]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
31db9f7c AB |
2 | /* |
3 | * Copyright (C) 2012 Alexander Block. All rights reserved. | |
31db9f7c AB |
4 | */ |
5 | ||
6 | #include <linux/bsearch.h> | |
7 | #include <linux/fs.h> | |
8 | #include <linux/file.h> | |
9 | #include <linux/sort.h> | |
10 | #include <linux/mount.h> | |
11 | #include <linux/xattr.h> | |
12 | #include <linux/posix_acl_xattr.h> | |
13 | #include <linux/radix-tree.h> | |
a1857ebe | 14 | #include <linux/vmalloc.h> |
ed84885d | 15 | #include <linux/string.h> |
2351f431 | 16 | #include <linux/compat.h> |
9678c543 | 17 | #include <linux/crc32c.h> |
31db9f7c AB |
18 | |
19 | #include "send.h" | |
20 | #include "backref.h" | |
21 | #include "locking.h" | |
22 | #include "disk-io.h" | |
23 | #include "btrfs_inode.h" | |
24 | #include "transaction.h" | |
ebb8765b | 25 | #include "compression.h" |
89efda52 | 26 | #include "xattr.h" |
31db9f7c | 27 | |
fd0ddbe2 FM |
28 | /* |
29 | * Maximum number of references an extent can have in order for us to attempt to | |
30 | * issue clone operations instead of write operations. This currently exists to | |
31 | * avoid hitting limitations of the backreference walking code (taking a lot of | |
32 | * time and using too much memory for extents with large number of references). | |
33 | */ | |
34 | #define SEND_MAX_EXTENT_REFS 64 | |
35 | ||
31db9f7c AB |
36 | /* |
37 | * A fs_path is a helper to dynamically build path names with unknown size. | |
38 | * It reallocates the internal buffer on demand. | |
39 | * It allows fast adding of path elements on the right side (normal path) and | |
40 | * fast adding to the left side (reversed path). A reversed path can also be | |
41 | * unreversed if needed. | |
42 | */ | |
43 | struct fs_path { | |
44 | union { | |
45 | struct { | |
46 | char *start; | |
47 | char *end; | |
31db9f7c AB |
48 | |
49 | char *buf; | |
1f5a7ff9 DS |
50 | unsigned short buf_len:15; |
51 | unsigned short reversed:1; | |
31db9f7c AB |
52 | char inline_buf[]; |
53 | }; | |
ace01050 DS |
54 | /* |
55 | * Average path length does not exceed 200 bytes, we'll have | |
56 | * better packing in the slab and higher chance to satisfy | |
57 | * a allocation later during send. | |
58 | */ | |
59 | char pad[256]; | |
31db9f7c AB |
60 | }; |
61 | }; | |
62 | #define FS_PATH_INLINE_SIZE \ | |
63 | (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) | |
64 | ||
65 | ||
66 | /* reused for each extent */ | |
67 | struct clone_root { | |
68 | struct btrfs_root *root; | |
69 | u64 ino; | |
70 | u64 offset; | |
71 | ||
72 | u64 found_refs; | |
73 | }; | |
74 | ||
75 | #define SEND_CTX_MAX_NAME_CACHE_SIZE 128 | |
76 | #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2) | |
77 | ||
78 | struct send_ctx { | |
79 | struct file *send_filp; | |
80 | loff_t send_off; | |
81 | char *send_buf; | |
82 | u32 send_size; | |
83 | u32 send_max_size; | |
84 | u64 total_send_size; | |
85 | u64 cmd_send_size[BTRFS_SEND_C_MAX + 1]; | |
cb95e7bf | 86 | u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ |
31db9f7c | 87 | |
31db9f7c AB |
88 | struct btrfs_root *send_root; |
89 | struct btrfs_root *parent_root; | |
90 | struct clone_root *clone_roots; | |
91 | int clone_roots_cnt; | |
92 | ||
93 | /* current state of the compare_tree call */ | |
94 | struct btrfs_path *left_path; | |
95 | struct btrfs_path *right_path; | |
96 | struct btrfs_key *cmp_key; | |
97 | ||
98 | /* | |
99 | * infos of the currently processed inode. In case of deleted inodes, | |
100 | * these are the values from the deleted inode. | |
101 | */ | |
102 | u64 cur_ino; | |
103 | u64 cur_inode_gen; | |
104 | int cur_inode_new; | |
105 | int cur_inode_new_gen; | |
106 | int cur_inode_deleted; | |
31db9f7c AB |
107 | u64 cur_inode_size; |
108 | u64 cur_inode_mode; | |
644d1940 | 109 | u64 cur_inode_rdev; |
16e7549f | 110 | u64 cur_inode_last_extent; |
ffa7c429 | 111 | u64 cur_inode_next_write_offset; |
46b2f459 | 112 | bool ignore_cur_inode; |
31db9f7c AB |
113 | |
114 | u64 send_progress; | |
115 | ||
116 | struct list_head new_refs; | |
117 | struct list_head deleted_refs; | |
118 | ||
119 | struct radix_tree_root name_cache; | |
120 | struct list_head name_cache_list; | |
121 | int name_cache_size; | |
122 | ||
2131bcd3 LB |
123 | struct file_ra_state ra; |
124 | ||
9f03740a FDBM |
125 | /* |
126 | * We process inodes by their increasing order, so if before an | |
127 | * incremental send we reverse the parent/child relationship of | |
128 | * directories such that a directory with a lower inode number was | |
129 | * the parent of a directory with a higher inode number, and the one | |
130 | * becoming the new parent got renamed too, we can't rename/move the | |
131 | * directory with lower inode number when we finish processing it - we | |
132 | * must process the directory with higher inode number first, then | |
133 | * rename/move it and then rename/move the directory with lower inode | |
134 | * number. Example follows. | |
135 | * | |
136 | * Tree state when the first send was performed: | |
137 | * | |
138 | * . | |
139 | * |-- a (ino 257) | |
140 | * |-- b (ino 258) | |
141 | * | | |
142 | * | | |
143 | * |-- c (ino 259) | |
144 | * | |-- d (ino 260) | |
145 | * | | |
146 | * |-- c2 (ino 261) | |
147 | * | |
148 | * Tree state when the second (incremental) send is performed: | |
149 | * | |
150 | * . | |
151 | * |-- a (ino 257) | |
152 | * |-- b (ino 258) | |
153 | * |-- c2 (ino 261) | |
154 | * |-- d2 (ino 260) | |
155 | * |-- cc (ino 259) | |
156 | * | |
157 | * The sequence of steps that lead to the second state was: | |
158 | * | |
159 | * mv /a/b/c/d /a/b/c2/d2 | |
160 | * mv /a/b/c /a/b/c2/d2/cc | |
161 | * | |
162 | * "c" has lower inode number, but we can't move it (2nd mv operation) | |
163 | * before we move "d", which has higher inode number. | |
164 | * | |
165 | * So we just memorize which move/rename operations must be performed | |
166 | * later when their respective parent is processed and moved/renamed. | |
167 | */ | |
168 | ||
169 | /* Indexed by parent directory inode number. */ | |
170 | struct rb_root pending_dir_moves; | |
171 | ||
172 | /* | |
173 | * Reverse index, indexed by the inode number of a directory that | |
174 | * is waiting for the move/rename of its immediate parent before its | |
175 | * own move/rename can be performed. | |
176 | */ | |
177 | struct rb_root waiting_dir_moves; | |
9dc44214 FM |
178 | |
179 | /* | |
180 | * A directory that is going to be rm'ed might have a child directory | |
181 | * which is in the pending directory moves index above. In this case, | |
182 | * the directory can only be removed after the move/rename of its child | |
183 | * is performed. Example: | |
184 | * | |
185 | * Parent snapshot: | |
186 | * | |
187 | * . (ino 256) | |
188 | * |-- a/ (ino 257) | |
189 | * |-- b/ (ino 258) | |
190 | * |-- c/ (ino 259) | |
191 | * | |-- x/ (ino 260) | |
192 | * | | |
193 | * |-- y/ (ino 261) | |
194 | * | |
195 | * Send snapshot: | |
196 | * | |
197 | * . (ino 256) | |
198 | * |-- a/ (ino 257) | |
199 | * |-- b/ (ino 258) | |
200 | * |-- YY/ (ino 261) | |
201 | * |-- x/ (ino 260) | |
202 | * | |
203 | * Sequence of steps that lead to the send snapshot: | |
204 | * rm -f /a/b/c/foo.txt | |
205 | * mv /a/b/y /a/b/YY | |
206 | * mv /a/b/c/x /a/b/YY | |
207 | * rmdir /a/b/c | |
208 | * | |
209 | * When the child is processed, its move/rename is delayed until its | |
210 | * parent is processed (as explained above), but all other operations | |
211 | * like update utimes, chown, chgrp, etc, are performed and the paths | |
212 | * that it uses for those operations must use the orphanized name of | |
213 | * its parent (the directory we're going to rm later), so we need to | |
214 | * memorize that name. | |
215 | * | |
216 | * Indexed by the inode number of the directory to be deleted. | |
217 | */ | |
218 | struct rb_root orphan_dirs; | |
9f03740a FDBM |
219 | }; |
220 | ||
221 | struct pending_dir_move { | |
222 | struct rb_node node; | |
223 | struct list_head list; | |
224 | u64 parent_ino; | |
225 | u64 ino; | |
226 | u64 gen; | |
227 | struct list_head update_refs; | |
228 | }; | |
229 | ||
230 | struct waiting_dir_move { | |
231 | struct rb_node node; | |
232 | u64 ino; | |
9dc44214 FM |
233 | /* |
234 | * There might be some directory that could not be removed because it | |
235 | * was waiting for this directory inode to be moved first. Therefore | |
236 | * after this directory is moved, we can try to rmdir the ino rmdir_ino. | |
237 | */ | |
238 | u64 rmdir_ino; | |
0b3f407e | 239 | u64 rmdir_gen; |
8b191a68 | 240 | bool orphanized; |
9dc44214 FM |
241 | }; |
242 | ||
243 | struct orphan_dir_info { | |
244 | struct rb_node node; | |
245 | u64 ino; | |
246 | u64 gen; | |
0f96f517 | 247 | u64 last_dir_index_offset; |
31db9f7c AB |
248 | }; |
249 | ||
250 | struct name_cache_entry { | |
251 | struct list_head list; | |
7e0926fe AB |
252 | /* |
253 | * radix_tree has only 32bit entries but we need to handle 64bit inums. | |
254 | * We use the lower 32bit of the 64bit inum to store it in the tree. If | |
255 | * more then one inum would fall into the same entry, we use radix_list | |
256 | * to store the additional entries. radix_list is also used to store | |
257 | * entries where two entries have the same inum but different | |
258 | * generations. | |
259 | */ | |
260 | struct list_head radix_list; | |
31db9f7c AB |
261 | u64 ino; |
262 | u64 gen; | |
263 | u64 parent_ino; | |
264 | u64 parent_gen; | |
265 | int ret; | |
266 | int need_later_update; | |
267 | int name_len; | |
268 | char name[]; | |
269 | }; | |
270 | ||
18d0f5c6 DS |
271 | #define ADVANCE 1 |
272 | #define ADVANCE_ONLY_NEXT -1 | |
273 | ||
274 | enum btrfs_compare_tree_result { | |
275 | BTRFS_COMPARE_TREE_NEW, | |
276 | BTRFS_COMPARE_TREE_DELETED, | |
277 | BTRFS_COMPARE_TREE_CHANGED, | |
278 | BTRFS_COMPARE_TREE_SAME, | |
279 | }; | |
18d0f5c6 | 280 | |
e67c718b | 281 | __cold |
95155585 FM |
282 | static void inconsistent_snapshot_error(struct send_ctx *sctx, |
283 | enum btrfs_compare_tree_result result, | |
284 | const char *what) | |
285 | { | |
286 | const char *result_string; | |
287 | ||
288 | switch (result) { | |
289 | case BTRFS_COMPARE_TREE_NEW: | |
290 | result_string = "new"; | |
291 | break; | |
292 | case BTRFS_COMPARE_TREE_DELETED: | |
293 | result_string = "deleted"; | |
294 | break; | |
295 | case BTRFS_COMPARE_TREE_CHANGED: | |
296 | result_string = "updated"; | |
297 | break; | |
298 | case BTRFS_COMPARE_TREE_SAME: | |
299 | ASSERT(0); | |
300 | result_string = "unchanged"; | |
301 | break; | |
302 | default: | |
303 | ASSERT(0); | |
304 | result_string = "unexpected"; | |
305 | } | |
306 | ||
307 | btrfs_err(sctx->send_root->fs_info, | |
308 | "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu", | |
309 | result_string, what, sctx->cmp_key->objectid, | |
310 | sctx->send_root->root_key.objectid, | |
311 | (sctx->parent_root ? | |
312 | sctx->parent_root->root_key.objectid : 0)); | |
313 | } | |
314 | ||
9f03740a FDBM |
315 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); |
316 | ||
9dc44214 FM |
317 | static struct waiting_dir_move * |
318 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino); | |
319 | ||
0b3f407e | 320 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen); |
9dc44214 | 321 | |
16e7549f JB |
322 | static int need_send_hole(struct send_ctx *sctx) |
323 | { | |
324 | return (sctx->parent_root && !sctx->cur_inode_new && | |
325 | !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && | |
326 | S_ISREG(sctx->cur_inode_mode)); | |
327 | } | |
328 | ||
31db9f7c AB |
329 | static void fs_path_reset(struct fs_path *p) |
330 | { | |
331 | if (p->reversed) { | |
332 | p->start = p->buf + p->buf_len - 1; | |
333 | p->end = p->start; | |
334 | *p->start = 0; | |
335 | } else { | |
336 | p->start = p->buf; | |
337 | p->end = p->start; | |
338 | *p->start = 0; | |
339 | } | |
340 | } | |
341 | ||
924794c9 | 342 | static struct fs_path *fs_path_alloc(void) |
31db9f7c AB |
343 | { |
344 | struct fs_path *p; | |
345 | ||
e780b0d1 | 346 | p = kmalloc(sizeof(*p), GFP_KERNEL); |
31db9f7c AB |
347 | if (!p) |
348 | return NULL; | |
349 | p->reversed = 0; | |
31db9f7c AB |
350 | p->buf = p->inline_buf; |
351 | p->buf_len = FS_PATH_INLINE_SIZE; | |
352 | fs_path_reset(p); | |
353 | return p; | |
354 | } | |
355 | ||
924794c9 | 356 | static struct fs_path *fs_path_alloc_reversed(void) |
31db9f7c AB |
357 | { |
358 | struct fs_path *p; | |
359 | ||
924794c9 | 360 | p = fs_path_alloc(); |
31db9f7c AB |
361 | if (!p) |
362 | return NULL; | |
363 | p->reversed = 1; | |
364 | fs_path_reset(p); | |
365 | return p; | |
366 | } | |
367 | ||
924794c9 | 368 | static void fs_path_free(struct fs_path *p) |
31db9f7c AB |
369 | { |
370 | if (!p) | |
371 | return; | |
ace01050 DS |
372 | if (p->buf != p->inline_buf) |
373 | kfree(p->buf); | |
31db9f7c AB |
374 | kfree(p); |
375 | } | |
376 | ||
377 | static int fs_path_len(struct fs_path *p) | |
378 | { | |
379 | return p->end - p->start; | |
380 | } | |
381 | ||
382 | static int fs_path_ensure_buf(struct fs_path *p, int len) | |
383 | { | |
384 | char *tmp_buf; | |
385 | int path_len; | |
386 | int old_buf_len; | |
387 | ||
388 | len++; | |
389 | ||
390 | if (p->buf_len >= len) | |
391 | return 0; | |
392 | ||
cfd4a535 CM |
393 | if (len > PATH_MAX) { |
394 | WARN_ON(1); | |
395 | return -ENOMEM; | |
396 | } | |
397 | ||
1b2782c8 DS |
398 | path_len = p->end - p->start; |
399 | old_buf_len = p->buf_len; | |
400 | ||
ace01050 DS |
401 | /* |
402 | * First time the inline_buf does not suffice | |
403 | */ | |
01a9a8a9 | 404 | if (p->buf == p->inline_buf) { |
e780b0d1 | 405 | tmp_buf = kmalloc(len, GFP_KERNEL); |
01a9a8a9 FM |
406 | if (tmp_buf) |
407 | memcpy(tmp_buf, p->buf, old_buf_len); | |
408 | } else { | |
e780b0d1 | 409 | tmp_buf = krealloc(p->buf, len, GFP_KERNEL); |
01a9a8a9 | 410 | } |
9c9ca00b DS |
411 | if (!tmp_buf) |
412 | return -ENOMEM; | |
413 | p->buf = tmp_buf; | |
414 | /* | |
415 | * The real size of the buffer is bigger, this will let the fast path | |
416 | * happen most of the time | |
417 | */ | |
418 | p->buf_len = ksize(p->buf); | |
ace01050 | 419 | |
31db9f7c AB |
420 | if (p->reversed) { |
421 | tmp_buf = p->buf + old_buf_len - path_len - 1; | |
422 | p->end = p->buf + p->buf_len - 1; | |
423 | p->start = p->end - path_len; | |
424 | memmove(p->start, tmp_buf, path_len + 1); | |
425 | } else { | |
426 | p->start = p->buf; | |
427 | p->end = p->start + path_len; | |
428 | } | |
429 | return 0; | |
430 | } | |
431 | ||
b23ab57d DS |
432 | static int fs_path_prepare_for_add(struct fs_path *p, int name_len, |
433 | char **prepared) | |
31db9f7c AB |
434 | { |
435 | int ret; | |
436 | int new_len; | |
437 | ||
438 | new_len = p->end - p->start + name_len; | |
439 | if (p->start != p->end) | |
440 | new_len++; | |
441 | ret = fs_path_ensure_buf(p, new_len); | |
442 | if (ret < 0) | |
443 | goto out; | |
444 | ||
445 | if (p->reversed) { | |
446 | if (p->start != p->end) | |
447 | *--p->start = '/'; | |
448 | p->start -= name_len; | |
b23ab57d | 449 | *prepared = p->start; |
31db9f7c AB |
450 | } else { |
451 | if (p->start != p->end) | |
452 | *p->end++ = '/'; | |
b23ab57d | 453 | *prepared = p->end; |
31db9f7c AB |
454 | p->end += name_len; |
455 | *p->end = 0; | |
456 | } | |
457 | ||
458 | out: | |
459 | return ret; | |
460 | } | |
461 | ||
462 | static int fs_path_add(struct fs_path *p, const char *name, int name_len) | |
463 | { | |
464 | int ret; | |
b23ab57d | 465 | char *prepared; |
31db9f7c | 466 | |
b23ab57d | 467 | ret = fs_path_prepare_for_add(p, name_len, &prepared); |
31db9f7c AB |
468 | if (ret < 0) |
469 | goto out; | |
b23ab57d | 470 | memcpy(prepared, name, name_len); |
31db9f7c AB |
471 | |
472 | out: | |
473 | return ret; | |
474 | } | |
475 | ||
476 | static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) | |
477 | { | |
478 | int ret; | |
b23ab57d | 479 | char *prepared; |
31db9f7c | 480 | |
b23ab57d | 481 | ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); |
31db9f7c AB |
482 | if (ret < 0) |
483 | goto out; | |
b23ab57d | 484 | memcpy(prepared, p2->start, p2->end - p2->start); |
31db9f7c AB |
485 | |
486 | out: | |
487 | return ret; | |
488 | } | |
489 | ||
490 | static int fs_path_add_from_extent_buffer(struct fs_path *p, | |
491 | struct extent_buffer *eb, | |
492 | unsigned long off, int len) | |
493 | { | |
494 | int ret; | |
b23ab57d | 495 | char *prepared; |
31db9f7c | 496 | |
b23ab57d | 497 | ret = fs_path_prepare_for_add(p, len, &prepared); |
31db9f7c AB |
498 | if (ret < 0) |
499 | goto out; | |
500 | ||
b23ab57d | 501 | read_extent_buffer(eb, prepared, off, len); |
31db9f7c AB |
502 | |
503 | out: | |
504 | return ret; | |
505 | } | |
506 | ||
31db9f7c AB |
507 | static int fs_path_copy(struct fs_path *p, struct fs_path *from) |
508 | { | |
509 | int ret; | |
510 | ||
511 | p->reversed = from->reversed; | |
512 | fs_path_reset(p); | |
513 | ||
514 | ret = fs_path_add_path(p, from); | |
515 | ||
516 | return ret; | |
517 | } | |
518 | ||
519 | ||
520 | static void fs_path_unreverse(struct fs_path *p) | |
521 | { | |
522 | char *tmp; | |
523 | int len; | |
524 | ||
525 | if (!p->reversed) | |
526 | return; | |
527 | ||
528 | tmp = p->start; | |
529 | len = p->end - p->start; | |
530 | p->start = p->buf; | |
531 | p->end = p->start + len; | |
532 | memmove(p->start, tmp, len + 1); | |
533 | p->reversed = 0; | |
534 | } | |
535 | ||
536 | static struct btrfs_path *alloc_path_for_send(void) | |
537 | { | |
538 | struct btrfs_path *path; | |
539 | ||
540 | path = btrfs_alloc_path(); | |
541 | if (!path) | |
542 | return NULL; | |
543 | path->search_commit_root = 1; | |
544 | path->skip_locking = 1; | |
3f8a18cc | 545 | path->need_commit_sem = 1; |
31db9f7c AB |
546 | return path; |
547 | } | |
548 | ||
48a3b636 | 549 | static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) |
31db9f7c AB |
550 | { |
551 | int ret; | |
31db9f7c AB |
552 | u32 pos = 0; |
553 | ||
31db9f7c | 554 | while (pos < len) { |
8e93157b | 555 | ret = kernel_write(filp, buf + pos, len - pos, off); |
31db9f7c AB |
556 | /* TODO handle that correctly */ |
557 | /*if (ret == -ERESTARTSYS) { | |
558 | continue; | |
559 | }*/ | |
560 | if (ret < 0) | |
8e93157b | 561 | return ret; |
31db9f7c | 562 | if (ret == 0) { |
8e93157b | 563 | return -EIO; |
31db9f7c AB |
564 | } |
565 | pos += ret; | |
566 | } | |
567 | ||
8e93157b | 568 | return 0; |
31db9f7c AB |
569 | } |
570 | ||
571 | static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) | |
572 | { | |
573 | struct btrfs_tlv_header *hdr; | |
574 | int total_len = sizeof(*hdr) + len; | |
575 | int left = sctx->send_max_size - sctx->send_size; | |
576 | ||
577 | if (unlikely(left < total_len)) | |
578 | return -EOVERFLOW; | |
579 | ||
580 | hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); | |
e2f896b3 DS |
581 | put_unaligned_le16(attr, &hdr->tlv_type); |
582 | put_unaligned_le16(len, &hdr->tlv_len); | |
31db9f7c AB |
583 | memcpy(hdr + 1, data, len); |
584 | sctx->send_size += total_len; | |
585 | ||
586 | return 0; | |
587 | } | |
588 | ||
95bc79d5 DS |
589 | #define TLV_PUT_DEFINE_INT(bits) \ |
590 | static int tlv_put_u##bits(struct send_ctx *sctx, \ | |
591 | u##bits attr, u##bits value) \ | |
592 | { \ | |
593 | __le##bits __tmp = cpu_to_le##bits(value); \ | |
594 | return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ | |
595 | } | |
31db9f7c | 596 | |
95bc79d5 | 597 | TLV_PUT_DEFINE_INT(64) |
31db9f7c AB |
598 | |
599 | static int tlv_put_string(struct send_ctx *sctx, u16 attr, | |
600 | const char *str, int len) | |
601 | { | |
602 | if (len == -1) | |
603 | len = strlen(str); | |
604 | return tlv_put(sctx, attr, str, len); | |
605 | } | |
606 | ||
607 | static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, | |
608 | const u8 *uuid) | |
609 | { | |
610 | return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); | |
611 | } | |
612 | ||
31db9f7c AB |
613 | static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, |
614 | struct extent_buffer *eb, | |
615 | struct btrfs_timespec *ts) | |
616 | { | |
617 | struct btrfs_timespec bts; | |
618 | read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); | |
619 | return tlv_put(sctx, attr, &bts, sizeof(bts)); | |
620 | } | |
621 | ||
622 | ||
895a72be | 623 | #define TLV_PUT(sctx, attrtype, data, attrlen) \ |
31db9f7c | 624 | do { \ |
895a72be | 625 | ret = tlv_put(sctx, attrtype, data, attrlen); \ |
31db9f7c AB |
626 | if (ret < 0) \ |
627 | goto tlv_put_failure; \ | |
628 | } while (0) | |
629 | ||
630 | #define TLV_PUT_INT(sctx, attrtype, bits, value) \ | |
631 | do { \ | |
632 | ret = tlv_put_u##bits(sctx, attrtype, value); \ | |
633 | if (ret < 0) \ | |
634 | goto tlv_put_failure; \ | |
635 | } while (0) | |
636 | ||
637 | #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) | |
638 | #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) | |
639 | #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) | |
640 | #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) | |
641 | #define TLV_PUT_STRING(sctx, attrtype, str, len) \ | |
642 | do { \ | |
643 | ret = tlv_put_string(sctx, attrtype, str, len); \ | |
644 | if (ret < 0) \ | |
645 | goto tlv_put_failure; \ | |
646 | } while (0) | |
647 | #define TLV_PUT_PATH(sctx, attrtype, p) \ | |
648 | do { \ | |
649 | ret = tlv_put_string(sctx, attrtype, p->start, \ | |
650 | p->end - p->start); \ | |
651 | if (ret < 0) \ | |
652 | goto tlv_put_failure; \ | |
653 | } while(0) | |
654 | #define TLV_PUT_UUID(sctx, attrtype, uuid) \ | |
655 | do { \ | |
656 | ret = tlv_put_uuid(sctx, attrtype, uuid); \ | |
657 | if (ret < 0) \ | |
658 | goto tlv_put_failure; \ | |
659 | } while (0) | |
31db9f7c AB |
660 | #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ |
661 | do { \ | |
662 | ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ | |
663 | if (ret < 0) \ | |
664 | goto tlv_put_failure; \ | |
665 | } while (0) | |
666 | ||
667 | static int send_header(struct send_ctx *sctx) | |
668 | { | |
669 | struct btrfs_stream_header hdr; | |
670 | ||
671 | strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); | |
672 | hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); | |
673 | ||
1bcea355 AJ |
674 | return write_buf(sctx->send_filp, &hdr, sizeof(hdr), |
675 | &sctx->send_off); | |
31db9f7c AB |
676 | } |
677 | ||
678 | /* | |
679 | * For each command/item we want to send to userspace, we call this function. | |
680 | */ | |
681 | static int begin_cmd(struct send_ctx *sctx, int cmd) | |
682 | { | |
683 | struct btrfs_cmd_header *hdr; | |
684 | ||
fae7f21c | 685 | if (WARN_ON(!sctx->send_buf)) |
31db9f7c | 686 | return -EINVAL; |
31db9f7c AB |
687 | |
688 | BUG_ON(sctx->send_size); | |
689 | ||
690 | sctx->send_size += sizeof(*hdr); | |
691 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
e2f896b3 | 692 | put_unaligned_le16(cmd, &hdr->cmd); |
31db9f7c AB |
693 | |
694 | return 0; | |
695 | } | |
696 | ||
697 | static int send_cmd(struct send_ctx *sctx) | |
698 | { | |
699 | int ret; | |
700 | struct btrfs_cmd_header *hdr; | |
701 | u32 crc; | |
702 | ||
703 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
e2f896b3 DS |
704 | put_unaligned_le32(sctx->send_size - sizeof(*hdr), &hdr->len); |
705 | put_unaligned_le32(0, &hdr->crc); | |
31db9f7c | 706 | |
65019df8 | 707 | crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); |
e2f896b3 | 708 | put_unaligned_le32(crc, &hdr->crc); |
31db9f7c | 709 | |
1bcea355 AJ |
710 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, |
711 | &sctx->send_off); | |
31db9f7c AB |
712 | |
713 | sctx->total_send_size += sctx->send_size; | |
e2f896b3 | 714 | sctx->cmd_send_size[get_unaligned_le16(&hdr->cmd)] += sctx->send_size; |
31db9f7c AB |
715 | sctx->send_size = 0; |
716 | ||
717 | return ret; | |
718 | } | |
719 | ||
720 | /* | |
721 | * Sends a move instruction to user space | |
722 | */ | |
723 | static int send_rename(struct send_ctx *sctx, | |
724 | struct fs_path *from, struct fs_path *to) | |
725 | { | |
04ab956e | 726 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
727 | int ret; |
728 | ||
04ab956e | 729 | btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start); |
31db9f7c AB |
730 | |
731 | ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); | |
732 | if (ret < 0) | |
733 | goto out; | |
734 | ||
735 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); | |
736 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); | |
737 | ||
738 | ret = send_cmd(sctx); | |
739 | ||
740 | tlv_put_failure: | |
741 | out: | |
742 | return ret; | |
743 | } | |
744 | ||
745 | /* | |
746 | * Sends a link instruction to user space | |
747 | */ | |
748 | static int send_link(struct send_ctx *sctx, | |
749 | struct fs_path *path, struct fs_path *lnk) | |
750 | { | |
04ab956e | 751 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
752 | int ret; |
753 | ||
04ab956e | 754 | btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start); |
31db9f7c AB |
755 | |
756 | ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); | |
757 | if (ret < 0) | |
758 | goto out; | |
759 | ||
760 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
761 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); | |
762 | ||
763 | ret = send_cmd(sctx); | |
764 | ||
765 | tlv_put_failure: | |
766 | out: | |
767 | return ret; | |
768 | } | |
769 | ||
770 | /* | |
771 | * Sends an unlink instruction to user space | |
772 | */ | |
773 | static int send_unlink(struct send_ctx *sctx, struct fs_path *path) | |
774 | { | |
04ab956e | 775 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
776 | int ret; |
777 | ||
04ab956e | 778 | btrfs_debug(fs_info, "send_unlink %s", path->start); |
31db9f7c AB |
779 | |
780 | ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); | |
781 | if (ret < 0) | |
782 | goto out; | |
783 | ||
784 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
785 | ||
786 | ret = send_cmd(sctx); | |
787 | ||
788 | tlv_put_failure: | |
789 | out: | |
790 | return ret; | |
791 | } | |
792 | ||
793 | /* | |
794 | * Sends a rmdir instruction to user space | |
795 | */ | |
796 | static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) | |
797 | { | |
04ab956e | 798 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
799 | int ret; |
800 | ||
04ab956e | 801 | btrfs_debug(fs_info, "send_rmdir %s", path->start); |
31db9f7c AB |
802 | |
803 | ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); | |
804 | if (ret < 0) | |
805 | goto out; | |
806 | ||
807 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
808 | ||
809 | ret = send_cmd(sctx); | |
810 | ||
811 | tlv_put_failure: | |
812 | out: | |
813 | return ret; | |
814 | } | |
815 | ||
816 | /* | |
817 | * Helper function to retrieve some fields from an inode item. | |
818 | */ | |
3f8a18cc JB |
819 | static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, |
820 | u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, | |
821 | u64 *gid, u64 *rdev) | |
31db9f7c AB |
822 | { |
823 | int ret; | |
824 | struct btrfs_inode_item *ii; | |
825 | struct btrfs_key key; | |
31db9f7c AB |
826 | |
827 | key.objectid = ino; | |
828 | key.type = BTRFS_INODE_ITEM_KEY; | |
829 | key.offset = 0; | |
830 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
31db9f7c | 831 | if (ret) { |
3f8a18cc JB |
832 | if (ret > 0) |
833 | ret = -ENOENT; | |
834 | return ret; | |
31db9f7c AB |
835 | } |
836 | ||
837 | ii = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
838 | struct btrfs_inode_item); | |
839 | if (size) | |
840 | *size = btrfs_inode_size(path->nodes[0], ii); | |
841 | if (gen) | |
842 | *gen = btrfs_inode_generation(path->nodes[0], ii); | |
843 | if (mode) | |
844 | *mode = btrfs_inode_mode(path->nodes[0], ii); | |
845 | if (uid) | |
846 | *uid = btrfs_inode_uid(path->nodes[0], ii); | |
847 | if (gid) | |
848 | *gid = btrfs_inode_gid(path->nodes[0], ii); | |
85a7b33b AB |
849 | if (rdev) |
850 | *rdev = btrfs_inode_rdev(path->nodes[0], ii); | |
31db9f7c | 851 | |
3f8a18cc JB |
852 | return ret; |
853 | } | |
854 | ||
855 | static int get_inode_info(struct btrfs_root *root, | |
856 | u64 ino, u64 *size, u64 *gen, | |
857 | u64 *mode, u64 *uid, u64 *gid, | |
858 | u64 *rdev) | |
859 | { | |
860 | struct btrfs_path *path; | |
861 | int ret; | |
862 | ||
863 | path = alloc_path_for_send(); | |
864 | if (!path) | |
865 | return -ENOMEM; | |
866 | ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, | |
867 | rdev); | |
31db9f7c AB |
868 | btrfs_free_path(path); |
869 | return ret; | |
870 | } | |
871 | ||
872 | typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, | |
873 | struct fs_path *p, | |
874 | void *ctx); | |
875 | ||
876 | /* | |
96b5bd77 JS |
877 | * Helper function to iterate the entries in ONE btrfs_inode_ref or |
878 | * btrfs_inode_extref. | |
31db9f7c AB |
879 | * The iterate callback may return a non zero value to stop iteration. This can |
880 | * be a negative value for error codes or 1 to simply stop it. | |
881 | * | |
96b5bd77 | 882 | * path must point to the INODE_REF or INODE_EXTREF when called. |
31db9f7c | 883 | */ |
924794c9 | 884 | static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
885 | struct btrfs_key *found_key, int resolve, |
886 | iterate_inode_ref_t iterate, void *ctx) | |
887 | { | |
96b5bd77 | 888 | struct extent_buffer *eb = path->nodes[0]; |
31db9f7c AB |
889 | struct btrfs_item *item; |
890 | struct btrfs_inode_ref *iref; | |
96b5bd77 | 891 | struct btrfs_inode_extref *extref; |
31db9f7c AB |
892 | struct btrfs_path *tmp_path; |
893 | struct fs_path *p; | |
96b5bd77 | 894 | u32 cur = 0; |
31db9f7c | 895 | u32 total; |
96b5bd77 | 896 | int slot = path->slots[0]; |
31db9f7c AB |
897 | u32 name_len; |
898 | char *start; | |
899 | int ret = 0; | |
96b5bd77 | 900 | int num = 0; |
31db9f7c | 901 | int index; |
96b5bd77 JS |
902 | u64 dir; |
903 | unsigned long name_off; | |
904 | unsigned long elem_size; | |
905 | unsigned long ptr; | |
31db9f7c | 906 | |
924794c9 | 907 | p = fs_path_alloc_reversed(); |
31db9f7c AB |
908 | if (!p) |
909 | return -ENOMEM; | |
910 | ||
911 | tmp_path = alloc_path_for_send(); | |
912 | if (!tmp_path) { | |
924794c9 | 913 | fs_path_free(p); |
31db9f7c AB |
914 | return -ENOMEM; |
915 | } | |
916 | ||
31db9f7c | 917 | |
96b5bd77 JS |
918 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
919 | ptr = (unsigned long)btrfs_item_ptr(eb, slot, | |
920 | struct btrfs_inode_ref); | |
dd3cc16b | 921 | item = btrfs_item_nr(slot); |
96b5bd77 JS |
922 | total = btrfs_item_size(eb, item); |
923 | elem_size = sizeof(*iref); | |
924 | } else { | |
925 | ptr = btrfs_item_ptr_offset(eb, slot); | |
926 | total = btrfs_item_size_nr(eb, slot); | |
927 | elem_size = sizeof(*extref); | |
928 | } | |
929 | ||
31db9f7c AB |
930 | while (cur < total) { |
931 | fs_path_reset(p); | |
932 | ||
96b5bd77 JS |
933 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
934 | iref = (struct btrfs_inode_ref *)(ptr + cur); | |
935 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
936 | name_off = (unsigned long)(iref + 1); | |
937 | index = btrfs_inode_ref_index(eb, iref); | |
938 | dir = found_key->offset; | |
939 | } else { | |
940 | extref = (struct btrfs_inode_extref *)(ptr + cur); | |
941 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
942 | name_off = (unsigned long)&extref->name; | |
943 | index = btrfs_inode_extref_index(eb, extref); | |
944 | dir = btrfs_inode_extref_parent(eb, extref); | |
945 | } | |
946 | ||
31db9f7c | 947 | if (resolve) { |
96b5bd77 JS |
948 | start = btrfs_ref_to_path(root, tmp_path, name_len, |
949 | name_off, eb, dir, | |
950 | p->buf, p->buf_len); | |
31db9f7c AB |
951 | if (IS_ERR(start)) { |
952 | ret = PTR_ERR(start); | |
953 | goto out; | |
954 | } | |
955 | if (start < p->buf) { | |
956 | /* overflow , try again with larger buffer */ | |
957 | ret = fs_path_ensure_buf(p, | |
958 | p->buf_len + p->buf - start); | |
959 | if (ret < 0) | |
960 | goto out; | |
96b5bd77 JS |
961 | start = btrfs_ref_to_path(root, tmp_path, |
962 | name_len, name_off, | |
963 | eb, dir, | |
964 | p->buf, p->buf_len); | |
31db9f7c AB |
965 | if (IS_ERR(start)) { |
966 | ret = PTR_ERR(start); | |
967 | goto out; | |
968 | } | |
969 | BUG_ON(start < p->buf); | |
970 | } | |
971 | p->start = start; | |
972 | } else { | |
96b5bd77 JS |
973 | ret = fs_path_add_from_extent_buffer(p, eb, name_off, |
974 | name_len); | |
31db9f7c AB |
975 | if (ret < 0) |
976 | goto out; | |
977 | } | |
978 | ||
96b5bd77 JS |
979 | cur += elem_size + name_len; |
980 | ret = iterate(num, dir, index, p, ctx); | |
31db9f7c AB |
981 | if (ret) |
982 | goto out; | |
31db9f7c AB |
983 | num++; |
984 | } | |
985 | ||
986 | out: | |
987 | btrfs_free_path(tmp_path); | |
924794c9 | 988 | fs_path_free(p); |
31db9f7c AB |
989 | return ret; |
990 | } | |
991 | ||
992 | typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, | |
993 | const char *name, int name_len, | |
994 | const char *data, int data_len, | |
