_MKDIR; |
e938c8ad | 2616 | } else if (S_ISLNK(mode)) { |
31db9f7c | 2617 | cmd = BTRFS_SEND_C_SYMLINK; |
e938c8ad | 2618 | } else if (S_ISCHR(mode) || S_ISBLK(mode)) { |
31db9f7c | 2619 | cmd = BTRFS_SEND_C_MKNOD; |
e938c8ad | 2620 | } else if (S_ISFIFO(mode)) { |
31db9f7c | 2621 | cmd = BTRFS_SEND_C_MKFIFO; |
e938c8ad | 2622 | } else if (S_ISSOCK(mode)) { |
31db9f7c | 2623 | cmd = BTRFS_SEND_C_MKSOCK; |
e938c8ad | 2624 | } else { |
f14d104d | 2625 | btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o", |
31db9f7c | 2626 | (int)(mode & S_IFMT)); |
ca6842bf | 2627 | ret = -EOPNOTSUPP; |
31db9f7c AB |
2628 | goto out; |
2629 | } | |
2630 | ||
2631 | ret = begin_cmd(sctx, cmd); | |
2632 | if (ret < 0) | |
2633 | goto out; | |
2634 | ||
1f4692da | 2635 | ret = gen_unique_name(sctx, ino, gen, p); |
31db9f7c AB |
2636 | if (ret < 0) |
2637 | goto out; | |
2638 | ||
2639 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
1f4692da | 2640 | TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); |
31db9f7c AB |
2641 | |
2642 | if (S_ISLNK(mode)) { | |
2643 | fs_path_reset(p); | |
924794c9 | 2644 | ret = read_symlink(sctx->send_root, ino, p); |
31db9f7c AB |
2645 | if (ret < 0) |
2646 | goto out; | |
2647 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); | |
2648 | } else if (S_ISCHR(mode) || S_ISBLK(mode) || | |
2649 | S_ISFIFO(mode) || S_ISSOCK(mode)) { | |
d79e5043 AJ |
2650 | TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); |
2651 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); | |
31db9f7c AB |
2652 | } |
2653 | ||
2654 | ret = send_cmd(sctx); | |
2655 | if (ret < 0) | |
2656 | goto out; | |
2657 | ||
2658 | ||
2659 | tlv_put_failure: | |
2660 | out: | |
924794c9 | 2661 | fs_path_free(p); |
31db9f7c AB |
2662 | return ret; |
2663 | } | |
2664 | ||
1f4692da AB |
2665 | /* |
2666 | * We need some special handling for inodes that get processed before the parent | |
2667 | * directory got created. See process_recorded_refs for details. | |
2668 | * This function does the check if we already created the dir out of order. | |
2669 | */ | |
2670 | static int did_create_dir(struct send_ctx *sctx, u64 dir) | |
2671 | { | |
2672 | int ret = 0; | |
2673 | struct btrfs_path *path = NULL; | |
2674 | struct btrfs_key key; | |
2675 | struct btrfs_key found_key; | |
2676 | struct btrfs_key di_key; | |
2677 | struct extent_buffer *eb; | |
2678 | struct btrfs_dir_item *di; | |
2679 | int slot; | |
2680 | ||
2681 | path = alloc_path_for_send(); | |
2682 | if (!path) { | |
2683 | ret = -ENOMEM; | |
2684 | goto out; | |
2685 | } | |
2686 | ||
2687 | key.objectid = dir; | |
2688 | key.type = BTRFS_DIR_INDEX_KEY; | |
2689 | key.offset = 0; | |
dff6d0ad FDBM |
2690 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); |
2691 | if (ret < 0) | |
2692 | goto out; | |
2693 | ||
1f4692da | 2694 | while (1) { |
dff6d0ad FDBM |
2695 | eb = path->nodes[0]; |
2696 | slot = path->slots[0]; | |
2697 | if (slot >= btrfs_header_nritems(eb)) { | |
2698 | ret = btrfs_next_leaf(sctx->send_root, path); | |
2699 | if (ret < 0) { | |
2700 | goto out; | |
2701 | } else if (ret > 0) { | |
2702 | ret = 0; | |
2703 | break; | |
2704 | } | |
2705 | continue; | |
1f4692da | 2706 | } |
dff6d0ad FDBM |
2707 | |
2708 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
2709 | if (found_key.objectid != key.objectid || | |
1f4692da AB |
2710 | found_key.type != key.type) { |
2711 | ret = 0; | |
2712 | goto out; | |
2713 | } | |
2714 | ||
2715 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); | |
2716 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2717 | ||
a0525414 JB |
2718 | if (di_key.type != BTRFS_ROOT_ITEM_KEY && |
2719 | di_key.objectid < sctx->send_progress) { | |
1f4692da AB |
2720 | ret = 1; |
2721 | goto out; | |
2722 | } | |
2723 | ||
dff6d0ad | 2724 | path->slots[0]++; |
1f4692da AB |
2725 | } |
2726 | ||
2727 | out: | |
2728 | btrfs_free_path(path); | |
2729 | return ret; | |
2730 | } | |
2731 | ||
2732 | /* | |
2733 | * Only creates the inode if it is: | |
2734 | * 1. Not a directory | |
2735 | * 2. Or a directory which was not created already due to out of order | |
2736 | * directories. See did_create_dir and process_recorded_refs for details. | |
2737 | */ | |
2738 | static int send_create_inode_if_needed(struct send_ctx *sctx) | |
2739 | { | |
2740 | int ret; | |
2741 | ||
2742 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
2743 | ret = did_create_dir(sctx, sctx->cur_ino); | |
2744 | if (ret < 0) | |
2745 | goto out; | |
2746 | if (ret) { | |
2747 | ret = 0; | |
2748 | goto out; | |
2749 | } | |
2750 | } | |
2751 | ||
2752 | ret = send_create_inode(sctx, sctx->cur_ino); | |
2753 | if (ret < 0) | |
2754 | goto out; | |
2755 | ||
2756 | out: | |
2757 | return ret; | |
2758 | } | |
2759 | ||
31db9f7c AB |
2760 | struct recorded_ref { |
2761 | struct list_head list; | |
31db9f7c AB |
2762 | char *name; |
2763 | struct fs_path *full_path; | |
2764 | u64 dir; | |
2765 | u64 dir_gen; | |
31db9f7c AB |
2766 | int name_len; |
2767 | }; | |
2768 | ||
fdb13889 FM |
2769 | static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) |
2770 | { | |
2771 | ref->full_path = path; | |
2772 | ref->name = (char *)kbasename(ref->full_path->start); | |
2773 | ref->name_len = ref->full_path->end - ref->name; | |
2774 | } | |
2775 | ||
31db9f7c AB |
2776 | /* |
2777 | * We need to process new refs before deleted refs, but compare_tree gives us | |
2778 | * everything mixed. So we first record all refs and later process them. | |
2779 | * This function is a helper to record one ref. | |
2780 | */ | |
a4d96d62 | 2781 | static int __record_ref(struct list_head *head, u64 dir, |
31db9f7c AB |
2782 | u64 dir_gen, struct fs_path *path) |
2783 | { | |
2784 | struct recorded_ref *ref; | |
31db9f7c | 2785 | |
e780b0d1 | 2786 | ref = kmalloc(sizeof(*ref), GFP_KERNEL); |
31db9f7c AB |
2787 | if (!ref) |
2788 | return -ENOMEM; | |
2789 | ||
2790 | ref->dir = dir; | |
2791 | ref->dir_gen = dir_gen; | |
fdb13889 | 2792 | set_ref_path(ref, path); |
31db9f7c AB |
2793 | list_add_tail(&ref->list, head); |
2794 | return 0; | |
2795 | } | |
2796 | ||
ba5e8f2e JB |
2797 | static int dup_ref(struct recorded_ref *ref, struct list_head *list) |
2798 | { | |
2799 | struct recorded_ref *new; | |
2800 | ||
e780b0d1 | 2801 | new = kmalloc(sizeof(*ref), GFP_KERNEL); |
ba5e8f2e JB |
2802 | if (!new) |
2803 | return -ENOMEM; | |
2804 | ||
2805 | new->dir = ref->dir; | |
2806 | new->dir_gen = ref->dir_gen; | |
2807 | new->full_path = NULL; | |
2808 | INIT_LIST_HEAD(&new->list); | |
2809 | list_add_tail(&new->list, list); | |
2810 | return 0; | |
2811 | } | |
2812 | ||
924794c9 | 2813 | static void __free_recorded_refs(struct list_head *head) |
31db9f7c AB |
2814 | { |
2815 | struct recorded_ref *cur; | |
31db9f7c | 2816 | |
e938c8ad AB |
2817 | while (!list_empty(head)) { |
2818 | cur = list_entry(head->next, struct recorded_ref, list); | |
924794c9 | 2819 | fs_path_free(cur->full_path); |
e938c8ad | 2820 | list_del(&cur->list); |
31db9f7c AB |
2821 | kfree(cur); |
2822 | } | |
31db9f7c AB |
2823 | } |
2824 | ||
2825 | static void free_recorded_refs(struct send_ctx *sctx) | |
2826 | { | |
924794c9 TI |
2827 | __free_recorded_refs(&sctx->new_refs); |
2828 | __free_recorded_refs(&sctx->deleted_refs); | |
31db9f7c AB |
2829 | } |
2830 | ||
2831 | /* | |
766702ef | 2832 | * Renames/moves a file/dir to its orphan name. Used when the first |
31db9f7c AB |
2833 | * ref of an unprocessed inode gets overwritten and for all non empty |
2834 | * directories. | |
2835 | */ | |
2836 | static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, | |
2837 | struct fs_path *path) | |
2838 | { | |
2839 | int ret; | |
2840 | struct fs_path *orphan; | |
2841 | ||
924794c9 | 2842 | orphan = fs_path_alloc(); |
31db9f7c AB |
2843 | if (!orphan) |
2844 | return -ENOMEM; | |
2845 | ||
2846 | ret = gen_unique_name(sctx, ino, gen, orphan); | |
2847 | if (ret < 0) | |
2848 | goto out; | |
2849 | ||
2850 | ret = send_rename(sctx, path, orphan); | |
2851 | ||
2852 | out: | |
924794c9 | 2853 | fs_path_free(orphan); |
31db9f7c AB |
2854 | return ret; |
2855 | } | |
2856 | ||
9dc44214 FM |
2857 | static struct orphan_dir_info * |
2858 | add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2859 | { | |
2860 | struct rb_node **p = &sctx->orphan_dirs.rb_node; | |
2861 | struct rb_node *parent = NULL; | |
2862 | struct orphan_dir_info *entry, *odi; | |
2863 | ||
e780b0d1 | 2864 | odi = kmalloc(sizeof(*odi), GFP_KERNEL); |
9dc44214 FM |
2865 | if (!odi) |
2866 | return ERR_PTR(-ENOMEM); | |
2867 | odi->ino = dir_ino; | |
2868 | odi->gen = 0; | |
2869 | ||
2870 | while (*p) { | |
2871 | parent = *p; | |
2872 | entry = rb_entry(parent, struct orphan_dir_info, node); | |
2873 | if (dir_ino < entry->ino) { | |
2874 | p = &(*p)->rb_left; | |
2875 | } else if (dir_ino > entry->ino) { | |
2876 | p = &(*p)->rb_right; | |
2877 | } else { | |
2878 | kfree(odi); | |
2879 | return entry; | |
2880 | } | |
2881 | } | |
2882 | ||
2883 | rb_link_node(&odi->node, parent, p); | |
2884 | rb_insert_color(&odi->node, &sctx->orphan_dirs); | |
2885 | return odi; | |
2886 | } | |
2887 | ||
2888 | static struct orphan_dir_info * | |
2889 | get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2890 | { | |
2891 | struct rb_node *n = sctx->orphan_dirs.rb_node; | |
2892 | struct orphan_dir_info *entry; | |
2893 | ||
2894 | while (n) { | |
2895 | entry = rb_entry(n, struct orphan_dir_info, node); | |
2896 | if (dir_ino < entry->ino) | |
2897 | n = n->rb_left; | |
2898 | else if (dir_ino > entry->ino) | |
2899 | n = n->rb_right; | |
2900 | else | |
2901 | return entry; | |
2902 | } | |
2903 | return NULL; | |
2904 | } | |
2905 | ||
2906 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino) | |
2907 | { | |
2908 | struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino); | |
2909 | ||
2910 | return odi != NULL; | |
2911 | } | |
2912 | ||
2913 | static void free_orphan_dir_info(struct send_ctx *sctx, | |
2914 | struct orphan_dir_info *odi) | |
2915 | { | |
2916 | if (!odi) | |
2917 | return; | |
2918 | rb_erase(&odi->node, &sctx->orphan_dirs); | |
2919 | kfree(odi); | |
2920 | } | |
2921 | ||
31db9f7c AB |
2922 | /* |
2923 | * Returns 1 if a directory can be removed at this point in time. | |
2924 | * We check this by iterating all dir items and checking if the inode behind | |
2925 | * the dir item was already processed. | |
2926 | */ | |
9dc44214 FM |
2927 | static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
2928 | u64 send_progress) | |
31db9f7c AB |
2929 | { |
2930 | int ret = 0; | |
2931 | struct btrfs_root *root = sctx->parent_root; | |
2932 | struct btrfs_path *path; | |
2933 | struct btrfs_key key; | |
2934 | struct btrfs_key found_key; | |
2935 | struct btrfs_key loc; | |
2936 | struct btrfs_dir_item *di; | |
2937 | ||
6d85ed05 AB |
2938 | /* |
2939 | * Don't try to rmdir the top/root subvolume dir. | |
2940 | */ | |
2941 | if (dir == BTRFS_FIRST_FREE_OBJECTID) | |
2942 | return 0; | |
2943 | ||
31db9f7c AB |
2944 | path = alloc_path_for_send(); |
2945 | if (!path) | |
2946 | return -ENOMEM; | |
2947 | ||
2948 | key.objectid = dir; | |
2949 | key.type = BTRFS_DIR_INDEX_KEY; | |
2950 | key.offset = 0; | |
dff6d0ad FDBM |
2951 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2952 | if (ret < 0) | |
2953 | goto out; | |
31db9f7c AB |
2954 | |
2955 | while (1) { | |
9dc44214 FM |
2956 | struct waiting_dir_move *dm; |
2957 | ||
dff6d0ad FDBM |
2958 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
2959 | ret = btrfs_next_leaf(root, path); | |
2960 | if (ret < 0) | |
2961 | goto out; | |
2962 | else if (ret > 0) | |
2963 | break; | |
2964 | continue; | |
31db9f7c | 2965 | } |
dff6d0ad FDBM |
2966 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2967 | path->slots[0]); | |
2968 | if (found_key.objectid != key.objectid || | |
2969 | found_key.type != key.type) | |
31db9f7c | 2970 | break; |
31db9f7c AB |
2971 | |
2972 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2973 | struct btrfs_dir_item); | |
2974 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
2975 | ||
9dc44214 FM |
2976 | dm = get_waiting_dir_move(sctx, loc.objectid); |
2977 | if (dm) { | |
2978 | struct orphan_dir_info *odi; | |
2979 | ||
2980 | odi = add_orphan_dir_info(sctx, dir); | |
2981 | if (IS_ERR(odi)) { | |
2982 | ret = PTR_ERR(odi); | |
2983 | goto out; | |
2984 | } | |
2985 | odi->gen = dir_gen; | |
2986 | dm->rmdir_ino = dir; | |
2987 | ret = 0; | |
2988 | goto out; | |
2989 | } | |
2990 | ||
31db9f7c | 2991 | if (loc.objectid > send_progress) { |
443f9d26 RK |
2992 | struct orphan_dir_info *odi; |
2993 | ||
2994 | odi = get_orphan_dir_info(sctx, dir); | |
2995 | free_orphan_dir_info(sctx, odi); | |
31db9f7c AB |
2996 | ret = 0; |
2997 | goto out; | |
2998 | } | |
2999 | ||
dff6d0ad | 3000 | path->slots[0]++; |
31db9f7c AB |
3001 | } |
3002 | ||
3003 | ret = 1; | |
3004 | ||
3005 | out: | |
3006 | btrfs_free_path(path); | |
3007 | return ret; | |
3008 | } | |
3009 | ||
9f03740a FDBM |
3010 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) |
3011 | { | |
9dc44214 | 3012 | struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); |
9f03740a | 3013 | |
9dc44214 | 3014 | return entry != NULL; |
9f03740a FDBM |
3015 | } |
3016 | ||
8b191a68 | 3017 | static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) |
9f03740a FDBM |
3018 | { |
3019 | struct rb_node **p = &sctx->waiting_dir_moves.rb_node; | |
3020 | struct rb_node *parent = NULL; | |
3021 | struct waiting_dir_move *entry, *dm; | |
3022 | ||
e780b0d1 | 3023 | dm = kmalloc(sizeof(*dm), GFP_KERNEL); |
9f03740a FDBM |
3024 | if (!dm) |
3025 | return -ENOMEM; | |
3026 | dm->ino = ino; | |
9dc44214 | 3027 | dm->rmdir_ino = 0; |
8b191a68 | 3028 | dm->orphanized = orphanized; |
9f03740a FDBM |
3029 | |
3030 | while (*p) { | |
3031 | parent = *p; | |
3032 | entry = rb_entry(parent, struct waiting_dir_move, node); | |
3033 | if (ino < entry->ino) { | |
3034 | p = &(*p)->rb_left; | |
3035 | } else if (ino > entry->ino) { | |
3036 | p = &(*p)->rb_right; | |
3037 | } else { | |
3038 | kfree(dm); | |
3039 | return -EEXIST; | |
3040 | } | |
3041 | } | |
3042 | ||
3043 | rb_link_node(&dm->node, parent, p); | |
3044 | rb_insert_color(&dm->node, &sctx->waiting_dir_moves); | |
3045 | return 0; | |
3046 | } | |
3047 | ||
9dc44214 FM |
3048 | static struct waiting_dir_move * |
3049 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
9f03740a FDBM |
3050 | { |
3051 | struct rb_node *n = sctx->waiting_dir_moves.rb_node; | |
3052 | struct waiting_dir_move *entry; | |
3053 | ||
3054 | while (n) { | |
3055 | entry = rb_entry(n, struct waiting_dir_move, node); | |
9dc44214 | 3056 | if (ino < entry->ino) |
9f03740a | 3057 | n = n->rb_left; |
9dc44214 | 3058 | else if (ino > entry->ino) |
9f03740a | 3059 | n = n->rb_right; |
9dc44214 FM |
3060 | else |
3061 | return entry; | |
9f03740a | 3062 | } |
9dc44214 FM |
3063 | return NULL; |
3064 | } | |
3065 | ||
3066 | static void free_waiting_dir_move(struct send_ctx *sctx, | |
3067 | struct waiting_dir_move *dm) | |
3068 | { | |
3069 | if (!dm) | |
3070 | return; | |
3071 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
3072 | kfree(dm); | |
9f03740a FDBM |
3073 | } |
3074 | ||
bfa7e1f8 FM |
3075 | static int add_pending_dir_move(struct send_ctx *sctx, |
3076 | u64 ino, | |
3077 | u64 ino_gen, | |
f959492f FM |
3078 | u64 parent_ino, |
3079 | struct list_head *new_refs, | |
84471e24 FM |
3080 | struct list_head *deleted_refs, |
3081 | const bool is_orphan) | |
9f03740a FDBM |
3082 | { |
3083 | struct rb_node **p = &sctx->pending_dir_moves.rb_node; | |
3084 | struct rb_node *parent = NULL; | |
73b802f4 | 3085 | struct pending_dir_move *entry = NULL, *pm; |
9f03740a FDBM |
3086 | struct recorded_ref *cur; |
3087 | int exists = 0; | |
3088 | int ret; | |
3089 | ||
e780b0d1 | 3090 | pm = kmalloc(sizeof(*pm), GFP_KERNEL); |
9f03740a FDBM |
3091 | if (!pm) |
3092 | return -ENOMEM; | |
3093 | pm->parent_ino = parent_ino; | |
bfa7e1f8 FM |
3094 | pm->ino = ino; |
3095 | pm->gen = ino_gen; | |
9f03740a FDBM |
3096 | INIT_LIST_HEAD(&pm->list); |
3097 | INIT_LIST_HEAD(&pm->update_refs); | |
3098 | RB_CLEAR_NODE(&pm->node); | |
3099 | ||
3100 | while (*p) { | |
3101 | parent = *p; | |
3102 | entry = rb_entry(parent, struct pending_dir_move, node); | |
3103 | if (parent_ino < entry->parent_ino) { | |
3104 | p = &(*p)->rb_left; | |
3105 | } else if (parent_ino > entry->parent_ino) { | |
3106 | p = &(*p)->rb_right; | |
3107 | } else { | |
3108 | exists = 1; | |
3109 | break; | |
3110 | } | |
3111 | } | |
3112 | ||
f959492f | 3113 | list_for_each_entry(cur, deleted_refs, list) { |
9f03740a FDBM |
3114 | ret = dup_ref(cur, &pm->update_refs); |
3115 | if (ret < 0) | |
3116 | goto out; | |
3117 | } | |
f959492f | 3118 | list_for_each_entry(cur, new_refs, list) { |
9f03740a FDBM |
3119 | ret = dup_ref(cur, &pm->update_refs); |
3120 | if (ret < 0) | |
3121 | goto out; | |
3122 | } | |
3123 | ||
8b191a68 | 3124 | ret = add_waiting_dir_move(sctx, pm->ino, is_orphan); |
9f03740a FDBM |
3125 | if (ret) |
3126 | goto out; | |
3127 | ||
3128 | if (exists) { | |
3129 | list_add_tail(&pm->list, &entry->list); | |
3130 | } else { | |
3131 | rb_link_node(&pm->node, parent, p); | |
3132 | rb_insert_color(&pm->node, &sctx->pending_dir_moves); | |
3133 | } | |
3134 | ret = 0; | |
3135 | out: | |
3136 | if (ret) { | |
3137 | __free_recorded_refs(&pm->update_refs); | |
3138 | kfree(pm); | |
3139 | } | |
3140 | return ret; | |
3141 | } | |
3142 | ||
3143 | static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, | |
3144 | u64 parent_ino) | |
3145 | { | |
3146 | struct rb_node *n = sctx->pending_dir_moves.rb_node; | |
3147 | struct pending_dir_move *entry; | |
3148 | ||
3149 | while (n) { | |
3150 | entry = rb_entry(n, struct pending_dir_move, node); | |
3151 | if (parent_ino < entry->parent_ino) | |
3152 | n = n->rb_left; | |
3153 | else if (parent_ino > entry->parent_ino) | |
3154 | n = n->rb_right; | |
3155 | else | |
3156 | return entry; | |
3157 | } | |
3158 | return NULL; | |
3159 | } | |
3160 | ||
801bec36 RK |
3161 | static int path_loop(struct send_ctx *sctx, struct fs_path *name, |
3162 | u64 ino, u64 gen, u64 *ancestor_ino) | |
3163 | { | |
