]>
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; |
14ecdce9 | 115 | bool ignore_cur_inode; |
31db9f7c AB |
116 | |
117 | u64 send_progress; | |
118 | ||
119 | struct list_head new_refs; | |
120 | struct list_head deleted_refs; | |
121 | ||
122 | struct radix_tree_root name_cache; | |
123 | struct list_head name_cache_list; | |
124 | int name_cache_size; | |
125 | ||
2131bcd3 LB |
126 | struct file_ra_state ra; |
127 | ||
31db9f7c | 128 | char *read_buf; |
9f03740a FDBM |
129 | |
130 | /* | |
131 | * We process inodes by their increasing order, so if before an | |
132 | * incremental send we reverse the parent/child relationship of | |
133 | * directories such that a directory with a lower inode number was | |
134 | * the parent of a directory with a higher inode number, and the one | |
135 | * becoming the new parent got renamed too, we can't rename/move the | |
136 | * directory with lower inode number when we finish processing it - we | |
137 | * must process the directory with higher inode number first, then | |
138 | * rename/move it and then rename/move the directory with lower inode | |
139 | * number. Example follows. | |
140 | * | |
141 | * Tree state when the first send was performed: | |
142 | * | |
143 | * . | |
144 | * |-- a (ino 257) | |
145 | * |-- b (ino 258) | |
146 | * | | |
147 | * | | |
148 | * |-- c (ino 259) | |
149 | * | |-- d (ino 260) | |
150 | * | | |
151 | * |-- c2 (ino 261) | |
152 | * | |
153 | * Tree state when the second (incremental) send is performed: | |
154 | * | |
155 | * . | |
156 | * |-- a (ino 257) | |
157 | * |-- b (ino 258) | |
158 | * |-- c2 (ino 261) | |
159 | * |-- d2 (ino 260) | |
160 | * |-- cc (ino 259) | |
161 | * | |
162 | * The sequence of steps that lead to the second state was: | |
163 | * | |
164 | * mv /a/b/c/d /a/b/c2/d2 | |
165 | * mv /a/b/c /a/b/c2/d2/cc | |
166 | * | |
167 | * "c" has lower inode number, but we can't move it (2nd mv operation) | |
168 | * before we move "d", which has higher inode number. | |
169 | * | |
170 | * So we just memorize which move/rename operations must be performed | |
171 | * later when their respective parent is processed and moved/renamed. | |
172 | */ | |
173 | ||
174 | /* Indexed by parent directory inode number. */ | |
175 | struct rb_root pending_dir_moves; | |
176 | ||
177 | /* | |
178 | * Reverse index, indexed by the inode number of a directory that | |
179 | * is waiting for the move/rename of its immediate parent before its | |
180 | * own move/rename can be performed. | |
181 | */ | |
182 | struct rb_root waiting_dir_moves; | |
9dc44214 FM |
183 | |
184 | /* | |
185 | * A directory that is going to be rm'ed might have a child directory | |
186 | * which is in the pending directory moves index above. In this case, | |
187 | * the directory can only be removed after the move/rename of its child | |
188 | * is performed. Example: | |
189 | * | |
190 | * Parent snapshot: | |
191 | * | |
192 | * . (ino 256) | |
193 | * |-- a/ (ino 257) | |
194 | * |-- b/ (ino 258) | |
195 | * |-- c/ (ino 259) | |
196 | * | |-- x/ (ino 260) | |
197 | * | | |
198 | * |-- y/ (ino 261) | |
199 | * | |
200 | * Send snapshot: | |
201 | * | |
202 | * . (ino 256) | |
203 | * |-- a/ (ino 257) | |
204 | * |-- b/ (ino 258) | |
205 | * |-- YY/ (ino 261) | |
206 | * |-- x/ (ino 260) | |
207 | * | |
208 | * Sequence of steps that lead to the send snapshot: | |
209 | * rm -f /a/b/c/foo.txt | |
210 | * mv /a/b/y /a/b/YY | |
211 | * mv /a/b/c/x /a/b/YY | |
212 | * rmdir /a/b/c | |
213 | * | |
214 | * When the child is processed, its move/rename is delayed until its | |
215 | * parent is processed (as explained above), but all other operations | |
216 | * like update utimes, chown, chgrp, etc, are performed and the paths | |
217 | * that it uses for those operations must use the orphanized name of | |
218 | * its parent (the directory we're going to rm later), so we need to | |
219 | * memorize that name. | |
220 | * | |
221 | * Indexed by the inode number of the directory to be deleted. | |
222 | */ | |
223 | struct rb_root orphan_dirs; | |
9f03740a FDBM |
224 | }; |
225 | ||
226 | struct pending_dir_move { | |
227 | struct rb_node node; | |
228 | struct list_head list; | |
229 | u64 parent_ino; | |
230 | u64 ino; | |
231 | u64 gen; | |
232 | struct list_head update_refs; | |
233 | }; | |
234 | ||
235 | struct waiting_dir_move { | |
236 | struct rb_node node; | |
237 | u64 ino; | |
9dc44214 FM |
238 | /* |
239 | * There might be some directory that could not be removed because it | |
240 | * was waiting for this directory inode to be moved first. Therefore | |
241 | * after this directory is moved, we can try to rmdir the ino rmdir_ino. | |
242 | */ | |
243 | u64 rmdir_ino; | |
8b191a68 | 244 | bool orphanized; |
9dc44214 FM |
245 | }; |
246 | ||
247 | struct orphan_dir_info { | |
248 | struct rb_node node; | |
249 | u64 ino; | |
250 | u64 gen; | |
31db9f7c AB |
251 | }; |
252 | ||
253 | struct name_cache_entry { | |
254 | struct list_head list; | |
7e0926fe AB |
255 | /* |
256 | * radix_tree has only 32bit entries but we need to handle 64bit inums. | |
257 | * We use the lower 32bit of the 64bit inum to store it in the tree. If | |
258 | * more then one inum would fall into the same entry, we use radix_list | |
259 | * to store the additional entries. radix_list is also used to store | |
260 | * entries where two entries have the same inum but different | |
261 | * generations. | |
262 | */ | |
263 | struct list_head radix_list; | |
31db9f7c AB |
264 | u64 ino; |
265 | u64 gen; | |
266 | u64 parent_ino; | |
267 | u64 parent_gen; | |
268 | int ret; | |
269 | int need_later_update; | |
270 | int name_len; | |
271 | char name[]; | |
272 | }; | |
273 | ||
95155585 FM |
274 | static void inconsistent_snapshot_error(struct send_ctx *sctx, |
275 | enum btrfs_compare_tree_result result, | |
276 | const char *what) | |
277 | { | |
278 | const char *result_string; | |
279 | ||
280 | switch (result) { | |
281 | case BTRFS_COMPARE_TREE_NEW: | |
282 | result_string = "new"; | |
283 | break; | |
284 | case BTRFS_COMPARE_TREE_DELETED: | |
285 | result_string = "deleted"; | |
286 | break; | |
287 | case BTRFS_COMPARE_TREE_CHANGED: | |
288 | result_string = "updated"; | |
289 | break; | |
290 | case BTRFS_COMPARE_TREE_SAME: | |
291 | ASSERT(0); | |
292 | result_string = "unchanged"; | |
293 | break; | |
294 | default: | |
295 | ASSERT(0); | |
296 | result_string = "unexpected"; | |
297 | } | |
298 | ||
299 | btrfs_err(sctx->send_root->fs_info, | |
300 | "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu", | |
301 | result_string, what, sctx->cmp_key->objectid, | |
302 | sctx->send_root->root_key.objectid, | |
303 | (sctx->parent_root ? | |
304 | sctx->parent_root->root_key.objectid : 0)); | |
305 | } | |
306 | ||
9f03740a FDBM |
307 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); |
308 | ||
9dc44214 FM |
309 | static struct waiting_dir_move * |
310 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino); | |
311 | ||
312 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino); | |
313 | ||
16e7549f JB |
314 | static int need_send_hole(struct send_ctx *sctx) |
315 | { | |
316 | return (sctx->parent_root && !sctx->cur_inode_new && | |
317 | !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && | |
318 | S_ISREG(sctx->cur_inode_mode)); | |
319 | } | |
320 | ||
31db9f7c AB |
321 | static void fs_path_reset(struct fs_path *p) |
322 | { | |
323 | if (p->reversed) { | |
324 | p->start = p->buf + p->buf_len - 1; | |
325 | p->end = p->start; | |
326 | *p->start = 0; | |
327 | } else { | |
328 | p->start = p->buf; | |
329 | p->end = p->start; | |
330 | *p->start = 0; | |
331 | } | |
332 | } | |
333 | ||
924794c9 | 334 | static struct fs_path *fs_path_alloc(void) |
31db9f7c AB |
335 | { |
336 | struct fs_path *p; | |
337 | ||
e780b0d1 | 338 | p = kmalloc(sizeof(*p), GFP_KERNEL); |
31db9f7c AB |
339 | if (!p) |
340 | return NULL; | |
341 | p->reversed = 0; | |
31db9f7c AB |
342 | p->buf = p->inline_buf; |
343 | p->buf_len = FS_PATH_INLINE_SIZE; | |
344 | fs_path_reset(p); | |
345 | return p; | |
346 | } | |
347 | ||
924794c9 | 348 | static struct fs_path *fs_path_alloc_reversed(void) |
31db9f7c AB |
349 | { |
350 | struct fs_path *p; | |
351 | ||
924794c9 | 352 | p = fs_path_alloc(); |
31db9f7c AB |
353 | if (!p) |
354 | return NULL; | |
355 | p->reversed = 1; | |
356 | fs_path_reset(p); | |
357 | return p; | |
358 | } | |
359 | ||
924794c9 | 360 | static void fs_path_free(struct fs_path *p) |
31db9f7c AB |
361 | { |
362 | if (!p) | |
363 | return; | |
ace01050 DS |
364 | if (p->buf != p->inline_buf) |
365 | kfree(p->buf); | |
31db9f7c AB |
366 | kfree(p); |
367 | } | |
368 | ||
369 | static int fs_path_len(struct fs_path *p) | |
370 | { | |
371 | return p->end - p->start; | |
372 | } | |
373 | ||
374 | static int fs_path_ensure_buf(struct fs_path *p, int len) | |
375 | { | |
376 | char *tmp_buf; | |
377 | int path_len; | |
378 | int old_buf_len; | |
379 | ||
380 | len++; | |
381 | ||
382 | if (p->buf_len >= len) | |
383 | return 0; | |
384 | ||
cfd4a535 CM |
385 | if (len > PATH_MAX) { |
386 | WARN_ON(1); | |
387 | return -ENOMEM; | |
388 | } | |
389 | ||
1b2782c8 DS |
390 | path_len = p->end - p->start; |
391 | old_buf_len = p->buf_len; | |
392 | ||
ace01050 DS |
393 | /* |
394 | * First time the inline_buf does not suffice | |
395 | */ | |
01a9a8a9 | 396 | if (p->buf == p->inline_buf) { |
e780b0d1 | 397 | tmp_buf = kmalloc(len, GFP_KERNEL); |
01a9a8a9 FM |
398 | if (tmp_buf) |
399 | memcpy(tmp_buf, p->buf, old_buf_len); | |
400 | } else { | |
e780b0d1 | 401 | tmp_buf = krealloc(p->buf, len, GFP_KERNEL); |
01a9a8a9 | 402 | } |
9c9ca00b DS |
403 | if (!tmp_buf) |
404 | return -ENOMEM; | |
405 | p->buf = tmp_buf; | |
406 | /* | |
407 | * The real size of the buffer is bigger, this will let the fast path | |
408 | * happen most of the time | |
409 | */ | |
410 | p->buf_len = ksize(p->buf); | |
ace01050 | 411 | |
31db9f7c AB |
412 | if (p->reversed) { |
413 | tmp_buf = p->buf + old_buf_len - path_len - 1; | |
414 | p->end = p->buf + p->buf_len - 1; | |
415 | p->start = p->end - path_len; | |
416 | memmove(p->start, tmp_buf, path_len + 1); | |
417 | } else { | |
418 | p->start = p->buf; | |
419 | p->end = p->start + path_len; | |
420 | } | |
421 | return 0; | |
422 | } | |
423 | ||
b23ab57d DS |
424 | static int fs_path_prepare_for_add(struct fs_path *p, int name_len, |
425 | char **prepared) | |
31db9f7c AB |
426 | { |
427 | int ret; | |
428 | int new_len; | |
429 | ||
430 | new_len = p->end - p->start + name_len; | |
431 | if (p->start != p->end) | |
432 | new_len++; | |
433 | ret = fs_path_ensure_buf(p, new_len); | |
434 | if (ret < 0) | |
435 | goto out; | |
436 | ||
437 | if (p->reversed) { | |
438 | if (p->start != p->end) | |
439 | *--p->start = '/'; | |
440 | p->start -= name_len; | |
b23ab57d | 441 | *prepared = p->start; |
31db9f7c AB |
442 | } else { |
443 | if (p->start != p->end) | |
444 | *p->end++ = '/'; | |
b23ab57d | 445 | *prepared = p->end; |
31db9f7c AB |
446 | p->end += name_len; |
447 | *p->end = 0; | |
448 | } | |
449 | ||
450 | out: | |
451 | return ret; | |
452 | } | |
453 | ||
454 | static int fs_path_add(struct fs_path *p, const char *name, int name_len) | |
455 | { | |
456 | int ret; | |
b23ab57d | 457 | char *prepared; |
31db9f7c | 458 | |
b23ab57d | 459 | ret = fs_path_prepare_for_add(p, name_len, &prepared); |
31db9f7c AB |
460 | if (ret < 0) |
461 | goto out; | |
b23ab57d | 462 | memcpy(prepared, name, name_len); |
31db9f7c AB |
463 | |
464 | out: | |
465 | return ret; | |
466 | } | |
467 | ||
468 | static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) | |
469 | { | |
470 | int ret; | |
b23ab57d | 471 | char *prepared; |
31db9f7c | 472 | |
b23ab57d | 473 | ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); |
31db9f7c AB |
474 | if (ret < 0) |
475 | goto out; | |
b23ab57d | 476 | memcpy(prepared, p2->start, p2->end - p2->start); |
31db9f7c AB |
477 | |
478 | out: | |
479 | return ret; | |
480 | } | |
481 | ||
482 | static int fs_path_add_from_extent_buffer(struct fs_path *p, | |
483 | struct extent_buffer *eb, | |
484 | unsigned long off, int len) | |
485 | { | |
486 | int ret; | |
b23ab57d | 487 | char *prepared; |
31db9f7c | 488 | |
b23ab57d | 489 | ret = fs_path_prepare_for_add(p, len, &prepared); |
31db9f7c AB |
490 | if (ret < 0) |
491 | goto out; | |
492 | ||
b23ab57d | 493 | read_extent_buffer(eb, prepared, off, len); |
31db9f7c AB |
494 | |
495 | out: | |
496 | return ret; | |
497 | } | |
498 | ||
31db9f7c AB |
499 | static int fs_path_copy(struct fs_path *p, struct fs_path *from) |
500 | { | |
501 | int ret; | |
502 | ||
503 | p->reversed = from->reversed; | |
504 | fs_path_reset(p); | |
505 | ||
506 | ret = fs_path_add_path(p, from); | |
507 | ||
508 | return ret; | |
509 | } | |
510 | ||
511 | ||
512 | static void fs_path_unreverse(struct fs_path *p) | |
513 | { | |
514 | char *tmp; | |
515 | int len; | |
516 | ||
517 | if (!p->reversed) | |
518 | return; | |
519 | ||
520 | tmp = p->start; | |
521 | len = p->end - p->start; | |
522 | p->start = p->buf; | |
523 | p->end = p->start + len; | |
524 | memmove(p->start, tmp, len + 1); | |
525 | p->reversed = 0; | |
526 | } | |
527 | ||
528 | static struct btrfs_path *alloc_path_for_send(void) | |
529 | { | |
530 | struct btrfs_path *path; | |
531 | ||
532 | path = btrfs_alloc_path(); | |
533 | if (!path) | |
534 | return NULL; | |
535 | path->search_commit_root = 1; | |
536 | path->skip_locking = 1; | |
3f8a18cc | 537 | path->need_commit_sem = 1; |
31db9f7c AB |
538 | return path; |
539 | } | |
540 | ||
48a3b636 | 541 | static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) |
31db9f7c AB |
542 | { |
543 | int ret; | |
31db9f7c AB |
544 | u32 pos = 0; |
545 | ||
31db9f7c | 546 | while (pos < len) { |
8e93157b | 547 | ret = kernel_write(filp, buf + pos, len - pos, off); |
31db9f7c AB |
548 | /* TODO handle that correctly */ |
549 | /*if (ret == -ERESTARTSYS) { | |
550 | continue; | |
551 | }*/ | |
552 | if (ret < 0) | |
8e93157b | 553 | return ret; |
31db9f7c | 554 | if (ret == 0) { |
8e93157b | 555 | return -EIO; |
31db9f7c AB |
556 | } |
557 | pos += ret; | |
558 | } | |
559 | ||
8e93157b | 560 | return 0; |
31db9f7c AB |
561 | } |
562 | ||
563 | static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) | |
564 | { | |
565 | struct btrfs_tlv_header *hdr; | |
566 | int total_len = sizeof(*hdr) + len; | |
567 | int left = sctx->send_max_size - sctx->send_size; | |
568 | ||
569 | if (unlikely(left < total_len)) | |
570 | return -EOVERFLOW; | |
571 | ||
572 | hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); | |
573 | hdr->tlv_type = cpu_to_le16(attr); | |
574 | hdr->tlv_len = cpu_to_le16(len); | |
575 | memcpy(hdr + 1, data, len); | |
576 | sctx->send_size += total_len; | |
577 | ||
578 | return 0; | |
579 | } | |
580 | ||
95bc79d5 DS |
581 | #define TLV_PUT_DEFINE_INT(bits) \ |
582 | static int tlv_put_u##bits(struct send_ctx *sctx, \ | |
583 | u##bits attr, u##bits value) \ | |
584 | { \ | |
585 | __le##bits __tmp = cpu_to_le##bits(value); \ | |
586 | return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ | |
587 | } | |
31db9f7c | 588 | |
95bc79d5 | 589 | TLV_PUT_DEFINE_INT(64) |
31db9f7c AB |
590 | |
591 | static int tlv_put_string(struct send_ctx *sctx, u16 attr, | |
592 | const char *str, int len) | |
593 | { | |
594 | if (len == -1) | |
595 | len = strlen(str); | |
596 | return tlv_put(sctx, attr, str, len); | |
597 | } | |
598 | ||
599 | static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, | |
600 | const u8 *uuid) | |
601 | { | |
602 | return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); | |
603 | } | |
604 | ||
31db9f7c AB |
605 | static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, |
606 | struct extent_buffer *eb, | |
607 | struct btrfs_timespec *ts) | |
608 | { | |
609 | struct btrfs_timespec bts; | |
610 | read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); | |
611 | return tlv_put(sctx, attr, &bts, sizeof(bts)); | |
612 | } | |
613 | ||
614 | ||
615 | #define TLV_PUT(sctx, attrtype, attrlen, data) \ | |
616 | do { \ | |
617 | ret = tlv_put(sctx, attrtype, attrlen, data); \ | |
618 | if (ret < 0) \ | |
619 | goto tlv_put_failure; \ | |
620 | } while (0) | |
621 | ||
622 | #define TLV_PUT_INT(sctx, attrtype, bits, value) \ | |
623 | do { \ | |
624 | ret = tlv_put_u##bits(sctx, attrtype, value); \ | |
625 | if (ret < 0) \ | |
626 | goto tlv_put_failure; \ | |
627 | } while (0) | |
628 | ||
629 | #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) | |
630 | #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) | |
631 | #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) | |
632 | #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) | |
633 | #define TLV_PUT_STRING(sctx, attrtype, str, len) \ | |
634 | do { \ | |
635 | ret = tlv_put_string(sctx, attrtype, str, len); \ | |
636 | if (ret < 0) \ | |
637 | goto tlv_put_failure; \ | |
638 | } while (0) | |
639 | #define TLV_PUT_PATH(sctx, attrtype, p) \ | |
640 | do { \ | |
641 | ret = tlv_put_string(sctx, attrtype, p->start, \ | |
642 | p->end - p->start); \ | |
643 | if (ret < 0) \ | |
644 | goto tlv_put_failure; \ | |
645 | } while(0) | |
646 | #define TLV_PUT_UUID(sctx, attrtype, uuid) \ | |
647 | do { \ | |
648 | ret = tlv_put_uuid(sctx, attrtype, uuid); \ | |
649 | if (ret < 0) \ | |
650 | goto tlv_put_failure; \ | |
651 | } while (0) | |
31db9f7c AB |
652 | #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ |
653 | do { \ | |
654 | ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ | |
655 | if (ret < 0) \ | |
656 | goto tlv_put_failure; \ | |
657 | } while (0) | |
658 | ||
659 | static int send_header(struct send_ctx *sctx) | |
660 | { | |
661 | struct btrfs_stream_header hdr; | |
662 | ||
663 | strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); | |
664 | hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); | |
665 | ||
1bcea355 AJ |
666 | return write_buf(sctx->send_filp, &hdr, sizeof(hdr), |
667 | &sctx->send_off); | |
31db9f7c AB |
668 | } |
669 | ||
670 | /* | |
671 | * For each command/item we want to send to userspace, we call this function. | |
672 | */ | |
673 | static int begin_cmd(struct send_ctx *sctx, int cmd) | |
674 | { | |
675 | struct btrfs_cmd_header *hdr; | |
676 | ||
fae7f21c | 677 | if (WARN_ON(!sctx->send_buf)) |
31db9f7c | 678 | return -EINVAL; |
31db9f7c AB |
679 | |
680 | BUG_ON(sctx->send_size); | |
681 | ||
682 | sctx->send_size += sizeof(*hdr); | |
683 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
684 | hdr->cmd = cpu_to_le16(cmd); | |
685 | ||
686 | return 0; | |
687 | } | |
688 | ||
689 | static int send_cmd(struct send_ctx *sctx) | |
690 | { | |
691 | int ret; | |
692 | struct btrfs_cmd_header *hdr; | |
693 | u32 crc; | |
694 | ||
695 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
696 | hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr)); | |
697 | hdr->crc = 0; | |
698 | ||
0b947aff | 699 | crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); |
31db9f7c AB |
700 | hdr->crc = cpu_to_le32(crc); |
701 | ||
1bcea355 AJ |
702 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, |
703 | &sctx->send_off); | |
31db9f7c AB |
704 | |
705 | sctx->total_send_size += sctx->send_size; | |
706 | sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size; | |
707 | sctx->send_size = 0; | |
708 | ||
709 | return ret; | |
710 | } | |
711 | ||
712 | /* | |
713 | * Sends a move instruction to user space | |
714 | */ | |
715 | static int send_rename(struct send_ctx *sctx, | |
716 | struct fs_path *from, struct fs_path *to) | |
717 | { | |
04ab956e | 718 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
719 | int ret; |
720 | ||
04ab956e | 721 | btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start); |
31db9f7c AB |
722 | |
723 | ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); | |
724 | if (ret < 0) | |
725 | goto out; | |
726 | ||
727 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); | |
728 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); | |
729 | ||
730 | ret = send_cmd(sctx); | |
731 | ||
732 | tlv_put_failure: | |
733 | out: | |
734 | return ret; | |
735 | } | |
736 | ||
737 | /* | |
738 | * Sends a link instruction to user space | |
739 | */ | |
740 | static int send_link(struct send_ctx *sctx, | |
741 | struct fs_path *path, struct fs_path *lnk) | |
742 | { | |
04ab956e | 743 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
744 | int ret; |
745 | ||
04ab956e | 746 | btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start); |
31db9f7c AB |
747 | |
748 | ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); | |
749 | if (ret < 0) | |
750 | goto out; | |
751 | ||
752 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
753 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); | |
754 | ||
755 | ret = send_cmd(sctx); | |
756 | ||
757 | tlv_put_failure: | |
758 | out: | |
759 | return ret; | |
760 | } | |
761 | ||
762 | /* | |
763 | * Sends an unlink instruction to user space | |
764 | */ | |
765 | static int send_unlink(struct send_ctx *sctx, struct fs_path *path) | |
766 | { | |
04ab956e | 767 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
768 | int ret; |
769 | ||
04ab956e | 770 | btrfs_debug(fs_info, "send_unlink %s", path->start); |
31db9f7c AB |
771 | |
772 | ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); | |
773 | if (ret < 0) | |
774 | goto out; | |
775 | ||
776 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
777 | ||
778 | ret = send_cmd(sctx); | |
779 | ||
780 | tlv_put_failure: | |
781 | out: | |
782 | return ret; | |
783 | } | |
784 | ||
785 | /* | |
786 | * Sends a rmdir instruction to user space | |
787 | */ | |
788 | static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) | |
789 | { | |
04ab956e | 790 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
791 | int ret; |
792 | ||
04ab956e | 793 | btrfs_debug(fs_info, "send_rmdir %s", path->start); |
31db9f7c AB |
794 | |
795 | ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); | |
796 | if (ret < 0) | |
797 | goto out; | |
798 | ||
799 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
800 | ||
801 | ret = send_cmd(sctx); | |
802 | ||
803 | tlv_put_failure: | |
804 | out: | |
805 | return ret; | |
806 | } | |
807 | ||
808 | /* | |
809 | * Helper function to retrieve some fields from an inode item. | |
810 | */ | |
3f8a18cc JB |
811 | static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, |
812 | u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, | |
813 | u64 *gid, u64 *rdev) | |
31db9f7c AB |
814 | { |
815 | int ret; | |
816 | struct btrfs_inode_item *ii; | |
817 | struct btrfs_key key; | |
31db9f7c AB |
818 | |
819 | key.objectid = ino; | |
820 | key.type = BTRFS_INODE_ITEM_KEY; | |
821 | key.offset = 0; | |
822 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
31db9f7c | 823 | if (ret) { |
3f8a18cc JB |
824 | if (ret > 0) |
825 | ret = -ENOENT; | |
826 | return ret; | |
31db9f7c AB |
827 | } |
828 | ||
829 | ii = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
830 | struct btrfs_inode_item); | |
831 | if (size) | |
832 | *size = btrfs_inode_size(path->nodes[0], ii); | |
833 | if (gen) | |
834 | *gen = btrfs_inode_generation(path->nodes[0], ii); | |
835 | if (mode) | |
836 | *mode = btrfs_inode_mode(path->nodes[0], ii); | |
837 | if (uid) | |
838 | *uid = btrfs_inode_uid(path->nodes[0], ii); | |
839 | if (gid) | |
840 | *gid = btrfs_inode_gid(path->nodes[0], ii); | |
85a7b33b AB |
841 | if (rdev) |
842 | *rdev = btrfs_inode_rdev(path->nodes[0], ii); | |
31db9f7c | 843 | |
3f8a18cc JB |
844 | return ret; |
845 | } | |
846 | ||
847 | static int get_inode_info(struct btrfs_root *root, | |
848 | u64 ino, u64 *size, u64 *gen, | |
849 | u64 *mode, u64 *uid, u64 *gid, | |
850 | u64 *rdev) | |
851 | { | |
852 | struct btrfs_path *path; | |
853 | int ret; | |
854 | ||
855 | path = alloc_path_for_send(); | |
856 | if (!path) | |
857 | return -ENOMEM; | |
858 | ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, | |
859 | rdev); | |
31db9f7c AB |
860 | btrfs_free_path(path); |
861 | return ret; | |
862 | } | |
863 | ||
864 | typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, | |
865 | struct fs_path *p, | |
866 | void *ctx); | |
867 | ||
868 | /* | |
96b5bd77 JS |
869 | * Helper function to iterate the entries in ONE btrfs_inode_ref or |
870 | * btrfs_inode_extref. | |
31db9f7c AB |
871 | * The iterate callback may return a non zero value to stop iteration. This can |
872 | * be a negative value for error codes or 1 to simply stop it. | |
873 | * | |
96b5bd77 | 874 | * path must point to the INODE_REF or INODE_EXTREF when called. |
31db9f7c | 875 | */ |
924794c9 | 876 | static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
877 | struct btrfs_key *found_key, int resolve, |
878 | iterate_inode_ref_t iterate, void *ctx) | |
879 | { | |
96b5bd77 | 880 | struct extent_buffer *eb = path->nodes[0]; |
31db9f7c AB |
881 | struct btrfs_item *item; |
882 | struct btrfs_inode_ref *iref; | |
96b5bd77 | 883 | struct btrfs_inode_extref *extref; |
31db9f7c AB |
884 | struct btrfs_path *tmp_path; |
885 | struct fs_path *p; | |
96b5bd77 | 886 | u32 cur = 0; |
31db9f7c | 887 | u32 total; |
96b5bd77 | 888 | int slot = path->slots[0]; |
31db9f7c AB |
889 | u32 name_len; |
890 | char *start; | |
891 | int ret = 0; | |
96b5bd77 | 892 | int num = 0; |
31db9f7c | 893 | int index; |
96b5bd77 JS |
894 | u64 dir; |
895 | unsigned long name_off; | |
896 | unsigned long elem_size; | |
897 | unsigned long ptr; | |
31db9f7c | 898 | |
924794c9 | 899 | p = fs_path_alloc_reversed(); |
31db9f7c AB |
900 | if (!p) |
901 | return -ENOMEM; | |
902 | ||
903 | tmp_path = alloc_path_for_send(); | |
904 | if (!tmp_path) { | |
924794c9 | 905 | fs_path_free(p); |
31db9f7c AB |
906 | return -ENOMEM; |
907 | } | |
908 | ||
31db9f7c | 909 | |
96b5bd77 JS |
910 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
911 | ptr = (unsigned long)btrfs_item_ptr(eb, slot, | |
912 | struct btrfs_inode_ref); | |
dd3cc16b | 913 | item = btrfs_item_nr(slot); |
96b5bd77 JS |
914 | total = btrfs_item_size(eb, item); |
915 | elem_size = sizeof(*iref); | |
916 | } else { | |
917 | ptr = btrfs_item_ptr_offset(eb, slot); | |
918 | total = btrfs_item_size_nr(eb, slot); | |
919 | elem_size = sizeof(*extref); | |
920 | } | |
921 | ||
31db9f7c AB |
922 | while (cur < total) { |
923 | fs_path_reset(p); | |
924 | ||
96b5bd77 JS |
925 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
926 | iref = (struct btrfs_inode_ref *)(ptr + cur); | |
927 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
928 | name_off = (unsigned long)(iref + 1); | |
929 | index = btrfs_inode_ref_index(eb, iref); | |
930 | dir = found_key->offset; | |
931 | } else { | |
932 | extref = (struct btrfs_inode_extref *)(ptr + cur); | |
933 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
934 | name_off = (unsigned long)&extref->name; | |
935 | index = btrfs_inode_extref_index(eb, extref); | |
936 | dir = btrfs_inode_extref_parent(eb, extref); | |
937 | } | |
938 | ||
31db9f7c | 939 | if (resolve) { |
96b5bd77 JS |
940 | start = btrfs_ref_to_path(root, tmp_path, name_len, |
941 | name_off, eb, dir, | |
942 | p->buf, p->buf_len); | |
31db9f7c AB |
943 | if (IS_ERR(start)) { |
944 | ret = PTR_ERR(start); | |
945 | goto out; | |
946 | } | |
947 | if (start < p->buf) { | |
948 | /* overflow , try again with larger buffer */ | |
949 | ret = fs_path_ensure_buf(p, | |
950 | p->buf_len + p->buf - start); | |
951 | if (ret < 0) | |
952 | goto out; | |
96b5bd77 JS |
953 | start = btrfs_ref_to_path(root, tmp_path, |
954 | name_len, name_off, | |
955 | eb, dir, | |
956 | p->buf, p->buf_len); | |
31db9f7c AB |
957 | if (IS_ERR(start)) { |
958 | ret = PTR_ERR(start); | |
959 | goto out; | |
960 | } | |
961 | BUG_ON(start < p->buf); | |
962 | } | |
963 | p->start = start; | |
964 | } else { | |
96b5bd77 JS |
965 | ret = fs_path_add_from_extent_buffer(p, eb, name_off, |
966 | name_len); | |
31db9f7c AB |
967 | if (ret < 0) |
968 | goto out; | |
969 | } | |
970 | ||
96b5bd77 JS |
971 | cur += elem_size + name_len; |
972 | ret = iterate(num, dir, index, p, ctx); | |
31db9f7c AB |
973 | if (ret) |
974 | goto out; | |
31db9f7c AB |
975 | num++; |
976 | } | |
977 | ||
978 | out: | |
979 | btrfs_free_path(tmp_path); | |
924794c9 | 980 | fs_path_free(p); |
31db9f7c AB |
981 | return ret; |
982 | } | |
983 | ||
984 | typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, | |
985 | const char *name, int name_len, | |
986 | const char *data, int data_len, | |
987 | u8 type, void *ctx); | |
988 | ||
989 | /* | |
990 | * Helper function to iterate the entries in ONE btrfs_dir_item. | |
991 | * The iterate callback may return a non zero value to stop iteration. This can | |
992 | * be a negative value for error codes or 1 to simply stop it. | |
993 | * | |
994 | * path must point to the dir item when called. | |
995 | */ | |
924794c9 | 996 | static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
997 | iterate_dir_item_t iterate, void *ctx) |
998 | { | |
999 | int ret = 0; | |
1000 | struct extent_buffer *eb; | |
1001 | struct btrfs_item *item; | |
1002 | struct btrfs_dir_item *di; | |
31db9f7c AB |
1003 | struct btrfs_key di_key; |
1004 | char *buf = NULL; | |
7e3ae33e | 1005 | int buf_len; |
31db9f7c AB |
1006 | u32 name_len; |
1007 | u32 data_len; | |
1008 | u32 cur; | |
1009 | u32 len; | |
1010 | u32 total; | |
1011 | int slot; | |
1012 | int num; | |
1013 | u8 type; | |
1014 | ||
4395e0c4 FM |
1015 | /* |
