]>
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" |
31db9f7c | 26 | |
31db9f7c AB |
27 | /* |
28 | * A fs_path is a helper to dynamically build path names with unknown size. | |
29 | * It reallocates the internal buffer on demand. | |
30 | * It allows fast adding of path elements on the right side (normal path) and | |
31 | * fast adding to the left side (reversed path). A reversed path can also be | |
32 | * unreversed if needed. | |
33 | */ | |
34 | struct fs_path { | |
35 | union { | |
36 | struct { | |
37 | char *start; | |
38 | char *end; | |
31db9f7c AB |
39 | |
40 | char *buf; | |
1f5a7ff9 DS |
41 | unsigned short buf_len:15; |
42 | unsigned short reversed:1; | |
31db9f7c AB |
43 | char inline_buf[]; |
44 | }; | |
ace01050 DS |
45 | /* |
46 | * Average path length does not exceed 200 bytes, we'll have | |
47 | * better packing in the slab and higher chance to satisfy | |
48 | * a allocation later during send. | |
49 | */ | |
50 | char pad[256]; | |
31db9f7c AB |
51 | }; |
52 | }; | |
53 | #define FS_PATH_INLINE_SIZE \ | |
54 | (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) | |
55 | ||
56 | ||
57 | /* reused for each extent */ | |
58 | struct clone_root { | |
59 | struct btrfs_root *root; | |
60 | u64 ino; | |
61 | u64 offset; | |
62 | ||
63 | u64 found_refs; | |
64 | }; | |
65 | ||
66 | #define SEND_CTX_MAX_NAME_CACHE_SIZE 128 | |
67 | #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2) | |
68 | ||
69 | struct send_ctx { | |
70 | struct file *send_filp; | |
71 | loff_t send_off; | |
72 | char *send_buf; | |
73 | u32 send_size; | |
74 | u32 send_max_size; | |
75 | u64 total_send_size; | |
76 | u64 cmd_send_size[BTRFS_SEND_C_MAX + 1]; | |
cb95e7bf | 77 | u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ |
31db9f7c | 78 | |
31db9f7c AB |
79 | struct btrfs_root *send_root; |
80 | struct btrfs_root *parent_root; | |
81 | struct clone_root *clone_roots; | |
82 | int clone_roots_cnt; | |
83 | ||
84 | /* current state of the compare_tree call */ | |
85 | struct btrfs_path *left_path; | |
86 | struct btrfs_path *right_path; | |
87 | struct btrfs_key *cmp_key; | |
88 | ||
89 | /* | |
90 | * infos of the currently processed inode. In case of deleted inodes, | |
91 | * these are the values from the deleted inode. | |
92 | */ | |
93 | u64 cur_ino; | |
94 | u64 cur_inode_gen; | |
95 | int cur_inode_new; | |
96 | int cur_inode_new_gen; | |
97 | int cur_inode_deleted; | |
31db9f7c AB |
98 | u64 cur_inode_size; |
99 | u64 cur_inode_mode; | |
644d1940 | 100 | u64 cur_inode_rdev; |
16e7549f | 101 | u64 cur_inode_last_extent; |
ffa7c429 | 102 | u64 cur_inode_next_write_offset; |
46b2f459 | 103 | bool ignore_cur_inode; |
31db9f7c AB |
104 | |
105 | u64 send_progress; | |
106 | ||
107 | struct list_head new_refs; | |
108 | struct list_head deleted_refs; | |
109 | ||
110 | struct radix_tree_root name_cache; | |
111 | struct list_head name_cache_list; | |
112 | int name_cache_size; | |
113 | ||
2131bcd3 LB |
114 | struct file_ra_state ra; |
115 | ||
31db9f7c | 116 | char *read_buf; |
9f03740a FDBM |
117 | |
118 | /* | |
119 | * We process inodes by their increasing order, so if before an | |
120 | * incremental send we reverse the parent/child relationship of | |
121 | * directories such that a directory with a lower inode number was | |
122 | * the parent of a directory with a higher inode number, and the one | |
123 | * becoming the new parent got renamed too, we can't rename/move the | |
124 | * directory with lower inode number when we finish processing it - we | |
125 | * must process the directory with higher inode number first, then | |
126 | * rename/move it and then rename/move the directory with lower inode | |