3164 | int ret = 0; | |
3165 | u64 parent_inode = 0; | |
3166 | u64 parent_gen = 0; | |
3167 | u64 start_ino = ino; | |
3168 | ||
3169 | *ancestor_ino = 0; | |
3170 | while (ino != BTRFS_FIRST_FREE_OBJECTID) { | |
3171 | fs_path_reset(name); | |
3172 | ||
3173 | if (is_waiting_for_rm(sctx, ino)) | |
3174 | break; | |
3175 | if (is_waiting_for_move(sctx, ino)) { | |
3176 | if (*ancestor_ino == 0) | |
3177 | *ancestor_ino = ino; | |
3178 | ret = get_first_ref(sctx->parent_root, ino, | |
3179 | &parent_inode, &parent_gen, name); | |
3180 | } else { | |
3181 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
3182 | &parent_inode, | |
3183 | &parent_gen, name); | |
3184 | if (ret > 0) { | |
3185 | ret = 0; | |
3186 | break; | |
3187 | } | |
3188 | } | |
3189 | if (ret < 0) | |
3190 | break; | |
3191 | if (parent_inode == start_ino) { | |
3192 | ret = 1; | |
3193 | if (*ancestor_ino == 0) | |
3194 | *ancestor_ino = ino; | |
3195 | break; | |
3196 | } | |
3197 | ino = parent_inode; | |
3198 | gen = parent_gen; | |
3199 | } | |
3200 | return ret; | |
3201 | } | |
3202 | ||
9f03740a FDBM |
3203 | static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) |
3204 | { | |
3205 | struct fs_path *from_path = NULL; | |
3206 | struct fs_path *to_path = NULL; | |
2b863a13 | 3207 | struct fs_path *name = NULL; |
9f03740a FDBM |
3208 | u64 orig_progress = sctx->send_progress; |
3209 | struct recorded_ref *cur; | |
2b863a13 | 3210 | u64 parent_ino, parent_gen; |
9dc44214 FM |
3211 | struct waiting_dir_move *dm = NULL; |
3212 | u64 rmdir_ino = 0; | |
801bec36 RK |
3213 | u64 ancestor; |
3214 | bool is_orphan; | |
9f03740a FDBM |
3215 | int ret; |
3216 | ||
2b863a13 | 3217 | name = fs_path_alloc(); |
9f03740a | 3218 | from_path = fs_path_alloc(); |
2b863a13 FM |
3219 | if (!name || !from_path) { |
3220 | ret = -ENOMEM; | |
3221 | goto out; | |
3222 | } | |
9f03740a | 3223 | |
9dc44214 FM |
3224 | dm = get_waiting_dir_move(sctx, pm->ino); |
3225 | ASSERT(dm); | |
3226 | rmdir_ino = dm->rmdir_ino; | |
801bec36 | 3227 | is_orphan = dm->orphanized; |
9dc44214 | 3228 | free_waiting_dir_move(sctx, dm); |
2b863a13 | 3229 | |
801bec36 | 3230 | if (is_orphan) { |
84471e24 FM |
3231 | ret = gen_unique_name(sctx, pm->ino, |
3232 | pm->gen, from_path); | |
3233 | } else { | |
3234 | ret = get_first_ref(sctx->parent_root, pm->ino, | |
3235 | &parent_ino, &parent_gen, name); | |
3236 | if (ret < 0) | |
3237 | goto out; | |
3238 | ret = get_cur_path(sctx, parent_ino, parent_gen, | |
3239 | from_path); | |
3240 | if (ret < 0) | |
3241 | goto out; | |
3242 | ret = fs_path_add_path(from_path, name); | |
3243 | } | |
c992ec94 FM |
3244 | if (ret < 0) |
3245 | goto out; | |
2b863a13 | 3246 | |
f959492f | 3247 | sctx->send_progress = sctx->cur_ino + 1; |
801bec36 | 3248 | ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor); |
7969e77a FM |
3249 | if (ret < 0) |
3250 | goto out; | |
801bec36 RK |
3251 | if (ret) { |
3252 | LIST_HEAD(deleted_refs); | |
3253 | ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID); | |
3254 | ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor, | |
3255 | &pm->update_refs, &deleted_refs, | |
3256 | is_orphan); | |
3257 | if (ret < 0) | |
3258 | goto out; | |
3259 | if (rmdir_ino) { | |
3260 | dm = get_waiting_dir_move(sctx, pm->ino); | |
3261 | ASSERT(dm); | |
3262 | dm->rmdir_ino = rmdir_ino; | |
3263 | } | |
3264 | goto out; | |
3265 | } | |
c992ec94 FM |
3266 | fs_path_reset(name); |
3267 | to_path = name; | |
2b863a13 | 3268 | name = NULL; |
9f03740a FDBM |
3269 | ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); |
3270 | if (ret < 0) | |
3271 | goto out; | |
3272 | ||
3273 | ret = send_rename(sctx, from_path, to_path); | |
3274 | if (ret < 0) | |
3275 | goto out; | |
3276 | ||
9dc44214 FM |
3277 | if (rmdir_ino) { |
3278 | struct orphan_dir_info *odi; | |
3279 | ||
3280 | odi = get_orphan_dir_info(sctx, rmdir_ino); | |
3281 | if (!odi) { | |
3282 | /* already deleted */ | |
3283 | goto finish; | |
3284 | } | |
99ea42dd | 3285 | ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino); |
9dc44214 FM |
3286 | if (ret < 0) |
3287 | goto out; | |
3288 | if (!ret) | |
3289 | goto finish; | |
3290 | ||
3291 | name = fs_path_alloc(); | |
3292 | if (!name) { | |
3293 | ret = -ENOMEM; | |
3294 | goto out; | |
3295 | } | |
3296 | ret = get_cur_path(sctx, rmdir_ino, odi->gen, name); | |
3297 | if (ret < 0) | |
3298 | goto out; | |
3299 | ret = send_rmdir(sctx, name); | |
3300 | if (ret < 0) | |
3301 | goto out; | |
3302 | free_orphan_dir_info(sctx, odi); | |
3303 | } | |
3304 | ||
3305 | finish: | |
9f03740a FDBM |
3306 | ret = send_utimes(sctx, pm->ino, pm->gen); |
3307 | if (ret < 0) | |
3308 | goto out; | |
3309 | ||
3310 | /* | |
3311 | * After rename/move, need to update the utimes of both new parent(s) | |
3312 | * and old parent(s). | |
3313 | */ | |
3314 | list_for_each_entry(cur, &pm->update_refs, list) { | |
764433a1 RK |
3315 | /* |
3316 | * The parent inode might have been deleted in the send snapshot | |
3317 | */ | |
3318 | ret = get_inode_info(sctx->send_root, cur->dir, NULL, | |
3319 | NULL, NULL, NULL, NULL, NULL); | |
3320 | if (ret == -ENOENT) { | |
3321 | ret = 0; | |
9dc44214 | 3322 | continue; |
764433a1 RK |
3323 | } |
3324 | if (ret < 0) | |
3325 | goto out; | |
3326 | ||
9f03740a FDBM |
3327 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
3328 | if (ret < 0) | |
3329 | goto out; | |
3330 | } | |
3331 | ||
3332 | out: | |
2b863a13 | 3333 | fs_path_free(name); |
9f03740a FDBM |
3334 | fs_path_free(from_path); |
3335 | fs_path_free(to_path); | |
3336 | sctx->send_progress = orig_progress; | |
3337 | ||
3338 | return ret; | |
3339 | } | |
3340 | ||
3341 | static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) | |
3342 | { | |
3343 | if (!list_empty(&m->list)) | |
3344 | list_del(&m->list); | |
3345 | if (!RB_EMPTY_NODE(&m->node)) | |
3346 | rb_erase(&m->node, &sctx->pending_dir_moves); | |
3347 | __free_recorded_refs(&m->update_refs); | |
3348 | kfree(m); | |
3349 | } | |
3350 | ||
3351 | static void tail_append_pending_moves(struct pending_dir_move *moves, | |
3352 | struct list_head *stack) | |
3353 | { | |
3354 | if (list_empty(&moves->list)) { | |
3355 | list_add_tail(&moves->list, stack); | |
3356 | } else { | |
3357 | LIST_HEAD(list); | |
3358 | list_splice_init(&moves->list, &list); | |
3359 | list_add_tail(&moves->list, stack); | |
3360 | list_splice_tail(&list, stack); | |
3361 | } | |
3362 | } | |
3363 | ||
3364 | static int apply_children_dir_moves(struct send_ctx *sctx) | |
3365 | { | |
3366 | struct pending_dir_move *pm; | |
3367 | struct list_head stack; | |
3368 | u64 parent_ino = sctx->cur_ino; | |
3369 | int ret = 0; | |
3370 | ||
3371 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3372 | if (!pm) | |
3373 | return 0; | |
3374 | ||
3375 | INIT_LIST_HEAD(&stack); | |
3376 | tail_append_pending_moves(pm, &stack); | |
3377 | ||
3378 | while (!list_empty(&stack)) { | |
3379 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3380 | parent_ino = pm->ino; | |
3381 | ret = apply_dir_move(sctx, pm); | |
3382 | free_pending_move(sctx, pm); | |
3383 | if (ret) | |
3384 | goto out; | |
3385 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3386 | if (pm) | |
3387 | tail_append_pending_moves(pm, &stack); | |
3388 | } | |
3389 | return 0; | |
3390 | ||
3391 | out: | |
3392 | while (!list_empty(&stack)) { | |
3393 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3394 | free_pending_move(sctx, pm); | |
3395 | } | |
3396 | return ret; | |
3397 | } | |
3398 | ||
84471e24 FM |
3399 | /* |
3400 | * We might need to delay a directory rename even when no ancestor directory | |
3401 | * (in the send root) with a higher inode number than ours (sctx->cur_ino) was | |
3402 | * renamed. This happens when we rename a directory to the old name (the name | |
3403 | * in the parent root) of some other unrelated directory that got its rename | |
3404 | * delayed due to some ancestor with higher number that got renamed. | |
3405 | * | |
3406 | * Example: | |
3407 | * | |
3408 | * Parent snapshot: | |
3409 | * . (ino 256) | |
3410 | * |---- a/ (ino 257) | |
3411 | * | |---- file (ino 260) | |
3412 | * | | |
3413 | * |---- b/ (ino 258) | |
3414 | * |---- c/ (ino 259) | |
3415 | * | |
3416 | * Send snapshot: | |
3417 | * . (ino 256) | |
3418 | * |---- a/ (ino 258) | |
3419 | * |---- x/ (ino 259) | |
3420 | * |---- y/ (ino 257) | |
3421 | * |----- file (ino 260) | |
3422 | * | |
3423 | * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257 | |
3424 | * from 'a' to 'x/y' happening first, which in turn depends on the rename of | |
3425 | * inode 259 from 'c' to 'x'. So the order of rename commands the send stream | |
3426 | * must issue is: | |
3427 | * | |
3428 | * 1 - rename 259 from 'c' to 'x' | |
3429 | * 2 - rename 257 from 'a' to 'x/y' | |
3430 | * 3 - rename 258 from 'b' to 'a' | |
3431 | * | |
3432 | * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can | |
3433 | * be done right away and < 0 on error. | |
3434 | */ | |
3435 | static int wait_for_dest_dir_move(struct send_ctx *sctx, | |
3436 | struct recorded_ref *parent_ref, | |
3437 | const bool is_orphan) | |
3438 | { | |
2ff7e61e | 3439 | struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info; |
84471e24 FM |
3440 | struct btrfs_path *path; |
3441 | struct btrfs_key key; | |
3442 | struct btrfs_key di_key; | |
3443 | struct btrfs_dir_item *di; | |
3444 | u64 left_gen; | |
3445 | u64 right_gen; | |
3446 | int ret = 0; | |
801bec36 | 3447 | struct waiting_dir_move *wdm; |
84471e24 FM |
3448 | |
3449 | if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) | |
3450 | return 0; | |
3451 | ||
3452 | path = alloc_path_for_send(); | |
3453 | if (!path) | |
3454 | return -ENOMEM; | |
3455 | ||
3456 | key.objectid = parent_ref->dir; | |
3457 | key.type = BTRFS_DIR_ITEM_KEY; | |
3458 | key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); | |
3459 | ||
3460 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
3461 | if (ret < 0) { | |
3462 | goto out; | |
3463 | } else if (ret > 0) { | |
3464 | ret = 0; | |
3465 | goto out; | |
3466 | } | |
3467 | ||
2ff7e61e JM |
3468 | di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name, |
3469 | parent_ref->name_len); | |
84471e24 FM |
3470 | if (!di) { |
3471 | ret = 0; | |
3472 | goto out; | |
3473 | } | |
3474 | /* | |
3475 | * di_key.objectid has the number of the inode that has a dentry in the | |
3476 | * parent directory with the same name that sctx->cur_ino is being | |
3477 | * renamed to. We need to check if that inode is in the send root as | |
3478 | * well and if it is currently marked as an inode with a pending rename, | |
3479 | * if it is, we need to delay the rename of sctx->cur_ino as well, so | |
3480 | * that it happens after that other inode is renamed. | |
3481 | */ | |
3482 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); | |
3483 | if (di_key.type != BTRFS_INODE_ITEM_KEY) { | |
3484 | ret = 0; | |
3485 | goto out; | |
3486 | } | |
3487 | ||
3488 | ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL, | |
3489 | &left_gen, NULL, NULL, NULL, NULL); | |
3490 | if (ret < 0) | |
3491 | goto out; | |
3492 | ret = get_inode_info(sctx->send_root, di_key.objectid, NULL, | |
3493 | &right_gen, NULL, NULL, NULL, NULL); | |
3494 | if (ret < 0) { | |
3495 | if (ret == -ENOENT) | |
3496 | ret = 0; | |
3497 | goto out; | |
3498 | } | |
3499 | ||
3500 | /* Different inode, no need to delay the rename of sctx->cur_ino */ | |
3501 | if (right_gen != left_gen) { | |
3502 | ret = 0; | |
3503 | goto out; | |
3504 | } | |
3505 | ||
801bec36 RK |
3506 | wdm = get_waiting_dir_move(sctx, di_key.objectid); |
3507 | if (wdm && !wdm->orphanized) { | |
84471e24 FM |
3508 | ret = add_pending_dir_move(sctx, |
3509 | sctx->cur_ino, | |
3510 | sctx->cur_inode_gen, | |
3511 | di_key.objectid, | |
3512 | &sctx->new_refs, | |
3513 | &sctx->deleted_refs, | |
3514 | is_orphan); | |
3515 | if (!ret) | |
3516 | ret = 1; | |
3517 | } | |
3518 | out: | |
3519 | btrfs_free_path(path); | |
3520 | return ret; | |
3521 | } | |
3522 | ||
80aa6027 | 3523 | /* |
ea37d599 FM |
3524 | * Check if inode ino2, or any of its ancestors, is inode ino1. |
3525 | * Return 1 if true, 0 if false and < 0 on error. | |
3526 | */ | |
3527 | static int check_ino_in_path(struct btrfs_root *root, | |
3528 | const u64 ino1, | |
3529 | const u64 ino1_gen, | |
3530 | const u64 ino2, | |
3531 | const u64 ino2_gen, | |
3532 | struct fs_path *fs_path) | |
3533 | { | |
3534 | u64 ino = ino2; | |
3535 | ||
3536 | if (ino1 == ino2) | |
3537 | return ino1_gen == ino2_gen; | |
3538 | ||
3539 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { | |
3540 | u64 parent; | |
3541 | u64 parent_gen; | |
3542 | int ret; | |
3543 | ||
3544 | fs_path_reset(fs_path); | |
3545 | ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path); | |
3546 | if (ret < 0) | |
3547 | return ret; | |
3548 | if (parent == ino1) | |
3549 | return parent_gen == ino1_gen; | |
3550 | ino = parent; | |
3551 | } | |
3552 | return 0; | |
3553 | } | |
3554 | ||
3555 | /* | |
3556 | * Check if ino ino1 is an ancestor of inode ino2 in the given root for any | |
3557 | * possible path (in case ino2 is not a directory and has multiple hard links). | |
80aa6027 FM |
3558 | * Return 1 if true, 0 if false and < 0 on error. |
3559 | */ | |
3560 | static int is_ancestor(struct btrfs_root *root, | |
3561 | const u64 ino1, | |
3562 | const u64 ino1_gen, | |
3563 | const u64 ino2, | |
3564 | struct fs_path *fs_path) | |
3565 | { | |
ea37d599 | 3566 | bool free_fs_path = false; |
72c3668f | 3567 | int ret = 0; |
ea37d599 FM |
3568 | struct btrfs_path *path = NULL; |
3569 | struct btrfs_key key; | |
72c3668f FM |
3570 | |
3571 | if (!fs_path) { | |
3572 | fs_path = fs_path_alloc(); | |
3573 | if (!fs_path) | |
3574 | return -ENOMEM; | |
ea37d599 | 3575 | free_fs_path = true; |
72c3668f | 3576 | } |
80aa6027 | 3577 | |
ea37d599 FM |
3578 | path = alloc_path_for_send(); |
3579 | if (!path) { | |
3580 | ret = -ENOMEM; | |
3581 | goto out; | |
3582 | } | |
80aa6027 | 3583 | |
ea37d599 FM |
3584 | key.objectid = ino2; |
3585 | key.type = BTRFS_INODE_REF_KEY; | |
3586 | key.offset = 0; | |
3587 | ||
3588 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3589 | if (ret < 0) | |
3590 | goto out; | |
3591 | ||
3592 | while (true) { | |
3593 | struct extent_buffer *leaf = path->nodes[0]; | |
3594 | int slot = path->slots[0]; | |
3595 | u32 cur_offset = 0; | |
3596 | u32 item_size; | |
3597 | ||
3598 | if (slot >= btrfs_header_nritems(leaf)) { | |
3599 | ret = btrfs_next_leaf(root, path); | |
3600 | if (ret < 0) | |
3601 | goto out; | |
3602 | if (ret > 0) | |
3603 | break; | |
3604 | continue; | |
72c3668f | 3605 | } |
ea37d599 FM |
3606 | |
3607 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
3608 | if (key.objectid != ino2) | |
3609 | break; | |
3610 | if (key.type != BTRFS_INODE_REF_KEY && | |
3611 | key.type != BTRFS_INODE_EXTREF_KEY) | |
3612 | break; | |
3613 | ||
3614 | item_size = btrfs_item_size_nr(leaf, slot); | |
3615 | while (cur_offset < item_size) { | |
3616 | u64 parent; | |
3617 | u64 parent_gen; | |
3618 | ||
3619 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
3620 | unsigned long ptr; | |
3621 | struct btrfs_inode_extref *extref; | |
3622 | ||
3623 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3624 | extref = (struct btrfs_inode_extref *) | |
3625 | (ptr + cur_offset); | |
3626 | parent = btrfs_inode_extref_parent(leaf, | |
3627 | extref); | |
3628 | cur_offset += sizeof(*extref); | |
3629 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
3630 | extref); | |
3631 | } else { | |
3632 | parent = key.offset; | |
3633 | cur_offset = item_size; | |
3634 | } | |
3635 | ||
3636 | ret = get_inode_info(root, parent, NULL, &parent_gen, | |
3637 | NULL, NULL, NULL, NULL); | |
3638 | if (ret < 0) | |
3639 | goto out; | |
3640 | ret = check_ino_in_path(root, ino1, ino1_gen, | |
3641 | parent, parent_gen, fs_path); | |
3642 | if (ret) | |
3643 | goto out; | |
80aa6027 | 3644 | } |
ea37d599 | 3645 | path->slots[0]++; |
80aa6027 | 3646 | } |
ea37d599 | 3647 | ret = 0; |
72c3668f | 3648 | out: |
ea37d599 FM |
3649 | btrfs_free_path(path); |
3650 | if (free_fs_path) | |
72c3668f FM |
3651 | fs_path_free(fs_path); |
3652 | return ret; | |
80aa6027 FM |
3653 | } |
3654 | ||
9f03740a | 3655 | static int wait_for_parent_move(struct send_ctx *sctx, |
8b191a68 FM |
3656 | struct recorded_ref *parent_ref, |
3657 | const bool is_orphan) | |
9f03740a | 3658 | { |
f959492f | 3659 | int ret = 0; |
9f03740a | 3660 | u64 ino = parent_ref->dir; |
fe9c798d | 3661 | u64 ino_gen = parent_ref->dir_gen; |
9f03740a | 3662 | u64 parent_ino_before, parent_ino_after; |
9f03740a FDBM |
3663 | struct fs_path *path_before = NULL; |
3664 | struct fs_path *path_after = NULL; | |
3665 | int len1, len2; | |
9f03740a FDBM |
3666 | |
3667 | path_after = fs_path_alloc(); | |
f959492f FM |
3668 | path_before = fs_path_alloc(); |
3669 | if (!path_after || !path_before) { | |
9f03740a FDBM |
3670 | ret = -ENOMEM; |
3671 | goto out; | |
3672 | } | |
3673 | ||
bfa7e1f8 | 3674 | /* |
f959492f FM |
3675 | * Our current directory inode may not yet be renamed/moved because some |
3676 | * ancestor (immediate or not) has to be renamed/moved first. So find if | |
3677 | * such ancestor exists and make sure our own rename/move happens after | |
80aa6027 FM |
3678 | * that ancestor is processed to avoid path build infinite loops (done |
3679 | * at get_cur_path()). | |
bfa7e1f8 | 3680 | */ |
f959492f | 3681 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { |
fe9c798d FM |
3682 | u64 parent_ino_after_gen; |
3683 | ||
f959492f | 3684 | if (is_waiting_for_move(sctx, ino)) { |
80aa6027 FM |
3685 | /* |
3686 | * If the current inode is an ancestor of ino in the | |
3687 | * parent root, we need to delay the rename of the | |
3688 | * current inode, otherwise don't delayed the rename | |
3689 | * because we can end up with a circular dependency | |
3690 | * of renames, resulting in some directories never | |
3691 | * getting the respective rename operations issued in | |
3692 | * the send stream or getting into infinite path build | |
3693 | * loops. | |
3694 | */ | |