995 | u8 type, void *ctx); | |
996 | ||
997 | /* | |
998 | * Helper function to iterate the entries in ONE btrfs_dir_item. | |
999 | * The iterate callback may return a non zero value to stop iteration. This can | |
1000 | * be a negative value for error codes or 1 to simply stop it. | |
1001 | * | |
1002 | * path must point to the dir item when called. | |
1003 | */ | |
924794c9 | 1004 | static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
1005 | iterate_dir_item_t iterate, void *ctx) |
1006 | { | |
1007 | int ret = 0; | |
1008 | struct extent_buffer *eb; | |
1009 | struct btrfs_item *item; | |
1010 | struct btrfs_dir_item *di; | |
31db9f7c AB |
1011 | struct btrfs_key di_key; |
1012 | char *buf = NULL; | |
7e3ae33e | 1013 | int buf_len; |
31db9f7c AB |
1014 | u32 name_len; |
1015 | u32 data_len; | |
1016 | u32 cur; | |
1017 | u32 len; | |
1018 | u32 total; | |
1019 | int slot; | |
1020 | int num; | |
1021 | u8 type; | |
1022 | ||
4395e0c4 FM |
1023 | /* |
1024 | * Start with a small buffer (1 page). If later we end up needing more | |
1025 | * space, which can happen for xattrs on a fs with a leaf size greater | |
1026 | * then the page size, attempt to increase the buffer. Typically xattr | |
1027 | * values are small. | |
1028 | */ | |
1029 | buf_len = PATH_MAX; | |
e780b0d1 | 1030 | buf = kmalloc(buf_len, GFP_KERNEL); |
31db9f7c AB |
1031 | if (!buf) { |
1032 | ret = -ENOMEM; | |
1033 | goto out; | |
1034 | } | |
1035 | ||
31db9f7c AB |
1036 | eb = path->nodes[0]; |
1037 | slot = path->slots[0]; | |
dd3cc16b | 1038 | item = btrfs_item_nr(slot); |
31db9f7c AB |
1039 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
1040 | cur = 0; | |
1041 | len = 0; | |
1042 | total = btrfs_item_size(eb, item); | |
1043 | ||
1044 | num = 0; | |
1045 | while (cur < total) { | |
1046 | name_len = btrfs_dir_name_len(eb, di); | |
1047 | data_len = btrfs_dir_data_len(eb, di); | |
1048 | type = btrfs_dir_type(eb, di); | |
1049 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1050 | ||
7e3ae33e FM |
1051 | if (type == BTRFS_FT_XATTR) { |
1052 | if (name_len > XATTR_NAME_MAX) { | |
1053 | ret = -ENAMETOOLONG; | |
1054 | goto out; | |
1055 | } | |
da17066c JM |
1056 | if (name_len + data_len > |
1057 | BTRFS_MAX_XATTR_SIZE(root->fs_info)) { | |
7e3ae33e FM |
1058 | ret = -E2BIG; |
1059 | goto out; | |
1060 | } | |
1061 | } else { | |
1062 | /* | |
1063 | * Path too long | |
1064 | */ | |
4395e0c4 | 1065 | if (name_len + data_len > PATH_MAX) { |
7e3ae33e FM |
1066 | ret = -ENAMETOOLONG; |
1067 | goto out; | |
1068 | } | |
31db9f7c AB |
1069 | } |
1070 | ||
4395e0c4 FM |
1071 | if (name_len + data_len > buf_len) { |
1072 | buf_len = name_len + data_len; | |
1073 | if (is_vmalloc_addr(buf)) { | |
1074 | vfree(buf); | |
1075 | buf = NULL; | |
1076 | } else { | |
1077 | char *tmp = krealloc(buf, buf_len, | |
e780b0d1 | 1078 | GFP_KERNEL | __GFP_NOWARN); |
4395e0c4 FM |
1079 | |
1080 | if (!tmp) | |
1081 | kfree(buf); | |
1082 | buf = tmp; | |
1083 | } | |
1084 | if (!buf) { | |
f11f7441 | 1085 | buf = kvmalloc(buf_len, GFP_KERNEL); |
4395e0c4 FM |
1086 | if (!buf) { |
1087 | ret = -ENOMEM; | |
1088 | goto out; | |
1089 | } | |
1090 | } | |
1091 | } | |
1092 | ||
31db9f7c AB |
1093 | read_extent_buffer(eb, buf, (unsigned long)(di + 1), |
1094 | name_len + data_len); | |
1095 | ||
1096 | len = sizeof(*di) + name_len + data_len; | |
1097 | di = (struct btrfs_dir_item *)((char *)di + len); | |
1098 | cur += len; | |
1099 | ||
1100 | ret = iterate(num, &di_key, buf, name_len, buf + name_len, | |
1101 | data_len, type, ctx); | |
1102 | if (ret < 0) | |
1103 | goto out; | |
1104 | if (ret) { | |
1105 | ret = 0; | |
1106 | goto out; | |
1107 | } | |
1108 | ||
1109 | num++; | |
1110 | } | |
1111 | ||
1112 | out: | |
4395e0c4 | 1113 | kvfree(buf); |
31db9f7c AB |
1114 | return ret; |
1115 | } | |
1116 | ||
1117 | static int __copy_first_ref(int num, u64 dir, int index, | |
1118 | struct fs_path *p, void *ctx) | |
1119 | { | |
1120 | int ret; | |
1121 | struct fs_path *pt = ctx; | |
1122 | ||
1123 | ret = fs_path_copy(pt, p); | |
1124 | if (ret < 0) | |
1125 | return ret; | |
1126 | ||
1127 | /* we want the first only */ | |
1128 | return 1; | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * Retrieve the first path of an inode. If an inode has more then one | |
1133 | * ref/hardlink, this is ignored. | |
1134 | */ | |
924794c9 | 1135 | static int get_inode_path(struct btrfs_root *root, |
31db9f7c AB |
1136 | u64 ino, struct fs_path *path) |
1137 | { | |
1138 | int ret; | |
1139 | struct btrfs_key key, found_key; | |
1140 | struct btrfs_path *p; | |
1141 | ||
1142 | p = alloc_path_for_send(); | |
1143 | if (!p) | |
1144 | return -ENOMEM; | |
1145 | ||
1146 | fs_path_reset(path); | |
1147 | ||
1148 | key.objectid = ino; | |
1149 | key.type = BTRFS_INODE_REF_KEY; | |
1150 | key.offset = 0; | |
1151 | ||
1152 | ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); | |
1153 | if (ret < 0) | |
1154 | goto out; | |
1155 | if (ret) { | |
1156 | ret = 1; | |
1157 | goto out; | |
1158 | } | |
1159 | btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); | |
1160 | if (found_key.objectid != ino || | |
96b5bd77 JS |
1161 | (found_key.type != BTRFS_INODE_REF_KEY && |
1162 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1163 | ret = -ENOENT; |
1164 | goto out; | |
1165 | } | |
1166 | ||
924794c9 TI |
1167 | ret = iterate_inode_ref(root, p, &found_key, 1, |
1168 | __copy_first_ref, path); | |
31db9f7c AB |
1169 | if (ret < 0) |
1170 | goto out; | |
1171 | ret = 0; | |
1172 | ||
1173 | out: | |
1174 | btrfs_free_path(p); | |
1175 | return ret; | |
1176 | } | |
1177 | ||
1178 | struct backref_ctx { | |
1179 | struct send_ctx *sctx; | |
1180 | ||
1181 | /* number of total found references */ | |
1182 | u64 found; | |
1183 | ||
1184 | /* | |
1185 | * used for clones found in send_root. clones found behind cur_objectid | |
1186 | * and cur_offset are not considered as allowed clones. | |
1187 | */ | |
1188 | u64 cur_objectid; | |
1189 | u64 cur_offset; | |
1190 | ||
1191 | /* may be truncated in case it's the last extent in a file */ | |
1192 | u64 extent_len; | |
1193 | ||
619d8c4e FM |
1194 | /* data offset in the file extent item */ |
1195 | u64 data_offset; | |
1196 | ||
31db9f7c | 1197 | /* Just to check for bugs in backref resolving */ |
ee849c04 | 1198 | int found_itself; |
31db9f7c AB |
1199 | }; |
1200 | ||
1201 | static int __clone_root_cmp_bsearch(const void *key, const void *elt) | |
1202 | { | |
995e01b7 | 1203 | u64 root = (u64)(uintptr_t)key; |
31db9f7c AB |
1204 | struct clone_root *cr = (struct clone_root *)elt; |
1205 | ||
4fd786e6 | 1206 | if (root < cr->root->root_key.objectid) |
31db9f7c | 1207 | return -1; |
4fd786e6 | 1208 | if (root > cr->root->root_key.objectid) |
31db9f7c AB |
1209 | return 1; |
1210 | return 0; | |
1211 | } | |
1212 | ||
1213 | static int __clone_root_cmp_sort(const void *e1, const void *e2) | |
1214 | { | |
1215 | struct clone_root *cr1 = (struct clone_root *)e1; | |
1216 | struct clone_root *cr2 = (struct clone_root *)e2; | |
1217 | ||
4fd786e6 | 1218 | if (cr1->root->root_key.objectid < cr2->root->root_key.objectid) |
31db9f7c | 1219 | return -1; |
4fd786e6 | 1220 | if (cr1->root->root_key.objectid > cr2->root->root_key.objectid) |
31db9f7c AB |
1221 | return 1; |
1222 | return 0; | |
1223 | } | |
1224 | ||
1225 | /* | |
1226 | * Called for every backref that is found for the current extent. | |
766702ef | 1227 | * Results are collected in sctx->clone_roots->ino/offset/found_refs |
31db9f7c AB |
1228 | */ |
1229 | static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) | |
1230 | { | |
1231 | struct backref_ctx *bctx = ctx_; | |
1232 | struct clone_root *found; | |
31db9f7c AB |
1233 | |
1234 | /* First check if the root is in the list of accepted clone sources */ | |
995e01b7 | 1235 | found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, |
31db9f7c AB |
1236 | bctx->sctx->clone_roots_cnt, |
1237 | sizeof(struct clone_root), | |
1238 | __clone_root_cmp_bsearch); | |
1239 | if (!found) | |
1240 | return 0; | |
1241 | ||
1242 | if (found->root == bctx->sctx->send_root && | |
1243 | ino == bctx->cur_objectid && | |
1244 | offset == bctx->cur_offset) { | |
ee849c04 | 1245 | bctx->found_itself = 1; |
31db9f7c AB |
1246 | } |
1247 | ||
31db9f7c AB |
1248 | /* |
1249 | * Make sure we don't consider clones from send_root that are | |
1250 | * behind the current inode/offset. | |
1251 | */ | |
1252 | if (found->root == bctx->sctx->send_root) { | |
1253 | /* | |
11f2069c FM |
1254 | * If the source inode was not yet processed we can't issue a |
1255 | * clone operation, as the source extent does not exist yet at | |
1256 | * the destination of the stream. | |
31db9f7c | 1257 | */ |
11f2069c FM |
1258 | if (ino > bctx->cur_objectid) |
1259 | return 0; | |
1260 | /* | |
1261 | * We clone from the inode currently being sent as long as the | |
1262 | * source extent is already processed, otherwise we could try | |
1263 | * to clone from an extent that does not exist yet at the | |
1264 | * destination of the stream. | |
1265 | */ | |
1266 | if (ino == bctx->cur_objectid && | |
9722b101 FM |
1267 | offset + bctx->extent_len > |
1268 | bctx->sctx->cur_inode_next_write_offset) | |
31db9f7c | 1269 | return 0; |
31db9f7c AB |
1270 | } |
1271 | ||
1272 | bctx->found++; | |
1273 | found->found_refs++; | |
1274 | if (ino < found->ino) { | |
1275 | found->ino = ino; | |
1276 | found->offset = offset; | |
1277 | } else if (found->ino == ino) { | |
1278 | /* | |
1279 | * same extent found more then once in the same file. | |
1280 | */ | |
1281 | if (found->offset > offset + bctx->extent_len) | |
1282 | found->offset = offset; | |
1283 | } | |
1284 | ||
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | /* | |
766702ef AB |
1289 | * Given an inode, offset and extent item, it finds a good clone for a clone |
1290 | * instruction. Returns -ENOENT when none could be found. The function makes | |
1291 | * sure that the returned clone is usable at the point where sending is at the | |
1292 | * moment. This means, that no clones are accepted which lie behind the current | |
1293 | * inode+offset. | |
1294 | * | |
31db9f7c AB |
1295 | * path must point to the extent item when called. |
1296 | */ | |
1297 | static int find_extent_clone(struct send_ctx *sctx, | |
1298 | struct btrfs_path *path, | |
1299 | u64 ino, u64 data_offset, | |
1300 | u64 ino_size, | |
1301 | struct clone_root **found) | |
1302 | { | |
04ab956e | 1303 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
1304 | int ret; |
1305 | int extent_type; | |
1306 | u64 logical; | |
74dd17fb | 1307 | u64 disk_byte; |
31db9f7c AB |
1308 | u64 num_bytes; |
1309 | u64 extent_item_pos; | |
69917e43 | 1310 | u64 flags = 0; |
31db9f7c AB |
1311 | struct btrfs_file_extent_item *fi; |
1312 | struct extent_buffer *eb = path->nodes[0]; | |
35075bb0 | 1313 | struct backref_ctx *backref_ctx = NULL; |
31db9f7c AB |
1314 | struct clone_root *cur_clone_root; |
1315 | struct btrfs_key found_key; | |
1316 | struct btrfs_path *tmp_path; | |
fd0ddbe2 | 1317 | struct btrfs_extent_item *ei; |
74dd17fb | 1318 | int compressed; |
31db9f7c AB |
1319 | u32 i; |
1320 | ||
1321 | tmp_path = alloc_path_for_send(); | |
1322 | if (!tmp_path) | |
1323 | return -ENOMEM; | |
1324 | ||
3f8a18cc JB |
1325 | /* We only use this path under the commit sem */ |
1326 | tmp_path->need_commit_sem = 0; | |
1327 | ||
e780b0d1 | 1328 | backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL); |
35075bb0 AB |
1329 | if (!backref_ctx) { |
1330 | ret = -ENOMEM; | |
1331 | goto out; | |
1332 | } | |
1333 | ||
31db9f7c AB |
1334 | if (data_offset >= ino_size) { |
1335 | /* | |
1336 | * There may be extents that lie behind the file's size. | |
1337 | * I at least had this in combination with snapshotting while | |
1338 | * writing large files. | |
1339 | */ | |
1340 | ret = 0; | |
1341 | goto out; | |
1342 | } | |
1343 | ||
1344 | fi = btrfs_item_ptr(eb, path->slots[0], | |
1345 | struct btrfs_file_extent_item); | |
1346 | extent_type = btrfs_file_extent_type(eb, fi); | |
1347 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1348 | ret = -ENOENT; | |
1349 | goto out; | |
1350 | } | |
74dd17fb | 1351 | compressed = btrfs_file_extent_compression(eb, fi); |
31db9f7c AB |
1352 | |
1353 | num_bytes = btrfs_file_extent_num_bytes(eb, fi); | |
74dd17fb CM |
1354 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
1355 | if (disk_byte == 0) { | |
31db9f7c AB |
1356 | ret = -ENOENT; |
1357 | goto out; | |
1358 | } | |
74dd17fb | 1359 | logical = disk_byte + btrfs_file_extent_offset(eb, fi); |
31db9f7c | 1360 | |
04ab956e JM |
1361 | down_read(&fs_info->commit_root_sem); |
1362 | ret = extent_from_logical(fs_info, disk_byte, tmp_path, | |
69917e43 | 1363 | &found_key, &flags); |
04ab956e | 1364 | up_read(&fs_info->commit_root_sem); |
31db9f7c AB |
1365 | |
1366 | if (ret < 0) | |
1367 | goto out; | |
69917e43 | 1368 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
31db9f7c AB |
1369 | ret = -EIO; |
1370 | goto out; | |
1371 | } | |
1372 | ||
fd0ddbe2 FM |
1373 | ei = btrfs_item_ptr(tmp_path->nodes[0], tmp_path->slots[0], |
1374 | struct btrfs_extent_item); | |
1375 | /* | |
1376 | * Backreference walking (iterate_extent_inodes() below) is currently | |
1377 | * too expensive when an extent has a large number of references, both | |
1378 | * in time spent and used memory. So for now just fallback to write | |
1379 | * operations instead of clone operations when an extent has more than | |
1380 | * a certain amount of references. | |
1381 | */ | |
1382 | if (btrfs_extent_refs(tmp_path->nodes[0], ei) > SEND_MAX_EXTENT_REFS) { | |
1383 | ret = -ENOENT; | |
1384 | goto out; | |
1385 | } | |
1386 | btrfs_release_path(tmp_path); | |
1387 | ||
31db9f7c AB |
1388 | /* |
1389 | * Setup the clone roots. | |
1390 | */ | |
1391 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1392 | cur_clone_root = sctx->clone_roots + i; | |
1393 | cur_clone_root->ino = (u64)-1; | |
1394 | cur_clone_root->offset = 0; | |
1395 | cur_clone_root->found_refs = 0; | |
1396 | } | |
1397 | ||
35075bb0 AB |
1398 | backref_ctx->sctx = sctx; |
1399 | backref_ctx->found = 0; | |
1400 | backref_ctx->cur_objectid = ino; | |
1401 | backref_ctx->cur_offset = data_offset; | |
1402 | backref_ctx->found_itself = 0; | |
1403 | backref_ctx->extent_len = num_bytes; | |
619d8c4e FM |
1404 | /* |
1405 | * For non-compressed extents iterate_extent_inodes() gives us extent | |
1406 | * offsets that already take into account the data offset, but not for | |
1407 | * compressed extents, since the offset is logical and not relative to | |
1408 | * the physical extent locations. We must take this into account to | |
1409 | * avoid sending clone offsets that go beyond the source file's size, | |
1410 | * which would result in the clone ioctl failing with -EINVAL on the | |
1411 | * receiving end. | |
1412 | */ | |
1413 | if (compressed == BTRFS_COMPRESS_NONE) | |
1414 | backref_ctx->data_offset = 0; | |
1415 | else | |
1416 | backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi); | |
31db9f7c AB |
1417 | |
1418 | /* | |
1419 | * The last extent of a file may be too large due to page alignment. | |
1420 | * We need to adjust extent_len in this case so that the checks in | |
1421 | * __iterate_backrefs work. | |
1422 | */ | |
1423 | if (data_offset + num_bytes >= ino_size) | |
35075bb0 | 1424 | backref_ctx->extent_len = ino_size - data_offset; |
31db9f7c AB |
1425 | |
1426 | /* | |
1427 | * Now collect all backrefs. | |
1428 | */ | |
74dd17fb CM |
1429 | if (compressed == BTRFS_COMPRESS_NONE) |
1430 | extent_item_pos = logical - found_key.objectid; | |
1431 | else | |
1432 | extent_item_pos = 0; | |
0b246afa JM |
1433 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
1434 | extent_item_pos, 1, __iterate_backrefs, | |
c995ab3c | 1435 | backref_ctx, false); |
74dd17fb | 1436 | |
31db9f7c AB |
1437 | if (ret < 0) |
1438 | goto out; | |
1439 | ||
35075bb0 | 1440 | if (!backref_ctx->found_itself) { |
31db9f7c AB |
1441 | /* found a bug in backref code? */ |
1442 | ret = -EIO; | |
04ab956e | 1443 | btrfs_err(fs_info, |
5d163e0e | 1444 | "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu", |
04ab956e | 1445 | ino, data_offset, disk_byte, found_key.objectid); |
31db9f7c AB |
1446 | goto out; |
1447 | } | |
1448 | ||
04ab956e JM |
1449 | btrfs_debug(fs_info, |
1450 | "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu", | |
1451 | data_offset, ino, num_bytes, logical); | |
31db9f7c | 1452 | |
35075bb0 | 1453 | if (!backref_ctx->found) |
04ab956e | 1454 | btrfs_debug(fs_info, "no clones found"); |
31db9f7c AB |
1455 | |
1456 | cur_clone_root = NULL; | |
1457 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1458 | if (sctx->clone_roots[i].found_refs) { | |
1459 | if (!cur_clone_root) | |
1460 | cur_clone_root = sctx->clone_roots + i; | |
1461 | else if (sctx->clone_roots[i].root == sctx->send_root) | |
1462 | /* prefer clones from send_root over others */ | |
1463 | cur_clone_root = sctx->clone_roots + i; | |
31db9f7c AB |
1464 | } |
1465 | ||
1466 | } | |
1467 | ||
1468 | if (cur_clone_root) { | |
1469 | *found = cur_clone_root; | |
1470 | ret = 0; | |
1471 | } else { | |
1472 | ret = -ENOENT; | |
1473 | } | |
1474 | ||
1475 | out: | |
1476 | btrfs_free_path(tmp_path); | |
35075bb0 | 1477 | kfree(backref_ctx); |
31db9f7c AB |
1478 | return ret; |
1479 | } | |
1480 | ||
924794c9 | 1481 | static int read_symlink(struct btrfs_root *root, |
31db9f7c AB |
1482 | u64 ino, |
1483 | struct fs_path *dest) | |
1484 | { | |
1485 | int ret; | |
1486 | struct btrfs_path *path; | |
1487 | struct btrfs_key key; | |
1488 | struct btrfs_file_extent_item *ei; | |
1489 | u8 type; | |
1490 | u8 compression; | |
1491 | unsigned long off; | |
1492 | int len; | |
1493 | ||
1494 | path = alloc_path_for_send(); | |
1495 | if (!path) | |
1496 | return -ENOMEM; | |
1497 | ||
1498 | key.objectid = ino; | |
1499 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1500 | key.offset = 0; | |
1501 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1502 | if (ret < 0) | |
1503 | goto out; | |
a879719b FM |
1504 | if (ret) { |
1505 | /* | |
1506 | * An empty symlink inode. Can happen in rare error paths when | |
1507 | * creating a symlink (transaction committed before the inode | |
1508 | * eviction handler removed the symlink inode items and a crash | |
1509 | * happened in between or the subvol was snapshoted in between). | |
1510 | * Print an informative message to dmesg/syslog so that the user | |
1511 | * can delete the symlink. | |
1512 | */ | |
1513 | btrfs_err(root->fs_info, | |
1514 | "Found empty symlink inode %llu at root %llu", | |
1515 | ino, root->root_key.objectid); | |
1516 | ret = -EIO; | |
1517 | goto out; | |
1518 | } | |
31db9f7c AB |
1519 | |
1520 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1521 | struct btrfs_file_extent_item); | |
1522 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
1523 | compression = btrfs_file_extent_compression(path->nodes[0], ei); | |
1524 | BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); | |
1525 | BUG_ON(compression); | |
1526 | ||
1527 | off = btrfs_file_extent_inline_start(ei); | |
e41ca589 | 1528 | len = btrfs_file_extent_ram_bytes(path->nodes[0], ei); |
31db9f7c AB |
1529 | |
1530 | ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); | |
31db9f7c AB |
1531 | |
1532 | out: | |
1533 | btrfs_free_path(path); | |
1534 | return ret; | |
1535 | } | |
1536 | ||
1537 | /* | |
1538 | * Helper function to generate a file name that is unique in the root of | |
1539 | * send_root and parent_root. This is used to generate names for orphan inodes. | |
1540 | */ | |
1541 | static int gen_unique_name(struct send_ctx *sctx, | |
1542 | u64 ino, u64 gen, | |
1543 | struct fs_path *dest) | |
1544 | { | |
1545 | int ret = 0; | |
1546 | struct btrfs_path *path; | |
1547 | struct btrfs_dir_item *di; | |
1548 | char tmp[64]; | |
1549 | int len; | |
1550 | u64 idx = 0; | |
1551 | ||
1552 | path = alloc_path_for_send(); | |
1553 | if (!path) | |
1554 | return -ENOMEM; | |
1555 | ||
1556 | while (1) { | |
f74b86d8 | 1557 | len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", |
31db9f7c | 1558 | ino, gen, idx); |
64792f25 | 1559 | ASSERT(len < sizeof(tmp)); |
31db9f7c AB |
1560 | |
1561 | di = btrfs_lookup_dir_item(NULL, sctx->send_root, | |
1562 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1563 | tmp, strlen(tmp), 0); | |
1564 | btrfs_release_path(path); | |
1565 | if (IS_ERR(di)) { | |
1566 | ret = PTR_ERR(di); | |
1567 | goto out; | |
1568 | } | |
1569 | if (di) { | |
1570 | /* not unique, try again */ | |
1571 | idx++; | |
1572 | continue; | |
1573 | } | |
1574 | ||
1575 | if (!sctx->parent_root) { | |
1576 | /* unique */ | |
1577 | ret = 0; | |
1578 | break; | |
1579 | } | |
1580 | ||
1581 | di = btrfs_lookup_dir_item(NULL, sctx->parent_root, | |
1582 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1583 | tmp, strlen(tmp), 0); | |
1584 | btrfs_release_path(path); | |
1585 | if (IS_ERR(di)) { | |
1586 | ret = PTR_ERR(di); | |
1587 | goto out; | |
1588 | } | |
1589 | if (di) { | |
1590 | /* not unique, try again */ | |
1591 | idx++; | |
1592 | continue; | |
1593 | } | |
1594 | /* unique */ | |
1595 | break; | |
1596 | } | |
1597 | ||
1598 | ret = fs_path_add(dest, tmp, strlen(tmp)); | |
1599 | ||
1600 | out: | |
1601 | btrfs_free_path(path); | |
1602 | return ret; | |
1603 | } | |
1604 | ||
1605 | enum inode_state { | |
1606 | inode_state_no_change, | |
1607 | inode_state_will_create, | |
1608 | inode_state_did_create, | |
1609 | inode_state_will_delete, | |
1610 | inode_state_did_delete, | |
1611 | }; | |
1612 | ||
1613 | static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) | |
1614 | { | |
1615 | int ret; | |
1616 | int left_ret; | |
1617 | int right_ret; | |
1618 | u64 left_gen; | |
1619 | u64 right_gen; | |
1620 | ||
1621 | ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, | |
85a7b33b | 1622 | NULL, NULL); |
31db9f7c AB |
1623 | if (ret < 0 && ret != -ENOENT) |
1624 | goto out; | |
1625 | left_ret = ret; | |
1626 | ||
1627 | if (!sctx->parent_root) { | |
1628 | right_ret = -ENOENT; | |
1629 | } else { | |
1630 | ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, | |
85a7b33b | 1631 | NULL, NULL, NULL, NULL); |
31db9f7c AB |
1632 | if (ret < 0 && ret != -ENOENT) |
1633 | goto out; | |
1634 | right_ret = ret; | |
1635 | } | |
1636 | ||
1637 | if (!left_ret && !right_ret) { | |
e938c8ad | 1638 | if (left_gen == gen && right_gen == gen) { |
31db9f7c | 1639 | ret = inode_state_no_change; |
e938c8ad | 1640 | } else if (left_gen == gen) { |
31db9f7c AB |
1641 | if (ino < sctx->send_progress) |
1642 | ret = inode_state_did_create; | |
1643 | else | |
1644 | ret = inode_state_will_create; | |
1645 | } else if (right_gen == gen) { | |
1646 | if (ino < sctx->send_progress) | |
1647 | ret = inode_state_did_delete; | |
1648 | else | |
1649 | ret = inode_state_will_delete; | |
1650 | } else { | |
1651 | ret = -ENOENT; | |
1652 | } | |
1653 | } else if (!left_ret) { | |
1654 | if (left_gen == gen) { | |
1655 | if (ino < sctx->send_progress) | |
1656 | ret = inode_state_did_create; | |
1657 | else | |
1658 | ret = inode_state_will_create; | |
1659 | } else { | |
1660 | ret = -ENOENT; | |
1661 | } | |
1662 | } else if (!right_ret) { | |
1663 | if (right_gen == gen) { | |
1664 | if (ino < sctx->send_progress) | |
1665 | ret = inode_state_did_delete; | |
1666 | else | |
1667 | ret = inode_state_will_delete; | |
1668 | } else { | |
1669 | ret = -ENOENT; | |
1670 | } | |
1671 | } else { | |
1672 | ret = -ENOENT; | |
1673 | } | |
1674 | ||
1675 | out: | |
1676 | return ret; | |
1677 | } | |
1678 | ||
1679 | static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) | |
1680 | { | |
1681 | int ret; | |
1682 | ||
4dd9920d RK |
1683 | if (ino == BTRFS_FIRST_FREE_OBJECTID) |
1684 | return 1; | |
1685 | ||
31db9f7c AB |
1686 | ret = get_cur_inode_state(sctx, ino, gen); |
1687 | if (ret < 0) | |
1688 | goto out; | |
1689 | ||
1690 | if (ret == inode_state_no_change || | |
1691 | ret == inode_state_did_create || | |
1692 | ret == inode_state_will_delete) | |
1693 | ret = 1; | |
1694 | else | |
1695 | ret = 0; | |
1696 | ||
1697 | out: | |
1698 | return ret; | |
1699 | } | |
1700 | ||
1701 | /* | |
1702 | * Helper function to lookup a dir item in a dir. | |
1703 | */ | |
1704 | static int lookup_dir_item_inode(struct btrfs_root *root, | |
1705 | u64 dir, const char *name, int name_len, | |
1706 | u64 *found_inode, | |
1707 | u8 *found_type) | |
1708 | { | |
1709 | int ret = 0; | |
1710 | struct btrfs_dir_item *di; | |
1711 | struct btrfs_key key; | |
1712 | struct btrfs_path *path; | |
1713 | ||
1714 | path = alloc_path_for_send(); | |
1715 | if (!path) | |
1716 | return -ENOMEM; | |
1717 | ||
1718 | di = btrfs_lookup_dir_item(NULL, root, path, | |
1719 | dir, name, name_len, 0); | |
3cf5068f LB |
1720 | if (IS_ERR_OR_NULL(di)) { |
1721 | ret = di ? PTR_ERR(di) : -ENOENT; | |
31db9f7c AB |
1722 | goto out; |
1723 | } | |
1724 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
1af56070 FM |
1725 | if (key.type == BTRFS_ROOT_ITEM_KEY) { |
1726 | ret = -ENOENT; | |
1727 | goto out; | |
1728 | } | |
31db9f7c AB |
1729 | *found_inode = key.objectid; |
1730 | *found_type = btrfs_dir_type(path->nodes[0], di); | |
1731 | ||
1732 | out: | |
1733 | btrfs_free_path(path); | |
1734 | return ret; | |
1735 | } | |
1736 | ||
766702ef AB |
1737 | /* |
1738 | * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, | |
1739 | * generation of the parent dir and the name of the dir entry. | |
1740 | */ | |
924794c9 | 1741 | static int get_first_ref(struct btrfs_root *root, u64 ino, |
31db9f7c AB |
1742 | u64 *dir, u64 *dir_gen, struct fs_path *name) |
1743 | { | |
1744 | int ret; | |
1745 | struct btrfs_key key; | |
1746 | struct btrfs_key found_key; | |
1747 | struct btrfs_path *path; | |
31db9f7c | 1748 | int len; |
96b5bd77 | 1749 | u64 parent_dir; |
31db9f7c AB |
1750 | |
1751 | path = alloc_path_for_send(); | |
1752 | if (!path) | |
1753 | return -ENOMEM; | |
1754 | ||
1755 | key.objectid = ino; | |
1756 | key.type = BTRFS_INODE_REF_KEY; | |
1757 | key.offset = 0; | |
1758 | ||
1759 | ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); | |
1760 | if (ret < 0) | |
1761 | goto out; | |
1762 | if (!ret) | |
1763 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1764 | path->slots[0]); | |
96b5bd77 JS |
1765 | if (ret || found_key.objectid != ino || |
1766 | (found_key.type != BTRFS_INODE_REF_KEY && | |
1767 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1768 | ret = -ENOENT; |
1769 | goto out; | |
1770 | } | |
1771 | ||
51a60253 | 1772 | if (found_key.type == BTRFS_INODE_REF_KEY) { |
96b5bd77 JS |
1773 | struct btrfs_inode_ref *iref; |
1774 | iref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1775 | struct btrfs_inode_ref); | |
1776 | len = btrfs_inode_ref_name_len(path->nodes[0], iref); | |
1777 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1778 | (unsigned long)(iref + 1), | |
1779 | len); | |
1780 | parent_dir = found_key.offset; | |
1781 | } else { | |
1782 | struct btrfs_inode_extref *extref; | |
1783 | extref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1784 | struct btrfs_inode_extref); | |
1785 | len = btrfs_inode_extref_name_len(path->nodes[0], extref); | |
1786 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1787 | (unsigned long)&extref->name, len); | |
1788 | parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); | |
1789 | } | |
31db9f7c AB |
1790 | if (ret < 0) |
1791 | goto out; | |
1792 | btrfs_release_path(path); | |
1793 | ||
b46ab97b FM |
1794 | if (dir_gen) { |
1795 | ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, | |
1796 | NULL, NULL, NULL); | |
1797 | if (ret < 0) | |
1798 | goto out; | |
1799 | } | |
31db9f7c | 1800 | |
96b5bd77 | 1801 | *dir = parent_dir; |
31db9f7c AB |
1802 | |
1803 | out: | |
1804 | btrfs_free_path(path); | |
1805 | return ret; | |
1806 | } | |
1807 | ||
924794c9 | 1808 | static int is_first_ref(struct btrfs_root *root, |
31db9f7c AB |
1809 | u64 ino, u64 dir, |
1810 | const char *name, int name_len) | |
1811 | { | |
1812 | int ret; | |
1813 | struct fs_path *tmp_name; | |
1814 | u64 tmp_dir; | |
31db9f7c | 1815 | |
924794c9 | 1816 | tmp_name = fs_path_alloc(); |
31db9f7c AB |
1817 | if (!tmp_name) |
1818 | return -ENOMEM; | |
1819 | ||
b46ab97b | 1820 | ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name); |
31db9f7c AB |
1821 | if (ret < 0) |
1822 | goto out; | |
1823 | ||
b9291aff | 1824 | if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { |
31db9f7c AB |
1825 | ret = 0; |
1826 | goto out; | |
1827 | } | |
1828 | ||
e938c8ad | 1829 | ret = !memcmp(tmp_name->start, name, name_len); |
31db9f7c AB |
1830 | |
1831 | out: | |
924794c9 | 1832 | fs_path_free(tmp_name); |
31db9f7c AB |
1833 | return ret; |
1834 | } | |
1835 | ||
766702ef AB |
1836 | /* |
1837 | * Used by process_recorded_refs to determine if a new ref would overwrite an | |
1838 | * already existing ref. In case it detects an overwrite, it returns the | |
1839 | * inode/gen in who_ino/who_gen. | |
1840 | * When an overwrite is detected, process_recorded_refs does proper orphanizing | |
1841 | * to make sure later references to the overwritten inode are possible. | |
1842 | * Orphanizing is however only required for the first ref of an inode. | |
1843 | * process_recorded_refs does an additional is_first_ref check to see if | |
1844 | * orphanizing is really required. | |
1845 | */ | |
31db9f7c AB |
1846 | static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
1847 | const char *name, int name_len, | |
f5962781 | 1848 | u64 *who_ino, u64 *who_gen, u64 *who_mode) |
31db9f7c AB |
1849 | { |
1850 | int ret = 0; | |
ebdad913 | 1851 | u64 gen; |
31db9f7c AB |
1852 | u64 other_inode = 0; |
1853 | u8 other_type = 0; | |
1854 | ||
1855 | if (!sctx->parent_root) | |
1856 | goto out; | |
1857 | ||
1858 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1859 | if (ret <= 0) | |
1860 | goto out; | |
1861 | ||
ebdad913 JB |
1862 | /* |
1863 | * If we have a parent root we need to verify that the parent dir was | |
01327610 | 1864 | * not deleted and then re-created, if it was then we have no overwrite |
ebdad913 JB |
1865 | * and we can just unlink this entry. |
1866 | */ | |
4dd9920d | 1867 | if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) { |
ebdad913 JB |
1868 | ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, |
1869 | NULL, NULL, NULL); | |
1870 | if (ret < 0 && ret != -ENOENT) | |
1871 | goto out; | |
1872 | if (ret) { | |
1873 | ret = 0; | |
1874 | goto out; | |
1875 | } | |
1876 | if (gen != dir_gen) | |
1877 | goto out; | |
1878 | } | |
1879 | ||
31db9f7c AB |
1880 | ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, |
1881 | &other_inode, &other_type); | |
1882 | if (ret < 0 && ret != -ENOENT) | |
1883 | goto out; | |
1884 | if (ret) { | |
1885 | ret = 0; | |
1886 | goto out; | |
1887 | } | |
1888 | ||
766702ef AB |
1889 | /* |
1890 | * Check if the overwritten ref was already processed. If yes, the ref | |
1891 | * was already unlinked/moved, so we can safely assume that we will not | |
1892 | * overwrite anything at this point in time. | |
1893 | */ | |
801bec36 RK |
1894 | if (other_inode > sctx->send_progress || |
1895 | is_waiting_for_move(sctx, other_inode)) { | |
31db9f7c | 1896 | ret = get_inode_info(sctx->parent_root, other_inode, NULL, |
f5962781 | 1897 | who_gen, who_mode, NULL, NULL, NULL); |
31db9f7c AB |
1898 | if (ret < 0) |
1899 | goto out; | |
1900 | ||
1901 | ret = 1; | |
1902 | *who_ino = other_inode; | |
1903 | } else { | |
1904 | ret = 0; | |
1905 | } | |
1906 | ||
1907 | out: | |
1908 | return ret; | |
1909 | } | |
1910 | ||
766702ef AB |
1911 | /* |
1912 | * Checks if the ref was overwritten by an already processed inode. This is | |
1913 | * used by __get_cur_name_and_parent to find out if the ref was orphanized and | |
1914 | * thus the orphan name needs be used. | |
1915 | * process_recorded_refs also uses it to avoid unlinking of refs that were | |
1916 | * overwritten. | |
1917 | */ | |
31db9f7c AB |
1918 | static int did_overwrite_ref(struct send_ctx *sctx, |
1919 | u64 dir, u64 dir_gen, | |
1920 | u64 ino, u64 ino_gen, | |
1921 | const char *name, int name_len) | |
1922 | { | |
1923 | int ret = 0; | |
1924 | u64 gen; | |
1925 | u64 ow_inode; | |
1926 | u8 other_type; | |
1927 | ||
1928 | if (!sctx->parent_root) | |
1929 | goto out; | |
1930 | ||
1931 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1932 | if (ret <= 0) | |
1933 | goto out; | |
1934 | ||
01914101 RK |
1935 | if (dir != BTRFS_FIRST_FREE_OBJECTID) { |
1936 | ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, | |
1937 | NULL, NULL, NULL); | |
1938 | if (ret < 0 && ret != -ENOENT) | |
1939 | goto out; | |
1940 | if (ret) { | |
1941 | ret = 0; | |
1942 | goto out; | |
1943 | } | |
1944 | if (gen != dir_gen) | |
1945 | goto out; | |
1946 | } | |
1947 | ||