127 | * number. Example follows. | |
128 | * | |
129 | * Tree state when the first send was performed: | |
130 | * | |
131 | * . | |
132 | * |-- a (ino 257) | |
133 | * |-- b (ino 258) | |
134 | * | | |
135 | * | | |
136 | * |-- c (ino 259) | |
137 | * | |-- d (ino 260) | |
138 | * | | |
139 | * |-- c2 (ino 261) | |
140 | * | |
141 | * Tree state when the second (incremental) send is performed: | |
142 | * | |
143 | * . | |
144 | * |-- a (ino 257) | |
145 | * |-- b (ino 258) | |
146 | * |-- c2 (ino 261) | |
147 | * |-- d2 (ino 260) | |
148 | * |-- cc (ino 259) | |
149 | * | |
150 | * The sequence of steps that lead to the second state was: | |
151 | * | |
152 | * mv /a/b/c/d /a/b/c2/d2 | |
153 | * mv /a/b/c /a/b/c2/d2/cc | |
154 | * | |
155 | * "c" has lower inode number, but we can't move it (2nd mv operation) | |
156 | * before we move "d", which has higher inode number. | |
157 | * | |
158 | * So we just memorize which move/rename operations must be performed | |
159 | * later when their respective parent is processed and moved/renamed. | |
160 | */ | |
161 | ||
162 | /* Indexed by parent directory inode number. */ | |
163 | struct rb_root pending_dir_moves; | |
164 | ||
165 | /* | |
166 | * Reverse index, indexed by the inode number of a directory that | |
167 | * is waiting for the move/rename of its immediate parent before its | |
168 | * own move/rename can be performed. | |
169 | */ | |
170 | struct rb_root waiting_dir_moves; | |
9dc44214 FM |
171 | |
172 | /* | |
173 | * A directory that is going to be rm'ed might have a child directory | |
174 | * which is in the pending directory moves index above. In this case, | |
175 | * the directory can only be removed after the move/rename of its child | |
176 | * is performed. Example: | |
177 | * | |
178 | * Parent snapshot: | |
179 | * | |
180 | * . (ino 256) | |
181 | * |-- a/ (ino 257) | |
182 | * |-- b/ (ino 258) | |
183 | * |-- c/ (ino 259) | |
184 | * | |-- x/ (ino 260) | |
185 | * | | |
186 | * |-- y/ (ino 261) | |
187 | * | |
188 | * Send snapshot: | |
189 | * | |
190 | * . (ino 256) | |
191 | * |-- a/ (ino 257) | |
192 | * |-- b/ (ino 258) | |
193 | * |-- YY/ (ino 261) | |
194 | * |-- x/ (ino 260) | |
195 | * | |
196 | * Sequence of steps that lead to the send snapshot: | |
197 | * rm -f /a/b/c/foo.txt | |
198 | * mv /a/b/y /a/b/YY | |
199 | * mv /a/b/c/x /a/b/YY | |
200 | * rmdir /a/b/c | |
201 | * | |
202 | * When the child is processed, its move/rename is delayed until its | |
203 | * parent is processed (as explained above), but all other operations | |
204 | * like update utimes, chown, chgrp, etc, are performed and the paths | |
205 | * that it uses for those operations must use the orphanized name of | |
206 | * its parent (the directory we're going to rm later), so we need to | |
207 | * memorize that name. | |
208 | * | |
209 | * Indexed by the inode number of the directory to be deleted. | |
210 | */ | |
211 | struct rb_root orphan_dirs; | |
9f03740a FDBM |
212 | }; |
213 | ||
214 | struct pending_dir_move { | |
215 | struct rb_node node; | |
216 | struct list_head list; | |
217 | u64 parent_ino; | |
218 | u64 ino; | |
219 | u64 gen; | |
220 | struct list_head update_refs; | |
221 | }; | |
222 | ||
223 | struct waiting_dir_move { | |
224 | struct rb_node node; | |
225 | u64 ino; | |
9dc44214 FM |
226 | /* |
227 | * There might be some directory that could not be removed because it | |
228 | * was waiting for this directory inode to be moved first. Therefore | |
229 | * after this directory is moved, we can try to rmdir the ino rmdir_ino. | |
230 | */ | |
231 | u64 rmdir_ino; | |
8b191a68 | 232 | bool orphanized; |
9dc44214 FM |
233 | }; |
234 | ||
235 | struct orphan_dir_info { | |