3695 | ret = is_ancestor(sctx->parent_root, | |
3696 | sctx->cur_ino, sctx->cur_inode_gen, | |
3697 | ino, path_before); | |
4122ea64 FM |
3698 | if (ret) |
3699 | break; | |
f959492f | 3700 | } |
bfa7e1f8 FM |
3701 | |
3702 | fs_path_reset(path_before); | |
3703 | fs_path_reset(path_after); | |
3704 | ||
3705 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
fe9c798d | 3706 | &parent_ino_after_gen, path_after); |
bfa7e1f8 FM |
3707 | if (ret < 0) |
3708 | goto out; | |
3709 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
3710 | NULL, path_before); | |
f959492f | 3711 | if (ret < 0 && ret != -ENOENT) { |
bfa7e1f8 | 3712 | goto out; |
f959492f | 3713 | } else if (ret == -ENOENT) { |
bf8e8ca6 | 3714 | ret = 0; |
f959492f | 3715 | break; |
bfa7e1f8 FM |
3716 | } |
3717 | ||
3718 | len1 = fs_path_len(path_before); | |
3719 | len2 = fs_path_len(path_after); | |
f959492f FM |
3720 | if (ino > sctx->cur_ino && |
3721 | (parent_ino_before != parent_ino_after || len1 != len2 || | |
3722 | memcmp(path_before->start, path_after->start, len1))) { | |
fe9c798d FM |
3723 | u64 parent_ino_gen; |
3724 | ||
3725 | ret = get_inode_info(sctx->parent_root, ino, NULL, | |
3726 | &parent_ino_gen, NULL, NULL, NULL, | |
3727 | NULL); | |
3728 | if (ret < 0) | |
3729 | goto out; | |
3730 | if (ino_gen == parent_ino_gen) { | |
3731 | ret = 1; | |
3732 | break; | |
3733 | } | |
bfa7e1f8 | 3734 | } |
bfa7e1f8 | 3735 | ino = parent_ino_after; |
fe9c798d | 3736 | ino_gen = parent_ino_after_gen; |
bfa7e1f8 FM |
3737 | } |
3738 | ||
9f03740a FDBM |
3739 | out: |
3740 | fs_path_free(path_before); | |
3741 | fs_path_free(path_after); | |
3742 | ||
f959492f FM |
3743 | if (ret == 1) { |
3744 | ret = add_pending_dir_move(sctx, | |
3745 | sctx->cur_ino, | |
3746 | sctx->cur_inode_gen, | |
3747 | ino, | |
3748 | &sctx->new_refs, | |
84471e24 | 3749 | &sctx->deleted_refs, |
8b191a68 | 3750 | is_orphan); |
f959492f FM |
3751 | if (!ret) |
3752 | ret = 1; | |
3753 | } | |
3754 | ||
9f03740a FDBM |
3755 | return ret; |
3756 | } | |
3757 | ||
f5962781 FM |
3758 | static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) |
3759 | { | |
3760 | int ret; | |
3761 | struct fs_path *new_path; | |
3762 | ||
3763 | /* | |
3764 | * Our reference's name member points to its full_path member string, so | |
3765 | * we use here a new path. | |
3766 | */ | |
3767 | new_path = fs_path_alloc(); | |
3768 | if (!new_path) | |
3769 | return -ENOMEM; | |
3770 | ||
3771 | ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path); | |
3772 | if (ret < 0) { | |
3773 | fs_path_free(new_path); | |
3774 | return ret; | |
3775 | } | |
3776 | ret = fs_path_add(new_path, ref->name, ref->name_len); | |
3777 | if (ret < 0) { | |
3778 | fs_path_free(new_path); | |
3779 | return ret; | |
3780 | } | |
3781 | ||
3782 | fs_path_free(ref->full_path); | |
3783 | set_ref_path(ref, new_path); | |
3784 | ||
3785 | return 0; | |
3786 | } | |
3787 | ||
31db9f7c AB |
3788 | /* |
3789 | * This does all the move/link/unlink/rmdir magic. | |
3790 | */ | |
9f03740a | 3791 | static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) |
31db9f7c | 3792 | { |
04ab956e | 3793 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
3794 | int ret = 0; |
3795 | struct recorded_ref *cur; | |
1f4692da | 3796 | struct recorded_ref *cur2; |
ba5e8f2e | 3797 | struct list_head check_dirs; |
31db9f7c | 3798 | struct fs_path *valid_path = NULL; |
b24baf69 | 3799 | u64 ow_inode = 0; |
31db9f7c | 3800 | u64 ow_gen; |
f5962781 | 3801 | u64 ow_mode; |
31db9f7c AB |
3802 | int did_overwrite = 0; |
3803 | int is_orphan = 0; | |
29d6d30f | 3804 | u64 last_dir_ino_rm = 0; |
84471e24 | 3805 | bool can_rename = true; |
f5962781 | 3806 | bool orphanized_dir = false; |
fdb13889 | 3807 | bool orphanized_ancestor = false; |
31db9f7c | 3808 | |
04ab956e | 3809 | btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino); |
31db9f7c | 3810 | |
6d85ed05 AB |
3811 | /* |
3812 | * This should never happen as the root dir always has the same ref | |
3813 | * which is always '..' | |
3814 | */ | |
3815 | BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); | |
ba5e8f2e | 3816 | INIT_LIST_HEAD(&check_dirs); |
6d85ed05 | 3817 | |
924794c9 | 3818 | valid_path = fs_path_alloc(); |
31db9f7c AB |
3819 | if (!valid_path) { |
3820 | ret = -ENOMEM; | |
3821 | goto out; | |
3822 | } | |
3823 | ||
31db9f7c AB |
3824 | /* |
3825 | * First, check if the first ref of the current inode was overwritten | |
3826 | * before. If yes, we know that the current inode was already orphanized | |
3827 | * and thus use the orphan name. If not, we can use get_cur_path to | |
3828 | * get the path of the first ref as it would like while receiving at | |
3829 | * this point in time. | |
3830 | * New inodes are always orphan at the beginning, so force to use the | |
3831 | * orphan name in this case. | |
3832 | * The first ref is stored in valid_path and will be updated if it | |
3833 | * gets moved around. | |
3834 | */ | |
3835 | if (!sctx->cur_inode_new) { | |
3836 | ret = did_overwrite_first_ref(sctx, sctx->cur_ino, | |
3837 | sctx->cur_inode_gen); | |
3838 | if (ret < 0) | |
3839 | goto out; | |
3840 | if (ret) | |
3841 | did_overwrite = 1; | |
3842 | } | |
3843 | if (sctx->cur_inode_new || did_overwrite) { | |
3844 | ret = gen_unique_name(sctx, sctx->cur_ino, | |
3845 | sctx->cur_inode_gen, valid_path); | |
3846 | if (ret < 0) | |
3847 | goto out; | |
3848 | is_orphan = 1; | |
3849 | } else { | |
3850 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
3851 | valid_path); | |
3852 | if (ret < 0) | |
3853 | goto out; | |
3854 | } | |
3855 | ||
3856 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
1f4692da AB |
3857 | /* |
3858 | * We may have refs where the parent directory does not exist | |
3859 | * yet. This happens if the parent directories inum is higher | |
3860 | * the the current inum. To handle this case, we create the | |
3861 | * parent directory out of order. But we need to check if this | |
3862 | * did already happen before due to other refs in the same dir. | |
3863 | */ | |
3864 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); | |
3865 | if (ret < 0) | |
3866 | goto out; | |
3867 | if (ret == inode_state_will_create) { | |
3868 | ret = 0; | |
3869 | /* | |
3870 | * First check if any of the current inodes refs did | |
3871 | * already create the dir. | |
3872 | */ | |
3873 | list_for_each_entry(cur2, &sctx->new_refs, list) { | |
3874 | if (cur == cur2) | |
3875 | break; | |
3876 | if (cur2->dir == cur->dir) { | |
3877 | ret = 1; | |
3878 | break; | |
3879 | } | |
3880 | } | |
3881 | ||
3882 | /* | |
3883 | * If that did not happen, check if a previous inode | |
3884 | * did already create the dir. | |
3885 | */ | |
3886 | if (!ret) | |
3887 | ret = did_create_dir(sctx, cur->dir); | |
3888 | if (ret < 0) | |
3889 | goto out; | |
3890 | if (!ret) { | |
3891 | ret = send_create_inode(sctx, cur->dir); | |
3892 | if (ret < 0) | |
3893 | goto out; | |
3894 | } | |
3895 | } | |
3896 | ||
31db9f7c AB |
3897 | /* |
3898 | * Check if this new ref would overwrite the first ref of | |
3899 | * another unprocessed inode. If yes, orphanize the | |
3900 | * overwritten inode. If we find an overwritten ref that is | |
3901 | * not the first ref, simply unlink it. | |
3902 | */ | |
3903 | ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
3904 | cur->name, cur->name_len, | |
f5962781 | 3905 | &ow_inode, &ow_gen, &ow_mode); |
31db9f7c AB |
3906 | if (ret < 0) |
3907 | goto out; | |
3908 | if (ret) { | |
924794c9 TI |
3909 | ret = is_first_ref(sctx->parent_root, |
3910 | ow_inode, cur->dir, cur->name, | |
3911 | cur->name_len); | |
31db9f7c AB |
3912 | if (ret < 0) |
3913 | goto out; | |
3914 | if (ret) { | |
8996a48c | 3915 | struct name_cache_entry *nce; |
801bec36 | 3916 | struct waiting_dir_move *wdm; |
8996a48c | 3917 | |
31db9f7c AB |
3918 | ret = orphanize_inode(sctx, ow_inode, ow_gen, |
3919 | cur->full_path); | |
3920 | if (ret < 0) | |
3921 | goto out; | |
f5962781 FM |
3922 | if (S_ISDIR(ow_mode)) |
3923 | orphanized_dir = true; | |
801bec36 RK |
3924 | |
3925 | /* | |
3926 | * If ow_inode has its rename operation delayed | |
3927 | * make sure that its orphanized name is used in | |
3928 | * the source path when performing its rename | |
3929 | * operation. | |
3930 | */ | |
3931 | if (is_waiting_for_move(sctx, ow_inode)) { | |
3932 | wdm = get_waiting_dir_move(sctx, | |
3933 | ow_inode); | |
3934 | ASSERT(wdm); | |
3935 | wdm->orphanized = true; | |
3936 | } | |
3937 | ||
8996a48c FM |
3938 | /* |
3939 | * Make sure we clear our orphanized inode's | |
3940 | * name from the name cache. This is because the | |
3941 | * inode ow_inode might be an ancestor of some | |
3942 | * other inode that will be orphanized as well | |
3943 | * later and has an inode number greater than | |
3944 | * sctx->send_progress. We need to prevent | |
3945 | * future name lookups from using the old name | |
3946 | * and get instead the orphan name. | |
3947 | */ | |
3948 | nce = name_cache_search(sctx, ow_inode, ow_gen); | |
3949 | if (nce) { | |
3950 | name_cache_delete(sctx, nce); | |
3951 | kfree(nce); | |
3952 | } | |
801bec36 RK |
3953 | |
3954 | /* | |
3955 | * ow_inode might currently be an ancestor of | |
3956 | * cur_ino, therefore compute valid_path (the | |
3957 | * current path of cur_ino) again because it | |
3958 | * might contain the pre-orphanization name of | |
3959 | * ow_inode, which is no longer valid. | |
3960 | */ | |
72c3668f FM |
3961 | ret = is_ancestor(sctx->parent_root, |
3962 | ow_inode, ow_gen, | |
3963 | sctx->cur_ino, NULL); | |
3964 | if (ret > 0) { | |
fdb13889 | 3965 | orphanized_ancestor = true; |
72c3668f FM |
3966 | fs_path_reset(valid_path); |
3967 | ret = get_cur_path(sctx, sctx->cur_ino, | |
3968 | sctx->cur_inode_gen, | |
3969 | valid_path); | |
3970 | } | |
801bec36 RK |
3971 | if (ret < 0) |
3972 | goto out; | |
31db9f7c AB |
3973 | } else { |
3974 | ret = send_unlink(sctx, cur->full_path); | |
3975 | if (ret < 0) | |
3976 | goto out; | |
3977 | } | |
3978 | } | |
3979 | ||
84471e24 FM |
3980 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { |
3981 | ret = wait_for_dest_dir_move(sctx, cur, is_orphan); | |
3982 | if (ret < 0) | |
3983 | goto out; | |
3984 | if (ret == 1) { | |
3985 | can_rename = false; | |
3986 | *pending_move = 1; | |
3987 | } | |
3988 | } | |
3989 | ||
8b191a68 FM |
3990 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root && |
3991 | can_rename) { | |
3992 | ret = wait_for_parent_move(sctx, cur, is_orphan); | |
3993 | if (ret < 0) | |
3994 | goto out; | |
3995 | if (ret == 1) { | |
3996 | can_rename = false; | |
3997 | *pending_move = 1; | |
3998 | } | |
3999 | } | |
4000 | ||
31db9f7c AB |
4001 | /* |
4002 | * link/move the ref to the new place. If we have an orphan | |
4003 | * inode, move it and update valid_path. If not, link or move | |
4004 | * it depending on the inode mode. | |
4005 | */ | |
84471e24 | 4006 | if (is_orphan && can_rename) { |
31db9f7c AB |
4007 | ret = send_rename(sctx, valid_path, cur->full_path); |
4008 | if (ret < 0) | |
4009 | goto out; | |
4010 | is_orphan = 0; | |
4011 | ret = fs_path_copy(valid_path, cur->full_path); | |
4012 | if (ret < 0) | |
4013 | goto out; | |
84471e24 | 4014 | } else if (can_rename) { |
31db9f7c AB |
4015 | if (S_ISDIR(sctx->cur_inode_mode)) { |
4016 | /* | |
4017 | * Dirs can't be linked, so move it. For moved | |
4018 | * dirs, we always have one new and one deleted | |
4019 | * ref. The deleted ref is ignored later. | |
4020 | */ | |
8b191a68 FM |
4021 | ret = send_rename(sctx, valid_path, |
4022 | cur->full_path); | |
4023 | if (!ret) | |
4024 | ret = fs_path_copy(valid_path, | |
4025 | cur->full_path); | |
31db9f7c AB |
4026 | if (ret < 0) |
4027 | goto out; | |
4028 | } else { | |
f5962781 FM |
4029 | /* |
4030 | * We might have previously orphanized an inode | |
4031 | * which is an ancestor of our current inode, | |
4032 | * so our reference's full path, which was | |
4033 | * computed before any such orphanizations, must | |
4034 | * be updated. | |
4035 | */ | |
4036 | if (orphanized_dir) { | |
4037 | ret = update_ref_path(sctx, cur); | |
4038 | if (ret < 0) | |
4039 | goto out; | |
4040 | } | |
31db9f7c AB |
4041 | ret = send_link(sctx, cur->full_path, |
4042 | valid_path); | |
4043 | if (ret < 0) | |
4044 | goto out; | |
4045 | } | |
4046 | } | |
ba5e8f2e | 4047 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4048 | if (ret < 0) |
4049 | goto out; | |
4050 | } | |
4051 | ||
4052 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { | |
4053 | /* | |
4054 | * Check if we can already rmdir the directory. If not, | |
4055 | * orphanize it. For every dir item inside that gets deleted | |
4056 | * later, we do this check again and rmdir it then if possible. | |
4057 | * See the use of check_dirs for more details. | |
4058 | */ | |
9dc44214 FM |
4059 | ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
4060 | sctx->cur_ino); | |
31db9f7c AB |
4061 | if (ret < 0) |
4062 | goto out; | |
4063 | if (ret) { | |
4064 | ret = send_rmdir(sctx, valid_path); | |
4065 | if (ret < 0) | |
4066 | goto out; | |
4067 | } else if (!is_orphan) { | |
4068 | ret = orphanize_inode(sctx, sctx->cur_ino, | |
4069 | sctx->cur_inode_gen, valid_path); | |
4070 | if (ret < 0) | |
4071 | goto out; | |
4072 | is_orphan = 1; | |
4073 | } | |
4074 | ||
4075 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
ba5e8f2e | 4076 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4077 | if (ret < 0) |
4078 | goto out; | |
4079 | } | |
ccf1626b AB |
4080 | } else if (S_ISDIR(sctx->cur_inode_mode) && |
4081 | !list_empty(&sctx->deleted_refs)) { | |
4082 | /* | |
4083 | * We have a moved dir. Add the old parent to check_dirs | |
4084 | */ | |
4085 | cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, | |
4086 | list); | |
ba5e8f2e | 4087 | ret = dup_ref(cur, &check_dirs); |
ccf1626b AB |
4088 | if (ret < 0) |
4089 | goto out; | |
31db9f7c AB |
4090 | } else if (!S_ISDIR(sctx->cur_inode_mode)) { |
4091 | /* | |
4092 | * We have a non dir inode. Go through all deleted refs and | |
4093 | * unlink them if they were not already overwritten by other | |
4094 | * inodes. | |
4095 | */ | |
4096 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
4097 | ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
4098 | sctx->cur_ino, sctx->cur_inode_gen, | |
4099 | cur->name, cur->name_len); | |
4100 | if (ret < 0) | |
4101 | goto out; | |
4102 | if (!ret) { | |
fdb13889 FM |
4103 | /* |
4104 | * If we orphanized any ancestor before, we need | |
4105 | * to recompute the full path for deleted names, | |
4106 | * since any such path was computed before we | |
4107 | * processed any references and orphanized any | |
4108 | * ancestor inode. | |
4109 | */ | |
4110 | if (orphanized_ancestor) { | |
f5962781 FM |
4111 | ret = update_ref_path(sctx, cur); |
4112 | if (ret < 0) | |
fdb13889 | 4113 | goto out; |
fdb13889 | 4114 | } |
1f4692da AB |
4115 | ret = send_unlink(sctx, cur->full_path); |
4116 | if (ret < 0) | |
4117 | goto out; | |
31db9f7c | 4118 | } |
ba5e8f2e | 4119 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4120 | if (ret < 0) |
4121 | goto out; | |
4122 | } | |
31db9f7c AB |
4123 | /* |
4124 | * If the inode is still orphan, unlink the orphan. This may | |
4125 | * happen when a previous inode did overwrite the first ref | |
4126 | * of this inode and no new refs were added for the current | |
766702ef AB |
4127 | * inode. Unlinking does not mean that the inode is deleted in |
4128 | * all cases. There may still be links to this inode in other | |
4129 | * places. | |
31db9f7c | 4130 | */ |
1f4692da | 4131 | if (is_orphan) { |
31db9f7c AB |
4132 | ret = send_unlink(sctx, valid_path); |
4133 | if (ret < 0) | |
4134 | goto out; | |
4135 | } | |
4136 | } | |
4137 | ||
4138 | /* | |
4139 | * We did collect all parent dirs where cur_inode was once located. We | |
4140 | * now go through all these dirs and check if they are pending for | |
4141 | * deletion and if it's finally possible to perform the rmdir now. | |
4142 | * We also update the inode stats of the parent dirs here. | |
4143 | */ | |
ba5e8f2e | 4144 | list_for_each_entry(cur, &check_dirs, list) { |
766702ef AB |
4145 | /* |
4146 | * In case we had refs into dirs that were not processed yet, | |
4147 | * we don't need to do the utime and rmdir logic for these dirs. | |
4148 | * The dir will be processed later. | |
4149 | */ | |
ba5e8f2e | 4150 | if (cur->dir > sctx->cur_ino) |
31db9f7c AB |
4151 | continue; |
4152 | ||
ba5e8f2e | 4153 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4154 | if (ret < 0) |
4155 | goto out; | |
4156 | ||
4157 | if (ret == inode_state_did_create || | |
4158 | ret == inode_state_no_change) { | |
4159 | /* TODO delayed utimes */ | |
ba5e8f2e | 4160 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4161 | if (ret < 0) |
4162 | goto out; | |
29d6d30f FM |
4163 | } else if (ret == inode_state_did_delete && |
4164 | cur->dir != last_dir_ino_rm) { | |
9dc44214 FM |
4165 | ret = can_rmdir(sctx, cur->dir, cur->dir_gen, |
4166 | sctx->cur_ino); | |
31db9f7c AB |
4167 | if (ret < 0) |
4168 | goto out; | |
4169 | if (ret) { | |
ba5e8f2e JB |
4170 | ret = get_cur_path(sctx, cur->dir, |
4171 | cur->dir_gen, valid_path); | |
31db9f7c AB |
4172 | if (ret < 0) |
4173 | goto out; | |
4174 | ret = send_rmdir(sctx, valid_path); | |
4175 | if (ret < 0) | |
4176 | goto out; | |
29d6d30f | 4177 | last_dir_ino_rm = cur->dir; |
31db9f7c AB |
4178 | } |
4179 | } | |
4180 | } | |
4181 | ||
31db9f7c AB |
4182 | ret = 0; |
4183 | ||
4184 | out: | |
ba5e8f2e | 4185 | __free_recorded_refs(&check_dirs); |
31db9f7c | 4186 | free_recorded_refs(sctx); |
924794c9 | 4187 | fs_path_free(valid_path); |
31db9f7c AB |
4188 | return ret; |
4189 | } | |
4190 | ||
a0357511 NB |
4191 | static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name, |
4192 | void *ctx, struct list_head *refs) | |
31db9f7c AB |
4193 | { |
4194 | int ret = 0; | |
4195 | struct send_ctx *sctx = ctx; | |
4196 | struct fs_path *p; | |
4197 | u64 gen; | |
4198 | ||
924794c9 | 4199 | p = fs_path_alloc(); |
31db9f7c AB |
4200 | if (!p) |
4201 | return -ENOMEM; | |
4202 | ||
a4d96d62 | 4203 | ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, |