1016 | * Start with a small buffer (1 page). If later we end up needing more | |
1017 | * space, which can happen for xattrs on a fs with a leaf size greater | |
1018 | * then the page size, attempt to increase the buffer. Typically xattr | |
1019 | * values are small. | |
1020 | */ | |
1021 | buf_len = PATH_MAX; | |
e780b0d1 | 1022 | buf = kmalloc(buf_len, GFP_KERNEL); |
31db9f7c AB |
1023 | if (!buf) { |
1024 | ret = -ENOMEM; | |
1025 | goto out; | |
1026 | } | |
1027 | ||
31db9f7c AB |
1028 | eb = path->nodes[0]; |
1029 | slot = path->slots[0]; | |
dd3cc16b | 1030 | item = btrfs_item_nr(slot); |
31db9f7c AB |
1031 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
1032 | cur = 0; | |
1033 | len = 0; | |
1034 | total = btrfs_item_size(eb, item); | |
1035 | ||
1036 | num = 0; | |
1037 | while (cur < total) { | |
1038 | name_len = btrfs_dir_name_len(eb, di); | |
1039 | data_len = btrfs_dir_data_len(eb, di); | |
1040 | type = btrfs_dir_type(eb, di); | |
1041 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1042 | ||
7e3ae33e FM |
1043 | if (type == BTRFS_FT_XATTR) { |
1044 | if (name_len > XATTR_NAME_MAX) { | |
1045 | ret = -ENAMETOOLONG; | |
1046 | goto out; | |
1047 | } | |
da17066c JM |
1048 | if (name_len + data_len > |
1049 | BTRFS_MAX_XATTR_SIZE(root->fs_info)) { | |
7e3ae33e FM |
1050 | ret = -E2BIG; |
1051 | goto out; | |
1052 | } | |
1053 | } else { | |
1054 | /* | |
1055 | * Path too long | |
1056 | */ | |
4395e0c4 | 1057 | if (name_len + data_len > PATH_MAX) { |
7e3ae33e FM |
1058 | ret = -ENAMETOOLONG; |
1059 | goto out; | |
1060 | } | |
31db9f7c AB |
1061 | } |
1062 | ||
59b0a7f2 SY |
1063 | ret = btrfs_is_name_len_valid(eb, path->slots[0], |
1064 | (unsigned long)(di + 1), name_len + data_len); | |
1065 | if (!ret) { | |
1066 | ret = -EIO; | |
1067 | goto out; | |
1068 | } | |
4395e0c4 FM |
1069 | if (name_len + data_len > buf_len) { |
1070 | buf_len = name_len + data_len; | |
1071 | if (is_vmalloc_addr(buf)) { | |
1072 | vfree(buf); | |
1073 | buf = NULL; | |
1074 | } else { | |
1075 | char *tmp = krealloc(buf, buf_len, | |
e780b0d1 | 1076 | GFP_KERNEL | __GFP_NOWARN); |
4395e0c4 FM |
1077 | |
1078 | if (!tmp) | |
1079 | kfree(buf); | |
1080 | buf = tmp; | |
1081 | } | |
1082 | if (!buf) { | |
f11f7441 | 1083 | buf = kvmalloc(buf_len, GFP_KERNEL); |
4395e0c4 FM |
1084 | if (!buf) { |
1085 | ret = -ENOMEM; | |
1086 | goto out; | |
1087 | } | |
1088 | } | |
1089 | } | |
1090 | ||
31db9f7c AB |
1091 | read_extent_buffer(eb, buf, (unsigned long)(di + 1), |
1092 | name_len + data_len); | |
1093 | ||
1094 | len = sizeof(*di) + name_len + data_len; | |
1095 | di = (struct btrfs_dir_item *)((char *)di + len); | |
1096 | cur += len; | |
1097 | ||
1098 | ret = iterate(num, &di_key, buf, name_len, buf + name_len, | |
1099 | data_len, type, ctx); | |
1100 | if (ret < 0) | |
1101 | goto out; | |
1102 | if (ret) { | |
1103 | ret = 0; | |
1104 | goto out; | |
1105 | } | |
1106 | ||
1107 | num++; | |
1108 | } | |
1109 | ||
1110 | out: | |
4395e0c4 | 1111 | kvfree(buf); |
31db9f7c AB |
1112 | return ret; |
1113 | } | |
1114 | ||
1115 | static int __copy_first_ref(int num, u64 dir, int index, | |
1116 | struct fs_path *p, void *ctx) | |
1117 | { | |
1118 | int ret; | |
1119 | struct fs_path *pt = ctx; | |
1120 | ||
1121 | ret = fs_path_copy(pt, p); | |
1122 | if (ret < 0) | |
1123 | return ret; | |
1124 | ||
1125 | /* we want the first only */ | |
1126 | return 1; | |
1127 | } | |
1128 | ||
1129 | /* | |
1130 | * Retrieve the first path of an inode. If an inode has more then one | |
1131 | * ref/hardlink, this is ignored. | |
1132 | */ | |
924794c9 | 1133 | static int get_inode_path(struct btrfs_root *root, |
31db9f7c AB |
1134 | u64 ino, struct fs_path *path) |
1135 | { | |
1136 | int ret; | |
1137 | struct btrfs_key key, found_key; | |
1138 | struct btrfs_path *p; | |
1139 | ||
1140 | p = alloc_path_for_send(); | |
1141 | if (!p) | |
1142 | return -ENOMEM; | |
1143 | ||
1144 | fs_path_reset(path); | |
1145 | ||
1146 | key.objectid = ino; | |
1147 | key.type = BTRFS_INODE_REF_KEY; | |
1148 | key.offset = 0; | |
1149 | ||
1150 | ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); | |
1151 | if (ret < 0) | |
1152 | goto out; | |
1153 | if (ret) { | |
1154 | ret = 1; | |
1155 | goto out; | |
1156 | } | |
1157 | btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); | |
1158 | if (found_key.objectid != ino || | |
96b5bd77 JS |
1159 | (found_key.type != BTRFS_INODE_REF_KEY && |
1160 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1161 | ret = -ENOENT; |
1162 | goto out; | |
1163 | } | |
1164 | ||
924794c9 TI |
1165 | ret = iterate_inode_ref(root, p, &found_key, 1, |
1166 | __copy_first_ref, path); | |
31db9f7c AB |
1167 | if (ret < 0) |
1168 | goto out; | |
1169 | ret = 0; | |
1170 | ||
1171 | out: | |
1172 | btrfs_free_path(p); | |
1173 | return ret; | |
1174 | } | |
1175 | ||
1176 | struct backref_ctx { | |
1177 | struct send_ctx *sctx; | |
1178 | ||
3f8a18cc | 1179 | struct btrfs_path *path; |
31db9f7c AB |
1180 | /* number of total found references */ |
1181 | u64 found; | |
1182 | ||
1183 | /* | |
1184 | * used for clones found in send_root. clones found behind cur_objectid | |
1185 | * and cur_offset are not considered as allowed clones. | |
1186 | */ | |
1187 | u64 cur_objectid; | |
1188 | u64 cur_offset; | |
1189 | ||
1190 | /* may be truncated in case it's the last extent in a file */ | |
1191 | u64 extent_len; | |
1192 | ||
619d8c4e FM |
1193 | /* data offset in the file extent item */ |
1194 | u64 data_offset; | |
1195 | ||
31db9f7c | 1196 | /* Just to check for bugs in backref resolving */ |
ee849c04 | 1197 | int found_itself; |
31db9f7c AB |
1198 | }; |
1199 | ||
1200 | static int __clone_root_cmp_bsearch(const void *key, const void *elt) | |
1201 | { | |
995e01b7 | 1202 | u64 root = (u64)(uintptr_t)key; |
31db9f7c AB |
1203 | struct clone_root *cr = (struct clone_root *)elt; |
1204 | ||
1205 | if (root < cr->root->objectid) | |
1206 | return -1; | |
1207 | if (root > cr->root->objectid) | |
1208 | return 1; | |
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | static int __clone_root_cmp_sort(const void *e1, const void *e2) | |
1213 | { | |
1214 | struct clone_root *cr1 = (struct clone_root *)e1; | |
1215 | struct clone_root *cr2 = (struct clone_root *)e2; | |
1216 | ||
1217 | if (cr1->root->objectid < cr2->root->objectid) | |
1218 | return -1; | |
1219 | if (cr1->root->objectid > cr2->root->objectid) | |
1220 | return 1; | |
1221 | return 0; | |
1222 | } | |
1223 | ||
1224 | /* | |
1225 | * Called for every backref that is found for the current extent. | |
766702ef | 1226 | * Results are collected in sctx->clone_roots->ino/offset/found_refs |
31db9f7c AB |
1227 | */ |
1228 | static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) | |
1229 | { | |
1230 | struct backref_ctx *bctx = ctx_; | |
1231 | struct clone_root *found; | |
1232 | int ret; | |
1233 | u64 i_size; | |
1234 | ||
1235 | /* First check if the root is in the list of accepted clone sources */ | |
995e01b7 | 1236 | found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, |
31db9f7c AB |
1237 | bctx->sctx->clone_roots_cnt, |
1238 | sizeof(struct clone_root), | |
1239 | __clone_root_cmp_bsearch); | |
1240 | if (!found) | |
1241 | return 0; | |
1242 | ||
1243 | if (found->root == bctx->sctx->send_root && | |
1244 | ino == bctx->cur_objectid && | |
1245 | offset == bctx->cur_offset) { | |
ee849c04 | 1246 | bctx->found_itself = 1; |
31db9f7c AB |
1247 | } |
1248 | ||
1249 | /* | |
766702ef | 1250 | * There are inodes that have extents that lie behind its i_size. Don't |
31db9f7c AB |
1251 | * accept clones from these extents. |
1252 | */ | |
3f8a18cc JB |
1253 | ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL, |
1254 | NULL, NULL, NULL); | |
1255 | btrfs_release_path(bctx->path); | |
31db9f7c AB |
1256 | if (ret < 0) |
1257 | return ret; | |
1258 | ||
619d8c4e | 1259 | if (offset + bctx->data_offset + bctx->extent_len > i_size) |
31db9f7c AB |
1260 | return 0; |
1261 | ||
1262 | /* | |
1263 | * Make sure we don't consider clones from send_root that are | |
1264 | * behind the current inode/offset. | |
1265 | */ | |
1266 | if (found->root == bctx->sctx->send_root) { | |
1267 | /* | |
1268 | * TODO for the moment we don't accept clones from the inode | |
1269 | * that is currently send. We may change this when | |
1270 | * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same | |
1271 | * file. | |
1272 | */ | |
1273 | if (ino >= bctx->cur_objectid) | |
1274 | return 0; | |
31db9f7c AB |
1275 | } |
1276 | ||
1277 | bctx->found++; | |
1278 | found->found_refs++; | |
1279 | if (ino < found->ino) { | |
1280 | found->ino = ino; | |
1281 | found->offset = offset; | |
1282 | } else if (found->ino == ino) { | |
1283 | /* | |
1284 | * same extent found more then once in the same file. | |
1285 | */ | |
1286 | if (found->offset > offset + bctx->extent_len) | |
1287 | found->offset = offset; | |
1288 | } | |
1289 | ||
1290 | return 0; | |
1291 | } | |
1292 | ||
1293 | /* | |
766702ef AB |
1294 | * Given an inode, offset and extent item, it finds a good clone for a clone |
1295 | * instruction. Returns -ENOENT when none could be found. The function makes | |
1296 | * sure that the returned clone is usable at the point where sending is at the | |
1297 | * moment. This means, that no clones are accepted which lie behind the current | |
1298 | * inode+offset. | |
1299 | * | |
31db9f7c AB |
1300 | * path must point to the extent item when called. |
1301 | */ | |
1302 | static int find_extent_clone(struct send_ctx *sctx, | |
1303 | struct btrfs_path *path, | |
1304 | u64 ino, u64 data_offset, | |
1305 | u64 ino_size, | |
1306 | struct clone_root **found) | |
1307 | { | |
04ab956e | 1308 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
1309 | int ret; |
1310 | int extent_type; | |
1311 | u64 logical; | |
74dd17fb | 1312 | u64 disk_byte; |
31db9f7c AB |
1313 | u64 num_bytes; |
1314 | u64 extent_item_pos; | |
69917e43 | 1315 | u64 flags = 0; |
31db9f7c AB |
1316 | struct btrfs_file_extent_item *fi; |
1317 | struct extent_buffer *eb = path->nodes[0]; | |
35075bb0 | 1318 | struct backref_ctx *backref_ctx = NULL; |
31db9f7c AB |
1319 | struct clone_root *cur_clone_root; |
1320 | struct btrfs_key found_key; | |
1321 | struct btrfs_path *tmp_path; | |
74dd17fb | 1322 | int compressed; |
31db9f7c AB |
1323 | u32 i; |
1324 | ||
1325 | tmp_path = alloc_path_for_send(); | |
1326 | if (!tmp_path) | |
1327 | return -ENOMEM; | |
1328 | ||
3f8a18cc JB |
1329 | /* We only use this path under the commit sem */ |
1330 | tmp_path->need_commit_sem = 0; | |
1331 | ||
e780b0d1 | 1332 | backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL); |
35075bb0 AB |
1333 | if (!backref_ctx) { |
1334 | ret = -ENOMEM; | |
1335 | goto out; | |
1336 | } | |
1337 | ||
3f8a18cc JB |
1338 | backref_ctx->path = tmp_path; |
1339 | ||
31db9f7c AB |
1340 | if (data_offset >= ino_size) { |
1341 | /* | |
1342 | * There may be extents that lie behind the file's size. | |
1343 | * I at least had this in combination with snapshotting while | |
1344 | * writing large files. | |
1345 | */ | |
1346 | ret = 0; | |
1347 | goto out; | |
1348 | } | |
1349 | ||
1350 | fi = btrfs_item_ptr(eb, path->slots[0], | |
1351 | struct btrfs_file_extent_item); | |
1352 | extent_type = btrfs_file_extent_type(eb, fi); | |
1353 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1354 | ret = -ENOENT; | |
1355 | goto out; | |
1356 | } | |
74dd17fb | 1357 | compressed = btrfs_file_extent_compression(eb, fi); |
31db9f7c AB |
1358 | |
1359 | num_bytes = btrfs_file_extent_num_bytes(eb, fi); | |
74dd17fb CM |
1360 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
1361 | if (disk_byte == 0) { | |
31db9f7c AB |
1362 | ret = -ENOENT; |
1363 | goto out; | |
1364 | } | |
74dd17fb | 1365 | logical = disk_byte + btrfs_file_extent_offset(eb, fi); |
31db9f7c | 1366 | |
04ab956e JM |
1367 | down_read(&fs_info->commit_root_sem); |
1368 | ret = extent_from_logical(fs_info, disk_byte, tmp_path, | |
69917e43 | 1369 | &found_key, &flags); |
04ab956e | 1370 | up_read(&fs_info->commit_root_sem); |
31db9f7c AB |
1371 | btrfs_release_path(tmp_path); |
1372 | ||
1373 | if (ret < 0) | |
1374 | goto out; | |
69917e43 | 1375 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
31db9f7c AB |
1376 | ret = -EIO; |
1377 | goto out; | |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * Setup the clone roots. | |
1382 | */ | |
1383 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1384 | cur_clone_root = sctx->clone_roots + i; | |
1385 | cur_clone_root->ino = (u64)-1; | |
1386 | cur_clone_root->offset = 0; | |
1387 | cur_clone_root->found_refs = 0; | |
1388 | } | |
1389 | ||
35075bb0 AB |
1390 | backref_ctx->sctx = sctx; |
1391 | backref_ctx->found = 0; | |
1392 | backref_ctx->cur_objectid = ino; | |
1393 | backref_ctx->cur_offset = data_offset; | |
1394 | backref_ctx->found_itself = 0; | |
1395 | backref_ctx->extent_len = num_bytes; | |
619d8c4e FM |
1396 | /* |
1397 | * For non-compressed extents iterate_extent_inodes() gives us extent | |
1398 | * offsets that already take into account the data offset, but not for | |
1399 | * compressed extents, since the offset is logical and not relative to | |
1400 | * the physical extent locations. We must take this into account to | |
1401 | * avoid sending clone offsets that go beyond the source file's size, | |
1402 | * which would result in the clone ioctl failing with -EINVAL on the | |
1403 | * receiving end. | |
1404 | */ | |
1405 | if (compressed == BTRFS_COMPRESS_NONE) | |
1406 | backref_ctx->data_offset = 0; | |
1407 | else | |
1408 | backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi); | |
31db9f7c AB |
1409 | |
1410 | /* | |
1411 | * The last extent of a file may be too large due to page alignment. | |
1412 | * We need to adjust extent_len in this case so that the checks in | |
1413 | * __iterate_backrefs work. | |
1414 | */ | |
1415 | if (data_offset + num_bytes >= ino_size) | |
35075bb0 | 1416 | backref_ctx->extent_len = ino_size - data_offset; |
31db9f7c AB |
1417 | |
1418 | /* | |
1419 | * Now collect all backrefs. | |
1420 | */ | |
74dd17fb CM |
1421 | if (compressed == BTRFS_COMPRESS_NONE) |
1422 | extent_item_pos = logical - found_key.objectid; | |
1423 | else | |
1424 | extent_item_pos = 0; | |
0b246afa JM |
1425 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
1426 | extent_item_pos, 1, __iterate_backrefs, | |
c995ab3c | 1427 | backref_ctx, false); |
74dd17fb | 1428 | |
31db9f7c AB |
1429 | if (ret < 0) |
1430 | goto out; | |
1431 | ||
35075bb0 | 1432 | if (!backref_ctx->found_itself) { |
31db9f7c AB |
1433 | /* found a bug in backref code? */ |
1434 | ret = -EIO; | |
04ab956e | 1435 | btrfs_err(fs_info, |
5d163e0e | 1436 | "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu", |
04ab956e | 1437 | ino, data_offset, disk_byte, found_key.objectid); |
31db9f7c AB |
1438 | goto out; |
1439 | } | |
1440 | ||
04ab956e JM |
1441 | btrfs_debug(fs_info, |
1442 | "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu", | |
1443 | data_offset, ino, num_bytes, logical); | |
31db9f7c | 1444 | |
35075bb0 | 1445 | if (!backref_ctx->found) |
04ab956e | 1446 | btrfs_debug(fs_info, "no clones found"); |
31db9f7c AB |
1447 | |
1448 | cur_clone_root = NULL; | |
1449 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1450 | if (sctx->clone_roots[i].found_refs) { | |
1451 | if (!cur_clone_root) | |
1452 | cur_clone_root = sctx->clone_roots + i; | |
1453 | else if (sctx->clone_roots[i].root == sctx->send_root) | |
1454 | /* prefer clones from send_root over others */ | |
1455 | cur_clone_root = sctx->clone_roots + i; | |
31db9f7c AB |
1456 | } |
1457 | ||
1458 | } | |
1459 | ||
1460 | if (cur_clone_root) { | |
1461 | *found = cur_clone_root; | |
1462 | ret = 0; | |
1463 | } else { | |
1464 | ret = -ENOENT; | |
1465 | } | |
1466 | ||
1467 | out: | |
1468 | btrfs_free_path(tmp_path); | |
35075bb0 | 1469 | kfree(backref_ctx); |
31db9f7c AB |
1470 | return ret; |
1471 | } | |
1472 | ||
924794c9 | 1473 | static int read_symlink(struct btrfs_root *root, |
31db9f7c AB |
1474 | u64 ino, |
1475 | struct fs_path *dest) | |
1476 | { | |
1477 | int ret; | |
1478 | struct btrfs_path *path; | |
1479 | struct btrfs_key key; | |
1480 | struct btrfs_file_extent_item *ei; | |
1481 | u8 type; | |
1482 | u8 compression; | |
1483 | unsigned long off; | |
1484 | int len; | |
1485 | ||
1486 | path = alloc_path_for_send(); | |
1487 | if (!path) | |
1488 | return -ENOMEM; | |
1489 | ||
1490 | key.objectid = ino; | |
1491 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1492 | key.offset = 0; | |
1493 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1494 | if (ret < 0) | |
1495 | goto out; | |
a879719b FM |
1496 | if (ret) { |
1497 | /* | |
1498 | * An empty symlink inode. Can happen in rare error paths when | |
1499 | * creating a symlink (transaction committed before the inode | |
1500 | * eviction handler removed the symlink inode items and a crash | |
1501 | * happened in between or the subvol was snapshoted in between). | |
1502 | * Print an informative message to dmesg/syslog so that the user | |
1503 | * can delete the symlink. | |
1504 | */ | |
1505 | btrfs_err(root->fs_info, | |
1506 | "Found empty symlink inode %llu at root %llu", | |
1507 | ino, root->root_key.objectid); | |
1508 | ret = -EIO; | |
1509 | goto out; | |
1510 | } | |
31db9f7c AB |
1511 | |
1512 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1513 | struct btrfs_file_extent_item); | |
1514 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
1515 | compression = btrfs_file_extent_compression(path->nodes[0], ei); | |
1516 | BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); | |
1517 | BUG_ON(compression); | |
1518 | ||
1519 | off = btrfs_file_extent_inline_start(ei); | |
514ac8ad | 1520 | len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei); |
31db9f7c AB |
1521 | |
1522 | ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); | |
31db9f7c AB |
1523 | |
1524 | out: | |
1525 | btrfs_free_path(path); | |
1526 | return ret; | |
1527 | } | |
1528 | ||
1529 | /* | |
1530 | * Helper function to generate a file name that is unique in the root of | |
1531 | * send_root and parent_root. This is used to generate names for orphan inodes. | |
1532 | */ | |
1533 | static int gen_unique_name(struct send_ctx *sctx, | |
1534 | u64 ino, u64 gen, | |
1535 | struct fs_path *dest) | |
1536 | { | |
1537 | int ret = 0; | |
1538 | struct btrfs_path *path; | |
1539 | struct btrfs_dir_item *di; | |
1540 | char tmp[64]; | |
1541 | int len; | |
1542 | u64 idx = 0; | |
1543 | ||
1544 | path = alloc_path_for_send(); | |
1545 | if (!path) | |
1546 | return -ENOMEM; | |
1547 | ||
1548 | while (1) { | |
f74b86d8 | 1549 | len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", |
31db9f7c | 1550 | ino, gen, idx); |
64792f25 | 1551 | ASSERT(len < sizeof(tmp)); |
31db9f7c AB |
1552 | |
1553 | di = btrfs_lookup_dir_item(NULL, sctx->send_root, | |
1554 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1555 | tmp, strlen(tmp), 0); | |
1556 | btrfs_release_path(path); | |
1557 | if (IS_ERR(di)) { | |
1558 | ret = PTR_ERR(di); | |
1559 | goto out; | |
1560 | } | |
1561 | if (di) { | |
1562 | /* not unique, try again */ | |
1563 | idx++; | |
1564 | continue; | |
1565 | } | |
1566 | ||
1567 | if (!sctx->parent_root) { | |
1568 | /* unique */ | |
1569 | ret = 0; | |
1570 | break; | |
1571 | } | |
1572 | ||
1573 | di = btrfs_lookup_dir_item(NULL, sctx->parent_root, | |
1574 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1575 | tmp, strlen(tmp), 0); | |
1576 | btrfs_release_path(path); | |
1577 | if (IS_ERR(di)) { | |
1578 | ret = PTR_ERR(di); | |
1579 | goto out; | |
1580 | } | |
1581 | if (di) { | |
1582 | /* not unique, try again */ | |
1583 | idx++; | |
1584 | continue; | |
1585 | } | |
1586 | /* unique */ | |
1587 | break; | |
1588 | } | |
1589 | ||
1590 | ret = fs_path_add(dest, tmp, strlen(tmp)); | |
1591 | ||
1592 | out: | |
1593 | btrfs_free_path(path); | |
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | enum inode_state { | |
1598 | inode_state_no_change, | |
1599 | inode_state_will_create, | |
1600 | inode_state_did_create, | |
1601 | inode_state_will_delete, | |
1602 | inode_state_did_delete, | |
1603 | }; | |
1604 | ||
1605 | static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) | |
1606 | { | |
1607 | int ret; | |
1608 | int left_ret; | |
1609 | int right_ret; | |
1610 | u64 left_gen; | |
1611 | u64 right_gen; | |
1612 | ||
1613 | ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, | |
85a7b33b | 1614 | NULL, NULL); |
31db9f7c AB |
1615 | if (ret < 0 && ret != -ENOENT) |
1616 | goto out; | |
1617 | left_ret = ret; | |
1618 | ||
1619 | if (!sctx->parent_root) { | |
1620 | right_ret = -ENOENT; | |
1621 | } else { | |
1622 | ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, | |
85a7b33b | 1623 | NULL, NULL, NULL, NULL); |
31db9f7c AB |
1624 | if (ret < 0 && ret != -ENOENT) |
1625 | goto out; | |
1626 | right_ret = ret; | |
1627 | } | |
1628 | ||
1629 | if (!left_ret && !right_ret) { | |
e938c8ad | 1630 | if (left_gen == gen && right_gen == gen) { |
31db9f7c | 1631 | ret = inode_state_no_change; |
e938c8ad | 1632 | } else if (left_gen == gen) { |
31db9f7c AB |
1633 | if (ino < sctx->send_progress) |
1634 | ret = inode_state_did_create; | |
1635 | else | |
1636 | ret = inode_state_will_create; | |
1637 | } else if (right_gen == gen) { | |
1638 | if (ino < sctx->send_progress) | |
1639 | ret = inode_state_did_delete; | |
1640 | else | |
1641 | ret = inode_state_will_delete; | |
1642 | } else { | |
1643 | ret = -ENOENT; | |
1644 | } | |
1645 | } else if (!left_ret) { | |
1646 | if (left_gen == gen) { | |
1647 | if (ino < sctx->send_progress) | |
1648 | ret = inode_state_did_create; | |
1649 | else | |
1650 | ret = inode_state_will_create; | |
1651 | } else { | |
1652 | ret = -ENOENT; | |
1653 | } | |
1654 | } else if (!right_ret) { | |
1655 | if (right_gen == gen) { | |
1656 | if (ino < sctx->send_progress) | |
1657 | ret = inode_state_did_delete; | |
1658 | else | |
1659 | ret = inode_state_will_delete; | |
1660 | } else { | |
1661 | ret = -ENOENT; | |
1662 | } | |
1663 | } else { | |
1664 | ret = -ENOENT; | |
1665 | } | |
1666 | ||
1667 | out: | |
1668 | return ret; | |
1669 | } | |
1670 | ||
1671 | static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) | |
1672 | { | |
1673 | int ret; | |
1674 | ||
4dd9920d RK |
1675 | if (ino == BTRFS_FIRST_FREE_OBJECTID) |
1676 | return 1; | |
1677 | ||
31db9f7c AB |
1678 | ret = get_cur_inode_state(sctx, ino, gen); |
1679 | if (ret < 0) | |
1680 | goto out; | |
1681 | ||
1682 | if (ret == inode_state_no_change || | |
1683 | ret == inode_state_did_create || | |
1684 | ret == inode_state_will_delete) | |
1685 | ret = 1; | |
1686 | else | |
1687 | ret = 0; | |
1688 | ||
1689 | out: | |
1690 | return ret; | |
1691 | } | |
1692 | ||
1693 | /* | |
1694 | * Helper function to lookup a dir item in a dir. | |
1695 | */ | |
1696 | static int lookup_dir_item_inode(struct btrfs_root *root, | |
1697 | u64 dir, const char *name, int name_len, | |
1698 | u64 *found_inode, | |
1699 | u8 *found_type) | |
1700 | { | |
1701 | int ret = 0; | |
1702 | struct btrfs_dir_item *di; | |
1703 | struct btrfs_key key; | |
1704 | struct btrfs_path *path; | |
1705 | ||
1706 | path = alloc_path_for_send(); | |
1707 | if (!path) | |
1708 | return -ENOMEM; | |
1709 | ||
1710 | di = btrfs_lookup_dir_item(NULL, root, path, | |
1711 | dir, name, name_len, 0); | |
1712 | if (!di) { | |
1713 | ret = -ENOENT; | |
1714 | goto out; | |
1715 | } | |
1716 | if (IS_ERR(di)) { | |
1717 | ret = PTR_ERR(di); | |
1718 | goto out; | |
1719 | } | |
1720 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
1af56070 FM |
1721 | if (key.type == BTRFS_ROOT_ITEM_KEY) { |
1722 | ret = -ENOENT; | |
1723 | goto out; | |
1724 | } | |
31db9f7c AB |
1725 | *found_inode = key.objectid; |
1726 | *found_type = btrfs_dir_type(path->nodes[0], di); | |
1727 | ||
1728 | out: | |
1729 | btrfs_free_path(path); | |
1730 | return ret; | |
1731 | } | |
1732 | ||
766702ef AB |
1733 | /* |
1734 | * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, | |
1735 | * generation of the parent dir and the name of the dir entry. | |
1736 | */ | |
924794c9 | 1737 | static int get_first_ref(struct btrfs_root *root, u64 ino, |
31db9f7c AB |
1738 | u64 *dir, u64 *dir_gen, struct fs_path *name) |
1739 | { | |
1740 | int ret; | |
1741 | struct btrfs_key key; | |
1742 | struct btrfs_key found_key; | |
1743 | struct btrfs_path *path; | |
31db9f7c | 1744 | int len; |
96b5bd77 | 1745 | u64 parent_dir; |
31db9f7c AB |
1746 | |
1747 | path = alloc_path_for_send(); | |
1748 | if (!path) | |
1749 | return -ENOMEM; | |
1750 | ||
1751 | key.objectid = ino; | |
1752 | key.type = BTRFS_INODE_REF_KEY; | |
1753 | key.offset = 0; | |
1754 | ||
1755 | ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); | |
1756 | if (ret < 0) | |
1757 | goto out; | |
1758 | if (!ret) | |
1759 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1760 | path->slots[0]); | |
96b5bd77 JS |
1761 | if (ret || found_key.objectid != ino || |
1762 | (found_key.type != BTRFS_INODE_REF_KEY && | |
1763 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1764 | ret = -ENOENT; |
1765 | goto out; | |
1766 | } | |
1767 | ||
51a60253 | 1768 | if (found_key.type == BTRFS_INODE_REF_KEY) { |
96b5bd77 JS |
1769 | struct btrfs_inode_ref *iref; |
1770 | iref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1771 | struct btrfs_inode_ref); | |
1772 | len = btrfs_inode_ref_name_len(path->nodes[0], iref); | |
1773 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1774 | (unsigned long)(iref + 1), | |
1775 | len); | |
1776 | parent_dir = found_key.offset; | |
1777 | } else { | |
1778 | struct btrfs_inode_extref *extref; | |
1779 | extref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1780 | struct btrfs_inode_extref); | |
1781 | len = btrfs_inode_extref_name_len(path->nodes[0], extref); | |
1782 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1783 | (unsigned long)&extref->name, len); | |
1784 | parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); | |
1785 | } | |
31db9f7c AB |
1786 | if (ret < 0) |
1787 | goto out; | |
1788 | btrfs_release_path(path); | |
1789 | ||
b46ab97b FM |
1790 | if (dir_gen) { |
1791 | ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, | |
1792 | NULL, NULL, NULL); | |
1793 | if (ret < 0) | |
1794 | goto out; | |
1795 | } | |
31db9f7c | 1796 | |
96b5bd77 | 1797 | *dir = parent_dir; |
31db9f7c AB |
1798 | |
1799 | out: | |
1800 | btrfs_free_path(path); | |
1801 | return ret; | |
1802 | } | |
1803 | ||
924794c9 | 1804 | static int is_first_ref(struct btrfs_root *root, |
31db9f7c AB |
1805 | u64 ino, u64 dir, |
1806 | const char *name, int name_len) | |
1807 | { | |
1808 | int ret; | |
1809 | struct fs_path *tmp_name; | |
1810 | u64 tmp_dir; | |
31db9f7c | 1811 | |
924794c9 | 1812 | tmp_name = fs_path_alloc(); |
31db9f7c AB |
1813 | if (!tmp_name) |
1814 | return -ENOMEM; | |
1815 | ||
b46ab97b | 1816 | ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name); |
31db9f7c AB |
1817 | if (ret < 0) |
1818 | goto out; | |
1819 | ||
b9291aff | 1820 | if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { |
31db9f7c AB |
1821 | ret = 0; |
1822 | goto out; | |
1823 | } | |
1824 | ||
e938c8ad | 1825 | ret = !memcmp(tmp_name->start, name, name_len); |
31db9f7c AB |
1826 | |
1827 | out: | |
924794c9 | 1828 | fs_path_free(tmp_name); |
31db9f7c AB |
1829 | return ret; |
1830 | } | |
1831 | ||
766702ef AB |
1832 | /* |
1833 | * Used by process_recorded_refs to determine if a new ref would overwrite an | |
1834 | * already existing ref. In case it detects an overwrite, it returns the | |
1835 | * inode/gen in who_ino/who_gen. | |
1836 | * When an overwrite is detected, process_recorded_refs does proper orphanizing | |
1837 | * to make sure later references to the overwritten inode are possible. | |
1838 | * Orphanizing is however only required for the first ref of an inode. | |
1839 | * process_recorded_refs does an additional is_first_ref check to see if | |
1840 | * orphanizing is really required. | |
1841 | */ | |
31db9f7c AB |
1842 | static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
1843 | const char *name, int name_len, | |
f5962781 | 1844 | u64 *who_ino, u64 *who_gen, u64 *who_mode) |
31db9f7c AB |
1845 | { |
1846 | int ret = 0; | |
ebdad913 | 1847 | u64 gen; |
31db9f7c AB |
1848 | u64 other_inode = 0; |
1849 | u8 other_type = 0; | |
1850 | ||
1851 | if (!sctx->parent_root) | |
1852 | goto out; | |
1853 | ||
1854 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1855 | if (ret <= 0) | |
1856 | goto out; | |
1857 | ||
ebdad913 JB |
1858 | /* |
1859 | * If we have a parent root we need to verify that the parent dir was | |
01327610 | 1860 | * not deleted and then re-created, if it was then we have no overwrite |
ebdad913 JB |
1861 | * and we can just unlink this entry. |
1862 | */ | |
4dd9920d | 1863 | if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) { |
ebdad913 JB |
1864 | ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, |
1865 | NULL, NULL, NULL); | |
1866 | if (ret < 0 && ret != -ENOENT) | |
1867 | goto out; | |
1868 | if (ret) { | |
1869 | ret = 0; | |
1870 | goto out; | |
1871 | } | |
1872 | if (gen != dir_gen) | |
1873 | goto out; | |
1874 | } | |
1875 | ||
31db9f7c AB |
1876 | ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, |
1877 | &other_inode, &other_type); | |
1878 | if (ret < 0 && ret != -ENOENT) | |