236 | struct rb_node node; | |
237 | u64 ino; | |
238 | u64 gen; | |
0f96f517 | 239 | u64 last_dir_index_offset; |
31db9f7c AB |
240 | }; |
241 | ||
242 | struct name_cache_entry { | |
243 | struct list_head list; | |
7e0926fe AB |
244 | /* |
245 | * radix_tree has only 32bit entries but we need to handle 64bit inums. | |
246 | * We use the lower 32bit of the 64bit inum to store it in the tree. If | |
247 | * more then one inum would fall into the same entry, we use radix_list | |
248 | * to store the additional entries. radix_list is also used to store | |
249 | * entries where two entries have the same inum but different | |
250 | * generations. | |
251 | */ | |
252 | struct list_head radix_list; | |
31db9f7c AB |
253 | u64 ino; |
254 | u64 gen; | |
255 | u64 parent_ino; | |
256 | u64 parent_gen; | |
257 | int ret; | |
258 | int need_later_update; | |
259 | int name_len; | |
260 | char name[]; | |
261 | }; | |
262 | ||
e67c718b | 263 | __cold |
95155585 FM |
264 | static void inconsistent_snapshot_error(struct send_ctx *sctx, |
265 | enum btrfs_compare_tree_result result, | |
266 | const char *what) | |
267 | { | |
268 | const char *result_string; | |
269 | ||
270 | switch (result) { | |
271 | case BTRFS_COMPARE_TREE_NEW: | |
272 | result_string = "new"; | |
273 | break; | |
274 | case BTRFS_COMPARE_TREE_DELETED: | |
275 | result_string = "deleted"; | |
276 | break; | |
277 | case BTRFS_COMPARE_TREE_CHANGED: | |
278 | result_string = "updated"; | |
279 | break; | |
280 | case BTRFS_COMPARE_TREE_SAME: | |
281 | ASSERT(0); | |
282 | result_string = "unchanged"; | |
283 | break; | |
284 | default: | |
285 | ASSERT(0); | |
286 | result_string = "unexpected"; | |
287 | } | |
288 | ||
289 | btrfs_err(sctx->send_root->fs_info, | |
290 | "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu", | |
291 | result_string, what, sctx->cmp_key->objectid, | |
292 | sctx->send_root->root_key.objectid, | |
293 | (sctx->parent_root ? | |
294 | sctx->parent_root->root_key.objectid : 0)); | |
295 | } | |
296 | ||
9f03740a FDBM |
297 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); |
298 | ||
9dc44214 FM |
299 | static struct waiting_dir_move * |
300 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino); | |
301 | ||
302 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino); | |
303 | ||
16e7549f JB |
304 | static int need_send_hole(struct send_ctx *sctx) |
305 | { | |
306 | return (sctx->parent_root && !sctx->cur_inode_new && | |
307 | !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && | |
308 | S_ISREG(sctx->cur_inode_mode)); | |
309 | } | |
310 | ||
31db9f7c AB |
311 | static void fs_path_reset(struct fs_path *p) |
312 | { | |
313 | if (p->reversed) { | |
314 | p->start = p->buf + p->buf_len - 1; | |
315 | p->end = p->start; | |
316 | *p->start = 0; | |
317 | } else { | |
318 | p->start = p->buf; | |
319 | p->end = p->start; | |
320 | *p->start = 0; | |
321 | } | |
322 | } | |
323 | ||
924794c9 | 324 | static struct fs_path *fs_path_alloc(void) |
31db9f7c AB |
325 | { |
326 | struct fs_path *p; | |
327 | ||
e780b0d1 | 328 | p = kmalloc(sizeof(*p), GFP_KERNEL); |
31db9f7c AB |
329 | if (!p) |
330 | return NULL; | |
331 | p->reversed = 0; | |
31db9f7c AB |
332 | p->buf = p->inline_buf; |
333 | p->buf_len = FS_PATH_INLINE_SIZE; | |
334 | fs_path_reset(p); | |
335 | return p; | |
336 | } | |
337 | ||
924794c9 | 338 | static struct fs_path *fs_path_alloc_reversed(void) |
31db9f7c AB |
339 | { |
340 | struct fs_path *p; | |
341 | ||
924794c9 | 342 | p = fs_path_alloc(); |
31db9f7c AB |
343 | if (!p) |
344 | return NULL; | |
345 | p->reversed = 1; | |
346 | fs_path_reset(p); | |
347 | return p; | |
348 | } | |
349 | ||