85a7b33b | 4204 | NULL, NULL); |
31db9f7c AB |
4205 | if (ret < 0) |
4206 | goto out; | |
4207 | ||
31db9f7c AB |
4208 | ret = get_cur_path(sctx, dir, gen, p); |
4209 | if (ret < 0) | |
4210 | goto out; | |
4211 | ret = fs_path_add_path(p, name); | |
4212 | if (ret < 0) | |
4213 | goto out; | |
4214 | ||
a4d96d62 | 4215 | ret = __record_ref(refs, dir, gen, p); |
31db9f7c AB |
4216 | |
4217 | out: | |
4218 | if (ret) | |
924794c9 | 4219 | fs_path_free(p); |
31db9f7c AB |
4220 | return ret; |
4221 | } | |
4222 | ||
a4d96d62 LB |
4223 | static int __record_new_ref(int num, u64 dir, int index, |
4224 | struct fs_path *name, | |
4225 | void *ctx) | |
4226 | { | |
4227 | struct send_ctx *sctx = ctx; | |
a0357511 | 4228 | return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs); |
a4d96d62 LB |
4229 | } |
4230 | ||
4231 | ||
31db9f7c AB |
4232 | static int __record_deleted_ref(int num, u64 dir, int index, |
4233 | struct fs_path *name, | |
4234 | void *ctx) | |
4235 | { | |
31db9f7c | 4236 | struct send_ctx *sctx = ctx; |
a0357511 NB |
4237 | return record_ref(sctx->parent_root, dir, name, ctx, |
4238 | &sctx->deleted_refs); | |
31db9f7c AB |
4239 | } |
4240 | ||
4241 | static int record_new_ref(struct send_ctx *sctx) | |
4242 | { | |
4243 | int ret; | |
4244 | ||
924794c9 TI |
4245 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
4246 | sctx->cmp_key, 0, __record_new_ref, sctx); | |
31db9f7c AB |
4247 | if (ret < 0) |
4248 | goto out; | |
4249 | ret = 0; | |
4250 | ||
4251 | out: | |
4252 | return ret; | |
4253 | } | |
4254 | ||
4255 | static int record_deleted_ref(struct send_ctx *sctx) | |
4256 | { | |
4257 | int ret; | |
4258 | ||
924794c9 TI |
4259 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
4260 | sctx->cmp_key, 0, __record_deleted_ref, sctx); | |
31db9f7c AB |
4261 | if (ret < 0) |
4262 | goto out; | |
4263 | ret = 0; | |
4264 | ||
4265 | out: | |
4266 | return ret; | |
4267 | } | |
4268 | ||
4269 | struct find_ref_ctx { | |
4270 | u64 dir; | |
ba5e8f2e JB |
4271 | u64 dir_gen; |
4272 | struct btrfs_root *root; | |
31db9f7c AB |
4273 | struct fs_path *name; |
4274 | int found_idx; | |
4275 | }; | |
4276 | ||
4277 | static int __find_iref(int num, u64 dir, int index, | |
4278 | struct fs_path *name, | |
4279 | void *ctx_) | |
4280 | { | |
4281 | struct find_ref_ctx *ctx = ctx_; | |
ba5e8f2e JB |
4282 | u64 dir_gen; |
4283 | int ret; | |
31db9f7c AB |
4284 | |
4285 | if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && | |
4286 | strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { | |
ba5e8f2e JB |
4287 | /* |
4288 | * To avoid doing extra lookups we'll only do this if everything | |
4289 | * else matches. | |
4290 | */ | |
4291 | ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, | |
4292 | NULL, NULL, NULL); | |
4293 | if (ret) | |
4294 | return ret; | |
4295 | if (dir_gen != ctx->dir_gen) | |
4296 | return 0; | |
31db9f7c AB |
4297 | ctx->found_idx = num; |
4298 | return 1; | |
4299 | } | |
4300 | return 0; | |
4301 | } | |
4302 | ||
924794c9 | 4303 | static int find_iref(struct btrfs_root *root, |
31db9f7c AB |
4304 | struct btrfs_path *path, |
4305 | struct btrfs_key *key, | |
ba5e8f2e | 4306 | u64 dir, u64 dir_gen, struct fs_path *name) |
31db9f7c AB |
4307 | { |
4308 | int ret; | |
4309 | struct find_ref_ctx ctx; | |
4310 | ||
4311 | ctx.dir = dir; | |
4312 | ctx.name = name; | |
ba5e8f2e | 4313 | ctx.dir_gen = dir_gen; |
31db9f7c | 4314 | ctx.found_idx = -1; |
ba5e8f2e | 4315 | ctx.root = root; |
31db9f7c | 4316 | |
924794c9 | 4317 | ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); |
31db9f7c AB |
4318 | if (ret < 0) |
4319 | return ret; | |
4320 | ||
4321 | if (ctx.found_idx == -1) | |
4322 | return -ENOENT; | |
4323 | ||
4324 | return ctx.found_idx; | |
4325 | } | |
4326 | ||
4327 | static int __record_changed_new_ref(int num, u64 dir, int index, | |
4328 | struct fs_path *name, | |
4329 | void *ctx) | |
4330 | { | |
ba5e8f2e | 4331 | u64 dir_gen; |
31db9f7c AB |
4332 | int ret; |
4333 | struct send_ctx *sctx = ctx; | |
4334 | ||
ba5e8f2e JB |
4335 | ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, |
4336 | NULL, NULL, NULL); | |
4337 | if (ret) | |
4338 | return ret; | |
4339 | ||
924794c9 | 4340 | ret = find_iref(sctx->parent_root, sctx->right_path, |
ba5e8f2e | 4341 | sctx->cmp_key, dir, dir_gen, name); |
31db9f7c AB |
4342 | if (ret == -ENOENT) |
4343 | ret = __record_new_ref(num, dir, index, name, sctx); | |
4344 | else if (ret > 0) | |
4345 | ret = 0; | |
4346 | ||
4347 | return ret; | |
4348 | } | |
4349 | ||
4350 | static int __record_changed_deleted_ref(int num, u64 dir, int index, | |
4351 | struct fs_path *name, | |
4352 | void *ctx) | |
4353 | { | |
ba5e8f2e | 4354 | u64 dir_gen; |
31db9f7c AB |
4355 | int ret; |
4356 | struct send_ctx *sctx = ctx; | |
4357 | ||
ba5e8f2e JB |
4358 | ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, |
4359 | NULL, NULL, NULL); | |
4360 | if (ret) | |
4361 | return ret; | |
4362 | ||
924794c9 | 4363 | ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, |
ba5e8f2e | 4364 | dir, dir_gen, name); |
31db9f7c AB |
4365 | if (ret == -ENOENT) |
4366 | ret = __record_deleted_ref(num, dir, index, name, sctx); | |
4367 | else if (ret > 0) | |
4368 | ret = 0; | |
4369 | ||
4370 | return ret; | |
4371 | } | |
4372 | ||
4373 | static int record_changed_ref(struct send_ctx *sctx) | |
4374 | { | |
4375 | int ret = 0; | |
4376 | ||
924794c9 | 4377 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
31db9f7c AB |
4378 | sctx->cmp_key, 0, __record_changed_new_ref, sctx); |
4379 | if (ret < 0) | |
4380 | goto out; | |
924794c9 | 4381 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
4382 | sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); |
4383 | if (ret < 0) | |
4384 | goto out; | |
4385 | ret = 0; | |
4386 | ||
4387 | out: | |
4388 | return ret; | |
4389 | } | |
4390 | ||
4391 | /* | |
4392 | * Record and process all refs at once. Needed when an inode changes the | |
4393 | * generation number, which means that it was deleted and recreated. | |
4394 | */ | |
4395 | static int process_all_refs(struct send_ctx *sctx, | |
4396 | enum btrfs_compare_tree_result cmd) | |
4397 | { | |
4398 | int ret; | |
4399 | struct btrfs_root *root; | |
4400 | struct btrfs_path *path; | |
4401 | struct btrfs_key key; | |
4402 | struct btrfs_key found_key; | |
4403 | struct extent_buffer *eb; | |
4404 | int slot; | |
4405 | iterate_inode_ref_t cb; | |
9f03740a | 4406 | int pending_move = 0; |
31db9f7c AB |
4407 | |
4408 | path = alloc_path_for_send(); | |
4409 | if (!path) | |
4410 | return -ENOMEM; | |
4411 | ||
4412 | if (cmd == BTRFS_COMPARE_TREE_NEW) { | |
4413 | root = sctx->send_root; | |
4414 | cb = __record_new_ref; | |
4415 | } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { | |
4416 | root = sctx->parent_root; | |
4417 | cb = __record_deleted_ref; | |
4418 | } else { | |
4d1a63b2 DS |
4419 | btrfs_err(sctx->send_root->fs_info, |
4420 | "Wrong command %d in process_all_refs", cmd); | |
4421 | ret = -EINVAL; | |
4422 | goto out; | |
31db9f7c AB |
4423 | } |
4424 | ||
4425 | key.objectid = sctx->cmp_key->objectid; | |
4426 | key.type = BTRFS_INODE_REF_KEY; | |
4427 | key.offset = 0; | |
dff6d0ad FDBM |
4428 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4429 | if (ret < 0) | |
4430 | goto out; | |
31db9f7c | 4431 | |
dff6d0ad | 4432 | while (1) { |
31db9f7c AB |
4433 | eb = path->nodes[0]; |
4434 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4435 | if (slot >= btrfs_header_nritems(eb)) { |
4436 | ret = btrfs_next_leaf(root, path); | |
4437 | if (ret < 0) | |
4438 | goto out; | |
4439 | else if (ret > 0) | |
4440 | break; | |
4441 | continue; | |
4442 | } | |
4443 | ||
31db9f7c AB |
4444 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
4445 | ||
4446 | if (found_key.objectid != key.objectid || | |
96b5bd77 JS |
4447 | (found_key.type != BTRFS_INODE_REF_KEY && |
4448 | found_key.type != BTRFS_INODE_EXTREF_KEY)) | |
31db9f7c | 4449 | break; |
31db9f7c | 4450 | |
924794c9 | 4451 | ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); |
31db9f7c AB |
4452 | if (ret < 0) |
4453 | goto out; | |
4454 | ||
dff6d0ad | 4455 | path->slots[0]++; |
31db9f7c | 4456 | } |
e938c8ad | 4457 | btrfs_release_path(path); |
31db9f7c | 4458 | |
3dc09ec8 JB |
4459 | /* |
4460 | * We don't actually care about pending_move as we are simply | |
4461 | * re-creating this inode and will be rename'ing it into place once we | |
4462 | * rename the parent directory. | |
4463 | */ | |
9f03740a | 4464 | ret = process_recorded_refs(sctx, &pending_move); |
31db9f7c AB |
4465 | out: |
4466 | btrfs_free_path(path); | |
4467 | return ret; | |
4468 | } | |
4469 | ||
4470 | static int send_set_xattr(struct send_ctx *sctx, | |
4471 | struct fs_path *path, | |
4472 | const char *name, int name_len, | |
4473 | const char *data, int data_len) | |
4474 | { | |
4475 | int ret = 0; | |
4476 | ||
4477 | ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); | |
4478 | if (ret < 0) | |
4479 | goto out; | |
4480 | ||
4481 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4482 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4483 | TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); | |
4484 | ||
4485 | ret = send_cmd(sctx); | |
4486 | ||
4487 | tlv_put_failure: | |
4488 | out: | |
4489 | return ret; | |
4490 | } | |
4491 | ||
4492 | static int send_remove_xattr(struct send_ctx *sctx, | |
4493 | struct fs_path *path, | |
4494 | const char *name, int name_len) | |
4495 | { | |
4496 | int ret = 0; | |
4497 | ||
4498 | ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); | |
4499 | if (ret < 0) | |
4500 | goto out; | |
4501 | ||
4502 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4503 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4504 | ||
4505 | ret = send_cmd(sctx); | |
4506 | ||
4507 | tlv_put_failure: | |
4508 | out: | |
4509 | return ret; | |
4510 | } | |
4511 | ||
4512 | static int __process_new_xattr(int num, struct btrfs_key *di_key, | |
4513 | const char *name, int name_len, | |
4514 | const char *data, int data_len, | |
4515 | u8 type, void *ctx) | |
4516 | { | |
4517 | int ret; | |
4518 | struct send_ctx *sctx = ctx; | |
4519 | struct fs_path *p; | |
2211d5ba | 4520 | struct posix_acl_xattr_header dummy_acl; |
31db9f7c | 4521 | |
924794c9 | 4522 | p = fs_path_alloc(); |
31db9f7c AB |
4523 | if (!p) |
4524 | return -ENOMEM; | |
4525 | ||
4526 | /* | |
01327610 | 4527 | * This hack is needed because empty acls are stored as zero byte |
31db9f7c | 4528 | * data in xattrs. Problem with that is, that receiving these zero byte |
01327610 | 4529 | * acls will fail later. To fix this, we send a dummy acl list that |
31db9f7c AB |
4530 | * only contains the version number and no entries. |
4531 | */ | |
4532 | if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || | |
4533 | !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { | |
4534 | if (data_len == 0) { | |
4535 | dummy_acl.a_version = | |
4536 | cpu_to_le32(POSIX_ACL_XATTR_VERSION); | |
4537 | data = (char *)&dummy_acl; | |
4538 | data_len = sizeof(dummy_acl); | |
4539 | } | |
4540 | } | |
4541 | ||
4542 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4543 | if (ret < 0) | |
4544 | goto out; | |
4545 | ||
4546 | ret = send_set_xattr(sctx, p, name, name_len, data, data_len); | |
4547 | ||
4548 | out: | |
924794c9 | 4549 | fs_path_free(p); |
31db9f7c AB |
4550 | return ret; |
4551 | } | |
4552 | ||
4553 | static int __process_deleted_xattr(int num, struct btrfs_key *di_key, | |
4554 | const char *name, int name_len, | |
4555 | const char *data, int data_len, | |
4556 | u8 type, void *ctx) | |
4557 | { | |
4558 | int ret; | |
4559 | struct send_ctx *sctx = ctx; | |
4560 | struct fs_path *p; | |
4561 | ||
924794c9 | 4562 | p = fs_path_alloc(); |
31db9f7c AB |
4563 | if (!p) |
4564 | return -ENOMEM; | |
4565 | ||
4566 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4567 | if (ret < 0) | |
4568 | goto out; | |
4569 | ||
4570 | ret = send_remove_xattr(sctx, p, name, name_len); | |
4571 | ||
4572 | out: | |
924794c9 | 4573 | fs_path_free(p); |
31db9f7c AB |
4574 | return ret; |
4575 | } | |
4576 | ||
4577 | static int process_new_xattr(struct send_ctx *sctx) | |
4578 | { | |
4579 | int ret = 0; | |
4580 | ||
924794c9 | 4581 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4582 | __process_new_xattr, sctx); |
31db9f7c AB |
4583 | |
4584 | return ret; | |
4585 | } | |
4586 | ||
4587 | static int process_deleted_xattr(struct send_ctx *sctx) | |
4588 | { | |
e2c89907 | 4589 | return iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4590 | __process_deleted_xattr, sctx); |
31db9f7c AB |
4591 | } |
4592 | ||
4593 | struct find_xattr_ctx { | |
4594 | const char *name; | |
4595 | int name_len; | |
4596 | int found_idx; | |
4597 | char *found_data; | |
4598 | int found_data_len; | |
4599 | }; | |
4600 | ||
4601 | static int __find_xattr(int num, struct btrfs_key *di_key, | |
4602 | const char *name, int name_len, | |
4603 | const char *data, int data_len, | |
4604 | u8 type, void *vctx) | |
4605 | { | |
4606 | struct find_xattr_ctx *ctx = vctx; | |
4607 | ||
4608 | if (name_len == ctx->name_len && | |
4609 | strncmp(name, ctx->name, name_len) == 0) { | |
4610 | ctx->found_idx = num; | |
4611 | ctx->found_data_len = data_len; | |
e780b0d1 | 4612 | ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); |
31db9f7c AB |
4613 | if (!ctx->found_data) |
4614 | return -ENOMEM; | |
31db9f7c AB |
4615 | return 1; |
4616 | } | |
4617 | return 0; | |
4618 | } | |
4619 | ||
924794c9 | 4620 | static int find_xattr(struct btrfs_root *root, |
31db9f7c AB |
4621 | struct btrfs_path *path, |
4622 | struct btrfs_key *key, | |
4623 | const char *name, int name_len, | |
4624 | char **data, int *data_len) | |
4625 | { | |
4626 | int ret; | |
4627 | struct find_xattr_ctx ctx; | |
4628 | ||
4629 | ctx.name = name; | |
4630 | ctx.name_len = name_len; | |
4631 | ctx.found_idx = -1; | |
4632 | ctx.found_data = NULL; | |
4633 | ctx.found_data_len = 0; | |
4634 | ||
a0357511 | 4635 | ret = iterate_dir_item(root, path, __find_xattr, &ctx); |
31db9f7c AB |
4636 | if (ret < 0) |
4637 | return ret; | |
4638 | ||
4639 | if (ctx.found_idx == -1) | |
4640 | return -ENOENT; | |
4641 | if (data) { | |
4642 | *data = ctx.found_data; | |
4643 | *data_len = ctx.found_data_len; | |
4644 | } else { | |
4645 | kfree(ctx.found_data); | |
4646 | } | |
4647 | return ctx.found_idx; | |
4648 | } | |
4649 | ||
4650 | ||
4651 | static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, | |
4652 | const char *name, int name_len, | |
4653 | const char *data, int data_len, | |
4654 | u8 type, void *ctx) | |
4655 | { | |
4656 | int ret; | |
4657 | struct send_ctx *sctx = ctx; | |
4658 | char *found_data = NULL; | |
4659 | int found_data_len = 0; | |
31db9f7c | 4660 | |
924794c9 TI |
4661 | ret = find_xattr(sctx->parent_root, sctx->right_path, |
4662 | sctx->cmp_key, name, name_len, &found_data, | |
4663 | &found_data_len); | |
31db9f7c AB |
4664 | if (ret == -ENOENT) { |
4665 | ret = __process_new_xattr(num, di_key, name, name_len, data, | |
4666 | data_len, type, ctx); | |
4667 | } else if (ret >= 0) { | |
4668 | if (data_len != found_data_len || | |
4669 | memcmp(data, found_data, data_len)) { | |
4670 | ret = __process_new_xattr(num, di_key, name, name_len, | |
4671 | data, data_len, type, ctx); | |
4672 | } else { | |
4673 | ret = 0; | |
4674 | } | |
4675 | } | |
4676 | ||
4677 | kfree(found_data); | |
31db9f7c AB |
4678 | return ret; |
4679 | } | |
4680 | ||
4681 | static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, | |
4682 | const char *name, int name_len, | |
4683 | const char *data, int data_len, | |
4684 | u8 type, void *ctx) | |
4685 | { | |
4686 | int ret; | |
4687 | struct send_ctx *sctx = ctx; | |
4688 | ||
924794c9 TI |
4689 | ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, |
4690 | name, name_len, NULL, NULL); | |
31db9f7c AB |
4691 | if (ret == -ENOENT) |
4692 | ret = __process_deleted_xattr(num, di_key, name, name_len, data, | |
4693 | data_len, type, ctx); | |
4694 | else if (ret >= 0) | |
4695 | ret = 0; | |
4696 | ||
4697 | return ret; | |
4698 | } | |
4699 | ||
4700 | static int process_changed_xattr(struct send_ctx *sctx) | |
4701 | { | |
4702 | int ret = 0; | |
4703 | ||
924794c9 | 4704 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4705 | __process_changed_new_xattr, sctx); |
31db9f7c AB |
4706 | if (ret < 0) |
4707 | goto out; | |
924794c9 | 4708 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4709 | __process_changed_deleted_xattr, sctx); |
31db9f7c AB |
4710 | |
4711 | out: | |
4712 | return ret; | |
4713 | } | |
4714 | ||
4715 | static int process_all_new_xattrs(struct send_ctx *sctx) | |
4716 | { | |
4717 | int ret; | |
4718 | struct btrfs_root *root; | |
4719 | struct btrfs_path *path; | |
4720 | struct btrfs_key key; | |
4721 | struct btrfs_key found_key; | |
4722 | struct extent_buffer *eb; | |
4723 | int slot; | |
4724 | ||
4725 | path = alloc_path_for_send(); | |
4726 | if (!path) | |
4727 | return -ENOMEM; | |
4728 | ||
4729 | root = sctx->send_root; | |
4730 | ||
4731 | key.objectid = sctx->cmp_key->objectid; | |
4732 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4733 | key.offset = 0; | |
dff6d0ad FDBM |
4734 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4735 | if (ret < 0) | |
4736 | goto out; | |
31db9f7c | 4737 | |
dff6d0ad | 4738 | while (1) { |
31db9f7c AB |
4739 | eb = path->nodes[0]; |
4740 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4741 | if (slot >= btrfs_header_nritems(eb)) { |
4742 | ret = btrfs_next_leaf(root, path); | |
4743 | if (ret < 0) { | |
4744 | goto out; | |
4745 | } else if (ret > 0) { | |
4746 | ret = 0; | |
4747 | break; | |
4748 | } | |
4749 | continue; | |
4750 | } | |
31db9f7c | 4751 | |
dff6d0ad | 4752 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
31db9f7c AB |
4753 | if (found_key.objectid != key.objectid || |
4754 | found_key.type != key.type) { | |
4755 | ret = 0; | |
4756 | goto out; | |
4757 | } | |
4758 | ||
a0357511 | 4759 | ret = iterate_dir_item(root, path, __process_new_xattr, sctx); |
31db9f7c AB |
4760 | if (ret < 0) |
4761 | goto out; | |
4762 | ||