31db9f7c AB |
1948 | /* check if the ref was overwritten by another ref */ |
1949 | ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, | |
1950 | &ow_inode, &other_type); | |
1951 | if (ret < 0 && ret != -ENOENT) | |
1952 | goto out; | |
1953 | if (ret) { | |
1954 | /* was never and will never be overwritten */ | |
1955 | ret = 0; | |
1956 | goto out; | |
1957 | } | |
1958 | ||
1959 | ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, | |
85a7b33b | 1960 | NULL, NULL); |
31db9f7c AB |
1961 | if (ret < 0) |
1962 | goto out; | |
1963 | ||
1964 | if (ow_inode == ino && gen == ino_gen) { | |
1965 | ret = 0; | |
1966 | goto out; | |
1967 | } | |
1968 | ||
8b191a68 FM |
1969 | /* |
1970 | * We know that it is or will be overwritten. Check this now. | |
1971 | * The current inode being processed might have been the one that caused | |
b786f16a FM |
1972 | * inode 'ino' to be orphanized, therefore check if ow_inode matches |
1973 | * the current inode being processed. | |
8b191a68 | 1974 | */ |
b786f16a FM |
1975 | if ((ow_inode < sctx->send_progress) || |
1976 | (ino != sctx->cur_ino && ow_inode == sctx->cur_ino && | |
1977 | gen == sctx->cur_inode_gen)) | |
31db9f7c AB |
1978 | ret = 1; |
1979 | else | |
1980 | ret = 0; | |
1981 | ||
1982 | out: | |
1983 | return ret; | |
1984 | } | |
1985 | ||
766702ef AB |
1986 | /* |
1987 | * Same as did_overwrite_ref, but also checks if it is the first ref of an inode | |
1988 | * that got overwritten. This is used by process_recorded_refs to determine | |
1989 | * if it has to use the path as returned by get_cur_path or the orphan name. | |
1990 | */ | |
31db9f7c AB |
1991 | static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) |
1992 | { | |
1993 | int ret = 0; | |
1994 | struct fs_path *name = NULL; | |
1995 | u64 dir; | |
1996 | u64 dir_gen; | |
1997 | ||
1998 | if (!sctx->parent_root) | |
1999 | goto out; | |
2000 | ||
924794c9 | 2001 | name = fs_path_alloc(); |
31db9f7c AB |
2002 | if (!name) |
2003 | return -ENOMEM; | |
2004 | ||
924794c9 | 2005 | ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); |
31db9f7c AB |
2006 | if (ret < 0) |
2007 | goto out; | |
2008 | ||
2009 | ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, | |
2010 | name->start, fs_path_len(name)); | |
31db9f7c AB |
2011 | |
2012 | out: | |
924794c9 | 2013 | fs_path_free(name); |
31db9f7c AB |
2014 | return ret; |
2015 | } | |
2016 | ||
766702ef AB |
2017 | /* |
2018 | * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, | |
2019 | * so we need to do some special handling in case we have clashes. This function | |
2020 | * takes care of this with the help of name_cache_entry::radix_list. | |
5dc67d0b | 2021 | * In case of error, nce is kfreed. |
766702ef | 2022 | */ |
31db9f7c AB |
2023 | static int name_cache_insert(struct send_ctx *sctx, |
2024 | struct name_cache_entry *nce) | |
2025 | { | |
2026 | int ret = 0; | |
7e0926fe AB |
2027 | struct list_head *nce_head; |
2028 | ||
2029 | nce_head = radix_tree_lookup(&sctx->name_cache, | |
2030 | (unsigned long)nce->ino); | |
2031 | if (!nce_head) { | |
e780b0d1 | 2032 | nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL); |
cfa7a9cc TI |
2033 | if (!nce_head) { |
2034 | kfree(nce); | |
31db9f7c | 2035 | return -ENOMEM; |
cfa7a9cc | 2036 | } |
7e0926fe | 2037 | INIT_LIST_HEAD(nce_head); |
31db9f7c | 2038 | |
7e0926fe | 2039 | ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); |
5dc67d0b AB |
2040 | if (ret < 0) { |
2041 | kfree(nce_head); | |
2042 | kfree(nce); | |
31db9f7c | 2043 | return ret; |
5dc67d0b | 2044 | } |
31db9f7c | 2045 | } |
7e0926fe | 2046 | list_add_tail(&nce->radix_list, nce_head); |
31db9f7c AB |
2047 | list_add_tail(&nce->list, &sctx->name_cache_list); |
2048 | sctx->name_cache_size++; | |
2049 | ||
2050 | return ret; | |
2051 | } | |
2052 | ||
2053 | static void name_cache_delete(struct send_ctx *sctx, | |
2054 | struct name_cache_entry *nce) | |
2055 | { | |
7e0926fe | 2056 | struct list_head *nce_head; |
31db9f7c | 2057 | |
7e0926fe AB |
2058 | nce_head = radix_tree_lookup(&sctx->name_cache, |
2059 | (unsigned long)nce->ino); | |
57fb8910 DS |
2060 | if (!nce_head) { |
2061 | btrfs_err(sctx->send_root->fs_info, | |
2062 | "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", | |
2063 | nce->ino, sctx->name_cache_size); | |
2064 | } | |
31db9f7c | 2065 | |
7e0926fe | 2066 | list_del(&nce->radix_list); |
31db9f7c | 2067 | list_del(&nce->list); |
31db9f7c | 2068 | sctx->name_cache_size--; |
7e0926fe | 2069 | |
57fb8910 DS |
2070 | /* |
2071 | * We may not get to the final release of nce_head if the lookup fails | |
2072 | */ | |
2073 | if (nce_head && list_empty(nce_head)) { | |
7e0926fe AB |
2074 | radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); |
2075 | kfree(nce_head); | |
2076 | } | |
31db9f7c AB |
2077 | } |
2078 | ||
2079 | static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, | |
2080 | u64 ino, u64 gen) | |
2081 | { | |
7e0926fe AB |
2082 | struct list_head *nce_head; |
2083 | struct name_cache_entry *cur; | |
31db9f7c | 2084 | |
7e0926fe AB |
2085 | nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); |
2086 | if (!nce_head) | |
31db9f7c AB |
2087 | return NULL; |
2088 | ||
7e0926fe AB |
2089 | list_for_each_entry(cur, nce_head, radix_list) { |
2090 | if (cur->ino == ino && cur->gen == gen) | |
2091 | return cur; | |
2092 | } | |
31db9f7c AB |
2093 | return NULL; |
2094 | } | |
2095 | ||
766702ef AB |
2096 | /* |
2097 | * Removes the entry from the list and adds it back to the end. This marks the | |
2098 | * entry as recently used so that name_cache_clean_unused does not remove it. | |
2099 | */ | |
31db9f7c AB |
2100 | static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) |
2101 | { | |
2102 | list_del(&nce->list); | |
2103 | list_add_tail(&nce->list, &sctx->name_cache_list); | |
2104 | } | |
2105 | ||
766702ef AB |
2106 | /* |
2107 | * Remove some entries from the beginning of name_cache_list. | |
2108 | */ | |
31db9f7c AB |
2109 | static void name_cache_clean_unused(struct send_ctx *sctx) |
2110 | { | |
2111 | struct name_cache_entry *nce; | |
2112 | ||
2113 | if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) | |
2114 | return; | |
2115 | ||
2116 | while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { | |
2117 | nce = list_entry(sctx->name_cache_list.next, | |
2118 | struct name_cache_entry, list); | |
2119 | name_cache_delete(sctx, nce); | |
2120 | kfree(nce); | |
2121 | } | |
2122 | } | |
2123 | ||
2124 | static void name_cache_free(struct send_ctx *sctx) | |
2125 | { | |
2126 | struct name_cache_entry *nce; | |
31db9f7c | 2127 | |
e938c8ad AB |
2128 | while (!list_empty(&sctx->name_cache_list)) { |
2129 | nce = list_entry(sctx->name_cache_list.next, | |
2130 | struct name_cache_entry, list); | |
31db9f7c | 2131 | name_cache_delete(sctx, nce); |
17589bd9 | 2132 | kfree(nce); |
31db9f7c AB |
2133 | } |
2134 | } | |
2135 | ||
766702ef AB |
2136 | /* |
2137 | * Used by get_cur_path for each ref up to the root. | |
2138 | * Returns 0 if it succeeded. | |
2139 | * Returns 1 if the inode is not existent or got overwritten. In that case, the | |
2140 | * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 | |
2141 | * is returned, parent_ino/parent_gen are not guaranteed to be valid. | |
2142 | * Returns <0 in case of error. | |
2143 | */ | |
31db9f7c AB |
2144 | static int __get_cur_name_and_parent(struct send_ctx *sctx, |
2145 | u64 ino, u64 gen, | |
2146 | u64 *parent_ino, | |
2147 | u64 *parent_gen, | |
2148 | struct fs_path *dest) | |
2149 | { | |
2150 | int ret; | |
2151 | int nce_ret; | |
31db9f7c AB |
2152 | struct name_cache_entry *nce = NULL; |
2153 | ||
766702ef AB |
2154 | /* |
2155 | * First check if we already did a call to this function with the same | |
2156 | * ino/gen. If yes, check if the cache entry is still up-to-date. If yes | |
2157 | * return the cached result. | |
2158 | */ | |
31db9f7c AB |
2159 | nce = name_cache_search(sctx, ino, gen); |
2160 | if (nce) { | |
2161 | if (ino < sctx->send_progress && nce->need_later_update) { | |
2162 | name_cache_delete(sctx, nce); | |
2163 | kfree(nce); | |
2164 | nce = NULL; | |
2165 | } else { | |
2166 | name_cache_used(sctx, nce); | |
2167 | *parent_ino = nce->parent_ino; | |
2168 | *parent_gen = nce->parent_gen; | |
2169 | ret = fs_path_add(dest, nce->name, nce->name_len); | |
2170 | if (ret < 0) | |
2171 | goto out; | |
2172 | ret = nce->ret; | |
2173 | goto out; | |
2174 | } | |
2175 | } | |
2176 | ||
766702ef AB |
2177 | /* |
2178 | * If the inode is not existent yet, add the orphan name and return 1. | |
2179 | * This should only happen for the parent dir that we determine in | |
2180 | * __record_new_ref | |
2181 | */ | |
31db9f7c AB |
2182 | ret = is_inode_existent(sctx, ino, gen); |
2183 | if (ret < 0) | |
2184 | goto out; | |
2185 | ||
2186 | if (!ret) { | |
2187 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2188 | if (ret < 0) | |
2189 | goto out; | |
2190 | ret = 1; | |
2191 | goto out_cache; | |
2192 | } | |
2193 | ||
766702ef AB |
2194 | /* |
2195 | * Depending on whether the inode was already processed or not, use | |
2196 | * send_root or parent_root for ref lookup. | |
2197 | */ | |
bf0d1f44 | 2198 | if (ino < sctx->send_progress) |
924794c9 TI |
2199 | ret = get_first_ref(sctx->send_root, ino, |
2200 | parent_ino, parent_gen, dest); | |
31db9f7c | 2201 | else |
924794c9 TI |
2202 | ret = get_first_ref(sctx->parent_root, ino, |
2203 | parent_ino, parent_gen, dest); | |
31db9f7c AB |
2204 | if (ret < 0) |
2205 | goto out; | |
2206 | ||
766702ef AB |
2207 | /* |
2208 | * Check if the ref was overwritten by an inode's ref that was processed | |
2209 | * earlier. If yes, treat as orphan and return 1. | |
2210 | */ | |
31db9f7c AB |
2211 | ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, |
2212 | dest->start, dest->end - dest->start); | |
2213 | if (ret < 0) | |
2214 | goto out; | |
2215 | if (ret) { | |
2216 | fs_path_reset(dest); | |
2217 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2218 | if (ret < 0) | |
2219 | goto out; | |
2220 | ret = 1; | |
2221 | } | |
2222 | ||
2223 | out_cache: | |
766702ef AB |
2224 | /* |
2225 | * Store the result of the lookup in the name cache. | |
2226 | */ | |
e780b0d1 | 2227 | nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); |
31db9f7c AB |
2228 | if (!nce) { |
2229 | ret = -ENOMEM; | |
2230 | goto out; | |
2231 | } | |
2232 | ||
2233 | nce->ino = ino; | |
2234 | nce->gen = gen; | |
2235 | nce->parent_ino = *parent_ino; | |
2236 | nce->parent_gen = *parent_gen; | |
2237 | nce->name_len = fs_path_len(dest); | |
2238 | nce->ret = ret; | |
2239 | strcpy(nce->name, dest->start); | |
31db9f7c AB |
2240 | |
2241 | if (ino < sctx->send_progress) | |
2242 | nce->need_later_update = 0; | |
2243 | else | |
2244 | nce->need_later_update = 1; | |
2245 | ||
2246 | nce_ret = name_cache_insert(sctx, nce); | |
2247 | if (nce_ret < 0) | |
2248 | ret = nce_ret; | |
2249 | name_cache_clean_unused(sctx); | |
2250 | ||
2251 | out: | |
31db9f7c AB |
2252 | return ret; |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * Magic happens here. This function returns the first ref to an inode as it | |
2257 | * would look like while receiving the stream at this point in time. | |
2258 | * We walk the path up to the root. For every inode in between, we check if it | |
2259 | * was already processed/sent. If yes, we continue with the parent as found | |
2260 | * in send_root. If not, we continue with the parent as found in parent_root. | |
2261 | * If we encounter an inode that was deleted at this point in time, we use the | |
2262 | * inodes "orphan" name instead of the real name and stop. Same with new inodes | |
2263 | * that were not created yet and overwritten inodes/refs. | |
2264 | * | |
52042d8e | 2265 | * When do we have orphan inodes: |
31db9f7c AB |
2266 | * 1. When an inode is freshly created and thus no valid refs are available yet |
2267 | * 2. When a directory lost all it's refs (deleted) but still has dir items | |
2268 | * inside which were not processed yet (pending for move/delete). If anyone | |
2269 | * tried to get the path to the dir items, it would get a path inside that | |
2270 | * orphan directory. | |
2271 | * 3. When an inode is moved around or gets new links, it may overwrite the ref | |
2272 | * of an unprocessed inode. If in that case the first ref would be | |
2273 | * overwritten, the overwritten inode gets "orphanized". Later when we | |
2274 | * process this overwritten inode, it is restored at a new place by moving | |
2275 | * the orphan inode. | |
2276 | * | |
2277 | * sctx->send_progress tells this function at which point in time receiving | |
2278 | * would be. | |
2279 | */ | |
2280 | static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, | |
2281 | struct fs_path *dest) | |
2282 | { | |
2283 | int ret = 0; | |
2284 | struct fs_path *name = NULL; | |
2285 | u64 parent_inode = 0; | |
2286 | u64 parent_gen = 0; | |
2287 | int stop = 0; | |
2288 | ||
924794c9 | 2289 | name = fs_path_alloc(); |
31db9f7c AB |
2290 | if (!name) { |
2291 | ret = -ENOMEM; | |
2292 | goto out; | |
2293 | } | |
2294 | ||
2295 | dest->reversed = 1; | |
2296 | fs_path_reset(dest); | |
2297 | ||
2298 | while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { | |
8b191a68 FM |
2299 | struct waiting_dir_move *wdm; |
2300 | ||
31db9f7c AB |
2301 | fs_path_reset(name); |
2302 | ||
0b3f407e | 2303 | if (is_waiting_for_rm(sctx, ino, gen)) { |
9dc44214 FM |
2304 | ret = gen_unique_name(sctx, ino, gen, name); |
2305 | if (ret < 0) | |
2306 | goto out; | |
2307 | ret = fs_path_add_path(dest, name); | |
2308 | break; | |
2309 | } | |
2310 | ||
8b191a68 FM |
2311 | wdm = get_waiting_dir_move(sctx, ino); |
2312 | if (wdm && wdm->orphanized) { | |
2313 | ret = gen_unique_name(sctx, ino, gen, name); | |
2314 | stop = 1; | |
2315 | } else if (wdm) { | |
bf0d1f44 FM |
2316 | ret = get_first_ref(sctx->parent_root, ino, |
2317 | &parent_inode, &parent_gen, name); | |
2318 | } else { | |
2319 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
2320 | &parent_inode, | |
2321 | &parent_gen, name); | |
2322 | if (ret) | |
2323 | stop = 1; | |
2324 | } | |
2325 | ||
31db9f7c AB |
2326 | if (ret < 0) |
2327 | goto out; | |
9f03740a | 2328 | |
31db9f7c AB |
2329 | ret = fs_path_add_path(dest, name); |
2330 | if (ret < 0) | |
2331 | goto out; | |
2332 | ||
2333 | ino = parent_inode; | |
2334 | gen = parent_gen; | |
2335 | } | |
2336 | ||
2337 | out: | |
924794c9 | 2338 | fs_path_free(name); |
31db9f7c AB |
2339 | if (!ret) |
2340 | fs_path_unreverse(dest); | |
2341 | return ret; | |
2342 | } | |
2343 | ||
31db9f7c AB |
2344 | /* |
2345 | * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace | |
2346 | */ | |
2347 | static int send_subvol_begin(struct send_ctx *sctx) | |
2348 | { | |
2349 | int ret; | |
2350 | struct btrfs_root *send_root = sctx->send_root; | |
2351 | struct btrfs_root *parent_root = sctx->parent_root; | |
2352 | struct btrfs_path *path; | |
2353 | struct btrfs_key key; | |
2354 | struct btrfs_root_ref *ref; | |
2355 | struct extent_buffer *leaf; | |
2356 | char *name = NULL; | |
2357 | int namelen; | |
2358 | ||
ffcfaf81 | 2359 | path = btrfs_alloc_path(); |
31db9f7c AB |
2360 | if (!path) |
2361 | return -ENOMEM; | |
2362 | ||
e780b0d1 | 2363 | name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); |
31db9f7c AB |
2364 | if (!name) { |
2365 | btrfs_free_path(path); | |
2366 | return -ENOMEM; | |
2367 | } | |
2368 | ||
4fd786e6 | 2369 | key.objectid = send_root->root_key.objectid; |
31db9f7c AB |
2370 | key.type = BTRFS_ROOT_BACKREF_KEY; |
2371 | key.offset = 0; | |
2372 | ||
2373 | ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, | |
2374 | &key, path, 1, 0); | |
2375 | if (ret < 0) | |
2376 | goto out; | |
2377 | if (ret) { | |
2378 | ret = -ENOENT; | |
2379 | goto out; | |
2380 | } | |
2381 | ||
2382 | leaf = path->nodes[0]; | |
2383 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2384 | if (key.type != BTRFS_ROOT_BACKREF_KEY || | |
4fd786e6 | 2385 | key.objectid != send_root->root_key.objectid) { |
31db9f7c AB |
2386 | ret = -ENOENT; |
2387 | goto out; | |
2388 | } | |
2389 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
2390 | namelen = btrfs_root_ref_name_len(leaf, ref); | |
2391 | read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); | |
2392 | btrfs_release_path(path); | |
2393 | ||
31db9f7c AB |
2394 | if (parent_root) { |
2395 | ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); | |
2396 | if (ret < 0) | |
2397 | goto out; | |
2398 | } else { | |
2399 | ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); | |
2400 | if (ret < 0) | |
2401 | goto out; | |
2402 | } | |
2403 | ||
2404 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); | |
b96b1db0 RR |
2405 | |
2406 | if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid)) | |
2407 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2408 | sctx->send_root->root_item.received_uuid); | |
2409 | else | |
2410 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2411 | sctx->send_root->root_item.uuid); | |
2412 | ||
31db9f7c | 2413 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, |
09e3a288 | 2414 | btrfs_root_ctransid(&sctx->send_root->root_item)); |
31db9f7c | 2415 | if (parent_root) { |
37b8d27d JB |
2416 | if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) |
2417 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2418 | parent_root->root_item.received_uuid); | |
2419 | else | |
2420 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2421 | parent_root->root_item.uuid); | |
31db9f7c | 2422 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
09e3a288 | 2423 | btrfs_root_ctransid(&sctx->parent_root->root_item)); |
31db9f7c AB |
2424 | } |
2425 | ||
2426 | ret = send_cmd(sctx); | |
2427 | ||
2428 | tlv_put_failure: | |
2429 | out: | |
2430 | btrfs_free_path(path); | |
2431 | kfree(name); | |
2432 | return ret; | |
2433 | } | |
2434 | ||
2435 | static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) | |
2436 | { | |
04ab956e | 2437 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2438 | int ret = 0; |
2439 | struct fs_path *p; | |
2440 | ||
04ab956e | 2441 | btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size); |
31db9f7c | 2442 | |
924794c9 | 2443 | p = fs_path_alloc(); |
31db9f7c AB |
2444 | if (!p) |
2445 | return -ENOMEM; | |
2446 | ||
2447 | ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); | |
2448 | if (ret < 0) | |
2449 | goto out; | |
2450 | ||
2451 | ret = get_cur_path(sctx, ino, gen, p); | |
2452 | if (ret < 0) | |
2453 | goto out; | |
2454 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2455 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); | |
2456 | ||
2457 | ret = send_cmd(sctx); | |
2458 | ||
2459 | tlv_put_failure: | |
2460 | out: | |
924794c9 | 2461 | fs_path_free(p); |
31db9f7c AB |
2462 | return ret; |
2463 | } | |
2464 | ||
2465 | static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) | |
2466 | { | |
04ab956e | 2467 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2468 | int ret = 0; |
2469 | struct fs_path *p; | |
2470 | ||
04ab956e | 2471 | btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode); |
31db9f7c | 2472 | |
924794c9 | 2473 | p = fs_path_alloc(); |
31db9f7c AB |
2474 | if (!p) |
2475 | return -ENOMEM; | |
2476 | ||
2477 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); | |
2478 | if (ret < 0) | |
2479 | goto out; | |
2480 | ||
2481 | ret = get_cur_path(sctx, ino, gen, p); | |
2482 | if (ret < 0) | |
2483 | goto out; | |
2484 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2485 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); | |
2486 | ||
2487 | ret = send_cmd(sctx); | |
2488 | ||
2489 | tlv_put_failure: | |
2490 | out: | |
924794c9 | 2491 | fs_path_free(p); |
31db9f7c AB |
2492 | return ret; |
2493 | } | |
2494 | ||
2495 | static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) | |
2496 | { | |
04ab956e | 2497 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2498 | int ret = 0; |
2499 | struct fs_path *p; | |
2500 | ||
04ab956e JM |
2501 | btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu", |
2502 | ino, uid, gid); | |
31db9f7c | 2503 | |
924794c9 | 2504 | p = fs_path_alloc(); |
31db9f7c AB |
2505 | if (!p) |
2506 | return -ENOMEM; | |
2507 | ||
2508 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); | |
2509 | if (ret < 0) | |
2510 | goto out; | |
2511 | ||
2512 | ret = get_cur_path(sctx, ino, gen, p); | |
2513 | if (ret < 0) | |
2514 | goto out; | |
2515 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2516 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); | |
2517 | TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); | |
2518 | ||
2519 | ret = send_cmd(sctx); | |
2520 | ||
2521 | tlv_put_failure: | |
2522 | out: | |
924794c9 | 2523 | fs_path_free(p); |
31db9f7c AB |
2524 | return ret; |
2525 | } | |
2526 | ||
2527 | static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) | |
2528 | { | |
04ab956e | 2529 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2530 | int ret = 0; |
2531 | struct fs_path *p = NULL; | |
2532 | struct btrfs_inode_item *ii; | |
2533 | struct btrfs_path *path = NULL; | |
2534 | struct extent_buffer *eb; | |
2535 | struct btrfs_key key; | |
2536 | int slot; | |
2537 | ||
04ab956e | 2538 | btrfs_debug(fs_info, "send_utimes %llu", ino); |
31db9f7c | 2539 | |
924794c9 | 2540 | p = fs_path_alloc(); |
31db9f7c AB |
2541 | if (!p) |
2542 | return -ENOMEM; | |
2543 | ||
2544 | path = alloc_path_for_send(); | |
2545 | if (!path) { | |
2546 | ret = -ENOMEM; | |
2547 | goto out; | |
2548 | } | |
2549 | ||
2550 | key.objectid = ino; | |
2551 | key.type = BTRFS_INODE_ITEM_KEY; | |
2552 | key.offset = 0; | |
2553 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); | |
15b253ea FM |
2554 | if (ret > 0) |
2555 | ret = -ENOENT; | |
31db9f7c AB |
2556 | if (ret < 0) |
2557 | goto out; | |
2558 | ||
2559 | eb = path->nodes[0]; | |
2560 | slot = path->slots[0]; | |
2561 | ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
2562 | ||
2563 | ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); | |
2564 | if (ret < 0) | |
2565 | goto out; | |
2566 | ||
2567 | ret = get_cur_path(sctx, ino, gen, p); | |
2568 | if (ret < 0) | |
2569 | goto out; | |
2570 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
a937b979 DS |
2571 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); |
2572 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); | |
2573 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); | |
766702ef | 2574 | /* TODO Add otime support when the otime patches get into upstream */ |
31db9f7c AB |
2575 | |
2576 | ret = send_cmd(sctx); | |
2577 | ||
2578 | tlv_put_failure: | |
2579 | out: | |
924794c9 | 2580 | fs_path_free(p); |
31db9f7c AB |
2581 | btrfs_free_path(path); |
2582 | return ret; | |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have | |
2587 | * a valid path yet because we did not process the refs yet. So, the inode | |
2588 | * is created as orphan. | |
2589 | */ | |
1f4692da | 2590 | static int send_create_inode(struct send_ctx *sctx, u64 ino) |
31db9f7c | 2591 | { |
04ab956e | 2592 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c | 2593 | int ret = 0; |
31db9f7c | 2594 | struct fs_path *p; |
31db9f7c | 2595 | int cmd; |
1f4692da | 2596 | u64 gen; |
31db9f7c | 2597 | u64 mode; |
1f4692da | 2598 | u64 rdev; |
31db9f7c | 2599 | |
04ab956e | 2600 | btrfs_debug(fs_info, "send_create_inode %llu", ino); |
31db9f7c | 2601 | |
924794c9 | 2602 | p = fs_path_alloc(); |
31db9f7c AB |
2603 | if (!p) |
2604 | return -ENOMEM; | |
2605 | ||
644d1940 LB |
2606 | if (ino != sctx->cur_ino) { |
2607 | ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, | |
2608 | NULL, NULL, &rdev); | |
2609 | if (ret < 0) | |
2610 | goto out; | |
2611 | } else { | |
2612 | gen = sctx->cur_inode_gen; | |
2613 | mode = sctx->cur_inode_mode; | |
2614 | rdev = sctx->cur_inode_rdev; | |
2615 | } | |
31db9f7c | 2616 | |
e938c8ad | 2617 | if (S_ISREG(mode)) { |
31db9f7c | 2618 | cmd = BTRFS_SEND_C_MKFILE; |
e938c8ad | 2619 | } else if (S_ISDIR(mode)) { |
31db9f7c | 2620 | cmd = BTRFS_SEND_C_MKDIR; |
e938c8ad | 2621 | } else if (S_ISLNK(mode)) { |
31db9f7c | 2622 | cmd = BTRFS_SEND_C_SYMLINK; |
e938c8ad | 2623 | } else if (S_ISCHR(mode) || S_ISBLK(mode)) { |
31db9f7c | 2624 | cmd = BTRFS_SEND_C_MKNOD; |
e938c8ad | 2625 | } else if (S_ISFIFO(mode)) { |
31db9f7c | 2626 | cmd = BTRFS_SEND_C_MKFIFO; |
e938c8ad | 2627 | } else if (S_ISSOCK(mode)) { |
31db9f7c | 2628 | cmd = BTRFS_SEND_C_MKSOCK; |
e938c8ad | 2629 | } else { |
f14d104d | 2630 | btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o", |
31db9f7c | 2631 | (int)(mode & S_IFMT)); |
ca6842bf | 2632 | ret = -EOPNOTSUPP; |
31db9f7c AB |
2633 | goto out; |
2634 | } | |
2635 | ||
2636 | ret = begin_cmd(sctx, cmd); | |
2637 | if (ret < 0) | |
2638 | goto out; | |
2639 | ||
1f4692da | 2640 | ret = gen_unique_name(sctx, ino, gen, p); |
31db9f7c AB |
2641 | if (ret < 0) |
2642 | goto out; | |
2643 | ||
2644 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
1f4692da | 2645 | TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); |
31db9f7c AB |
2646 | |
2647 | if (S_ISLNK(mode)) { | |
2648 | fs_path_reset(p); | |
924794c9 | 2649 | ret = read_symlink(sctx->send_root, ino, p); |
31db9f7c AB |
2650 | if (ret < 0) |
2651 | goto out; | |
2652 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); | |
2653 | } else if (S_ISCHR(mode) || S_ISBLK(mode) || | |
2654 | S_ISFIFO(mode) || S_ISSOCK(mode)) { | |
d79e5043 AJ |
2655 | TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); |
2656 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); | |
31db9f7c AB |
2657 | } |
2658 | ||
2659 | ret = send_cmd(sctx); | |
2660 | if (ret < 0) | |
2661 | goto out; | |
2662 | ||
2663 | ||
2664 | tlv_put_failure: | |
2665 | out: | |
924794c9 | 2666 | fs_path_free(p); |
31db9f7c AB |
2667 | return ret; |
2668 | } | |
2669 | ||
1f4692da AB |
2670 | /* |
2671 | * We need some special handling for inodes that get processed before the parent | |
2672 | * directory got created. See process_recorded_refs for details. | |
2673 | * This function does the check if we already created the dir out of order. | |
2674 | */ | |
2675 | static int did_create_dir(struct send_ctx *sctx, u64 dir) | |
2676 | { | |
2677 | int ret = 0; | |
2678 | struct btrfs_path *path = NULL; | |
2679 | struct btrfs_key key; | |
2680 | struct btrfs_key found_key; | |
2681 | struct btrfs_key di_key; | |
2682 | struct extent_buffer *eb; | |
2683 | struct btrfs_dir_item *di; | |
2684 | int slot; | |
2685 | ||
2686 | path = alloc_path_for_send(); | |
2687 | if (!path) { | |
2688 | ret = -ENOMEM; | |
2689 | goto out; | |
2690 | } | |
2691 | ||
2692 | key.objectid = dir; | |
2693 | key.type = BTRFS_DIR_INDEX_KEY; | |
2694 | key.offset = 0; | |
dff6d0ad FDBM |
2695 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); |
2696 | if (ret < 0) | |
2697 | goto out; | |
2698 | ||
1f4692da | 2699 | while (1) { |
dff6d0ad FDBM |
2700 | eb = path->nodes[0]; |
2701 | slot = path->slots[0]; | |
2702 | if (slot >= btrfs_header_nritems(eb)) { | |
2703 | ret = btrfs_next_leaf(sctx->send_root, path); | |
2704 | if (ret < 0) { | |
2705 | goto out; | |
2706 | } else if (ret > 0) { | |
2707 | ret = 0; | |
2708 | break; | |
2709 | } | |
2710 | continue; | |
1f4692da | 2711 | } |
dff6d0ad FDBM |
2712 | |
2713 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
2714 | if (found_key.objectid != key.objectid || | |
1f4692da AB |
2715 | found_key.type != key.type) { |
2716 | ret = 0; | |
2717 | goto out; | |
2718 | } | |
2719 | ||
2720 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); | |
2721 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2722 | ||
a0525414 JB |
2723 | if (di_key.type != BTRFS_ROOT_ITEM_KEY && |
2724 | di_key.objectid < sctx->send_progress) { | |
1f4692da AB |
2725 | ret = 1; |
2726 | goto out; | |
2727 | } | |
2728 | ||
dff6d0ad | 2729 | path->slots[0]++; |
1f4692da AB |
2730 | } |
2731 | ||
2732 | out: | |
2733 | btrfs_free_path(path); | |
2734 | return ret; | |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * Only creates the inode if it is: | |
2739 | * 1. Not a directory | |
2740 | * 2. Or a directory which was not created already due to out of order | |
2741 | * directories. See did_create_dir and process_recorded_refs for details. | |
2742 | */ | |
2743 | static int send_create_inode_if_needed(struct send_ctx *sctx) | |
2744 | { | |
2745 | int ret; | |
2746 | ||
2747 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
2748 | ret = did_create_dir(sctx, sctx->cur_ino); | |
2749 | if (ret < 0) | |
2750 | goto out; | |
2751 | if (ret) { | |
2752 | ret = 0; | |
2753 | goto out; | |
2754 | } | |
2755 | } | |
2756 | ||
2757 | ret = send_create_inode(sctx, sctx->cur_ino); | |
2758 | if (ret < 0) | |
2759 | goto out; | |
2760 | ||
2761 | out: | |
2762 | return ret; | |
2763 | } | |
2764 | ||
31db9f7c AB |
2765 | struct recorded_ref { |
2766 | struct list_head list; | |
31db9f7c AB |
2767 | char *name; |
2768 | struct fs_path *full_path; | |
2769 | u64 dir; | |
2770 | u64 dir_gen; | |
31db9f7c AB |
2771 | int name_len; |
2772 | }; | |
2773 | ||
fdb13889 FM |
2774 | static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) |
2775 | { | |
2776 | ref->full_path = path; | |
2777 | ref->name = (char *)kbasename(ref->full_path->start); | |
2778 | ref->name_len = ref->full_path->end - ref->name; | |
2779 | } | |
2780 | ||
31db9f7c AB |
2781 | /* |
2782 | * We need to process new refs before deleted refs, but compare_tree gives us | |
2783 | * everything mixed. So we first record all refs and later process them. | |
2784 | * This function is a helper to record one ref. | |
2785 | */ | |
a4d96d62 | 2786 | static int __record_ref(struct list_head *head, u64 dir, |
31db9f7c AB |
2787 | u64 dir_gen, struct fs_path *path) |
2788 | { | |
2789 | struct recorded_ref *ref; | |
31db9f7c | 2790 | |
e780b0d1 | 2791 | ref = kmalloc(sizeof(*ref), GFP_KERNEL); |
31db9f7c AB |
2792 | if (!ref) |
2793 | return -ENOMEM; | |
2794 | ||
2795 | ref->dir = dir; | |
2796 | ref->dir_gen = dir_gen; | |
fdb13889 | 2797 | set_ref_path(ref, path); |
31db9f7c AB |
2798 | list_add_tail(&ref->list, head); |
2799 | return 0; | |
2800 | } | |
2801 | ||
ba5e8f2e JB |
2802 | static int dup_ref(struct recorded_ref *ref, struct list_head *list) |
2803 | { | |
2804 | struct recorded_ref *new; | |
2805 | ||
e780b0d1 | 2806 | new = kmalloc(sizeof(*ref), GFP_KERNEL); |
ba5e8f2e JB |
2807 | if (!new) |
2808 | return -ENOMEM; | |
2809 | ||
2810 | new->dir = ref->dir; | |
2811 | new->dir_gen = ref->dir_gen; | |
2812 | new->full_path = NULL; | |
2813 | INIT_LIST_HEAD(&new->list); | |
2814 | list_add_tail(&new->list, list); | |
2815 | return 0; | |
2816 | } | |
2817 | ||
924794c9 | 2818 | static void __free_recorded_refs(struct list_head *head) |
31db9f7c AB |
2819 | { |
2820 | struct recorded_ref *cur; | |
31db9f7c | 2821 | |
e938c8ad AB |
2822 | while (!list_empty(head)) { |
2823 | cur = list_entry(head->next, struct recorded_ref, list); | |
924794c9 | 2824 | fs_path_free(cur->full_path); |
e938c8ad | 2825 | list_del(&cur->list); |
31db9f7c AB |
2826 | kfree(cur); |
2827 | } | |
31db9f7c AB |
2828 | } |
2829 | ||
2830 | static void free_recorded_refs(struct send_ctx *sctx) | |
2831 | { | |
924794c9 TI |
2832 | __free_recorded_refs(&sctx->new_refs); |
2833 | __free_recorded_refs(&sctx->deleted_refs); | |
31db9f7c AB |
2834 | } |
2835 | ||
2836 | /* | |
766702ef | 2837 | * Renames/moves a file/dir to its orphan name. Used when the first |
31db9f7c AB |
2838 | * ref of an unprocessed inode gets overwritten and for all non empty |
2839 | * directories. | |
2840 | */ | |
2841 | static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, | |
2842 | struct fs_path *path) | |
2843 | { | |
2844 | int ret; | |
2845 | struct fs_path *orphan; | |
2846 | ||
924794c9 | 2847 | orphan = fs_path_alloc(); |
31db9f7c AB |
2848 | if (!orphan) |
2849 | return -ENOMEM; | |
2850 | ||
2851 | ret = gen_unique_name(sctx, ino, gen, orphan); | |
2852 | if (ret < 0) | |
2853 | goto out; | |
2854 | ||
2855 | ret = send_rename(sctx, path, orphan); | |
2856 | ||
2857 | out: | |
924794c9 | 2858 | fs_path_free(orphan); |
31db9f7c AB |
2859 | return ret; |
2860 | } | |
2861 | ||
0b3f407e FM |
2862 | static struct orphan_dir_info *add_orphan_dir_info(struct send_ctx *sctx, |
2863 | u64 dir_ino, u64 dir_gen) | |
9dc44214 FM |
2864 | { |
2865 | struct rb_node **p = &sctx->orphan_dirs.rb_node; | |
2866 | struct rb_node *parent = NULL; | |
2867 | struct orphan_dir_info *entry, *odi; | |
2868 | ||
9dc44214 FM |
2869 | while (*p) { |
2870 | parent = *p; | |
2871 | entry = rb_entry(parent, struct orphan_dir_info, node); | |
0b3f407e | 2872 | if (dir_ino < entry->ino) |
9dc44214 | 2873 | p = &(*p)->rb_left; |
0b3f407e | 2874 | else if (dir_ino > entry->ino) |
9dc44214 | 2875 | p = &(*p)->rb_right; |
0b3f407e FM |
2876 | else if (dir_gen < entry->gen) |
2877 | p = &(*p)->rb_left; | |