924794c9 | 350 | static void fs_path_free(struct fs_path *p) |
31db9f7c AB |
351 | { |
352 | if (!p) | |
353 | return; | |
ace01050 DS |
354 | if (p->buf != p->inline_buf) |
355 | kfree(p->buf); | |
31db9f7c AB |
356 | kfree(p); |
357 | } | |
358 | ||
359 | static int fs_path_len(struct fs_path *p) | |
360 | { | |
361 | return p->end - p->start; | |
362 | } | |
363 | ||
364 | static int fs_path_ensure_buf(struct fs_path *p, int len) | |
365 | { | |
366 | char *tmp_buf; | |
367 | int path_len; | |
368 | int old_buf_len; | |
369 | ||
370 | len++; | |
371 | ||
372 | if (p->buf_len >= len) | |
373 | return 0; | |
374 | ||
cfd4a535 CM |
375 | if (len > PATH_MAX) { |
376 | WARN_ON(1); | |
377 | return -ENOMEM; | |
378 | } | |
379 | ||
1b2782c8 DS |
380 | path_len = p->end - p->start; |
381 | old_buf_len = p->buf_len; | |
382 | ||
ace01050 DS |
383 | /* |
384 | * First time the inline_buf does not suffice | |
385 | */ | |
01a9a8a9 | 386 | if (p->buf == p->inline_buf) { |
e780b0d1 | 387 | tmp_buf = kmalloc(len, GFP_KERNEL); |
01a9a8a9 FM |
388 | if (tmp_buf) |
389 | memcpy(tmp_buf, p->buf, old_buf_len); | |
390 | } else { | |
e780b0d1 | 391 | tmp_buf = krealloc(p->buf, len, GFP_KERNEL); |
01a9a8a9 | 392 | } |
9c9ca00b DS |
393 | if (!tmp_buf) |
394 | return -ENOMEM; | |
395 | p->buf = tmp_buf; | |
396 | /* | |
397 | * The real size of the buffer is bigger, this will let the fast path | |
398 | * happen most of the time | |
399 | */ | |
400 | p->buf_len = ksize(p->buf); | |
ace01050 | 401 | |
31db9f7c AB |
402 | if (p->reversed) { |
403 | tmp_buf = p->buf + old_buf_len - path_len - 1; | |
404 | p->end = p->buf + p->buf_len - 1; | |
405 | p->start = p->end - path_len; | |
406 | memmove(p->start, tmp_buf, path_len + 1); | |
407 | } else { | |
408 | p->start = p->buf; | |
409 | p->end = p->start + path_len; | |
410 | } | |
411 | return 0; | |
412 | } | |
413 | ||
b23ab57d DS |
414 | static int fs_path_prepare_for_add(struct fs_path *p, int name_len, |
415 | char **prepared) | |
31db9f7c AB |
416 | { |
417 | int ret; | |
418 | int new_len; | |
419 | ||
420 | new_len = p->end - p->start + name_len; | |
421 | if (p->start != p->end) | |
422 | new_len++; | |
423 | ret = fs_path_ensure_buf(p, new_len); | |
424 | if (ret < 0) | |
425 | goto out; | |
426 | ||
427 | if (p->reversed) { | |
428 | if (p->start != p->end) | |
429 | *--p->start = '/'; | |
430 | p->start -= name_len; | |
b23ab57d | 431 | *prepared = p->start; |
31db9f7c AB |
432 | } else { |
433 | if (p->start != p->end) | |
434 | *p->end++ = '/'; | |
b23ab57d | 435 | *prepared = p->end; |
31db9f7c AB |
436 | p->end += name_len; |
437 | *p->end = 0; | |
438 | } | |
439 | ||
440 | out: | |
441 | return ret; | |
442 | } | |
443 | ||
444 | static int fs_path_add(struct fs_path *p, const char *name, int name_len) | |
445 | { | |
446 | int ret; | |
b23ab57d | 447 | char *prepared; |
31db9f7c | 448 | |
b23ab57d | 449 | ret = fs_path_prepare_for_add(p, name_len, &prepared); |
31db9f7c AB |
450 | if (ret < 0) |
451 | goto out; | |
b23ab57d | 452 | memcpy(prepared, name, name_len); |
31db9f7c AB |
453 | |
454 | out: | |
455 | return ret; | |
456 | } | |
457 | ||
458 | static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) | |
459 | { | |
460 | int ret; | |
b23ab57d | 461 | char *prepared; |
31db9f7c | 462 | |
b23ab57d | 463 | ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); |
31db9f7c AB |
464 | if (ret < 0) |
465 | goto out; | |
b23ab57d | 466 | memcpy(prepared, p2->start, p2->end - p2->start); |
31db9f7c AB |
467 | |
468 | out: | |
469 | return ret; | |
470 | } | |
471 | ||
472 | static int fs_path_add_from_extent_buffer(struct fs_path *p, | |
473 | struct extent_buffer *eb, | |