dff6d0ad | 4763 | path->slots[0]++; |
31db9f7c AB |
4764 | } |
4765 | ||
4766 | out: | |
4767 | btrfs_free_path(path); | |
4768 | return ret; | |
4769 | } | |
4770 | ||
ed259095 JB |
4771 | static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len) |
4772 | { | |
4773 | struct btrfs_root *root = sctx->send_root; | |
4774 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4775 | struct inode *inode; | |
4776 | struct page *page; | |
4777 | char *addr; | |
4778 | struct btrfs_key key; | |
09cbfeaf | 4779 | pgoff_t index = offset >> PAGE_SHIFT; |
ed259095 | 4780 | pgoff_t last_index; |
09cbfeaf | 4781 | unsigned pg_offset = offset & ~PAGE_MASK; |
ed259095 JB |
4782 | ssize_t ret = 0; |
4783 | ||
4784 | key.objectid = sctx->cur_ino; | |
4785 | key.type = BTRFS_INODE_ITEM_KEY; | |
4786 | key.offset = 0; | |
4787 | ||
4788 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
4789 | if (IS_ERR(inode)) | |
4790 | return PTR_ERR(inode); | |
4791 | ||
4792 | if (offset + len > i_size_read(inode)) { | |
4793 | if (offset > i_size_read(inode)) | |
4794 | len = 0; | |
4795 | else | |
4796 | len = offset - i_size_read(inode); | |
4797 | } | |
4798 | if (len == 0) | |
4799 | goto out; | |
4800 | ||
09cbfeaf | 4801 | last_index = (offset + len - 1) >> PAGE_SHIFT; |
2131bcd3 LB |
4802 | |
4803 | /* initial readahead */ | |
4804 | memset(&sctx->ra, 0, sizeof(struct file_ra_state)); | |
4805 | file_ra_state_init(&sctx->ra, inode->i_mapping); | |
2131bcd3 | 4806 | |
ed259095 JB |
4807 | while (index <= last_index) { |
4808 | unsigned cur_len = min_t(unsigned, len, | |
09cbfeaf | 4809 | PAGE_SIZE - pg_offset); |
eef16ba2 KH |
4810 | |
4811 | page = find_lock_page(inode->i_mapping, index); | |
ed259095 | 4812 | if (!page) { |
eef16ba2 KH |
4813 | page_cache_sync_readahead(inode->i_mapping, &sctx->ra, |
4814 | NULL, index, last_index + 1 - index); | |
4815 | ||
4816 | page = find_or_create_page(inode->i_mapping, index, | |
4817 | GFP_KERNEL); | |
4818 | if (!page) { | |
4819 | ret = -ENOMEM; | |
4820 | break; | |
4821 | } | |
4822 | } | |
4823 | ||
4824 | if (PageReadahead(page)) { | |
4825 | page_cache_async_readahead(inode->i_mapping, &sctx->ra, | |
4826 | NULL, page, index, last_index + 1 - index); | |
ed259095 JB |
4827 | } |
4828 | ||
4829 | if (!PageUptodate(page)) { | |
4830 | btrfs_readpage(NULL, page); | |
4831 | lock_page(page); | |
4832 | if (!PageUptodate(page)) { | |
4833 | unlock_page(page); | |
09cbfeaf | 4834 | put_page(page); |
ed259095 JB |
4835 | ret = -EIO; |
4836 | break; | |
4837 | } | |
4838 | } | |
4839 | ||
4840 | addr = kmap(page); | |
4841 | memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len); | |
4842 | kunmap(page); | |
4843 | unlock_page(page); | |
09cbfeaf | 4844 | put_page(page); |
ed259095 JB |
4845 | index++; |
4846 | pg_offset = 0; | |
4847 | len -= cur_len; | |
4848 | ret += cur_len; | |
4849 | } | |
4850 | out: | |
4851 | iput(inode); | |
4852 | return ret; | |
4853 | } | |
4854 | ||
31db9f7c AB |
4855 | /* |
4856 | * Read some bytes from the current inode/file and send a write command to | |
4857 | * user space. | |
4858 | */ | |
4859 | static int send_write(struct send_ctx *sctx, u64 offset, u32 len) | |
4860 | { | |
04ab956e | 4861 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
4862 | int ret = 0; |
4863 | struct fs_path *p; | |
ed259095 | 4864 | ssize_t num_read = 0; |
31db9f7c | 4865 | |
924794c9 | 4866 | p = fs_path_alloc(); |
31db9f7c AB |
4867 | if (!p) |
4868 | return -ENOMEM; | |
4869 | ||
04ab956e | 4870 | btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len); |
31db9f7c | 4871 | |
ed259095 JB |
4872 | num_read = fill_read_buf(sctx, offset, len); |
4873 | if (num_read <= 0) { | |
4874 | if (num_read < 0) | |
4875 | ret = num_read; | |
31db9f7c | 4876 | goto out; |
ed259095 | 4877 | } |
31db9f7c AB |
4878 | |
4879 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
4880 | if (ret < 0) | |
4881 | goto out; | |
4882 | ||
4883 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4884 | if (ret < 0) | |
4885 | goto out; | |
4886 | ||
4887 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4888 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
e938c8ad | 4889 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read); |
31db9f7c AB |
4890 | |
4891 | ret = send_cmd(sctx); | |
4892 | ||
4893 | tlv_put_failure: | |
4894 | out: | |
924794c9 | 4895 | fs_path_free(p); |
31db9f7c AB |
4896 | if (ret < 0) |
4897 | return ret; | |
e938c8ad | 4898 | return num_read; |
31db9f7c AB |
4899 | } |
4900 | ||
4901 | /* | |
4902 | * Send a clone command to user space. | |
4903 | */ | |
4904 | static int send_clone(struct send_ctx *sctx, | |
4905 | u64 offset, u32 len, | |
4906 | struct clone_root *clone_root) | |
4907 | { | |
4908 | int ret = 0; | |
31db9f7c AB |
4909 | struct fs_path *p; |
4910 | u64 gen; | |
4911 | ||
04ab956e JM |
4912 | btrfs_debug(sctx->send_root->fs_info, |
4913 | "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu", | |
4914 | offset, len, clone_root->root->objectid, clone_root->ino, | |
4915 | clone_root->offset); | |
31db9f7c | 4916 | |
924794c9 | 4917 | p = fs_path_alloc(); |
31db9f7c AB |
4918 | if (!p) |
4919 | return -ENOMEM; | |
4920 | ||
4921 | ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); | |
4922 | if (ret < 0) | |
4923 | goto out; | |
4924 | ||
4925 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4926 | if (ret < 0) | |
4927 | goto out; | |
4928 | ||
4929 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4930 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); | |
4931 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4932 | ||
e938c8ad | 4933 | if (clone_root->root == sctx->send_root) { |
31db9f7c | 4934 | ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, |
85a7b33b | 4935 | &gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
4936 | if (ret < 0) |
4937 | goto out; | |
4938 | ret = get_cur_path(sctx, clone_root->ino, gen, p); | |
4939 | } else { | |
924794c9 | 4940 | ret = get_inode_path(clone_root->root, clone_root->ino, p); |
31db9f7c AB |
4941 | } |
4942 | if (ret < 0) | |
4943 | goto out; | |
4944 | ||
37b8d27d JB |
4945 | /* |
4946 | * If the parent we're using has a received_uuid set then use that as | |
4947 | * our clone source as that is what we will look for when doing a | |
4948 | * receive. | |
4949 | * | |
4950 | * This covers the case that we create a snapshot off of a received | |
4951 | * subvolume and then use that as the parent and try to receive on a | |
4952 | * different host. | |
4953 | */ | |
4954 | if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid)) | |
4955 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4956 | clone_root->root->root_item.received_uuid); | |
4957 | else | |
4958 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4959 | clone_root->root->root_item.uuid); | |
31db9f7c | 4960 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 4961 | le64_to_cpu(clone_root->root->root_item.ctransid)); |
31db9f7c AB |
4962 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); |
4963 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, | |
4964 | clone_root->offset); | |
4965 | ||
4966 | ret = send_cmd(sctx); | |
4967 | ||
4968 | tlv_put_failure: | |
4969 | out: | |
924794c9 | 4970 | fs_path_free(p); |
31db9f7c AB |
4971 | return ret; |
4972 | } | |
4973 | ||
cb95e7bf MF |
4974 | /* |
4975 | * Send an update extent command to user space. | |
4976 | */ | |
4977 | static int send_update_extent(struct send_ctx *sctx, | |
4978 | u64 offset, u32 len) | |
4979 | { | |
4980 | int ret = 0; | |
4981 | struct fs_path *p; | |
4982 | ||
924794c9 | 4983 | p = fs_path_alloc(); |
cb95e7bf MF |
4984 | if (!p) |
4985 | return -ENOMEM; | |
4986 | ||
4987 | ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); | |
4988 | if (ret < 0) | |
4989 | goto out; | |
4990 | ||
4991 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4992 | if (ret < 0) | |
4993 | goto out; | |
4994 | ||
4995 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4996 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4997 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); | |
4998 | ||
4999 | ret = send_cmd(sctx); | |
5000 | ||
5001 | tlv_put_failure: | |
5002 | out: | |
924794c9 | 5003 | fs_path_free(p); |
cb95e7bf MF |
5004 | return ret; |
5005 | } | |
5006 | ||
16e7549f JB |
5007 | static int send_hole(struct send_ctx *sctx, u64 end) |
5008 | { | |
5009 | struct fs_path *p = NULL; | |
5010 | u64 offset = sctx->cur_inode_last_extent; | |
5011 | u64 len; | |
5012 | int ret = 0; | |
5013 | ||
5014 | p = fs_path_alloc(); | |
5015 | if (!p) | |
5016 | return -ENOMEM; | |
c715e155 FM |
5017 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); |
5018 | if (ret < 0) | |
5019 | goto tlv_put_failure; | |
16e7549f JB |
5020 | memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE); |
5021 | while (offset < end) { | |
5022 | len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE); | |
5023 | ||
5024 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
16e7549f JB |
5025 | if (ret < 0) |
5026 | break; | |
5027 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5028 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5029 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len); | |
5030 | ret = send_cmd(sctx); | |
5031 | if (ret < 0) | |
5032 | break; | |
5033 | offset += len; | |
5034 | } | |
5035 | tlv_put_failure: | |
5036 | fs_path_free(p); | |
5037 | return ret; | |
5038 | } | |
5039 | ||
d906d49f FM |
5040 | static int send_extent_data(struct send_ctx *sctx, |
5041 | const u64 offset, | |
5042 | const u64 len) | |
5043 | { | |
5044 | u64 sent = 0; | |
5045 | ||
5046 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) | |
5047 | return send_update_extent(sctx, offset, len); | |
5048 | ||
5049 | while (sent < len) { | |
5050 | u64 size = len - sent; | |
5051 | int ret; | |
5052 | ||
5053 | if (size > BTRFS_SEND_READ_SIZE) | |
5054 | size = BTRFS_SEND_READ_SIZE; | |
5055 | ret = send_write(sctx, offset + sent, size); | |
5056 | if (ret < 0) | |
5057 | return ret; | |
5058 | if (!ret) | |
5059 | break; | |
5060 | sent += ret; | |
5061 | } | |
5062 | return 0; | |
5063 | } | |
5064 | ||
5065 | static int clone_range(struct send_ctx *sctx, | |
5066 | struct clone_root *clone_root, | |
5067 | const u64 disk_byte, | |
5068 | u64 data_offset, | |
5069 | u64 offset, | |
5070 | u64 len) | |
5071 | { | |
5072 | struct btrfs_path *path; | |
5073 | struct btrfs_key key; | |
5074 | int ret; | |
5075 | ||
72610b1b FM |
5076 | /* |
5077 | * Prevent cloning from a zero offset with a length matching the sector | |
5078 | * size because in some scenarios this will make the receiver fail. | |
5079 | * | |
5080 | * For example, if in the source filesystem the extent at offset 0 | |
5081 | * has a length of sectorsize and it was written using direct IO, then | |
5082 | * it can never be an inline extent (even if compression is enabled). | |
5083 | * Then this extent can be cloned in the original filesystem to a non | |
5084 | * zero file offset, but it may not be possible to clone in the | |
5085 | * destination filesystem because it can be inlined due to compression | |
5086 | * on the destination filesystem (as the receiver's write operations are | |
5087 | * always done using buffered IO). The same happens when the original | |
5088 | * filesystem does not have compression enabled but the destination | |
5089 | * filesystem has. | |
5090 | */ | |
5091 | if (clone_root->offset == 0 && | |
5092 | len == sctx->send_root->fs_info->sectorsize) | |
5093 | return send_extent_data(sctx, offset, len); | |
5094 | ||
d906d49f FM |
5095 | path = alloc_path_for_send(); |
5096 | if (!path) | |
5097 | return -ENOMEM; | |
5098 | ||
5099 | /* | |
5100 | * We can't send a clone operation for the entire range if we find | |
5101 | * extent items in the respective range in the source file that | |
5102 | * refer to different extents or if we find holes. | |
5103 | * So check for that and do a mix of clone and regular write/copy | |
5104 | * operations if needed. | |
5105 | * | |
5106 | * Example: | |
5107 | * | |
5108 | * mkfs.btrfs -f /dev/sda | |
5109 | * mount /dev/sda /mnt | |
5110 | * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo | |
5111 | * cp --reflink=always /mnt/foo /mnt/bar | |
5112 | * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo | |
5113 | * btrfs subvolume snapshot -r /mnt /mnt/snap | |
5114 | * | |
5115 | * If when we send the snapshot and we are processing file bar (which | |
5116 | * has a higher inode number than foo) we blindly send a clone operation | |
5117 | * for the [0, 100K[ range from foo to bar, the receiver ends up getting | |
5118 | * a file bar that matches the content of file foo - iow, doesn't match | |
5119 | * the content from bar in the original filesystem. | |
5120 | */ | |
5121 | key.objectid = clone_root->ino; | |
5122 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5123 | key.offset = clone_root->offset; | |
5124 | ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); | |
5125 | if (ret < 0) | |
5126 | goto out; | |
5127 | if (ret > 0 && path->slots[0] > 0) { | |
5128 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
5129 | if (key.objectid == clone_root->ino && | |
5130 | key.type == BTRFS_EXTENT_DATA_KEY) | |
5131 | path->slots[0]--; | |
5132 | } | |
5133 | ||
5134 | while (true) { | |
5135 | struct extent_buffer *leaf = path->nodes[0]; | |
5136 | int slot = path->slots[0]; | |
5137 | struct btrfs_file_extent_item *ei; | |
5138 | u8 type; | |
5139 | u64 ext_len; | |
5140 | u64 clone_len; | |
5141 | ||
5142 | if (slot >= btrfs_header_nritems(leaf)) { | |
5143 | ret = btrfs_next_leaf(clone_root->root, path); | |
5144 | if (ret < 0) | |
5145 | goto out; | |
5146 | else if (ret > 0) | |
5147 | break; | |
5148 | continue; | |
5149 | } | |
5150 | ||
5151 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5152 | ||
5153 | /* | |
5154 | * We might have an implicit trailing hole (NO_HOLES feature | |
5155 | * enabled). We deal with it after leaving this loop. | |
5156 | */ | |
5157 | if (key.objectid != clone_root->ino || | |
5158 | key.type != BTRFS_EXTENT_DATA_KEY) | |
5159 | break; | |
5160 | ||
5161 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5162 | type = btrfs_file_extent_type(leaf, ei); | |
5163 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
5164 | ext_len = btrfs_file_extent_inline_len(leaf, slot, ei); | |
09cbfeaf | 5165 | ext_len = PAGE_ALIGN(ext_len); |
d906d49f FM |
5166 | } else { |
5167 | ext_len = btrfs_file_extent_num_bytes(leaf, ei); | |
5168 | } | |
5169 | ||
5170 | if (key.offset + ext_len <= clone_root->offset) | |
5171 | goto next; | |
5172 | ||
5173 | if (key.offset > clone_root->offset) { | |
5174 | /* Implicit hole, NO_HOLES feature enabled. */ | |
5175 | u64 hole_len = key.offset - clone_root->offset; | |
5176 | ||
5177 | if (hole_len > len) | |
5178 | hole_len = len; | |
5179 | ret = send_extent_data(sctx, offset, hole_len); | |
5180 | if (ret < 0) | |
5181 | goto out; | |
5182 | ||
5183 | len -= hole_len; | |
5184 | if (len == 0) | |
5185 | break; | |
5186 | offset += hole_len; | |
5187 | clone_root->offset += hole_len; | |
5188 | data_offset += hole_len; | |
5189 | } | |
5190 | ||
5191 | if (key.offset >= clone_root->offset + len) | |
5192 | break; | |
5193 | ||
5194 | clone_len = min_t(u64, ext_len, len); | |
5195 | ||
5196 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && | |
5197 | btrfs_file_extent_offset(leaf, ei) == data_offset) | |
5198 | ret = send_clone(sctx, offset, clone_len, clone_root); | |
5199 | else | |
5200 | ret = send_extent_data(sctx, offset, clone_len); | |
5201 | ||
5202 | if (ret < 0) | |
5203 | goto out; | |
5204 | ||
5205 | len -= clone_len; | |
5206 | if (len == 0) | |
5207 | break; | |
5208 | offset += clone_len; | |
5209 | clone_root->offset += clone_len; | |
5210 | data_offset += clone_len; | |
5211 | next: | |
5212 | path->slots[0]++; | |
5213 | } | |
5214 | ||
5215 | if (len > 0) | |
5216 | ret = send_extent_data(sctx, offset, len); | |
5217 | else | |
5218 | ret = 0; | |
5219 | out: | |
5220 | btrfs_free_path(path); | |
5221 | return ret; | |
5222 | } | |
5223 | ||
31db9f7c AB |
5224 | static int send_write_or_clone(struct send_ctx *sctx, |
5225 | struct btrfs_path *path, | |
5226 | struct btrfs_key *key, | |
5227 | struct clone_root *clone_root) | |
5228 | { | |
5229 | int ret = 0; | |
5230 | struct btrfs_file_extent_item *ei; | |
5231 | u64 offset = key->offset; | |
31db9f7c | 5232 | u64 len; |
31db9f7c | 5233 | u8 type; |
28e5dd8f | 5234 | u64 bs = sctx->send_root->fs_info->sb->s_blocksize; |
31db9f7c AB |
5235 | |
5236 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5237 | struct btrfs_file_extent_item); | |
5238 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
74dd17fb | 5239 | if (type == BTRFS_FILE_EXTENT_INLINE) { |
514ac8ad CM |
5240 | len = btrfs_file_extent_inline_len(path->nodes[0], |
5241 | path->slots[0], ei); | |
74dd17fb CM |
5242 | /* |
5243 | * it is possible the inline item won't cover the whole page, | |
5244 | * but there may be items after this page. Make | |
5245 | * sure to send the whole thing | |
5246 | */ | |
09cbfeaf | 5247 | len = PAGE_ALIGN(len); |
74dd17fb | 5248 | } else { |
31db9f7c | 5249 | len = btrfs_file_extent_num_bytes(path->nodes[0], ei); |
74dd17fb | 5250 | } |
31db9f7c AB |
5251 | |
5252 | if (offset + len > sctx->cur_inode_size) | |
5253 | len = sctx->cur_inode_size - offset; | |
5254 | if (len == 0) { | |
5255 | ret = 0; | |
5256 | goto out; | |
5257 | } | |
5258 | ||
28e5dd8f | 5259 | if (clone_root && IS_ALIGNED(offset + len, bs)) { |
d906d49f FM |
5260 | u64 disk_byte; |
5261 | u64 data_offset; | |
5262 | ||
5263 | disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); | |
5264 | data_offset = btrfs_file_extent_offset(path->nodes[0], ei); | |
5265 | ret = clone_range(sctx, clone_root, disk_byte, data_offset, | |
5266 | offset, len); | |
cb95e7bf | 5267 | } else { |
d906d49f | 5268 | ret = send_extent_data(sctx, offset, len); |
31db9f7c | 5269 | } |
31db9f7c AB |
5270 | out: |
5271 | return ret; | |
5272 | } | |
5273 | ||
5274 | static int is_extent_unchanged(struct send_ctx *sctx, | |
5275 | struct btrfs_path *left_path, | |