2878 | else if (dir_gen > entry->gen) | |
2879 | p = &(*p)->rb_right; | |
2880 | else | |
9dc44214 | 2881 | return entry; |
9dc44214 FM |
2882 | } |
2883 | ||
35c8eda1 RK |
2884 | odi = kmalloc(sizeof(*odi), GFP_KERNEL); |
2885 | if (!odi) | |
2886 | return ERR_PTR(-ENOMEM); | |
2887 | odi->ino = dir_ino; | |
0b3f407e | 2888 | odi->gen = dir_gen; |
0f96f517 | 2889 | odi->last_dir_index_offset = 0; |
35c8eda1 | 2890 | |
9dc44214 FM |
2891 | rb_link_node(&odi->node, parent, p); |
2892 | rb_insert_color(&odi->node, &sctx->orphan_dirs); | |
2893 | return odi; | |
2894 | } | |
2895 | ||
0b3f407e FM |
2896 | static struct orphan_dir_info *get_orphan_dir_info(struct send_ctx *sctx, |
2897 | u64 dir_ino, u64 gen) | |
9dc44214 FM |
2898 | { |
2899 | struct rb_node *n = sctx->orphan_dirs.rb_node; | |
2900 | struct orphan_dir_info *entry; | |
2901 | ||
2902 | while (n) { | |
2903 | entry = rb_entry(n, struct orphan_dir_info, node); | |
2904 | if (dir_ino < entry->ino) | |
2905 | n = n->rb_left; | |
2906 | else if (dir_ino > entry->ino) | |
2907 | n = n->rb_right; | |
0b3f407e FM |
2908 | else if (gen < entry->gen) |
2909 | n = n->rb_left; | |
2910 | else if (gen > entry->gen) | |
2911 | n = n->rb_right; | |
9dc44214 FM |
2912 | else |
2913 | return entry; | |
2914 | } | |
2915 | return NULL; | |
2916 | } | |
2917 | ||
0b3f407e | 2918 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen) |
9dc44214 | 2919 | { |
0b3f407e | 2920 | struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino, gen); |
9dc44214 FM |
2921 | |
2922 | return odi != NULL; | |
2923 | } | |
2924 | ||
2925 | static void free_orphan_dir_info(struct send_ctx *sctx, | |
2926 | struct orphan_dir_info *odi) | |
2927 | { | |
2928 | if (!odi) | |
2929 | return; | |
2930 | rb_erase(&odi->node, &sctx->orphan_dirs); | |
2931 | kfree(odi); | |
2932 | } | |
2933 | ||
31db9f7c AB |
2934 | /* |
2935 | * Returns 1 if a directory can be removed at this point in time. | |
2936 | * We check this by iterating all dir items and checking if the inode behind | |
2937 | * the dir item was already processed. | |
2938 | */ | |
9dc44214 FM |
2939 | static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
2940 | u64 send_progress) | |
31db9f7c AB |
2941 | { |
2942 | int ret = 0; | |
2943 | struct btrfs_root *root = sctx->parent_root; | |
2944 | struct btrfs_path *path; | |
2945 | struct btrfs_key key; | |
2946 | struct btrfs_key found_key; | |
2947 | struct btrfs_key loc; | |
2948 | struct btrfs_dir_item *di; | |
0f96f517 | 2949 | struct orphan_dir_info *odi = NULL; |
31db9f7c | 2950 | |
6d85ed05 AB |
2951 | /* |
2952 | * Don't try to rmdir the top/root subvolume dir. | |
2953 | */ | |
2954 | if (dir == BTRFS_FIRST_FREE_OBJECTID) | |
2955 | return 0; | |
2956 | ||
31db9f7c AB |
2957 | path = alloc_path_for_send(); |
2958 | if (!path) | |
2959 | return -ENOMEM; | |
2960 | ||
2961 | key.objectid = dir; | |
2962 | key.type = BTRFS_DIR_INDEX_KEY; | |
2963 | key.offset = 0; | |
0f96f517 | 2964 | |
0b3f407e | 2965 | odi = get_orphan_dir_info(sctx, dir, dir_gen); |
0f96f517 RK |
2966 | if (odi) |
2967 | key.offset = odi->last_dir_index_offset; | |
2968 | ||
dff6d0ad FDBM |
2969 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2970 | if (ret < 0) | |
2971 | goto out; | |
31db9f7c AB |
2972 | |
2973 | while (1) { | |
9dc44214 FM |
2974 | struct waiting_dir_move *dm; |
2975 | ||
dff6d0ad FDBM |
2976 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
2977 | ret = btrfs_next_leaf(root, path); | |
2978 | if (ret < 0) | |
2979 | goto out; | |
2980 | else if (ret > 0) | |
2981 | break; | |
2982 | continue; | |
31db9f7c | 2983 | } |
dff6d0ad FDBM |
2984 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2985 | path->slots[0]); | |
2986 | if (found_key.objectid != key.objectid || | |
2987 | found_key.type != key.type) | |
31db9f7c | 2988 | break; |
31db9f7c AB |
2989 | |
2990 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2991 | struct btrfs_dir_item); | |
2992 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
2993 | ||
9dc44214 FM |
2994 | dm = get_waiting_dir_move(sctx, loc.objectid); |
2995 | if (dm) { | |
0b3f407e | 2996 | odi = add_orphan_dir_info(sctx, dir, dir_gen); |
9dc44214 FM |
2997 | if (IS_ERR(odi)) { |
2998 | ret = PTR_ERR(odi); | |
2999 | goto out; | |
3000 | } | |
3001 | odi->gen = dir_gen; | |
0f96f517 | 3002 | odi->last_dir_index_offset = found_key.offset; |
9dc44214 | 3003 | dm->rmdir_ino = dir; |
0b3f407e | 3004 | dm->rmdir_gen = dir_gen; |
9dc44214 FM |
3005 | ret = 0; |
3006 | goto out; | |
3007 | } | |
3008 | ||
31db9f7c | 3009 | if (loc.objectid > send_progress) { |
0b3f407e | 3010 | odi = add_orphan_dir_info(sctx, dir, dir_gen); |
0f96f517 RK |
3011 | if (IS_ERR(odi)) { |
3012 | ret = PTR_ERR(odi); | |
3013 | goto out; | |
3014 | } | |
3015 | odi->gen = dir_gen; | |
3016 | odi->last_dir_index_offset = found_key.offset; | |
31db9f7c AB |
3017 | ret = 0; |
3018 | goto out; | |
3019 | } | |
3020 | ||
dff6d0ad | 3021 | path->slots[0]++; |
31db9f7c | 3022 | } |
0f96f517 | 3023 | free_orphan_dir_info(sctx, odi); |
31db9f7c AB |
3024 | |
3025 | ret = 1; | |
3026 | ||
3027 | out: | |
3028 | btrfs_free_path(path); | |
3029 | return ret; | |
3030 | } | |
3031 | ||
9f03740a FDBM |
3032 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) |
3033 | { | |
9dc44214 | 3034 | struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); |
9f03740a | 3035 | |
9dc44214 | 3036 | return entry != NULL; |
9f03740a FDBM |
3037 | } |
3038 | ||
8b191a68 | 3039 | static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) |
9f03740a FDBM |
3040 | { |
3041 | struct rb_node **p = &sctx->waiting_dir_moves.rb_node; | |
3042 | struct rb_node *parent = NULL; | |
3043 | struct waiting_dir_move *entry, *dm; | |
3044 | ||
e780b0d1 | 3045 | dm = kmalloc(sizeof(*dm), GFP_KERNEL); |
9f03740a FDBM |
3046 | if (!dm) |
3047 | return -ENOMEM; | |
3048 | dm->ino = ino; | |
9dc44214 | 3049 | dm->rmdir_ino = 0; |
0b3f407e | 3050 | dm->rmdir_gen = 0; |
8b191a68 | 3051 | dm->orphanized = orphanized; |
9f03740a FDBM |
3052 | |
3053 | while (*p) { | |
3054 | parent = *p; | |
3055 | entry = rb_entry(parent, struct waiting_dir_move, node); | |
3056 | if (ino < entry->ino) { | |
3057 | p = &(*p)->rb_left; | |
3058 | } else if (ino > entry->ino) { | |
3059 | p = &(*p)->rb_right; | |
3060 | } else { | |
3061 | kfree(dm); | |
3062 | return -EEXIST; | |
3063 | } | |
3064 | } | |
3065 | ||
3066 | rb_link_node(&dm->node, parent, p); | |
3067 | rb_insert_color(&dm->node, &sctx->waiting_dir_moves); | |
3068 | return 0; | |
3069 | } | |
3070 | ||
9dc44214 FM |
3071 | static struct waiting_dir_move * |
3072 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
9f03740a FDBM |
3073 | { |
3074 | struct rb_node *n = sctx->waiting_dir_moves.rb_node; | |
3075 | struct waiting_dir_move *entry; | |
3076 | ||
3077 | while (n) { | |
3078 | entry = rb_entry(n, struct waiting_dir_move, node); | |
9dc44214 | 3079 | if (ino < entry->ino) |
9f03740a | 3080 | n = n->rb_left; |
9dc44214 | 3081 | else if (ino > entry->ino) |
9f03740a | 3082 | n = n->rb_right; |
9dc44214 FM |
3083 | else |
3084 | return entry; | |
9f03740a | 3085 | } |
9dc44214 FM |
3086 | return NULL; |
3087 | } | |
3088 | ||
3089 | static void free_waiting_dir_move(struct send_ctx *sctx, | |
3090 | struct waiting_dir_move *dm) | |
3091 | { | |
3092 | if (!dm) | |
3093 | return; | |
3094 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
3095 | kfree(dm); | |
9f03740a FDBM |
3096 | } |
3097 | ||
bfa7e1f8 FM |
3098 | static int add_pending_dir_move(struct send_ctx *sctx, |
3099 | u64 ino, | |
3100 | u64 ino_gen, | |
f959492f FM |
3101 | u64 parent_ino, |
3102 | struct list_head *new_refs, | |
84471e24 FM |
3103 | struct list_head *deleted_refs, |
3104 | const bool is_orphan) | |
9f03740a FDBM |
3105 | { |
3106 | struct rb_node **p = &sctx->pending_dir_moves.rb_node; | |
3107 | struct rb_node *parent = NULL; | |
73b802f4 | 3108 | struct pending_dir_move *entry = NULL, *pm; |
9f03740a FDBM |
3109 | struct recorded_ref *cur; |
3110 | int exists = 0; | |
3111 | int ret; | |
3112 | ||
e780b0d1 | 3113 | pm = kmalloc(sizeof(*pm), GFP_KERNEL); |
9f03740a FDBM |
3114 | if (!pm) |
3115 | return -ENOMEM; | |
3116 | pm->parent_ino = parent_ino; | |
bfa7e1f8 FM |
3117 | pm->ino = ino; |
3118 | pm->gen = ino_gen; | |
9f03740a FDBM |
3119 | INIT_LIST_HEAD(&pm->list); |
3120 | INIT_LIST_HEAD(&pm->update_refs); | |
3121 | RB_CLEAR_NODE(&pm->node); | |
3122 | ||
3123 | while (*p) { | |
3124 | parent = *p; | |
3125 | entry = rb_entry(parent, struct pending_dir_move, node); | |
3126 | if (parent_ino < entry->parent_ino) { | |
3127 | p = &(*p)->rb_left; | |
3128 | } else if (parent_ino > entry->parent_ino) { | |
3129 | p = &(*p)->rb_right; | |
3130 | } else { | |
3131 | exists = 1; | |
3132 | break; | |
3133 | } | |
3134 | } | |
3135 | ||
f959492f | 3136 | list_for_each_entry(cur, deleted_refs, list) { |
9f03740a FDBM |
3137 | ret = dup_ref(cur, &pm->update_refs); |
3138 | if (ret < 0) | |
3139 | goto out; | |
3140 | } | |
f959492f | 3141 | list_for_each_entry(cur, new_refs, list) { |
9f03740a FDBM |
3142 | ret = dup_ref(cur, &pm->update_refs); |
3143 | if (ret < 0) | |
3144 | goto out; | |
3145 | } | |
3146 | ||
8b191a68 | 3147 | ret = add_waiting_dir_move(sctx, pm->ino, is_orphan); |
9f03740a FDBM |
3148 | if (ret) |
3149 | goto out; | |
3150 | ||
3151 | if (exists) { | |
3152 | list_add_tail(&pm->list, &entry->list); | |
3153 | } else { | |
3154 | rb_link_node(&pm->node, parent, p); | |
3155 | rb_insert_color(&pm->node, &sctx->pending_dir_moves); | |
3156 | } | |
3157 | ret = 0; | |
3158 | out: | |
3159 | if (ret) { | |
3160 | __free_recorded_refs(&pm->update_refs); | |
3161 | kfree(pm); | |
3162 | } | |
3163 | return ret; | |
3164 | } | |
3165 | ||
3166 | static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, | |
3167 | u64 parent_ino) | |
3168 | { | |
3169 | struct rb_node *n = sctx->pending_dir_moves.rb_node; | |
3170 | struct pending_dir_move *entry; | |
3171 | ||
3172 | while (n) { | |
3173 | entry = rb_entry(n, struct pending_dir_move, node); | |
3174 | if (parent_ino < entry->parent_ino) | |
3175 | n = n->rb_left; | |
3176 | else if (parent_ino > entry->parent_ino) | |
3177 | n = n->rb_right; | |
3178 | else | |
3179 | return entry; | |
3180 | } | |
3181 | return NULL; | |
3182 | } | |
3183 | ||
801bec36 RK |
3184 | static int path_loop(struct send_ctx *sctx, struct fs_path *name, |
3185 | u64 ino, u64 gen, u64 *ancestor_ino) | |
3186 | { | |
3187 | int ret = 0; | |
3188 | u64 parent_inode = 0; | |
3189 | u64 parent_gen = 0; | |
3190 | u64 start_ino = ino; | |
3191 | ||
3192 | *ancestor_ino = 0; | |
3193 | while (ino != BTRFS_FIRST_FREE_OBJECTID) { | |
3194 | fs_path_reset(name); | |
3195 | ||
0b3f407e | 3196 | if (is_waiting_for_rm(sctx, ino, gen)) |
801bec36 RK |
3197 | break; |
3198 | if (is_waiting_for_move(sctx, ino)) { | |
3199 | if (*ancestor_ino == 0) | |
3200 | *ancestor_ino = ino; | |
3201 | ret = get_first_ref(sctx->parent_root, ino, | |
3202 | &parent_inode, &parent_gen, name); | |
3203 | } else { | |
3204 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
3205 | &parent_inode, | |
3206 | &parent_gen, name); | |
3207 | if (ret > 0) { | |
3208 | ret = 0; | |
3209 | break; | |
3210 | } | |
3211 | } | |
3212 | if (ret < 0) | |
3213 | break; | |
3214 | if (parent_inode == start_ino) { | |
3215 | ret = 1; | |
3216 | if (*ancestor_ino == 0) | |
3217 | *ancestor_ino = ino; | |
3218 | break; | |
3219 | } | |
3220 | ino = parent_inode; | |
3221 | gen = parent_gen; | |
3222 | } | |
3223 | return ret; | |
3224 | } | |
3225 | ||
9f03740a FDBM |
3226 | static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) |
3227 | { | |
3228 | struct fs_path *from_path = NULL; | |
3229 | struct fs_path *to_path = NULL; | |
2b863a13 | 3230 | struct fs_path *name = NULL; |
9f03740a FDBM |
3231 | u64 orig_progress = sctx->send_progress; |
3232 | struct recorded_ref *cur; | |
2b863a13 | 3233 | u64 parent_ino, parent_gen; |
9dc44214 FM |
3234 | struct waiting_dir_move *dm = NULL; |
3235 | u64 rmdir_ino = 0; | |
0b3f407e | 3236 | u64 rmdir_gen; |
801bec36 RK |
3237 | u64 ancestor; |
3238 | bool is_orphan; | |
9f03740a FDBM |
3239 | int ret; |
3240 | ||
2b863a13 | 3241 | name = fs_path_alloc(); |
9f03740a | 3242 | from_path = fs_path_alloc(); |
2b863a13 FM |
3243 | if (!name || !from_path) { |
3244 | ret = -ENOMEM; | |
3245 | goto out; | |
3246 | } | |
9f03740a | 3247 | |
9dc44214 FM |
3248 | dm = get_waiting_dir_move(sctx, pm->ino); |
3249 | ASSERT(dm); | |
3250 | rmdir_ino = dm->rmdir_ino; | |
0b3f407e | 3251 | rmdir_gen = dm->rmdir_gen; |
801bec36 | 3252 | is_orphan = dm->orphanized; |
9dc44214 | 3253 | free_waiting_dir_move(sctx, dm); |
2b863a13 | 3254 | |
801bec36 | 3255 | if (is_orphan) { |
84471e24 FM |
3256 | ret = gen_unique_name(sctx, pm->ino, |
3257 | pm->gen, from_path); | |
3258 | } else { | |
3259 | ret = get_first_ref(sctx->parent_root, pm->ino, | |
3260 | &parent_ino, &parent_gen, name); | |
3261 | if (ret < 0) | |
3262 | goto out; | |
3263 | ret = get_cur_path(sctx, parent_ino, parent_gen, | |
3264 | from_path); | |
3265 | if (ret < 0) | |
3266 | goto out; | |
3267 | ret = fs_path_add_path(from_path, name); | |
3268 | } | |
c992ec94 FM |
3269 | if (ret < 0) |
3270 | goto out; | |
2b863a13 | 3271 | |
f959492f | 3272 | sctx->send_progress = sctx->cur_ino + 1; |
801bec36 | 3273 | ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor); |
7969e77a FM |
3274 | if (ret < 0) |
3275 | goto out; | |
801bec36 RK |
3276 | if (ret) { |
3277 | LIST_HEAD(deleted_refs); | |
3278 | ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID); | |
3279 | ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor, | |
3280 | &pm->update_refs, &deleted_refs, | |
3281 | is_orphan); | |
3282 | if (ret < 0) | |
3283 | goto out; | |
3284 | if (rmdir_ino) { | |
3285 | dm = get_waiting_dir_move(sctx, pm->ino); | |
3286 | ASSERT(dm); | |
3287 | dm->rmdir_ino = rmdir_ino; | |
0b3f407e | 3288 | dm->rmdir_gen = rmdir_gen; |
801bec36 RK |
3289 | } |
3290 | goto out; | |
3291 | } | |
c992ec94 FM |
3292 | fs_path_reset(name); |
3293 | to_path = name; | |
2b863a13 | 3294 | name = NULL; |
9f03740a FDBM |
3295 | ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); |
3296 | if (ret < 0) | |
3297 | goto out; | |
3298 | ||
3299 | ret = send_rename(sctx, from_path, to_path); | |
3300 | if (ret < 0) | |
3301 | goto out; | |
3302 | ||
9dc44214 FM |
3303 | if (rmdir_ino) { |
3304 | struct orphan_dir_info *odi; | |
0f96f517 | 3305 | u64 gen; |
9dc44214 | 3306 | |
0b3f407e | 3307 | odi = get_orphan_dir_info(sctx, rmdir_ino, rmdir_gen); |
9dc44214 FM |
3308 | if (!odi) { |
3309 | /* already deleted */ | |
3310 | goto finish; | |
3311 | } | |
0f96f517 RK |
3312 | gen = odi->gen; |
3313 | ||
3314 | ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino); | |
9dc44214 FM |
3315 | if (ret < 0) |
3316 | goto out; | |
3317 | if (!ret) | |
3318 | goto finish; | |
3319 | ||
3320 | name = fs_path_alloc(); | |
3321 | if (!name) { | |
3322 | ret = -ENOMEM; | |
3323 | goto out; | |
3324 | } | |
0f96f517 | 3325 | ret = get_cur_path(sctx, rmdir_ino, gen, name); |
9dc44214 FM |
3326 | if (ret < 0) |
3327 | goto out; | |
3328 | ret = send_rmdir(sctx, name); | |
3329 | if (ret < 0) | |
3330 | goto out; | |
9dc44214 FM |
3331 | } |
3332 | ||
3333 | finish: | |
9f03740a FDBM |
3334 | ret = send_utimes(sctx, pm->ino, pm->gen); |
3335 | if (ret < 0) | |
3336 | goto out; | |
3337 | ||
3338 | /* | |
3339 | * After rename/move, need to update the utimes of both new parent(s) | |
3340 | * and old parent(s). | |
3341 | */ | |
3342 | list_for_each_entry(cur, &pm->update_refs, list) { | |
764433a1 RK |
3343 | /* |
3344 | * The parent inode might have been deleted in the send snapshot | |
3345 | */ | |
3346 | ret = get_inode_info(sctx->send_root, cur->dir, NULL, | |
3347 | NULL, NULL, NULL, NULL, NULL); | |
3348 | if (ret == -ENOENT) { | |
3349 | ret = 0; | |
9dc44214 | 3350 | continue; |
764433a1 RK |
3351 | } |
3352 | if (ret < 0) | |
3353 | goto out; | |
3354 | ||
9f03740a FDBM |
3355 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
3356 | if (ret < 0) | |
3357 | goto out; | |
3358 | } | |
3359 | ||
3360 | out: | |
2b863a13 | 3361 | fs_path_free(name); |
9f03740a FDBM |
3362 | fs_path_free(from_path); |
3363 | fs_path_free(to_path); | |
3364 | sctx->send_progress = orig_progress; | |
3365 | ||
3366 | return ret; | |
3367 | } | |
3368 | ||
3369 | static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) | |
3370 | { | |
3371 | if (!list_empty(&m->list)) | |
3372 | list_del(&m->list); | |
3373 | if (!RB_EMPTY_NODE(&m->node)) | |
3374 | rb_erase(&m->node, &sctx->pending_dir_moves); | |
3375 | __free_recorded_refs(&m->update_refs); | |
3376 | kfree(m); | |
3377 | } | |
3378 | ||
a4390aee RK |
3379 | static void tail_append_pending_moves(struct send_ctx *sctx, |
3380 | struct pending_dir_move *moves, | |
9f03740a FDBM |
3381 | struct list_head *stack) |
3382 | { | |
3383 | if (list_empty(&moves->list)) { | |
3384 | list_add_tail(&moves->list, stack); | |
3385 | } else { | |
3386 | LIST_HEAD(list); | |
3387 | list_splice_init(&moves->list, &list); | |
3388 | list_add_tail(&moves->list, stack); | |
3389 | list_splice_tail(&list, stack); | |
3390 | } | |
a4390aee RK |
3391 | if (!RB_EMPTY_NODE(&moves->node)) { |
3392 | rb_erase(&moves->node, &sctx->pending_dir_moves); | |
3393 | RB_CLEAR_NODE(&moves->node); | |
3394 | } | |
9f03740a FDBM |
3395 | } |
3396 | ||
3397 | static int apply_children_dir_moves(struct send_ctx *sctx) | |
3398 | { | |
3399 | struct pending_dir_move *pm; | |
3400 | struct list_head stack; | |
3401 | u64 parent_ino = sctx->cur_ino; | |
3402 | int ret = 0; | |
3403 | ||
3404 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3405 | if (!pm) | |
3406 | return 0; | |
3407 | ||
3408 | INIT_LIST_HEAD(&stack); | |
a4390aee | 3409 | tail_append_pending_moves(sctx, pm, &stack); |
9f03740a FDBM |
3410 | |
3411 | while (!list_empty(&stack)) { | |
3412 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3413 | parent_ino = pm->ino; | |
3414 | ret = apply_dir_move(sctx, pm); | |
3415 | free_pending_move(sctx, pm); | |
3416 | if (ret) | |
3417 | goto out; | |
3418 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3419 | if (pm) | |
a4390aee | 3420 | tail_append_pending_moves(sctx, pm, &stack); |
9f03740a FDBM |
3421 | } |
3422 | return 0; | |
3423 | ||
3424 | out: | |
3425 | while (!list_empty(&stack)) { | |
3426 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3427 | free_pending_move(sctx, pm); | |
3428 | } | |
3429 | return ret; | |
3430 | } | |
3431 | ||
84471e24 FM |
3432 | /* |
3433 | * We might need to delay a directory rename even when no ancestor directory | |
3434 | * (in the send root) with a higher inode number than ours (sctx->cur_ino) was | |
3435 | * renamed. This happens when we rename a directory to the old name (the name | |
3436 | * in the parent root) of some other unrelated directory that got its rename | |
3437 | * delayed due to some ancestor with higher number that got renamed. | |
3438 | * | |
3439 | * Example: | |
3440 | * | |
3441 | * Parent snapshot: | |
3442 | * . (ino 256) | |
3443 | * |---- a/ (ino 257) | |
3444 | * | |---- file (ino 260) | |
3445 | * | | |
3446 | * |---- b/ (ino 258) | |
3447 | * |---- c/ (ino 259) | |
3448 | * | |
3449 | * Send snapshot: | |
3450 | * . (ino 256) | |
3451 | * |---- a/ (ino 258) | |
3452 | * |---- x/ (ino 259) | |
3453 | * |---- y/ (ino 257) | |
3454 | * |----- file (ino 260) | |
3455 | * | |
3456 | * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257 | |
3457 | * from 'a' to 'x/y' happening first, which in turn depends on the rename of | |
3458 | * inode 259 from 'c' to 'x'. So the order of rename commands the send stream | |
3459 | * must issue is: | |
3460 | * | |
3461 | * 1 - rename 259 from 'c' to 'x' | |
3462 | * 2 - rename 257 from 'a' to 'x/y' | |
3463 | * 3 - rename 258 from 'b' to 'a' | |
3464 | * | |
3465 | * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can | |
3466 | * be done right away and < 0 on error. | |
3467 | */ | |
3468 | static int wait_for_dest_dir_move(struct send_ctx *sctx, | |
3469 | struct recorded_ref *parent_ref, | |
3470 | const bool is_orphan) | |
3471 | { | |
2ff7e61e | 3472 | struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info; |
84471e24 FM |
3473 | struct btrfs_path *path; |
3474 | struct btrfs_key key; | |
3475 | struct btrfs_key di_key; | |
3476 | struct btrfs_dir_item *di; | |
3477 | u64 left_gen; | |
3478 | u64 right_gen; | |
3479 | int ret = 0; | |
801bec36 | 3480 | struct waiting_dir_move *wdm; |
84471e24 FM |
3481 | |
3482 | if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) | |
3483 | return 0; | |
3484 | ||
3485 | path = alloc_path_for_send(); | |
3486 | if (!path) | |
3487 | return -ENOMEM; | |
3488 | ||
3489 | key.objectid = parent_ref->dir; | |
3490 | key.type = BTRFS_DIR_ITEM_KEY; | |
3491 | key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); | |
3492 | ||
3493 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
3494 | if (ret < 0) { | |
3495 | goto out; | |
3496 | } else if (ret > 0) { | |
3497 | ret = 0; | |
3498 | goto out; | |
3499 | } | |
3500 | ||
2ff7e61e JM |
3501 | di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name, |
3502 | parent_ref->name_len); | |
84471e24 FM |
3503 | if (!di) { |
3504 | ret = 0; | |
3505 | goto out; | |
3506 | } | |
3507 | /* | |
3508 | * di_key.objectid has the number of the inode that has a dentry in the | |
3509 | * parent directory with the same name that sctx->cur_ino is being | |
3510 | * renamed to. We need to check if that inode is in the send root as | |
3511 | * well and if it is currently marked as an inode with a pending rename, | |
3512 | * if it is, we need to delay the rename of sctx->cur_ino as well, so | |
3513 | * that it happens after that other inode is renamed. | |
3514 | */ | |
3515 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); | |
3516 | if (di_key.type != BTRFS_INODE_ITEM_KEY) { | |
3517 | ret = 0; | |
3518 | goto out; | |
3519 | } | |
3520 | ||
3521 | ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL, | |
3522 | &left_gen, NULL, NULL, NULL, NULL); | |
3523 | if (ret < 0) | |
3524 | goto out; | |
3525 | ret = get_inode_info(sctx->send_root, di_key.objectid, NULL, | |
3526 | &right_gen, NULL, NULL, NULL, NULL); | |
3527 | if (ret < 0) { | |
3528 | if (ret == -ENOENT) | |
3529 | ret = 0; | |
3530 | goto out; | |
3531 | } | |
3532 | ||
3533 | /* Different inode, no need to delay the rename of sctx->cur_ino */ | |
3534 | if (right_gen != left_gen) { | |
3535 | ret = 0; | |
3536 | goto out; | |
3537 | } | |
3538 | ||
801bec36 RK |
3539 | wdm = get_waiting_dir_move(sctx, di_key.objectid); |
3540 | if (wdm && !wdm->orphanized) { | |
84471e24 FM |
3541 | ret = add_pending_dir_move(sctx, |
3542 | sctx->cur_ino, | |
3543 | sctx->cur_inode_gen, | |
3544 | di_key.objectid, | |
3545 | &sctx->new_refs, | |
3546 | &sctx->deleted_refs, | |
3547 | is_orphan); | |
3548 | if (!ret) | |
3549 | ret = 1; | |
3550 | } | |
3551 | out: | |
3552 | btrfs_free_path(path); | |
3553 | return ret; | |
3554 | } | |
3555 | ||
80aa6027 | 3556 | /* |
ea37d599 FM |
3557 | * Check if inode ino2, or any of its ancestors, is inode ino1. |
3558 | * Return 1 if true, 0 if false and < 0 on error. | |
3559 | */ | |
3560 | static int check_ino_in_path(struct btrfs_root *root, | |
3561 | const u64 ino1, | |
3562 | const u64 ino1_gen, | |
3563 | const u64 ino2, | |
3564 | const u64 ino2_gen, | |
3565 | struct fs_path *fs_path) | |
3566 | { | |
3567 | u64 ino = ino2; | |
3568 | ||
3569 | if (ino1 == ino2) | |
3570 | return ino1_gen == ino2_gen; | |
3571 | ||
3572 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { | |
3573 | u64 parent; | |
3574 | u64 parent_gen; | |
3575 | int ret; | |
3576 | ||
3577 | fs_path_reset(fs_path); | |
3578 | ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path); | |
3579 | if (ret < 0) | |
3580 | return ret; | |
3581 | if (parent == ino1) | |
3582 | return parent_gen == ino1_gen; | |
3583 | ino = parent; | |
3584 | } | |
3585 | return 0; | |
3586 | } | |
3587 | ||
3588 | /* | |
3589 | * Check if ino ino1 is an ancestor of inode ino2 in the given root for any | |
3590 | * possible path (in case ino2 is not a directory and has multiple hard links). | |
80aa6027 FM |
3591 | * Return 1 if true, 0 if false and < 0 on error. |
3592 | */ | |
3593 | static int is_ancestor(struct btrfs_root *root, | |
3594 | const u64 ino1, | |
3595 | const u64 ino1_gen, | |
3596 | const u64 ino2, | |
3597 | struct fs_path *fs_path) | |
3598 | { | |
ea37d599 | 3599 | bool free_fs_path = false; |
72c3668f | 3600 | int ret = 0; |
ea37d599 FM |
3601 | struct btrfs_path *path = NULL; |
3602 | struct btrfs_key key; | |
72c3668f FM |
3603 | |
3604 | if (!fs_path) { | |
3605 | fs_path = fs_path_alloc(); | |
3606 | if (!fs_path) | |
3607 | return -ENOMEM; | |
ea37d599 | 3608 | free_fs_path = true; |
72c3668f | 3609 | } |
80aa6027 | 3610 | |
ea37d599 FM |
3611 | path = alloc_path_for_send(); |
3612 | if (!path) { | |
3613 | ret = -ENOMEM; | |
3614 | goto out; | |
3615 | } | |
80aa6027 | 3616 | |
ea37d599 FM |
3617 | key.objectid = ino2; |
3618 | key.type = BTRFS_INODE_REF_KEY; | |
3619 | key.offset = 0; | |
3620 | ||
3621 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3622 | if (ret < 0) | |
3623 | goto out; | |
3624 | ||
3625 | while (true) { | |
3626 | struct extent_buffer *leaf = path->nodes[0]; | |
3627 | int slot = path->slots[0]; | |
3628 | u32 cur_offset = 0; | |
3629 | u32 item_size; | |
3630 | ||
3631 | if (slot >= btrfs_header_nritems(leaf)) { | |
3632 | ret = btrfs_next_leaf(root, path); | |
3633 | if (ret < 0) | |
3634 | goto out; | |
3635 | if (ret > 0) | |
3636 | break; | |
3637 | continue; | |
72c3668f | 3638 | } |
ea37d599 FM |
3639 | |
3640 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
3641 | if (key.objectid != ino2) | |
3642 | break; | |
3643 | if (key.type != BTRFS_INODE_REF_KEY && | |
3644 | key.type != BTRFS_INODE_EXTREF_KEY) | |
3645 | break; | |
3646 | ||
3647 | item_size = btrfs_item_size_nr(leaf, slot); | |
3648 | while (cur_offset < item_size) { | |
3649 | u64 parent; | |
3650 | u64 parent_gen; | |
3651 | ||
3652 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
3653 | unsigned long ptr; | |
3654 | struct btrfs_inode_extref *extref; | |
3655 | ||
3656 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3657 | extref = (struct btrfs_inode_extref *) | |
3658 | (ptr + cur_offset); | |
3659 | parent = btrfs_inode_extref_parent(leaf, | |
3660 | extref); | |
3661 | cur_offset += sizeof(*extref); | |
3662 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
3663 | extref); | |
3664 | } else { | |
3665 | parent = key.offset; | |
3666 | cur_offset = item_size; | |
3667 | } | |
3668 | ||
3669 | ret = get_inode_info(root, parent, NULL, &parent_gen, | |
3670 | NULL, NULL, NULL, NULL); | |
3671 | if (ret < 0) | |
3672 | goto out; | |
3673 | ret = check_ino_in_path(root, ino1, ino1_gen, | |
3674 | parent, parent_gen, fs_path); | |
3675 | if (ret) | |
3676 | goto out; | |
80aa6027 | 3677 | } |
ea37d599 | 3678 | path->slots[0]++; |
80aa6027 | 3679 | } |
ea37d599 | 3680 | ret = 0; |
72c3668f | 3681 | out: |
ea37d599 FM |
3682 | btrfs_free_path(path); |
3683 | if (free_fs_path) | |
72c3668f FM |
3684 | fs_path_free(fs_path); |
3685 | return ret; | |
80aa6027 FM |
3686 | } |
3687 | ||
9f03740a | 3688 | static int wait_for_parent_move(struct send_ctx *sctx, |
8b191a68 FM |
3689 | struct recorded_ref *parent_ref, |
3690 | const bool is_orphan) | |
9f03740a | 3691 | { |
f959492f | 3692 | int ret = 0; |
9f03740a | 3693 | u64 ino = parent_ref->dir; |
fe9c798d | 3694 | u64 ino_gen = parent_ref->dir_gen; |
9f03740a | 3695 | u64 parent_ino_before, parent_ino_after; |
9f03740a FDBM |
3696 | struct fs_path *path_before = NULL; |
3697 | struct fs_path *path_after = NULL; | |
3698 | int len1, len2; | |
9f03740a FDBM |
3699 | |
3700 | path_after = fs_path_alloc(); | |
f959492f FM |
3701 | path_before = fs_path_alloc(); |
3702 | if (!path_after || !path_before) { | |
9f03740a FDBM |
3703 | ret = -ENOMEM; |
3704 | goto out; | |
3705 | } | |
3706 | ||
bfa7e1f8 | 3707 | /* |
f959492f FM |
3708 | * Our current directory inode may not yet be renamed/moved because some |
3709 | * ancestor (immediate or not) has to be renamed/moved first. So find if | |
3710 | * such ancestor exists and make sure our own rename/move happens after | |
80aa6027 FM |
3711 | * that ancestor is processed to avoid path build infinite loops (done |
3712 | * at get_cur_path()). | |
bfa7e1f8 | 3713 | */ |
f959492f | 3714 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { |
fe9c798d FM |
3715 | u64 parent_ino_after_gen; |
3716 | ||
f959492f | 3717 | if (is_waiting_for_move(sctx, ino)) { |
80aa6027 FM |
3718 | /* |
3719 | * If the current inode is an ancestor of ino in the | |
3720 | * parent root, we need to delay the rename of the | |
3721 | * current inode, otherwise don't delayed the rename | |
3722 | * because we can end up with a circular dependency | |
3723 | * of renames, resulting in some directories never | |
3724 | * getting the respective rename operations issued in | |
3725 | * the send stream or getting into infinite path build | |
3726 | * loops. | |
3727 | */ | |
3728 | ret = is_ancestor(sctx->parent_root, | |
3729 | sctx->cur_ino, sctx->cur_inode_gen, | |
3730 | ino, path_before); | |
4122ea64 FM |
3731 | if (ret) |
3732 | break; | |
f959492f | 3733 | } |
bfa7e1f8 FM |
3734 | |
3735 | fs_path_reset(path_before); | |
3736 | fs_path_reset(path_after); | |
3737 | ||
3738 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
fe9c798d | 3739 | &parent_ino_after_gen, path_after); |
bfa7e1f8 FM |
3740 | if (ret < 0) |
3741 | goto out; | |
3742 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
3743 | NULL, path_before); | |
f959492f | 3744 | if (ret < 0 && ret != -ENOENT) { |
bfa7e1f8 | 3745 | goto out; |
f959492f | 3746 | } else if (ret == -ENOENT) { |
bf8e8ca6 | 3747 | ret = 0; |
f959492f | 3748 | break; |
bfa7e1f8 FM |
3749 | } |
3750 | ||
3751 | len1 = fs_path_len(path_before); | |
3752 | len2 = fs_path_len(path_after); | |
f959492f FM |
3753 | if (ino > sctx->cur_ino && |
3754 | (parent_ino_before != parent_ino_after || len1 != len2 || | |
3755 | memcmp(path_before->start, path_after->start, len1))) { | |
fe9c798d FM |
3756 | u64 parent_ino_gen; |
3757 | ||
3758 | ret = get_inode_info(sctx->parent_root, ino, NULL, | |
3759 | &parent_ino_gen, NULL, NULL, NULL, | |
3760 | NULL); | |
3761 | if (ret < 0) | |
3762 | goto out; | |
3763 | if (ino_gen == parent_ino_gen) { | |
3764 | ret = 1; | |
3765 | break; | |
3766 | } | |
bfa7e1f8 | 3767 | } |
bfa7e1f8 | 3768 | ino = parent_ino_after; |
fe9c798d | 3769 | ino_gen = parent_ino_after_gen; |
bfa7e1f8 FM |
3770 | } |
3771 | ||
9f03740a FDBM |
3772 | out: |
3773 | fs_path_free(path_before); | |
3774 | fs_path_free(path_after); | |
3775 | ||
f959492f FM |
3776 | if (ret == 1) { |
3777 | ret = add_pending_dir_move(sctx, | |
3778 | sctx->cur_ino, | |
3779 | sctx->cur_inode_gen, | |
3780 | ino, | |
3781 | &sctx->new_refs, | |
84471e24 | 3782 | &sctx->deleted_refs, |
8b191a68 | 3783 | is_orphan); |
f959492f FM |
3784 | if (!ret) |
3785 | ret = 1; | |
3786 | } | |
3787 | ||
9f03740a FDBM |
3788 | return ret; |
3789 | } | |
3790 | ||
f5962781 FM |
3791 | static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) |
3792 | { | |
3793 | int ret; | |
3794 | struct fs_path *new_path; | |
3795 | ||
3796 | /* | |
3797 | * Our reference's name member points to its full_path member string, so | |
3798 | * we use here a new path. | |
3799 | */ | |
3800 | new_path = fs_path_alloc(); | |
3801 | if (!new_path) | |
3802 | return -ENOMEM; | |
3803 | ||
3804 | ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path); | |
3805 | if (ret < 0) { | |
3806 | fs_path_free(new_path); | |
3807 | return ret; | |
3808 | } | |
3809 | ret = fs_path_add(new_path, ref->name, ref->name_len); | |
3810 | if (ret < 0) { | |
3811 | fs_path_free(new_path); | |
3812 | return ret; | |
3813 | } | |
3814 | ||
3815 | fs_path_free(ref->full_path); | |
3816 | set_ref_path(ref, new_path); | |
3817 | ||
3818 | return 0; | |
3819 | } | |
3820 | ||
9c2b4e03 FM |
3821 | /* |
3822 | * When processing the new references for an inode we may orphanize an existing | |
3823 | * directory inode because its old name conflicts with one of the new references | |
3824 | * of the current inode. Later, when processing another new reference of our | |
3825 | * inode, we might need to orphanize another inode, but the path we have in the | |
3826 | * reference reflects the pre-orphanization name of the directory we previously | |
3827 | * orphanized. For example: | |
3828 | * | |
3829 | * parent snapshot looks like: | |
3830 | * | |
3831 | * . (ino 256) | |
3832 | * |----- f1 (ino 257) | |
3833 | * |----- f2 (ino 258) | |
3834 | * |----- d1/ (ino 259) | |
3835 | * |----- d2/ (ino 260) | |
3836 | * | |
3837 | * send snapshot looks like: | |
3838 | * | |
3839 | * . (ino 256) | |