1879 | goto out; | |
1880 | if (ret) { | |
1881 | ret = 0; | |
1882 | goto out; | |
1883 | } | |
1884 | ||
766702ef AB |
1885 | /* |
1886 | * Check if the overwritten ref was already processed. If yes, the ref | |
1887 | * was already unlinked/moved, so we can safely assume that we will not | |
1888 | * overwrite anything at this point in time. | |
1889 | */ | |
801bec36 RK |
1890 | if (other_inode > sctx->send_progress || |
1891 | is_waiting_for_move(sctx, other_inode)) { | |
31db9f7c | 1892 | ret = get_inode_info(sctx->parent_root, other_inode, NULL, |
f5962781 | 1893 | who_gen, who_mode, NULL, NULL, NULL); |
31db9f7c AB |
1894 | if (ret < 0) |
1895 | goto out; | |
1896 | ||
1897 | ret = 1; | |
1898 | *who_ino = other_inode; | |
1899 | } else { | |
1900 | ret = 0; | |
1901 | } | |
1902 | ||
1903 | out: | |
1904 | return ret; | |
1905 | } | |
1906 | ||
766702ef AB |
1907 | /* |
1908 | * Checks if the ref was overwritten by an already processed inode. This is | |
1909 | * used by __get_cur_name_and_parent to find out if the ref was orphanized and | |
1910 | * thus the orphan name needs be used. | |
1911 | * process_recorded_refs also uses it to avoid unlinking of refs that were | |
1912 | * overwritten. | |
1913 | */ | |
31db9f7c AB |
1914 | static int did_overwrite_ref(struct send_ctx *sctx, |
1915 | u64 dir, u64 dir_gen, | |
1916 | u64 ino, u64 ino_gen, | |
1917 | const char *name, int name_len) | |
1918 | { | |
1919 | int ret = 0; | |
1920 | u64 gen; | |
1921 | u64 ow_inode; | |
1922 | u8 other_type; | |
1923 | ||
1924 | if (!sctx->parent_root) | |
1925 | goto out; | |
1926 | ||
1927 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1928 | if (ret <= 0) | |
1929 | goto out; | |
1930 | ||
01914101 RK |
1931 | if (dir != BTRFS_FIRST_FREE_OBJECTID) { |
1932 | ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, | |
1933 | NULL, NULL, NULL); | |
1934 | if (ret < 0 && ret != -ENOENT) | |
1935 | goto out; | |
1936 | if (ret) { | |
1937 | ret = 0; | |
1938 | goto out; | |
1939 | } | |
1940 | if (gen != dir_gen) | |
1941 | goto out; | |
1942 | } | |
1943 | ||
31db9f7c AB |
1944 | /* check if the ref was overwritten by another ref */ |
1945 | ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, | |
1946 | &ow_inode, &other_type); | |
1947 | if (ret < 0 && ret != -ENOENT) | |
1948 | goto out; | |
1949 | if (ret) { | |
1950 | /* was never and will never be overwritten */ | |
1951 | ret = 0; | |
1952 | goto out; | |
1953 | } | |
1954 | ||
1955 | ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, | |
85a7b33b | 1956 | NULL, NULL); |
31db9f7c AB |
1957 | if (ret < 0) |
1958 | goto out; | |
1959 | ||
1960 | if (ow_inode == ino && gen == ino_gen) { | |
1961 | ret = 0; | |
1962 | goto out; | |
1963 | } | |
1964 | ||
8b191a68 FM |
1965 | /* |
1966 | * We know that it is or will be overwritten. Check this now. | |
1967 | * The current inode being processed might have been the one that caused | |
b786f16a FM |
1968 | * inode 'ino' to be orphanized, therefore check if ow_inode matches |
1969 | * the current inode being processed. | |
8b191a68 | 1970 | */ |
b786f16a FM |
1971 | if ((ow_inode < sctx->send_progress) || |
1972 | (ino != sctx->cur_ino && ow_inode == sctx->cur_ino && | |
1973 | gen == sctx->cur_inode_gen)) | |
31db9f7c AB |
1974 | ret = 1; |
1975 | else | |
1976 | ret = 0; | |
1977 | ||
1978 | out: | |
1979 | return ret; | |
1980 | } | |
1981 | ||
766702ef AB |
1982 | /* |
1983 | * Same as did_overwrite_ref, but also checks if it is the first ref of an inode | |
1984 | * that got overwritten. This is used by process_recorded_refs to determine | |
1985 | * if it has to use the path as returned by get_cur_path or the orphan name. | |
1986 | */ | |
31db9f7c AB |
1987 | static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) |
1988 | { | |
1989 | int ret = 0; | |
1990 | struct fs_path *name = NULL; | |
1991 | u64 dir; | |
1992 | u64 dir_gen; | |
1993 | ||
1994 | if (!sctx->parent_root) | |
1995 | goto out; | |
1996 | ||
924794c9 | 1997 | name = fs_path_alloc(); |
31db9f7c AB |
1998 | if (!name) |
1999 | return -ENOMEM; | |
2000 | ||
924794c9 | 2001 | ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); |
31db9f7c AB |
2002 | if (ret < 0) |
2003 | goto out; | |
2004 | ||
2005 | ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, | |
2006 | name->start, fs_path_len(name)); | |
31db9f7c AB |
2007 | |
2008 | out: | |
924794c9 | 2009 | fs_path_free(name); |
31db9f7c AB |
2010 | return ret; |
2011 | } | |
2012 | ||
766702ef AB |
2013 | /* |
2014 | * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, | |
2015 | * so we need to do some special handling in case we have clashes. This function | |
2016 | * takes care of this with the help of name_cache_entry::radix_list. | |
5dc67d0b | 2017 | * In case of error, nce is kfreed. |
766702ef | 2018 | */ |
31db9f7c AB |
2019 | static int name_cache_insert(struct send_ctx *sctx, |
2020 | struct name_cache_entry *nce) | |
2021 | { | |
2022 | int ret = 0; | |
7e0926fe AB |
2023 | struct list_head *nce_head; |
2024 | ||
2025 | nce_head = radix_tree_lookup(&sctx->name_cache, | |
2026 | (unsigned long)nce->ino); | |
2027 | if (!nce_head) { | |
e780b0d1 | 2028 | nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL); |
cfa7a9cc TI |
2029 | if (!nce_head) { |
2030 | kfree(nce); | |
31db9f7c | 2031 | return -ENOMEM; |
cfa7a9cc | 2032 | } |
7e0926fe | 2033 | INIT_LIST_HEAD(nce_head); |
31db9f7c | 2034 | |
7e0926fe | 2035 | ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); |
5dc67d0b AB |
2036 | if (ret < 0) { |
2037 | kfree(nce_head); | |
2038 | kfree(nce); | |
31db9f7c | 2039 | return ret; |
5dc67d0b | 2040 | } |
31db9f7c | 2041 | } |
7e0926fe | 2042 | list_add_tail(&nce->radix_list, nce_head); |
31db9f7c AB |
2043 | list_add_tail(&nce->list, &sctx->name_cache_list); |
2044 | sctx->name_cache_size++; | |
2045 | ||
2046 | return ret; | |
2047 | } | |
2048 | ||
2049 | static void name_cache_delete(struct send_ctx *sctx, | |
2050 | struct name_cache_entry *nce) | |
2051 | { | |
7e0926fe | 2052 | struct list_head *nce_head; |
31db9f7c | 2053 | |
7e0926fe AB |
2054 | nce_head = radix_tree_lookup(&sctx->name_cache, |
2055 | (unsigned long)nce->ino); | |
57fb8910 DS |
2056 | if (!nce_head) { |
2057 | btrfs_err(sctx->send_root->fs_info, | |
2058 | "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", | |
2059 | nce->ino, sctx->name_cache_size); | |
2060 | } | |
31db9f7c | 2061 | |
7e0926fe | 2062 | list_del(&nce->radix_list); |
31db9f7c | 2063 | list_del(&nce->list); |
31db9f7c | 2064 | sctx->name_cache_size--; |
7e0926fe | 2065 | |
57fb8910 DS |
2066 | /* |
2067 | * We may not get to the final release of nce_head if the lookup fails | |
2068 | */ | |
2069 | if (nce_head && list_empty(nce_head)) { | |
7e0926fe AB |
2070 | radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); |
2071 | kfree(nce_head); | |
2072 | } | |
31db9f7c AB |
2073 | } |
2074 | ||
2075 | static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, | |
2076 | u64 ino, u64 gen) | |
2077 | { | |
7e0926fe AB |
2078 | struct list_head *nce_head; |
2079 | struct name_cache_entry *cur; | |
31db9f7c | 2080 | |
7e0926fe AB |
2081 | nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); |
2082 | if (!nce_head) | |
31db9f7c AB |
2083 | return NULL; |
2084 | ||
7e0926fe AB |
2085 | list_for_each_entry(cur, nce_head, radix_list) { |
2086 | if (cur->ino == ino && cur->gen == gen) | |
2087 | return cur; | |
2088 | } | |
31db9f7c AB |
2089 | return NULL; |
2090 | } | |
2091 | ||
766702ef AB |
2092 | /* |
2093 | * Removes the entry from the list and adds it back to the end. This marks the | |
2094 | * entry as recently used so that name_cache_clean_unused does not remove it. | |
2095 | */ | |
31db9f7c AB |
2096 | static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) |
2097 | { | |
2098 | list_del(&nce->list); | |
2099 | list_add_tail(&nce->list, &sctx->name_cache_list); | |
2100 | } | |
2101 | ||
766702ef AB |
2102 | /* |
2103 | * Remove some entries from the beginning of name_cache_list. | |
2104 | */ | |
31db9f7c AB |
2105 | static void name_cache_clean_unused(struct send_ctx *sctx) |
2106 | { | |
2107 | struct name_cache_entry *nce; | |
2108 | ||
2109 | if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) | |
2110 | return; | |
2111 | ||
2112 | while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { | |
2113 | nce = list_entry(sctx->name_cache_list.next, | |
2114 | struct name_cache_entry, list); | |
2115 | name_cache_delete(sctx, nce); | |
2116 | kfree(nce); | |
2117 | } | |
2118 | } | |
2119 | ||
2120 | static void name_cache_free(struct send_ctx *sctx) | |
2121 | { | |
2122 | struct name_cache_entry *nce; | |
31db9f7c | 2123 | |
e938c8ad AB |
2124 | while (!list_empty(&sctx->name_cache_list)) { |
2125 | nce = list_entry(sctx->name_cache_list.next, | |
2126 | struct name_cache_entry, list); | |
31db9f7c | 2127 | name_cache_delete(sctx, nce); |
17589bd9 | 2128 | kfree(nce); |
31db9f7c AB |
2129 | } |
2130 | } | |
2131 | ||
766702ef AB |
2132 | /* |
2133 | * Used by get_cur_path for each ref up to the root. | |
2134 | * Returns 0 if it succeeded. | |
2135 | * Returns 1 if the inode is not existent or got overwritten. In that case, the | |
2136 | * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 | |
2137 | * is returned, parent_ino/parent_gen are not guaranteed to be valid. | |
2138 | * Returns <0 in case of error. | |
2139 | */ | |
31db9f7c AB |
2140 | static int __get_cur_name_and_parent(struct send_ctx *sctx, |
2141 | u64 ino, u64 gen, | |
2142 | u64 *parent_ino, | |
2143 | u64 *parent_gen, | |
2144 | struct fs_path *dest) | |
2145 | { | |
2146 | int ret; | |
2147 | int nce_ret; | |
31db9f7c AB |
2148 | struct name_cache_entry *nce = NULL; |
2149 | ||
766702ef AB |
2150 | /* |
2151 | * First check if we already did a call to this function with the same | |
2152 | * ino/gen. If yes, check if the cache entry is still up-to-date. If yes | |
2153 | * return the cached result. | |
2154 | */ | |
31db9f7c AB |
2155 | nce = name_cache_search(sctx, ino, gen); |
2156 | if (nce) { | |
2157 | if (ino < sctx->send_progress && nce->need_later_update) { | |
2158 | name_cache_delete(sctx, nce); | |
2159 | kfree(nce); | |
2160 | nce = NULL; | |
2161 | } else { | |
2162 | name_cache_used(sctx, nce); | |
2163 | *parent_ino = nce->parent_ino; | |
2164 | *parent_gen = nce->parent_gen; | |
2165 | ret = fs_path_add(dest, nce->name, nce->name_len); | |
2166 | if (ret < 0) | |
2167 | goto out; | |
2168 | ret = nce->ret; | |
2169 | goto out; | |
2170 | } | |
2171 | } | |
2172 | ||
766702ef AB |
2173 | /* |
2174 | * If the inode is not existent yet, add the orphan name and return 1. | |
2175 | * This should only happen for the parent dir that we determine in | |
2176 | * __record_new_ref | |
2177 | */ | |
31db9f7c AB |
2178 | ret = is_inode_existent(sctx, ino, gen); |
2179 | if (ret < 0) | |
2180 | goto out; | |
2181 | ||
2182 | if (!ret) { | |
2183 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2184 | if (ret < 0) | |
2185 | goto out; | |
2186 | ret = 1; | |
2187 | goto out_cache; | |
2188 | } | |
2189 | ||
766702ef AB |
2190 | /* |
2191 | * Depending on whether the inode was already processed or not, use | |
2192 | * send_root or parent_root for ref lookup. | |
2193 | */ | |
bf0d1f44 | 2194 | if (ino < sctx->send_progress) |
924794c9 TI |
2195 | ret = get_first_ref(sctx->send_root, ino, |
2196 | parent_ino, parent_gen, dest); | |
31db9f7c | 2197 | else |
924794c9 TI |
2198 | ret = get_first_ref(sctx->parent_root, ino, |
2199 | parent_ino, parent_gen, dest); | |
31db9f7c AB |
2200 | if (ret < 0) |
2201 | goto out; | |
2202 | ||
766702ef AB |
2203 | /* |
2204 | * Check if the ref was overwritten by an inode's ref that was processed | |
2205 | * earlier. If yes, treat as orphan and return 1. | |
2206 | */ | |
31db9f7c AB |
2207 | ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, |
2208 | dest->start, dest->end - dest->start); | |
2209 | if (ret < 0) | |
2210 | goto out; | |
2211 | if (ret) { | |
2212 | fs_path_reset(dest); | |
2213 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2214 | if (ret < 0) | |
2215 | goto out; | |
2216 | ret = 1; | |
2217 | } | |
2218 | ||
2219 | out_cache: | |
766702ef AB |
2220 | /* |
2221 | * Store the result of the lookup in the name cache. | |
2222 | */ | |
e780b0d1 | 2223 | nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); |
31db9f7c AB |
2224 | if (!nce) { |
2225 | ret = -ENOMEM; | |
2226 | goto out; | |
2227 | } | |
2228 | ||
2229 | nce->ino = ino; | |
2230 | nce->gen = gen; | |
2231 | nce->parent_ino = *parent_ino; | |
2232 | nce->parent_gen = *parent_gen; | |
2233 | nce->name_len = fs_path_len(dest); | |
2234 | nce->ret = ret; | |
2235 | strcpy(nce->name, dest->start); | |
31db9f7c AB |
2236 | |
2237 | if (ino < sctx->send_progress) | |
2238 | nce->need_later_update = 0; | |
2239 | else | |
2240 | nce->need_later_update = 1; | |
2241 | ||
2242 | nce_ret = name_cache_insert(sctx, nce); | |
2243 | if (nce_ret < 0) | |
2244 | ret = nce_ret; | |
2245 | name_cache_clean_unused(sctx); | |
2246 | ||
2247 | out: | |
31db9f7c AB |
2248 | return ret; |
2249 | } | |
2250 | ||
2251 | /* | |
2252 | * Magic happens here. This function returns the first ref to an inode as it | |
2253 | * would look like while receiving the stream at this point in time. | |
2254 | * We walk the path up to the root. For every inode in between, we check if it | |
2255 | * was already processed/sent. If yes, we continue with the parent as found | |
2256 | * in send_root. If not, we continue with the parent as found in parent_root. | |
2257 | * If we encounter an inode that was deleted at this point in time, we use the | |
2258 | * inodes "orphan" name instead of the real name and stop. Same with new inodes | |
2259 | * that were not created yet and overwritten inodes/refs. | |
2260 | * | |
2261 | * When do we have have orphan inodes: | |
2262 | * 1. When an inode is freshly created and thus no valid refs are available yet | |
2263 | * 2. When a directory lost all it's refs (deleted) but still has dir items | |
2264 | * inside which were not processed yet (pending for move/delete). If anyone | |
2265 | * tried to get the path to the dir items, it would get a path inside that | |
2266 | * orphan directory. | |
2267 | * 3. When an inode is moved around or gets new links, it may overwrite the ref | |
2268 | * of an unprocessed inode. If in that case the first ref would be | |
2269 | * overwritten, the overwritten inode gets "orphanized". Later when we | |
2270 | * process this overwritten inode, it is restored at a new place by moving | |
2271 | * the orphan inode. | |
2272 | * | |
2273 | * sctx->send_progress tells this function at which point in time receiving | |
2274 | * would be. | |
2275 | */ | |
2276 | static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, | |
2277 | struct fs_path *dest) | |
2278 | { | |
2279 | int ret = 0; | |
2280 | struct fs_path *name = NULL; | |
2281 | u64 parent_inode = 0; | |
2282 | u64 parent_gen = 0; | |
2283 | int stop = 0; | |
2284 | ||
924794c9 | 2285 | name = fs_path_alloc(); |
31db9f7c AB |
2286 | if (!name) { |
2287 | ret = -ENOMEM; | |
2288 | goto out; | |
2289 | } | |
2290 | ||
2291 | dest->reversed = 1; | |
2292 | fs_path_reset(dest); | |
2293 | ||
2294 | while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { | |
8b191a68 FM |
2295 | struct waiting_dir_move *wdm; |
2296 | ||
31db9f7c AB |
2297 | fs_path_reset(name); |
2298 | ||
9dc44214 FM |
2299 | if (is_waiting_for_rm(sctx, ino)) { |
2300 | ret = gen_unique_name(sctx, ino, gen, name); | |
2301 | if (ret < 0) | |
2302 | goto out; | |
2303 | ret = fs_path_add_path(dest, name); | |
2304 | break; | |
2305 | } | |
2306 | ||
8b191a68 FM |
2307 | wdm = get_waiting_dir_move(sctx, ino); |
2308 | if (wdm && wdm->orphanized) { | |
2309 | ret = gen_unique_name(sctx, ino, gen, name); | |
2310 | stop = 1; | |
2311 | } else if (wdm) { | |
bf0d1f44 FM |
2312 | ret = get_first_ref(sctx->parent_root, ino, |
2313 | &parent_inode, &parent_gen, name); | |
2314 | } else { | |
2315 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
2316 | &parent_inode, | |
2317 | &parent_gen, name); | |
2318 | if (ret) | |
2319 | stop = 1; | |
2320 | } | |
2321 | ||
31db9f7c AB |
2322 | if (ret < 0) |
2323 | goto out; | |
9f03740a | 2324 | |
31db9f7c AB |
2325 | ret = fs_path_add_path(dest, name); |
2326 | if (ret < 0) | |
2327 | goto out; | |
2328 | ||
2329 | ino = parent_inode; | |
2330 | gen = parent_gen; | |
2331 | } | |
2332 | ||
2333 | out: | |
924794c9 | 2334 | fs_path_free(name); |
31db9f7c AB |
2335 | if (!ret) |
2336 | fs_path_unreverse(dest); | |
2337 | return ret; | |
2338 | } | |
2339 | ||
31db9f7c AB |
2340 | /* |
2341 | * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace | |
2342 | */ | |
2343 | static int send_subvol_begin(struct send_ctx *sctx) | |
2344 | { | |
2345 | int ret; | |
2346 | struct btrfs_root *send_root = sctx->send_root; | |
2347 | struct btrfs_root *parent_root = sctx->parent_root; | |
2348 | struct btrfs_path *path; | |
2349 | struct btrfs_key key; | |
2350 | struct btrfs_root_ref *ref; | |
2351 | struct extent_buffer *leaf; | |
2352 | char *name = NULL; | |
2353 | int namelen; | |
2354 | ||
ffcfaf81 | 2355 | path = btrfs_alloc_path(); |
31db9f7c AB |
2356 | if (!path) |
2357 | return -ENOMEM; | |
2358 | ||
e780b0d1 | 2359 | name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); |
31db9f7c AB |
2360 | if (!name) { |
2361 | btrfs_free_path(path); | |
2362 | return -ENOMEM; | |
2363 | } | |
2364 | ||
2365 | key.objectid = send_root->objectid; | |
2366 | key.type = BTRFS_ROOT_BACKREF_KEY; | |
2367 | key.offset = 0; | |
2368 | ||
2369 | ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, | |
2370 | &key, path, 1, 0); | |
2371 | if (ret < 0) | |
2372 | goto out; | |
2373 | if (ret) { | |
2374 | ret = -ENOENT; | |
2375 | goto out; | |
2376 | } | |
2377 | ||
2378 | leaf = path->nodes[0]; | |
2379 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2380 | if (key.type != BTRFS_ROOT_BACKREF_KEY || | |
2381 | key.objectid != send_root->objectid) { | |
2382 | ret = -ENOENT; | |
2383 | goto out; | |
2384 | } | |
2385 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
2386 | namelen = btrfs_root_ref_name_len(leaf, ref); | |
2387 | read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); | |
2388 | btrfs_release_path(path); | |
2389 | ||
31db9f7c AB |
2390 | if (parent_root) { |
2391 | ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); | |
2392 | if (ret < 0) | |
2393 | goto out; | |
2394 | } else { | |
2395 | ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); | |
2396 | if (ret < 0) | |
2397 | goto out; | |
2398 | } | |
2399 | ||
2400 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); | |
b96b1db0 RR |
2401 | |
2402 | if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid)) | |
2403 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2404 | sctx->send_root->root_item.received_uuid); | |
2405 | else | |
2406 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2407 | sctx->send_root->root_item.uuid); | |
2408 | ||
31db9f7c | 2409 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, |
5a0f4e2c | 2410 | le64_to_cpu(sctx->send_root->root_item.ctransid)); |
31db9f7c | 2411 | if (parent_root) { |
37b8d27d JB |
2412 | if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) |
2413 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2414 | parent_root->root_item.received_uuid); | |
2415 | else | |
2416 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2417 | parent_root->root_item.uuid); | |
31db9f7c | 2418 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 2419 | le64_to_cpu(sctx->parent_root->root_item.ctransid)); |
31db9f7c AB |
2420 | } |
2421 | ||
2422 | ret = send_cmd(sctx); | |
2423 | ||
2424 | tlv_put_failure: | |
2425 | out: | |
2426 | btrfs_free_path(path); | |
2427 | kfree(name); | |
2428 | return ret; | |
2429 | } | |
2430 | ||
2431 | static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) | |
2432 | { | |
04ab956e | 2433 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2434 | int ret = 0; |
2435 | struct fs_path *p; | |
2436 | ||
04ab956e | 2437 | btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size); |
31db9f7c | 2438 | |
924794c9 | 2439 | p = fs_path_alloc(); |
31db9f7c AB |
2440 | if (!p) |
2441 | return -ENOMEM; | |
2442 | ||
2443 | ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); | |
2444 | if (ret < 0) | |
2445 | goto out; | |
2446 | ||
2447 | ret = get_cur_path(sctx, ino, gen, p); | |
2448 | if (ret < 0) | |
2449 | goto out; | |
2450 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2451 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); | |
2452 | ||
2453 | ret = send_cmd(sctx); | |
2454 | ||
2455 | tlv_put_failure: | |
2456 | out: | |
924794c9 | 2457 | fs_path_free(p); |
31db9f7c AB |
2458 | return ret; |
2459 | } | |
2460 | ||
2461 | static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) | |
2462 | { | |
04ab956e | 2463 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2464 | int ret = 0; |
2465 | struct fs_path *p; | |
2466 | ||
04ab956e | 2467 | btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode); |
31db9f7c | 2468 | |
924794c9 | 2469 | p = fs_path_alloc(); |
31db9f7c AB |
2470 | if (!p) |
2471 | return -ENOMEM; | |
2472 | ||
2473 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); | |
2474 | if (ret < 0) | |
2475 | goto out; | |
2476 | ||
2477 | ret = get_cur_path(sctx, ino, gen, p); | |
2478 | if (ret < 0) | |
2479 | goto out; | |
2480 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2481 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); | |
2482 | ||
2483 | ret = send_cmd(sctx); | |
2484 | ||
2485 | tlv_put_failure: | |
2486 | out: | |
924794c9 | 2487 | fs_path_free(p); |
31db9f7c AB |
2488 | return ret; |
2489 | } | |
2490 | ||
2491 | static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) | |
2492 | { | |
04ab956e | 2493 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2494 | int ret = 0; |
2495 | struct fs_path *p; | |
2496 | ||
04ab956e JM |
2497 | btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu", |
2498 | ino, uid, gid); | |
31db9f7c | 2499 | |
924794c9 | 2500 | p = fs_path_alloc(); |
31db9f7c AB |
2501 | if (!p) |
2502 | return -ENOMEM; | |
2503 | ||
2504 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); | |
2505 | if (ret < 0) | |
2506 | goto out; | |
2507 | ||
2508 | ret = get_cur_path(sctx, ino, gen, p); | |
2509 | if (ret < 0) | |
2510 | goto out; | |
2511 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2512 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); | |
2513 | TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); | |
2514 | ||
2515 | ret = send_cmd(sctx); | |
2516 | ||
2517 | tlv_put_failure: | |
2518 | out: | |
924794c9 | 2519 | fs_path_free(p); |
31db9f7c AB |
2520 | return ret; |
2521 | } | |
2522 | ||
2523 | static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) | |
2524 | { | |
04ab956e | 2525 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2526 | int ret = 0; |
2527 | struct fs_path *p = NULL; | |
2528 | struct btrfs_inode_item *ii; | |
2529 | struct btrfs_path *path = NULL; | |
2530 | struct extent_buffer *eb; | |
2531 | struct btrfs_key key; | |
2532 | int slot; | |
2533 | ||
04ab956e | 2534 | btrfs_debug(fs_info, "send_utimes %llu", ino); |
31db9f7c | 2535 | |
924794c9 | 2536 | p = fs_path_alloc(); |
31db9f7c AB |
2537 | if (!p) |
2538 | return -ENOMEM; | |
2539 | ||
2540 | path = alloc_path_for_send(); | |
2541 | if (!path) { | |
2542 | ret = -ENOMEM; | |
2543 | goto out; | |
2544 | } | |
2545 | ||
2546 | key.objectid = ino; | |
2547 | key.type = BTRFS_INODE_ITEM_KEY; | |
2548 | key.offset = 0; | |
2549 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); | |
15b253ea FM |
2550 | if (ret > 0) |
2551 | ret = -ENOENT; | |
31db9f7c AB |
2552 | if (ret < 0) |
2553 | goto out; | |
2554 | ||
2555 | eb = path->nodes[0]; | |
2556 | slot = path->slots[0]; | |
2557 | ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
2558 | ||
2559 | ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); | |
2560 | if (ret < 0) | |
2561 | goto out; | |
2562 | ||
2563 | ret = get_cur_path(sctx, ino, gen, p); | |
2564 | if (ret < 0) | |
2565 | goto out; | |
2566 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
a937b979 DS |
2567 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); |
2568 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); | |
2569 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); | |
766702ef | 2570 | /* TODO Add otime support when the otime patches get into upstream */ |
31db9f7c AB |
2571 | |
2572 | ret = send_cmd(sctx); | |
2573 | ||
2574 | tlv_put_failure: | |
2575 | out: | |
924794c9 | 2576 | fs_path_free(p); |
31db9f7c AB |
2577 | btrfs_free_path(path); |
2578 | return ret; | |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have | |
2583 | * a valid path yet because we did not process the refs yet. So, the inode | |
2584 | * is created as orphan. | |
2585 | */ | |
1f4692da | 2586 | static int send_create_inode(struct send_ctx *sctx, u64 ino) |
31db9f7c | 2587 | { |
04ab956e | 2588 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c | 2589 | int ret = 0; |
31db9f7c | 2590 | struct fs_path *p; |
31db9f7c | 2591 | int cmd; |
1f4692da | 2592 | u64 gen; |
31db9f7c | 2593 | u64 mode; |
1f4692da | 2594 | u64 rdev; |
31db9f7c | 2595 | |
04ab956e | 2596 | btrfs_debug(fs_info, "send_create_inode %llu", ino); |
31db9f7c | 2597 | |
924794c9 | 2598 | p = fs_path_alloc(); |
31db9f7c AB |
2599 | if (!p) |
2600 | return -ENOMEM; | |
2601 | ||
644d1940 LB |
2602 | if (ino != sctx->cur_ino) { |
2603 | ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, | |
2604 | NULL, NULL, &rdev); | |
2605 | if (ret < 0) | |
2606 | goto out; | |
2607 | } else { | |
2608 | gen = sctx->cur_inode_gen; | |
2609 | mode = sctx->cur_inode_mode; | |
2610 | rdev = sctx->cur_inode_rdev; | |
2611 | } | |
31db9f7c | 2612 | |
e938c8ad | 2613 | if (S_ISREG(mode)) { |
31db9f7c | 2614 | cmd = BTRFS_SEND_C_MKFILE; |
e938c8ad | 2615 | } else if (S_ISDIR(mode)) { |
31db9f7c | 2616 | cmd = BTRFS_SEND_C_MKDIR; |
e938c8ad | 2617 | } else if (S_ISLNK(mode)) { |
31db9f7c | 2618 | cmd = BTRFS_SEND_C_SYMLINK; |
e938c8ad | 2619 | } else if (S_ISCHR(mode) || S_ISBLK(mode)) { |
31db9f7c | 2620 | cmd = BTRFS_SEND_C_MKNOD; |
e938c8ad | 2621 | } else if (S_ISFIFO(mode)) { |
31db9f7c | 2622 | cmd = BTRFS_SEND_C_MKFIFO; |
e938c8ad | 2623 | } else if (S_ISSOCK(mode)) { |
31db9f7c | 2624 | cmd = BTRFS_SEND_C_MKSOCK; |
e938c8ad | 2625 | } else { |
f14d104d | 2626 | btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o", |
31db9f7c | 2627 | (int)(mode & S_IFMT)); |
ca6842bf | 2628 | ret = -EOPNOTSUPP; |
31db9f7c AB |
2629 | goto out; |
2630 | } | |
2631 | ||
2632 | ret = begin_cmd(sctx, cmd); | |
2633 | if (ret < 0) | |
2634 | goto out; | |
2635 | ||
1f4692da | 2636 | ret = gen_unique_name(sctx, ino, gen, p); |
31db9f7c AB |
2637 | if (ret < 0) |
2638 | goto out; | |
2639 | ||
2640 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
1f4692da | 2641 | TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); |
31db9f7c AB |
2642 | |
2643 | if (S_ISLNK(mode)) { | |
2644 | fs_path_reset(p); | |
924794c9 | 2645 | ret = read_symlink(sctx->send_root, ino, p); |
31db9f7c AB |
2646 | if (ret < 0) |
2647 | goto out; | |
2648 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); | |
2649 | } else if (S_ISCHR(mode) || S_ISBLK(mode) || | |
2650 | S_ISFIFO(mode) || S_ISSOCK(mode)) { | |
d79e5043 AJ |
2651 | TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); |
2652 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); | |
31db9f7c AB |
2653 | } |
2654 | ||
2655 | ret = send_cmd(sctx); | |
2656 | if (ret < 0) | |
2657 | goto out; | |
2658 | ||
2659 | ||
2660 | tlv_put_failure: | |
2661 | out: | |
924794c9 | 2662 | fs_path_free(p); |
31db9f7c AB |
2663 | return ret; |
2664 | } | |
2665 | ||
1f4692da AB |
2666 | /* |
2667 | * We need some special handling for inodes that get processed before the parent | |
2668 | * directory got created. See process_recorded_refs for details. | |
2669 | * This function does the check if we already created the dir out of order. | |
2670 | */ | |
2671 | static int did_create_dir(struct send_ctx *sctx, u64 dir) | |
2672 | { | |
2673 | int ret = 0; | |
2674 | struct btrfs_path *path = NULL; | |
2675 | struct btrfs_key key; | |
2676 | struct btrfs_key found_key; | |
2677 | struct btrfs_key di_key; | |
2678 | struct extent_buffer *eb; | |
2679 | struct btrfs_dir_item *di; | |
2680 | int slot; | |
2681 | ||
2682 | path = alloc_path_for_send(); | |
2683 | if (!path) { | |
2684 | ret = -ENOMEM; | |
2685 | goto out; | |
2686 | } | |
2687 | ||
2688 | key.objectid = dir; | |
2689 | key.type = BTRFS_DIR_INDEX_KEY; | |
2690 | key.offset = 0; | |
dff6d0ad FDBM |
2691 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); |
2692 | if (ret < 0) | |
2693 | goto out; | |
2694 | ||
1f4692da | 2695 | while (1) { |
dff6d0ad FDBM |
2696 | eb = path->nodes[0]; |
2697 | slot = path->slots[0]; | |
2698 | if (slot >= btrfs_header_nritems(eb)) { | |
2699 | ret = btrfs_next_leaf(sctx->send_root, path); | |
2700 | if (ret < 0) { | |
2701 | goto out; | |
2702 | } else if (ret > 0) { | |
2703 | ret = 0; | |
2704 | break; | |
2705 | } | |
2706 | continue; | |
1f4692da | 2707 | } |
dff6d0ad FDBM |
2708 | |
2709 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
2710 | if (found_key.objectid != key.objectid || | |
1f4692da AB |