5276 | struct btrfs_key *ekey) | |
5277 | { | |
5278 | int ret = 0; | |
5279 | struct btrfs_key key; | |
5280 | struct btrfs_path *path = NULL; | |
5281 | struct extent_buffer *eb; | |
5282 | int slot; | |
5283 | struct btrfs_key found_key; | |
5284 | struct btrfs_file_extent_item *ei; | |
5285 | u64 left_disknr; | |
5286 | u64 right_disknr; | |
5287 | u64 left_offset; | |
5288 | u64 right_offset; | |
5289 | u64 left_offset_fixed; | |
5290 | u64 left_len; | |
5291 | u64 right_len; | |
74dd17fb CM |
5292 | u64 left_gen; |
5293 | u64 right_gen; | |
31db9f7c AB |
5294 | u8 left_type; |
5295 | u8 right_type; | |
5296 | ||
5297 | path = alloc_path_for_send(); | |
5298 | if (!path) | |
5299 | return -ENOMEM; | |
5300 | ||
5301 | eb = left_path->nodes[0]; | |
5302 | slot = left_path->slots[0]; | |
31db9f7c AB |
5303 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
5304 | left_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
5305 | |
5306 | if (left_type != BTRFS_FILE_EXTENT_REG) { | |
5307 | ret = 0; | |
5308 | goto out; | |
5309 | } | |
74dd17fb CM |
5310 | left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5311 | left_len = btrfs_file_extent_num_bytes(eb, ei); | |
5312 | left_offset = btrfs_file_extent_offset(eb, ei); | |
5313 | left_gen = btrfs_file_extent_generation(eb, ei); | |
31db9f7c AB |
5314 | |
5315 | /* | |
5316 | * Following comments will refer to these graphics. L is the left | |
5317 | * extents which we are checking at the moment. 1-8 are the right | |
5318 | * extents that we iterate. | |
5319 | * | |
5320 | * |-----L-----| | |
5321 | * |-1-|-2a-|-3-|-4-|-5-|-6-| | |
5322 | * | |
5323 | * |-----L-----| | |
5324 | * |--1--|-2b-|...(same as above) | |
5325 | * | |
5326 | * Alternative situation. Happens on files where extents got split. | |
5327 | * |-----L-----| | |
5328 | * |-----------7-----------|-6-| | |
5329 | * | |
5330 | * Alternative situation. Happens on files which got larger. | |
5331 | * |-----L-----| | |
5332 | * |-8-| | |
5333 | * Nothing follows after 8. | |
5334 | */ | |
5335 | ||
5336 | key.objectid = ekey->objectid; | |
5337 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5338 | key.offset = ekey->offset; | |
5339 | ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); | |
5340 | if (ret < 0) | |
5341 | goto out; | |
5342 | if (ret) { | |
5343 | ret = 0; | |
5344 | goto out; | |
5345 | } | |
5346 | ||
5347 | /* | |
5348 | * Handle special case where the right side has no extents at all. | |
5349 | */ | |
5350 | eb = path->nodes[0]; | |
5351 | slot = path->slots[0]; | |
5352 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5353 | if (found_key.objectid != key.objectid || | |
5354 | found_key.type != key.type) { | |
57cfd462 JB |
5355 | /* If we're a hole then just pretend nothing changed */ |
5356 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5357 | goto out; |
5358 | } | |
5359 | ||
5360 | /* | |
5361 | * We're now on 2a, 2b or 7. | |
5362 | */ | |
5363 | key = found_key; | |
5364 | while (key.offset < ekey->offset + left_len) { | |
5365 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
5366 | right_type = btrfs_file_extent_type(eb, ei); | |
e1cbfd7b FM |
5367 | if (right_type != BTRFS_FILE_EXTENT_REG && |
5368 | right_type != BTRFS_FILE_EXTENT_INLINE) { | |
31db9f7c AB |
5369 | ret = 0; |
5370 | goto out; | |
5371 | } | |
5372 | ||
e1cbfd7b FM |
5373 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { |
5374 | right_len = btrfs_file_extent_inline_len(eb, slot, ei); | |
5375 | right_len = PAGE_ALIGN(right_len); | |
5376 | } else { | |
5377 | right_len = btrfs_file_extent_num_bytes(eb, ei); | |
5378 | } | |
007d31f7 | 5379 | |
31db9f7c AB |
5380 | /* |
5381 | * Are we at extent 8? If yes, we know the extent is changed. | |
5382 | * This may only happen on the first iteration. | |
5383 | */ | |
d8347fa4 | 5384 | if (found_key.offset + right_len <= ekey->offset) { |
57cfd462 JB |
5385 | /* If we're a hole just pretend nothing changed */ |
5386 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5387 | goto out; |
5388 | } | |
5389 | ||
e1cbfd7b FM |
5390 | /* |
5391 | * We just wanted to see if when we have an inline extent, what | |
5392 | * follows it is a regular extent (wanted to check the above | |
5393 | * condition for inline extents too). This should normally not | |
5394 | * happen but it's possible for example when we have an inline | |
5395 | * compressed extent representing data with a size matching | |
5396 | * the page size (currently the same as sector size). | |
5397 | */ | |
5398 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { | |
5399 | ret = 0; | |
5400 | goto out; | |
5401 | } | |
5402 | ||
24e52b11 FM |
5403 | right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5404 | right_offset = btrfs_file_extent_offset(eb, ei); | |
5405 | right_gen = btrfs_file_extent_generation(eb, ei); | |
5406 | ||
31db9f7c AB |
5407 | left_offset_fixed = left_offset; |
5408 | if (key.offset < ekey->offset) { | |
5409 | /* Fix the right offset for 2a and 7. */ | |
5410 | right_offset += ekey->offset - key.offset; | |
5411 | } else { | |
5412 | /* Fix the left offset for all behind 2a and 2b */ | |
5413 | left_offset_fixed += key.offset - ekey->offset; | |
5414 | } | |
5415 | ||
5416 | /* | |
5417 | * Check if we have the same extent. | |
5418 | */ | |
3954096d | 5419 | if (left_disknr != right_disknr || |
74dd17fb CM |
5420 | left_offset_fixed != right_offset || |
5421 | left_gen != right_gen) { | |
31db9f7c AB |
5422 | ret = 0; |
5423 | goto out; | |
5424 | } | |
5425 | ||
5426 | /* | |
5427 | * Go to the next extent. | |
5428 | */ | |
5429 | ret = btrfs_next_item(sctx->parent_root, path); | |
5430 | if (ret < 0) | |
5431 | goto out; | |
5432 | if (!ret) { | |
5433 | eb = path->nodes[0]; | |
5434 | slot = path->slots[0]; | |
5435 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5436 | } | |
5437 | if (ret || found_key.objectid != key.objectid || | |
5438 | found_key.type != key.type) { | |
5439 | key.offset += right_len; | |
5440 | break; | |
adaa4b8e JS |
5441 | } |
5442 | if (found_key.offset != key.offset + right_len) { | |
5443 | ret = 0; | |
5444 | goto out; | |
31db9f7c AB |
5445 | } |
5446 | key = found_key; | |
5447 | } | |
5448 | ||
5449 | /* | |
5450 | * We're now behind the left extent (treat as unchanged) or at the end | |
5451 | * of the right side (treat as changed). | |
5452 | */ | |
5453 | if (key.offset >= ekey->offset + left_len) | |
5454 | ret = 1; | |
5455 | else | |
5456 | ret = 0; | |
5457 | ||
5458 | ||
5459 | out: | |
5460 | btrfs_free_path(path); | |
5461 | return ret; | |
5462 | } | |
5463 | ||
16e7549f JB |
5464 | static int get_last_extent(struct send_ctx *sctx, u64 offset) |
5465 | { | |
5466 | struct btrfs_path *path; | |
5467 | struct btrfs_root *root = sctx->send_root; | |
5468 | struct btrfs_file_extent_item *fi; | |
5469 | struct btrfs_key key; | |
5470 | u64 extent_end; | |
5471 | u8 type; | |
5472 | int ret; | |
5473 | ||
5474 | path = alloc_path_for_send(); | |
5475 | if (!path) | |
5476 | return -ENOMEM; | |
5477 | ||
5478 | sctx->cur_inode_last_extent = 0; | |
5479 | ||
5480 | key.objectid = sctx->cur_ino; | |
5481 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5482 | key.offset = offset; | |
5483 | ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); | |
5484 | if (ret < 0) | |
5485 | goto out; | |
5486 | ret = 0; | |
5487 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
5488 | if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
5489 | goto out; | |
5490 | ||
5491 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5492 | struct btrfs_file_extent_item); | |
5493 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5494 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
5495 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
5496 | path->slots[0], fi); | |
16e7549f | 5497 | extent_end = ALIGN(key.offset + size, |
da17066c | 5498 | sctx->send_root->fs_info->sectorsize); |
16e7549f JB |
5499 | } else { |
5500 | extent_end = key.offset + | |
5501 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5502 | } | |
5503 | sctx->cur_inode_last_extent = extent_end; | |
5504 | out: | |
5505 | btrfs_free_path(path); | |
5506 | return ret; | |
5507 | } | |
5508 | ||
82bfb2e7 FM |
5509 | static int range_is_hole_in_parent(struct send_ctx *sctx, |
5510 | const u64 start, | |
5511 | const u64 end) | |
5512 | { | |
5513 | struct btrfs_path *path; | |
5514 | struct btrfs_key key; | |
5515 | struct btrfs_root *root = sctx->parent_root; | |
5516 | u64 search_start = start; | |
5517 | int ret; | |
5518 | ||
5519 | path = alloc_path_for_send(); | |
5520 | if (!path) | |
5521 | return -ENOMEM; | |
5522 | ||
5523 | key.objectid = sctx->cur_ino; | |
5524 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5525 | key.offset = search_start; | |
5526 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5527 | if (ret < 0) | |
5528 | goto out; | |
5529 | if (ret > 0 && path->slots[0] > 0) | |
5530 | path->slots[0]--; | |
5531 | ||
5532 | while (search_start < end) { | |
5533 | struct extent_buffer *leaf = path->nodes[0]; | |
5534 | int slot = path->slots[0]; | |
5535 | struct btrfs_file_extent_item *fi; | |
5536 | u64 extent_end; | |
5537 | ||
5538 | if (slot >= btrfs_header_nritems(leaf)) { | |
5539 | ret = btrfs_next_leaf(root, path); | |
5540 | if (ret < 0) | |
5541 | goto out; | |
5542 | else if (ret > 0) | |
5543 | break; | |
5544 | continue; | |
5545 | } | |
5546 | ||
5547 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5548 | if (key.objectid < sctx->cur_ino || | |
5549 | key.type < BTRFS_EXTENT_DATA_KEY) | |
5550 | goto next; | |
5551 | if (key.objectid > sctx->cur_ino || | |
5552 | key.type > BTRFS_EXTENT_DATA_KEY || | |
5553 | key.offset >= end) | |
5554 | break; | |
5555 | ||
5556 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5557 | if (btrfs_file_extent_type(leaf, fi) == | |
5558 | BTRFS_FILE_EXTENT_INLINE) { | |
5559 | u64 size = btrfs_file_extent_inline_len(leaf, slot, fi); | |
5560 | ||
5561 | extent_end = ALIGN(key.offset + size, | |
5562 | root->fs_info->sectorsize); | |
5563 | } else { | |
5564 | extent_end = key.offset + | |
5565 | btrfs_file_extent_num_bytes(leaf, fi); | |
5566 | } | |
5567 | if (extent_end <= start) | |
5568 | goto next; | |
5569 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) { | |
5570 | search_start = extent_end; | |
5571 | goto next; | |
5572 | } | |
5573 | ret = 0; | |
5574 | goto out; | |
5575 | next: | |
5576 | path->slots[0]++; | |
5577 | } | |
5578 | ret = 1; | |
5579 | out: | |
5580 | btrfs_free_path(path); | |
5581 | return ret; | |
5582 | } | |
5583 | ||
16e7549f JB |
5584 | static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, |
5585 | struct btrfs_key *key) | |
5586 | { | |
5587 | struct btrfs_file_extent_item *fi; | |
5588 | u64 extent_end; | |
5589 | u8 type; | |
5590 | int ret = 0; | |
5591 | ||
5592 | if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) | |
5593 | return 0; | |
5594 | ||
5595 | if (sctx->cur_inode_last_extent == (u64)-1) { | |
5596 | ret = get_last_extent(sctx, key->offset - 1); | |
5597 | if (ret) | |
5598 | return ret; | |
5599 | } | |
5600 | ||
5601 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5602 | struct btrfs_file_extent_item); | |
5603 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5604 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
5605 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
5606 | path->slots[0], fi); | |
16e7549f | 5607 | extent_end = ALIGN(key->offset + size, |
da17066c | 5608 | sctx->send_root->fs_info->sectorsize); |
16e7549f JB |
5609 | } else { |
5610 | extent_end = key->offset + | |
5611 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5612 | } | |
bf54f412 FDBM |
5613 | |
5614 | if (path->slots[0] == 0 && | |
5615 | sctx->cur_inode_last_extent < key->offset) { | |
5616 | /* | |
5617 | * We might have skipped entire leafs that contained only | |
5618 | * file extent items for our current inode. These leafs have | |
5619 | * a generation number smaller (older) than the one in the | |
5620 | * current leaf and the leaf our last extent came from, and | |
5621 | * are located between these 2 leafs. | |
5622 | */ | |
5623 | ret = get_last_extent(sctx, key->offset - 1); | |
5624 | if (ret) | |
5625 | return ret; | |
5626 | } | |
5627 | ||
82bfb2e7 FM |
5628 | if (sctx->cur_inode_last_extent < key->offset) { |
5629 | ret = range_is_hole_in_parent(sctx, | |
5630 | sctx->cur_inode_last_extent, | |
5631 | key->offset); | |
5632 | if (ret < 0) | |
5633 | return ret; | |
5634 | else if (ret == 0) | |
5635 | ret = send_hole(sctx, key->offset); | |
5636 | else | |
5637 | ret = 0; | |
5638 | } | |
16e7549f JB |
5639 | sctx->cur_inode_last_extent = extent_end; |
5640 | return ret; | |
5641 | } | |
5642 | ||
31db9f7c AB |
5643 | static int process_extent(struct send_ctx *sctx, |
5644 | struct btrfs_path *path, | |
5645 | struct btrfs_key *key) | |
5646 | { | |
31db9f7c | 5647 | struct clone_root *found_clone = NULL; |
57cfd462 | 5648 | int ret = 0; |
31db9f7c AB |
5649 | |
5650 | if (S_ISLNK(sctx->cur_inode_mode)) | |
5651 | return 0; | |
5652 | ||
5653 | if (sctx->parent_root && !sctx->cur_inode_new) { | |
5654 | ret = is_extent_unchanged(sctx, path, key); | |
5655 | if (ret < 0) | |
5656 | goto out; | |
5657 | if (ret) { | |
5658 | ret = 0; | |
16e7549f | 5659 | goto out_hole; |
31db9f7c | 5660 | } |
57cfd462 JB |
5661 | } else { |
5662 | struct btrfs_file_extent_item *ei; | |
5663 | u8 type; | |
5664 | ||
5665 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5666 | struct btrfs_file_extent_item); | |
5667 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
5668 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
5669 | type == BTRFS_FILE_EXTENT_REG) { | |
5670 | /* | |
5671 | * The send spec does not have a prealloc command yet, | |
5672 | * so just leave a hole for prealloc'ed extents until | |
5673 | * we have enough commands queued up to justify rev'ing | |
5674 | * the send spec. | |
5675 | */ | |
5676 | if (type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5677 | ret = 0; | |
5678 | goto out; | |
5679 | } | |
5680 | ||
5681 | /* Have a hole, just skip it. */ | |
5682 | if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { | |
5683 | ret = 0; | |
5684 | goto out; | |
5685 | } | |
5686 | } | |
31db9f7c AB |
5687 | } |
5688 | ||
5689 | ret = find_extent_clone(sctx, path, key->objectid, key->offset, | |
5690 | sctx->cur_inode_size, &found_clone); | |
5691 | if (ret != -ENOENT && ret < 0) | |
5692 | goto out; | |
5693 | ||
5694 | ret = send_write_or_clone(sctx, path, key, found_clone); | |
16e7549f JB |
5695 | if (ret) |
5696 | goto out; | |
5697 | out_hole: | |
5698 | ret = maybe_send_hole(sctx, path, key); | |
31db9f7c AB |
5699 | out: |
5700 | return ret; | |
5701 | } | |
5702 | ||
5703 | static int process_all_extents(struct send_ctx *sctx) | |
5704 | { | |
5705 | int ret; | |
5706 | struct btrfs_root *root; | |
5707 | struct btrfs_path *path; | |
5708 | struct btrfs_key key; | |
5709 | struct btrfs_key found_key; | |
5710 | struct extent_buffer *eb; | |
5711 | int slot; | |
5712 | ||
5713 | root = sctx->send_root; | |
5714 | path = alloc_path_for_send(); | |
5715 | if (!path) | |
5716 | return -ENOMEM; | |
5717 | ||
5718 | key.objectid = sctx->cmp_key->objectid; | |
5719 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5720 | key.offset = 0; | |
7fdd29d0 FDBM |
5721 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5722 | if (ret < 0) | |
5723 | goto out; | |
31db9f7c | 5724 | |
7fdd29d0 | 5725 | while (1) { |
31db9f7c AB |
5726 | eb = path->nodes[0]; |
5727 | slot = path->slots[0]; | |
7fdd29d0 FDBM |
5728 | |
5729 | if (slot >= btrfs_header_nritems(eb)) { | |
5730 | ret = btrfs_next_leaf(root, path); | |
5731 | if (ret < 0) { | |
5732 | goto out; | |
5733 | } else if (ret > 0) { | |
5734 | ret = 0; | |
5735 | break; | |
5736 | } | |
5737 | continue; | |
5738 | } | |
5739 | ||
31db9f7c AB |
5740 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
5741 | ||
5742 | if (found_key.objectid != key.objectid || | |
5743 | found_key.type != key.type) { | |
5744 | ret = 0; | |
5745 | goto out; | |
5746 | } | |
5747 | ||
5748 | ret = process_extent(sctx, path, &found_key); | |
5749 | if (ret < 0) | |
5750 | goto out; | |
5751 | ||
7fdd29d0 | 5752 | path->slots[0]++; |
31db9f7c AB |
5753 | } |
5754 | ||
5755 | out: | |
5756 | btrfs_free_path(path); | |
5757 | return ret; | |
5758 | } | |
5759 | ||
9f03740a FDBM |
5760 | static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, |
5761 | int *pending_move, | |
5762 | int *refs_processed) | |
31db9f7c AB |
5763 | { |
5764 | int ret = 0; | |
5765 | ||
5766 | if (sctx->cur_ino == 0) | |
5767 | goto out; | |
5768 | if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && | |
96b5bd77 | 5769 | sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) |
31db9f7c AB |
5770 | goto out; |
5771 | if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) | |
5772 | goto out; | |
5773 | ||
9f03740a | 5774 | ret = process_recorded_refs(sctx, pending_move); |
e479d9bb AB |
5775 | if (ret < 0) |
5776 | goto out; | |
5777 | ||
9f03740a | 5778 | *refs_processed = 1; |
31db9f7c AB |
5779 | out: |
5780 | return ret; | |
5781 | } | |
5782 | ||
5783 | static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) | |
5784 | { | |
5785 | int ret = 0; | |
5786 | u64 left_mode; | |
5787 | u64 left_uid; | |
5788 | u64 left_gid; | |
5789 | u64 right_mode; | |
5790 | u64 right_uid; | |
5791 | u64 right_gid; | |
5792 | int need_chmod = 0; | |
5793 | int need_chown = 0; | |
9f03740a FDBM |
5794 | int pending_move = 0; |
5795 | int refs_processed = 0; | |
31db9f7c | 5796 | |
9f03740a FDBM |
5797 | ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, |
5798 | &refs_processed); | |
31db9f7c AB |
5799 | if (ret < 0) |
5800 | goto out; | |
5801 | ||
9f03740a FDBM |
5802 | /* |
5803 | * We have processed the refs and thus need to advance send_progress. | |
5804 | * Now, calls to get_cur_xxx will take the updated refs of the current | |
5805 | * inode into account. | |
5806 | * | |
5807 | * On the other hand, if our current inode is a directory and couldn't | |
5808 | * be moved/renamed because its parent was renamed/moved too and it has | |