3840 | * |----- d1 (ino 258) | |
3841 | * |----- f2/ (ino 259) | |
3842 | * |----- f2_link/ (ino 260) | |
3843 | * | |----- f1 (ino 257) | |
3844 | * | | |
3845 | * |----- d2 (ino 258) | |
3846 | * | |
3847 | * When processing inode 257 we compute the name for inode 259 as "d1", and we | |
3848 | * cache it in the name cache. Later when we start processing inode 258, when | |
3849 | * collecting all its new references we set a full path of "d1/d2" for its new | |
3850 | * reference with name "d2". When we start processing the new references we | |
3851 | * start by processing the new reference with name "d1", and this results in | |
3852 | * orphanizing inode 259, since its old reference causes a conflict. Then we | |
3853 | * move on the next new reference, with name "d2", and we find out we must | |
3854 | * orphanize inode 260, as its old reference conflicts with ours - but for the | |
3855 | * orphanization we use a source path corresponding to the path we stored in the | |
3856 | * new reference, which is "d1/d2" and not "o259-6-0/d2" - this makes the | |
3857 | * receiver fail since the path component "d1/" no longer exists, it was renamed | |
3858 | * to "o259-6-0/" when processing the previous new reference. So in this case we | |
3859 | * must recompute the path in the new reference and use it for the new | |
3860 | * orphanization operation. | |
3861 | */ | |
3862 | static int refresh_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) | |
3863 | { | |
3864 | char *name; | |
3865 | int ret; | |
3866 | ||
3867 | name = kmemdup(ref->name, ref->name_len, GFP_KERNEL); | |
3868 | if (!name) | |
3869 | return -ENOMEM; | |
3870 | ||
3871 | fs_path_reset(ref->full_path); | |
3872 | ret = get_cur_path(sctx, ref->dir, ref->dir_gen, ref->full_path); | |
3873 | if (ret < 0) | |
3874 | goto out; | |
3875 | ||
3876 | ret = fs_path_add(ref->full_path, name, ref->name_len); | |
3877 | if (ret < 0) | |
3878 | goto out; | |
3879 | ||
3880 | /* Update the reference's base name pointer. */ | |
3881 | set_ref_path(ref, ref->full_path); | |
3882 | out: | |
3883 | kfree(name); | |
3884 | return ret; | |
3885 | } | |
3886 | ||
31db9f7c AB |
3887 | /* |
3888 | * This does all the move/link/unlink/rmdir magic. | |
3889 | */ | |
9f03740a | 3890 | static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) |
31db9f7c | 3891 | { |
04ab956e | 3892 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
3893 | int ret = 0; |
3894 | struct recorded_ref *cur; | |
1f4692da | 3895 | struct recorded_ref *cur2; |
ba5e8f2e | 3896 | struct list_head check_dirs; |
31db9f7c | 3897 | struct fs_path *valid_path = NULL; |
b24baf69 | 3898 | u64 ow_inode = 0; |
31db9f7c | 3899 | u64 ow_gen; |
f5962781 | 3900 | u64 ow_mode; |
31db9f7c AB |
3901 | int did_overwrite = 0; |
3902 | int is_orphan = 0; | |
29d6d30f | 3903 | u64 last_dir_ino_rm = 0; |
84471e24 | 3904 | bool can_rename = true; |
f5962781 | 3905 | bool orphanized_dir = false; |
fdb13889 | 3906 | bool orphanized_ancestor = false; |
31db9f7c | 3907 | |
04ab956e | 3908 | btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino); |
31db9f7c | 3909 | |
6d85ed05 AB |
3910 | /* |
3911 | * This should never happen as the root dir always has the same ref | |
3912 | * which is always '..' | |
3913 | */ | |
3914 | BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); | |
ba5e8f2e | 3915 | INIT_LIST_HEAD(&check_dirs); |
6d85ed05 | 3916 | |
924794c9 | 3917 | valid_path = fs_path_alloc(); |
31db9f7c AB |
3918 | if (!valid_path) { |
3919 | ret = -ENOMEM; | |
3920 | goto out; | |
3921 | } | |
3922 | ||
31db9f7c AB |
3923 | /* |
3924 | * First, check if the first ref of the current inode was overwritten | |
3925 | * before. If yes, we know that the current inode was already orphanized | |
3926 | * and thus use the orphan name. If not, we can use get_cur_path to | |
3927 | * get the path of the first ref as it would like while receiving at | |
3928 | * this point in time. | |
3929 | * New inodes are always orphan at the beginning, so force to use the | |
3930 | * orphan name in this case. | |
3931 | * The first ref is stored in valid_path and will be updated if it | |
3932 | * gets moved around. | |
3933 | */ | |
3934 | if (!sctx->cur_inode_new) { | |
3935 | ret = did_overwrite_first_ref(sctx, sctx->cur_ino, | |
3936 | sctx->cur_inode_gen); | |
3937 | if (ret < 0) | |
3938 | goto out; | |
3939 | if (ret) | |
3940 | did_overwrite = 1; | |
3941 | } | |
3942 | if (sctx->cur_inode_new || did_overwrite) { | |
3943 | ret = gen_unique_name(sctx, sctx->cur_ino, | |
3944 | sctx->cur_inode_gen, valid_path); | |
3945 | if (ret < 0) | |
3946 | goto out; | |
3947 | is_orphan = 1; | |
3948 | } else { | |
3949 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
3950 | valid_path); | |
3951 | if (ret < 0) | |
3952 | goto out; | |
3953 | } | |
3954 | ||
98272bb7 FM |
3955 | /* |
3956 | * Before doing any rename and link operations, do a first pass on the | |
3957 | * new references to orphanize any unprocessed inodes that may have a | |
3958 | * reference that conflicts with one of the new references of the current | |
3959 | * inode. This needs to happen first because a new reference may conflict | |
3960 | * with the old reference of a parent directory, so we must make sure | |
3961 | * that the path used for link and rename commands don't use an | |
3962 | * orphanized name when an ancestor was not yet orphanized. | |
3963 | * | |
3964 | * Example: | |
3965 | * | |
3966 | * Parent snapshot: | |
3967 | * | |
3968 | * . (ino 256) | |
3969 | * |----- testdir/ (ino 259) | |
3970 | * | |----- a (ino 257) | |
3971 | * | | |
3972 | * |----- b (ino 258) | |
3973 | * | |
3974 | * Send snapshot: | |
3975 | * | |
3976 | * . (ino 256) | |
3977 | * |----- testdir_2/ (ino 259) | |
3978 | * | |----- a (ino 260) | |
3979 | * | | |
3980 | * |----- testdir (ino 257) | |
3981 | * |----- b (ino 257) | |
3982 | * |----- b2 (ino 258) | |
3983 | * | |
3984 | * Processing the new reference for inode 257 with name "b" may happen | |
3985 | * before processing the new reference with name "testdir". If so, we | |
3986 | * must make sure that by the time we send a link command to create the | |
3987 | * hard link "b", inode 259 was already orphanized, since the generated | |
3988 | * path in "valid_path" already contains the orphanized name for 259. | |
3989 | * We are processing inode 257, so only later when processing 259 we do | |
3990 | * the rename operation to change its temporary (orphanized) name to | |
3991 | * "testdir_2". | |
3992 | */ | |
31db9f7c | 3993 | list_for_each_entry(cur, &sctx->new_refs, list) { |
1f4692da AB |
3994 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); |
3995 | if (ret < 0) | |
3996 | goto out; | |
98272bb7 FM |
3997 | if (ret == inode_state_will_create) |
3998 | continue; | |
1f4692da | 3999 | |
31db9f7c | 4000 | /* |
98272bb7 FM |
4001 | * Check if this new ref would overwrite the first ref of another |
4002 | * unprocessed inode. If yes, orphanize the overwritten inode. | |
4003 | * If we find an overwritten ref that is not the first ref, | |
4004 | * simply unlink it. | |
31db9f7c AB |
4005 | */ |
4006 | ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
4007 | cur->name, cur->name_len, | |
f5962781 | 4008 | &ow_inode, &ow_gen, &ow_mode); |
31db9f7c AB |
4009 | if (ret < 0) |
4010 | goto out; | |
4011 | if (ret) { | |
924794c9 TI |
4012 | ret = is_first_ref(sctx->parent_root, |
4013 | ow_inode, cur->dir, cur->name, | |
4014 | cur->name_len); | |
31db9f7c AB |
4015 | if (ret < 0) |
4016 | goto out; | |
4017 | if (ret) { | |
8996a48c | 4018 | struct name_cache_entry *nce; |
801bec36 | 4019 | struct waiting_dir_move *wdm; |
8996a48c | 4020 | |
9c2b4e03 FM |
4021 | if (orphanized_dir) { |
4022 | ret = refresh_ref_path(sctx, cur); | |
4023 | if (ret < 0) | |
4024 | goto out; | |
4025 | } | |
4026 | ||
31db9f7c AB |
4027 | ret = orphanize_inode(sctx, ow_inode, ow_gen, |
4028 | cur->full_path); | |
4029 | if (ret < 0) | |
4030 | goto out; | |
f5962781 FM |
4031 | if (S_ISDIR(ow_mode)) |
4032 | orphanized_dir = true; | |
801bec36 RK |
4033 | |
4034 | /* | |
4035 | * If ow_inode has its rename operation delayed | |
4036 | * make sure that its orphanized name is used in | |
4037 | * the source path when performing its rename | |
4038 | * operation. | |
4039 | */ | |
4040 | if (is_waiting_for_move(sctx, ow_inode)) { | |
4041 | wdm = get_waiting_dir_move(sctx, | |
4042 | ow_inode); | |
4043 | ASSERT(wdm); | |
4044 | wdm->orphanized = true; | |
4045 | } | |
4046 | ||
8996a48c FM |
4047 | /* |
4048 | * Make sure we clear our orphanized inode's | |
4049 | * name from the name cache. This is because the | |
4050 | * inode ow_inode might be an ancestor of some | |
4051 | * other inode that will be orphanized as well | |
4052 | * later and has an inode number greater than | |
4053 | * sctx->send_progress. We need to prevent | |
4054 | * future name lookups from using the old name | |
4055 | * and get instead the orphan name. | |
4056 | */ | |
4057 | nce = name_cache_search(sctx, ow_inode, ow_gen); | |
4058 | if (nce) { | |
4059 | name_cache_delete(sctx, nce); | |
4060 | kfree(nce); | |
4061 | } | |
801bec36 RK |
4062 | |
4063 | /* | |
4064 | * ow_inode might currently be an ancestor of | |
4065 | * cur_ino, therefore compute valid_path (the | |
4066 | * current path of cur_ino) again because it | |
4067 | * might contain the pre-orphanization name of | |
4068 | * ow_inode, which is no longer valid. | |
4069 | */ | |
72c3668f FM |
4070 | ret = is_ancestor(sctx->parent_root, |
4071 | ow_inode, ow_gen, | |
4072 | sctx->cur_ino, NULL); | |
4073 | if (ret > 0) { | |
fdb13889 | 4074 | orphanized_ancestor = true; |
72c3668f FM |
4075 | fs_path_reset(valid_path); |
4076 | ret = get_cur_path(sctx, sctx->cur_ino, | |
4077 | sctx->cur_inode_gen, | |
4078 | valid_path); | |
4079 | } | |
801bec36 RK |
4080 | if (ret < 0) |
4081 | goto out; | |
31db9f7c AB |
4082 | } else { |
4083 | ret = send_unlink(sctx, cur->full_path); | |
4084 | if (ret < 0) | |
4085 | goto out; | |
4086 | } | |
4087 | } | |
4088 | ||
98272bb7 FM |
4089 | } |
4090 | ||
4091 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
4092 | /* | |
4093 | * We may have refs where the parent directory does not exist | |
4094 | * yet. This happens if the parent directories inum is higher | |
4095 | * than the current inum. To handle this case, we create the | |
4096 | * parent directory out of order. But we need to check if this | |
4097 | * did already happen before due to other refs in the same dir. | |
4098 | */ | |
4099 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); | |
4100 | if (ret < 0) | |
4101 | goto out; | |
4102 | if (ret == inode_state_will_create) { | |
4103 | ret = 0; | |
4104 | /* | |
4105 | * First check if any of the current inodes refs did | |
4106 | * already create the dir. | |
4107 | */ | |
4108 | list_for_each_entry(cur2, &sctx->new_refs, list) { | |
4109 | if (cur == cur2) | |
4110 | break; | |
4111 | if (cur2->dir == cur->dir) { | |
4112 | ret = 1; | |
4113 | break; | |
4114 | } | |
4115 | } | |
4116 | ||
4117 | /* | |
4118 | * If that did not happen, check if a previous inode | |
4119 | * did already create the dir. | |
4120 | */ | |
4121 | if (!ret) | |
4122 | ret = did_create_dir(sctx, cur->dir); | |
4123 | if (ret < 0) | |
4124 | goto out; | |
4125 | if (!ret) { | |
4126 | ret = send_create_inode(sctx, cur->dir); | |
4127 | if (ret < 0) | |
4128 | goto out; | |
4129 | } | |
4130 | } | |
4131 | ||
84471e24 FM |
4132 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { |
4133 | ret = wait_for_dest_dir_move(sctx, cur, is_orphan); | |
4134 | if (ret < 0) | |
4135 | goto out; | |
4136 | if (ret == 1) { | |
4137 | can_rename = false; | |
4138 | *pending_move = 1; | |
4139 | } | |
4140 | } | |
4141 | ||
8b191a68 FM |
4142 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root && |
4143 | can_rename) { | |
4144 | ret = wait_for_parent_move(sctx, cur, is_orphan); | |
4145 | if (ret < 0) | |
4146 | goto out; | |
4147 | if (ret == 1) { | |
4148 | can_rename = false; | |
4149 | *pending_move = 1; | |
4150 | } | |
4151 | } | |
4152 | ||
31db9f7c AB |
4153 | /* |
4154 | * link/move the ref to the new place. If we have an orphan | |
4155 | * inode, move it and update valid_path. If not, link or move | |
4156 | * it depending on the inode mode. | |
4157 | */ | |
84471e24 | 4158 | if (is_orphan && can_rename) { |
31db9f7c AB |
4159 | ret = send_rename(sctx, valid_path, cur->full_path); |
4160 | if (ret < 0) | |
4161 | goto out; | |
4162 | is_orphan = 0; | |
4163 | ret = fs_path_copy(valid_path, cur->full_path); | |
4164 | if (ret < 0) | |
4165 | goto out; | |
84471e24 | 4166 | } else if (can_rename) { |
31db9f7c AB |
4167 | if (S_ISDIR(sctx->cur_inode_mode)) { |
4168 | /* | |
4169 | * Dirs can't be linked, so move it. For moved | |
4170 | * dirs, we always have one new and one deleted | |
4171 | * ref. The deleted ref is ignored later. | |
4172 | */ | |
8b191a68 FM |
4173 | ret = send_rename(sctx, valid_path, |
4174 | cur->full_path); | |
4175 | if (!ret) | |
4176 | ret = fs_path_copy(valid_path, | |
4177 | cur->full_path); | |
31db9f7c AB |
4178 | if (ret < 0) |
4179 | goto out; | |
4180 | } else { | |
f5962781 FM |
4181 | /* |
4182 | * We might have previously orphanized an inode | |
4183 | * which is an ancestor of our current inode, | |
4184 | * so our reference's full path, which was | |
4185 | * computed before any such orphanizations, must | |
4186 | * be updated. | |
4187 | */ | |
4188 | if (orphanized_dir) { | |
4189 | ret = update_ref_path(sctx, cur); | |
4190 | if (ret < 0) | |
4191 | goto out; | |
4192 | } | |
31db9f7c AB |
4193 | ret = send_link(sctx, cur->full_path, |
4194 | valid_path); | |
4195 | if (ret < 0) | |
4196 | goto out; | |
4197 | } | |
4198 | } | |
ba5e8f2e | 4199 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4200 | if (ret < 0) |
4201 | goto out; | |
4202 | } | |
4203 | ||
4204 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { | |
4205 | /* | |
4206 | * Check if we can already rmdir the directory. If not, | |
4207 | * orphanize it. For every dir item inside that gets deleted | |
4208 | * later, we do this check again and rmdir it then if possible. | |
4209 | * See the use of check_dirs for more details. | |
4210 | */ | |
9dc44214 FM |
4211 | ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
4212 | sctx->cur_ino); | |
31db9f7c AB |
4213 | if (ret < 0) |
4214 | goto out; | |
4215 | if (ret) { | |
4216 | ret = send_rmdir(sctx, valid_path); | |
4217 | if (ret < 0) | |
4218 | goto out; | |
4219 | } else if (!is_orphan) { | |
4220 | ret = orphanize_inode(sctx, sctx->cur_ino, | |
4221 | sctx->cur_inode_gen, valid_path); | |
4222 | if (ret < 0) | |
4223 | goto out; | |
4224 | is_orphan = 1; | |
4225 | } | |
4226 | ||
4227 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
ba5e8f2e | 4228 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4229 | if (ret < 0) |
4230 | goto out; | |
4231 | } | |
ccf1626b AB |
4232 | } else if (S_ISDIR(sctx->cur_inode_mode) && |
4233 | !list_empty(&sctx->deleted_refs)) { | |
4234 | /* | |
4235 | * We have a moved dir. Add the old parent to check_dirs | |
4236 | */ | |
4237 | cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, | |
4238 | list); | |
ba5e8f2e | 4239 | ret = dup_ref(cur, &check_dirs); |
ccf1626b AB |
4240 | if (ret < 0) |
4241 | goto out; | |
31db9f7c AB |
4242 | } else if (!S_ISDIR(sctx->cur_inode_mode)) { |
4243 | /* | |
4244 | * We have a non dir inode. Go through all deleted refs and | |
4245 | * unlink them if they were not already overwritten by other | |
4246 | * inodes. | |
4247 | */ | |
4248 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
4249 | ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
4250 | sctx->cur_ino, sctx->cur_inode_gen, | |
4251 | cur->name, cur->name_len); | |
4252 | if (ret < 0) | |
4253 | goto out; | |
4254 | if (!ret) { | |
fdb13889 FM |
4255 | /* |
4256 | * If we orphanized any ancestor before, we need | |
4257 | * to recompute the full path for deleted names, | |
4258 | * since any such path was computed before we | |
4259 | * processed any references and orphanized any | |
4260 | * ancestor inode. | |
4261 | */ | |
4262 | if (orphanized_ancestor) { | |
f5962781 FM |
4263 | ret = update_ref_path(sctx, cur); |
4264 | if (ret < 0) | |
fdb13889 | 4265 | goto out; |
fdb13889 | 4266 | } |
1f4692da AB |
4267 | ret = send_unlink(sctx, cur->full_path); |
4268 | if (ret < 0) | |
4269 | goto out; | |
31db9f7c | 4270 | } |
ba5e8f2e | 4271 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4272 | if (ret < 0) |
4273 | goto out; | |
4274 | } | |
31db9f7c AB |
4275 | /* |
4276 | * If the inode is still orphan, unlink the orphan. This may | |
4277 | * happen when a previous inode did overwrite the first ref | |
4278 | * of this inode and no new refs were added for the current | |
766702ef AB |
4279 | * inode. Unlinking does not mean that the inode is deleted in |
4280 | * all cases. There may still be links to this inode in other | |
4281 | * places. | |
31db9f7c | 4282 | */ |
1f4692da | 4283 | if (is_orphan) { |
31db9f7c AB |
4284 | ret = send_unlink(sctx, valid_path); |
4285 | if (ret < 0) | |
4286 | goto out; | |
4287 | } | |
4288 | } | |
4289 | ||
4290 | /* | |
4291 | * We did collect all parent dirs where cur_inode was once located. We | |
4292 | * now go through all these dirs and check if they are pending for | |
4293 | * deletion and if it's finally possible to perform the rmdir now. | |
4294 | * We also update the inode stats of the parent dirs here. | |
4295 | */ | |
ba5e8f2e | 4296 | list_for_each_entry(cur, &check_dirs, list) { |
766702ef AB |
4297 | /* |
4298 | * In case we had refs into dirs that were not processed yet, | |
4299 | * we don't need to do the utime and rmdir logic for these dirs. | |
4300 | * The dir will be processed later. | |
4301 | */ | |
ba5e8f2e | 4302 | if (cur->dir > sctx->cur_ino) |
31db9f7c AB |
4303 | continue; |
4304 | ||
ba5e8f2e | 4305 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4306 | if (ret < 0) |
4307 | goto out; | |
4308 | ||
4309 | if (ret == inode_state_did_create || | |
4310 | ret == inode_state_no_change) { | |
4311 | /* TODO delayed utimes */ | |
ba5e8f2e | 4312 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4313 | if (ret < 0) |
4314 | goto out; | |
29d6d30f FM |
4315 | } else if (ret == inode_state_did_delete && |
4316 | cur->dir != last_dir_ino_rm) { | |
9dc44214 FM |
4317 | ret = can_rmdir(sctx, cur->dir, cur->dir_gen, |
4318 | sctx->cur_ino); | |
31db9f7c AB |
4319 | if (ret < 0) |
4320 | goto out; | |
4321 | if (ret) { | |
ba5e8f2e JB |
4322 | ret = get_cur_path(sctx, cur->dir, |
4323 | cur->dir_gen, valid_path); | |
31db9f7c AB |
4324 | if (ret < 0) |
4325 | goto out; | |
4326 | ret = send_rmdir(sctx, valid_path); | |
4327 | if (ret < 0) | |
4328 | goto out; | |
29d6d30f | 4329 | last_dir_ino_rm = cur->dir; |
31db9f7c AB |
4330 | } |
4331 | } | |
4332 | } | |
4333 | ||
31db9f7c AB |
4334 | ret = 0; |
4335 | ||
4336 | out: | |
ba5e8f2e | 4337 | __free_recorded_refs(&check_dirs); |
31db9f7c | 4338 | free_recorded_refs(sctx); |
924794c9 | 4339 | fs_path_free(valid_path); |
31db9f7c AB |
4340 | return ret; |
4341 | } | |
4342 | ||
a0357511 NB |
4343 | static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name, |
4344 | void *ctx, struct list_head *refs) | |
31db9f7c AB |
4345 | { |
4346 | int ret = 0; | |
4347 | struct send_ctx *sctx = ctx; | |
4348 | struct fs_path *p; | |
4349 | u64 gen; | |
4350 | ||
924794c9 | 4351 | p = fs_path_alloc(); |
31db9f7c AB |
4352 | if (!p) |
4353 | return -ENOMEM; | |
4354 | ||
a4d96d62 | 4355 | ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, |
85a7b33b | 4356 | NULL, NULL); |
31db9f7c AB |
4357 | if (ret < 0) |
4358 | goto out; | |
4359 | ||
31db9f7c AB |
4360 | ret = get_cur_path(sctx, dir, gen, p); |
4361 | if (ret < 0) | |
4362 | goto out; | |
4363 | ret = fs_path_add_path(p, name); | |
4364 | if (ret < 0) | |
4365 | goto out; | |
4366 | ||
a4d96d62 | 4367 | ret = __record_ref(refs, dir, gen, p); |
31db9f7c AB |
4368 | |
4369 | out: | |
4370 | if (ret) | |
924794c9 | 4371 | fs_path_free(p); |
31db9f7c AB |
4372 | return ret; |
4373 | } | |
4374 | ||
a4d96d62 LB |
4375 | static int __record_new_ref(int num, u64 dir, int index, |
4376 | struct fs_path *name, | |
4377 | void *ctx) | |
4378 | { | |
4379 | struct send_ctx *sctx = ctx; | |
a0357511 | 4380 | return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs); |
a4d96d62 LB |
4381 | } |
4382 | ||
4383 | ||
31db9f7c AB |
4384 | static int __record_deleted_ref(int num, u64 dir, int index, |
4385 | struct fs_path *name, | |
4386 | void *ctx) | |
4387 | { | |
31db9f7c | 4388 | struct send_ctx *sctx = ctx; |
a0357511 NB |
4389 | return record_ref(sctx->parent_root, dir, name, ctx, |
4390 | &sctx->deleted_refs); | |
31db9f7c AB |
4391 | } |
4392 | ||
4393 | static int record_new_ref(struct send_ctx *sctx) | |
4394 | { | |
4395 | int ret; | |
4396 | ||
924794c9 TI |
4397 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
4398 | sctx->cmp_key, 0, __record_new_ref, sctx); | |
31db9f7c AB |
4399 | if (ret < 0) |
4400 | goto out; | |
4401 | ret = 0; | |
4402 | ||
4403 | out: | |
4404 | return ret; | |
4405 | } | |
4406 | ||
4407 | static int record_deleted_ref(struct send_ctx *sctx) | |
4408 | { | |
4409 | int ret; | |
4410 | ||
924794c9 TI |
4411 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
4412 | sctx->cmp_key, 0, __record_deleted_ref, sctx); | |
31db9f7c AB |
4413 | if (ret < 0) |
4414 | goto out; | |
4415 | ret = 0; | |
4416 | ||
4417 | out: | |
4418 | return ret; | |
4419 | } | |
4420 | ||
4421 | struct find_ref_ctx { | |
4422 | u64 dir; | |
ba5e8f2e JB |
4423 | u64 dir_gen; |
4424 | struct btrfs_root *root; | |
31db9f7c AB |
4425 | struct fs_path *name; |
4426 | int found_idx; | |
4427 | }; | |
4428 | ||
4429 | static int __find_iref(int num, u64 dir, int index, | |
4430 | struct fs_path *name, | |
4431 | void *ctx_) | |
4432 | { | |
4433 | struct find_ref_ctx *ctx = ctx_; | |
ba5e8f2e JB |
4434 | u64 dir_gen; |
4435 | int ret; | |
31db9f7c AB |
4436 | |
4437 | if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && | |
4438 | strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { | |
ba5e8f2e JB |
4439 | /* |
4440 | * To avoid doing extra lookups we'll only do this if everything | |
4441 | * else matches. | |
4442 | */ | |
4443 | ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, | |
4444 | NULL, NULL, NULL); | |
4445 | if (ret) | |
4446 | return ret; | |
4447 | if (dir_gen != ctx->dir_gen) | |
4448 | return 0; | |
31db9f7c AB |
4449 | ctx->found_idx = num; |
4450 | return 1; | |
4451 | } | |
4452 | return 0; | |
4453 | } | |
4454 | ||
924794c9 | 4455 | static int find_iref(struct btrfs_root *root, |
31db9f7c AB |
4456 | struct btrfs_path *path, |
4457 | struct btrfs_key *key, | |
ba5e8f2e | 4458 | u64 dir, u64 dir_gen, struct fs_path *name) |
31db9f7c AB |
4459 | { |
4460 | int ret; | |
4461 | struct find_ref_ctx ctx; | |
4462 | ||
4463 | ctx.dir = dir; | |
4464 | ctx.name = name; | |
ba5e8f2e | 4465 | ctx.dir_gen = dir_gen; |
31db9f7c | 4466 | ctx.found_idx = -1; |
ba5e8f2e | 4467 | ctx.root = root; |
31db9f7c | 4468 | |
924794c9 | 4469 | ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); |
31db9f7c AB |
4470 | if (ret < 0) |
4471 | return ret; | |
4472 | ||
4473 | if (ctx.found_idx == -1) | |
4474 | return -ENOENT; | |
4475 | ||
4476 | return ctx.found_idx; | |
4477 | } | |
4478 | ||
4479 | static int __record_changed_new_ref(int num, u64 dir, int index, | |
4480 | struct fs_path *name, | |
4481 | void *ctx) | |
4482 | { | |
ba5e8f2e | 4483 | u64 dir_gen; |
31db9f7c AB |
4484 | int ret; |
4485 | struct send_ctx *sctx = ctx; | |
4486 | ||
ba5e8f2e JB |
4487 | ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, |
4488 | NULL, NULL, NULL); | |
4489 | if (ret) | |
4490 | return ret; | |
4491 | ||
924794c9 | 4492 | ret = find_iref(sctx->parent_root, sctx->right_path, |
ba5e8f2e | 4493 | sctx->cmp_key, dir, dir_gen, name); |
31db9f7c AB |
4494 | if (ret == -ENOENT) |
4495 | ret = __record_new_ref(num, dir, index, name, sctx); | |
4496 | else if (ret > 0) | |
4497 | ret = 0; | |
4498 | ||
4499 | return ret; | |
4500 | } | |
4501 | ||
4502 | static int __record_changed_deleted_ref(int num, u64 dir, int index, | |
4503 | struct fs_path *name, | |
4504 | void *ctx) | |
4505 | { | |
ba5e8f2e | 4506 | u64 dir_gen; |
31db9f7c AB |
4507 | int ret; |
4508 | struct send_ctx *sctx = ctx; | |
4509 | ||
ba5e8f2e JB |
4510 | ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, |
4511 | NULL, NULL, NULL); | |
4512 | if (ret) | |
4513 | return ret; | |
4514 | ||
924794c9 | 4515 | ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, |
ba5e8f2e | 4516 | dir, dir_gen, name); |
31db9f7c AB |
4517 | if (ret == -ENOENT) |
4518 | ret = __record_deleted_ref(num, dir, index, name, sctx); | |
4519 | else if (ret > 0) | |
4520 | ret = 0; | |
4521 | ||
4522 | return ret; | |
4523 | } | |
4524 | ||
4525 | static int record_changed_ref(struct send_ctx *sctx) | |
4526 | { | |
4527 | int ret = 0; | |
4528 | ||
924794c9 | 4529 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
31db9f7c AB |
4530 | sctx->cmp_key, 0, __record_changed_new_ref, sctx); |
4531 | if (ret < 0) | |
4532 | goto out; | |
924794c9 | 4533 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
4534 | sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); |
4535 | if (ret < 0) | |
4536 | goto out; | |
4537 | ret = 0; | |
4538 | ||
4539 | out: | |
4540 | return ret; | |
4541 | } | |
4542 | ||
4543 | /* | |
4544 | * Record and process all refs at once. Needed when an inode changes the | |
4545 | * generation number, which means that it was deleted and recreated. | |
4546 | */ | |
4547 | static int process_all_refs(struct send_ctx *sctx, | |
4548 | enum btrfs_compare_tree_result cmd) | |
4549 | { | |
4550 | int ret; | |
4551 | struct btrfs_root *root; | |
4552 | struct btrfs_path *path; | |
4553 | struct btrfs_key key; | |
4554 | struct btrfs_key found_key; | |
4555 | struct extent_buffer *eb; | |
4556 | int slot; | |
4557 | iterate_inode_ref_t cb; | |
9f03740a | 4558 | int pending_move = 0; |
31db9f7c AB |
4559 | |
4560 | path = alloc_path_for_send(); | |
4561 | if (!path) | |
4562 | return -ENOMEM; | |
4563 | ||
4564 | if (cmd == BTRFS_COMPARE_TREE_NEW) { | |
4565 | root = sctx->send_root; | |
4566 | cb = __record_new_ref; | |
4567 | } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { | |
4568 | root = sctx->parent_root; | |
4569 | cb = __record_deleted_ref; | |
4570 | } else { | |
4d1a63b2 DS |
4571 | btrfs_err(sctx->send_root->fs_info, |
4572 | "Wrong command %d in process_all_refs", cmd); | |
4573 | ret = -EINVAL; | |
4574 | goto out; | |
31db9f7c AB |
4575 | } |
4576 | ||
4577 | key.objectid = sctx->cmp_key->objectid; | |
4578 | key.type = BTRFS_INODE_REF_KEY; | |
4579 | key.offset = 0; | |
dff6d0ad FDBM |
4580 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4581 | if (ret < 0) | |
4582 | goto out; | |
31db9f7c | 4583 | |
dff6d0ad | 4584 | while (1) { |
31db9f7c AB |
4585 | eb = path->nodes[0]; |
4586 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4587 | if (slot >= btrfs_header_nritems(eb)) { |
4588 | ret = btrfs_next_leaf(root, path); | |
4589 | if (ret < 0) | |
4590 | goto out; | |
4591 | else if (ret > 0) | |
4592 | break; | |
4593 | continue; | |
4594 | } | |
4595 | ||
31db9f7c AB |
4596 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
4597 | ||
4598 | if (found_key.objectid != key.objectid || | |
96b5bd77 JS |
4599 | (found_key.type != BTRFS_INODE_REF_KEY && |
4600 | found_key.type != BTRFS_INODE_EXTREF_KEY)) | |
31db9f7c | 4601 | break; |
31db9f7c | 4602 | |
924794c9 | 4603 | ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); |
31db9f7c AB |
4604 | if (ret < 0) |
4605 | goto out; | |
4606 | ||
dff6d0ad | 4607 | path->slots[0]++; |
31db9f7c | 4608 | } |
e938c8ad | 4609 | btrfs_release_path(path); |
31db9f7c | 4610 | |
3dc09ec8 JB |
4611 | /* |
4612 | * We don't actually care about pending_move as we are simply | |
4613 | * re-creating this inode and will be rename'ing it into place once we | |
4614 | * rename the parent directory. | |
4615 | */ | |
9f03740a | 4616 | ret = process_recorded_refs(sctx, &pending_move); |
31db9f7c AB |
4617 | out: |
4618 | btrfs_free_path(path); | |
4619 | return ret; | |
4620 | } | |
4621 | ||
4622 | static int send_set_xattr(struct send_ctx *sctx, | |
4623 | struct fs_path *path, | |
4624 | const char *name, int name_len, | |
4625 | const char *data, int data_len) | |
4626 | { | |
4627 | int ret = 0; | |
4628 | ||
4629 | ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); | |
4630 | if (ret < 0) | |
4631 | goto out; | |
4632 | ||
4633 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4634 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4635 | TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); | |
4636 | ||
4637 | ret = send_cmd(sctx); | |
4638 | ||
4639 | tlv_put_failure: | |
4640 | out: | |
4641 | return ret; | |
4642 | } | |
4643 | ||
4644 | static int send_remove_xattr(struct send_ctx *sctx, | |
4645 | struct fs_path *path, | |
4646 | const char *name, int name_len) | |
4647 | { | |
4648 | int ret = 0; | |
4649 | ||
4650 | ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); | |
4651 | if (ret < 0) | |
4652 | goto out; | |
4653 | ||
4654 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4655 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4656 | ||
4657 | ret = send_cmd(sctx); | |
4658 | ||
4659 | tlv_put_failure: | |
4660 | out: | |
4661 | return ret; | |
4662 | } | |
4663 | ||
4664 | static int __process_new_xattr(int num, struct btrfs_key *di_key, | |
4665 | const char *name, int name_len, | |
4666 | const char *data, int data_len, | |
4667 | u8 type, void *ctx) | |
4668 | { | |
4669 | int ret; | |
4670 | struct send_ctx *sctx = ctx; | |
4671 | struct fs_path *p; | |
2211d5ba | 4672 | struct posix_acl_xattr_header dummy_acl; |
31db9f7c | 4673 | |
89efda52 MPS |
4674 | /* Capabilities are emitted by finish_inode_if_needed */ |
4675 | if (!strncmp(name, XATTR_NAME_CAPS, name_len)) | |
4676 | return 0; | |
4677 | ||
924794c9 | 4678 | p = fs_path_alloc(); |
31db9f7c AB |
4679 | if (!p) |
4680 | return -ENOMEM; | |
4681 | ||
4682 | /* | |
01327610 | 4683 | * This hack is needed because empty acls are stored as zero byte |
31db9f7c | 4684 | * data in xattrs. Problem with that is, that receiving these zero byte |
01327610 | 4685 | * acls will fail later. To fix this, we send a dummy acl list that |
31db9f7c AB |
4686 | * only contains the version number and no entries. |
4687 | */ | |
4688 | if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || | |
4689 | !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { | |
4690 | if (data_len == 0) { | |
4691 | dummy_acl.a_version = | |
4692 | cpu_to_le32(POSIX_ACL_XATTR_VERSION); | |
4693 | data = (char *)&dummy_acl; | |
4694 | data_len = sizeof(dummy_acl); | |
4695 | } | |
4696 | } | |
4697 | ||
4698 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4699 | if (ret < 0) | |
4700 | goto out; | |
4701 | ||
4702 | ret = send_set_xattr(sctx, p, name, name_len, data, data_len); | |
4703 | ||
4704 | out: | |
924794c9 | 4705 | fs_path_free(p); |
31db9f7c AB |
4706 | return ret; |
4707 | } | |
4708 | ||
4709 | static int __process_deleted_xattr(int num, struct btrfs_key *di_key, | |
4710 | const char *name, int name_len, | |
4711 | const char *data, int data_len, | |
4712 | u8 type, void *ctx) | |
4713 | { | |
4714 | int ret; | |
4715 | struct send_ctx *sctx = ctx; | |
4716 | struct fs_path *p; | |
4717 | ||
924794c9 | 4718 | p = fs_path_alloc(); |
31db9f7c AB |
4719 | if (!p) |
4720 | return -ENOMEM; | |
4721 | ||
4722 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4723 | if (ret < 0) | |
4724 | goto out; | |
4725 | ||
4726 | ret = send_remove_xattr(sctx, p, name, name_len); | |
4727 | ||
4728 | out: | |
924794c9 | 4729 | fs_path_free(p); |
31db9f7c AB |
4730 | return ret; |
4731 | } | |
4732 | ||
4733 | static int process_new_xattr(struct send_ctx *sctx) | |
4734 | { | |
4735 | int ret = 0; | |
4736 | ||
924794c9 | 4737 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4738 | __process_new_xattr, sctx); |
31db9f7c AB |
4739 | |
4740 | return ret; | |
4741 | } | |
4742 | ||
4743 | static int process_deleted_xattr(struct send_ctx *sctx) | |
4744 | { | |
e2c89907 | 4745 | return iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4746 | __process_deleted_xattr, sctx); |
31db9f7c AB |
4747 | } |
4748 | ||
4749 | struct find_xattr_ctx { | |
4750 | const char *name; | |
4751 | int name_len; | |
4752 | int found_idx; | |
4753 | char *found_data; | |
4754 | int found_data_len; | |
4755 | }; | |
4756 | ||
4757 | static int __find_xattr(int num, struct btrfs_key *di_key, | |
4758 | const char *name, int name_len, | |
4759 | const char *data, int data_len, | |
4760 | u8 type, void *vctx) | |
4761 | { | |
4762 | struct find_xattr_ctx *ctx = vctx; | |
4763 | ||
4764 | if (name_len == ctx->name_len && | |