2711 | found_key.type != key.type) { |
2712 | ret = 0; | |
2713 | goto out; | |
2714 | } | |
2715 | ||
2716 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); | |
2717 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2718 | ||
a0525414 JB |
2719 | if (di_key.type != BTRFS_ROOT_ITEM_KEY && |
2720 | di_key.objectid < sctx->send_progress) { | |
1f4692da AB |
2721 | ret = 1; |
2722 | goto out; | |
2723 | } | |
2724 | ||
dff6d0ad | 2725 | path->slots[0]++; |
1f4692da AB |
2726 | } |
2727 | ||
2728 | out: | |
2729 | btrfs_free_path(path); | |
2730 | return ret; | |
2731 | } | |
2732 | ||
2733 | /* | |
2734 | * Only creates the inode if it is: | |
2735 | * 1. Not a directory | |
2736 | * 2. Or a directory which was not created already due to out of order | |
2737 | * directories. See did_create_dir and process_recorded_refs for details. | |
2738 | */ | |
2739 | static int send_create_inode_if_needed(struct send_ctx *sctx) | |
2740 | { | |
2741 | int ret; | |
2742 | ||
2743 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
2744 | ret = did_create_dir(sctx, sctx->cur_ino); | |
2745 | if (ret < 0) | |
2746 | goto out; | |
2747 | if (ret) { | |
2748 | ret = 0; | |
2749 | goto out; | |
2750 | } | |
2751 | } | |
2752 | ||
2753 | ret = send_create_inode(sctx, sctx->cur_ino); | |
2754 | if (ret < 0) | |
2755 | goto out; | |
2756 | ||
2757 | out: | |
2758 | return ret; | |
2759 | } | |
2760 | ||
31db9f7c AB |
2761 | struct recorded_ref { |
2762 | struct list_head list; | |
31db9f7c AB |
2763 | char *name; |
2764 | struct fs_path *full_path; | |
2765 | u64 dir; | |
2766 | u64 dir_gen; | |
31db9f7c AB |
2767 | int name_len; |
2768 | }; | |
2769 | ||
fdb13889 FM |
2770 | static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) |
2771 | { | |
2772 | ref->full_path = path; | |
2773 | ref->name = (char *)kbasename(ref->full_path->start); | |
2774 | ref->name_len = ref->full_path->end - ref->name; | |
2775 | } | |
2776 | ||
31db9f7c AB |
2777 | /* |
2778 | * We need to process new refs before deleted refs, but compare_tree gives us | |
2779 | * everything mixed. So we first record all refs and later process them. | |
2780 | * This function is a helper to record one ref. | |
2781 | */ | |
a4d96d62 | 2782 | static int __record_ref(struct list_head *head, u64 dir, |
31db9f7c AB |
2783 | u64 dir_gen, struct fs_path *path) |
2784 | { | |
2785 | struct recorded_ref *ref; | |
31db9f7c | 2786 | |
e780b0d1 | 2787 | ref = kmalloc(sizeof(*ref), GFP_KERNEL); |
31db9f7c AB |
2788 | if (!ref) |
2789 | return -ENOMEM; | |
2790 | ||
2791 | ref->dir = dir; | |
2792 | ref->dir_gen = dir_gen; | |
fdb13889 | 2793 | set_ref_path(ref, path); |
31db9f7c AB |
2794 | list_add_tail(&ref->list, head); |
2795 | return 0; | |
2796 | } | |
2797 | ||
ba5e8f2e JB |
2798 | static int dup_ref(struct recorded_ref *ref, struct list_head *list) |
2799 | { | |
2800 | struct recorded_ref *new; | |
2801 | ||
e780b0d1 | 2802 | new = kmalloc(sizeof(*ref), GFP_KERNEL); |
ba5e8f2e JB |
2803 | if (!new) |
2804 | return -ENOMEM; | |
2805 | ||
2806 | new->dir = ref->dir; | |
2807 | new->dir_gen = ref->dir_gen; | |
2808 | new->full_path = NULL; | |
2809 | INIT_LIST_HEAD(&new->list); | |
2810 | list_add_tail(&new->list, list); | |
2811 | return 0; | |
2812 | } | |
2813 | ||
924794c9 | 2814 | static void __free_recorded_refs(struct list_head *head) |
31db9f7c AB |
2815 | { |
2816 | struct recorded_ref *cur; | |
31db9f7c | 2817 | |
e938c8ad AB |
2818 | while (!list_empty(head)) { |
2819 | cur = list_entry(head->next, struct recorded_ref, list); | |
924794c9 | 2820 | fs_path_free(cur->full_path); |
e938c8ad | 2821 | list_del(&cur->list); |
31db9f7c AB |
2822 | kfree(cur); |
2823 | } | |
31db9f7c AB |
2824 | } |
2825 | ||
2826 | static void free_recorded_refs(struct send_ctx *sctx) | |
2827 | { | |
924794c9 TI |
2828 | __free_recorded_refs(&sctx->new_refs); |
2829 | __free_recorded_refs(&sctx->deleted_refs); | |
31db9f7c AB |
2830 | } |
2831 | ||
2832 | /* | |
766702ef | 2833 | * Renames/moves a file/dir to its orphan name. Used when the first |
31db9f7c AB |
2834 | * ref of an unprocessed inode gets overwritten and for all non empty |
2835 | * directories. | |
2836 | */ | |
2837 | static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, | |
2838 | struct fs_path *path) | |
2839 | { | |
2840 | int ret; | |
2841 | struct fs_path *orphan; | |
2842 | ||
924794c9 | 2843 | orphan = fs_path_alloc(); |
31db9f7c AB |
2844 | if (!orphan) |
2845 | return -ENOMEM; | |
2846 | ||
2847 | ret = gen_unique_name(sctx, ino, gen, orphan); | |
2848 | if (ret < 0) | |
2849 | goto out; | |
2850 | ||
2851 | ret = send_rename(sctx, path, orphan); | |
2852 | ||
2853 | out: | |
924794c9 | 2854 | fs_path_free(orphan); |
31db9f7c AB |
2855 | return ret; |
2856 | } | |
2857 | ||
9dc44214 FM |
2858 | static struct orphan_dir_info * |
2859 | add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2860 | { | |
2861 | struct rb_node **p = &sctx->orphan_dirs.rb_node; | |
2862 | struct rb_node *parent = NULL; | |
2863 | struct orphan_dir_info *entry, *odi; | |
2864 | ||
e780b0d1 | 2865 | odi = kmalloc(sizeof(*odi), GFP_KERNEL); |
9dc44214 FM |
2866 | if (!odi) |
2867 | return ERR_PTR(-ENOMEM); | |
2868 | odi->ino = dir_ino; | |
2869 | odi->gen = 0; | |
2870 | ||
2871 | while (*p) { | |
2872 | parent = *p; | |
2873 | entry = rb_entry(parent, struct orphan_dir_info, node); | |
2874 | if (dir_ino < entry->ino) { | |
2875 | p = &(*p)->rb_left; | |
2876 | } else if (dir_ino > entry->ino) { | |
2877 | p = &(*p)->rb_right; | |
2878 | } else { | |
2879 | kfree(odi); | |
2880 | return entry; | |
2881 | } | |
2882 | } | |
2883 | ||
2884 | rb_link_node(&odi->node, parent, p); | |
2885 | rb_insert_color(&odi->node, &sctx->orphan_dirs); | |
2886 | return odi; | |
2887 | } | |
2888 | ||
2889 | static struct orphan_dir_info * | |
2890 | get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2891 | { | |
2892 | struct rb_node *n = sctx->orphan_dirs.rb_node; | |
2893 | struct orphan_dir_info *entry; | |
2894 | ||
2895 | while (n) { | |
2896 | entry = rb_entry(n, struct orphan_dir_info, node); | |
2897 | if (dir_ino < entry->ino) | |
2898 | n = n->rb_left; | |
2899 | else if (dir_ino > entry->ino) | |
2900 | n = n->rb_right; | |
2901 | else | |
2902 | return entry; | |
2903 | } | |
2904 | return NULL; | |
2905 | } | |
2906 | ||
2907 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino) | |
2908 | { | |
2909 | struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino); | |
2910 | ||
2911 | return odi != NULL; | |
2912 | } | |
2913 | ||
2914 | static void free_orphan_dir_info(struct send_ctx *sctx, | |
2915 | struct orphan_dir_info *odi) | |
2916 | { | |
2917 | if (!odi) | |
2918 | return; | |
2919 | rb_erase(&odi->node, &sctx->orphan_dirs); | |
2920 | kfree(odi); | |
2921 | } | |
2922 | ||
31db9f7c AB |
2923 | /* |
2924 | * Returns 1 if a directory can be removed at this point in time. | |
2925 | * We check this by iterating all dir items and checking if the inode behind | |
2926 | * the dir item was already processed. | |
2927 | */ | |
9dc44214 FM |
2928 | static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
2929 | u64 send_progress) | |
31db9f7c AB |
2930 | { |
2931 | int ret = 0; | |
2932 | struct btrfs_root *root = sctx->parent_root; | |
2933 | struct btrfs_path *path; | |
2934 | struct btrfs_key key; | |
2935 | struct btrfs_key found_key; | |
2936 | struct btrfs_key loc; | |
2937 | struct btrfs_dir_item *di; | |
2938 | ||
6d85ed05 AB |
2939 | /* |
2940 | * Don't try to rmdir the top/root subvolume dir. | |
2941 | */ | |
2942 | if (dir == BTRFS_FIRST_FREE_OBJECTID) | |
2943 | return 0; | |
2944 | ||
31db9f7c AB |
2945 | path = alloc_path_for_send(); |
2946 | if (!path) | |
2947 | return -ENOMEM; | |
2948 | ||
2949 | key.objectid = dir; | |
2950 | key.type = BTRFS_DIR_INDEX_KEY; | |
2951 | key.offset = 0; | |
dff6d0ad FDBM |
2952 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2953 | if (ret < 0) | |
2954 | goto out; | |
31db9f7c AB |
2955 | |
2956 | while (1) { | |
9dc44214 FM |
2957 | struct waiting_dir_move *dm; |
2958 | ||
dff6d0ad FDBM |
2959 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
2960 | ret = btrfs_next_leaf(root, path); | |
2961 | if (ret < 0) | |
2962 | goto out; | |
2963 | else if (ret > 0) | |
2964 | break; | |
2965 | continue; | |
31db9f7c | 2966 | } |
dff6d0ad FDBM |
2967 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2968 | path->slots[0]); | |
2969 | if (found_key.objectid != key.objectid || | |
2970 | found_key.type != key.type) | |
31db9f7c | 2971 | break; |
31db9f7c AB |
2972 | |
2973 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2974 | struct btrfs_dir_item); | |
2975 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
2976 | ||
9dc44214 FM |
2977 | dm = get_waiting_dir_move(sctx, loc.objectid); |
2978 | if (dm) { | |
2979 | struct orphan_dir_info *odi; | |
2980 | ||
2981 | odi = add_orphan_dir_info(sctx, dir); | |
2982 | if (IS_ERR(odi)) { | |
2983 | ret = PTR_ERR(odi); | |
2984 | goto out; | |
2985 | } | |
2986 | odi->gen = dir_gen; | |
2987 | dm->rmdir_ino = dir; | |
2988 | ret = 0; | |
2989 | goto out; | |
2990 | } | |
2991 | ||
31db9f7c | 2992 | if (loc.objectid > send_progress) { |
443f9d26 RK |
2993 | struct orphan_dir_info *odi; |
2994 | ||
2995 | odi = get_orphan_dir_info(sctx, dir); | |
2996 | free_orphan_dir_info(sctx, odi); | |
31db9f7c AB |
2997 | ret = 0; |
2998 | goto out; | |
2999 | } | |
3000 | ||
dff6d0ad | 3001 | path->slots[0]++; |
31db9f7c AB |
3002 | } |
3003 | ||
3004 | ret = 1; | |
3005 | ||
3006 | out: | |
3007 | btrfs_free_path(path); | |
3008 | return ret; | |
3009 | } | |
3010 | ||
9f03740a FDBM |
3011 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) |
3012 | { | |
9dc44214 | 3013 | struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); |
9f03740a | 3014 | |
9dc44214 | 3015 | return entry != NULL; |
9f03740a FDBM |
3016 | } |
3017 | ||
8b191a68 | 3018 | static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) |
9f03740a FDBM |
3019 | { |
3020 | struct rb_node **p = &sctx->waiting_dir_moves.rb_node; | |
3021 | struct rb_node *parent = NULL; | |
3022 | struct waiting_dir_move *entry, *dm; | |
3023 | ||
e780b0d1 | 3024 | dm = kmalloc(sizeof(*dm), GFP_KERNEL); |
9f03740a FDBM |
3025 | if (!dm) |
3026 | return -ENOMEM; | |
3027 | dm->ino = ino; | |
9dc44214 | 3028 | dm->rmdir_ino = 0; |
8b191a68 | 3029 | dm->orphanized = orphanized; |
9f03740a FDBM |
3030 | |
3031 | while (*p) { | |
3032 | parent = *p; | |
3033 | entry = rb_entry(parent, struct waiting_dir_move, node); | |
3034 | if (ino < entry->ino) { | |
3035 | p = &(*p)->rb_left; | |
3036 | } else if (ino > entry->ino) { | |
3037 | p = &(*p)->rb_right; | |
3038 | } else { | |
3039 | kfree(dm); | |
3040 | return -EEXIST; | |
3041 | } | |
3042 | } | |
3043 | ||
3044 | rb_link_node(&dm->node, parent, p); | |
3045 | rb_insert_color(&dm->node, &sctx->waiting_dir_moves); | |
3046 | return 0; | |
3047 | } | |
3048 | ||
9dc44214 FM |
3049 | static struct waiting_dir_move * |
3050 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
9f03740a FDBM |
3051 | { |
3052 | struct rb_node *n = sctx->waiting_dir_moves.rb_node; | |
3053 | struct waiting_dir_move *entry; | |
3054 | ||
3055 | while (n) { | |
3056 | entry = rb_entry(n, struct waiting_dir_move, node); | |
9dc44214 | 3057 | if (ino < entry->ino) |
9f03740a | 3058 | n = n->rb_left; |
9dc44214 | 3059 | else if (ino > entry->ino) |
9f03740a | 3060 | n = n->rb_right; |
9dc44214 FM |
3061 | else |
3062 | return entry; | |
9f03740a | 3063 | } |
9dc44214 FM |
3064 | return NULL; |
3065 | } | |
3066 | ||
3067 | static void free_waiting_dir_move(struct send_ctx *sctx, | |
3068 | struct waiting_dir_move *dm) | |
3069 | { | |
3070 | if (!dm) | |
3071 | return; | |
3072 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
3073 | kfree(dm); | |
9f03740a FDBM |
3074 | } |
3075 | ||
bfa7e1f8 FM |
3076 | static int add_pending_dir_move(struct send_ctx *sctx, |
3077 | u64 ino, | |
3078 | u64 ino_gen, | |
f959492f FM |
3079 | u64 parent_ino, |
3080 | struct list_head *new_refs, | |
84471e24 FM |
3081 | struct list_head *deleted_refs, |
3082 | const bool is_orphan) | |
9f03740a FDBM |
3083 | { |
3084 | struct rb_node **p = &sctx->pending_dir_moves.rb_node; | |
3085 | struct rb_node *parent = NULL; | |
73b802f4 | 3086 | struct pending_dir_move *entry = NULL, *pm; |
9f03740a FDBM |
3087 | struct recorded_ref *cur; |
3088 | int exists = 0; | |
3089 | int ret; | |
3090 | ||
e780b0d1 | 3091 | pm = kmalloc(sizeof(*pm), GFP_KERNEL); |
9f03740a FDBM |
3092 | if (!pm) |
3093 | return -ENOMEM; | |
3094 | pm->parent_ino = parent_ino; | |
bfa7e1f8 FM |
3095 | pm->ino = ino; |
3096 | pm->gen = ino_gen; | |
9f03740a FDBM |
3097 | INIT_LIST_HEAD(&pm->list); |
3098 | INIT_LIST_HEAD(&pm->update_refs); | |
3099 | RB_CLEAR_NODE(&pm->node); | |
3100 | ||
3101 | while (*p) { | |
3102 | parent = *p; | |
3103 | entry = rb_entry(parent, struct pending_dir_move, node); | |
3104 | if (parent_ino < entry->parent_ino) { | |
3105 | p = &(*p)->rb_left; | |
3106 | } else if (parent_ino > entry->parent_ino) { | |
3107 | p = &(*p)->rb_right; | |
3108 | } else { | |
3109 | exists = 1; | |
3110 | break; | |
3111 | } | |
3112 | } | |
3113 | ||
f959492f | 3114 | list_for_each_entry(cur, deleted_refs, list) { |
9f03740a FDBM |
3115 | ret = dup_ref(cur, &pm->update_refs); |
3116 | if (ret < 0) | |
3117 | goto out; | |
3118 | } | |
f959492f | 3119 | list_for_each_entry(cur, new_refs, list) { |
9f03740a FDBM |
3120 | ret = dup_ref(cur, &pm->update_refs); |
3121 | if (ret < 0) | |
3122 | goto out; | |
3123 | } | |
3124 | ||
8b191a68 | 3125 | ret = add_waiting_dir_move(sctx, pm->ino, is_orphan); |
9f03740a FDBM |
3126 | if (ret) |
3127 | goto out; | |
3128 | ||
3129 | if (exists) { | |
3130 | list_add_tail(&pm->list, &entry->list); | |
3131 | } else { | |
3132 | rb_link_node(&pm->node, parent, p); | |
3133 | rb_insert_color(&pm->node, &sctx->pending_dir_moves); | |
3134 | } | |
3135 | ret = 0; | |
3136 | out: | |
3137 | if (ret) { | |
3138 | __free_recorded_refs(&pm->update_refs); | |
3139 | kfree(pm); | |
3140 | } | |
3141 | return ret; | |
3142 | } | |
3143 | ||
3144 | static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, | |
3145 | u64 parent_ino) | |
3146 | { | |
3147 | struct rb_node *n = sctx->pending_dir_moves.rb_node; | |
3148 | struct pending_dir_move *entry; | |
3149 | ||
3150 | while (n) { | |
3151 | entry = rb_entry(n, struct pending_dir_move, node); | |
3152 | if (parent_ino < entry->parent_ino) | |
3153 | n = n->rb_left; | |
3154 | else if (parent_ino > entry->parent_ino) | |
3155 | n = n->rb_right; | |
3156 | else | |
3157 | return entry; | |
3158 | } | |
3159 | return NULL; | |
3160 | } | |
3161 | ||
801bec36 RK |
3162 | static int path_loop(struct send_ctx *sctx, struct fs_path *name, |
3163 | u64 ino, u64 gen, u64 *ancestor_ino) | |
3164 | { | |
3165 | int ret = 0; | |
3166 | u64 parent_inode = 0; | |
3167 | u64 parent_gen = 0; | |
3168 | u64 start_ino = ino; | |
3169 | ||
3170 | *ancestor_ino = 0; | |
3171 | while (ino != BTRFS_FIRST_FREE_OBJECTID) { | |
3172 | fs_path_reset(name); | |
3173 | ||
3174 | if (is_waiting_for_rm(sctx, ino)) | |
3175 | break; | |
3176 | if (is_waiting_for_move(sctx, ino)) { | |
3177 | if (*ancestor_ino == 0) | |
3178 | *ancestor_ino = ino; | |
3179 | ret = get_first_ref(sctx->parent_root, ino, | |
3180 | &parent_inode, &parent_gen, name); | |
3181 | } else { | |
3182 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
3183 | &parent_inode, | |
3184 | &parent_gen, name); | |
3185 | if (ret > 0) { | |
3186 | ret = 0; | |
3187 | break; | |
3188 | } | |
3189 | } | |
3190 | if (ret < 0) | |
3191 | break; | |
3192 | if (parent_inode == start_ino) { | |
3193 | ret = 1; | |
3194 | if (*ancestor_ino == 0) | |
3195 | *ancestor_ino = ino; | |
3196 | break; | |
3197 | } | |
3198 | ino = parent_inode; | |
3199 | gen = parent_gen; | |
3200 | } | |
3201 | return ret; | |
3202 | } | |
3203 | ||
9f03740a FDBM |
3204 | static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) |
3205 | { | |
3206 | struct fs_path *from_path = NULL; | |
3207 | struct fs_path *to_path = NULL; | |
2b863a13 | 3208 | struct fs_path *name = NULL; |
9f03740a FDBM |
3209 | u64 orig_progress = sctx->send_progress; |
3210 | struct recorded_ref *cur; | |
2b863a13 | 3211 | u64 parent_ino, parent_gen; |
9dc44214 FM |
3212 | struct waiting_dir_move *dm = NULL; |
3213 | u64 rmdir_ino = 0; | |
801bec36 RK |
3214 | u64 ancestor; |
3215 | bool is_orphan; | |
9f03740a FDBM |
3216 | int ret; |
3217 | ||
2b863a13 | 3218 | name = fs_path_alloc(); |
9f03740a | 3219 | from_path = fs_path_alloc(); |
2b863a13 FM |
3220 | if (!name || !from_path) { |
3221 | ret = -ENOMEM; | |
3222 | goto out; | |
3223 | } | |
9f03740a | 3224 | |
9dc44214 FM |
3225 | dm = get_waiting_dir_move(sctx, pm->ino); |
3226 | ASSERT(dm); | |
3227 | rmdir_ino = dm->rmdir_ino; | |
801bec36 | 3228 | is_orphan = dm->orphanized; |
9dc44214 | 3229 | free_waiting_dir_move(sctx, dm); |
2b863a13 | 3230 | |
801bec36 | 3231 | if (is_orphan) { |
84471e24 FM |
3232 | ret = gen_unique_name(sctx, pm->ino, |
3233 | pm->gen, from_path); | |
3234 | } else { | |
3235 | ret = get_first_ref(sctx->parent_root, pm->ino, | |
3236 | &parent_ino, &parent_gen, name); | |
3237 | if (ret < 0) | |
3238 | goto out; | |
3239 | ret = get_cur_path(sctx, parent_ino, parent_gen, | |
3240 | from_path); | |
3241 | if (ret < 0) | |
3242 | goto out; | |
3243 | ret = fs_path_add_path(from_path, name); | |
3244 | } | |
c992ec94 FM |
3245 | if (ret < 0) |
3246 | goto out; | |
2b863a13 | 3247 | |
f959492f | 3248 | sctx->send_progress = sctx->cur_ino + 1; |
801bec36 | 3249 | ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor); |
7969e77a FM |
3250 | if (ret < 0) |
3251 | goto out; | |
801bec36 RK |
3252 | if (ret) { |
3253 | LIST_HEAD(deleted_refs); | |
3254 | ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID); | |
3255 | ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor, | |
3256 | &pm->update_refs, &deleted_refs, | |
3257 | is_orphan); | |
3258 | if (ret < 0) | |
3259 | goto out; | |
3260 | if (rmdir_ino) { | |
3261 | dm = get_waiting_dir_move(sctx, pm->ino); | |
3262 | ASSERT(dm); | |
3263 | dm->rmdir_ino = rmdir_ino; | |
3264 | } | |
3265 | goto out; | |
3266 | } | |
c992ec94 FM |
3267 | fs_path_reset(name); |
3268 | to_path = name; | |
2b863a13 | 3269 | name = NULL; |
9f03740a FDBM |
3270 | ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); |
3271 | if (ret < 0) | |
3272 | goto out; | |
3273 | ||
3274 | ret = send_rename(sctx, from_path, to_path); | |
3275 | if (ret < 0) | |
3276 | goto out; | |
3277 | ||
9dc44214 FM |
3278 | if (rmdir_ino) { |
3279 | struct orphan_dir_info *odi; | |
3280 | ||
3281 | odi = get_orphan_dir_info(sctx, rmdir_ino); | |
3282 | if (!odi) { | |
3283 | /* already deleted */ | |
3284 | goto finish; | |
3285 | } | |
99ea42dd | 3286 | ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino); |
9dc44214 FM |
3287 | if (ret < 0) |
3288 | goto out; | |
3289 | if (!ret) | |
3290 | goto finish; | |
3291 | ||
3292 | name = fs_path_alloc(); | |
3293 | if (!name) { | |
3294 | ret = -ENOMEM; | |
3295 | goto out; | |
3296 | } | |
3297 | ret = get_cur_path(sctx, rmdir_ino, odi->gen, name); | |
3298 | if (ret < 0) | |
3299 | goto out; | |
3300 | ret = send_rmdir(sctx, name); | |
3301 | if (ret < 0) | |
3302 | goto out; | |
3303 | free_orphan_dir_info(sctx, odi); | |
3304 | } | |
3305 | ||
3306 | finish: | |
9f03740a FDBM |
3307 | ret = send_utimes(sctx, pm->ino, pm->gen); |
3308 | if (ret < 0) | |
3309 | goto out; | |
3310 | ||
3311 | /* | |
3312 | * After rename/move, need to update the utimes of both new parent(s) | |
3313 | * and old parent(s). | |
3314 | */ | |
3315 | list_for_each_entry(cur, &pm->update_refs, list) { | |
764433a1 RK |
3316 | /* |
3317 | * The parent inode might have been deleted in the send snapshot | |
3318 | */ | |
3319 | ret = get_inode_info(sctx->send_root, cur->dir, NULL, | |
3320 | NULL, NULL, NULL, NULL, NULL); | |
3321 | if (ret == -ENOENT) { | |
3322 | ret = 0; | |
9dc44214 | 3323 | continue; |
764433a1 RK |
3324 | } |
3325 | if (ret < 0) | |
3326 | goto out; | |
3327 | ||
9f03740a FDBM |
3328 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
3329 | if (ret < 0) | |
3330 | goto out; | |
3331 | } | |
3332 | ||
3333 | out: | |
2b863a13 | 3334 | fs_path_free(name); |
9f03740a FDBM |
3335 | fs_path_free(from_path); |
3336 | fs_path_free(to_path); | |
3337 | sctx->send_progress = orig_progress; | |
3338 | ||
3339 | return ret; | |
3340 | } | |
3341 | ||
3342 | static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) | |
3343 | { | |
3344 | if (!list_empty(&m->list)) | |
3345 | list_del(&m->list); | |
3346 | if (!RB_EMPTY_NODE(&m->node)) | |
3347 | rb_erase(&m->node, &sctx->pending_dir_moves); | |
3348 | __free_recorded_refs(&m->update_refs); | |
3349 | kfree(m); | |
3350 | } | |
3351 | ||
983cdcb1 RK |
3352 | static void tail_append_pending_moves(struct send_ctx *sctx, |
3353 | struct pending_dir_move *moves, | |
9f03740a FDBM |
3354 | struct list_head *stack) |
3355 | { | |
3356 | if (list_empty(&moves->list)) { | |
3357 | list_add_tail(&moves->list, stack); | |
3358 | } else { | |
3359 | LIST_HEAD(list); | |
3360 | list_splice_init(&moves->list, &list); | |
3361 | list_add_tail(&moves->list, stack); | |
3362 | list_splice_tail(&list, stack); | |
3363 | } | |
983cdcb1 RK |
3364 | if (!RB_EMPTY_NODE(&moves->node)) { |
3365 | rb_erase(&moves->node, &sctx->pending_dir_moves); | |
3366 | RB_CLEAR_NODE(&moves->node); | |
3367 | } | |
9f03740a FDBM |
3368 | } |
3369 | ||
3370 | static int apply_children_dir_moves(struct send_ctx *sctx) | |
3371 | { | |
3372 | struct pending_dir_move *pm; | |
3373 | struct list_head stack; | |
3374 | u64 parent_ino = sctx->cur_ino; | |
3375 | int ret = 0; | |
3376 | ||
3377 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3378 | if (!pm) | |
3379 | return 0; | |
3380 | ||
3381 | INIT_LIST_HEAD(&stack); | |
983cdcb1 | 3382 | tail_append_pending_moves(sctx, pm, &stack); |
9f03740a FDBM |
3383 | |
3384 | while (!list_empty(&stack)) { | |
3385 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3386 | parent_ino = pm->ino; | |
3387 | ret = apply_dir_move(sctx, pm); | |
3388 | free_pending_move(sctx, pm); | |
3389 | if (ret) | |
3390 | goto out; | |
3391 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3392 | if (pm) | |
983cdcb1 | 3393 | tail_append_pending_moves(sctx, pm, &stack); |
9f03740a FDBM |
3394 | } |
3395 | return 0; | |
3396 | ||
3397 | out: | |
3398 | while (!list_empty(&stack)) { | |
3399 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3400 | free_pending_move(sctx, pm); | |
3401 | } | |
3402 | return ret; | |
3403 | } | |
3404 | ||
84471e24 FM |
3405 | /* |
3406 | * We might need to delay a directory rename even when no ancestor directory | |
3407 | * (in the send root) with a higher inode number than ours (sctx->cur_ino) was | |
3408 | * renamed. This happens when we rename a directory to the old name (the name | |
3409 | * in the parent root) of some other unrelated directory that got its rename | |
3410 | * delayed due to some ancestor with higher number that got renamed. | |
3411 | * | |
3412 | * Example: | |
3413 | * | |
3414 | * Parent snapshot: | |
3415 | * . (ino 256) | |
3416 | * |---- a/ (ino 257) | |
3417 | * | |---- file (ino 260) | |
3418 | * | | |
3419 | * |---- b/ (ino 258) | |
3420 | * |---- c/ (ino 259) | |
3421 | * | |
3422 | * Send snapshot: | |
3423 | * . (ino 256) | |
3424 | * |---- a/ (ino 258) | |
3425 | * |---- x/ (ino 259) | |
3426 | * |---- y/ (ino 257) | |
3427 | * |----- file (ino 260) | |
3428 | * | |
3429 | * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257 | |
3430 | * from 'a' to 'x/y' happening first, which in turn depends on the rename of | |
3431 | * inode 259 from 'c' to 'x'. So the order of rename commands the send stream | |
3432 | * must issue is: | |
3433 | * | |
3434 | * 1 - rename 259 from 'c' to 'x' | |
3435 | * 2 - rename 257 from 'a' to 'x/y' | |
3436 | * 3 - rename 258 from 'b' to 'a' | |
3437 | * | |
3438 | * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can | |
3439 | * be done right away and < 0 on error. | |
3440 | */ | |
3441 | static int wait_for_dest_dir_move(struct send_ctx *sctx, | |
3442 | struct recorded_ref *parent_ref, | |
3443 | const bool is_orphan) | |
3444 | { | |
2ff7e61e | 3445 | struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info; |
84471e24 FM |
3446 | struct btrfs_path *path; |
3447 | struct btrfs_key key; | |
3448 | struct btrfs_key di_key; | |
3449 | struct btrfs_dir_item *di; | |
3450 | u64 left_gen; | |
3451 | u64 right_gen; | |
3452 | int ret = 0; | |
801bec36 | 3453 | struct waiting_dir_move *wdm; |
84471e24 FM |
3454 | |
3455 | if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) | |
3456 | return 0; | |
3457 | ||
3458 | path = alloc_path_for_send(); | |
3459 | if (!path) | |
3460 | return -ENOMEM; | |
3461 | ||
3462 | key.objectid = parent_ref->dir; | |
3463 | key.type = BTRFS_DIR_ITEM_KEY; | |
3464 | key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); | |
3465 | ||
3466 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
3467 | if (ret < 0) { | |
3468 | goto out; | |
3469 | } else if (ret > 0) { | |
3470 | ret = 0; | |
3471 | goto out; | |
3472 | } | |
3473 | ||
2ff7e61e JM |
3474 | di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name, |
3475 | parent_ref->name_len); | |
84471e24 FM |
3476 | if (!di) { |
3477 | ret = 0; | |
3478 | goto out; | |
3479 | } | |
3480 | /* | |
3481 | * di_key.objectid has the number of the inode that has a dentry in the | |
3482 | * parent directory with the same name that sctx->cur_ino is being | |
3483 | * renamed to. We need to check if that inode is in the send root as | |
3484 | * well and if it is currently marked as an inode with a pending rename, | |
3485 | * if it is, we need to delay the rename of sctx->cur_ino as well, so | |
3486 | * that it happens after that other inode is renamed. | |
3487 | */ | |
3488 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); | |
3489 | if (di_key.type != BTRFS_INODE_ITEM_KEY) { | |
3490 | ret = 0; | |
3491 | goto out; | |
3492 | } | |
3493 | ||
3494 | ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL, | |
3495 | &left_gen, NULL, NULL, NULL, NULL); | |
3496 | if (ret < 0) | |
3497 | goto out; | |
3498 | ret = get_inode_info(sctx->send_root, di_key.objectid, NULL, | |
3499 | &right_gen, NULL, NULL, NULL, NULL); | |
3500 | if (ret < 0) { | |
3501 | if (ret == -ENOENT) | |
3502 | ret = 0; | |
3503 | goto out; | |
3504 | } | |
3505 | ||
3506 | /* Different inode, no need to delay the rename of sctx->cur_ino */ | |
3507 | if (right_gen != left_gen) { | |
3508 | ret = 0; | |
3509 | goto out; | |
3510 | } | |
3511 | ||
801bec36 RK |
3512 | wdm = get_waiting_dir_move(sctx, di_key.objectid); |
3513 | if (wdm && !wdm->orphanized) { | |
84471e24 FM |
3514 | ret = add_pending_dir_move(sctx, |
3515 | sctx->cur_ino, | |
3516 | sctx->cur_inode_gen, | |
3517 | di_key.objectid, | |
3518 | &sctx->new_refs, | |
3519 | &sctx->deleted_refs, | |
3520 | is_orphan); | |
3521 | if (!ret) | |
3522 | ret = 1; | |
3523 | } | |
3524 | out: | |
3525 | btrfs_free_path(path); | |
3526 | return ret; | |
3527 | } | |
3528 | ||
80aa6027 | 3529 | /* |
ea37d599 FM |
3530 | * Check if inode ino2, or any of its ancestors, is inode ino1. |
3531 | * Return 1 if true, 0 if false and < 0 on error. | |
3532 | */ | |
3533 | static int check_ino_in_path(struct btrfs_root *root, | |
3534 | const u64 ino1, | |
3535 | const u64 ino1_gen, | |
3536 | const u64 ino2, | |
3537 | const u64 ino2_gen, | |
3538 | struct fs_path *fs_path) | |
3539 | { | |
3540 | u64 ino = ino2; | |
3541 | ||
3542 | if (ino1 == ino2) | |
3543 | return ino1_gen == ino2_gen; | |
3544 | ||
3545 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { | |
3546 | u64 parent; | |
3547 | u64 parent_gen; | |
3548 | int ret; | |
3549 | ||
3550 | fs_path_reset(fs_path); | |
3551 | ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path); | |
3552 | if (ret < 0) | |
3553 | return ret; | |
3554 | if (parent == ino1) | |
3555 | return parent_gen == ino1_gen; | |
3556 | ino = parent; | |
3557 | } | |
3558 | return 0; | |
3559 | } | |
3560 | ||
3561 | /* | |
3562 | * Check if ino ino1 is an ancestor of inode ino2 in the given root for any | |
3563 | * possible path (in case ino2 is not a directory and has multiple hard links). | |
80aa6027 FM |
3564 | * Return 1 if true, 0 if false and < 0 on error. |
3565 | */ | |
3566 | static int is_ancestor(struct btrfs_root *root, | |
3567 | const u64 ino1, | |
3568 | const u64 ino1_gen, | |
3569 | const u64 ino2, | |
3570 | struct fs_path *fs_path) | |
3571 | { | |
ea37d599 | 3572 | bool free_fs_path = false; |
72c3668f | 3573 | int ret = 0; |
ea37d599 FM |
3574 | struct btrfs_path *path = NULL; |
3575 | struct btrfs_key key; | |
72c3668f FM |
3576 | |
3577 | if (!fs_path) { | |
3578 | fs_path = fs_path_alloc(); | |
3579 | if (!fs_path) | |
3580 | return -ENOMEM; | |
ea37d599 | 3581 | free_fs_path = true; |
72c3668f | 3582 | } |
80aa6027 | 3583 | |
ea37d599 FM |
3584 | path = alloc_path_for_send(); |
3585 | if (!path) { | |
3586 | ret = -ENOMEM; | |
3587 | goto out; | |
3588 | } | |
80aa6027 | 3589 | |
ea37d599 FM |
3590 | key.objectid = ino2; |
3591 | key.type = BTRFS_INODE_REF_KEY; | |
3592 | key.offset = 0; | |
3593 | ||
3594 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3595 | if (ret < 0) | |
3596 | goto out; | |
3597 | ||
3598 | while (true) { | |