5809 | * a higher inode number, we can only move/rename our current inode | |
5810 | * after we moved/renamed its parent. Therefore in this case operate on | |
5811 | * the old path (pre move/rename) of our current inode, and the | |
5812 | * move/rename will be performed later. | |
5813 | */ | |
5814 | if (refs_processed && !pending_move) | |
5815 | sctx->send_progress = sctx->cur_ino + 1; | |
5816 | ||
31db9f7c AB |
5817 | if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) |
5818 | goto out; | |
5819 | if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) | |
5820 | goto out; | |
5821 | ||
5822 | ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, | |
85a7b33b | 5823 | &left_mode, &left_uid, &left_gid, NULL); |
31db9f7c AB |
5824 | if (ret < 0) |
5825 | goto out; | |
5826 | ||
e2d044fe AL |
5827 | if (!sctx->parent_root || sctx->cur_inode_new) { |
5828 | need_chown = 1; | |
5829 | if (!S_ISLNK(sctx->cur_inode_mode)) | |
31db9f7c | 5830 | need_chmod = 1; |
e2d044fe AL |
5831 | } else { |
5832 | ret = get_inode_info(sctx->parent_root, sctx->cur_ino, | |
5833 | NULL, NULL, &right_mode, &right_uid, | |
5834 | &right_gid, NULL); | |
5835 | if (ret < 0) | |
5836 | goto out; | |
31db9f7c | 5837 | |
e2d044fe AL |
5838 | if (left_uid != right_uid || left_gid != right_gid) |
5839 | need_chown = 1; | |
5840 | if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) | |
5841 | need_chmod = 1; | |
31db9f7c AB |
5842 | } |
5843 | ||
5844 | if (S_ISREG(sctx->cur_inode_mode)) { | |
16e7549f | 5845 | if (need_send_hole(sctx)) { |
766b5e5a FM |
5846 | if (sctx->cur_inode_last_extent == (u64)-1 || |
5847 | sctx->cur_inode_last_extent < | |
5848 | sctx->cur_inode_size) { | |
16e7549f JB |
5849 | ret = get_last_extent(sctx, (u64)-1); |
5850 | if (ret) | |
5851 | goto out; | |
5852 | } | |
5853 | if (sctx->cur_inode_last_extent < | |
5854 | sctx->cur_inode_size) { | |
5855 | ret = send_hole(sctx, sctx->cur_inode_size); | |
5856 | if (ret) | |
5857 | goto out; | |
5858 | } | |
5859 | } | |
31db9f7c AB |
5860 | ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
5861 | sctx->cur_inode_size); | |
5862 | if (ret < 0) | |
5863 | goto out; | |
5864 | } | |
5865 | ||
5866 | if (need_chown) { | |
5867 | ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5868 | left_uid, left_gid); | |
5869 | if (ret < 0) | |
5870 | goto out; | |
5871 | } | |
5872 | if (need_chmod) { | |
5873 | ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5874 | left_mode); | |
5875 | if (ret < 0) | |
5876 | goto out; | |
5877 | } | |
5878 | ||
5879 | /* | |
9f03740a FDBM |
5880 | * If other directory inodes depended on our current directory |
5881 | * inode's move/rename, now do their move/rename operations. | |
31db9f7c | 5882 | */ |
9f03740a FDBM |
5883 | if (!is_waiting_for_move(sctx, sctx->cur_ino)) { |
5884 | ret = apply_children_dir_moves(sctx); | |
5885 | if (ret) | |
5886 | goto out; | |
fcbd2154 FM |
5887 | /* |
5888 | * Need to send that every time, no matter if it actually | |
5889 | * changed between the two trees as we have done changes to | |
5890 | * the inode before. If our inode is a directory and it's | |
5891 | * waiting to be moved/renamed, we will send its utimes when | |
5892 | * it's moved/renamed, therefore we don't need to do it here. | |
5893 | */ | |
5894 | sctx->send_progress = sctx->cur_ino + 1; | |
5895 | ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
5896 | if (ret < 0) | |
5897 | goto out; | |
9f03740a FDBM |
5898 | } |
5899 | ||
31db9f7c AB |
5900 | out: |
5901 | return ret; | |
5902 | } | |
5903 | ||
5904 | static int changed_inode(struct send_ctx *sctx, | |
5905 | enum btrfs_compare_tree_result result) | |
5906 | { | |
5907 | int ret = 0; | |
5908 | struct btrfs_key *key = sctx->cmp_key; | |
5909 | struct btrfs_inode_item *left_ii = NULL; | |
5910 | struct btrfs_inode_item *right_ii = NULL; | |
5911 | u64 left_gen = 0; | |
5912 | u64 right_gen = 0; | |
5913 | ||
31db9f7c AB |
5914 | sctx->cur_ino = key->objectid; |
5915 | sctx->cur_inode_new_gen = 0; | |
16e7549f | 5916 | sctx->cur_inode_last_extent = (u64)-1; |
e479d9bb AB |
5917 | |
5918 | /* | |
5919 | * Set send_progress to current inode. This will tell all get_cur_xxx | |
5920 | * functions that the current inode's refs are not updated yet. Later, | |
5921 | * when process_recorded_refs is finished, it is set to cur_ino + 1. | |
5922 | */ | |
31db9f7c AB |
5923 | sctx->send_progress = sctx->cur_ino; |
5924 | ||
5925 | if (result == BTRFS_COMPARE_TREE_NEW || | |
5926 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
5927 | left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], | |
5928 | sctx->left_path->slots[0], | |
5929 | struct btrfs_inode_item); | |
5930 | left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], | |
5931 | left_ii); | |
5932 | } else { | |
5933 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
5934 | sctx->right_path->slots[0], | |
5935 | struct btrfs_inode_item); | |
5936 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
5937 | right_ii); | |
5938 | } | |
5939 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
5940 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
5941 | sctx->right_path->slots[0], | |
5942 | struct btrfs_inode_item); | |
5943 | ||
5944 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
5945 | right_ii); | |
6d85ed05 AB |
5946 | |
5947 | /* | |
5948 | * The cur_ino = root dir case is special here. We can't treat | |
5949 | * the inode as deleted+reused because it would generate a | |
5950 | * stream that tries to delete/mkdir the root dir. | |
5951 | */ | |
5952 | if (left_gen != right_gen && | |
5953 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) | |
31db9f7c AB |
5954 | sctx->cur_inode_new_gen = 1; |
5955 | } | |
5956 | ||
5957 | if (result == BTRFS_COMPARE_TREE_NEW) { | |
5958 | sctx->cur_inode_gen = left_gen; | |
5959 | sctx->cur_inode_new = 1; | |
5960 | sctx->cur_inode_deleted = 0; | |
5961 | sctx->cur_inode_size = btrfs_inode_size( | |
5962 | sctx->left_path->nodes[0], left_ii); | |
5963 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5964 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
5965 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
5966 | sctx->left_path->nodes[0], left_ii); | |
31db9f7c | 5967 | if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) |
1f4692da | 5968 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
5969 | } else if (result == BTRFS_COMPARE_TREE_DELETED) { |
5970 | sctx->cur_inode_gen = right_gen; | |
5971 | sctx->cur_inode_new = 0; | |
5972 | sctx->cur_inode_deleted = 1; | |
5973 | sctx->cur_inode_size = btrfs_inode_size( | |
5974 | sctx->right_path->nodes[0], right_ii); | |
5975 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5976 | sctx->right_path->nodes[0], right_ii); | |
5977 | } else if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
766702ef AB |
5978 | /* |
5979 | * We need to do some special handling in case the inode was | |
5980 | * reported as changed with a changed generation number. This | |
5981 | * means that the original inode was deleted and new inode | |
5982 | * reused the same inum. So we have to treat the old inode as | |
5983 | * deleted and the new one as new. | |
5984 | */ | |
31db9f7c | 5985 | if (sctx->cur_inode_new_gen) { |
766702ef AB |
5986 | /* |
5987 | * First, process the inode as if it was deleted. | |
5988 | */ | |
31db9f7c AB |
5989 | sctx->cur_inode_gen = right_gen; |
5990 | sctx->cur_inode_new = 0; | |
5991 | sctx->cur_inode_deleted = 1; | |
5992 | sctx->cur_inode_size = btrfs_inode_size( | |
5993 | sctx->right_path->nodes[0], right_ii); | |
5994 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5995 | sctx->right_path->nodes[0], right_ii); | |
5996 | ret = process_all_refs(sctx, | |
5997 | BTRFS_COMPARE_TREE_DELETED); | |
5998 | if (ret < 0) | |
5999 | goto out; | |
6000 | ||
766702ef AB |
6001 | /* |
6002 | * Now process the inode as if it was new. | |
6003 | */ | |
31db9f7c AB |
6004 | sctx->cur_inode_gen = left_gen; |
6005 | sctx->cur_inode_new = 1; | |
6006 | sctx->cur_inode_deleted = 0; | |
6007 | sctx->cur_inode_size = btrfs_inode_size( | |
6008 | sctx->left_path->nodes[0], left_ii); | |
6009 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6010 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6011 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6012 | sctx->left_path->nodes[0], left_ii); | |
1f4692da | 6013 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6014 | if (ret < 0) |
6015 | goto out; | |
6016 | ||
6017 | ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); | |
6018 | if (ret < 0) | |
6019 | goto out; | |
e479d9bb AB |
6020 | /* |
6021 | * Advance send_progress now as we did not get into | |
6022 | * process_recorded_refs_if_needed in the new_gen case. | |
6023 | */ | |
6024 | sctx->send_progress = sctx->cur_ino + 1; | |
766702ef AB |
6025 | |
6026 | /* | |
6027 | * Now process all extents and xattrs of the inode as if | |
6028 | * they were all new. | |
6029 | */ | |
31db9f7c AB |
6030 | ret = process_all_extents(sctx); |
6031 | if (ret < 0) | |
6032 | goto out; | |
6033 | ret = process_all_new_xattrs(sctx); | |
6034 | if (ret < 0) | |
6035 | goto out; | |
6036 | } else { | |
6037 | sctx->cur_inode_gen = left_gen; | |
6038 | sctx->cur_inode_new = 0; | |
6039 | sctx->cur_inode_new_gen = 0; | |
6040 | sctx->cur_inode_deleted = 0; | |
6041 | sctx->cur_inode_size = btrfs_inode_size( | |
6042 | sctx->left_path->nodes[0], left_ii); | |
6043 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6044 | sctx->left_path->nodes[0], left_ii); | |
6045 | } | |
6046 | } | |
6047 | ||
6048 | out: | |
6049 | return ret; | |
6050 | } | |
6051 | ||
766702ef AB |
6052 | /* |
6053 | * We have to process new refs before deleted refs, but compare_trees gives us | |
6054 | * the new and deleted refs mixed. To fix this, we record the new/deleted refs | |
6055 | * first and later process them in process_recorded_refs. | |
6056 | * For the cur_inode_new_gen case, we skip recording completely because | |
6057 | * changed_inode did already initiate processing of refs. The reason for this is | |
6058 | * that in this case, compare_tree actually compares the refs of 2 different | |
6059 | * inodes. To fix this, process_all_refs is used in changed_inode to handle all | |
6060 | * refs of the right tree as deleted and all refs of the left tree as new. | |
6061 | */ | |
31db9f7c AB |
6062 | static int changed_ref(struct send_ctx *sctx, |
6063 | enum btrfs_compare_tree_result result) | |
6064 | { | |
6065 | int ret = 0; | |
6066 | ||
95155585 FM |
6067 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6068 | inconsistent_snapshot_error(sctx, result, "reference"); | |
6069 | return -EIO; | |
6070 | } | |
31db9f7c AB |
6071 | |
6072 | if (!sctx->cur_inode_new_gen && | |
6073 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { | |
6074 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6075 | ret = record_new_ref(sctx); | |
6076 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6077 | ret = record_deleted_ref(sctx); | |
6078 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6079 | ret = record_changed_ref(sctx); | |
6080 | } | |
6081 | ||
6082 | return ret; | |
6083 | } | |
6084 | ||
766702ef AB |
6085 | /* |
6086 | * Process new/deleted/changed xattrs. We skip processing in the | |
6087 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6088 | * of xattrs. The reason is the same as in changed_ref | |
6089 | */ | |
31db9f7c AB |
6090 | static int changed_xattr(struct send_ctx *sctx, |
6091 | enum btrfs_compare_tree_result result) | |
6092 | { | |
6093 | int ret = 0; | |
6094 | ||
95155585 FM |
6095 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6096 | inconsistent_snapshot_error(sctx, result, "xattr"); | |
6097 | return -EIO; | |
6098 | } | |
31db9f7c AB |
6099 | |
6100 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6101 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6102 | ret = process_new_xattr(sctx); | |
6103 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6104 | ret = process_deleted_xattr(sctx); | |
6105 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6106 | ret = process_changed_xattr(sctx); | |
6107 | } | |
6108 | ||
6109 | return ret; | |
6110 | } | |
6111 | ||
766702ef AB |
6112 | /* |
6113 | * Process new/deleted/changed extents. We skip processing in the | |
6114 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6115 | * of extents. The reason is the same as in changed_ref | |
6116 | */ | |
31db9f7c AB |
6117 | static int changed_extent(struct send_ctx *sctx, |
6118 | enum btrfs_compare_tree_result result) | |
6119 | { | |
6120 | int ret = 0; | |
6121 | ||
95155585 | 6122 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
d5e84fd8 FM |
6123 | |
6124 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
6125 | struct extent_buffer *leaf_l; | |
6126 | struct extent_buffer *leaf_r; | |
6127 | struct btrfs_file_extent_item *ei_l; | |
6128 | struct btrfs_file_extent_item *ei_r; | |
6129 | ||
6130 | leaf_l = sctx->left_path->nodes[0]; | |
6131 | leaf_r = sctx->right_path->nodes[0]; | |
6132 | ei_l = btrfs_item_ptr(leaf_l, | |
6133 | sctx->left_path->slots[0], | |
6134 | struct btrfs_file_extent_item); | |
6135 | ei_r = btrfs_item_ptr(leaf_r, | |
6136 | sctx->right_path->slots[0], | |
6137 | struct btrfs_file_extent_item); | |
6138 | ||
6139 | /* | |
6140 | * We may have found an extent item that has changed | |
6141 | * only its disk_bytenr field and the corresponding | |
6142 | * inode item was not updated. This case happens due to | |
6143 | * very specific timings during relocation when a leaf | |
6144 | * that contains file extent items is COWed while | |
6145 | * relocation is ongoing and its in the stage where it | |
6146 | * updates data pointers. So when this happens we can | |
6147 | * safely ignore it since we know it's the same extent, | |
6148 | * but just at different logical and physical locations | |
6149 | * (when an extent is fully replaced with a new one, we | |
6150 | * know the generation number must have changed too, | |
6151 | * since snapshot creation implies committing the current | |
6152 | * transaction, and the inode item must have been updated | |
6153 | * as well). | |
6154 | * This replacement of the disk_bytenr happens at | |
6155 | * relocation.c:replace_file_extents() through | |
6156 | * relocation.c:btrfs_reloc_cow_block(). | |
6157 | */ | |
6158 | if (btrfs_file_extent_generation(leaf_l, ei_l) == | |
6159 | btrfs_file_extent_generation(leaf_r, ei_r) && | |
6160 | btrfs_file_extent_ram_bytes(leaf_l, ei_l) == | |
6161 | btrfs_file_extent_ram_bytes(leaf_r, ei_r) && | |
6162 | btrfs_file_extent_compression(leaf_l, ei_l) == | |
6163 | btrfs_file_extent_compression(leaf_r, ei_r) && | |
6164 | btrfs_file_extent_encryption(leaf_l, ei_l) == | |
6165 | btrfs_file_extent_encryption(leaf_r, ei_r) && | |
6166 | btrfs_file_extent_other_encoding(leaf_l, ei_l) == | |
6167 | btrfs_file_extent_other_encoding(leaf_r, ei_r) && | |
6168 | btrfs_file_extent_type(leaf_l, ei_l) == | |
6169 | btrfs_file_extent_type(leaf_r, ei_r) && | |
6170 | btrfs_file_extent_disk_bytenr(leaf_l, ei_l) != | |
6171 | btrfs_file_extent_disk_bytenr(leaf_r, ei_r) && | |
6172 | btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) == | |
6173 | btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) && | |
6174 | btrfs_file_extent_offset(leaf_l, ei_l) == | |
6175 | btrfs_file_extent_offset(leaf_r, ei_r) && | |
6176 | btrfs_file_extent_num_bytes(leaf_l, ei_l) == | |
6177 | btrfs_file_extent_num_bytes(leaf_r, ei_r)) | |
6178 | return 0; | |
6179 | } | |
6180 | ||
95155585 FM |
6181 | inconsistent_snapshot_error(sctx, result, "extent"); |
6182 | return -EIO; | |
6183 | } | |
31db9f7c AB |
6184 | |
6185 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6186 | if (result != BTRFS_COMPARE_TREE_DELETED) | |
6187 | ret = process_extent(sctx, sctx->left_path, | |
6188 | sctx->cmp_key); | |
6189 | } | |
6190 | ||
6191 | return ret; | |
6192 | } | |
6193 | ||
ba5e8f2e JB |
6194 | static int dir_changed(struct send_ctx *sctx, u64 dir) |
6195 | { | |
6196 | u64 orig_gen, new_gen; | |
6197 | int ret; | |
6198 | ||
6199 | ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, | |
6200 | NULL, NULL); | |
6201 | if (ret) | |
6202 | return ret; | |
6203 | ||
6204 | ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, | |
6205 | NULL, NULL, NULL); | |
6206 | if (ret) | |
6207 | return ret; | |
6208 | ||
6209 | return (orig_gen != new_gen) ? 1 : 0; | |
6210 | } | |
6211 | ||
6212 | static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, | |
6213 | struct btrfs_key *key) | |
6214 | { | |
6215 | struct btrfs_inode_extref *extref; | |
6216 | struct extent_buffer *leaf; | |
6217 | u64 dirid = 0, last_dirid = 0; | |
6218 | unsigned long ptr; | |
6219 | u32 item_size; | |
6220 | u32 cur_offset = 0; | |
6221 | int ref_name_len; | |
6222 | int ret = 0; | |
6223 | ||
6224 | /* Easy case, just check this one dirid */ | |
6225 | if (key->type == BTRFS_INODE_REF_KEY) { | |
6226 | dirid = key->offset; | |
6227 | ||
6228 | ret = dir_changed(sctx, dirid); | |
6229 | goto out; | |
6230 | } | |
6231 | ||
6232 | leaf = path->nodes[0]; | |
6233 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6234 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
6235 | while (cur_offset < item_size) { | |
6236 | extref = (struct btrfs_inode_extref *)(ptr + | |
6237 | cur_offset); | |
6238 | dirid = btrfs_inode_extref_parent(leaf, extref); | |
6239 | ref_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
6240 | cur_offset += ref_name_len + sizeof(*extref); | |
6241 | if (dirid == last_dirid) | |
6242 | continue; | |
6243 | ret = dir_changed(sctx, dirid); | |
6244 | if (ret) | |
6245 | break; | |
6246 | last_dirid = dirid; | |
6247 | } | |
6248 | out: | |
6249 | return ret; | |
6250 | } | |
6251 | ||
766702ef AB |
6252 | /* |
6253 | * Updates compare related fields in sctx and simply forwards to the actual | |
6254 | * changed_xxx functions. | |
6255 | */ | |
ee8c494f | 6256 | static int changed_cb(struct btrfs_path *left_path, |
31db9f7c AB |
6257 | struct btrfs_path *right_path, |
6258 | struct btrfs_key *key, | |