4765 | strncmp(name, ctx->name, name_len) == 0) { | |
4766 | ctx->found_idx = num; | |
4767 | ctx->found_data_len = data_len; | |
e780b0d1 | 4768 | ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); |
31db9f7c AB |
4769 | if (!ctx->found_data) |
4770 | return -ENOMEM; | |
31db9f7c AB |
4771 | return 1; |
4772 | } | |
4773 | return 0; | |
4774 | } | |
4775 | ||
924794c9 | 4776 | static int find_xattr(struct btrfs_root *root, |
31db9f7c AB |
4777 | struct btrfs_path *path, |
4778 | struct btrfs_key *key, | |
4779 | const char *name, int name_len, | |
4780 | char **data, int *data_len) | |
4781 | { | |
4782 | int ret; | |
4783 | struct find_xattr_ctx ctx; | |
4784 | ||
4785 | ctx.name = name; | |
4786 | ctx.name_len = name_len; | |
4787 | ctx.found_idx = -1; | |
4788 | ctx.found_data = NULL; | |
4789 | ctx.found_data_len = 0; | |
4790 | ||
a0357511 | 4791 | ret = iterate_dir_item(root, path, __find_xattr, &ctx); |
31db9f7c AB |
4792 | if (ret < 0) |
4793 | return ret; | |
4794 | ||
4795 | if (ctx.found_idx == -1) | |
4796 | return -ENOENT; | |
4797 | if (data) { | |
4798 | *data = ctx.found_data; | |
4799 | *data_len = ctx.found_data_len; | |
4800 | } else { | |
4801 | kfree(ctx.found_data); | |
4802 | } | |
4803 | return ctx.found_idx; | |
4804 | } | |
4805 | ||
4806 | ||
4807 | static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, | |
4808 | const char *name, int name_len, | |
4809 | const char *data, int data_len, | |
4810 | u8 type, void *ctx) | |
4811 | { | |
4812 | int ret; | |
4813 | struct send_ctx *sctx = ctx; | |
4814 | char *found_data = NULL; | |
4815 | int found_data_len = 0; | |
31db9f7c | 4816 | |
924794c9 TI |
4817 | ret = find_xattr(sctx->parent_root, sctx->right_path, |
4818 | sctx->cmp_key, name, name_len, &found_data, | |
4819 | &found_data_len); | |
31db9f7c AB |
4820 | if (ret == -ENOENT) { |
4821 | ret = __process_new_xattr(num, di_key, name, name_len, data, | |
4822 | data_len, type, ctx); | |
4823 | } else if (ret >= 0) { | |
4824 | if (data_len != found_data_len || | |
4825 | memcmp(data, found_data, data_len)) { | |
4826 | ret = __process_new_xattr(num, di_key, name, name_len, | |
4827 | data, data_len, type, ctx); | |
4828 | } else { | |
4829 | ret = 0; | |
4830 | } | |
4831 | } | |
4832 | ||
4833 | kfree(found_data); | |
31db9f7c AB |
4834 | return ret; |
4835 | } | |
4836 | ||
4837 | static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, | |
4838 | const char *name, int name_len, | |
4839 | const char *data, int data_len, | |
4840 | u8 type, void *ctx) | |
4841 | { | |
4842 | int ret; | |
4843 | struct send_ctx *sctx = ctx; | |
4844 | ||
924794c9 TI |
4845 | ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, |
4846 | name, name_len, NULL, NULL); | |
31db9f7c AB |
4847 | if (ret == -ENOENT) |
4848 | ret = __process_deleted_xattr(num, di_key, name, name_len, data, | |
4849 | data_len, type, ctx); | |
4850 | else if (ret >= 0) | |
4851 | ret = 0; | |
4852 | ||
4853 | return ret; | |
4854 | } | |
4855 | ||
4856 | static int process_changed_xattr(struct send_ctx *sctx) | |
4857 | { | |
4858 | int ret = 0; | |
4859 | ||
924794c9 | 4860 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4861 | __process_changed_new_xattr, sctx); |
31db9f7c AB |
4862 | if (ret < 0) |
4863 | goto out; | |
924794c9 | 4864 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4865 | __process_changed_deleted_xattr, sctx); |
31db9f7c AB |
4866 | |
4867 | out: | |
4868 | return ret; | |
4869 | } | |
4870 | ||
4871 | static int process_all_new_xattrs(struct send_ctx *sctx) | |
4872 | { | |
4873 | int ret; | |
4874 | struct btrfs_root *root; | |
4875 | struct btrfs_path *path; | |
4876 | struct btrfs_key key; | |
4877 | struct btrfs_key found_key; | |
4878 | struct extent_buffer *eb; | |
4879 | int slot; | |
4880 | ||
4881 | path = alloc_path_for_send(); | |
4882 | if (!path) | |
4883 | return -ENOMEM; | |
4884 | ||
4885 | root = sctx->send_root; | |
4886 | ||
4887 | key.objectid = sctx->cmp_key->objectid; | |
4888 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4889 | key.offset = 0; | |
dff6d0ad FDBM |
4890 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4891 | if (ret < 0) | |
4892 | goto out; | |
31db9f7c | 4893 | |
dff6d0ad | 4894 | while (1) { |
31db9f7c AB |
4895 | eb = path->nodes[0]; |
4896 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4897 | if (slot >= btrfs_header_nritems(eb)) { |
4898 | ret = btrfs_next_leaf(root, path); | |
4899 | if (ret < 0) { | |
4900 | goto out; | |
4901 | } else if (ret > 0) { | |
4902 | ret = 0; | |
4903 | break; | |
4904 | } | |
4905 | continue; | |
4906 | } | |
31db9f7c | 4907 | |
dff6d0ad | 4908 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
31db9f7c AB |
4909 | if (found_key.objectid != key.objectid || |
4910 | found_key.type != key.type) { | |
4911 | ret = 0; | |
4912 | goto out; | |
4913 | } | |
4914 | ||
a0357511 | 4915 | ret = iterate_dir_item(root, path, __process_new_xattr, sctx); |
31db9f7c AB |
4916 | if (ret < 0) |
4917 | goto out; | |
4918 | ||
dff6d0ad | 4919 | path->slots[0]++; |
31db9f7c AB |
4920 | } |
4921 | ||
4922 | out: | |
4923 | btrfs_free_path(path); | |
4924 | return ret; | |
4925 | } | |
4926 | ||
8c7d9fe0 OS |
4927 | static inline u64 max_send_read_size(const struct send_ctx *sctx) |
4928 | { | |
4929 | return sctx->send_max_size - SZ_16K; | |
4930 | } | |
4931 | ||
4932 | static int put_data_header(struct send_ctx *sctx, u32 len) | |
4933 | { | |
4934 | struct btrfs_tlv_header *hdr; | |
4935 | ||
4936 | if (sctx->send_max_size - sctx->send_size < sizeof(*hdr) + len) | |
4937 | return -EOVERFLOW; | |
4938 | hdr = (struct btrfs_tlv_header *)(sctx->send_buf + sctx->send_size); | |
4939 | put_unaligned_le16(BTRFS_SEND_A_DATA, &hdr->tlv_type); | |
4940 | put_unaligned_le16(len, &hdr->tlv_len); | |
4941 | sctx->send_size += sizeof(*hdr); | |
4942 | return 0; | |
4943 | } | |
4944 | ||
4945 | static int put_file_data(struct send_ctx *sctx, u64 offset, u32 len) | |
ed259095 JB |
4946 | { |
4947 | struct btrfs_root *root = sctx->send_root; | |
4948 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4949 | struct inode *inode; | |
4950 | struct page *page; | |
4951 | char *addr; | |
09cbfeaf | 4952 | pgoff_t index = offset >> PAGE_SHIFT; |
ed259095 | 4953 | pgoff_t last_index; |
7073017a | 4954 | unsigned pg_offset = offset_in_page(offset); |
8c7d9fe0 OS |
4955 | int ret; |
4956 | ||
4957 | ret = put_data_header(sctx, len); | |
4958 | if (ret) | |
4959 | return ret; | |
ed259095 | 4960 | |
0202e83f | 4961 | inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root); |
ed259095 JB |
4962 | if (IS_ERR(inode)) |
4963 | return PTR_ERR(inode); | |
4964 | ||
09cbfeaf | 4965 | last_index = (offset + len - 1) >> PAGE_SHIFT; |
2131bcd3 LB |
4966 | |
4967 | /* initial readahead */ | |
4968 | memset(&sctx->ra, 0, sizeof(struct file_ra_state)); | |
4969 | file_ra_state_init(&sctx->ra, inode->i_mapping); | |
2131bcd3 | 4970 | |
ed259095 JB |
4971 | while (index <= last_index) { |
4972 | unsigned cur_len = min_t(unsigned, len, | |
09cbfeaf | 4973 | PAGE_SIZE - pg_offset); |
eef16ba2 KH |
4974 | |
4975 | page = find_lock_page(inode->i_mapping, index); | |
ed259095 | 4976 | if (!page) { |
eef16ba2 KH |
4977 | page_cache_sync_readahead(inode->i_mapping, &sctx->ra, |
4978 | NULL, index, last_index + 1 - index); | |
4979 | ||
4980 | page = find_or_create_page(inode->i_mapping, index, | |
4981 | GFP_KERNEL); | |
4982 | if (!page) { | |
4983 | ret = -ENOMEM; | |
4984 | break; | |
4985 | } | |
4986 | } | |
4987 | ||
4988 | if (PageReadahead(page)) { | |
4989 | page_cache_async_readahead(inode->i_mapping, &sctx->ra, | |
4990 | NULL, page, index, last_index + 1 - index); | |
ed259095 JB |
4991 | } |
4992 | ||
4993 | if (!PageUptodate(page)) { | |
4994 | btrfs_readpage(NULL, page); | |
4995 | lock_page(page); | |
4996 | if (!PageUptodate(page)) { | |
4997 | unlock_page(page); | |
09cbfeaf | 4998 | put_page(page); |
ed259095 JB |
4999 | ret = -EIO; |
5000 | break; | |
5001 | } | |
5002 | } | |
5003 | ||
5004 | addr = kmap(page); | |
8c7d9fe0 OS |
5005 | memcpy(sctx->send_buf + sctx->send_size, addr + pg_offset, |
5006 | cur_len); | |
ed259095 JB |
5007 | kunmap(page); |
5008 | unlock_page(page); | |
09cbfeaf | 5009 | put_page(page); |
ed259095 JB |
5010 | index++; |
5011 | pg_offset = 0; | |
5012 | len -= cur_len; | |
8c7d9fe0 | 5013 | sctx->send_size += cur_len; |
ed259095 | 5014 | } |
ed259095 JB |
5015 | iput(inode); |
5016 | return ret; | |
5017 | } | |
5018 | ||
31db9f7c AB |
5019 | /* |
5020 | * Read some bytes from the current inode/file and send a write command to | |
5021 | * user space. | |
5022 | */ | |
5023 | static int send_write(struct send_ctx *sctx, u64 offset, u32 len) | |
5024 | { | |
04ab956e | 5025 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
5026 | int ret = 0; |
5027 | struct fs_path *p; | |
31db9f7c | 5028 | |
924794c9 | 5029 | p = fs_path_alloc(); |
31db9f7c AB |
5030 | if (!p) |
5031 | return -ENOMEM; | |
5032 | ||
04ab956e | 5033 | btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len); |
31db9f7c | 5034 | |
31db9f7c AB |
5035 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); |
5036 | if (ret < 0) | |
5037 | goto out; | |
5038 | ||
5039 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
5040 | if (ret < 0) | |
5041 | goto out; | |
5042 | ||
5043 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5044 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
8c7d9fe0 OS |
5045 | ret = put_file_data(sctx, offset, len); |
5046 | if (ret < 0) | |
5047 | goto out; | |
31db9f7c AB |
5048 | |
5049 | ret = send_cmd(sctx); | |
5050 | ||
5051 | tlv_put_failure: | |
5052 | out: | |
924794c9 | 5053 | fs_path_free(p); |
a9b2e0de | 5054 | return ret; |
31db9f7c AB |
5055 | } |
5056 | ||
5057 | /* | |
5058 | * Send a clone command to user space. | |
5059 | */ | |
5060 | static int send_clone(struct send_ctx *sctx, | |
5061 | u64 offset, u32 len, | |
5062 | struct clone_root *clone_root) | |
5063 | { | |
5064 | int ret = 0; | |
31db9f7c AB |
5065 | struct fs_path *p; |
5066 | u64 gen; | |
5067 | ||
04ab956e JM |
5068 | btrfs_debug(sctx->send_root->fs_info, |
5069 | "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu", | |
4fd786e6 MT |
5070 | offset, len, clone_root->root->root_key.objectid, |
5071 | clone_root->ino, clone_root->offset); | |
31db9f7c | 5072 | |
924794c9 | 5073 | p = fs_path_alloc(); |
31db9f7c AB |
5074 | if (!p) |
5075 | return -ENOMEM; | |
5076 | ||
5077 | ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); | |
5078 | if (ret < 0) | |
5079 | goto out; | |
5080 | ||
5081 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
5082 | if (ret < 0) | |
5083 | goto out; | |
5084 | ||
5085 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5086 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); | |
5087 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5088 | ||
e938c8ad | 5089 | if (clone_root->root == sctx->send_root) { |
31db9f7c | 5090 | ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, |
85a7b33b | 5091 | &gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
5092 | if (ret < 0) |
5093 | goto out; | |
5094 | ret = get_cur_path(sctx, clone_root->ino, gen, p); | |
5095 | } else { | |
924794c9 | 5096 | ret = get_inode_path(clone_root->root, clone_root->ino, p); |
31db9f7c AB |
5097 | } |
5098 | if (ret < 0) | |
5099 | goto out; | |
5100 | ||
37b8d27d JB |
5101 | /* |
5102 | * If the parent we're using has a received_uuid set then use that as | |
5103 | * our clone source as that is what we will look for when doing a | |
5104 | * receive. | |
5105 | * | |
5106 | * This covers the case that we create a snapshot off of a received | |
5107 | * subvolume and then use that as the parent and try to receive on a | |
5108 | * different host. | |
5109 | */ | |
5110 | if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid)) | |
5111 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
5112 | clone_root->root->root_item.received_uuid); | |
5113 | else | |
5114 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
5115 | clone_root->root->root_item.uuid); | |
31db9f7c | 5116 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
09e3a288 | 5117 | btrfs_root_ctransid(&clone_root->root->root_item)); |
31db9f7c AB |
5118 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); |
5119 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, | |
5120 | clone_root->offset); | |
5121 | ||
5122 | ret = send_cmd(sctx); | |
5123 | ||
5124 | tlv_put_failure: | |
5125 | out: | |
924794c9 | 5126 | fs_path_free(p); |
31db9f7c AB |
5127 | return ret; |
5128 | } | |
5129 | ||
cb95e7bf MF |
5130 | /* |
5131 | * Send an update extent command to user space. | |
5132 | */ | |
5133 | static int send_update_extent(struct send_ctx *sctx, | |
5134 | u64 offset, u32 len) | |
5135 | { | |
5136 | int ret = 0; | |
5137 | struct fs_path *p; | |
5138 | ||
924794c9 | 5139 | p = fs_path_alloc(); |
cb95e7bf MF |
5140 | if (!p) |
5141 | return -ENOMEM; | |
5142 | ||
5143 | ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); | |
5144 | if (ret < 0) | |
5145 | goto out; | |
5146 | ||
5147 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
5148 | if (ret < 0) | |
5149 | goto out; | |
5150 | ||
5151 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5152 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5153 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); | |
5154 | ||
5155 | ret = send_cmd(sctx); | |
5156 | ||
5157 | tlv_put_failure: | |
5158 | out: | |
924794c9 | 5159 | fs_path_free(p); |
cb95e7bf MF |
5160 | return ret; |
5161 | } | |
5162 | ||
16e7549f JB |
5163 | static int send_hole(struct send_ctx *sctx, u64 end) |
5164 | { | |
5165 | struct fs_path *p = NULL; | |
8c7d9fe0 | 5166 | u64 read_size = max_send_read_size(sctx); |
16e7549f | 5167 | u64 offset = sctx->cur_inode_last_extent; |
16e7549f JB |
5168 | int ret = 0; |
5169 | ||
22d3151c FM |
5170 | /* |
5171 | * A hole that starts at EOF or beyond it. Since we do not yet support | |
5172 | * fallocate (for extent preallocation and hole punching), sending a | |
5173 | * write of zeroes starting at EOF or beyond would later require issuing | |
5174 | * a truncate operation which would undo the write and achieve nothing. | |
5175 | */ | |
5176 | if (offset >= sctx->cur_inode_size) | |
5177 | return 0; | |
5178 | ||
6b1f72e5 FM |
5179 | /* |
5180 | * Don't go beyond the inode's i_size due to prealloc extents that start | |
5181 | * after the i_size. | |
5182 | */ | |
5183 | end = min_t(u64, end, sctx->cur_inode_size); | |
5184 | ||
d4dfc0f4 FM |
5185 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) |
5186 | return send_update_extent(sctx, offset, end - offset); | |
5187 | ||
16e7549f JB |
5188 | p = fs_path_alloc(); |
5189 | if (!p) | |
5190 | return -ENOMEM; | |
c715e155 FM |
5191 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); |
5192 | if (ret < 0) | |
5193 | goto tlv_put_failure; | |
16e7549f | 5194 | while (offset < end) { |
8c7d9fe0 | 5195 | u64 len = min(end - offset, read_size); |
16e7549f JB |
5196 | |
5197 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
16e7549f JB |
5198 | if (ret < 0) |
5199 | break; | |
5200 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5201 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
8c7d9fe0 OS |
5202 | ret = put_data_header(sctx, len); |
5203 | if (ret < 0) | |
5204 | break; | |
5205 | memset(sctx->send_buf + sctx->send_size, 0, len); | |
5206 | sctx->send_size += len; | |
16e7549f JB |
5207 | ret = send_cmd(sctx); |
5208 | if (ret < 0) | |
5209 | break; | |
5210 | offset += len; | |
5211 | } | |
ffa7c429 | 5212 | sctx->cur_inode_next_write_offset = offset; |
16e7549f JB |
5213 | tlv_put_failure: |
5214 | fs_path_free(p); | |
5215 | return ret; | |
5216 | } | |
5217 | ||
d906d49f FM |
5218 | static int send_extent_data(struct send_ctx *sctx, |
5219 | const u64 offset, | |
5220 | const u64 len) | |
5221 | { | |
8c7d9fe0 | 5222 | u64 read_size = max_send_read_size(sctx); |
d906d49f FM |
5223 | u64 sent = 0; |
5224 | ||
5225 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) | |
5226 | return send_update_extent(sctx, offset, len); | |
5227 | ||
5228 | while (sent < len) { | |
8c7d9fe0 | 5229 | u64 size = min(len - sent, read_size); |
d906d49f FM |
5230 | int ret; |
5231 | ||
d906d49f FM |
5232 | ret = send_write(sctx, offset + sent, size); |
5233 | if (ret < 0) | |
5234 | return ret; | |
a9b2e0de | 5235 | sent += size; |
d906d49f FM |
5236 | } |
5237 | return 0; | |
5238 | } | |
5239 | ||
89efda52 MPS |
5240 | /* |
5241 | * Search for a capability xattr related to sctx->cur_ino. If the capability is | |
5242 | * found, call send_set_xattr function to emit it. | |
5243 | * | |
5244 | * Return 0 if there isn't a capability, or when the capability was emitted | |
5245 | * successfully, or < 0 if an error occurred. | |
5246 | */ | |
5247 | static int send_capabilities(struct send_ctx *sctx) | |
5248 | { | |
5249 | struct fs_path *fspath = NULL; | |
5250 | struct btrfs_path *path; | |
5251 | struct btrfs_dir_item *di; | |
5252 | struct extent_buffer *leaf; | |
5253 | unsigned long data_ptr; | |
5254 | char *buf = NULL; | |
5255 | int buf_len; | |
5256 | int ret = 0; | |
5257 | ||
5258 | path = alloc_path_for_send(); | |
5259 | if (!path) | |
5260 | return -ENOMEM; | |
5261 | ||
5262 | di = btrfs_lookup_xattr(NULL, sctx->send_root, path, sctx->cur_ino, | |
5263 | XATTR_NAME_CAPS, strlen(XATTR_NAME_CAPS), 0); | |
5264 | if (!di) { | |
5265 | /* There is no xattr for this inode */ | |
5266 | goto out; | |
5267 | } else if (IS_ERR(di)) { | |
5268 | ret = PTR_ERR(di); | |
5269 | goto out; | |
5270 | } | |
5271 | ||
5272 | leaf = path->nodes[0]; | |
5273 | buf_len = btrfs_dir_data_len(leaf, di); | |
5274 | ||
5275 | fspath = fs_path_alloc(); | |
5276 | buf = kmalloc(buf_len, GFP_KERNEL); | |
5277 | if (!fspath || !buf) { | |
5278 | ret = -ENOMEM; | |
5279 | goto out; | |
5280 | } | |
5281 | ||
5282 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath); | |
5283 | if (ret < 0) | |
5284 | goto out; | |
5285 | ||
5286 | data_ptr = (unsigned long)(di + 1) + btrfs_dir_name_len(leaf, di); | |
5287 | read_extent_buffer(leaf, buf, data_ptr, buf_len); | |
5288 | ||
5289 | ret = send_set_xattr(sctx, fspath, XATTR_NAME_CAPS, | |
5290 | strlen(XATTR_NAME_CAPS), buf, buf_len); | |
5291 | out: | |
5292 | kfree(buf); | |
5293 | fs_path_free(fspath); | |
5294 | btrfs_free_path(path); | |
5295 | return ret; | |
5296 | } | |
5297 | ||
d906d49f FM |
5298 | static int clone_range(struct send_ctx *sctx, |
5299 | struct clone_root *clone_root, | |
5300 | const u64 disk_byte, | |
5301 | u64 data_offset, | |
5302 | u64 offset, | |
5303 | u64 len) | |
5304 | { | |
5305 | struct btrfs_path *path; | |
5306 | struct btrfs_key key; | |
5307 | int ret; | |
431d3988 | 5308 | u64 clone_src_i_size = 0; |
d906d49f | 5309 | |
72610b1b FM |
5310 | /* |
5311 | * Prevent cloning from a zero offset with a length matching the sector | |
5312 | * size because in some scenarios this will make the receiver fail. | |
5313 | * | |
5314 | * For example, if in the source filesystem the extent at offset 0 | |
5315 | * has a length of sectorsize and it was written using direct IO, then | |
5316 | * it can never be an inline extent (even if compression is enabled). | |
5317 | * Then this extent can be cloned in the original filesystem to a non | |
5318 | * zero file offset, but it may not be possible to clone in the | |
5319 | * destination filesystem because it can be inlined due to compression | |
5320 | * on the destination filesystem (as the receiver's write operations are | |
5321 | * always done using buffered IO). The same happens when the original | |
5322 | * filesystem does not have compression enabled but the destination | |
5323 | * filesystem has. | |
5324 | */ | |
5325 | if (clone_root->offset == 0 && | |
5326 | len == sctx->send_root->fs_info->sectorsize) | |
5327 | return send_extent_data(sctx, offset, len); | |
5328 | ||
d906d49f FM |
5329 | path = alloc_path_for_send(); |
5330 | if (!path) | |
5331 | return -ENOMEM; | |
5332 | ||
040ee612 RK |
5333 | /* |
5334 | * There are inodes that have extents that lie behind its i_size. Don't | |
5335 | * accept clones from these extents. | |
5336 | */ | |
5337 | ret = __get_inode_info(clone_root->root, path, clone_root->ino, | |
5338 | &clone_src_i_size, NULL, NULL, NULL, NULL, NULL); | |
5339 | btrfs_release_path(path); | |
5340 | if (ret < 0) | |
5341 | goto out; | |
5342 | ||
d906d49f FM |
5343 | /* |
5344 | * We can't send a clone operation for the entire range if we find | |
5345 | * extent items in the respective range in the source file that | |
5346 | * refer to different extents or if we find holes. | |
5347 | * So check for that and do a mix of clone and regular write/copy | |
5348 | * operations if needed. | |
5349 | * | |
5350 | * Example: | |
5351 | * | |
5352 | * mkfs.btrfs -f /dev/sda | |
5353 | * mount /dev/sda /mnt | |
5354 | * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo | |
5355 | * cp --reflink=always /mnt/foo /mnt/bar | |
5356 | * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo | |
5357 | * btrfs subvolume snapshot -r /mnt /mnt/snap | |
5358 | * | |
5359 | * If when we send the snapshot and we are processing file bar (which | |
5360 | * has a higher inode number than foo) we blindly send a clone operation | |
5361 | * for the [0, 100K[ range from foo to bar, the receiver ends up getting | |
5362 | * a file bar that matches the content of file foo - iow, doesn't match | |
5363 | * the content from bar in the original filesystem. | |
5364 | */ | |
5365 | key.objectid = clone_root->ino; | |
5366 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5367 | key.offset = clone_root->offset; | |
5368 | ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); | |
5369 | if (ret < 0) | |
5370 | goto out; | |
5371 | if (ret > 0 && path->slots[0] > 0) { | |
5372 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
5373 | if (key.objectid == clone_root->ino && | |
5374 | key.type == BTRFS_EXTENT_DATA_KEY) | |
5375 | path->slots[0]--; | |
5376 | } | |
5377 | ||
5378 | while (true) { | |
5379 | struct extent_buffer *leaf = path->nodes[0]; | |
5380 | int slot = path->slots[0]; | |
5381 | struct btrfs_file_extent_item *ei; | |
5382 | u8 type; | |
5383 | u64 ext_len; | |
5384 | u64 clone_len; | |
040ee612 | 5385 | u64 clone_data_offset; |
d906d49f FM |
5386 | |
5387 | if (slot >= btrfs_header_nritems(leaf)) { | |
5388 | ret = btrfs_next_leaf(clone_root->root, path); | |
5389 | if (ret < 0) | |
5390 | goto out; | |
5391 | else if (ret > 0) | |
5392 | break; | |
5393 | continue; | |
5394 | } | |
5395 | ||
5396 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5397 | ||
5398 | /* | |
5399 | * We might have an implicit trailing hole (NO_HOLES feature | |
5400 | * enabled). We deal with it after leaving this loop. | |
5401 | */ | |
5402 | if (key.objectid != clone_root->ino || | |
5403 | key.type != BTRFS_EXTENT_DATA_KEY) | |
5404 | break; | |
5405 | ||
5406 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5407 | type = btrfs_file_extent_type(leaf, ei); | |
5408 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 5409 | ext_len = btrfs_file_extent_ram_bytes(leaf, ei); |
09cbfeaf | 5410 | ext_len = PAGE_ALIGN(ext_len); |
d906d49f FM |
5411 | } else { |
5412 | ext_len = btrfs_file_extent_num_bytes(leaf, ei); | |
5413 | } | |
5414 | ||
5415 | if (key.offset + ext_len <= clone_root->offset) | |
5416 | goto next; | |
5417 | ||
5418 | if (key.offset > clone_root->offset) { | |
5419 | /* Implicit hole, NO_HOLES feature enabled. */ | |
5420 | u64 hole_len = key.offset - clone_root->offset; | |
5421 | ||
5422 | if (hole_len > len) | |
5423 | hole_len = len; | |
5424 | ret = send_extent_data(sctx, offset, hole_len); | |
5425 | if (ret < 0) | |
5426 | goto out; | |
5427 | ||
5428 | len -= hole_len; | |
5429 | if (len == 0) | |
5430 | break; | |
5431 | offset += hole_len; | |
5432 | clone_root->offset += hole_len; | |
5433 | data_offset += hole_len; | |
5434 | } | |
5435 | ||
5436 | if (key.offset >= clone_root->offset + len) | |
5437 | break; | |
5438 | ||
040ee612 RK |
5439 | if (key.offset >= clone_src_i_size) |
5440 | break; | |
5441 | ||
5442 | if (key.offset + ext_len > clone_src_i_size) | |
5443 | ext_len = clone_src_i_size - key.offset; | |
5444 | ||
5445 | clone_data_offset = btrfs_file_extent_offset(leaf, ei); | |
5446 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) { | |
5447 | clone_root->offset = key.offset; | |
5448 | if (clone_data_offset < data_offset && | |
5449 | clone_data_offset + ext_len > data_offset) { | |
5450 | u64 extent_offset; | |
5451 | ||
5452 | extent_offset = data_offset - clone_data_offset; | |
5453 | ext_len -= extent_offset; | |
5454 | clone_data_offset += extent_offset; | |
5455 | clone_root->offset += extent_offset; | |
5456 | } | |
5457 | } | |
5458 | ||
d906d49f FM |
5459 | clone_len = min_t(u64, ext_len, len); |
5460 | ||
5461 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && | |
3c850b45 FM |
5462 | clone_data_offset == data_offset) { |
5463 | const u64 src_end = clone_root->offset + clone_len; | |
5464 | const u64 sectorsize = SZ_64K; | |
5465 | ||
5466 | /* | |
5467 | * We can't clone the last block, when its size is not | |
5468 | * sector size aligned, into the middle of a file. If we | |
5469 | * do so, the receiver will get a failure (-EINVAL) when | |
5470 | * trying to clone or will silently corrupt the data in | |
5471 | * the destination file if it's on a kernel without the | |
5472 | * fix introduced by commit ac765f83f1397646 | |
5473 | * ("Btrfs: fix data corruption due to cloning of eof | |
5474 | * block). | |
5475 | * | |
5476 | * So issue a clone of the aligned down range plus a | |
5477 | * regular write for the eof block, if we hit that case. | |
5478 | * | |
5479 | * Also, we use the maximum possible sector size, 64K, | |
5480 | * because we don't know what's the sector size of the | |
5481 | * filesystem that receives the stream, so we have to | |
5482 | * assume the largest possible sector size. | |
5483 | */ | |
5484 | if (src_end == clone_src_i_size && | |
5485 | !IS_ALIGNED(src_end, sectorsize) && | |
5486 | offset + clone_len < sctx->cur_inode_size) { | |
5487 | u64 slen; | |
5488 | ||
5489 | slen = ALIGN_DOWN(src_end - clone_root->offset, | |
5490 | sectorsize); | |
5491 | if (slen > 0) { | |
5492 | ret = send_clone(sctx, offset, slen, | |
5493 | clone_root); | |
5494 | if (ret < 0) | |
5495 | goto out; | |
5496 | } | |
5497 | ret = send_extent_data(sctx, offset + slen, | |
5498 | clone_len - slen); | |
5499 | } else { | |
5500 | ret = send_clone(sctx, offset, clone_len, | |
5501 | clone_root); | |
5502 | } | |
5503 | } else { | |
d906d49f | 5504 | ret = send_extent_data(sctx, offset, clone_len); |
3c850b45 | 5505 | } |
d906d49f FM |
5506 | |
5507 | if (ret < 0) | |
5508 | goto out; | |
5509 | ||
5510 | len -= clone_len; | |
5511 | if (len == 0) | |
5512 | break; | |
5513 | offset += clone_len; | |
5514 | clone_root->offset += clone_len; | |
5515 | data_offset += clone_len; | |
5516 | next: | |
5517 | path->slots[0]++; | |
5518 | } | |
5519 | ||
5520 | if (len > 0) | |
5521 | ret = send_extent_data(sctx, offset, len); | |
5522 | else | |
5523 | ret = 0; | |
5524 | out: | |
5525 | btrfs_free_path(path); | |
5526 | return ret; | |
5527 | } | |
5528 | ||
31db9f7c AB |
5529 | static int send_write_or_clone(struct send_ctx *sctx, |
5530 | struct btrfs_path *path, | |
5531 | struct btrfs_key *key, | |
5532 | struct clone_root *clone_root) | |
5533 | { | |
5534 | int ret = 0; | |
31db9f7c | 5535 | u64 offset = key->offset; |
c9a949af | 5536 | u64 end; |
28e5dd8f | 5537 | u64 bs = sctx->send_root->fs_info->sb->s_blocksize; |
31db9f7c | 5538 | |
c9a949af OS |
5539 | end = min_t(u64, btrfs_file_extent_end(path), sctx->cur_inode_size); |
5540 | if (offset >= end) | |
5541 | return 0; | |
31db9f7c | 5542 | |
c9a949af OS |
5543 | if (clone_root && IS_ALIGNED(end, bs)) { |
5544 | struct btrfs_file_extent_item *ei; | |
d906d49f FM |
5545 | u64 disk_byte; |
5546 | u64 data_offset; | |
5547 | ||
c9a949af OS |
5548 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], |
5549 | struct btrfs_file_extent_item); | |
d906d49f FM |
5550 | disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); |
5551 | data_offset = btrfs_file_extent_offset(path->nodes[0], ei); | |
5552 | ret = clone_range(sctx, clone_root, disk_byte, data_offset, | |
c9a949af | 5553 | offset, end - offset); |
cb95e7bf | 5554 | } else { |
c9a949af | 5555 | ret = send_extent_data(sctx, offset, end - offset); |
31db9f7c | 5556 | } |
c9a949af | 5557 | sctx->cur_inode_next_write_offset = end; |
31db9f7c AB |
5558 | return ret; |
5559 | } | |
5560 | ||
5561 | static int is_extent_unchanged(struct send_ctx *sctx, | |
5562 | struct btrfs_path *left_path, | |
5563 | struct btrfs_key *ekey) | |
5564 | { | |
5565 | int ret = 0; | |
5566 | struct btrfs_key key; | |
5567 | struct btrfs_path *path = NULL; | |
5568 | struct extent_buffer *eb; | |
5569 | int slot; | |
5570 | struct btrfs_key found_key; | |
5571 | struct btrfs_file_extent_item *ei; | |
5572 | u64 left_disknr; | |
5573 | u64 right_disknr; | |
5574 | u64 left_offset; | |
5575 | u64 right_offset; | |
5576 | u64 left_offset_fixed; | |
5577 | u64 left_len; | |
5578 | u64 right_len; | |
74dd17fb CM |
5579 | u64 left_gen; |
5580 | u64 right_gen; | |
31db9f7c AB |
5581 | u8 left_type; |
5582 | u8 right_type; | |
5583 | ||
5584 | path = alloc_path_for_send(); | |
5585 | if (!path) | |
5586 | return -ENOMEM; | |
5587 | ||
5588 | eb = left_path->nodes[0]; | |
5589 | slot = left_path->slots[0]; | |
31db9f7c AB |
5590 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
5591 | left_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
5592 | |
5593 | if (left_type != BTRFS_FILE_EXTENT_REG) { | |
5594 | ret = 0; | |
5595 | goto out; | |
5596 | } | |
74dd17fb CM |
5597 | left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5598 | left_len = btrfs_file_extent_num_bytes(eb, ei); | |
5599 | left_offset = btrfs_file_extent_offset(eb, ei); | |
5600 | left_gen = btrfs_file_extent_generation(eb, ei); | |
31db9f7c AB |
5601 | |
5602 | /* | |
5603 | * Following comments will refer to these graphics. L is the left | |
5604 | * extents which we are checking at the moment. 1-8 are the right | |
5605 | * extents that we iterate. | |
5606 | * | |
5607 | * |-----L-----| | |
5608 | * |-1-|-2a-|-3-|-4-|-5-|-6-| | |
5609 | * | |
5610 | * |-----L-----| | |
5611 | * |--1--|-2b-|...(same as above) | |
5612 | * | |
5613 | * Alternative situation. Happens on files where extents got split. | |
5614 | * |-----L-----| | |
5615 | * |-----------7-----------|-6-| | |
5616 | * | |
5617 | * Alternative situation. Happens on files which got larger. | |
5618 | * |-----L-----| | |
5619 | * |-8-| | |
5620 | * Nothing follows after 8. | |
5621 | */ | |
5622 | ||
5623 | key.objectid = ekey->objectid; | |
5624 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5625 | key.offset = ekey->offset; | |
5626 | ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); | |
5627 | if (ret < 0) | |
5628 | goto out; | |
5629 | if (ret) { | |
5630 | ret = 0; | |
5631 | goto out; | |
5632 | } | |
5633 | ||
5634 | /* | |
5635 | * Handle special case where the right side has no extents at all. | |
5636 | */ | |
5637 | eb = path->nodes[0]; | |
5638 | slot = path->slots[0]; | |
5639 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5640 | if (found_key.objectid != key.objectid || | |
5641 | found_key.type != key.type) { | |
57cfd462 JB |
5642 | /* If we're a hole then just pretend nothing changed */ |
5643 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5644 | goto out; |
5645 | } | |
5646 | ||
5647 | /* | |
5648 | * We're now on 2a, 2b or 7. | |
5649 | */ | |
5650 | key = found_key; | |
5651 | while (key.offset < ekey->offset + left_len) { | |
5652 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
5653 | right_type = btrfs_file_extent_type(eb, ei); | |
e1cbfd7b FM |
5654 | if (right_type != BTRFS_FILE_EXTENT_REG && |
5655 | right_type != BTRFS_FILE_EXTENT_INLINE) { | |
31db9f7c AB |
5656 | ret = 0; |
5657 | goto out; | |
5658 | } | |
5659 | ||
e1cbfd7b | 5660 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 | 5661 | right_len = btrfs_file_extent_ram_bytes(eb, ei); |
e1cbfd7b FM |
5662 | right_len = PAGE_ALIGN(right_len); |
5663 | } else { | |
5664 | right_len = btrfs_file_extent_num_bytes(eb, ei); | |
5665 | } | |
007d31f7 | 5666 | |
31db9f7c AB |
5667 | /* |
5668 | * Are we at extent 8? If yes, we know the extent is changed. | |
5669 | * This may only happen on the first iteration. | |
5670 | */ | |
d8347fa4 | 5671 | if (found_key.offset + right_len <= ekey->offset) { |
57cfd462 JB |
5672 | /* If we're a hole just pretend nothing changed */ |
5673 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5674 | goto out; |
5675 | } | |
5676 | ||
e1cbfd7b FM |
5677 | /* |
5678 | * We just wanted to see if when we have an inline extent, what | |