3599 | struct extent_buffer *leaf = path->nodes[0]; | |
3600 | int slot = path->slots[0]; | |
3601 | u32 cur_offset = 0; | |
3602 | u32 item_size; | |
3603 | ||
3604 | if (slot >= btrfs_header_nritems(leaf)) { | |
3605 | ret = btrfs_next_leaf(root, path); | |
3606 | if (ret < 0) | |
3607 | goto out; | |
3608 | if (ret > 0) | |
3609 | break; | |
3610 | continue; | |
72c3668f | 3611 | } |
ea37d599 FM |
3612 | |
3613 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
3614 | if (key.objectid != ino2) | |
3615 | break; | |
3616 | if (key.type != BTRFS_INODE_REF_KEY && | |
3617 | key.type != BTRFS_INODE_EXTREF_KEY) | |
3618 | break; | |
3619 | ||
3620 | item_size = btrfs_item_size_nr(leaf, slot); | |
3621 | while (cur_offset < item_size) { | |
3622 | u64 parent; | |
3623 | u64 parent_gen; | |
3624 | ||
3625 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
3626 | unsigned long ptr; | |
3627 | struct btrfs_inode_extref *extref; | |
3628 | ||
3629 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3630 | extref = (struct btrfs_inode_extref *) | |
3631 | (ptr + cur_offset); | |
3632 | parent = btrfs_inode_extref_parent(leaf, | |
3633 | extref); | |
3634 | cur_offset += sizeof(*extref); | |
3635 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
3636 | extref); | |
3637 | } else { | |
3638 | parent = key.offset; | |
3639 | cur_offset = item_size; | |
3640 | } | |
3641 | ||
3642 | ret = get_inode_info(root, parent, NULL, &parent_gen, | |
3643 | NULL, NULL, NULL, NULL); | |
3644 | if (ret < 0) | |
3645 | goto out; | |
3646 | ret = check_ino_in_path(root, ino1, ino1_gen, | |
3647 | parent, parent_gen, fs_path); | |
3648 | if (ret) | |
3649 | goto out; | |
80aa6027 | 3650 | } |
ea37d599 | 3651 | path->slots[0]++; |
80aa6027 | 3652 | } |
ea37d599 | 3653 | ret = 0; |
72c3668f | 3654 | out: |
ea37d599 FM |
3655 | btrfs_free_path(path); |
3656 | if (free_fs_path) | |
72c3668f FM |
3657 | fs_path_free(fs_path); |
3658 | return ret; | |
80aa6027 FM |
3659 | } |
3660 | ||
9f03740a | 3661 | static int wait_for_parent_move(struct send_ctx *sctx, |
8b191a68 FM |
3662 | struct recorded_ref *parent_ref, |
3663 | const bool is_orphan) | |
9f03740a | 3664 | { |
f959492f | 3665 | int ret = 0; |
9f03740a | 3666 | u64 ino = parent_ref->dir; |
fe9c798d | 3667 | u64 ino_gen = parent_ref->dir_gen; |
9f03740a | 3668 | u64 parent_ino_before, parent_ino_after; |
9f03740a FDBM |
3669 | struct fs_path *path_before = NULL; |
3670 | struct fs_path *path_after = NULL; | |
3671 | int len1, len2; | |
9f03740a FDBM |
3672 | |
3673 | path_after = fs_path_alloc(); | |
f959492f FM |
3674 | path_before = fs_path_alloc(); |
3675 | if (!path_after || !path_before) { | |
9f03740a FDBM |
3676 | ret = -ENOMEM; |
3677 | goto out; | |
3678 | } | |
3679 | ||
bfa7e1f8 | 3680 | /* |
f959492f FM |
3681 | * Our current directory inode may not yet be renamed/moved because some |
3682 | * ancestor (immediate or not) has to be renamed/moved first. So find if | |
3683 | * such ancestor exists and make sure our own rename/move happens after | |
80aa6027 FM |
3684 | * that ancestor is processed to avoid path build infinite loops (done |
3685 | * at get_cur_path()). | |
bfa7e1f8 | 3686 | */ |
f959492f | 3687 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { |
fe9c798d FM |
3688 | u64 parent_ino_after_gen; |
3689 | ||
f959492f | 3690 | if (is_waiting_for_move(sctx, ino)) { |
80aa6027 FM |
3691 | /* |
3692 | * If the current inode is an ancestor of ino in the | |
3693 | * parent root, we need to delay the rename of the | |
3694 | * current inode, otherwise don't delayed the rename | |
3695 | * because we can end up with a circular dependency | |
3696 | * of renames, resulting in some directories never | |
3697 | * getting the respective rename operations issued in | |
3698 | * the send stream or getting into infinite path build | |
3699 | * loops. | |
3700 | */ | |
3701 | ret = is_ancestor(sctx->parent_root, | |
3702 | sctx->cur_ino, sctx->cur_inode_gen, | |
3703 | ino, path_before); | |
4122ea64 FM |
3704 | if (ret) |
3705 | break; | |
f959492f | 3706 | } |
bfa7e1f8 FM |
3707 | |
3708 | fs_path_reset(path_before); | |
3709 | fs_path_reset(path_after); | |
3710 | ||
3711 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
fe9c798d | 3712 | &parent_ino_after_gen, path_after); |
bfa7e1f8 FM |
3713 | if (ret < 0) |
3714 | goto out; | |
3715 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
3716 | NULL, path_before); | |
f959492f | 3717 | if (ret < 0 && ret != -ENOENT) { |
bfa7e1f8 | 3718 | goto out; |
f959492f | 3719 | } else if (ret == -ENOENT) { |
bf8e8ca6 | 3720 | ret = 0; |
f959492f | 3721 | break; |
bfa7e1f8 FM |
3722 | } |
3723 | ||
3724 | len1 = fs_path_len(path_before); | |
3725 | len2 = fs_path_len(path_after); | |
f959492f FM |
3726 | if (ino > sctx->cur_ino && |
3727 | (parent_ino_before != parent_ino_after || len1 != len2 || | |
3728 | memcmp(path_before->start, path_after->start, len1))) { | |
fe9c798d FM |
3729 | u64 parent_ino_gen; |
3730 | ||
3731 | ret = get_inode_info(sctx->parent_root, ino, NULL, | |
3732 | &parent_ino_gen, NULL, NULL, NULL, | |
3733 | NULL); | |
3734 | if (ret < 0) | |
3735 | goto out; | |
3736 | if (ino_gen == parent_ino_gen) { | |
3737 | ret = 1; | |
3738 | break; | |
3739 | } | |
bfa7e1f8 | 3740 | } |
bfa7e1f8 | 3741 | ino = parent_ino_after; |
fe9c798d | 3742 | ino_gen = parent_ino_after_gen; |
bfa7e1f8 FM |
3743 | } |
3744 | ||
9f03740a FDBM |
3745 | out: |
3746 | fs_path_free(path_before); | |
3747 | fs_path_free(path_after); | |
3748 | ||
f959492f FM |
3749 | if (ret == 1) { |
3750 | ret = add_pending_dir_move(sctx, | |
3751 | sctx->cur_ino, | |
3752 | sctx->cur_inode_gen, | |
3753 | ino, | |
3754 | &sctx->new_refs, | |
84471e24 | 3755 | &sctx->deleted_refs, |
8b191a68 | 3756 | is_orphan); |
f959492f FM |
3757 | if (!ret) |
3758 | ret = 1; | |
3759 | } | |
3760 | ||
9f03740a FDBM |
3761 | return ret; |
3762 | } | |
3763 | ||
f5962781 FM |
3764 | static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) |
3765 | { | |
3766 | int ret; | |
3767 | struct fs_path *new_path; | |
3768 | ||
3769 | /* | |
3770 | * Our reference's name member points to its full_path member string, so | |
3771 | * we use here a new path. | |
3772 | */ | |
3773 | new_path = fs_path_alloc(); | |
3774 | if (!new_path) | |
3775 | return -ENOMEM; | |
3776 | ||
3777 | ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path); | |
3778 | if (ret < 0) { | |
3779 | fs_path_free(new_path); | |
3780 | return ret; | |
3781 | } | |
3782 | ret = fs_path_add(new_path, ref->name, ref->name_len); | |
3783 | if (ret < 0) { | |
3784 | fs_path_free(new_path); | |
3785 | return ret; | |
3786 | } | |
3787 | ||
3788 | fs_path_free(ref->full_path); | |
3789 | set_ref_path(ref, new_path); | |
3790 | ||
3791 | return 0; | |
3792 | } | |
3793 | ||
31db9f7c AB |
3794 | /* |
3795 | * This does all the move/link/unlink/rmdir magic. | |
3796 | */ | |
9f03740a | 3797 | static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) |
31db9f7c | 3798 | { |
04ab956e | 3799 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
3800 | int ret = 0; |
3801 | struct recorded_ref *cur; | |
1f4692da | 3802 | struct recorded_ref *cur2; |
ba5e8f2e | 3803 | struct list_head check_dirs; |
31db9f7c | 3804 | struct fs_path *valid_path = NULL; |
b24baf69 | 3805 | u64 ow_inode = 0; |
31db9f7c | 3806 | u64 ow_gen; |
f5962781 | 3807 | u64 ow_mode; |
31db9f7c AB |
3808 | int did_overwrite = 0; |
3809 | int is_orphan = 0; | |
29d6d30f | 3810 | u64 last_dir_ino_rm = 0; |
84471e24 | 3811 | bool can_rename = true; |
f5962781 | 3812 | bool orphanized_dir = false; |
fdb13889 | 3813 | bool orphanized_ancestor = false; |
31db9f7c | 3814 | |
04ab956e | 3815 | btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino); |
31db9f7c | 3816 | |
6d85ed05 AB |
3817 | /* |
3818 | * This should never happen as the root dir always has the same ref | |
3819 | * which is always '..' | |
3820 | */ | |
3821 | BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); | |
ba5e8f2e | 3822 | INIT_LIST_HEAD(&check_dirs); |
6d85ed05 | 3823 | |
924794c9 | 3824 | valid_path = fs_path_alloc(); |
31db9f7c AB |
3825 | if (!valid_path) { |
3826 | ret = -ENOMEM; | |
3827 | goto out; | |
3828 | } | |
3829 | ||
31db9f7c AB |
3830 | /* |
3831 | * First, check if the first ref of the current inode was overwritten | |
3832 | * before. If yes, we know that the current inode was already orphanized | |
3833 | * and thus use the orphan name. If not, we can use get_cur_path to | |
3834 | * get the path of the first ref as it would like while receiving at | |
3835 | * this point in time. | |
3836 | * New inodes are always orphan at the beginning, so force to use the | |
3837 | * orphan name in this case. | |
3838 | * The first ref is stored in valid_path and will be updated if it | |
3839 | * gets moved around. | |
3840 | */ | |
3841 | if (!sctx->cur_inode_new) { | |
3842 | ret = did_overwrite_first_ref(sctx, sctx->cur_ino, | |
3843 | sctx->cur_inode_gen); | |
3844 | if (ret < 0) | |
3845 | goto out; | |
3846 | if (ret) | |
3847 | did_overwrite = 1; | |
3848 | } | |
3849 | if (sctx->cur_inode_new || did_overwrite) { | |
3850 | ret = gen_unique_name(sctx, sctx->cur_ino, | |
3851 | sctx->cur_inode_gen, valid_path); | |
3852 | if (ret < 0) | |
3853 | goto out; | |
3854 | is_orphan = 1; | |
3855 | } else { | |
3856 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
3857 | valid_path); | |
3858 | if (ret < 0) | |
3859 | goto out; | |
3860 | } | |
3861 | ||
3862 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
1f4692da AB |
3863 | /* |
3864 | * We may have refs where the parent directory does not exist | |
3865 | * yet. This happens if the parent directories inum is higher | |
3866 | * the the current inum. To handle this case, we create the | |
3867 | * parent directory out of order. But we need to check if this | |
3868 | * did already happen before due to other refs in the same dir. | |
3869 | */ | |
3870 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); | |
3871 | if (ret < 0) | |
3872 | goto out; | |
3873 | if (ret == inode_state_will_create) { | |
3874 | ret = 0; | |
3875 | /* | |
3876 | * First check if any of the current inodes refs did | |
3877 | * already create the dir. | |
3878 | */ | |
3879 | list_for_each_entry(cur2, &sctx->new_refs, list) { | |
3880 | if (cur == cur2) | |
3881 | break; | |
3882 | if (cur2->dir == cur->dir) { | |
3883 | ret = 1; | |
3884 | break; | |
3885 | } | |
3886 | } | |
3887 | ||
3888 | /* | |
3889 | * If that did not happen, check if a previous inode | |
3890 | * did already create the dir. | |
3891 | */ | |
3892 | if (!ret) | |
3893 | ret = did_create_dir(sctx, cur->dir); | |
3894 | if (ret < 0) | |
3895 | goto out; | |
3896 | if (!ret) { | |
3897 | ret = send_create_inode(sctx, cur->dir); | |
3898 | if (ret < 0) | |
3899 | goto out; | |
3900 | } | |
3901 | } | |
3902 | ||
31db9f7c AB |
3903 | /* |
3904 | * Check if this new ref would overwrite the first ref of | |
3905 | * another unprocessed inode. If yes, orphanize the | |
3906 | * overwritten inode. If we find an overwritten ref that is | |
3907 | * not the first ref, simply unlink it. | |
3908 | */ | |
3909 | ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
3910 | cur->name, cur->name_len, | |
f5962781 | 3911 | &ow_inode, &ow_gen, &ow_mode); |
31db9f7c AB |
3912 | if (ret < 0) |
3913 | goto out; | |
3914 | if (ret) { | |
924794c9 TI |
3915 | ret = is_first_ref(sctx->parent_root, |
3916 | ow_inode, cur->dir, cur->name, | |
3917 | cur->name_len); | |
31db9f7c AB |
3918 | if (ret < 0) |
3919 | goto out; | |
3920 | if (ret) { | |
8996a48c | 3921 | struct name_cache_entry *nce; |
801bec36 | 3922 | struct waiting_dir_move *wdm; |
8996a48c | 3923 | |
31db9f7c AB |
3924 | ret = orphanize_inode(sctx, ow_inode, ow_gen, |
3925 | cur->full_path); | |
3926 | if (ret < 0) | |
3927 | goto out; | |
f5962781 FM |
3928 | if (S_ISDIR(ow_mode)) |
3929 | orphanized_dir = true; | |
801bec36 RK |
3930 | |
3931 | /* | |
3932 | * If ow_inode has its rename operation delayed | |
3933 | * make sure that its orphanized name is used in | |
3934 | * the source path when performing its rename | |
3935 | * operation. | |
3936 | */ | |
3937 | if (is_waiting_for_move(sctx, ow_inode)) { | |
3938 | wdm = get_waiting_dir_move(sctx, | |
3939 | ow_inode); | |
3940 | ASSERT(wdm); | |
3941 | wdm->orphanized = true; | |
3942 | } | |
3943 | ||
8996a48c FM |
3944 | /* |
3945 | * Make sure we clear our orphanized inode's | |
3946 | * name from the name cache. This is because the | |
3947 | * inode ow_inode might be an ancestor of some | |
3948 | * other inode that will be orphanized as well | |
3949 | * later and has an inode number greater than | |
3950 | * sctx->send_progress. We need to prevent | |
3951 | * future name lookups from using the old name | |
3952 | * and get instead the orphan name. | |
3953 | */ | |
3954 | nce = name_cache_search(sctx, ow_inode, ow_gen); | |
3955 | if (nce) { | |
3956 | name_cache_delete(sctx, nce); | |
3957 | kfree(nce); | |
3958 | } | |
801bec36 RK |
3959 | |
3960 | /* | |
3961 | * ow_inode might currently be an ancestor of | |
3962 | * cur_ino, therefore compute valid_path (the | |
3963 | * current path of cur_ino) again because it | |
3964 | * might contain the pre-orphanization name of | |
3965 | * ow_inode, which is no longer valid. | |
3966 | */ | |
72c3668f FM |
3967 | ret = is_ancestor(sctx->parent_root, |
3968 | ow_inode, ow_gen, | |
3969 | sctx->cur_ino, NULL); | |
3970 | if (ret > 0) { | |
fdb13889 | 3971 | orphanized_ancestor = true; |
72c3668f FM |
3972 | fs_path_reset(valid_path); |
3973 | ret = get_cur_path(sctx, sctx->cur_ino, | |
3974 | sctx->cur_inode_gen, | |
3975 | valid_path); | |
3976 | } | |
801bec36 RK |
3977 | if (ret < 0) |
3978 | goto out; | |
31db9f7c AB |
3979 | } else { |
3980 | ret = send_unlink(sctx, cur->full_path); | |
3981 | if (ret < 0) | |
3982 | goto out; | |
3983 | } | |
3984 | } | |
3985 | ||
84471e24 FM |
3986 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { |
3987 | ret = wait_for_dest_dir_move(sctx, cur, is_orphan); | |
3988 | if (ret < 0) | |
3989 | goto out; | |
3990 | if (ret == 1) { | |
3991 | can_rename = false; | |
3992 | *pending_move = 1; | |
3993 | } | |
3994 | } | |
3995 | ||
8b191a68 FM |
3996 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root && |
3997 | can_rename) { | |
3998 | ret = wait_for_parent_move(sctx, cur, is_orphan); | |
3999 | if (ret < 0) | |
4000 | goto out; | |
4001 | if (ret == 1) { | |
4002 | can_rename = false; | |
4003 | *pending_move = 1; | |
4004 | } | |
4005 | } | |
4006 | ||
31db9f7c AB |
4007 | /* |
4008 | * link/move the ref to the new place. If we have an orphan | |
4009 | * inode, move it and update valid_path. If not, link or move | |
4010 | * it depending on the inode mode. | |
4011 | */ | |
84471e24 | 4012 | if (is_orphan && can_rename) { |
31db9f7c AB |
4013 | ret = send_rename(sctx, valid_path, cur->full_path); |
4014 | if (ret < 0) | |
4015 | goto out; | |
4016 | is_orphan = 0; | |
4017 | ret = fs_path_copy(valid_path, cur->full_path); | |
4018 | if (ret < 0) | |
4019 | goto out; | |
84471e24 | 4020 | } else if (can_rename) { |
31db9f7c AB |
4021 | if (S_ISDIR(sctx->cur_inode_mode)) { |
4022 | /* | |
4023 | * Dirs can't be linked, so move it. For moved | |
4024 | * dirs, we always have one new and one deleted | |
4025 | * ref. The deleted ref is ignored later. | |
4026 | */ | |
8b191a68 FM |
4027 | ret = send_rename(sctx, valid_path, |
4028 | cur->full_path); | |
4029 | if (!ret) | |
4030 | ret = fs_path_copy(valid_path, | |
4031 | cur->full_path); | |
31db9f7c AB |
4032 | if (ret < 0) |
4033 | goto out; | |
4034 | } else { | |
f5962781 FM |
4035 | /* |
4036 | * We might have previously orphanized an inode | |
4037 | * which is an ancestor of our current inode, | |
4038 | * so our reference's full path, which was | |
4039 | * computed before any such orphanizations, must | |
4040 | * be updated. | |
4041 | */ | |
4042 | if (orphanized_dir) { | |
4043 | ret = update_ref_path(sctx, cur); | |
4044 | if (ret < 0) | |
4045 | goto out; | |
4046 | } | |
31db9f7c AB |
4047 | ret = send_link(sctx, cur->full_path, |
4048 | valid_path); | |
4049 | if (ret < 0) | |
4050 | goto out; | |
4051 | } | |
4052 | } | |
ba5e8f2e | 4053 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4054 | if (ret < 0) |
4055 | goto out; | |
4056 | } | |
4057 | ||
4058 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { | |
4059 | /* | |
4060 | * Check if we can already rmdir the directory. If not, | |
4061 | * orphanize it. For every dir item inside that gets deleted | |
4062 | * later, we do this check again and rmdir it then if possible. | |
4063 | * See the use of check_dirs for more details. | |
4064 | */ | |
9dc44214 FM |
4065 | ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
4066 | sctx->cur_ino); | |
31db9f7c AB |
4067 | if (ret < 0) |
4068 | goto out; | |
4069 | if (ret) { | |
4070 | ret = send_rmdir(sctx, valid_path); | |
4071 | if (ret < 0) | |
4072 | goto out; | |
4073 | } else if (!is_orphan) { | |
4074 | ret = orphanize_inode(sctx, sctx->cur_ino, | |
4075 | sctx->cur_inode_gen, valid_path); | |
4076 | if (ret < 0) | |
4077 | goto out; | |
4078 | is_orphan = 1; | |
4079 | } | |
4080 | ||
4081 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
ba5e8f2e | 4082 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4083 | if (ret < 0) |
4084 | goto out; | |
4085 | } | |
ccf1626b AB |
4086 | } else if (S_ISDIR(sctx->cur_inode_mode) && |
4087 | !list_empty(&sctx->deleted_refs)) { | |
4088 | /* | |
4089 | * We have a moved dir. Add the old parent to check_dirs | |
4090 | */ | |
4091 | cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, | |
4092 | list); | |
ba5e8f2e | 4093 | ret = dup_ref(cur, &check_dirs); |
ccf1626b AB |
4094 | if (ret < 0) |
4095 | goto out; | |
31db9f7c AB |
4096 | } else if (!S_ISDIR(sctx->cur_inode_mode)) { |
4097 | /* | |
4098 | * We have a non dir inode. Go through all deleted refs and | |
4099 | * unlink them if they were not already overwritten by other | |
4100 | * inodes. | |
4101 | */ | |
4102 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
4103 | ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
4104 | sctx->cur_ino, sctx->cur_inode_gen, | |
4105 | cur->name, cur->name_len); | |
4106 | if (ret < 0) | |
4107 | goto out; | |
4108 | if (!ret) { | |
fdb13889 FM |
4109 | /* |
4110 | * If we orphanized any ancestor before, we need | |
4111 | * to recompute the full path for deleted names, | |
4112 | * since any such path was computed before we | |
4113 | * processed any references and orphanized any | |
4114 | * ancestor inode. | |
4115 | */ | |
4116 | if (orphanized_ancestor) { | |
f5962781 FM |
4117 | ret = update_ref_path(sctx, cur); |
4118 | if (ret < 0) | |
fdb13889 | 4119 | goto out; |
fdb13889 | 4120 | } |
1f4692da AB |
4121 | ret = send_unlink(sctx, cur->full_path); |
4122 | if (ret < 0) | |
4123 | goto out; | |
31db9f7c | 4124 | } |
ba5e8f2e | 4125 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4126 | if (ret < 0) |
4127 | goto out; | |
4128 | } | |
31db9f7c AB |
4129 | /* |
4130 | * If the inode is still orphan, unlink the orphan. This may | |
4131 | * happen when a previous inode did overwrite the first ref | |
4132 | * of this inode and no new refs were added for the current | |
766702ef AB |
4133 | * inode. Unlinking does not mean that the inode is deleted in |
4134 | * all cases. There may still be links to this inode in other | |
4135 | * places. | |
31db9f7c | 4136 | */ |
1f4692da | 4137 | if (is_orphan) { |
31db9f7c AB |
4138 | ret = send_unlink(sctx, valid_path); |
4139 | if (ret < 0) | |
4140 | goto out; | |
4141 | } | |
4142 | } | |
4143 | ||
4144 | /* | |
4145 | * We did collect all parent dirs where cur_inode was once located. We | |
4146 | * now go through all these dirs and check if they are pending for | |
4147 | * deletion and if it's finally possible to perform the rmdir now. | |
4148 | * We also update the inode stats of the parent dirs here. | |
4149 | */ | |
ba5e8f2e | 4150 | list_for_each_entry(cur, &check_dirs, list) { |
766702ef AB |
4151 | /* |
4152 | * In case we had refs into dirs that were not processed yet, | |
4153 | * we don't need to do the utime and rmdir logic for these dirs. | |
4154 | * The dir will be processed later. | |
4155 | */ | |
ba5e8f2e | 4156 | if (cur->dir > sctx->cur_ino) |
31db9f7c AB |
4157 | continue; |
4158 | ||
ba5e8f2e | 4159 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4160 | if (ret < 0) |
4161 | goto out; | |
4162 | ||
4163 | if (ret == inode_state_did_create || | |
4164 | ret == inode_state_no_change) { | |
4165 | /* TODO delayed utimes */ | |
ba5e8f2e | 4166 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4167 | if (ret < 0) |
4168 | goto out; | |
29d6d30f FM |
4169 | } else if (ret == inode_state_did_delete && |
4170 | cur->dir != last_dir_ino_rm) { | |
9dc44214 FM |
4171 | ret = can_rmdir(sctx, cur->dir, cur->dir_gen, |
4172 | sctx->cur_ino); | |
31db9f7c AB |
4173 | if (ret < 0) |
4174 | goto out; | |
4175 | if (ret) { | |
ba5e8f2e JB |
4176 | ret = get_cur_path(sctx, cur->dir, |
4177 | cur->dir_gen, valid_path); | |
31db9f7c AB |
4178 | if (ret < 0) |
4179 | goto out; | |
4180 | ret = send_rmdir(sctx, valid_path); | |
4181 | if (ret < 0) | |
4182 | goto out; | |
29d6d30f | 4183 | last_dir_ino_rm = cur->dir; |
31db9f7c AB |
4184 | } |
4185 | } | |
4186 | } | |
4187 | ||
31db9f7c AB |
4188 | ret = 0; |
4189 | ||
4190 | out: | |
ba5e8f2e | 4191 | __free_recorded_refs(&check_dirs); |
31db9f7c | 4192 | free_recorded_refs(sctx); |
924794c9 | 4193 | fs_path_free(valid_path); |
31db9f7c AB |
4194 | return ret; |
4195 | } | |
4196 | ||
a0357511 NB |
4197 | static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name, |
4198 | void *ctx, struct list_head *refs) | |
31db9f7c AB |
4199 | { |
4200 | int ret = 0; | |
4201 | struct send_ctx *sctx = ctx; | |
4202 | struct fs_path *p; | |
4203 | u64 gen; | |
4204 | ||
924794c9 | 4205 | p = fs_path_alloc(); |
31db9f7c AB |
4206 | if (!p) |
4207 | return -ENOMEM; | |
4208 | ||
a4d96d62 | 4209 | ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, |
85a7b33b | 4210 | NULL, NULL); |
31db9f7c AB |
4211 | if (ret < 0) |
4212 | goto out; | |
4213 | ||
31db9f7c AB |
4214 | ret = get_cur_path(sctx, dir, gen, p); |
4215 | if (ret < 0) | |
4216 | goto out; | |
4217 | ret = fs_path_add_path(p, name); | |
4218 | if (ret < 0) | |
4219 | goto out; | |
4220 | ||
a4d96d62 | 4221 | ret = __record_ref(refs, dir, gen, p); |
31db9f7c AB |
4222 | |
4223 | out: | |
4224 | if (ret) | |
924794c9 | 4225 | fs_path_free(p); |
31db9f7c AB |
4226 | return ret; |
4227 | } | |
4228 | ||
a4d96d62 LB |
4229 | static int __record_new_ref(int num, u64 dir, int index, |
4230 | struct fs_path *name, | |
4231 | void *ctx) | |
4232 | { | |
4233 | struct send_ctx *sctx = ctx; | |
a0357511 | 4234 | return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs); |
a4d96d62 LB |
4235 | } |
4236 | ||
4237 | ||
31db9f7c AB |
4238 | static int __record_deleted_ref(int num, u64 dir, int index, |
4239 | struct fs_path *name, | |
4240 | void *ctx) | |
4241 | { | |
31db9f7c | 4242 | struct send_ctx *sctx = ctx; |
a0357511 NB |
4243 | return record_ref(sctx->parent_root, dir, name, ctx, |
4244 | &sctx->deleted_refs); | |
31db9f7c AB |
4245 | } |
4246 | ||
4247 | static int record_new_ref(struct send_ctx *sctx) | |
4248 | { | |
4249 | int ret; | |
4250 | ||
924794c9 TI |
4251 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
4252 | sctx->cmp_key, 0, __record_new_ref, sctx); | |
31db9f7c AB |
4253 | if (ret < 0) |
4254 | goto out; | |
4255 | ret = 0; | |
4256 | ||
4257 | out: | |
4258 | return ret; | |
4259 | } | |
4260 | ||
4261 | static int record_deleted_ref(struct send_ctx *sctx) | |
4262 | { | |
4263 | int ret; | |
4264 | ||
924794c9 TI |
4265 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
4266 | sctx->cmp_key, 0, __record_deleted_ref, sctx); | |
31db9f7c AB |
4267 | if (ret < 0) |
4268 | goto out; | |
4269 | ret = 0; | |
4270 | ||
4271 | out: | |
4272 | return ret; | |
4273 | } | |
4274 | ||
4275 | struct find_ref_ctx { | |
4276 | u64 dir; | |
ba5e8f2e JB |
4277 | u64 dir_gen; |
4278 | struct btrfs_root *root; | |
31db9f7c AB |
4279 | struct fs_path *name; |
4280 | int found_idx; | |
4281 | }; | |
4282 | ||
4283 | static int __find_iref(int num, u64 dir, int index, | |
4284 | struct fs_path *name, | |
4285 | void *ctx_) | |
4286 | { | |
4287 | struct find_ref_ctx *ctx = ctx_; | |
ba5e8f2e JB |
4288 | u64 dir_gen; |
4289 | int ret; | |
31db9f7c AB |
4290 | |
4291 | if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && | |
4292 | strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { | |
ba5e8f2e JB |
4293 | /* |
4294 | * To avoid doing extra lookups we'll only do this if everything | |
4295 | * else matches. | |
4296 | */ | |
4297 | ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, | |
4298 | NULL, NULL, NULL); | |
4299 | if (ret) | |
4300 | return ret; | |
4301 | if (dir_gen != ctx->dir_gen) | |
4302 | return 0; | |
31db9f7c AB |
4303 | ctx->found_idx = num; |
4304 | return 1; | |
4305 | } | |
4306 | return 0; | |
4307 | } | |
4308 | ||
924794c9 | 4309 | static int find_iref(struct btrfs_root *root, |
31db9f7c AB |
4310 | struct btrfs_path *path, |
4311 | struct btrfs_key *key, | |
ba5e8f2e | 4312 | u64 dir, u64 dir_gen, struct fs_path *name) |
31db9f7c AB |
4313 | { |
4314 | int ret; | |
4315 | struct find_ref_ctx ctx; | |
4316 | ||
4317 | ctx.dir = dir; | |
4318 | ctx.name = name; | |
ba5e8f2e | 4319 | ctx.dir_gen = dir_gen; |
31db9f7c | 4320 | ctx.found_idx = -1; |
ba5e8f2e | 4321 | ctx.root = root; |
31db9f7c | 4322 | |
924794c9 | 4323 | ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); |
31db9f7c AB |
4324 | if (ret < 0) |
4325 | return ret; | |
4326 | ||
4327 | if (ctx.found_idx == -1) | |
4328 | return -ENOENT; | |
4329 | ||
4330 | return ctx.found_idx; | |
4331 | } | |
4332 | ||
4333 | static int __record_changed_new_ref(int num, u64 dir, int index, | |
4334 | struct fs_path *name, | |
4335 | void *ctx) | |
4336 | { | |
ba5e8f2e | 4337 | u64 dir_gen; |
31db9f7c AB |
4338 | int ret; |
4339 | struct send_ctx *sctx = ctx; | |
4340 | ||
ba5e8f2e JB |
4341 | ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, |
4342 | NULL, NULL, NULL); | |
4343 | if (ret) | |
4344 | return ret; | |
4345 | ||
924794c9 | 4346 | ret = find_iref(sctx->parent_root, sctx->right_path, |
ba5e8f2e | 4347 | sctx->cmp_key, dir, dir_gen, name); |
31db9f7c AB |
4348 | if (ret == -ENOENT) |
4349 | ret = __record_new_ref(num, dir, index, name, sctx); | |
4350 | else if (ret > 0) | |
4351 | ret = 0; | |
4352 | ||
4353 | return ret; | |
4354 | } | |
4355 | ||
4356 | static int __record_changed_deleted_ref(int num, u64 dir, int index, | |
4357 | struct fs_path *name, | |
4358 | void *ctx) | |
4359 | { | |
ba5e8f2e | 4360 | u64 dir_gen; |
31db9f7c AB |
4361 | int ret; |
4362 | struct send_ctx *sctx = ctx; | |
4363 | ||
ba5e8f2e JB |
4364 | ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, |
4365 | NULL, NULL, NULL); | |
4366 | if (ret) | |
4367 | return ret; | |
4368 | ||
924794c9 | 4369 | ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, |
ba5e8f2e | 4370 | dir, dir_gen, name); |
31db9f7c AB |
4371 | if (ret == -ENOENT) |
4372 | ret = __record_deleted_ref(num, dir, index, name, sctx); | |
4373 | else if (ret > 0) | |
4374 | ret = 0; | |
4375 | ||
4376 | return ret; | |
4377 | } | |
4378 | ||
4379 | static int record_changed_ref(struct send_ctx *sctx) | |
4380 | { | |
4381 | int ret = 0; | |
4382 | ||
924794c9 | 4383 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
31db9f7c AB |
4384 | sctx->cmp_key, 0, __record_changed_new_ref, sctx); |
4385 | if (ret < 0) | |
4386 | goto out; | |
924794c9 | 4387 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
4388 | sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); |
4389 | if (ret < 0) | |
4390 | goto out; | |
4391 | ret = 0; | |
4392 | ||
4393 | out: | |
4394 | return ret; | |
4395 | } | |
4396 | ||
4397 | /* | |
4398 | * Record and process all refs at once. Needed when an inode changes the | |
4399 | * generation number, which means that it was deleted and recreated. | |
4400 | */ | |
4401 | static int process_all_refs(struct send_ctx *sctx, | |
4402 | enum btrfs_compare_tree_result cmd) | |
4403 | { | |
4404 | int ret; | |
4405 | struct btrfs_root *root; | |
4406 | struct btrfs_path *path; | |
4407 | struct btrfs_key key; | |
4408 | struct btrfs_key found_key; | |
4409 | struct extent_buffer *eb; | |
4410 | int slot; | |
4411 | iterate_inode_ref_t cb; | |
9f03740a | 4412 | int pending_move = 0; |
31db9f7c AB |
4413 | |
4414 | path = alloc_path_for_send(); | |
4415 | if (!path) | |
4416 | return -ENOMEM; | |
4417 | ||
4418 | if (cmd == BTRFS_COMPARE_TREE_NEW) { | |
4419 | root = sctx->send_root; | |
4420 | cb = __record_new_ref; | |
4421 | } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { | |
4422 | root = sctx->parent_root; | |
4423 | cb = __record_deleted_ref; | |
4424 | } else { | |
4d1a63b2 DS |
4425 | btrfs_err(sctx->send_root->fs_info, |
4426 | "Wrong command %d in process_all_refs", cmd); | |
4427 | ret = -EINVAL; | |
4428 | goto out; | |
31db9f7c AB |
4429 | } |
4430 | ||
4431 | key.objectid = sctx->cmp_key->objectid; | |
4432 | key.type = BTRFS_INODE_REF_KEY; | |
4433 | key.offset = 0; | |
dff6d0ad FDBM |
4434 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4435 | if (ret < 0) | |
4436 | goto out; | |
31db9f7c | 4437 | |
dff6d0ad | 4438 | while (1) { |
31db9f7c AB |