6259 | enum btrfs_compare_tree_result result, | |
6260 | void *ctx) | |
6261 | { | |
6262 | int ret = 0; | |
6263 | struct send_ctx *sctx = ctx; | |
6264 | ||
ba5e8f2e | 6265 | if (result == BTRFS_COMPARE_TREE_SAME) { |
16e7549f JB |
6266 | if (key->type == BTRFS_INODE_REF_KEY || |
6267 | key->type == BTRFS_INODE_EXTREF_KEY) { | |
6268 | ret = compare_refs(sctx, left_path, key); | |
6269 | if (!ret) | |
6270 | return 0; | |
6271 | if (ret < 0) | |
6272 | return ret; | |
6273 | } else if (key->type == BTRFS_EXTENT_DATA_KEY) { | |
6274 | return maybe_send_hole(sctx, left_path, key); | |
6275 | } else { | |
ba5e8f2e | 6276 | return 0; |
16e7549f | 6277 | } |
ba5e8f2e JB |
6278 | result = BTRFS_COMPARE_TREE_CHANGED; |
6279 | ret = 0; | |
6280 | } | |
6281 | ||
31db9f7c AB |
6282 | sctx->left_path = left_path; |
6283 | sctx->right_path = right_path; | |
6284 | sctx->cmp_key = key; | |
6285 | ||
6286 | ret = finish_inode_if_needed(sctx, 0); | |
6287 | if (ret < 0) | |
6288 | goto out; | |
6289 | ||
2981e225 AB |
6290 | /* Ignore non-FS objects */ |
6291 | if (key->objectid == BTRFS_FREE_INO_OBJECTID || | |
6292 | key->objectid == BTRFS_FREE_SPACE_OBJECTID) | |
6293 | goto out; | |
6294 | ||
31db9f7c AB |
6295 | if (key->type == BTRFS_INODE_ITEM_KEY) |
6296 | ret = changed_inode(sctx, result); | |
96b5bd77 JS |
6297 | else if (key->type == BTRFS_INODE_REF_KEY || |
6298 | key->type == BTRFS_INODE_EXTREF_KEY) | |
31db9f7c AB |
6299 | ret = changed_ref(sctx, result); |
6300 | else if (key->type == BTRFS_XATTR_ITEM_KEY) | |
6301 | ret = changed_xattr(sctx, result); | |
6302 | else if (key->type == BTRFS_EXTENT_DATA_KEY) | |
6303 | ret = changed_extent(sctx, result); | |
6304 | ||
6305 | out: | |
6306 | return ret; | |
6307 | } | |
6308 | ||
6309 | static int full_send_tree(struct send_ctx *sctx) | |
6310 | { | |
6311 | int ret; | |
31db9f7c AB |
6312 | struct btrfs_root *send_root = sctx->send_root; |
6313 | struct btrfs_key key; | |
6314 | struct btrfs_key found_key; | |
6315 | struct btrfs_path *path; | |
6316 | struct extent_buffer *eb; | |
6317 | int slot; | |
31db9f7c AB |
6318 | |
6319 | path = alloc_path_for_send(); | |
6320 | if (!path) | |
6321 | return -ENOMEM; | |
6322 | ||
31db9f7c AB |
6323 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; |
6324 | key.type = BTRFS_INODE_ITEM_KEY; | |
6325 | key.offset = 0; | |
6326 | ||
31db9f7c AB |
6327 | ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); |
6328 | if (ret < 0) | |
6329 | goto out; | |
6330 | if (ret) | |
6331 | goto out_finish; | |
6332 | ||
6333 | while (1) { | |
31db9f7c AB |
6334 | eb = path->nodes[0]; |
6335 | slot = path->slots[0]; | |
6336 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6337 | ||
ee8c494f NB |
6338 | ret = changed_cb(path, NULL, &found_key, |
6339 | BTRFS_COMPARE_TREE_NEW, sctx); | |
31db9f7c AB |
6340 | if (ret < 0) |
6341 | goto out; | |
6342 | ||
6343 | key.objectid = found_key.objectid; | |
6344 | key.type = found_key.type; | |
6345 | key.offset = found_key.offset + 1; | |
6346 | ||
6347 | ret = btrfs_next_item(send_root, path); | |
6348 | if (ret < 0) | |
6349 | goto out; | |
6350 | if (ret) { | |
6351 | ret = 0; | |
6352 | break; | |
6353 | } | |
6354 | } | |
6355 | ||
6356 | out_finish: | |
6357 | ret = finish_inode_if_needed(sctx, 1); | |
6358 | ||
6359 | out: | |
6360 | btrfs_free_path(path); | |
31db9f7c AB |
6361 | return ret; |
6362 | } | |
6363 | ||
6364 | static int send_subvol(struct send_ctx *sctx) | |
6365 | { | |
6366 | int ret; | |
6367 | ||
c2c71324 SB |
6368 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { |
6369 | ret = send_header(sctx); | |
6370 | if (ret < 0) | |
6371 | goto out; | |
6372 | } | |
31db9f7c AB |
6373 | |
6374 | ret = send_subvol_begin(sctx); | |
6375 | if (ret < 0) | |
6376 | goto out; | |
6377 | ||
6378 | if (sctx->parent_root) { | |
6379 | ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, | |
6380 | changed_cb, sctx); | |
6381 | if (ret < 0) | |
6382 | goto out; | |
6383 | ret = finish_inode_if_needed(sctx, 1); | |
6384 | if (ret < 0) | |
6385 | goto out; | |
6386 | } else { | |
6387 | ret = full_send_tree(sctx); | |
6388 | if (ret < 0) | |
6389 | goto out; | |
6390 | } | |
6391 | ||
6392 | out: | |
31db9f7c AB |
6393 | free_recorded_refs(sctx); |
6394 | return ret; | |
6395 | } | |
6396 | ||
e5fa8f86 FM |
6397 | /* |
6398 | * If orphan cleanup did remove any orphans from a root, it means the tree | |
6399 | * was modified and therefore the commit root is not the same as the current | |
6400 | * root anymore. This is a problem, because send uses the commit root and | |
6401 | * therefore can see inode items that don't exist in the current root anymore, | |
6402 | * and for example make calls to btrfs_iget, which will do tree lookups based | |
6403 | * on the current root and not on the commit root. Those lookups will fail, | |
6404 | * returning a -ESTALE error, and making send fail with that error. So make | |
6405 | * sure a send does not see any orphans we have just removed, and that it will | |
6406 | * see the same inodes regardless of whether a transaction commit happened | |
6407 | * before it started (meaning that the commit root will be the same as the | |
6408 | * current root) or not. | |
6409 | */ | |
6410 | static int ensure_commit_roots_uptodate(struct send_ctx *sctx) | |
6411 | { | |
6412 | int i; | |
6413 | struct btrfs_trans_handle *trans = NULL; | |
6414 | ||
6415 | again: | |
6416 | if (sctx->parent_root && | |
6417 | sctx->parent_root->node != sctx->parent_root->commit_root) | |
6418 | goto commit_trans; | |
6419 | ||
6420 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6421 | if (sctx->clone_roots[i].root->node != | |
6422 | sctx->clone_roots[i].root->commit_root) | |
6423 | goto commit_trans; | |
6424 | ||
6425 | if (trans) | |
3a45bb20 | 6426 | return btrfs_end_transaction(trans); |
e5fa8f86 FM |
6427 | |
6428 | return 0; | |
6429 | ||
6430 | commit_trans: | |
6431 | /* Use any root, all fs roots will get their commit roots updated. */ | |
6432 | if (!trans) { | |
6433 | trans = btrfs_join_transaction(sctx->send_root); | |
6434 | if (IS_ERR(trans)) | |
6435 | return PTR_ERR(trans); | |
6436 | goto again; | |
6437 | } | |
6438 | ||
3a45bb20 | 6439 | return btrfs_commit_transaction(trans); |
e5fa8f86 FM |
6440 | } |
6441 | ||
66ef7d65 DS |
6442 | static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) |
6443 | { | |
6444 | spin_lock(&root->root_item_lock); | |
6445 | root->send_in_progress--; | |
6446 | /* | |
6447 | * Not much left to do, we don't know why it's unbalanced and | |
6448 | * can't blindly reset it to 0. | |
6449 | */ | |
6450 | if (root->send_in_progress < 0) | |
6451 | btrfs_err(root->fs_info, | |
0b246afa JM |
6452 | "send_in_progres unbalanced %d root %llu", |
6453 | root->send_in_progress, root->root_key.objectid); | |
66ef7d65 DS |
6454 | spin_unlock(&root->root_item_lock); |
6455 | } | |
6456 | ||
2351f431 | 6457 | long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) |
31db9f7c AB |
6458 | { |
6459 | int ret = 0; | |
0b246afa JM |
6460 | struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root; |
6461 | struct btrfs_fs_info *fs_info = send_root->fs_info; | |
31db9f7c | 6462 | struct btrfs_root *clone_root; |
31db9f7c | 6463 | struct btrfs_key key; |
31db9f7c AB |
6464 | struct send_ctx *sctx = NULL; |
6465 | u32 i; | |
6466 | u64 *clone_sources_tmp = NULL; | |
2c686537 | 6467 | int clone_sources_to_rollback = 0; |
e55d1153 | 6468 | unsigned alloc_size; |
896c14f9 | 6469 | int sort_clone_roots = 0; |
18f687d5 | 6470 | int index; |
31db9f7c AB |
6471 | |
6472 | if (!capable(CAP_SYS_ADMIN)) | |
6473 | return -EPERM; | |
6474 | ||
2c686537 DS |
6475 | /* |
6476 | * The subvolume must remain read-only during send, protect against | |
521e0546 | 6477 | * making it RW. This also protects against deletion. |
2c686537 DS |
6478 | */ |
6479 | spin_lock(&send_root->root_item_lock); | |
6480 | send_root->send_in_progress++; | |
6481 | spin_unlock(&send_root->root_item_lock); | |
6482 | ||
139f807a JB |
6483 | /* |
6484 | * This is done when we lookup the root, it should already be complete | |
6485 | * by the time we get here. | |
6486 | */ | |
6487 | WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE); | |
6488 | ||
2c686537 DS |
6489 | /* |
6490 | * Userspace tools do the checks and warn the user if it's | |
6491 | * not RO. | |
6492 | */ | |
6493 | if (!btrfs_root_readonly(send_root)) { | |
6494 | ret = -EPERM; | |
6495 | goto out; | |
6496 | } | |
6497 | ||
457ae726 DC |
6498 | /* |
6499 | * Check that we don't overflow at later allocations, we request | |
6500 | * clone_sources_count + 1 items, and compare to unsigned long inside | |
6501 | * access_ok. | |
6502 | */ | |
f5ecec3c | 6503 | if (arg->clone_sources_count > |
457ae726 | 6504 | ULONG_MAX / sizeof(struct clone_root) - 1) { |
f5ecec3c DC |
6505 | ret = -EINVAL; |
6506 | goto out; | |
6507 | } | |
6508 | ||
31db9f7c | 6509 | if (!access_ok(VERIFY_READ, arg->clone_sources, |
700ff4f0 DC |
6510 | sizeof(*arg->clone_sources) * |
6511 | arg->clone_sources_count)) { | |
31db9f7c AB |
6512 | ret = -EFAULT; |
6513 | goto out; | |
6514 | } | |
6515 | ||
c2c71324 | 6516 | if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { |
cb95e7bf MF |
6517 | ret = -EINVAL; |
6518 | goto out; | |
6519 | } | |
6520 | ||
e780b0d1 | 6521 | sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL); |
31db9f7c AB |
6522 | if (!sctx) { |
6523 | ret = -ENOMEM; | |
6524 | goto out; | |
6525 | } | |
6526 | ||
6527 | INIT_LIST_HEAD(&sctx->new_refs); | |
6528 | INIT_LIST_HEAD(&sctx->deleted_refs); | |
e780b0d1 | 6529 | INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL); |
31db9f7c AB |
6530 | INIT_LIST_HEAD(&sctx->name_cache_list); |
6531 | ||
cb95e7bf MF |
6532 | sctx->flags = arg->flags; |
6533 | ||
31db9f7c | 6534 | sctx->send_filp = fget(arg->send_fd); |
ecc7ada7 TI |
6535 | if (!sctx->send_filp) { |
6536 | ret = -EBADF; | |
31db9f7c AB |
6537 | goto out; |
6538 | } | |
6539 | ||
31db9f7c | 6540 | sctx->send_root = send_root; |
521e0546 DS |
6541 | /* |
6542 | * Unlikely but possible, if the subvolume is marked for deletion but | |
6543 | * is slow to remove the directory entry, send can still be started | |
6544 | */ | |
6545 | if (btrfs_root_dead(sctx->send_root)) { | |
6546 | ret = -EPERM; | |
6547 | goto out; | |
6548 | } | |
6549 | ||
31db9f7c AB |
6550 | sctx->clone_roots_cnt = arg->clone_sources_count; |
6551 | ||
6552 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE; | |
752ade68 | 6553 | sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); |
31db9f7c | 6554 | if (!sctx->send_buf) { |
752ade68 MH |
6555 | ret = -ENOMEM; |
6556 | goto out; | |
31db9f7c AB |
6557 | } |
6558 | ||
752ade68 | 6559 | sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL); |
31db9f7c | 6560 | if (!sctx->read_buf) { |
752ade68 MH |
6561 | ret = -ENOMEM; |
6562 | goto out; | |
31db9f7c AB |
6563 | } |
6564 | ||
9f03740a FDBM |
6565 | sctx->pending_dir_moves = RB_ROOT; |
6566 | sctx->waiting_dir_moves = RB_ROOT; | |
9dc44214 | 6567 | sctx->orphan_dirs = RB_ROOT; |
9f03740a | 6568 | |
e55d1153 DS |
6569 | alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1); |
6570 | ||
818e010b | 6571 | sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 6572 | if (!sctx->clone_roots) { |
818e010b DS |
6573 | ret = -ENOMEM; |
6574 | goto out; | |
31db9f7c AB |
6575 | } |
6576 | ||
e55d1153 DS |
6577 | alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources); |
6578 | ||
31db9f7c | 6579 | if (arg->clone_sources_count) { |
752ade68 | 6580 | clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 6581 | if (!clone_sources_tmp) { |
752ade68 MH |
6582 | ret = -ENOMEM; |
6583 | goto out; | |
31db9f7c AB |
6584 | } |
6585 | ||
6586 | ret = copy_from_user(clone_sources_tmp, arg->clone_sources, | |
e55d1153 | 6587 | alloc_size); |
31db9f7c AB |
6588 | if (ret) { |
6589 | ret = -EFAULT; | |
6590 | goto out; | |
6591 | } | |
6592 | ||
6593 | for (i = 0; i < arg->clone_sources_count; i++) { | |
6594 | key.objectid = clone_sources_tmp[i]; | |
6595 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6596 | key.offset = (u64)-1; | |
18f687d5 WS |
6597 | |
6598 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6599 | ||
31db9f7c | 6600 | clone_root = btrfs_read_fs_root_no_name(fs_info, &key); |
31db9f7c | 6601 | if (IS_ERR(clone_root)) { |
18f687d5 | 6602 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
31db9f7c AB |
6603 | ret = PTR_ERR(clone_root); |
6604 | goto out; | |
6605 | } | |
2c686537 | 6606 | spin_lock(&clone_root->root_item_lock); |
5cc2b17e FM |
6607 | if (!btrfs_root_readonly(clone_root) || |
6608 | btrfs_root_dead(clone_root)) { | |
2c686537 | 6609 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 | 6610 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6611 | ret = -EPERM; |
6612 | goto out; | |
6613 | } | |
2f1f465a | 6614 | clone_root->send_in_progress++; |
2c686537 | 6615 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 WS |
6616 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
6617 | ||
31db9f7c | 6618 | sctx->clone_roots[i].root = clone_root; |
2f1f465a | 6619 | clone_sources_to_rollback = i + 1; |
31db9f7c | 6620 | } |
2f91306a | 6621 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6622 | clone_sources_tmp = NULL; |
6623 | } | |
6624 | ||
6625 | if (arg->parent_root) { | |
6626 | key.objectid = arg->parent_root; | |
6627 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6628 | key.offset = (u64)-1; | |
18f687d5 WS |
6629 | |
6630 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6631 | ||
31db9f7c | 6632 | sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key); |
b1b19596 | 6633 | if (IS_ERR(sctx->parent_root)) { |
18f687d5 | 6634 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
b1b19596 | 6635 | ret = PTR_ERR(sctx->parent_root); |
31db9f7c AB |
6636 | goto out; |
6637 | } | |
18f687d5 | 6638 | |
2c686537 DS |
6639 | spin_lock(&sctx->parent_root->root_item_lock); |
6640 | sctx->parent_root->send_in_progress++; | |
521e0546 DS |
6641 | if (!btrfs_root_readonly(sctx->parent_root) || |
6642 | btrfs_root_dead(sctx->parent_root)) { | |
2c686537 | 6643 | spin_unlock(&sctx->parent_root->root_item_lock); |
18f687d5 | 6644 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6645 | ret = -EPERM; |
6646 | goto out; | |
6647 | } | |
6648 | spin_unlock(&sctx->parent_root->root_item_lock); | |
18f687d5 WS |
6649 | |
6650 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
31db9f7c AB |
6651 | } |
6652 | ||
6653 | /* | |
6654 | * Clones from send_root are allowed, but only if the clone source | |
6655 | * is behind the current send position. This is checked while searching | |
6656 | * for possible clone sources. | |
6657 | */ | |
6658 | sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root; | |
6659 | ||
6660 | /* We do a bsearch later */ | |
6661 | sort(sctx->clone_roots, sctx->clone_roots_cnt, | |
6662 | sizeof(*sctx->clone_roots), __clone_root_cmp_sort, | |
6663 | NULL); | |
896c14f9 | 6664 | sort_clone_roots = 1; |
31db9f7c | 6665 | |
e5fa8f86 FM |
6666 | ret = ensure_commit_roots_uptodate(sctx); |
6667 | if (ret) | |
6668 | goto out; | |
6669 | ||
2755a0de | 6670 | current->journal_info = BTRFS_SEND_TRANS_STUB; |
31db9f7c | 6671 | ret = send_subvol(sctx); |
a26e8c9f | 6672 | current->journal_info = NULL; |
31db9f7c AB |
6673 | if (ret < 0) |
6674 | goto out; | |
6675 | ||
c2c71324 SB |
6676 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { |
6677 | ret = begin_cmd(sctx, BTRFS_SEND_C_END); | |
6678 | if (ret < 0) | |
6679 | goto out; | |
6680 | ret = send_cmd(sctx); | |
6681 | if (ret < 0) | |
6682 | goto out; | |
6683 | } | |
31db9f7c AB |
6684 | |
6685 | out: | |
9f03740a FDBM |
6686 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); |
6687 | while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { | |
6688 | struct rb_node *n; | |
6689 | struct pending_dir_move *pm; | |
6690 | ||
6691 | n = rb_first(&sctx->pending_dir_moves); | |
6692 | pm = rb_entry(n, struct pending_dir_move, node); | |
6693 | while (!list_empty(&pm->list)) { | |
6694 | struct pending_dir_move *pm2; | |
6695 | ||
6696 | pm2 = list_first_entry(&pm->list, | |
6697 | struct pending_dir_move, list); | |
6698 | free_pending_move(sctx, pm2); | |
6699 | } | |
6700 | free_pending_move(sctx, pm); | |
6701 | } | |
6702 | ||
6703 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); | |
6704 | while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { | |
6705 | struct rb_node *n; | |
6706 | struct waiting_dir_move *dm; | |
6707 | ||
6708 | n = rb_first(&sctx->waiting_dir_moves); | |
6709 | dm = rb_entry(n, struct waiting_dir_move, node); | |
6710 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
6711 | kfree(dm); | |
6712 | } | |
6713 | ||
9dc44214 FM |
6714 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); |
6715 | while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { | |
6716 | struct rb_node *n; | |
6717 | struct orphan_dir_info *odi; | |
6718 | ||
6719 | n = rb_first(&sctx->orphan_dirs); | |
6720 | odi = rb_entry(n, struct orphan_dir_info, node); | |
6721 | free_orphan_dir_info(sctx, odi); | |
6722 | } | |
6723 | ||
896c14f9 WS |
6724 | if (sort_clone_roots) { |
6725 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6726 | btrfs_root_dec_send_in_progress( | |
6727 | sctx->clone_roots[i].root); | |
6728 | } else { | |
6729 | for (i = 0; sctx && i < clone_sources_to_rollback; i++) | |
6730 | btrfs_root_dec_send_in_progress( | |
6731 | sctx->clone_roots[i].root); | |
6732 | ||
6733 | btrfs_root_dec_send_in_progress(send_root); | |
6734 | } | |
66ef7d65 DS |
6735 | if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) |
6736 | btrfs_root_dec_send_in_progress(sctx->parent_root); | |
2c686537 | 6737 | |
2f91306a | 6738 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6739 | |
6740 | if (sctx) { | |
6741 | if (sctx->send_filp) | |
6742 | fput(sctx->send_filp); | |
6743 | ||
c03d01f3 | 6744 | kvfree(sctx->clone_roots); |
6ff48ce0 | 6745 | kvfree(sctx->send_buf); |
eb5b75fe | 6746 | kvfree(sctx->read_buf); |
31db9f7c AB |
6747 | |
6748 | name_cache_free(sctx); | |
6749 | ||
6750 | kfree(sctx); | |
6751 | } | |
6752 | ||
6753 | return ret; | |
6754 | } |