5679 | * follows it is a regular extent (wanted to check the above | |
5680 | * condition for inline extents too). This should normally not | |
5681 | * happen but it's possible for example when we have an inline | |
5682 | * compressed extent representing data with a size matching | |
5683 | * the page size (currently the same as sector size). | |
5684 | */ | |
5685 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { | |
5686 | ret = 0; | |
5687 | goto out; | |
5688 | } | |
5689 | ||
24e52b11 FM |
5690 | right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5691 | right_offset = btrfs_file_extent_offset(eb, ei); | |
5692 | right_gen = btrfs_file_extent_generation(eb, ei); | |
5693 | ||
31db9f7c AB |
5694 | left_offset_fixed = left_offset; |
5695 | if (key.offset < ekey->offset) { | |
5696 | /* Fix the right offset for 2a and 7. */ | |
5697 | right_offset += ekey->offset - key.offset; | |
5698 | } else { | |
5699 | /* Fix the left offset for all behind 2a and 2b */ | |
5700 | left_offset_fixed += key.offset - ekey->offset; | |
5701 | } | |
5702 | ||
5703 | /* | |
5704 | * Check if we have the same extent. | |
5705 | */ | |
3954096d | 5706 | if (left_disknr != right_disknr || |
74dd17fb CM |
5707 | left_offset_fixed != right_offset || |
5708 | left_gen != right_gen) { | |
31db9f7c AB |
5709 | ret = 0; |
5710 | goto out; | |
5711 | } | |
5712 | ||
5713 | /* | |
5714 | * Go to the next extent. | |
5715 | */ | |
5716 | ret = btrfs_next_item(sctx->parent_root, path); | |
5717 | if (ret < 0) | |
5718 | goto out; | |
5719 | if (!ret) { | |
5720 | eb = path->nodes[0]; | |
5721 | slot = path->slots[0]; | |
5722 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5723 | } | |
5724 | if (ret || found_key.objectid != key.objectid || | |
5725 | found_key.type != key.type) { | |
5726 | key.offset += right_len; | |
5727 | break; | |
adaa4b8e JS |
5728 | } |
5729 | if (found_key.offset != key.offset + right_len) { | |
5730 | ret = 0; | |
5731 | goto out; | |
31db9f7c AB |
5732 | } |
5733 | key = found_key; | |
5734 | } | |
5735 | ||
5736 | /* | |
5737 | * We're now behind the left extent (treat as unchanged) or at the end | |
5738 | * of the right side (treat as changed). | |
5739 | */ | |
5740 | if (key.offset >= ekey->offset + left_len) | |
5741 | ret = 1; | |
5742 | else | |
5743 | ret = 0; | |
5744 | ||
5745 | ||
5746 | out: | |
5747 | btrfs_free_path(path); | |
5748 | return ret; | |
5749 | } | |
5750 | ||
16e7549f JB |
5751 | static int get_last_extent(struct send_ctx *sctx, u64 offset) |
5752 | { | |
5753 | struct btrfs_path *path; | |
5754 | struct btrfs_root *root = sctx->send_root; | |
16e7549f | 5755 | struct btrfs_key key; |
16e7549f JB |
5756 | int ret; |
5757 | ||
5758 | path = alloc_path_for_send(); | |
5759 | if (!path) | |
5760 | return -ENOMEM; | |
5761 | ||
5762 | sctx->cur_inode_last_extent = 0; | |
5763 | ||
5764 | key.objectid = sctx->cur_ino; | |
5765 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5766 | key.offset = offset; | |
5767 | ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); | |
5768 | if (ret < 0) | |
5769 | goto out; | |
5770 | ret = 0; | |
5771 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
5772 | if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
5773 | goto out; | |
5774 | ||
a5eeb3d1 | 5775 | sctx->cur_inode_last_extent = btrfs_file_extent_end(path); |
16e7549f JB |
5776 | out: |
5777 | btrfs_free_path(path); | |
5778 | return ret; | |
5779 | } | |
5780 | ||
82bfb2e7 FM |
5781 | static int range_is_hole_in_parent(struct send_ctx *sctx, |
5782 | const u64 start, | |
5783 | const u64 end) | |
5784 | { | |
5785 | struct btrfs_path *path; | |
5786 | struct btrfs_key key; | |
5787 | struct btrfs_root *root = sctx->parent_root; | |
5788 | u64 search_start = start; | |
5789 | int ret; | |
5790 | ||
5791 | path = alloc_path_for_send(); | |
5792 | if (!path) | |
5793 | return -ENOMEM; | |
5794 | ||
5795 | key.objectid = sctx->cur_ino; | |
5796 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5797 | key.offset = search_start; | |
5798 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5799 | if (ret < 0) | |
5800 | goto out; | |
5801 | if (ret > 0 && path->slots[0] > 0) | |
5802 | path->slots[0]--; | |
5803 | ||
5804 | while (search_start < end) { | |
5805 | struct extent_buffer *leaf = path->nodes[0]; | |
5806 | int slot = path->slots[0]; | |
5807 | struct btrfs_file_extent_item *fi; | |
5808 | u64 extent_end; | |
5809 | ||
5810 | if (slot >= btrfs_header_nritems(leaf)) { | |
5811 | ret = btrfs_next_leaf(root, path); | |
5812 | if (ret < 0) | |
5813 | goto out; | |
5814 | else if (ret > 0) | |
5815 | break; | |
5816 | continue; | |
5817 | } | |
5818 | ||
5819 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5820 | if (key.objectid < sctx->cur_ino || | |
5821 | key.type < BTRFS_EXTENT_DATA_KEY) | |
5822 | goto next; | |
5823 | if (key.objectid > sctx->cur_ino || | |
5824 | key.type > BTRFS_EXTENT_DATA_KEY || | |
5825 | key.offset >= end) | |
5826 | break; | |
5827 | ||
5828 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
a5eeb3d1 | 5829 | extent_end = btrfs_file_extent_end(path); |
82bfb2e7 FM |
5830 | if (extent_end <= start) |
5831 | goto next; | |
5832 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) { | |
5833 | search_start = extent_end; | |
5834 | goto next; | |
5835 | } | |
5836 | ret = 0; | |
5837 | goto out; | |
5838 | next: | |
5839 | path->slots[0]++; | |
5840 | } | |
5841 | ret = 1; | |
5842 | out: | |
5843 | btrfs_free_path(path); | |
5844 | return ret; | |
5845 | } | |
5846 | ||
16e7549f JB |
5847 | static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, |
5848 | struct btrfs_key *key) | |
5849 | { | |
16e7549f JB |
5850 | int ret = 0; |
5851 | ||
5852 | if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) | |
5853 | return 0; | |
5854 | ||
5855 | if (sctx->cur_inode_last_extent == (u64)-1) { | |
5856 | ret = get_last_extent(sctx, key->offset - 1); | |
5857 | if (ret) | |
5858 | return ret; | |
5859 | } | |
5860 | ||
bf54f412 FDBM |
5861 | if (path->slots[0] == 0 && |
5862 | sctx->cur_inode_last_extent < key->offset) { | |
5863 | /* | |
5864 | * We might have skipped entire leafs that contained only | |
5865 | * file extent items for our current inode. These leafs have | |
5866 | * a generation number smaller (older) than the one in the | |
5867 | * current leaf and the leaf our last extent came from, and | |
5868 | * are located between these 2 leafs. | |
5869 | */ | |
5870 | ret = get_last_extent(sctx, key->offset - 1); | |
5871 | if (ret) | |
5872 | return ret; | |
5873 | } | |
5874 | ||
82bfb2e7 FM |
5875 | if (sctx->cur_inode_last_extent < key->offset) { |
5876 | ret = range_is_hole_in_parent(sctx, | |
5877 | sctx->cur_inode_last_extent, | |
5878 | key->offset); | |
5879 | if (ret < 0) | |
5880 | return ret; | |
5881 | else if (ret == 0) | |
5882 | ret = send_hole(sctx, key->offset); | |
5883 | else | |
5884 | ret = 0; | |
5885 | } | |
a5eeb3d1 | 5886 | sctx->cur_inode_last_extent = btrfs_file_extent_end(path); |
16e7549f JB |
5887 | return ret; |
5888 | } | |
5889 | ||
31db9f7c AB |
5890 | static int process_extent(struct send_ctx *sctx, |
5891 | struct btrfs_path *path, | |
5892 | struct btrfs_key *key) | |
5893 | { | |
31db9f7c | 5894 | struct clone_root *found_clone = NULL; |
57cfd462 | 5895 | int ret = 0; |
31db9f7c AB |
5896 | |
5897 | if (S_ISLNK(sctx->cur_inode_mode)) | |
5898 | return 0; | |
5899 | ||
5900 | if (sctx->parent_root && !sctx->cur_inode_new) { | |
5901 | ret = is_extent_unchanged(sctx, path, key); | |
5902 | if (ret < 0) | |
5903 | goto out; | |
5904 | if (ret) { | |
5905 | ret = 0; | |
16e7549f | 5906 | goto out_hole; |
31db9f7c | 5907 | } |
57cfd462 JB |
5908 | } else { |
5909 | struct btrfs_file_extent_item *ei; | |
5910 | u8 type; | |
5911 | ||
5912 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5913 | struct btrfs_file_extent_item); | |
5914 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
5915 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
5916 | type == BTRFS_FILE_EXTENT_REG) { | |
5917 | /* | |
5918 | * The send spec does not have a prealloc command yet, | |
5919 | * so just leave a hole for prealloc'ed extents until | |
5920 | * we have enough commands queued up to justify rev'ing | |
5921 | * the send spec. | |
5922 | */ | |
5923 | if (type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5924 | ret = 0; | |
5925 | goto out; | |
5926 | } | |
5927 | ||
5928 | /* Have a hole, just skip it. */ | |
5929 | if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { | |
5930 | ret = 0; | |
5931 | goto out; | |
5932 | } | |
5933 | } | |
31db9f7c AB |
5934 | } |
5935 | ||
5936 | ret = find_extent_clone(sctx, path, key->objectid, key->offset, | |
5937 | sctx->cur_inode_size, &found_clone); | |
5938 | if (ret != -ENOENT && ret < 0) | |
5939 | goto out; | |
5940 | ||
5941 | ret = send_write_or_clone(sctx, path, key, found_clone); | |
16e7549f JB |
5942 | if (ret) |
5943 | goto out; | |
5944 | out_hole: | |
5945 | ret = maybe_send_hole(sctx, path, key); | |
31db9f7c AB |
5946 | out: |
5947 | return ret; | |
5948 | } | |
5949 | ||
5950 | static int process_all_extents(struct send_ctx *sctx) | |
5951 | { | |
5952 | int ret; | |
5953 | struct btrfs_root *root; | |
5954 | struct btrfs_path *path; | |
5955 | struct btrfs_key key; | |
5956 | struct btrfs_key found_key; | |
5957 | struct extent_buffer *eb; | |
5958 | int slot; | |
5959 | ||
5960 | root = sctx->send_root; | |
5961 | path = alloc_path_for_send(); | |
5962 | if (!path) | |
5963 | return -ENOMEM; | |
5964 | ||
5965 | key.objectid = sctx->cmp_key->objectid; | |
5966 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5967 | key.offset = 0; | |
7fdd29d0 FDBM |
5968 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5969 | if (ret < 0) | |
5970 | goto out; | |
31db9f7c | 5971 | |
7fdd29d0 | 5972 | while (1) { |
31db9f7c AB |
5973 | eb = path->nodes[0]; |
5974 | slot = path->slots[0]; | |
7fdd29d0 FDBM |
5975 | |
5976 | if (slot >= btrfs_header_nritems(eb)) { | |
5977 | ret = btrfs_next_leaf(root, path); | |
5978 | if (ret < 0) { | |
5979 | goto out; | |
5980 | } else if (ret > 0) { | |
5981 | ret = 0; | |
5982 | break; | |
5983 | } | |
5984 | continue; | |
5985 | } | |
5986 | ||
31db9f7c AB |
5987 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
5988 | ||
5989 | if (found_key.objectid != key.objectid || | |
5990 | found_key.type != key.type) { | |
5991 | ret = 0; | |
5992 | goto out; | |
5993 | } | |
5994 | ||
5995 | ret = process_extent(sctx, path, &found_key); | |
5996 | if (ret < 0) | |
5997 | goto out; | |
5998 | ||
7fdd29d0 | 5999 | path->slots[0]++; |
31db9f7c AB |
6000 | } |
6001 | ||
6002 | out: | |
6003 | btrfs_free_path(path); | |
6004 | return ret; | |
6005 | } | |
6006 | ||
9f03740a FDBM |
6007 | static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, |
6008 | int *pending_move, | |
6009 | int *refs_processed) | |
31db9f7c AB |
6010 | { |
6011 | int ret = 0; | |
6012 | ||
6013 | if (sctx->cur_ino == 0) | |
6014 | goto out; | |
6015 | if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && | |
96b5bd77 | 6016 | sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) |
31db9f7c AB |
6017 | goto out; |
6018 | if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) | |
6019 | goto out; | |
6020 | ||
9f03740a | 6021 | ret = process_recorded_refs(sctx, pending_move); |
e479d9bb AB |
6022 | if (ret < 0) |
6023 | goto out; | |
6024 | ||
9f03740a | 6025 | *refs_processed = 1; |
31db9f7c AB |
6026 | out: |
6027 | return ret; | |
6028 | } | |
6029 | ||
6030 | static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) | |
6031 | { | |
6032 | int ret = 0; | |
6033 | u64 left_mode; | |
6034 | u64 left_uid; | |
6035 | u64 left_gid; | |
6036 | u64 right_mode; | |
6037 | u64 right_uid; | |
6038 | u64 right_gid; | |
6039 | int need_chmod = 0; | |
6040 | int need_chown = 0; | |
ffa7c429 | 6041 | int need_truncate = 1; |
9f03740a FDBM |
6042 | int pending_move = 0; |
6043 | int refs_processed = 0; | |
31db9f7c | 6044 | |
46b2f459 FM |
6045 | if (sctx->ignore_cur_inode) |
6046 | return 0; | |
6047 | ||
9f03740a FDBM |
6048 | ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, |
6049 | &refs_processed); | |
31db9f7c AB |
6050 | if (ret < 0) |
6051 | goto out; | |
6052 | ||
9f03740a FDBM |
6053 | /* |
6054 | * We have processed the refs and thus need to advance send_progress. | |
6055 | * Now, calls to get_cur_xxx will take the updated refs of the current | |
6056 | * inode into account. | |
6057 | * | |
6058 | * On the other hand, if our current inode is a directory and couldn't | |
6059 | * be moved/renamed because its parent was renamed/moved too and it has | |
6060 | * a higher inode number, we can only move/rename our current inode | |
6061 | * after we moved/renamed its parent. Therefore in this case operate on | |
6062 | * the old path (pre move/rename) of our current inode, and the | |
6063 | * move/rename will be performed later. | |
6064 | */ | |
6065 | if (refs_processed && !pending_move) | |
6066 | sctx->send_progress = sctx->cur_ino + 1; | |
6067 | ||
31db9f7c AB |
6068 | if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) |
6069 | goto out; | |
6070 | if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) | |
6071 | goto out; | |
6072 | ||
6073 | ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, | |
85a7b33b | 6074 | &left_mode, &left_uid, &left_gid, NULL); |
31db9f7c AB |
6075 | if (ret < 0) |
6076 | goto out; | |
6077 | ||
e2d044fe AL |
6078 | if (!sctx->parent_root || sctx->cur_inode_new) { |
6079 | need_chown = 1; | |
6080 | if (!S_ISLNK(sctx->cur_inode_mode)) | |
31db9f7c | 6081 | need_chmod = 1; |
ffa7c429 FM |
6082 | if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size) |
6083 | need_truncate = 0; | |
e2d044fe | 6084 | } else { |
ffa7c429 FM |
6085 | u64 old_size; |
6086 | ||
e2d044fe | 6087 | ret = get_inode_info(sctx->parent_root, sctx->cur_ino, |
ffa7c429 | 6088 | &old_size, NULL, &right_mode, &right_uid, |
e2d044fe AL |
6089 | &right_gid, NULL); |
6090 | if (ret < 0) | |
6091 | goto out; | |
31db9f7c | 6092 | |
e2d044fe AL |
6093 | if (left_uid != right_uid || left_gid != right_gid) |
6094 | need_chown = 1; | |
6095 | if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) | |
6096 | need_chmod = 1; | |
ffa7c429 FM |
6097 | if ((old_size == sctx->cur_inode_size) || |
6098 | (sctx->cur_inode_size > old_size && | |
6099 | sctx->cur_inode_next_write_offset == sctx->cur_inode_size)) | |
6100 | need_truncate = 0; | |
31db9f7c AB |
6101 | } |
6102 | ||
6103 | if (S_ISREG(sctx->cur_inode_mode)) { | |
16e7549f | 6104 | if (need_send_hole(sctx)) { |
766b5e5a FM |
6105 | if (sctx->cur_inode_last_extent == (u64)-1 || |
6106 | sctx->cur_inode_last_extent < | |
6107 | sctx->cur_inode_size) { | |
16e7549f JB |
6108 | ret = get_last_extent(sctx, (u64)-1); |
6109 | if (ret) | |
6110 | goto out; | |
6111 | } | |
6112 | if (sctx->cur_inode_last_extent < | |
6113 | sctx->cur_inode_size) { | |
6114 | ret = send_hole(sctx, sctx->cur_inode_size); | |
6115 | if (ret) | |
6116 | goto out; | |
6117 | } | |
6118 | } | |
ffa7c429 FM |
6119 | if (need_truncate) { |
6120 | ret = send_truncate(sctx, sctx->cur_ino, | |
6121 | sctx->cur_inode_gen, | |
6122 | sctx->cur_inode_size); | |
6123 | if (ret < 0) | |
6124 | goto out; | |
6125 | } | |
31db9f7c AB |
6126 | } |
6127 | ||
6128 | if (need_chown) { | |
6129 | ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
6130 | left_uid, left_gid); | |
6131 | if (ret < 0) | |
6132 | goto out; | |
6133 | } | |
6134 | if (need_chmod) { | |
6135 | ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
6136 | left_mode); | |
6137 | if (ret < 0) | |
6138 | goto out; | |
6139 | } | |
6140 | ||
89efda52 MPS |
6141 | ret = send_capabilities(sctx); |
6142 | if (ret < 0) | |
6143 | goto out; | |
6144 | ||
31db9f7c | 6145 | /* |
9f03740a FDBM |
6146 | * If other directory inodes depended on our current directory |
6147 | * inode's move/rename, now do their move/rename operations. | |
31db9f7c | 6148 | */ |
9f03740a FDBM |
6149 | if (!is_waiting_for_move(sctx, sctx->cur_ino)) { |
6150 | ret = apply_children_dir_moves(sctx); | |
6151 | if (ret) | |
6152 | goto out; | |
fcbd2154 FM |
6153 | /* |
6154 | * Need to send that every time, no matter if it actually | |
6155 | * changed between the two trees as we have done changes to | |
6156 | * the inode before. If our inode is a directory and it's | |
6157 | * waiting to be moved/renamed, we will send its utimes when | |
6158 | * it's moved/renamed, therefore we don't need to do it here. | |
6159 | */ | |
6160 | sctx->send_progress = sctx->cur_ino + 1; | |
6161 | ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
6162 | if (ret < 0) | |
6163 | goto out; | |
9f03740a FDBM |
6164 | } |
6165 | ||
31db9f7c AB |
6166 | out: |
6167 | return ret; | |
6168 | } | |
6169 | ||
46b2f459 FM |
6170 | struct parent_paths_ctx { |
6171 | struct list_head *refs; | |
6172 | struct send_ctx *sctx; | |
6173 | }; | |
6174 | ||
6175 | static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name, | |
6176 | void *ctx) | |
6177 | { | |
6178 | struct parent_paths_ctx *ppctx = ctx; | |
6179 | ||
6180 | return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx, | |
6181 | ppctx->refs); | |
6182 | } | |
6183 | ||
6184 | /* | |
6185 | * Issue unlink operations for all paths of the current inode found in the | |
6186 | * parent snapshot. | |
6187 | */ | |
6188 | static int btrfs_unlink_all_paths(struct send_ctx *sctx) | |
6189 | { | |
6190 | LIST_HEAD(deleted_refs); | |
6191 | struct btrfs_path *path; | |
6192 | struct btrfs_key key; | |
6193 | struct parent_paths_ctx ctx; | |
6194 | int ret; | |
6195 | ||
6196 | path = alloc_path_for_send(); | |
6197 | if (!path) | |
6198 | return -ENOMEM; | |
6199 | ||
6200 | key.objectid = sctx->cur_ino; | |
6201 | key.type = BTRFS_INODE_REF_KEY; | |
6202 | key.offset = 0; | |
6203 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
6204 | if (ret < 0) | |
6205 | goto out; | |
6206 | ||
6207 | ctx.refs = &deleted_refs; | |
6208 | ctx.sctx = sctx; | |
6209 | ||
6210 | while (true) { | |
6211 | struct extent_buffer *eb = path->nodes[0]; | |
6212 | int slot = path->slots[0]; | |
6213 | ||
6214 | if (slot >= btrfs_header_nritems(eb)) { | |
6215 | ret = btrfs_next_leaf(sctx->parent_root, path); | |
6216 | if (ret < 0) | |
6217 | goto out; | |
6218 | else if (ret > 0) | |
6219 | break; | |
6220 | continue; | |
6221 | } | |
6222 | ||
6223 | btrfs_item_key_to_cpu(eb, &key, slot); | |
6224 | if (key.objectid != sctx->cur_ino) | |
6225 | break; | |
6226 | if (key.type != BTRFS_INODE_REF_KEY && | |
6227 | key.type != BTRFS_INODE_EXTREF_KEY) | |
6228 | break; | |
6229 | ||
6230 | ret = iterate_inode_ref(sctx->parent_root, path, &key, 1, | |
6231 | record_parent_ref, &ctx); | |
6232 | if (ret < 0) | |
6233 | goto out; | |
6234 | ||
6235 | path->slots[0]++; | |
6236 | } | |
6237 | ||
6238 | while (!list_empty(&deleted_refs)) { | |
6239 | struct recorded_ref *ref; | |
6240 | ||
6241 | ref = list_first_entry(&deleted_refs, struct recorded_ref, list); | |
6242 | ret = send_unlink(sctx, ref->full_path); | |
6243 | if (ret < 0) | |
6244 | goto out; | |
6245 | fs_path_free(ref->full_path); | |
6246 | list_del(&ref->list); | |
6247 | kfree(ref); | |
6248 | } | |
6249 | ret = 0; | |
6250 | out: | |
6251 | btrfs_free_path(path); | |
6252 | if (ret) | |
6253 | __free_recorded_refs(&deleted_refs); | |
6254 | return ret; | |
6255 | } | |
6256 | ||
31db9f7c AB |
6257 | static int changed_inode(struct send_ctx *sctx, |
6258 | enum btrfs_compare_tree_result result) | |
6259 | { | |
6260 | int ret = 0; | |
6261 | struct btrfs_key *key = sctx->cmp_key; | |
6262 | struct btrfs_inode_item *left_ii = NULL; | |
6263 | struct btrfs_inode_item *right_ii = NULL; | |
6264 | u64 left_gen = 0; | |
6265 | u64 right_gen = 0; | |
6266 | ||
31db9f7c AB |
6267 | sctx->cur_ino = key->objectid; |
6268 | sctx->cur_inode_new_gen = 0; | |
16e7549f | 6269 | sctx->cur_inode_last_extent = (u64)-1; |
ffa7c429 | 6270 | sctx->cur_inode_next_write_offset = 0; |
46b2f459 | 6271 | sctx->ignore_cur_inode = false; |
e479d9bb AB |
6272 | |
6273 | /* | |
6274 | * Set send_progress to current inode. This will tell all get_cur_xxx | |
6275 | * functions that the current inode's refs are not updated yet. Later, | |
6276 | * when process_recorded_refs is finished, it is set to cur_ino + 1. | |
6277 | */ | |
31db9f7c AB |
6278 | sctx->send_progress = sctx->cur_ino; |
6279 | ||
6280 | if (result == BTRFS_COMPARE_TREE_NEW || | |
6281 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
6282 | left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], | |
6283 | sctx->left_path->slots[0], | |
6284 | struct btrfs_inode_item); | |
6285 | left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], | |
6286 | left_ii); | |
6287 | } else { | |
6288 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6289 | sctx->right_path->slots[0], | |
6290 | struct btrfs_inode_item); | |
6291 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6292 | right_ii); | |
6293 | } | |
6294 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
6295 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6296 | sctx->right_path->slots[0], | |
6297 | struct btrfs_inode_item); | |
6298 | ||
6299 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6300 | right_ii); | |
6d85ed05 AB |
6301 | |
6302 | /* | |
6303 | * The cur_ino = root dir case is special here. We can't treat | |
6304 | * the inode as deleted+reused because it would generate a | |
6305 | * stream that tries to delete/mkdir the root dir. | |
6306 | */ | |
6307 | if (left_gen != right_gen && | |
6308 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) | |
31db9f7c AB |
6309 | sctx->cur_inode_new_gen = 1; |
6310 | } | |
6311 | ||
46b2f459 FM |
6312 | /* |
6313 | * Normally we do not find inodes with a link count of zero (orphans) | |
6314 | * because the most common case is to create a snapshot and use it | |
6315 | * for a send operation. However other less common use cases involve | |
6316 | * using a subvolume and send it after turning it to RO mode just | |
6317 | * after deleting all hard links of a file while holding an open | |
6318 | * file descriptor against it or turning a RO snapshot into RW mode, | |
6319 | * keep an open file descriptor against a file, delete it and then | |
6320 | * turn the snapshot back to RO mode before using it for a send | |
6321 | * operation. So if we find such cases, ignore the inode and all its | |
6322 | * items completely if it's a new inode, or if it's a changed inode | |
6323 | * make sure all its previous paths (from the parent snapshot) are all | |
6324 | * unlinked and all other the inode items are ignored. | |
6325 | */ | |
6326 | if (result == BTRFS_COMPARE_TREE_NEW || | |
6327 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
6328 | u32 nlinks; | |
6329 | ||
6330 | nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii); | |
6331 | if (nlinks == 0) { | |
6332 | sctx->ignore_cur_inode = true; | |
6333 | if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6334 | ret = btrfs_unlink_all_paths(sctx); | |
6335 | goto out; | |
6336 | } | |
6337 | } | |
6338 | ||
31db9f7c AB |
6339 | if (result == BTRFS_COMPARE_TREE_NEW) { |
6340 | sctx->cur_inode_gen = left_gen; | |
6341 | sctx->cur_inode_new = 1; | |
6342 | sctx->cur_inode_deleted = 0; | |
6343 | sctx->cur_inode_size = btrfs_inode_size( | |
6344 | sctx->left_path->nodes[0], left_ii); | |
6345 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6346 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6347 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6348 | sctx->left_path->nodes[0], left_ii); | |
31db9f7c | 6349 | if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) |
1f4692da | 6350 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6351 | } else if (result == BTRFS_COMPARE_TREE_DELETED) { |
6352 | sctx->cur_inode_gen = right_gen; | |
6353 | sctx->cur_inode_new = 0; | |
6354 | sctx->cur_inode_deleted = 1; | |
6355 | sctx->cur_inode_size = btrfs_inode_size( | |
6356 | sctx->right_path->nodes[0], right_ii); | |
6357 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6358 | sctx->right_path->nodes[0], right_ii); | |
6359 | } else if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
766702ef AB |
6360 | /* |
6361 | * We need to do some special handling in case the inode was | |
6362 | * reported as changed with a changed generation number. This | |
6363 | * means that the original inode was deleted and new inode | |
6364 | * reused the same inum. So we have to treat the old inode as | |
6365 | * deleted and the new one as new. | |
6366 | */ | |
31db9f7c | 6367 | if (sctx->cur_inode_new_gen) { |
766702ef AB |
6368 | /* |
6369 | * First, process the inode as if it was deleted. | |
6370 | */ | |
31db9f7c AB |
6371 | sctx->cur_inode_gen = right_gen; |
6372 | sctx->cur_inode_new = 0; | |
6373 | sctx->cur_inode_deleted = 1; | |
6374 | sctx->cur_inode_size = btrfs_inode_size( | |
6375 | sctx->right_path->nodes[0], right_ii); | |
6376 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6377 | sctx->right_path->nodes[0], right_ii); | |
6378 | ret = process_all_refs(sctx, | |
6379 | BTRFS_COMPARE_TREE_DELETED); | |
6380 | if (ret < 0) | |
6381 | goto out; | |
6382 | ||
766702ef AB |
6383 | /* |
6384 | * Now process the inode as if it was new. | |
6385 | */ | |
31db9f7c AB |
6386 | sctx->cur_inode_gen = left_gen; |
6387 | sctx->cur_inode_new = 1; | |
6388 | sctx->cur_inode_deleted = 0; | |
6389 | sctx->cur_inode_size = btrfs_inode_size( | |
6390 | sctx->left_path->nodes[0], left_ii); | |
6391 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6392 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6393 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6394 | sctx->left_path->nodes[0], left_ii); | |
1f4692da | 6395 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6396 | if (ret < 0) |
6397 | goto out; | |
6398 | ||
6399 | ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); | |
6400 | if (ret < 0) | |
6401 | goto out; | |
e479d9bb AB |
6402 | /* |
6403 | * Advance send_progress now as we did not get into | |
6404 | * process_recorded_refs_if_needed in the new_gen case. | |
6405 | */ | |
6406 | sctx->send_progress = sctx->cur_ino + 1; | |
766702ef AB |
6407 | |
6408 | /* | |
6409 | * Now process all extents and xattrs of the inode as if | |
6410 | * they were all new. | |
6411 | */ | |
31db9f7c AB |
6412 | ret = process_all_extents(sctx); |
6413 | if (ret < 0) | |
6414 | goto out; | |
6415 | ret = process_all_new_xattrs(sctx); | |
6416 | if (ret < 0) | |
6417 | goto out; | |
6418 | } else { | |
6419 | sctx->cur_inode_gen = left_gen; | |
6420 | sctx->cur_inode_new = 0; | |
6421 | sctx->cur_inode_new_gen = 0; | |
6422 | sctx->cur_inode_deleted = 0; | |
6423 | sctx->cur_inode_size = btrfs_inode_size( | |
6424 | sctx->left_path->nodes[0], left_ii); | |
6425 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6426 | sctx->left_path->nodes[0], left_ii); | |
6427 | } | |
6428 | } | |
6429 | ||
6430 | out: | |
6431 | return ret; | |
6432 | } | |
6433 | ||
766702ef AB |
6434 | /* |
6435 | * We have to process new refs before deleted refs, but compare_trees gives us | |
6436 | * the new and deleted refs mixed. To fix this, we record the new/deleted refs | |
6437 | * first and later process them in process_recorded_refs. | |
6438 | * For the cur_inode_new_gen case, we skip recording completely because | |
6439 | * changed_inode did already initiate processing of refs. The reason for this is | |
6440 | * that in this case, compare_tree actually compares the refs of 2 different | |
6441 | * inodes. To fix this, process_all_refs is used in changed_inode to handle all | |
6442 | * refs of the right tree as deleted and all refs of the left tree as new. | |
6443 | */ | |
31db9f7c AB |
6444 | static int changed_ref(struct send_ctx *sctx, |
6445 | enum btrfs_compare_tree_result result) | |
6446 | { | |
6447 | int ret = 0; | |
6448 | ||
95155585 FM |
6449 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6450 | inconsistent_snapshot_error(sctx, result, "reference"); | |
6451 | return -EIO; | |
6452 | } | |
31db9f7c AB |
6453 | |
6454 | if (!sctx->cur_inode_new_gen && | |
6455 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { | |
6456 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6457 | ret = record_new_ref(sctx); | |
6458 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6459 | ret = record_deleted_ref(sctx); | |
6460 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6461 | ret = record_changed_ref(sctx); | |
6462 | } | |
6463 | ||
6464 | return ret; | |
6465 | } | |
6466 | ||
766702ef AB |
6467 | /* |
6468 | * Process new/deleted/changed xattrs. We skip processing in the | |
6469 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6470 | * of xattrs. The reason is the same as in changed_ref | |
6471 | */ | |
31db9f7c AB |
6472 | static int changed_xattr(struct send_ctx *sctx, |
6473 | enum btrfs_compare_tree_result result) | |
6474 | { | |
6475 | int ret = 0; | |
6476 | ||
95155585 FM |
6477 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6478 | inconsistent_snapshot_error(sctx, result, "xattr"); | |
6479 | return -EIO; | |
6480 | } | |
31db9f7c AB |
6481 | |
6482 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6483 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6484 | ret = process_new_xattr(sctx); | |
6485 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6486 | ret = process_deleted_xattr(sctx); | |
6487 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6488 | ret = process_changed_xattr(sctx); | |
6489 | } | |
6490 | ||
6491 | return ret; | |
6492 | } | |
6493 | ||
766702ef AB |
6494 | /* |
6495 | * Process new/deleted/changed extents. We skip processing in the | |
6496 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6497 | * of extents. The reason is the same as in changed_ref | |
6498 | */ | |
31db9f7c AB |
6499 | static int changed_extent(struct send_ctx *sctx, |
6500 | enum btrfs_compare_tree_result result) | |
6501 | { | |
6502 | int ret = 0; | |
6503 | ||
b4f9a1a8 FM |
6504 | /* |
6505 | * We have found an extent item that changed without the inode item | |
6506 | * having changed. This can happen either after relocation (where the | |
6507 | * disk_bytenr of an extent item is replaced at | |
6508 | * relocation.c:replace_file_extents()) or after deduplication into a | |
6509 | * file in both the parent and send snapshots (where an extent item can | |
6510 | * get modified or replaced with a new one). Note that deduplication | |
6511 | * updates the inode item, but it only changes the iversion (sequence | |
6512 | * field in the inode item) of the inode, so if a file is deduplicated | |
6513 | * the same amount of times in both the parent and send snapshots, its | |
6514 | * iversion becames the same in both snapshots, whence the inode item is | |
6515 | * the same on both snapshots. | |
6516 | */ | |
6517 | if (sctx->cur_ino != sctx->cmp_key->objectid) | |
6518 | return 0; | |
31db9f7c AB |
6519 | |
6520 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6521 | if (result != BTRFS_COMPARE_TREE_DELETED) | |
6522 | ret = process_extent(sctx, sctx->left_path, | |
6523 | sctx->cmp_key); | |
6524 | } | |
6525 | ||
6526 | return ret; | |
6527 | } | |
6528 | ||
ba5e8f2e JB |
6529 | static int dir_changed(struct send_ctx *sctx, u64 dir) |
6530 | { | |
6531 | u64 orig_gen, new_gen; | |
6532 | int ret; | |
6533 | ||
6534 | ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, | |
6535 | NULL, NULL); | |
6536 | if (ret) | |
6537 | return ret; | |
6538 | ||
6539 | ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, | |
6540 | NULL, NULL, NULL); | |
6541 | if (ret) | |
6542 | return ret; | |
6543 | ||
6544 | return (orig_gen != new_gen) ? 1 : 0; | |
6545 | } | |
6546 | ||
6547 | static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, | |
6548 | struct btrfs_key *key) | |
6549 | { | |
6550 | struct btrfs_inode_extref *extref; | |
6551 | struct extent_buffer *leaf; | |
6552 | u64 dirid = 0, last_dirid = 0; | |
6553 | unsigned long ptr; | |
6554 | u32 item_size; | |
6555 | u32 cur_offset = 0; | |
6556 | int ref_name_len; | |
6557 | int ret = 0; | |
6558 | ||
6559 | /* Easy case, just check this one dirid */ | |
6560 | if (key->type == BTRFS_INODE_REF_KEY) { | |
6561 | dirid = key->offset; | |
6562 | ||
6563 | ret = dir_changed(sctx, dirid); | |
6564 | goto out; | |
6565 | } | |
6566 | ||
6567 | leaf = path->nodes[0]; | |
6568 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6569 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
6570 | while (cur_offset < item_size) { | |
6571 | extref = (struct btrfs_inode_extref *)(ptr + | |
6572 | cur_offset); | |
6573 | dirid = btrfs_inode_extref_parent(leaf, extref); | |
6574 | ref_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
6575 | cur_offset += ref_name_len + sizeof(*extref); | |
6576 | if (dirid == last_dirid) | |
6577 | continue; | |
6578 | ret = dir_changed(sctx, dirid); | |