4439 | eb = path->nodes[0]; |
4440 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4441 | if (slot >= btrfs_header_nritems(eb)) { |
4442 | ret = btrfs_next_leaf(root, path); | |
4443 | if (ret < 0) | |
4444 | goto out; | |
4445 | else if (ret > 0) | |
4446 | break; | |
4447 | continue; | |
4448 | } | |
4449 | ||
31db9f7c AB |
4450 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
4451 | ||
4452 | if (found_key.objectid != key.objectid || | |
96b5bd77 JS |
4453 | (found_key.type != BTRFS_INODE_REF_KEY && |
4454 | found_key.type != BTRFS_INODE_EXTREF_KEY)) | |
31db9f7c | 4455 | break; |
31db9f7c | 4456 | |
924794c9 | 4457 | ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); |
31db9f7c AB |
4458 | if (ret < 0) |
4459 | goto out; | |
4460 | ||
dff6d0ad | 4461 | path->slots[0]++; |
31db9f7c | 4462 | } |
e938c8ad | 4463 | btrfs_release_path(path); |
31db9f7c | 4464 | |
3dc09ec8 JB |
4465 | /* |
4466 | * We don't actually care about pending_move as we are simply | |
4467 | * re-creating this inode and will be rename'ing it into place once we | |
4468 | * rename the parent directory. | |
4469 | */ | |
9f03740a | 4470 | ret = process_recorded_refs(sctx, &pending_move); |
31db9f7c AB |
4471 | out: |
4472 | btrfs_free_path(path); | |
4473 | return ret; | |
4474 | } | |
4475 | ||
4476 | static int send_set_xattr(struct send_ctx *sctx, | |
4477 | struct fs_path *path, | |
4478 | const char *name, int name_len, | |
4479 | const char *data, int data_len) | |
4480 | { | |
4481 | int ret = 0; | |
4482 | ||
4483 | ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); | |
4484 | if (ret < 0) | |
4485 | goto out; | |
4486 | ||
4487 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4488 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4489 | TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); | |
4490 | ||
4491 | ret = send_cmd(sctx); | |
4492 | ||
4493 | tlv_put_failure: | |
4494 | out: | |
4495 | return ret; | |
4496 | } | |
4497 | ||
4498 | static int send_remove_xattr(struct send_ctx *sctx, | |
4499 | struct fs_path *path, | |
4500 | const char *name, int name_len) | |
4501 | { | |
4502 | int ret = 0; | |
4503 | ||
4504 | ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); | |
4505 | if (ret < 0) | |
4506 | goto out; | |
4507 | ||
4508 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4509 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4510 | ||
4511 | ret = send_cmd(sctx); | |
4512 | ||
4513 | tlv_put_failure: | |
4514 | out: | |
4515 | return ret; | |
4516 | } | |
4517 | ||
4518 | static int __process_new_xattr(int num, struct btrfs_key *di_key, | |
4519 | const char *name, int name_len, | |
4520 | const char *data, int data_len, | |
4521 | u8 type, void *ctx) | |
4522 | { | |
4523 | int ret; | |
4524 | struct send_ctx *sctx = ctx; | |
4525 | struct fs_path *p; | |
2211d5ba | 4526 | struct posix_acl_xattr_header dummy_acl; |
31db9f7c | 4527 | |
924794c9 | 4528 | p = fs_path_alloc(); |
31db9f7c AB |
4529 | if (!p) |
4530 | return -ENOMEM; | |
4531 | ||
4532 | /* | |
01327610 | 4533 | * This hack is needed because empty acls are stored as zero byte |
31db9f7c | 4534 | * data in xattrs. Problem with that is, that receiving these zero byte |
01327610 | 4535 | * acls will fail later. To fix this, we send a dummy acl list that |
31db9f7c AB |
4536 | * only contains the version number and no entries. |
4537 | */ | |
4538 | if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || | |
4539 | !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { | |
4540 | if (data_len == 0) { | |
4541 | dummy_acl.a_version = | |
4542 | cpu_to_le32(POSIX_ACL_XATTR_VERSION); | |
4543 | data = (char *)&dummy_acl; | |
4544 | data_len = sizeof(dummy_acl); | |
4545 | } | |
4546 | } | |
4547 | ||
4548 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4549 | if (ret < 0) | |
4550 | goto out; | |
4551 | ||
4552 | ret = send_set_xattr(sctx, p, name, name_len, data, data_len); | |
4553 | ||
4554 | out: | |
924794c9 | 4555 | fs_path_free(p); |
31db9f7c AB |
4556 | return ret; |
4557 | } | |
4558 | ||
4559 | static int __process_deleted_xattr(int num, struct btrfs_key *di_key, | |
4560 | const char *name, int name_len, | |
4561 | const char *data, int data_len, | |
4562 | u8 type, void *ctx) | |
4563 | { | |
4564 | int ret; | |
4565 | struct send_ctx *sctx = ctx; | |
4566 | struct fs_path *p; | |
4567 | ||
924794c9 | 4568 | p = fs_path_alloc(); |
31db9f7c AB |
4569 | if (!p) |
4570 | return -ENOMEM; | |
4571 | ||
4572 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4573 | if (ret < 0) | |
4574 | goto out; | |
4575 | ||
4576 | ret = send_remove_xattr(sctx, p, name, name_len); | |
4577 | ||
4578 | out: | |
924794c9 | 4579 | fs_path_free(p); |
31db9f7c AB |
4580 | return ret; |
4581 | } | |
4582 | ||
4583 | static int process_new_xattr(struct send_ctx *sctx) | |
4584 | { | |
4585 | int ret = 0; | |
4586 | ||
924794c9 | 4587 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4588 | __process_new_xattr, sctx); |
31db9f7c AB |
4589 | |
4590 | return ret; | |
4591 | } | |
4592 | ||
4593 | static int process_deleted_xattr(struct send_ctx *sctx) | |
4594 | { | |
e2c89907 | 4595 | return iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4596 | __process_deleted_xattr, sctx); |
31db9f7c AB |
4597 | } |
4598 | ||
4599 | struct find_xattr_ctx { | |
4600 | const char *name; | |
4601 | int name_len; | |
4602 | int found_idx; | |
4603 | char *found_data; | |
4604 | int found_data_len; | |
4605 | }; | |
4606 | ||
4607 | static int __find_xattr(int num, struct btrfs_key *di_key, | |
4608 | const char *name, int name_len, | |
4609 | const char *data, int data_len, | |
4610 | u8 type, void *vctx) | |
4611 | { | |
4612 | struct find_xattr_ctx *ctx = vctx; | |
4613 | ||
4614 | if (name_len == ctx->name_len && | |
4615 | strncmp(name, ctx->name, name_len) == 0) { | |
4616 | ctx->found_idx = num; | |
4617 | ctx->found_data_len = data_len; | |
e780b0d1 | 4618 | ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); |
31db9f7c AB |
4619 | if (!ctx->found_data) |
4620 | return -ENOMEM; | |
31db9f7c AB |
4621 | return 1; |
4622 | } | |
4623 | return 0; | |
4624 | } | |
4625 | ||
924794c9 | 4626 | static int find_xattr(struct btrfs_root *root, |
31db9f7c AB |
4627 | struct btrfs_path *path, |
4628 | struct btrfs_key *key, | |
4629 | const char *name, int name_len, | |
4630 | char **data, int *data_len) | |
4631 | { | |
4632 | int ret; | |
4633 | struct find_xattr_ctx ctx; | |
4634 | ||
4635 | ctx.name = name; | |
4636 | ctx.name_len = name_len; | |
4637 | ctx.found_idx = -1; | |
4638 | ctx.found_data = NULL; | |
4639 | ctx.found_data_len = 0; | |
4640 | ||
a0357511 | 4641 | ret = iterate_dir_item(root, path, __find_xattr, &ctx); |
31db9f7c AB |
4642 | if (ret < 0) |
4643 | return ret; | |
4644 | ||
4645 | if (ctx.found_idx == -1) | |
4646 | return -ENOENT; | |
4647 | if (data) { | |
4648 | *data = ctx.found_data; | |
4649 | *data_len = ctx.found_data_len; | |
4650 | } else { | |
4651 | kfree(ctx.found_data); | |
4652 | } | |
4653 | return ctx.found_idx; | |
4654 | } | |
4655 | ||
4656 | ||
4657 | static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, | |
4658 | const char *name, int name_len, | |
4659 | const char *data, int data_len, | |
4660 | u8 type, void *ctx) | |
4661 | { | |
4662 | int ret; | |
4663 | struct send_ctx *sctx = ctx; | |
4664 | char *found_data = NULL; | |
4665 | int found_data_len = 0; | |
31db9f7c | 4666 | |
924794c9 TI |
4667 | ret = find_xattr(sctx->parent_root, sctx->right_path, |
4668 | sctx->cmp_key, name, name_len, &found_data, | |
4669 | &found_data_len); | |
31db9f7c AB |
4670 | if (ret == -ENOENT) { |
4671 | ret = __process_new_xattr(num, di_key, name, name_len, data, | |
4672 | data_len, type, ctx); | |
4673 | } else if (ret >= 0) { | |
4674 | if (data_len != found_data_len || | |
4675 | memcmp(data, found_data, data_len)) { | |
4676 | ret = __process_new_xattr(num, di_key, name, name_len, | |
4677 | data, data_len, type, ctx); | |
4678 | } else { | |
4679 | ret = 0; | |
4680 | } | |
4681 | } | |
4682 | ||
4683 | kfree(found_data); | |
31db9f7c AB |
4684 | return ret; |
4685 | } | |
4686 | ||
4687 | static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, | |
4688 | const char *name, int name_len, | |
4689 | const char *data, int data_len, | |
4690 | u8 type, void *ctx) | |
4691 | { | |
4692 | int ret; | |
4693 | struct send_ctx *sctx = ctx; | |
4694 | ||
924794c9 TI |
4695 | ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, |
4696 | name, name_len, NULL, NULL); | |
31db9f7c AB |
4697 | if (ret == -ENOENT) |
4698 | ret = __process_deleted_xattr(num, di_key, name, name_len, data, | |
4699 | data_len, type, ctx); | |
4700 | else if (ret >= 0) | |
4701 | ret = 0; | |
4702 | ||
4703 | return ret; | |
4704 | } | |
4705 | ||
4706 | static int process_changed_xattr(struct send_ctx *sctx) | |
4707 | { | |
4708 | int ret = 0; | |
4709 | ||
924794c9 | 4710 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4711 | __process_changed_new_xattr, sctx); |
31db9f7c AB |
4712 | if (ret < 0) |
4713 | goto out; | |
924794c9 | 4714 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4715 | __process_changed_deleted_xattr, sctx); |
31db9f7c AB |
4716 | |
4717 | out: | |
4718 | return ret; | |
4719 | } | |
4720 | ||
4721 | static int process_all_new_xattrs(struct send_ctx *sctx) | |
4722 | { | |
4723 | int ret; | |
4724 | struct btrfs_root *root; | |
4725 | struct btrfs_path *path; | |
4726 | struct btrfs_key key; | |
4727 | struct btrfs_key found_key; | |
4728 | struct extent_buffer *eb; | |
4729 | int slot; | |
4730 | ||
4731 | path = alloc_path_for_send(); | |
4732 | if (!path) | |
4733 | return -ENOMEM; | |
4734 | ||
4735 | root = sctx->send_root; | |
4736 | ||
4737 | key.objectid = sctx->cmp_key->objectid; | |
4738 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4739 | key.offset = 0; | |
dff6d0ad FDBM |
4740 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4741 | if (ret < 0) | |
4742 | goto out; | |
31db9f7c | 4743 | |
dff6d0ad | 4744 | while (1) { |
31db9f7c AB |
4745 | eb = path->nodes[0]; |
4746 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4747 | if (slot >= btrfs_header_nritems(eb)) { |
4748 | ret = btrfs_next_leaf(root, path); | |
4749 | if (ret < 0) { | |
4750 | goto out; | |
4751 | } else if (ret > 0) { | |
4752 | ret = 0; | |
4753 | break; | |
4754 | } | |
4755 | continue; | |
4756 | } | |
31db9f7c | 4757 | |
dff6d0ad | 4758 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
31db9f7c AB |
4759 | if (found_key.objectid != key.objectid || |
4760 | found_key.type != key.type) { | |
4761 | ret = 0; | |
4762 | goto out; | |
4763 | } | |
4764 | ||
a0357511 | 4765 | ret = iterate_dir_item(root, path, __process_new_xattr, sctx); |
31db9f7c AB |
4766 | if (ret < 0) |
4767 | goto out; | |
4768 | ||
dff6d0ad | 4769 | path->slots[0]++; |
31db9f7c AB |
4770 | } |
4771 | ||
4772 | out: | |
4773 | btrfs_free_path(path); | |
4774 | return ret; | |
4775 | } | |
4776 | ||
ed259095 JB |
4777 | static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len) |
4778 | { | |
4779 | struct btrfs_root *root = sctx->send_root; | |
4780 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4781 | struct inode *inode; | |
4782 | struct page *page; | |
4783 | char *addr; | |
4784 | struct btrfs_key key; | |
09cbfeaf | 4785 | pgoff_t index = offset >> PAGE_SHIFT; |
ed259095 | 4786 | pgoff_t last_index; |
09cbfeaf | 4787 | unsigned pg_offset = offset & ~PAGE_MASK; |
ed259095 JB |
4788 | ssize_t ret = 0; |
4789 | ||
4790 | key.objectid = sctx->cur_ino; | |
4791 | key.type = BTRFS_INODE_ITEM_KEY; | |
4792 | key.offset = 0; | |
4793 | ||
4794 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
4795 | if (IS_ERR(inode)) | |
4796 | return PTR_ERR(inode); | |
4797 | ||
4798 | if (offset + len > i_size_read(inode)) { | |
4799 | if (offset > i_size_read(inode)) | |
4800 | len = 0; | |
4801 | else | |
4802 | len = offset - i_size_read(inode); | |
4803 | } | |
4804 | if (len == 0) | |
4805 | goto out; | |
4806 | ||
09cbfeaf | 4807 | last_index = (offset + len - 1) >> PAGE_SHIFT; |
2131bcd3 LB |
4808 | |
4809 | /* initial readahead */ | |
4810 | memset(&sctx->ra, 0, sizeof(struct file_ra_state)); | |
4811 | file_ra_state_init(&sctx->ra, inode->i_mapping); | |
2131bcd3 | 4812 | |
ed259095 JB |
4813 | while (index <= last_index) { |
4814 | unsigned cur_len = min_t(unsigned, len, | |
09cbfeaf | 4815 | PAGE_SIZE - pg_offset); |
eef16ba2 KH |
4816 | |
4817 | page = find_lock_page(inode->i_mapping, index); | |
ed259095 | 4818 | if (!page) { |
eef16ba2 KH |
4819 | page_cache_sync_readahead(inode->i_mapping, &sctx->ra, |
4820 | NULL, index, last_index + 1 - index); | |
4821 | ||
4822 | page = find_or_create_page(inode->i_mapping, index, | |
4823 | GFP_KERNEL); | |
4824 | if (!page) { | |
4825 | ret = -ENOMEM; | |
4826 | break; | |
4827 | } | |
4828 | } | |
4829 | ||
4830 | if (PageReadahead(page)) { | |
4831 | page_cache_async_readahead(inode->i_mapping, &sctx->ra, | |
4832 | NULL, page, index, last_index + 1 - index); | |
ed259095 JB |
4833 | } |
4834 | ||
4835 | if (!PageUptodate(page)) { | |
4836 | btrfs_readpage(NULL, page); | |
4837 | lock_page(page); | |
4838 | if (!PageUptodate(page)) { | |
4839 | unlock_page(page); | |
09cbfeaf | 4840 | put_page(page); |
ed259095 JB |
4841 | ret = -EIO; |
4842 | break; | |
4843 | } | |
4844 | } | |
4845 | ||
4846 | addr = kmap(page); | |
4847 | memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len); | |
4848 | kunmap(page); | |
4849 | unlock_page(page); | |
09cbfeaf | 4850 | put_page(page); |
ed259095 JB |
4851 | index++; |
4852 | pg_offset = 0; | |
4853 | len -= cur_len; | |
4854 | ret += cur_len; | |
4855 | } | |
4856 | out: | |
4857 | iput(inode); | |
4858 | return ret; | |
4859 | } | |
4860 | ||
31db9f7c AB |
4861 | /* |
4862 | * Read some bytes from the current inode/file and send a write command to | |
4863 | * user space. | |
4864 | */ | |
4865 | static int send_write(struct send_ctx *sctx, u64 offset, u32 len) | |
4866 | { | |
04ab956e | 4867 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
4868 | int ret = 0; |
4869 | struct fs_path *p; | |
ed259095 | 4870 | ssize_t num_read = 0; |
31db9f7c | 4871 | |
924794c9 | 4872 | p = fs_path_alloc(); |
31db9f7c AB |
4873 | if (!p) |
4874 | return -ENOMEM; | |
4875 | ||
04ab956e | 4876 | btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len); |
31db9f7c | 4877 | |
ed259095 JB |
4878 | num_read = fill_read_buf(sctx, offset, len); |
4879 | if (num_read <= 0) { | |
4880 | if (num_read < 0) | |
4881 | ret = num_read; | |
31db9f7c | 4882 | goto out; |
ed259095 | 4883 | } |
31db9f7c AB |
4884 | |
4885 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
4886 | if (ret < 0) | |
4887 | goto out; | |
4888 | ||
4889 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4890 | if (ret < 0) | |
4891 | goto out; | |
4892 | ||
4893 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4894 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
e938c8ad | 4895 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read); |
31db9f7c AB |
4896 | |
4897 | ret = send_cmd(sctx); | |
4898 | ||
4899 | tlv_put_failure: | |
4900 | out: | |
924794c9 | 4901 | fs_path_free(p); |
31db9f7c AB |
4902 | if (ret < 0) |
4903 | return ret; | |
e938c8ad | 4904 | return num_read; |
31db9f7c AB |
4905 | } |
4906 | ||
4907 | /* | |
4908 | * Send a clone command to user space. | |
4909 | */ | |
4910 | static int send_clone(struct send_ctx *sctx, | |
4911 | u64 offset, u32 len, | |
4912 | struct clone_root *clone_root) | |
4913 | { | |
4914 | int ret = 0; | |
31db9f7c AB |
4915 | struct fs_path *p; |
4916 | u64 gen; | |
4917 | ||
04ab956e JM |
4918 | btrfs_debug(sctx->send_root->fs_info, |
4919 | "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu", | |
4920 | offset, len, clone_root->root->objectid, clone_root->ino, | |
4921 | clone_root->offset); | |
31db9f7c | 4922 | |
924794c9 | 4923 | p = fs_path_alloc(); |
31db9f7c AB |
4924 | if (!p) |
4925 | return -ENOMEM; | |
4926 | ||
4927 | ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); | |
4928 | if (ret < 0) | |
4929 | goto out; | |
4930 | ||
4931 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4932 | if (ret < 0) | |
4933 | goto out; | |
4934 | ||
4935 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4936 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); | |
4937 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4938 | ||
e938c8ad | 4939 | if (clone_root->root == sctx->send_root) { |
31db9f7c | 4940 | ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, |
85a7b33b | 4941 | &gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
4942 | if (ret < 0) |
4943 | goto out; | |
4944 | ret = get_cur_path(sctx, clone_root->ino, gen, p); | |
4945 | } else { | |
924794c9 | 4946 | ret = get_inode_path(clone_root->root, clone_root->ino, p); |
31db9f7c AB |
4947 | } |
4948 | if (ret < 0) | |
4949 | goto out; | |
4950 | ||
37b8d27d JB |
4951 | /* |
4952 | * If the parent we're using has a received_uuid set then use that as | |
4953 | * our clone source as that is what we will look for when doing a | |
4954 | * receive. | |
4955 | * | |
4956 | * This covers the case that we create a snapshot off of a received | |
4957 | * subvolume and then use that as the parent and try to receive on a | |
4958 | * different host. | |
4959 | */ | |
4960 | if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid)) | |
4961 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4962 | clone_root->root->root_item.received_uuid); | |
4963 | else | |
4964 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4965 | clone_root->root->root_item.uuid); | |
31db9f7c | 4966 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 4967 | le64_to_cpu(clone_root->root->root_item.ctransid)); |
31db9f7c AB |
4968 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); |
4969 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, | |
4970 | clone_root->offset); | |
4971 | ||
4972 | ret = send_cmd(sctx); | |
4973 | ||
4974 | tlv_put_failure: | |
4975 | out: | |
924794c9 | 4976 | fs_path_free(p); |
31db9f7c AB |
4977 | return ret; |
4978 | } | |
4979 | ||
cb95e7bf MF |
4980 | /* |
4981 | * Send an update extent command to user space. | |
4982 | */ | |
4983 | static int send_update_extent(struct send_ctx *sctx, | |
4984 | u64 offset, u32 len) | |
4985 | { | |
4986 | int ret = 0; | |
4987 | struct fs_path *p; | |
4988 | ||
924794c9 | 4989 | p = fs_path_alloc(); |
cb95e7bf MF |
4990 | if (!p) |
4991 | return -ENOMEM; | |
4992 | ||
4993 | ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); | |
4994 | if (ret < 0) | |
4995 | goto out; | |
4996 | ||
4997 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4998 | if (ret < 0) | |
4999 | goto out; | |
5000 | ||
5001 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5002 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5003 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); | |
5004 | ||
5005 | ret = send_cmd(sctx); | |
5006 | ||
5007 | tlv_put_failure: | |
5008 | out: | |
924794c9 | 5009 | fs_path_free(p); |
cb95e7bf MF |
5010 | return ret; |
5011 | } | |
5012 | ||
16e7549f JB |
5013 | static int send_hole(struct send_ctx *sctx, u64 end) |
5014 | { | |
5015 | struct fs_path *p = NULL; | |
5016 | u64 offset = sctx->cur_inode_last_extent; | |
5017 | u64 len; | |
5018 | int ret = 0; | |
5019 | ||
74f880be FM |
5020 | /* |
5021 | * A hole that starts at EOF or beyond it. Since we do not yet support | |
5022 | * fallocate (for extent preallocation and hole punching), sending a | |
5023 | * write of zeroes starting at EOF or beyond would later require issuing | |
5024 | * a truncate operation which would undo the write and achieve nothing. | |
5025 | */ | |
5026 | if (offset >= sctx->cur_inode_size) | |
5027 | return 0; | |
5028 | ||
80ebbef6 FM |
5029 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) |
5030 | return send_update_extent(sctx, offset, end - offset); | |
5031 | ||
16e7549f JB |
5032 | p = fs_path_alloc(); |
5033 | if (!p) | |
5034 | return -ENOMEM; | |
c715e155 FM |
5035 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); |
5036 | if (ret < 0) | |
5037 | goto tlv_put_failure; | |
16e7549f JB |
5038 | memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE); |
5039 | while (offset < end) { | |
5040 | len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE); | |
5041 | ||
5042 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
16e7549f JB |
5043 | if (ret < 0) |
5044 | break; | |
5045 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5046 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5047 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len); | |
5048 | ret = send_cmd(sctx); | |
5049 | if (ret < 0) | |
5050 | break; | |
5051 | offset += len; | |
5052 | } | |
5053 | tlv_put_failure: | |
5054 | fs_path_free(p); | |
5055 | return ret; | |
5056 | } | |
5057 | ||
d906d49f FM |
5058 | static int send_extent_data(struct send_ctx *sctx, |
5059 | const u64 offset, | |
5060 | const u64 len) | |
5061 | { | |
5062 | u64 sent = 0; | |
5063 | ||
5064 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) | |
5065 | return send_update_extent(sctx, offset, len); | |
5066 | ||
5067 | while (sent < len) { | |
5068 | u64 size = len - sent; | |
5069 | int ret; | |
5070 | ||
5071 | if (size > BTRFS_SEND_READ_SIZE) | |
5072 | size = BTRFS_SEND_READ_SIZE; | |
5073 | ret = send_write(sctx, offset + sent, size); | |
5074 | if (ret < 0) | |
5075 | return ret; | |
5076 | if (!ret) | |
5077 | break; | |
5078 | sent += ret; | |
5079 | } | |
5080 | return 0; | |
5081 | } | |
5082 | ||
5083 | static int clone_range(struct send_ctx *sctx, | |
5084 | struct clone_root *clone_root, | |
5085 | const u64 disk_byte, | |
5086 | u64 data_offset, | |
5087 | u64 offset, | |
5088 | u64 len) | |
5089 | { | |
5090 | struct btrfs_path *path; | |
5091 | struct btrfs_key key; | |
5092 | int ret; | |
5093 | ||
72610b1b FM |
5094 | /* |
5095 | * Prevent cloning from a zero offset with a length matching the sector | |
5096 | * size because in some scenarios this will make the receiver fail. | |
5097 | * | |
5098 | * For example, if in the source filesystem the extent at offset 0 | |
5099 | * has a length of sectorsize and it was written using direct IO, then | |
5100 | * it can never be an inline extent (even if compression is enabled). | |
5101 | * Then this extent can be cloned in the original filesystem to a non | |
5102 | * zero file offset, but it may not be possible to clone in the | |
5103 | * destination filesystem because it can be inlined due to compression | |
5104 | * on the destination filesystem (as the receiver's write operations are | |
5105 | * always done using buffered IO). The same happens when the original | |
5106 | * filesystem does not have compression enabled but the destination | |
5107 | * filesystem has. | |
5108 | */ | |
5109 | if (clone_root->offset == 0 && | |
5110 | len == sctx->send_root->fs_info->sectorsize) | |
5111 | return send_extent_data(sctx, offset, len); | |
5112 | ||
d906d49f FM |
5113 | path = alloc_path_for_send(); |
5114 | if (!path) | |
5115 | return -ENOMEM; | |
5116 | ||
5117 | /* | |
5118 | * We can't send a clone operation for the entire range if we find | |
5119 | * extent items in the respective range in the source file that | |
5120 | * refer to different extents or if we find holes. | |
5121 | * So check for that and do a mix of clone and regular write/copy | |
5122 | * operations if needed. | |
5123 | * | |
5124 | * Example: | |
5125 | * | |
5126 | * mkfs.btrfs -f /dev/sda | |
5127 | * mount /dev/sda /mnt | |
5128 | * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo | |
5129 | * cp --reflink=always /mnt/foo /mnt/bar | |
5130 | * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo | |
5131 | * btrfs subvolume snapshot -r /mnt /mnt/snap | |
5132 | * | |
5133 | * If when we send the snapshot and we are processing file bar (which | |
5134 | * has a higher inode number than foo) we blindly send a clone operation | |
5135 | * for the [0, 100K[ range from foo to bar, the receiver ends up getting | |
5136 | * a file bar that matches the content of file foo - iow, doesn't match | |
5137 | * the content from bar in the original filesystem. | |
5138 | */ | |
5139 | key.objectid = clone_root->ino; | |
5140 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5141 | key.offset = clone_root->offset; | |
5142 | ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); | |
5143 | if (ret < 0) | |
5144 | goto out; | |
5145 | if (ret > 0 && path->slots[0] > 0) { | |
5146 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
5147 | if (key.objectid == clone_root->ino && | |
5148 | key.type == BTRFS_EXTENT_DATA_KEY) | |
5149 | path->slots[0]--; | |
5150 | } | |
5151 | ||
5152 | while (true) { | |
5153 | struct extent_buffer *leaf = path->nodes[0]; | |
5154 | int slot = path->slots[0]; | |
5155 | struct btrfs_file_extent_item *ei; | |
5156 | u8 type; | |
5157 | u64 ext_len; | |
5158 | u64 clone_len; | |
5159 | ||
5160 | if (slot >= btrfs_header_nritems(leaf)) { | |
5161 | ret = btrfs_next_leaf(clone_root->root, path); | |
5162 | if (ret < 0) | |
5163 | goto out; | |
5164 | else if (ret > 0) | |
5165 | break; | |
5166 | continue; | |
5167 | } | |
5168 | ||
5169 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5170 | ||
5171 | /* | |
5172 | * We might have an implicit trailing hole (NO_HOLES feature | |
5173 | * enabled). We deal with it after leaving this loop. | |
5174 | */ | |
5175 | if (key.objectid != clone_root->ino || | |
5176 | key.type != BTRFS_EXTENT_DATA_KEY) | |
5177 | break; | |
5178 | ||
5179 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5180 | type = btrfs_file_extent_type(leaf, ei); | |
5181 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
5182 | ext_len = btrfs_file_extent_inline_len(leaf, slot, ei); | |
09cbfeaf | 5183 | ext_len = PAGE_ALIGN(ext_len); |
d906d49f FM |
5184 | } else { |
5185 | ext_len = btrfs_file_extent_num_bytes(leaf, ei); | |
5186 | } | |
5187 | ||
5188 | if (key.offset + ext_len <= clone_root->offset) | |
5189 | goto next; | |
5190 | ||
5191 | if (key.offset > clone_root->offset) { | |
5192 | /* Implicit hole, NO_HOLES feature enabled. */ | |
5193 | u64 hole_len = key.offset - clone_root->offset; | |
5194 | ||
5195 | if (hole_len > len) | |
5196 | hole_len = len; | |
5197 | ret = send_extent_data(sctx, offset, hole_len); | |
5198 | if (ret < 0) | |
5199 | goto out; | |
5200 | ||
5201 | len -= hole_len; | |
5202 | if (len == 0) | |
5203 | break; | |
5204 | offset += hole_len; | |
5205 | clone_root->offset += hole_len; | |
5206 | data_offset += hole_len; | |
5207 | } | |
5208 | ||
5209 | if (key.offset >= clone_root->offset + len) | |
5210 | break; | |
5211 | ||
5212 | clone_len = min_t(u64, ext_len, len); | |
5213 | ||
5214 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && | |
5215 | btrfs_file_extent_offset(leaf, ei) == data_offset) | |
5216 | ret = send_clone(sctx, offset, clone_len, clone_root); | |
5217 | else | |
5218 | ret = send_extent_data(sctx, offset, clone_len); | |
5219 | ||
5220 | if (ret < 0) | |
5221 | goto out; | |
5222 | ||
5223 | len -= clone_len; | |
5224 | if (len == 0) | |
5225 | break; | |
5226 | offset += clone_len; | |
5227 | clone_root->offset += clone_len; | |
5228 | data_offset += clone_len; | |
5229 | next: | |
5230 | path->slots[0]++; | |
5231 | } | |
5232 | ||
5233 | if (len > 0) | |
5234 | ret = send_extent_data(sctx, offset, len); | |
5235 | else | |
5236 | ret = 0; | |
5237 | out: | |
5238 | btrfs_free_path(path); | |
5239 | return ret; | |
5240 | } | |
5241 | ||
31db9f7c AB |
5242 | static int send_write_or_clone(struct send_ctx *sctx, |
5243 | struct btrfs_path *path, | |
5244 | struct btrfs_key *key, | |
5245 | struct clone_root *clone_root) | |
5246 | { | |
5247 | int ret = 0; | |
5248 | struct btrfs_file_extent_item *ei; | |
5249 | u64 offset = key->offset; | |
31db9f7c | 5250 | u64 len; |
31db9f7c | 5251 | u8 type; |
28e5dd8f | 5252 | u64 bs = sctx->send_root->fs_info->sb->s_blocksize; |
31db9f7c AB |
5253 | |
5254 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5255 | struct btrfs_file_extent_item); | |
5256 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
74dd17fb | 5257 | if (type == BTRFS_FILE_EXTENT_INLINE) { |
514ac8ad CM |
5258 | len = btrfs_file_extent_inline_len(path->nodes[0], |
5259 | path->slots[0], ei); | |
74dd17fb CM |
5260 | /* |
5261 | * it is possible the inline item won't cover the whole page, | |
5262 | * but there may be items after this page. Make | |
5263 | * sure to send the whole thing | |
5264 | */ | |
09cbfeaf | 5265 | len = PAGE_ALIGN(len); |
74dd17fb | 5266 | } else { |
31db9f7c | 5267 | len = btrfs_file_extent_num_bytes(path->nodes[0], ei); |
74dd17fb | 5268 | } |
31db9f7c AB |
5269 | |
5270 | if (offset + len > sctx->cur_inode_size) | |
5271 | len = sctx->cur_inode_size - offset; | |
5272 | if (len == 0) { | |
5273 | ret = 0; | |
5274 | goto out; | |
5275 | } | |
5276 | ||
28e5dd8f | 5277 | if (clone_root && IS_ALIGNED(offset + len, bs)) { |
d906d49f FM |
5278 | u64 disk_byte; |
5279 | u64 data_offset; | |
5280 | ||
5281 | disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); | |
5282 | data_offset = btrfs_file_extent_offset(path->nodes[0], ei); | |
5283 | ret = clone_range(sctx, clone_root, disk_byte, data_offset, | |
5284 | offset, len); | |
cb95e7bf | 5285 | } else { |
d906d49f | 5286 | ret = send_extent_data(sctx, offset, len); |
31db9f7c | 5287 | } |
31db9f7c AB |
5288 | out: |
5289 | return ret; | |