6579 | if (ret) | |
6580 | break; | |
6581 | last_dirid = dirid; | |
6582 | } | |
6583 | out: | |
6584 | return ret; | |
6585 | } | |
6586 | ||
766702ef AB |
6587 | /* |
6588 | * Updates compare related fields in sctx and simply forwards to the actual | |
6589 | * changed_xxx functions. | |
6590 | */ | |
ee8c494f | 6591 | static int changed_cb(struct btrfs_path *left_path, |
31db9f7c AB |
6592 | struct btrfs_path *right_path, |
6593 | struct btrfs_key *key, | |
6594 | enum btrfs_compare_tree_result result, | |
6595 | void *ctx) | |
6596 | { | |
6597 | int ret = 0; | |
6598 | struct send_ctx *sctx = ctx; | |
6599 | ||
ba5e8f2e | 6600 | if (result == BTRFS_COMPARE_TREE_SAME) { |
16e7549f JB |
6601 | if (key->type == BTRFS_INODE_REF_KEY || |
6602 | key->type == BTRFS_INODE_EXTREF_KEY) { | |
6603 | ret = compare_refs(sctx, left_path, key); | |
6604 | if (!ret) | |
6605 | return 0; | |
6606 | if (ret < 0) | |
6607 | return ret; | |
6608 | } else if (key->type == BTRFS_EXTENT_DATA_KEY) { | |
6609 | return maybe_send_hole(sctx, left_path, key); | |
6610 | } else { | |
ba5e8f2e | 6611 | return 0; |
16e7549f | 6612 | } |
ba5e8f2e JB |
6613 | result = BTRFS_COMPARE_TREE_CHANGED; |
6614 | ret = 0; | |
6615 | } | |
6616 | ||
31db9f7c AB |
6617 | sctx->left_path = left_path; |
6618 | sctx->right_path = right_path; | |
6619 | sctx->cmp_key = key; | |
6620 | ||
6621 | ret = finish_inode_if_needed(sctx, 0); | |
6622 | if (ret < 0) | |
6623 | goto out; | |
6624 | ||
2981e225 AB |
6625 | /* Ignore non-FS objects */ |
6626 | if (key->objectid == BTRFS_FREE_INO_OBJECTID || | |
6627 | key->objectid == BTRFS_FREE_SPACE_OBJECTID) | |
6628 | goto out; | |
6629 | ||
46b2f459 | 6630 | if (key->type == BTRFS_INODE_ITEM_KEY) { |
31db9f7c | 6631 | ret = changed_inode(sctx, result); |
46b2f459 FM |
6632 | } else if (!sctx->ignore_cur_inode) { |
6633 | if (key->type == BTRFS_INODE_REF_KEY || | |
6634 | key->type == BTRFS_INODE_EXTREF_KEY) | |
6635 | ret = changed_ref(sctx, result); | |
6636 | else if (key->type == BTRFS_XATTR_ITEM_KEY) | |
6637 | ret = changed_xattr(sctx, result); | |
6638 | else if (key->type == BTRFS_EXTENT_DATA_KEY) | |
6639 | ret = changed_extent(sctx, result); | |
6640 | } | |
31db9f7c AB |
6641 | |
6642 | out: | |
6643 | return ret; | |
6644 | } | |
6645 | ||
6646 | static int full_send_tree(struct send_ctx *sctx) | |
6647 | { | |
6648 | int ret; | |
31db9f7c AB |
6649 | struct btrfs_root *send_root = sctx->send_root; |
6650 | struct btrfs_key key; | |
31db9f7c AB |
6651 | struct btrfs_path *path; |
6652 | struct extent_buffer *eb; | |
6653 | int slot; | |
31db9f7c AB |
6654 | |
6655 | path = alloc_path_for_send(); | |
6656 | if (!path) | |
6657 | return -ENOMEM; | |
6658 | ||
31db9f7c AB |
6659 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; |
6660 | key.type = BTRFS_INODE_ITEM_KEY; | |
6661 | key.offset = 0; | |
6662 | ||
31db9f7c AB |
6663 | ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); |
6664 | if (ret < 0) | |
6665 | goto out; | |
6666 | if (ret) | |
6667 | goto out_finish; | |
6668 | ||
6669 | while (1) { | |
31db9f7c AB |
6670 | eb = path->nodes[0]; |
6671 | slot = path->slots[0]; | |
ca5d2ba1 | 6672 | btrfs_item_key_to_cpu(eb, &key, slot); |
31db9f7c | 6673 | |
ca5d2ba1 | 6674 | ret = changed_cb(path, NULL, &key, |
ee8c494f | 6675 | BTRFS_COMPARE_TREE_NEW, sctx); |
31db9f7c AB |
6676 | if (ret < 0) |
6677 | goto out; | |
6678 | ||
31db9f7c AB |
6679 | ret = btrfs_next_item(send_root, path); |
6680 | if (ret < 0) | |
6681 | goto out; | |
6682 | if (ret) { | |
6683 | ret = 0; | |
6684 | break; | |
6685 | } | |
6686 | } | |
6687 | ||
6688 | out_finish: | |
6689 | ret = finish_inode_if_needed(sctx, 1); | |
6690 | ||
6691 | out: | |
6692 | btrfs_free_path(path); | |
31db9f7c AB |
6693 | return ret; |
6694 | } | |
6695 | ||
18d0f5c6 DS |
6696 | static int tree_move_down(struct btrfs_path *path, int *level) |
6697 | { | |
6698 | struct extent_buffer *eb; | |
6699 | ||
6700 | BUG_ON(*level == 0); | |
6701 | eb = btrfs_read_node_slot(path->nodes[*level], path->slots[*level]); | |
6702 | if (IS_ERR(eb)) | |
6703 | return PTR_ERR(eb); | |
6704 | ||
6705 | path->nodes[*level - 1] = eb; | |
6706 | path->slots[*level - 1] = 0; | |
6707 | (*level)--; | |
6708 | return 0; | |
6709 | } | |
6710 | ||
6711 | static int tree_move_next_or_upnext(struct btrfs_path *path, | |
6712 | int *level, int root_level) | |
6713 | { | |
6714 | int ret = 0; | |
6715 | int nritems; | |
6716 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
6717 | ||
6718 | path->slots[*level]++; | |
6719 | ||
6720 | while (path->slots[*level] >= nritems) { | |
6721 | if (*level == root_level) | |
6722 | return -1; | |
6723 | ||
6724 | /* move upnext */ | |
6725 | path->slots[*level] = 0; | |
6726 | free_extent_buffer(path->nodes[*level]); | |
6727 | path->nodes[*level] = NULL; | |
6728 | (*level)++; | |
6729 | path->slots[*level]++; | |
6730 | ||
6731 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
6732 | ret = 1; | |
6733 | } | |
6734 | return ret; | |
6735 | } | |
6736 | ||
6737 | /* | |
6738 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
6739 | * or down. | |
6740 | */ | |
6741 | static int tree_advance(struct btrfs_path *path, | |
6742 | int *level, int root_level, | |
6743 | int allow_down, | |
6744 | struct btrfs_key *key) | |
6745 | { | |
6746 | int ret; | |
6747 | ||
6748 | if (*level == 0 || !allow_down) { | |
6749 | ret = tree_move_next_or_upnext(path, level, root_level); | |
6750 | } else { | |
6751 | ret = tree_move_down(path, level); | |
6752 | } | |
6753 | if (ret >= 0) { | |
6754 | if (*level == 0) | |
6755 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
6756 | path->slots[*level]); | |
6757 | else | |
6758 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
6759 | path->slots[*level]); | |
6760 | } | |
6761 | return ret; | |
6762 | } | |
6763 | ||
6764 | static int tree_compare_item(struct btrfs_path *left_path, | |
6765 | struct btrfs_path *right_path, | |
6766 | char *tmp_buf) | |
6767 | { | |
6768 | int cmp; | |
6769 | int len1, len2; | |
6770 | unsigned long off1, off2; | |
6771 | ||
6772 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
6773 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
6774 | if (len1 != len2) | |
6775 | return 1; | |
6776 | ||
6777 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
6778 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
6779 | right_path->slots[0]); | |
6780 | ||
6781 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
6782 | ||
6783 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
6784 | if (cmp) | |
6785 | return 1; | |
6786 | return 0; | |
6787 | } | |
6788 | ||
6789 | /* | |
6790 | * This function compares two trees and calls the provided callback for | |
6791 | * every changed/new/deleted item it finds. | |
6792 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
6793 | * the compare pretty fast on snapshotted subvolumes. | |
6794 | * | |
6795 | * This currently works on commit roots only. As commit roots are read only, | |
6796 | * we don't do any locking. The commit roots are protected with transactions. | |
6797 | * Transactions are ended and rejoined when a commit is tried in between. | |
6798 | * | |
6799 | * This function checks for modifications done to the trees while comparing. | |
6800 | * If it detects a change, it aborts immediately. | |
6801 | */ | |
6802 | static int btrfs_compare_trees(struct btrfs_root *left_root, | |
1b51d6fc | 6803 | struct btrfs_root *right_root, void *ctx) |
18d0f5c6 DS |
6804 | { |
6805 | struct btrfs_fs_info *fs_info = left_root->fs_info; | |
6806 | int ret; | |
6807 | int cmp; | |
6808 | struct btrfs_path *left_path = NULL; | |
6809 | struct btrfs_path *right_path = NULL; | |
6810 | struct btrfs_key left_key; | |
6811 | struct btrfs_key right_key; | |
6812 | char *tmp_buf = NULL; | |
6813 | int left_root_level; | |
6814 | int right_root_level; | |
6815 | int left_level; | |
6816 | int right_level; | |
6817 | int left_end_reached; | |
6818 | int right_end_reached; | |
6819 | int advance_left; | |
6820 | int advance_right; | |
6821 | u64 left_blockptr; | |
6822 | u64 right_blockptr; | |
6823 | u64 left_gen; | |
6824 | u64 right_gen; | |
6825 | ||
6826 | left_path = btrfs_alloc_path(); | |
6827 | if (!left_path) { | |
6828 | ret = -ENOMEM; | |
6829 | goto out; | |
6830 | } | |
6831 | right_path = btrfs_alloc_path(); | |
6832 | if (!right_path) { | |
6833 | ret = -ENOMEM; | |
6834 | goto out; | |
6835 | } | |
6836 | ||
6837 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); | |
6838 | if (!tmp_buf) { | |
6839 | ret = -ENOMEM; | |
6840 | goto out; | |
6841 | } | |
6842 | ||
6843 | left_path->search_commit_root = 1; | |
6844 | left_path->skip_locking = 1; | |
6845 | right_path->search_commit_root = 1; | |
6846 | right_path->skip_locking = 1; | |
6847 | ||
6848 | /* | |
6849 | * Strategy: Go to the first items of both trees. Then do | |
6850 | * | |
6851 | * If both trees are at level 0 | |
6852 | * Compare keys of current items | |
6853 | * If left < right treat left item as new, advance left tree | |
6854 | * and repeat | |
6855 | * If left > right treat right item as deleted, advance right tree | |
6856 | * and repeat | |
6857 | * If left == right do deep compare of items, treat as changed if | |
6858 | * needed, advance both trees and repeat | |
6859 | * If both trees are at the same level but not at level 0 | |
6860 | * Compare keys of current nodes/leafs | |
6861 | * If left < right advance left tree and repeat | |
6862 | * If left > right advance right tree and repeat | |
6863 | * If left == right compare blockptrs of the next nodes/leafs | |
6864 | * If they match advance both trees but stay at the same level | |
6865 | * and repeat | |
6866 | * If they don't match advance both trees while allowing to go | |
6867 | * deeper and repeat | |
6868 | * If tree levels are different | |
6869 | * Advance the tree that needs it and repeat | |
6870 | * | |
6871 | * Advancing a tree means: | |
6872 | * If we are at level 0, try to go to the next slot. If that's not | |
6873 | * possible, go one level up and repeat. Stop when we found a level | |
6874 | * where we could go to the next slot. We may at this point be on a | |
6875 | * node or a leaf. | |
6876 | * | |
6877 | * If we are not at level 0 and not on shared tree blocks, go one | |
6878 | * level deeper. | |
6879 | * | |
6880 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
6881 | * the right if possible or go up and right. | |
6882 | */ | |
6883 | ||
6884 | down_read(&fs_info->commit_root_sem); | |
6885 | left_level = btrfs_header_level(left_root->commit_root); | |
6886 | left_root_level = left_level; | |
6887 | left_path->nodes[left_level] = | |
6888 | btrfs_clone_extent_buffer(left_root->commit_root); | |
6889 | if (!left_path->nodes[left_level]) { | |
6890 | up_read(&fs_info->commit_root_sem); | |
6891 | ret = -ENOMEM; | |
6892 | goto out; | |
6893 | } | |
6894 | ||
6895 | right_level = btrfs_header_level(right_root->commit_root); | |
6896 | right_root_level = right_level; | |
6897 | right_path->nodes[right_level] = | |
6898 | btrfs_clone_extent_buffer(right_root->commit_root); | |
6899 | if (!right_path->nodes[right_level]) { | |
6900 | up_read(&fs_info->commit_root_sem); | |
6901 | ret = -ENOMEM; | |
6902 | goto out; | |
6903 | } | |
6904 | up_read(&fs_info->commit_root_sem); | |
6905 | ||
6906 | if (left_level == 0) | |
6907 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
6908 | &left_key, left_path->slots[left_level]); | |
6909 | else | |
6910 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
6911 | &left_key, left_path->slots[left_level]); | |
6912 | if (right_level == 0) | |
6913 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
6914 | &right_key, right_path->slots[right_level]); | |
6915 | else | |
6916 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
6917 | &right_key, right_path->slots[right_level]); | |
6918 | ||
6919 | left_end_reached = right_end_reached = 0; | |
6920 | advance_left = advance_right = 0; | |
6921 | ||
6922 | while (1) { | |
6af112b1 | 6923 | cond_resched(); |
18d0f5c6 DS |
6924 | if (advance_left && !left_end_reached) { |
6925 | ret = tree_advance(left_path, &left_level, | |
6926 | left_root_level, | |
6927 | advance_left != ADVANCE_ONLY_NEXT, | |
6928 | &left_key); | |
6929 | if (ret == -1) | |
6930 | left_end_reached = ADVANCE; | |
6931 | else if (ret < 0) | |
6932 | goto out; | |
6933 | advance_left = 0; | |
6934 | } | |
6935 | if (advance_right && !right_end_reached) { | |
6936 | ret = tree_advance(right_path, &right_level, | |
6937 | right_root_level, | |
6938 | advance_right != ADVANCE_ONLY_NEXT, | |
6939 | &right_key); | |
6940 | if (ret == -1) | |
6941 | right_end_reached = ADVANCE; | |
6942 | else if (ret < 0) | |
6943 | goto out; | |
6944 | advance_right = 0; | |
6945 | } | |
6946 | ||
6947 | if (left_end_reached && right_end_reached) { | |
6948 | ret = 0; | |
6949 | goto out; | |
6950 | } else if (left_end_reached) { | |
6951 | if (right_level == 0) { | |
6952 | ret = changed_cb(left_path, right_path, | |
6953 | &right_key, | |
6954 | BTRFS_COMPARE_TREE_DELETED, | |
6955 | ctx); | |
6956 | if (ret < 0) | |
6957 | goto out; | |
6958 | } | |
6959 | advance_right = ADVANCE; | |
6960 | continue; | |
6961 | } else if (right_end_reached) { | |
6962 | if (left_level == 0) { | |
6963 | ret = changed_cb(left_path, right_path, | |
6964 | &left_key, | |
6965 | BTRFS_COMPARE_TREE_NEW, | |
6966 | ctx); | |
6967 | if (ret < 0) | |
6968 | goto out; | |
6969 | } | |
6970 | advance_left = ADVANCE; | |
6971 | continue; | |
6972 | } | |
6973 | ||
6974 | if (left_level == 0 && right_level == 0) { | |
6975 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
6976 | if (cmp < 0) { | |
6977 | ret = changed_cb(left_path, right_path, | |
6978 | &left_key, | |
6979 | BTRFS_COMPARE_TREE_NEW, | |
6980 | ctx); | |
6981 | if (ret < 0) | |
6982 | goto out; | |
6983 | advance_left = ADVANCE; | |
6984 | } else if (cmp > 0) { | |
6985 | ret = changed_cb(left_path, right_path, | |
6986 | &right_key, | |
6987 | BTRFS_COMPARE_TREE_DELETED, | |
6988 | ctx); | |
6989 | if (ret < 0) | |
6990 | goto out; | |
6991 | advance_right = ADVANCE; | |
6992 | } else { | |
6993 | enum btrfs_compare_tree_result result; | |
6994 | ||
6995 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); | |
6996 | ret = tree_compare_item(left_path, right_path, | |
6997 | tmp_buf); | |
6998 | if (ret) | |
6999 | result = BTRFS_COMPARE_TREE_CHANGED; | |
7000 | else | |
7001 | result = BTRFS_COMPARE_TREE_SAME; | |
7002 | ret = changed_cb(left_path, right_path, | |
7003 | &left_key, result, ctx); | |
7004 | if (ret < 0) | |
7005 | goto out; | |
7006 | advance_left = ADVANCE; | |
7007 | advance_right = ADVANCE; | |
7008 | } | |
7009 | } else if (left_level == right_level) { | |
7010 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
7011 | if (cmp < 0) { | |
7012 | advance_left = ADVANCE; | |
7013 | } else if (cmp > 0) { | |
7014 | advance_right = ADVANCE; | |
7015 | } else { | |
7016 | left_blockptr = btrfs_node_blockptr( | |
7017 | left_path->nodes[left_level], | |
7018 | left_path->slots[left_level]); | |
7019 | right_blockptr = btrfs_node_blockptr( | |
7020 | right_path->nodes[right_level], | |
7021 | right_path->slots[right_level]); | |
7022 | left_gen = btrfs_node_ptr_generation( | |
7023 | left_path->nodes[left_level], | |
7024 | left_path->slots[left_level]); | |
7025 | right_gen = btrfs_node_ptr_generation( | |
7026 | right_path->nodes[right_level], | |
7027 | right_path->slots[right_level]); | |
7028 | if (left_blockptr == right_blockptr && | |
7029 | left_gen == right_gen) { | |
7030 | /* | |
7031 | * As we're on a shared block, don't | |
7032 | * allow to go deeper. | |
7033 | */ | |
7034 | advance_left = ADVANCE_ONLY_NEXT; | |
7035 | advance_right = ADVANCE_ONLY_NEXT; | |
7036 | } else { | |
7037 | advance_left = ADVANCE; | |
7038 | advance_right = ADVANCE; | |
7039 | } | |
7040 | } | |
7041 | } else if (left_level < right_level) { | |
7042 | advance_right = ADVANCE; | |
7043 | } else { | |
7044 | advance_left = ADVANCE; | |
7045 | } | |
7046 | } | |
7047 | ||
7048 | out: | |
7049 | btrfs_free_path(left_path); | |
7050 | btrfs_free_path(right_path); | |
7051 | kvfree(tmp_buf); | |
7052 | return ret; | |
7053 | } | |
7054 | ||
31db9f7c AB |
7055 | static int send_subvol(struct send_ctx *sctx) |
7056 | { | |
7057 | int ret; | |
7058 | ||
c2c71324 SB |
7059 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { |
7060 | ret = send_header(sctx); | |
7061 | if (ret < 0) | |
7062 | goto out; | |
7063 | } | |
31db9f7c AB |
7064 | |
7065 | ret = send_subvol_begin(sctx); | |
7066 | if (ret < 0) | |
7067 | goto out; | |
7068 | ||
7069 | if (sctx->parent_root) { | |
1b51d6fc | 7070 | ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, sctx); |
31db9f7c AB |
7071 | if (ret < 0) |
7072 | goto out; | |
7073 | ret = finish_inode_if_needed(sctx, 1); | |
7074 | if (ret < 0) | |
7075 | goto out; | |
7076 | } else { | |
7077 | ret = full_send_tree(sctx); | |
7078 | if (ret < 0) | |
7079 | goto out; | |
7080 | } | |
7081 | ||
7082 | out: | |
31db9f7c AB |
7083 | free_recorded_refs(sctx); |
7084 | return ret; | |
7085 | } | |
7086 | ||
e5fa8f86 FM |
7087 | /* |
7088 | * If orphan cleanup did remove any orphans from a root, it means the tree | |
7089 | * was modified and therefore the commit root is not the same as the current | |
7090 | * root anymore. This is a problem, because send uses the commit root and | |
7091 | * therefore can see inode items that don't exist in the current root anymore, | |
7092 | * and for example make calls to btrfs_iget, which will do tree lookups based | |
7093 | * on the current root and not on the commit root. Those lookups will fail, | |
7094 | * returning a -ESTALE error, and making send fail with that error. So make | |
7095 | * sure a send does not see any orphans we have just removed, and that it will | |
7096 | * see the same inodes regardless of whether a transaction commit happened | |
7097 | * before it started (meaning that the commit root will be the same as the | |
7098 | * current root) or not. | |
7099 | */ | |
7100 | static int ensure_commit_roots_uptodate(struct send_ctx *sctx) | |
7101 | { | |
7102 | int i; | |
7103 | struct btrfs_trans_handle *trans = NULL; | |
7104 | ||
7105 | again: | |
7106 | if (sctx->parent_root && | |
7107 | sctx->parent_root->node != sctx->parent_root->commit_root) | |
7108 | goto commit_trans; | |
7109 | ||
7110 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
7111 | if (sctx->clone_roots[i].root->node != | |
7112 | sctx->clone_roots[i].root->commit_root) | |
7113 | goto commit_trans; | |
7114 | ||
7115 | if (trans) | |
3a45bb20 | 7116 | return btrfs_end_transaction(trans); |
e5fa8f86 FM |
7117 | |
7118 | return 0; | |
7119 | ||
7120 | commit_trans: | |
7121 | /* Use any root, all fs roots will get their commit roots updated. */ | |
7122 | if (!trans) { | |
7123 | trans = btrfs_join_transaction(sctx->send_root); | |
7124 | if (IS_ERR(trans)) | |
7125 | return PTR_ERR(trans); | |
7126 | goto again; | |
7127 | } | |
7128 | ||
3a45bb20 | 7129 | return btrfs_commit_transaction(trans); |
e5fa8f86 FM |
7130 | } |
7131 | ||
9f89d5de FM |
7132 | /* |
7133 | * Make sure any existing dellaloc is flushed for any root used by a send | |
7134 | * operation so that we do not miss any data and we do not race with writeback | |
7135 | * finishing and changing a tree while send is using the tree. This could | |
7136 | * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and | |
7137 | * a send operation then uses the subvolume. | |
7138 | * After flushing delalloc ensure_commit_roots_uptodate() must be called. | |
7139 | */ | |
7140 | static int flush_delalloc_roots(struct send_ctx *sctx) | |
7141 | { | |
7142 | struct btrfs_root *root = sctx->parent_root; | |
7143 | int ret; | |
7144 | int i; | |
7145 | ||
7146 | if (root) { | |
7147 | ret = btrfs_start_delalloc_snapshot(root); | |
7148 | if (ret) | |
7149 | return ret; | |
7150 | btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX); | |
7151 | } | |
7152 | ||
7153 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
7154 | root = sctx->clone_roots[i].root; | |
7155 | ret = btrfs_start_delalloc_snapshot(root); | |
7156 | if (ret) | |
7157 | return ret; | |
7158 | btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX); | |
7159 | } | |
7160 | ||
7161 | return 0; | |
7162 | } | |
7163 | ||
66ef7d65 DS |
7164 | static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) |
7165 | { | |
7166 | spin_lock(&root->root_item_lock); | |
7167 | root->send_in_progress--; | |
7168 | /* | |
7169 | * Not much left to do, we don't know why it's unbalanced and | |
7170 | * can't blindly reset it to 0. | |
7171 | */ | |
7172 | if (root->send_in_progress < 0) | |
7173 | btrfs_err(root->fs_info, | |
f5686e3a | 7174 | "send_in_progress unbalanced %d root %llu", |
0b246afa | 7175 | root->send_in_progress, root->root_key.objectid); |
66ef7d65 DS |
7176 | spin_unlock(&root->root_item_lock); |
7177 | } | |
7178 | ||
62d54f3a FM |
7179 | static void dedupe_in_progress_warn(const struct btrfs_root *root) |
7180 | { | |
7181 | btrfs_warn_rl(root->fs_info, | |
7182 | "cannot use root %llu for send while deduplications on it are in progress (%d in progress)", | |
7183 | root->root_key.objectid, root->dedupe_in_progress); | |
7184 | } | |
7185 | ||
2351f431 | 7186 | long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) |
31db9f7c AB |
7187 | { |
7188 | int ret = 0; | |
0b246afa JM |
7189 | struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root; |
7190 | struct btrfs_fs_info *fs_info = send_root->fs_info; | |
31db9f7c | 7191 | struct btrfs_root *clone_root; |
31db9f7c AB |
7192 | struct send_ctx *sctx = NULL; |
7193 | u32 i; | |
7194 | u64 *clone_sources_tmp = NULL; | |
2c686537 | 7195 | int clone_sources_to_rollback = 0; |
bae12df9 | 7196 | size_t alloc_size; |
896c14f9 | 7197 | int sort_clone_roots = 0; |
31db9f7c AB |
7198 | |
7199 | if (!capable(CAP_SYS_ADMIN)) | |
7200 | return -EPERM; | |
7201 | ||
2c686537 DS |
7202 | /* |
7203 | * The subvolume must remain read-only during send, protect against | |
521e0546 | 7204 | * making it RW. This also protects against deletion. |
2c686537 DS |
7205 | */ |
7206 | spin_lock(&send_root->root_item_lock); | |
62d54f3a FM |
7207 | if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) { |
7208 | dedupe_in_progress_warn(send_root); | |
7209 | spin_unlock(&send_root->root_item_lock); | |
7210 | return -EAGAIN; | |
7211 | } | |
2c686537 DS |
7212 | send_root->send_in_progress++; |
7213 | spin_unlock(&send_root->root_item_lock); | |
7214 | ||
2c686537 DS |
7215 | /* |
7216 | * Userspace tools do the checks and warn the user if it's | |
7217 | * not RO. | |
7218 | */ | |
7219 | if (!btrfs_root_readonly(send_root)) { | |
7220 | ret = -EPERM; | |
7221 | goto out; | |
7222 | } | |
7223 | ||
457ae726 DC |
7224 | /* |
7225 | * Check that we don't overflow at later allocations, we request | |
7226 | * clone_sources_count + 1 items, and compare to unsigned long inside | |
7227 | * access_ok. | |
7228 | */ | |
f5ecec3c | 7229 | if (arg->clone_sources_count > |
457ae726 | 7230 | ULONG_MAX / sizeof(struct clone_root) - 1) { |
f5ecec3c DC |
7231 | ret = -EINVAL; |
7232 | goto out; | |
7233 | } | |
7234 | ||
c2c71324 | 7235 | if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { |
cb95e7bf MF |
7236 | ret = -EINVAL; |
7237 | goto out; | |
7238 | } | |
7239 | ||
e780b0d1 | 7240 | sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL); |
31db9f7c AB |
7241 | if (!sctx) { |
7242 | ret = -ENOMEM; | |
7243 | goto out; | |
7244 | } | |
7245 | ||
7246 | INIT_LIST_HEAD(&sctx->new_refs); | |
7247 | INIT_LIST_HEAD(&sctx->deleted_refs); | |
e780b0d1 | 7248 | INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL); |
31db9f7c AB |
7249 | INIT_LIST_HEAD(&sctx->name_cache_list); |
7250 | ||
cb95e7bf MF |
7251 | sctx->flags = arg->flags; |
7252 | ||
31db9f7c | 7253 | sctx->send_filp = fget(arg->send_fd); |
ecc7ada7 TI |
7254 | if (!sctx->send_filp) { |
7255 | ret = -EBADF; | |
31db9f7c AB |
7256 | goto out; |
7257 | } | |
7258 | ||
31db9f7c | 7259 | sctx->send_root = send_root; |
521e0546 DS |
7260 | /* |
7261 | * Unlikely but possible, if the subvolume is marked for deletion but | |
7262 | * is slow to remove the directory entry, send can still be started | |
7263 | */ | |
7264 | if (btrfs_root_dead(sctx->send_root)) { | |
7265 | ret = -EPERM; | |
7266 | goto out; | |
7267 | } | |
7268 | ||
31db9f7c AB |
7269 | sctx->clone_roots_cnt = arg->clone_sources_count; |
7270 | ||
7271 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE; | |
752ade68 | 7272 | sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); |
31db9f7c | 7273 | if (!sctx->send_buf) { |
752ade68 MH |
7274 | ret = -ENOMEM; |
7275 | goto out; | |
31db9f7c AB |
7276 | } |
7277 | ||
9f03740a FDBM |
7278 | sctx->pending_dir_moves = RB_ROOT; |
7279 | sctx->waiting_dir_moves = RB_ROOT; | |
9dc44214 | 7280 | sctx->orphan_dirs = RB_ROOT; |
9f03740a | 7281 | |
bae12df9 DE |
7282 | sctx->clone_roots = kvcalloc(sizeof(*sctx->clone_roots), |
7283 | arg->clone_sources_count + 1, | |
7284 | GFP_KERNEL); | |
31db9f7c | 7285 | if (!sctx->clone_roots) { |
818e010b DS |
7286 | ret = -ENOMEM; |
7287 | goto out; | |
31db9f7c AB |
7288 | } |
7289 | ||
bae12df9 DE |
7290 | alloc_size = array_size(sizeof(*arg->clone_sources), |
7291 | arg->clone_sources_count); | |
e55d1153 | 7292 | |
31db9f7c | 7293 | if (arg->clone_sources_count) { |
752ade68 | 7294 | clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 7295 | if (!clone_sources_tmp) { |
752ade68 MH |
7296 | ret = -ENOMEM; |
7297 | goto out; | |
31db9f7c AB |
7298 | } |
7299 | ||
7300 | ret = copy_from_user(clone_sources_tmp, arg->clone_sources, | |
e55d1153 | 7301 | alloc_size); |
31db9f7c AB |
7302 | if (ret) { |
7303 | ret = -EFAULT; | |
7304 | goto out; | |
7305 | } | |
7306 | ||
7307 | for (i = 0; i < arg->clone_sources_count; i++) { | |
56e9357a DS |
7308 | clone_root = btrfs_get_fs_root(fs_info, |
7309 | clone_sources_tmp[i], true); | |
31db9f7c AB |
7310 | if (IS_ERR(clone_root)) { |
7311 | ret = PTR_ERR(clone_root); | |
7312 | goto out; | |
7313 | } | |
2c686537 | 7314 | spin_lock(&clone_root->root_item_lock); |
5cc2b17e FM |
7315 | if (!btrfs_root_readonly(clone_root) || |
7316 | btrfs_root_dead(clone_root)) { | |
2c686537 | 7317 | spin_unlock(&clone_root->root_item_lock); |
00246528 | 7318 | btrfs_put_root(clone_root); |
2c686537 DS |
7319 | ret = -EPERM; |
7320 | goto out; | |
7321 | } | |
62d54f3a FM |
7322 | if (clone_root->dedupe_in_progress) { |
7323 | dedupe_in_progress_warn(clone_root); | |
7324 | spin_unlock(&clone_root->root_item_lock); | |
00246528 | 7325 | btrfs_put_root(clone_root); |
62d54f3a FM |
7326 | ret = -EAGAIN; |
7327 | goto out; | |
7328 | } | |
2f1f465a | 7329 | clone_root->send_in_progress++; |
2c686537 | 7330 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 | 7331 | |
31db9f7c | 7332 | sctx->clone_roots[i].root = clone_root; |
2f1f465a | 7333 | clone_sources_to_rollback = i + 1; |
31db9f7c | 7334 | } |
2f91306a | 7335 | kvfree(clone_sources_tmp); |
31db9f7c AB |
7336 | clone_sources_tmp = NULL; |
7337 | } | |
7338 | ||
7339 | if (arg->parent_root) { | |
56e9357a DS |
7340 | sctx->parent_root = btrfs_get_fs_root(fs_info, arg->parent_root, |
7341 | true); | |
b1b19596 SB |
7342 | if (IS_ERR(sctx->parent_root)) { |
7343 | ret = PTR_ERR(sctx->parent_root); | |
31db9f7c AB |
7344 | goto out; |
7345 | } | |
18f687d5 | 7346 | |
2c686537 DS |
7347 | spin_lock(&sctx->parent_root->root_item_lock); |
7348 | sctx->parent_root->send_in_progress++; | |
521e0546 DS |
7349 | if (!btrfs_root_readonly(sctx->parent_root) || |
7350 | btrfs_root_dead(sctx->parent_root)) { | |
2c686537 DS |
7351 | spin_unlock(&sctx->parent_root->root_item_lock); |
7352 | ret = -EPERM; | |
7353 | goto out; | |
7354 | } | |
62d54f3a FM |
7355 | if (sctx->parent_root->dedupe_in_progress) { |
7356 | dedupe_in_progress_warn(sctx->parent_root); | |
7357 | spin_unlock(&sctx->parent_root->root_item_lock); | |
62d54f3a FM |
7358 | ret = -EAGAIN; |
7359 | goto out; | |
7360 | } | |
2c686537 | 7361 | spin_unlock(&sctx->parent_root->root_item_lock); |
31db9f7c AB |
7362 | } |
7363 | ||
7364 | /* | |
7365 | * Clones from send_root are allowed, but only if the clone source | |
7366 | * is behind the current send position. This is checked while searching | |
7367 | * for possible clone sources. | |
7368 | */ | |
6f9a3da5 | 7369 | sctx->clone_roots[sctx->clone_roots_cnt++].root = |
00246528 | 7370 | btrfs_grab_root(sctx->send_root); |
31db9f7c AB |
7371 | |
7372 | /* We do a bsearch later */ | |
7373 | sort(sctx->clone_roots, sctx->clone_roots_cnt, | |
7374 | sizeof(*sctx->clone_roots), __clone_root_cmp_sort, | |
7375 | NULL); | |
896c14f9 | 7376 | sort_clone_roots = 1; |
31db9f7c | 7377 | |
9f89d5de FM |
7378 | ret = flush_delalloc_roots(sctx); |
7379 | if (ret) | |
7380 | goto out; | |
7381 | ||
e5fa8f86 FM |
7382 | ret = ensure_commit_roots_uptodate(sctx); |
7383 | if (ret) | |
7384 | goto out; | |
7385 | ||
9e967495 FM |
7386 | mutex_lock(&fs_info->balance_mutex); |
7387 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { | |
7388 | mutex_unlock(&fs_info->balance_mutex); | |
7389 | btrfs_warn_rl(fs_info, | |
7390 | "cannot run send because a balance operation is in progress"); | |
7391 | ret = -EAGAIN; | |
7392 | goto out; | |
7393 | } | |
7394 | fs_info->send_in_progress++; | |
7395 | mutex_unlock(&fs_info->balance_mutex); | |
7396 | ||
2755a0de | 7397 | current->journal_info = BTRFS_SEND_TRANS_STUB; |
31db9f7c | 7398 | ret = send_subvol(sctx); |
a26e8c9f | 7399 | current->journal_info = NULL; |
9e967495 FM |
7400 | mutex_lock(&fs_info->balance_mutex); |
7401 | fs_info->send_in_progress--; | |
7402 | mutex_unlock(&fs_info->balance_mutex); | |
31db9f7c AB |
7403 | if (ret < 0) |
7404 | goto out; | |
7405 | ||
c2c71324 SB |
7406 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { |
7407 | ret = begin_cmd(sctx, BTRFS_SEND_C_END); | |
7408 | if (ret < 0) | |
7409 | goto out; | |
7410 | ret = send_cmd(sctx); | |
7411 | if (ret < 0) | |
7412 | goto out; | |
7413 | } | |
31db9f7c AB |
7414 | |
7415 | out: | |
9f03740a FDBM |
7416 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); |
7417 | while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { | |
7418 | struct rb_node *n; | |
7419 | struct pending_dir_move *pm; | |
7420 | ||
7421 | n = rb_first(&sctx->pending_dir_moves); | |
7422 | pm = rb_entry(n, struct pending_dir_move, node); | |
7423 | while (!list_empty(&pm->list)) { | |
7424 | struct pending_dir_move *pm2; | |
7425 | ||
7426 | pm2 = list_first_entry(&pm->list, | |
7427 | struct pending_dir_move, list); | |
7428 | free_pending_move(sctx, pm2); | |
7429 | } | |
7430 | free_pending_move(sctx, pm); | |
7431 | } | |
7432 | ||
7433 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); | |
7434 | while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { | |
7435 | struct rb_node *n; | |
7436 | struct waiting_dir_move *dm; | |
7437 | ||
7438 | n = rb_first(&sctx->waiting_dir_moves); | |
7439 | dm = rb_entry(n, struct waiting_dir_move, node); | |
7440 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
7441 | kfree(dm); | |
7442 | } | |
7443 | ||
9dc44214 FM |
7444 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); |
7445 | while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { | |
7446 | struct rb_node *n; | |
7447 | struct orphan_dir_info *odi; | |
7448 | ||
7449 | n = rb_first(&sctx->orphan_dirs); | |
7450 | odi = rb_entry(n, struct orphan_dir_info, node); | |
7451 | free_orphan_dir_info(sctx, odi); | |
7452 | } | |
7453 | ||
896c14f9 | 7454 | if (sort_clone_roots) { |
6f9a3da5 | 7455 | for (i = 0; i < sctx->clone_roots_cnt; i++) { |
896c14f9 WS |
7456 | btrfs_root_dec_send_in_progress( |
7457 | sctx->clone_roots[i].root); | |
00246528 | 7458 | btrfs_put_root(sctx->clone_roots[i].root); |
6f9a3da5 | 7459 | } |
896c14f9 | 7460 | } else { |
6f9a3da5 | 7461 | for (i = 0; sctx && i < clone_sources_to_rollback; i++) { |
896c14f9 WS |
7462 | btrfs_root_dec_send_in_progress( |
7463 | sctx->clone_roots[i].root); | |
00246528 | 7464 | btrfs_put_root(sctx->clone_roots[i].root); |
6f9a3da5 | 7465 | } |
896c14f9 WS |
7466 | |
7467 | btrfs_root_dec_send_in_progress(send_root); | |
7468 | } | |
6f9a3da5 | 7469 | if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) { |
66ef7d65 | 7470 | btrfs_root_dec_send_in_progress(sctx->parent_root); |
00246528 | 7471 | btrfs_put_root(sctx->parent_root); |
6f9a3da5 | 7472 | } |
2c686537 | 7473 | |
2f91306a | 7474 | kvfree(clone_sources_tmp); |
31db9f7c AB |
7475 | |
7476 | if (sctx) { | |
7477 | if (sctx->send_filp) | |
7478 | fput(sctx->send_filp); | |
7479 | ||
c03d01f3 | 7480 | kvfree(sctx->clone_roots); |
6ff48ce0 | 7481 | kvfree(sctx->send_buf); |
31db9f7c AB |
7482 | |
7483 | name_cache_free(sctx); | |
7484 | ||
7485 | kfree(sctx); | |
7486 | } | |
7487 | ||
7488 | return ret; | |
7489 | } |