5290 | } | |
5291 | ||
5292 | static int is_extent_unchanged(struct send_ctx *sctx, | |
5293 | struct btrfs_path *left_path, | |
5294 | struct btrfs_key *ekey) | |
5295 | { | |
5296 | int ret = 0; | |
5297 | struct btrfs_key key; | |
5298 | struct btrfs_path *path = NULL; | |
5299 | struct extent_buffer *eb; | |
5300 | int slot; | |
5301 | struct btrfs_key found_key; | |
5302 | struct btrfs_file_extent_item *ei; | |
5303 | u64 left_disknr; | |
5304 | u64 right_disknr; | |
5305 | u64 left_offset; | |
5306 | u64 right_offset; | |
5307 | u64 left_offset_fixed; | |
5308 | u64 left_len; | |
5309 | u64 right_len; | |
74dd17fb CM |
5310 | u64 left_gen; |
5311 | u64 right_gen; | |
31db9f7c AB |
5312 | u8 left_type; |
5313 | u8 right_type; | |
5314 | ||
5315 | path = alloc_path_for_send(); | |
5316 | if (!path) | |
5317 | return -ENOMEM; | |
5318 | ||
5319 | eb = left_path->nodes[0]; | |
5320 | slot = left_path->slots[0]; | |
31db9f7c AB |
5321 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
5322 | left_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
5323 | |
5324 | if (left_type != BTRFS_FILE_EXTENT_REG) { | |
5325 | ret = 0; | |
5326 | goto out; | |
5327 | } | |
74dd17fb CM |
5328 | left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5329 | left_len = btrfs_file_extent_num_bytes(eb, ei); | |
5330 | left_offset = btrfs_file_extent_offset(eb, ei); | |
5331 | left_gen = btrfs_file_extent_generation(eb, ei); | |
31db9f7c AB |
5332 | |
5333 | /* | |
5334 | * Following comments will refer to these graphics. L is the left | |
5335 | * extents which we are checking at the moment. 1-8 are the right | |
5336 | * extents that we iterate. | |
5337 | * | |
5338 | * |-----L-----| | |
5339 | * |-1-|-2a-|-3-|-4-|-5-|-6-| | |
5340 | * | |
5341 | * |-----L-----| | |
5342 | * |--1--|-2b-|...(same as above) | |
5343 | * | |
5344 | * Alternative situation. Happens on files where extents got split. | |
5345 | * |-----L-----| | |
5346 | * |-----------7-----------|-6-| | |
5347 | * | |
5348 | * Alternative situation. Happens on files which got larger. | |
5349 | * |-----L-----| | |
5350 | * |-8-| | |
5351 | * Nothing follows after 8. | |
5352 | */ | |
5353 | ||
5354 | key.objectid = ekey->objectid; | |
5355 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5356 | key.offset = ekey->offset; | |
5357 | ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); | |
5358 | if (ret < 0) | |
5359 | goto out; | |
5360 | if (ret) { | |
5361 | ret = 0; | |
5362 | goto out; | |
5363 | } | |
5364 | ||
5365 | /* | |
5366 | * Handle special case where the right side has no extents at all. | |
5367 | */ | |
5368 | eb = path->nodes[0]; | |
5369 | slot = path->slots[0]; | |
5370 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5371 | if (found_key.objectid != key.objectid || | |
5372 | found_key.type != key.type) { | |
57cfd462 JB |
5373 | /* If we're a hole then just pretend nothing changed */ |
5374 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5375 | goto out; |
5376 | } | |
5377 | ||
5378 | /* | |
5379 | * We're now on 2a, 2b or 7. | |
5380 | */ | |
5381 | key = found_key; | |
5382 | while (key.offset < ekey->offset + left_len) { | |
5383 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
5384 | right_type = btrfs_file_extent_type(eb, ei); | |
e1cbfd7b FM |
5385 | if (right_type != BTRFS_FILE_EXTENT_REG && |
5386 | right_type != BTRFS_FILE_EXTENT_INLINE) { | |
31db9f7c AB |
5387 | ret = 0; |
5388 | goto out; | |
5389 | } | |
5390 | ||
e1cbfd7b FM |
5391 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { |
5392 | right_len = btrfs_file_extent_inline_len(eb, slot, ei); | |
5393 | right_len = PAGE_ALIGN(right_len); | |
5394 | } else { | |
5395 | right_len = btrfs_file_extent_num_bytes(eb, ei); | |
5396 | } | |
007d31f7 | 5397 | |
31db9f7c AB |
5398 | /* |
5399 | * Are we at extent 8? If yes, we know the extent is changed. | |
5400 | * This may only happen on the first iteration. | |
5401 | */ | |
d8347fa4 | 5402 | if (found_key.offset + right_len <= ekey->offset) { |
57cfd462 JB |
5403 | /* If we're a hole just pretend nothing changed */ |
5404 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5405 | goto out; |
5406 | } | |
5407 | ||
e1cbfd7b FM |
5408 | /* |
5409 | * We just wanted to see if when we have an inline extent, what | |
5410 | * follows it is a regular extent (wanted to check the above | |
5411 | * condition for inline extents too). This should normally not | |
5412 | * happen but it's possible for example when we have an inline | |
5413 | * compressed extent representing data with a size matching | |
5414 | * the page size (currently the same as sector size). | |
5415 | */ | |
5416 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { | |
5417 | ret = 0; | |
5418 | goto out; | |
5419 | } | |
5420 | ||
24e52b11 FM |
5421 | right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5422 | right_offset = btrfs_file_extent_offset(eb, ei); | |
5423 | right_gen = btrfs_file_extent_generation(eb, ei); | |
5424 | ||
31db9f7c AB |
5425 | left_offset_fixed = left_offset; |
5426 | if (key.offset < ekey->offset) { | |
5427 | /* Fix the right offset for 2a and 7. */ | |
5428 | right_offset += ekey->offset - key.offset; | |
5429 | } else { | |
5430 | /* Fix the left offset for all behind 2a and 2b */ | |
5431 | left_offset_fixed += key.offset - ekey->offset; | |
5432 | } | |
5433 | ||
5434 | /* | |
5435 | * Check if we have the same extent. | |
5436 | */ | |
3954096d | 5437 | if (left_disknr != right_disknr || |
74dd17fb CM |
5438 | left_offset_fixed != right_offset || |
5439 | left_gen != right_gen) { | |
31db9f7c AB |
5440 | ret = 0; |
5441 | goto out; | |
5442 | } | |
5443 | ||
5444 | /* | |
5445 | * Go to the next extent. | |
5446 | */ | |
5447 | ret = btrfs_next_item(sctx->parent_root, path); | |
5448 | if (ret < 0) | |
5449 | goto out; | |
5450 | if (!ret) { | |
5451 | eb = path->nodes[0]; | |
5452 | slot = path->slots[0]; | |
5453 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5454 | } | |
5455 | if (ret || found_key.objectid != key.objectid || | |
5456 | found_key.type != key.type) { | |
5457 | key.offset += right_len; | |
5458 | break; | |
adaa4b8e JS |
5459 | } |
5460 | if (found_key.offset != key.offset + right_len) { | |
5461 | ret = 0; | |
5462 | goto out; | |
31db9f7c AB |
5463 | } |
5464 | key = found_key; | |
5465 | } | |
5466 | ||
5467 | /* | |
5468 | * We're now behind the left extent (treat as unchanged) or at the end | |
5469 | * of the right side (treat as changed). | |
5470 | */ | |
5471 | if (key.offset >= ekey->offset + left_len) | |
5472 | ret = 1; | |
5473 | else | |
5474 | ret = 0; | |
5475 | ||
5476 | ||
5477 | out: | |
5478 | btrfs_free_path(path); | |
5479 | return ret; | |
5480 | } | |
5481 | ||
16e7549f JB |
5482 | static int get_last_extent(struct send_ctx *sctx, u64 offset) |
5483 | { | |
5484 | struct btrfs_path *path; | |
5485 | struct btrfs_root *root = sctx->send_root; | |
5486 | struct btrfs_file_extent_item *fi; | |
5487 | struct btrfs_key key; | |
5488 | u64 extent_end; | |
5489 | u8 type; | |
5490 | int ret; | |
5491 | ||
5492 | path = alloc_path_for_send(); | |
5493 | if (!path) | |
5494 | return -ENOMEM; | |
5495 | ||
5496 | sctx->cur_inode_last_extent = 0; | |
5497 | ||
5498 | key.objectid = sctx->cur_ino; | |
5499 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5500 | key.offset = offset; | |
5501 | ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); | |
5502 | if (ret < 0) | |
5503 | goto out; | |
5504 | ret = 0; | |
5505 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
5506 | if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
5507 | goto out; | |
5508 | ||
5509 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5510 | struct btrfs_file_extent_item); | |
5511 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5512 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
5513 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
5514 | path->slots[0], fi); | |
16e7549f | 5515 | extent_end = ALIGN(key.offset + size, |
da17066c | 5516 | sctx->send_root->fs_info->sectorsize); |
16e7549f JB |
5517 | } else { |
5518 | extent_end = key.offset + | |
5519 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5520 | } | |
5521 | sctx->cur_inode_last_extent = extent_end; | |
5522 | out: | |
5523 | btrfs_free_path(path); | |
5524 | return ret; | |
5525 | } | |
5526 | ||
82bfb2e7 FM |
5527 | static int range_is_hole_in_parent(struct send_ctx *sctx, |
5528 | const u64 start, | |
5529 | const u64 end) | |
5530 | { | |
5531 | struct btrfs_path *path; | |
5532 | struct btrfs_key key; | |
5533 | struct btrfs_root *root = sctx->parent_root; | |
5534 | u64 search_start = start; | |
5535 | int ret; | |
5536 | ||
5537 | path = alloc_path_for_send(); | |
5538 | if (!path) | |
5539 | return -ENOMEM; | |
5540 | ||
5541 | key.objectid = sctx->cur_ino; | |
5542 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5543 | key.offset = search_start; | |
5544 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5545 | if (ret < 0) | |
5546 | goto out; | |
5547 | if (ret > 0 && path->slots[0] > 0) | |
5548 | path->slots[0]--; | |
5549 | ||
5550 | while (search_start < end) { | |
5551 | struct extent_buffer *leaf = path->nodes[0]; | |
5552 | int slot = path->slots[0]; | |
5553 | struct btrfs_file_extent_item *fi; | |
5554 | u64 extent_end; | |
5555 | ||
5556 | if (slot >= btrfs_header_nritems(leaf)) { | |
5557 | ret = btrfs_next_leaf(root, path); | |
5558 | if (ret < 0) | |
5559 | goto out; | |
5560 | else if (ret > 0) | |
5561 | break; | |
5562 | continue; | |
5563 | } | |
5564 | ||
5565 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5566 | if (key.objectid < sctx->cur_ino || | |
5567 | key.type < BTRFS_EXTENT_DATA_KEY) | |
5568 | goto next; | |
5569 | if (key.objectid > sctx->cur_ino || | |
5570 | key.type > BTRFS_EXTENT_DATA_KEY || | |
5571 | key.offset >= end) | |
5572 | break; | |
5573 | ||
5574 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5575 | if (btrfs_file_extent_type(leaf, fi) == | |
5576 | BTRFS_FILE_EXTENT_INLINE) { | |
5577 | u64 size = btrfs_file_extent_inline_len(leaf, slot, fi); | |
5578 | ||
5579 | extent_end = ALIGN(key.offset + size, | |
5580 | root->fs_info->sectorsize); | |
5581 | } else { | |
5582 | extent_end = key.offset + | |
5583 | btrfs_file_extent_num_bytes(leaf, fi); | |
5584 | } | |
5585 | if (extent_end <= start) | |
5586 | goto next; | |
5587 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) { | |
5588 | search_start = extent_end; | |
5589 | goto next; | |
5590 | } | |
5591 | ret = 0; | |
5592 | goto out; | |
5593 | next: | |
5594 | path->slots[0]++; | |
5595 | } | |
5596 | ret = 1; | |
5597 | out: | |
5598 | btrfs_free_path(path); | |
5599 | return ret; | |
5600 | } | |
5601 | ||
16e7549f JB |
5602 | static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, |
5603 | struct btrfs_key *key) | |
5604 | { | |
5605 | struct btrfs_file_extent_item *fi; | |
5606 | u64 extent_end; | |
5607 | u8 type; | |
5608 | int ret = 0; | |
5609 | ||
5610 | if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) | |
5611 | return 0; | |
5612 | ||
5613 | if (sctx->cur_inode_last_extent == (u64)-1) { | |
5614 | ret = get_last_extent(sctx, key->offset - 1); | |
5615 | if (ret) | |
5616 | return ret; | |
5617 | } | |
5618 | ||
5619 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5620 | struct btrfs_file_extent_item); | |
5621 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5622 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
5623 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
5624 | path->slots[0], fi); | |
16e7549f | 5625 | extent_end = ALIGN(key->offset + size, |
da17066c | 5626 | sctx->send_root->fs_info->sectorsize); |
16e7549f JB |
5627 | } else { |
5628 | extent_end = key->offset + | |
5629 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5630 | } | |
bf54f412 FDBM |
5631 | |
5632 | if (path->slots[0] == 0 && | |
5633 | sctx->cur_inode_last_extent < key->offset) { | |
5634 | /* | |
5635 | * We might have skipped entire leafs that contained only | |
5636 | * file extent items for our current inode. These leafs have | |
5637 | * a generation number smaller (older) than the one in the | |
5638 | * current leaf and the leaf our last extent came from, and | |
5639 | * are located between these 2 leafs. | |
5640 | */ | |
5641 | ret = get_last_extent(sctx, key->offset - 1); | |
5642 | if (ret) | |
5643 | return ret; | |
5644 | } | |
5645 | ||
82bfb2e7 FM |
5646 | if (sctx->cur_inode_last_extent < key->offset) { |
5647 | ret = range_is_hole_in_parent(sctx, | |
5648 | sctx->cur_inode_last_extent, | |
5649 | key->offset); | |
5650 | if (ret < 0) | |
5651 | return ret; | |
5652 | else if (ret == 0) | |
5653 | ret = send_hole(sctx, key->offset); | |
5654 | else | |
5655 | ret = 0; | |
5656 | } | |
16e7549f JB |
5657 | sctx->cur_inode_last_extent = extent_end; |
5658 | return ret; | |
5659 | } | |
5660 | ||
31db9f7c AB |
5661 | static int process_extent(struct send_ctx *sctx, |
5662 | struct btrfs_path *path, | |
5663 | struct btrfs_key *key) | |
5664 | { | |
31db9f7c | 5665 | struct clone_root *found_clone = NULL; |
57cfd462 | 5666 | int ret = 0; |
31db9f7c AB |
5667 | |
5668 | if (S_ISLNK(sctx->cur_inode_mode)) | |
5669 | return 0; | |
5670 | ||
5671 | if (sctx->parent_root && !sctx->cur_inode_new) { | |
5672 | ret = is_extent_unchanged(sctx, path, key); | |
5673 | if (ret < 0) | |
5674 | goto out; | |
5675 | if (ret) { | |
5676 | ret = 0; | |
16e7549f | 5677 | goto out_hole; |
31db9f7c | 5678 | } |
57cfd462 JB |
5679 | } else { |
5680 | struct btrfs_file_extent_item *ei; | |
5681 | u8 type; | |
5682 | ||
5683 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5684 | struct btrfs_file_extent_item); | |
5685 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
5686 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
5687 | type == BTRFS_FILE_EXTENT_REG) { | |
5688 | /* | |
5689 | * The send spec does not have a prealloc command yet, | |
5690 | * so just leave a hole for prealloc'ed extents until | |
5691 | * we have enough commands queued up to justify rev'ing | |
5692 | * the send spec. | |
5693 | */ | |
5694 | if (type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5695 | ret = 0; | |
5696 | goto out; | |
5697 | } | |
5698 | ||
5699 | /* Have a hole, just skip it. */ | |
5700 | if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { | |
5701 | ret = 0; | |
5702 | goto out; | |
5703 | } | |
5704 | } | |
31db9f7c AB |
5705 | } |
5706 | ||
5707 | ret = find_extent_clone(sctx, path, key->objectid, key->offset, | |
5708 | sctx->cur_inode_size, &found_clone); | |
5709 | if (ret != -ENOENT && ret < 0) | |
5710 | goto out; | |
5711 | ||
5712 | ret = send_write_or_clone(sctx, path, key, found_clone); | |
16e7549f JB |
5713 | if (ret) |
5714 | goto out; | |
5715 | out_hole: | |
5716 | ret = maybe_send_hole(sctx, path, key); | |
31db9f7c AB |
5717 | out: |
5718 | return ret; | |
5719 | } | |
5720 | ||
5721 | static int process_all_extents(struct send_ctx *sctx) | |
5722 | { | |
5723 | int ret; | |
5724 | struct btrfs_root *root; | |
5725 | struct btrfs_path *path; | |
5726 | struct btrfs_key key; | |
5727 | struct btrfs_key found_key; | |
5728 | struct extent_buffer *eb; | |
5729 | int slot; | |
5730 | ||
5731 | root = sctx->send_root; | |
5732 | path = alloc_path_for_send(); | |
5733 | if (!path) | |
5734 | return -ENOMEM; | |
5735 | ||
5736 | key.objectid = sctx->cmp_key->objectid; | |
5737 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5738 | key.offset = 0; | |
7fdd29d0 FDBM |
5739 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5740 | if (ret < 0) | |
5741 | goto out; | |
31db9f7c | 5742 | |
7fdd29d0 | 5743 | while (1) { |
31db9f7c AB |
5744 | eb = path->nodes[0]; |
5745 | slot = path->slots[0]; | |
7fdd29d0 FDBM |
5746 | |
5747 | if (slot >= btrfs_header_nritems(eb)) { | |
5748 | ret = btrfs_next_leaf(root, path); | |
5749 | if (ret < 0) { | |
5750 | goto out; | |
5751 | } else if (ret > 0) { | |
5752 | ret = 0; | |
5753 | break; | |
5754 | } | |
5755 | continue; | |
5756 | } | |
5757 | ||
31db9f7c AB |
5758 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
5759 | ||
5760 | if (found_key.objectid != key.objectid || | |
5761 | found_key.type != key.type) { | |
5762 | ret = 0; | |
5763 | goto out; | |
5764 | } | |
5765 | ||
5766 | ret = process_extent(sctx, path, &found_key); | |
5767 | if (ret < 0) | |
5768 | goto out; | |
5769 | ||
7fdd29d0 | 5770 | path->slots[0]++; |
31db9f7c AB |
5771 | } |
5772 | ||
5773 | out: | |
5774 | btrfs_free_path(path); | |
5775 | return ret; | |
5776 | } | |
5777 | ||
9f03740a FDBM |
5778 | static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, |
5779 | int *pending_move, | |
5780 | int *refs_processed) | |
31db9f7c AB |
5781 | { |
5782 | int ret = 0; | |
5783 | ||
5784 | if (sctx->cur_ino == 0) | |
5785 | goto out; | |
5786 | if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && | |
96b5bd77 | 5787 | sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) |
31db9f7c AB |
5788 | goto out; |
5789 | if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) | |
5790 | goto out; | |
5791 | ||
9f03740a | 5792 | ret = process_recorded_refs(sctx, pending_move); |
e479d9bb AB |
5793 | if (ret < 0) |
5794 | goto out; | |
5795 | ||
9f03740a | 5796 | *refs_processed = 1; |
31db9f7c AB |
5797 | out: |
5798 | return ret; | |
5799 | } | |
5800 | ||
5801 | static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) | |
5802 | { | |
5803 | int ret = 0; | |
5804 | u64 left_mode; | |
5805 | u64 left_uid; | |
5806 | u64 left_gid; | |
5807 | u64 right_mode; | |
5808 | u64 right_uid; | |
5809 | u64 right_gid; | |
5810 | int need_chmod = 0; | |
5811 | int need_chown = 0; | |
9f03740a FDBM |
5812 | int pending_move = 0; |
5813 | int refs_processed = 0; | |
31db9f7c | 5814 | |
14ecdce9 FM |
5815 | if (sctx->ignore_cur_inode) |
5816 | return 0; | |
5817 | ||
9f03740a FDBM |
5818 | ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, |
5819 | &refs_processed); | |
31db9f7c AB |
5820 | if (ret < 0) |
5821 | goto out; | |
5822 | ||
9f03740a FDBM |
5823 | /* |
5824 | * We have processed the refs and thus need to advance send_progress. | |
5825 | * Now, calls to get_cur_xxx will take the updated refs of the current | |
5826 | * inode into account. | |
5827 | * | |
5828 | * On the other hand, if our current inode is a directory and couldn't | |
5829 | * be moved/renamed because its parent was renamed/moved too and it has | |
5830 | * a higher inode number, we can only move/rename our current inode | |
5831 | * after we moved/renamed its parent. Therefore in this case operate on | |
5832 | * the old path (pre move/rename) of our current inode, and the | |
5833 | * move/rename will be performed later. | |
5834 | */ | |
5835 | if (refs_processed && !pending_move) | |
5836 | sctx->send_progress = sctx->cur_ino + 1; | |
5837 | ||
31db9f7c AB |
5838 | if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) |
5839 | goto out; | |
5840 | if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) | |
5841 | goto out; | |
5842 | ||
5843 | ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, | |
85a7b33b | 5844 | &left_mode, &left_uid, &left_gid, NULL); |
31db9f7c AB |
5845 | if (ret < 0) |
5846 | goto out; | |
5847 | ||
e2d044fe AL |
5848 | if (!sctx->parent_root || sctx->cur_inode_new) { |
5849 | need_chown = 1; | |
5850 | if (!S_ISLNK(sctx->cur_inode_mode)) | |
31db9f7c | 5851 | need_chmod = 1; |
e2d044fe AL |
5852 | } else { |
5853 | ret = get_inode_info(sctx->parent_root, sctx->cur_ino, | |
5854 | NULL, NULL, &right_mode, &right_uid, | |
5855 | &right_gid, NULL); | |
5856 | if (ret < 0) | |
5857 | goto out; | |
31db9f7c | 5858 | |
e2d044fe AL |
5859 | if (left_uid != right_uid || left_gid != right_gid) |
5860 | need_chown = 1; | |
5861 | if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) | |
5862 | need_chmod = 1; | |
31db9f7c AB |
5863 | } |
5864 | ||
5865 | if (S_ISREG(sctx->cur_inode_mode)) { | |
16e7549f | 5866 | if (need_send_hole(sctx)) { |
766b5e5a FM |
5867 | if (sctx->cur_inode_last_extent == (u64)-1 || |
5868 | sctx->cur_inode_last_extent < | |
5869 | sctx->cur_inode_size) { | |
16e7549f JB |
5870 | ret = get_last_extent(sctx, (u64)-1); |
5871 | if (ret) | |
5872 | goto out; | |
5873 | } | |
5874 | if (sctx->cur_inode_last_extent < | |
5875 | sctx->cur_inode_size) { | |
5876 | ret = send_hole(sctx, sctx->cur_inode_size); | |
5877 | if (ret) | |
5878 | goto out; | |
5879 | } | |
5880 | } | |
31db9f7c AB |
5881 | ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
5882 | sctx->cur_inode_size); | |
5883 | if (ret < 0) | |
5884 | goto out; | |
5885 | } | |
5886 | ||
5887 | if (need_chown) { | |
5888 | ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5889 | left_uid, left_gid); | |
5890 | if (ret < 0) | |
5891 | goto out; | |
5892 | } | |
5893 | if (need_chmod) { | |
5894 | ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5895 | left_mode); | |
5896 | if (ret < 0) | |
5897 | goto out; | |
5898 | } | |
5899 | ||
5900 | /* | |
9f03740a FDBM |
5901 | * If other directory inodes depended on our current directory |
5902 | * inode's move/rename, now do their move/rename operations. | |
31db9f7c | 5903 | */ |
9f03740a FDBM |
5904 | if (!is_waiting_for_move(sctx, sctx->cur_ino)) { |
5905 | ret = apply_children_dir_moves(sctx); | |
5906 | if (ret) | |
5907 | goto out; | |
fcbd2154 FM |
5908 | /* |
5909 | * Need to send that every time, no matter if it actually | |
5910 | * changed between the two trees as we have done changes to | |
5911 | * the inode before. If our inode is a directory and it's | |
5912 | * waiting to be moved/renamed, we will send its utimes when | |
5913 | * it's moved/renamed, therefore we don't need to do it here. | |
5914 | */ | |
5915 | sctx->send_progress = sctx->cur_ino + 1; | |
5916 | ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
5917 | if (ret < 0) | |
5918 | goto out; | |
9f03740a FDBM |
5919 | } |
5920 | ||
31db9f7c AB |
5921 | out: |
5922 | return ret; | |
5923 | } | |
5924 | ||
14ecdce9 FM |
5925 | struct parent_paths_ctx { |
5926 | struct list_head *refs; | |
5927 | struct send_ctx *sctx; | |
5928 | }; | |
5929 | ||
5930 | static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name, | |
5931 | void *ctx) | |
5932 | { | |
5933 | struct parent_paths_ctx *ppctx = ctx; | |
5934 | ||
5935 | return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx, | |
5936 | ppctx->refs); | |
5937 | } | |
5938 | ||
5939 | /* | |
5940 | * Issue unlink operations for all paths of the current inode found in the | |
5941 | * parent snapshot. | |
5942 | */ | |
5943 | static int btrfs_unlink_all_paths(struct send_ctx *sctx) | |
5944 | { | |
5945 | LIST_HEAD(deleted_refs); | |
5946 | struct btrfs_path *path; | |
5947 | struct btrfs_key key; | |
5948 | struct parent_paths_ctx ctx; | |
5949 | int ret; | |
5950 | ||
5951 | path = alloc_path_for_send(); | |
5952 | if (!path) | |
5953 | return -ENOMEM; | |
5954 | ||
5955 | key.objectid = sctx->cur_ino; | |
5956 | key.type = BTRFS_INODE_REF_KEY; | |
5957 | key.offset = 0; | |
5958 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
5959 | if (ret < 0) | |
5960 | goto out; | |
5961 | ||
5962 | ctx.refs = &deleted_refs; | |
5963 | ctx.sctx = sctx; | |
5964 | ||
5965 | while (true) { | |
5966 | struct extent_buffer *eb = path->nodes[0]; | |
5967 | int slot = path->slots[0]; | |
5968 | ||
5969 | if (slot >= btrfs_header_nritems(eb)) { | |
5970 | ret = btrfs_next_leaf(sctx->parent_root, path); | |
5971 | if (ret < 0) | |
5972 | goto out; | |
5973 | else if (ret > 0) | |
5974 | break; | |
5975 | continue; | |
5976 | } | |
5977 | ||
5978 | btrfs_item_key_to_cpu(eb, &key, slot); | |
5979 | if (key.objectid != sctx->cur_ino) | |
5980 | break; | |
5981 | if (key.type != BTRFS_INODE_REF_KEY && | |
5982 | key.type != BTRFS_INODE_EXTREF_KEY) | |
5983 | break; | |
5984 | ||
5985 | ret = iterate_inode_ref(sctx->parent_root, path, &key, 1, | |
5986 | record_parent_ref, &ctx); | |
5987 | if (ret < 0) | |
5988 | goto out; | |
5989 | ||
5990 | path->slots[0]++; | |
5991 | } | |
5992 | ||
5993 | while (!list_empty(&deleted_refs)) { | |
5994 | struct recorded_ref *ref; | |
5995 | ||
5996 | ref = list_first_entry(&deleted_refs, struct recorded_ref, list); | |
5997 | ret = send_unlink(sctx, ref->full_path); | |
5998 | if (ret < 0) | |
5999 | goto out; | |
6000 | fs_path_free(ref->full_path); | |
6001 | list_del(&ref->list); | |
6002 | kfree(ref); | |
6003 | } | |
6004 | ret = 0; | |
6005 | out: | |
6006 | btrfs_free_path(path); | |
6007 | if (ret) | |
6008 | __free_recorded_refs(&deleted_refs); | |
6009 | return ret; | |
6010 | } | |
6011 | ||
31db9f7c AB |
6012 | static int changed_inode(struct send_ctx *sctx, |
6013 | enum btrfs_compare_tree_result result) | |
6014 | { | |
6015 | int ret = 0; | |
6016 | struct btrfs_key *key = sctx->cmp_key; | |
6017 | struct btrfs_inode_item *left_ii = NULL; | |
6018 | struct btrfs_inode_item *right_ii = NULL; | |
6019 | u64 left_gen = 0; | |
6020 | u64 right_gen = 0; | |
6021 | ||
31db9f7c AB |
6022 | sctx->cur_ino = key->objectid; |
6023 | sctx->cur_inode_new_gen = 0; | |
16e7549f | 6024 | sctx->cur_inode_last_extent = (u64)-1; |
14ecdce9 | 6025 | sctx->ignore_cur_inode = false; |
e479d9bb AB |
6026 | |
6027 | /* | |
6028 | * Set send_progress to current inode. This will tell all get_cur_xxx | |
6029 | * functions that the current inode's refs are not updated yet. Later, | |
6030 | * when process_recorded_refs is finished, it is set to cur_ino + 1. | |
6031 | */ | |
31db9f7c AB |
6032 | sctx->send_progress = sctx->cur_ino; |
6033 | ||
6034 | if (result == BTRFS_COMPARE_TREE_NEW || | |
6035 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
6036 | left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], | |
6037 | sctx->left_path->slots[0], | |
6038 | struct btrfs_inode_item); | |
6039 | left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], | |
6040 | left_ii); | |
6041 | } else { | |
6042 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6043 | sctx->right_path->slots[0], | |
6044 | struct btrfs_inode_item); | |
6045 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6046 | right_ii); | |
6047 | } | |
6048 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
6049 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6050 | sctx->right_path->slots[0], | |
6051 | struct btrfs_inode_item); | |
6052 | ||
6053 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6054 | right_ii); | |
6d85ed05 AB |
6055 | |
6056 | /* | |
6057 | * The cur_ino = root dir case is special here. We can't treat | |
6058 | * the inode as deleted+reused because it would generate a | |
6059 | * stream that tries to delete/mkdir the root dir. | |
6060 | */ | |
6061 | if (left_gen != right_gen && | |
6062 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) | |
31db9f7c AB |
6063 | sctx->cur_inode_new_gen = 1; |
6064 | } | |
6065 | ||
14ecdce9 FM |
6066 | /* |
6067 | * Normally we do not find inodes with a link count of zero (orphans) | |
6068 | * because the most common case is to create a snapshot and use it | |
6069 | * for a send operation. However other less common use cases involve | |
6070 | * using a subvolume and send it after turning it to RO mode just | |
6071 | * after deleting all hard links of a file while holding an open | |
6072 | * file descriptor against it or turning a RO snapshot into RW mode, | |
6073 | * keep an open file descriptor against a file, delete it and then | |
6074 | * turn the snapshot back to RO mode before using it for a send | |
6075 | * operation. So if we find such cases, ignore the inode and all its | |
6076 | * items completely if it's a new inode, or if it's a changed inode | |
6077 | * make sure all its previous paths (from the parent snapshot) are all | |
6078 | * unlinked and all other the inode items are ignored. | |
6079 | */ | |
6080 | if (result == BTRFS_COMPARE_TREE_NEW || | |
6081 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
6082 | u32 nlinks; | |
6083 | ||
6084 | nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii); | |
6085 | if (nlinks == 0) { | |
6086 | sctx->ignore_cur_inode = true; | |
6087 | if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6088 | ret = btrfs_unlink_all_paths(sctx); | |
6089 | goto out; | |
6090 | } | |
6091 | } | |
6092 | ||
31db9f7c AB |
6093 | if (result == BTRFS_COMPARE_TREE_NEW) { |
6094 | sctx->cur_inode_gen = left_gen; | |
6095 | sctx->cur_inode_new = 1; | |
6096 | sctx->cur_inode_deleted = 0; | |
6097 | sctx->cur_inode_size = btrfs_inode_size( | |
6098 | sctx->left_path->nodes[0], left_ii); | |
6099 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6100 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6101 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6102 | sctx->left_path->nodes[0], left_ii); | |
31db9f7c | 6103 | if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) |
1f4692da | 6104 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6105 | } else if (result == BTRFS_COMPARE_TREE_DELETED) { |
6106 | sctx->cur_inode_gen = right_gen; | |
6107 | sctx->cur_inode_new = 0; | |
6108 | sctx->cur_inode_deleted = 1; | |
6109 | sctx->cur_inode_size = btrfs_inode_size( | |
6110 | sctx->right_path->nodes[0], right_ii); | |
6111 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6112 | sctx->right_path->nodes[0], right_ii); | |
6113 | } else if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
766702ef AB |
6114 | /* |
6115 | * We need to do some special handling in case the inode was | |
6116 | * reported as changed with a changed generation number. This | |
6117 | * means that the original inode was deleted and new inode | |
6118 | * reused the same inum. So we have to treat the old inode as | |
6119 | * deleted and the new one as new. | |
6120 | */ | |
31db9f7c | 6121 | if (sctx->cur_inode_new_gen) { |
766702ef AB |
6122 | /* |
6123 | * First, process the inode as if it was deleted. | |
6124 | */ | |
31db9f7c AB |
6125 | sctx->cur_inode_gen = right_gen; |
6126 | sctx->cur_inode_new = 0; | |
6127 | sctx->cur_inode_deleted = 1; | |
6128 | sctx->cur_inode_size = btrfs_inode_size( | |
6129 | sctx->right_path->nodes[0], right_ii); | |
6130 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6131 | sctx->right_path->nodes[0], right_ii); | |
6132 | ret = process_all_refs(sctx, | |
6133 | BTRFS_COMPARE_TREE_DELETED); | |
6134 | if (ret < 0) | |
6135 | goto out; | |
6136 | ||
766702ef AB |
6137 | /* |
6138 | * Now process the inode as if it was new. | |
6139 | */ | |
31db9f7c AB |
6140 | sctx->cur_inode_gen = left_gen; |
6141 | sctx->cur_inode_new = 1; | |
6142 | sctx->cur_inode_deleted = 0; | |
6143 | sctx->cur_inode_size = btrfs_inode_size( | |
6144 | sctx->left_path->nodes[0], left_ii); | |
6145 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6146 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6147 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6148 | sctx->left_path->nodes[0], left_ii); | |
1f4692da | 6149 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6150 | if (ret < 0) |
6151 | goto out; | |
6152 | ||
6153 | ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); | |
6154 | if (ret < 0) | |
6155 | goto out; | |
e479d9bb AB |
6156 | /* |
6157 | * Advance send_progress now as we did not get into | |
6158 | * process_recorded_refs_if_needed in the new_gen case. | |
6159 | */ | |
6160 | sctx->send_progress = sctx->cur_ino + 1; | |
766702ef AB |
6161 | |
6162 | /* | |
6163 | * Now process all extents and xattrs of the inode as if | |
6164 | * they were all new. | |
6165 | */ | |
31db9f7c AB |
6166 | ret = process_all_extents(sctx); |
6167 | if (ret < 0) | |
6168 | goto out; | |
6169 | ret = process_all_new_xattrs(sctx); | |
6170 | if (ret < 0) | |
6171 | goto out; | |
6172 | } else { | |
6173 | sctx->cur_inode_gen = left_gen; | |
6174 | sctx->cur_inode_new = 0; | |
6175 | sctx->cur_inode_new_gen = 0; | |
6176 | sctx->cur_inode_deleted = 0; | |
6177 | sctx->cur_inode_size = btrfs_inode_size( | |
6178 | sctx->left_path->nodes[0], left_ii); | |
6179 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6180 | sctx->left_path->nodes[0], left_ii); | |
6181 | } | |
6182 | } | |
6183 | ||
6184 | out: | |
6185 | return ret; | |
6186 | } | |
6187 | ||
766702ef AB |
6188 | /* |
6189 | * We have to process new refs before deleted refs, but compare_trees gives us | |
6190 | * the new and deleted refs mixed. To fix this, we record the new/deleted refs | |
6191 | * first and later process them in process_recorded_refs. | |
6192 | * For the cur_inode_new_gen case, we skip recording completely because | |
6193 | * changed_inode did already initiate processing of refs. The reason for this is | |
6194 | * that in this case, compare_tree actually compares the refs of 2 different | |
6195 | * inodes. To fix this, process_all_refs is used in changed_inode to handle all | |
6196 | * refs of the right tree as deleted and all refs of the left tree as new. | |
6197 | */ | |
31db9f7c AB |
6198 | static int changed_ref(struct send_ctx *sctx, |
6199 | enum btrfs_compare_tree_result result) | |
6200 | { | |
6201 | int ret = 0; | |
6202 | ||
95155585 FM |
6203 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6204 | inconsistent_snapshot_error(sctx, result, "reference"); | |
6205 | return -EIO; | |
6206 | } | |
31db9f7c AB |
6207 | |
6208 | if (!sctx->cur_inode_new_gen && | |
6209 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { | |
6210 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6211 | ret = record_new_ref(sctx); | |
6212 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6213 | ret = record_deleted_ref(sctx); | |
6214 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6215 | ret = record_changed_ref(sctx); | |
6216 | } | |
6217 | ||
6218 | return ret; | |
6219 | } | |
6220 | ||
766702ef AB |
6221 | /* |
6222 | * Process new/deleted/changed xattrs. We skip processing in the | |
6223 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6224 | * of xattrs. The reason is the same as in changed_ref | |
6225 | */ | |
31db9f7c AB |
6226 | static int changed_xattr(struct send_ctx *sctx, |
6227 | enum btrfs_compare_tree_result result) | |
6228 | { | |
6229 | int ret = 0; | |
6230 | ||
95155585 FM |
6231 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6232 | inconsistent_snapshot_error(sctx, result, "xattr"); | |
6233 | return -EIO; | |
6234 | } | |
31db9f7c AB |
6235 | |
6236 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6237 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6238 | ret = process_new_xattr(sctx); | |
6239 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6240 | ret = process_deleted_xattr(sctx); | |
6241 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6242 | ret = process_changed_xattr(sctx); | |
6243 | } | |
6244 | ||
6245 | return ret; | |
6246 | } | |
6247 | ||
766702ef AB |
6248 | /* |
6249 | * Process new/deleted/changed extents. We skip processing in the | |
6250 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6251 | * of extents. The reason is the same as in changed_ref | |
6252 | */ | |
31db9f7c AB |
6253 | static int changed_extent(struct send_ctx *sctx, |
6254 | enum btrfs_compare_tree_result result) | |
6255 | { | |
6256 | int ret = 0; | |
6257 | ||
95155585 | 6258 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
d5e84fd8 FM |
6259 | |
6260 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
6261 | struct extent_buffer *leaf_l; | |
6262 | struct extent_buffer *leaf_r; | |
6263 | struct btrfs_file_extent_item *ei_l; | |
6264 | struct btrfs_file_extent_item *ei_r; | |
6265 | ||
6266 | leaf_l = sctx->left_path->nodes[0]; | |
6267 | leaf_r = sctx->right_path->nodes[0]; | |
6268 | ei_l = btrfs_item_ptr(leaf_l, | |
6269 | sctx->left_path->slots[0], | |
6270 | struct btrfs_file_extent_item); | |
6271 | ei_r = btrfs_item_ptr(leaf_r, | |
6272 | sctx->right_path->slots[0], | |
6273 | struct btrfs_file_extent_item); | |
6274 | ||
6275 | /* | |
6276 | * We may have found an extent item that has changed | |
6277 | * only its disk_bytenr field and the corresponding | |
6278 | * inode item was not updated. This case happens due to | |
6279 | * very specific timings during relocation when a leaf | |
6280 | * that contains file extent items is COWed while | |
6281 | * relocation is ongoing and its in the stage where it | |
6282 | * updates data pointers. So when this happens we can | |
6283 | * safely ignore it since we know it's the same extent, | |
6284 | * but just at different logical and physical locations | |
6285 | * (when an extent is fully replaced with a new one, we | |
6286 | * know the generation number must have changed too, | |
6287 | * since snapshot creation implies committing the current | |
6288 | * transaction, and the inode item must have been updated | |
6289 | * as well). | |
6290 | * This replacement of the disk_bytenr happens at | |
6291 | * relocation.c:replace_file_extents() through | |
6292 | * relocation.c:btrfs_reloc_cow_block(). | |
6293 | */ | |
6294 | if (btrfs_file_extent_generation(leaf_l, ei_l) == | |
6295 | btrfs_file_extent_generation(leaf_r, ei_r) && | |
6296 | btrfs_file_extent_ram_bytes(leaf_l, ei_l) == | |
6297 | btrfs_file_extent_ram_bytes(leaf_r, ei_r) && | |
6298 | btrfs_file_extent_compression(leaf_l, ei_l) == | |
6299 | btrfs_file_extent_compression(leaf_r, ei_r) && | |
6300 | btrfs_file_extent_encryption(leaf_l, ei_l) == | |
6301 | btrfs_file_extent_encryption(leaf_r, ei_r) && | |
6302 | btrfs_file_extent_other_encoding(leaf_l, ei_l) == | |
6303 | btrfs_file_extent_other_encoding(leaf_r, ei_r) && | |
6304 | btrfs_file_extent_type(leaf_l, ei_l) == | |
6305 | btrfs_file_extent_type(leaf_r, ei_r) && | |
6306 | btrfs_file_extent_disk_bytenr(leaf_l, ei_l) != | |
6307 | btrfs_file_extent_disk_bytenr(leaf_r, ei_r) && | |
6308 | btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) == | |
6309 | btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) && | |
6310 | btrfs_file_extent_offset(leaf_l, ei_l) == | |
6311 | btrfs_file_extent_offset(leaf_r, ei_r) && | |
6312 | btrfs_file_extent_num_bytes(leaf_l, ei_l) == | |
6313 | btrfs_file_extent_num_bytes(leaf_r, ei_r)) | |
6314 | return 0; | |
6315 | } | |
6316 | ||
95155585 FM |
6317 | inconsistent_snapshot_error(sctx, result, "extent"); |
6318 | return -EIO; | |
6319 | } | |
31db9f7c AB |
6320 | |
6321 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6322 | if (result != BTRFS_COMPARE_TREE_DELETED) | |
6323 | ret = process_extent(sctx, sctx->left_path, | |
6324 | sctx->cmp_key); | |
6325 | } | |
6326 | ||
6327 | return ret; | |
6328 | } | |
6329 | ||
ba5e8f2e JB |
6330 | static int dir_changed(struct send_ctx *sctx, u64 dir) |
6331 | { | |
6332 | u64 orig_gen, new_gen; | |
6333 | int ret; | |
6334 | ||
6335 | ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, | |
6336 | NULL, NULL); | |
6337 | if (ret) | |
6338 | return ret; | |
6339 | ||
6340 | ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, | |
6341 | NULL, NULL, NULL); | |
6342 | if (ret) | |
6343 | return ret; | |
6344 | ||
6345 | return (orig_gen != new_gen) ? 1 : 0; | |
6346 | } | |
6347 | ||
6348 | static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, | |
6349 | struct btrfs_key *key) | |
6350 | { | |
6351 | struct btrfs_inode_extref *extref; | |
6352 | struct extent_buffer *leaf; | |
6353 | u64 dirid = 0, last_dirid = 0; | |
6354 | unsigned long ptr; | |
6355 | u32 item_size; | |
6356 | u32 cur_offset = 0; | |
6357 | int ref_name_len; | |
6358 | int ret = 0; | |
6359 | ||
6360 | /* Easy case, just check this one dirid */ | |
6361 | if (key->type == BTRFS_INODE_REF_KEY) { | |
6362 | dirid = key->offset; | |
6363 | ||
6364 | ret = dir_changed(sctx, dirid); | |
6365 | goto out; | |
6366 | } | |
6367 | ||
6368 | leaf = path->nodes[0]; | |
6369 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6370 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
6371 | while (cur_offset < item_size) { | |
6372 | extref = (struct btrfs_inode_extref *)(ptr + | |
6373 | cur_offset); | |
6374 | dirid = btrfs_inode_extref_parent(leaf, extref); | |
6375 | ref_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
6376 | cur_offset += ref_name_len + sizeof(*extref); | |
6377 | if (dirid == last_dirid) | |
6378 | continue; | |
6379 | ret = dir_changed(sctx, dirid); | |
6380 | if (ret) | |
6381 | break; | |
6382 | last_dirid = dirid; | |
6383 | } | |
6384 | out: | |
6385 | return ret; | |
6386 | } | |
6387 | ||
766702ef AB |
6388 | /* |
6389 | * Updates compare related fields in sctx and simply forwards to the actual | |
6390 | * changed_xxx functions. | |
6391 | */ | |
ee8c494f | 6392 | static int changed_cb(struct btrfs_path *left_path, |
31db9f7c AB |
6393 | struct btrfs_path *right_path, |
6394 | struct btrfs_key *key, | |
6395 | enum btrfs_compare_tree_result result, | |
6396 | void *ctx) | |
6397 | { | |
6398 | int ret = 0; | |
6399 | struct send_ctx *sctx = ctx; | |
6400 | ||
ba5e8f2e | 6401 | if (result == BTRFS_COMPARE_TREE_SAME) { |
16e7549f JB |
6402 | if (key->type == BTRFS_INODE_REF_KEY || |
6403 | key->type == BTRFS_INODE_EXTREF_KEY) { | |
6404 | ret = compare_refs(sctx, left_path, key); | |
6405 | if (!ret) | |
6406 | return 0; | |
6407 | if (ret < 0) | |
6408 | return ret; | |
6409 | } else if (key->type == BTRFS_EXTENT_DATA_KEY) { | |
6410 | return maybe_send_hole(sctx, left_path, key); | |
6411 | } else { | |
ba5e8f2e | 6412 | return 0; |
16e7549f | 6413 | } |
ba5e8f2e JB |
6414 | result = BTRFS_COMPARE_TREE_CHANGED; |
6415 | ret = 0; | |
6416 | } | |
6417 | ||
31db9f7c AB |
6418 | sctx->left_path = left_path; |
6419 | sctx->right_path = right_path; | |
6420 | sctx->cmp_key = key; | |
6421 | ||
6422 | ret = finish_inode_if_needed(sctx, 0); | |
6423 | if (ret < 0) | |
6424 | goto out; | |
6425 | ||
2981e225 AB |
6426 | /* Ignore non-FS objects */ |
6427 | if (key->objectid == BTRFS_FREE_INO_OBJECTID || | |
6428 | key->objectid == BTRFS_FREE_SPACE_OBJECTID) | |
6429 | goto out; | |
6430 | ||
14ecdce9 | 6431 | if (key->type == BTRFS_INODE_ITEM_KEY) { |
31db9f7c | 6432 | ret = changed_inode(sctx, result); |
14ecdce9 FM |
6433 | } else if (!sctx->ignore_cur_inode) { |
6434 | if (key->type == BTRFS_INODE_REF_KEY || | |
6435 | key->type == BTRFS_INODE_EXTREF_KEY) | |
6436 | ret = changed_ref(sctx, result); | |
6437 | else if (key->type == BTRFS_XATTR_ITEM_KEY) | |
6438 | ret = changed_xattr(sctx, result); | |
6439 | else if (key->type == BTRFS_EXTENT_DATA_KEY) | |
6440 | ret = changed_extent(sctx, result); | |
6441 | } | |
31db9f7c AB |
6442 | |
6443 | out: | |
6444 | return ret; | |
6445 | } | |
6446 | ||
6447 | static int full_send_tree(struct send_ctx *sctx) | |
6448 | { | |
6449 | int ret; | |
31db9f7c AB |
6450 | struct btrfs_root *send_root = sctx->send_root; |
6451 | struct btrfs_key key; | |
6452 | struct btrfs_key found_key; | |
6453 | struct btrfs_path *path; | |
6454 | struct extent_buffer *eb; | |
6455 | int slot; | |
31db9f7c AB |
6456 | |
6457 | path = alloc_path_for_send(); | |
6458 | if (!path) | |
6459 | return -ENOMEM; | |
6460 | ||
31db9f7c AB |
6461 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; |
6462 | key.type = BTRFS_INODE_ITEM_KEY; | |
6463 | key.offset = 0; | |
6464 | ||
31db9f7c AB |
6465 | ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); |
6466 | if (ret < 0) | |
6467 | goto out; | |
6468 | if (ret) | |
6469 | goto out_finish; | |
6470 | ||
6471 | while (1) { | |
31db9f7c AB |
6472 | eb = path->nodes[0]; |
6473 | slot = path->slots[0]; | |
6474 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6475 | ||
ee8c494f NB |
6476 | ret = changed_cb(path, NULL, &found_key, |
6477 | BTRFS_COMPARE_TREE_NEW, sctx); | |
31db9f7c AB |
6478 | if (ret < 0) |
6479 | goto out; | |
6480 | ||
6481 | key.objectid = found_key.objectid; | |
6482 | key.type = found_key.type; | |
6483 | key.offset = found_key.offset + 1; | |
6484 | ||
6485 | ret = btrfs_next_item(send_root, path); | |
6486 | if (ret < 0) | |
6487 | goto out; | |
6488 | if (ret) { | |
6489 | ret = 0; | |
6490 | break; | |
6491 | } | |
6492 | } | |
6493 | ||
6494 | out_finish: | |
6495 | ret = finish_inode_if_needed(sctx, 1); | |
6496 | ||
6497 | out: | |
6498 | btrfs_free_path(path); | |
31db9f7c AB |
6499 | return ret; |
6500 | } | |
6501 | ||
6502 | static int send_subvol(struct send_ctx *sctx) | |
6503 | { | |
6504 | int ret; | |
6505 | ||
c2c71324 SB |
6506 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { |
6507 | ret = send_header(sctx); | |
6508 | if (ret < 0) | |
6509 | goto out; | |
6510 | } | |
31db9f7c AB |
6511 | |
6512 | ret = send_subvol_begin(sctx); | |
6513 | if (ret < 0) | |
6514 | goto out; | |
6515 | ||
6516 | if (sctx->parent_root) { | |
6517 | ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, | |
6518 | changed_cb, sctx); | |
6519 | if (ret < 0) | |
6520 | goto out; | |
6521 | ret = finish_inode_if_needed(sctx, 1); | |
6522 | if (ret < 0) | |
6523 | goto out; | |
6524 | } else { | |
6525 | ret = full_send_tree(sctx); | |
6526 | if (ret < 0) | |
6527 | goto out; | |
6528 | } | |
6529 | ||
6530 | out: | |
31db9f7c AB |
6531 | free_recorded_refs(sctx); |
6532 | return ret; | |
6533 | } | |
6534 | ||
e5fa8f86 FM |
6535 | /* |
6536 | * If orphan cleanup did remove any orphans from a root, it means the tree | |
6537 | * was modified and therefore the commit root is not the same as the current | |
6538 | * root anymore. This is a problem, because send uses the commit root and | |
6539 | * therefore can see inode items that don't exist in the current root anymore, | |
6540 | * and for example make calls to btrfs_iget, which will do tree lookups based | |
6541 | * on the current root and not on the commit root. Those lookups will fail, | |
6542 | * returning a -ESTALE error, and making send fail with that error. So make | |
6543 | * sure a send does not see any orphans we have just removed, and that it will | |
6544 | * see the same inodes regardless of whether a transaction commit happened | |
6545 | * before it started (meaning that the commit root will be the same as the | |
6546 | * current root) or not. | |
6547 | */ | |
6548 | static int ensure_commit_roots_uptodate(struct send_ctx *sctx) | |
6549 | { | |
6550 | int i; | |
6551 | struct btrfs_trans_handle *trans = NULL; | |
6552 | ||
6553 | again: | |
6554 | if (sctx->parent_root && | |
6555 | sctx->parent_root->node != sctx->parent_root->commit_root) | |
6556 | goto commit_trans; | |
6557 | ||
6558 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6559 | if (sctx->clone_roots[i].root->node != | |
6560 | sctx->clone_roots[i].root->commit_root) | |
6561 | goto commit_trans; | |
6562 | ||
6563 | if (trans) | |
3a45bb20 | 6564 | return btrfs_end_transaction(trans); |
e5fa8f86 FM |
6565 | |
6566 | return 0; | |
6567 | ||
6568 | commit_trans: | |
6569 | /* Use any root, all fs roots will get their commit roots updated. */ | |
6570 | if (!trans) { | |
6571 | trans = btrfs_join_transaction(sctx->send_root); | |
6572 | if (IS_ERR(trans)) | |
6573 | return PTR_ERR(trans); | |
6574 | goto again; | |
6575 | } | |
6576 | ||
3a45bb20 | 6577 | return btrfs_commit_transaction(trans); |
e5fa8f86 FM |
6578 | } |
6579 | ||
66ef7d65 DS |
6580 | static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) |
6581 | { | |
6582 | spin_lock(&root->root_item_lock); | |
6583 | root->send_in_progress--; | |
6584 | /* | |
6585 | * Not much left to do, we don't know why it's unbalanced and | |
6586 | * can't blindly reset it to 0. | |
6587 | */ | |
6588 | if (root->send_in_progress < 0) | |
6589 | btrfs_err(root->fs_info, | |
0b246afa JM |
6590 | "send_in_progres unbalanced %d root %llu", |
6591 | root->send_in_progress, root->root_key.objectid); | |
66ef7d65 DS |
6592 | spin_unlock(&root->root_item_lock); |
6593 | } | |
6594 | ||
2351f431 | 6595 | long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) |
31db9f7c AB |
6596 | { |
6597 | int ret = 0; | |
0b246afa JM |
6598 | struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root; |
6599 | struct btrfs_fs_info *fs_info = send_root->fs_info; | |
31db9f7c | 6600 | struct btrfs_root *clone_root; |
31db9f7c | 6601 | struct btrfs_key key; |
31db9f7c AB |
6602 | struct send_ctx *sctx = NULL; |
6603 | u32 i; | |
6604 | u64 *clone_sources_tmp = NULL; | |
2c686537 | 6605 | int clone_sources_to_rollback = 0; |
e55d1153 | 6606 | unsigned alloc_size; |
896c14f9 | 6607 | int sort_clone_roots = 0; |
18f687d5 | 6608 | int index; |
31db9f7c AB |
6609 | |
6610 | if (!capable(CAP_SYS_ADMIN)) | |
6611 | return -EPERM; | |
6612 | ||
2c686537 DS |
6613 | /* |
6614 | * The subvolume must remain read-only during send, protect against | |
521e0546 | 6615 | * making it RW. This also protects against deletion. |
2c686537 DS |
6616 | */ |
6617 | spin_lock(&send_root->root_item_lock); | |
6618 | send_root->send_in_progress++; | |
6619 | spin_unlock(&send_root->root_item_lock); | |
6620 | ||
139f807a JB |
6621 | /* |
6622 | * This is done when we lookup the root, it should already be complete | |
6623 | * by the time we get here. | |
6624 | */ | |
6625 | WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE); | |
6626 | ||
2c686537 DS |
6627 | /* |
6628 | * Userspace tools do the checks and warn the user if it's | |
6629 | * not RO. | |
6630 | */ | |
6631 | if (!btrfs_root_readonly(send_root)) { | |
6632 | ret = -EPERM; | |
6633 | goto out; | |
6634 | } | |
6635 | ||
457ae726 DC |
6636 | /* |
6637 | * Check that we don't overflow at later allocations, we request | |
6638 | * clone_sources_count + 1 items, and compare to unsigned long inside | |
6639 | * access_ok. | |
6640 | */ | |
f5ecec3c | 6641 | if (arg->clone_sources_count > |
457ae726 | 6642 | ULONG_MAX / sizeof(struct clone_root) - 1) { |
f5ecec3c DC |
6643 | ret = -EINVAL; |
6644 | goto out; | |
6645 | } | |
6646 | ||
31db9f7c | 6647 | if (!access_ok(VERIFY_READ, arg->clone_sources, |
700ff4f0 DC |
6648 | sizeof(*arg->clone_sources) * |
6649 | arg->clone_sources_count)) { | |
31db9f7c AB |
6650 | ret = -EFAULT; |
6651 | goto out; | |
6652 | } | |
6653 | ||
c2c71324 | 6654 | if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { |
cb95e7bf MF |
6655 | ret = -EINVAL; |
6656 | goto out; | |
6657 | } | |
6658 | ||
e780b0d1 | 6659 | sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL); |
31db9f7c AB |
6660 | if (!sctx) { |
6661 | ret = -ENOMEM; | |
6662 | goto out; | |
6663 | } | |
6664 | ||
6665 | INIT_LIST_HEAD(&sctx->new_refs); | |
6666 | INIT_LIST_HEAD(&sctx->deleted_refs); | |
e780b0d1 | 6667 | INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL); |
31db9f7c AB |
6668 | INIT_LIST_HEAD(&sctx->name_cache_list); |
6669 | ||
cb95e7bf MF |
6670 | sctx->flags = arg->flags; |
6671 | ||
31db9f7c | 6672 | sctx->send_filp = fget(arg->send_fd); |
ecc7ada7 TI |
6673 | if (!sctx->send_filp) { |
6674 | ret = -EBADF; | |
31db9f7c AB |
6675 | goto out; |
6676 | } | |
6677 | ||
31db9f7c | 6678 | sctx->send_root = send_root; |
521e0546 DS |
6679 | /* |
6680 | * Unlikely but possible, if the subvolume is marked for deletion but | |
6681 | * is slow to remove the directory entry, send can still be started | |
6682 | */ | |
6683 | if (btrfs_root_dead(sctx->send_root)) { | |
6684 | ret = -EPERM; | |
6685 | goto out; | |
6686 | } | |
6687 | ||
31db9f7c AB |
6688 | sctx->clone_roots_cnt = arg->clone_sources_count; |
6689 | ||
6690 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE; | |
752ade68 | 6691 | sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); |
31db9f7c | 6692 | if (!sctx->send_buf) { |
752ade68 MH |
6693 | ret = -ENOMEM; |
6694 | goto out; | |
31db9f7c AB |
6695 | } |
6696 | ||
752ade68 | 6697 | sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL); |
31db9f7c | 6698 | if (!sctx->read_buf) { |
752ade68 MH |
6699 | ret = -ENOMEM; |
6700 | goto out; | |
31db9f7c AB |
6701 | } |
6702 | ||
9f03740a FDBM |
6703 | sctx->pending_dir_moves = RB_ROOT; |
6704 | sctx->waiting_dir_moves = RB_ROOT; | |
9dc44214 | 6705 | sctx->orphan_dirs = RB_ROOT; |
9f03740a | 6706 | |
e55d1153 DS |
6707 | alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1); |
6708 | ||
818e010b | 6709 | sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 6710 | if (!sctx->clone_roots) { |
818e010b DS |
6711 | ret = -ENOMEM; |
6712 | goto out; | |
31db9f7c AB |
6713 | } |
6714 | ||
e55d1153 DS |
6715 | alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources); |
6716 | ||
31db9f7c | 6717 | if (arg->clone_sources_count) { |
752ade68 | 6718 | clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 6719 | if (!clone_sources_tmp) { |
752ade68 MH |
6720 | ret = -ENOMEM; |
6721 | goto out; | |
31db9f7c AB |
6722 | } |
6723 | ||
6724 | ret = copy_from_user(clone_sources_tmp, arg->clone_sources, | |
e55d1153 | 6725 | alloc_size); |
31db9f7c AB |
6726 | if (ret) { |
6727 | ret = -EFAULT; | |
6728 | goto out; | |
6729 | } | |
6730 | ||
6731 | for (i = 0; i < arg->clone_sources_count; i++) { | |
6732 | key.objectid = clone_sources_tmp[i]; | |
6733 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6734 | key.offset = (u64)-1; | |
18f687d5 WS |
6735 | |
6736 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6737 | ||
31db9f7c | 6738 | clone_root = btrfs_read_fs_root_no_name(fs_info, &key); |
31db9f7c | 6739 | if (IS_ERR(clone_root)) { |
18f687d5 | 6740 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
31db9f7c AB |
6741 | ret = PTR_ERR(clone_root); |
6742 | goto out; | |
6743 | } | |
2c686537 | 6744 | spin_lock(&clone_root->root_item_lock); |
5cc2b17e FM |
6745 | if (!btrfs_root_readonly(clone_root) || |
6746 | btrfs_root_dead(clone_root)) { | |
2c686537 | 6747 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 | 6748 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6749 | ret = -EPERM; |
6750 | goto out; | |
6751 | } | |
2f1f465a | 6752 | clone_root->send_in_progress++; |
2c686537 | 6753 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 WS |
6754 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
6755 | ||
31db9f7c | 6756 | sctx->clone_roots[i].root = clone_root; |
2f1f465a | 6757 | clone_sources_to_rollback = i + 1; |
31db9f7c | 6758 | } |
2f91306a | 6759 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6760 | clone_sources_tmp = NULL; |
6761 | } | |
6762 | ||
6763 | if (arg->parent_root) { | |
6764 | key.objectid = arg->parent_root; | |
6765 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6766 | key.offset = (u64)-1; | |
18f687d5 WS |
6767 | |
6768 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6769 | ||
31db9f7c | 6770 | sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key); |
b1b19596 | 6771 | if (IS_ERR(sctx->parent_root)) { |
18f687d5 | 6772 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
b1b19596 | 6773 | ret = PTR_ERR(sctx->parent_root); |
31db9f7c AB |
6774 | goto out; |
6775 | } | |
18f687d5 | 6776 | |
2c686537 DS |
6777 | spin_lock(&sctx->parent_root->root_item_lock); |
6778 | sctx->parent_root->send_in_progress++; | |
521e0546 DS |
6779 | if (!btrfs_root_readonly(sctx->parent_root) || |
6780 | btrfs_root_dead(sctx->parent_root)) { | |
2c686537 | 6781 | spin_unlock(&sctx->parent_root->root_item_lock); |
18f687d5 | 6782 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6783 | ret = -EPERM; |
6784 | goto out; | |
6785 | } | |
6786 | spin_unlock(&sctx->parent_root->root_item_lock); | |
18f687d5 WS |
6787 | |
6788 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
31db9f7c AB |
6789 | } |
6790 | ||
6791 | /* | |
6792 | * Clones from send_root are allowed, but only if the clone source | |
6793 | * is behind the current send position. This is checked while searching | |
6794 | * for possible clone sources. | |
6795 | */ | |
6796 | sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root; | |
6797 | ||
6798 | /* We do a bsearch later */ | |
6799 | sort(sctx->clone_roots, sctx->clone_roots_cnt, | |
6800 | sizeof(*sctx->clone_roots), __clone_root_cmp_sort, | |
6801 | NULL); | |
896c14f9 | 6802 | sort_clone_roots = 1; |
31db9f7c | 6803 | |
e5fa8f86 FM |
6804 | ret = ensure_commit_roots_uptodate(sctx); |
6805 | if (ret) | |
6806 | goto out; | |
6807 | ||
2755a0de | 6808 | current->journal_info = BTRFS_SEND_TRANS_STUB; |
31db9f7c | 6809 | ret = send_subvol(sctx); |
a26e8c9f | 6810 | current->journal_info = NULL; |
31db9f7c AB |
6811 | if (ret < 0) |
6812 | goto out; | |
6813 | ||
c2c71324 SB |
6814 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { |
6815 | ret = begin_cmd(sctx, BTRFS_SEND_C_END); | |
6816 | if (ret < 0) | |
6817 | goto out; | |
6818 | ret = send_cmd(sctx); | |
6819 | if (ret < 0) | |
6820 | goto out; | |
6821 | } | |
31db9f7c AB |
6822 | |
6823 | out: | |
9f03740a FDBM |
6824 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); |
6825 | while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { | |
6826 | struct rb_node *n; | |
6827 | struct pending_dir_move *pm; | |
6828 | ||
6829 | n = rb_first(&sctx->pending_dir_moves); | |
6830 | pm = rb_entry(n, struct pending_dir_move, node); | |
6831 | while (!list_empty(&pm->list)) { | |
6832 | struct pending_dir_move *pm2; | |
6833 | ||
6834 | pm2 = list_first_entry(&pm->list, | |
6835 | struct pending_dir_move, list); | |
6836 | free_pending_move(sctx, pm2); | |
6837 | } | |
6838 | free_pending_move(sctx, pm); | |
6839 | } | |
6840 | ||
6841 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); | |
6842 | while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { | |
6843 | struct rb_node *n; | |
6844 | struct waiting_dir_move *dm; | |
6845 | ||
6846 | n = rb_first(&sctx->waiting_dir_moves); | |
6847 | dm = rb_entry(n, struct waiting_dir_move, node); | |
6848 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
6849 | kfree(dm); | |
6850 | } | |
6851 | ||
9dc44214 FM |
6852 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); |
6853 | while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { | |
6854 | struct rb_node *n; | |
6855 | struct orphan_dir_info *odi; | |
6856 | ||
6857 | n = rb_first(&sctx->orphan_dirs); | |
6858 | odi = rb_entry(n, struct orphan_dir_info, node); | |
6859 | free_orphan_dir_info(sctx, odi); | |
6860 | } | |
6861 | ||
896c14f9 WS |
6862 | if (sort_clone_roots) { |
6863 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6864 | btrfs_root_dec_send_in_progress( | |
6865 | sctx->clone_roots[i].root); | |
6866 | } else { | |
6867 | for (i = 0; sctx && i < clone_sources_to_rollback; i++) | |
6868 | btrfs_root_dec_send_in_progress( | |
6869 | sctx->clone_roots[i].root); | |
6870 | ||
6871 | btrfs_root_dec_send_in_progress(send_root); | |
6872 | } | |
66ef7d65 DS |
6873 | if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) |
6874 | btrfs_root_dec_send_in_progress(sctx->parent_root); | |
2c686537 | 6875 | |
2f91306a | 6876 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6877 | |
6878 | if (sctx) { | |
6879 | if (sctx->send_filp) | |
6880 | fput(sctx->send_filp); | |
6881 | ||
c03d01f3 | 6882 | kvfree(sctx->clone_roots); |
6ff48ce0 | 6883 | kvfree(sctx->send_buf); |
eb5b75fe | 6884 | kvfree(sctx->read_buf); |
31db9f7c AB |
6885 | |
6886 | name_cache_free(sctx); | |
6887 | ||
6888 | kfree(sctx); | |
6889 | } | |
6890 | ||
6891 | return ret; | |
6892 | } |