474 | unsigned long off, int len) | |
475 | { | |
476 | int ret; | |
b23ab57d | 477 | char *prepared; |
31db9f7c | 478 | |
b23ab57d | 479 | ret = fs_path_prepare_for_add(p, len, &prepared); |
31db9f7c AB |
480 | if (ret < 0) |
481 | goto out; | |
482 | ||
b23ab57d | 483 | read_extent_buffer(eb, prepared, off, len); |
31db9f7c AB |
484 | |
485 | out: | |
486 | return ret; | |
487 | } | |
488 | ||
31db9f7c AB |
489 | static int fs_path_copy(struct fs_path *p, struct fs_path *from) |
490 | { | |
491 | int ret; | |
492 | ||
493 | p->reversed = from->reversed; | |
494 | fs_path_reset(p); | |
495 | ||
496 | ret = fs_path_add_path(p, from); | |
497 | ||
498 | return ret; | |
499 | } | |
500 | ||
501 | ||
502 | static void fs_path_unreverse(struct fs_path *p) | |
503 | { | |
504 | char *tmp; | |
505 | int len; | |
506 | ||
507 | if (!p->reversed) | |
508 | return; | |
509 | ||
510 | tmp = p->start; | |
511 | len = p->end - p->start; | |
512 | p->start = p->buf; | |
513 | p->end = p->start + len; | |
514 | memmove(p->start, tmp, len + 1); | |
515 | p->reversed = 0; | |
516 | } | |
517 | ||
518 | static struct btrfs_path *alloc_path_for_send(void) | |
519 | { | |
520 | struct btrfs_path *path; | |
521 | ||
522 | path = btrfs_alloc_path(); | |
523 | if (!path) | |
524 | return NULL; | |
525 | path->search_commit_root = 1; | |
526 | path->skip_locking = 1; | |
3f8a18cc | 527 | path->need_commit_sem = 1; |
31db9f7c AB |
528 | return path; |
529 | } | |
530 | ||
48a3b636 | 531 | static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) |
31db9f7c AB |
532 | { |
533 | int ret; | |
31db9f7c AB |
534 | u32 pos = 0; |
535 | ||
31db9f7c | 536 | while (pos < len) { |
8e93157b | 537 | ret = kernel_write(filp, buf + pos, len - pos, off); |
31db9f7c AB |
538 | /* TODO handle that correctly */ |
539 | /*if (ret == -ERESTARTSYS) { | |
540 | continue; | |
541 | }*/ | |
542 | if (ret < 0) | |
8e93157b | 543 | return ret; |
31db9f7c | 544 | if (ret == 0) { |
8e93157b | 545 | return -EIO; |
31db9f7c AB |
546 | } |
547 | pos += ret; | |
548 | } | |
549 | ||
8e93157b | 550 | return 0; |
31db9f7c AB |
551 | } |
552 | ||
553 | static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) | |
554 | { | |
555 | struct btrfs_tlv_header *hdr; | |
556 | int total_len = sizeof(*hdr) + len; | |
557 | int left = sctx->send_max_size - sctx->send_size; | |
558 | ||
559 | if (unlikely(left < total_len)) | |
560 | return -EOVERFLOW; | |
561 | ||
562 | hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); | |
563 | hdr->tlv_type = cpu_to_le16(attr); | |
564 | hdr->tlv_len = cpu_to_le16(len); | |
565 | memcpy(hdr + 1, data, len); | |
566 | sctx->send_size += total_len; | |
567 | ||
568 | return 0; | |
569 | } | |
570 | ||
95bc79d5 DS |
571 | #define TLV_PUT_DEFINE_INT(bits) \ |
572 | static int tlv_put_u##bits(struct send_ctx *sctx, \ | |
573 | u##bits attr, u##bits value) \ | |
574 | { \ | |
575 | __le##bits __tmp = cpu_to_le##bits(value); \ | |
576 | return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ | |
577 | } | |
31db9f7c | 578 | |
95bc79d5 | 579 | TLV_PUT_DEFINE_INT(64) |
31db9f7c AB |
580 | |
581 | static int tlv_put_string(struct send_ctx *sctx, u16 attr, | |
582 | const char *str, int len) | |
583 | { | |
584 | if (len == -1) | |
585 | len = strlen(str); | |
586 | return tlv_put(sctx, attr, str, len); | |
587 | } | |
588 | ||
589 | static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, | |
590 | const u8 *uuid) | |
591 | { | |
592 | return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); | |
593 | } | |
594 | ||
31db9f7c AB |
595 | static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, |
596 | struct extent_buffer *eb, | |
597 | struct btrfs_timespec *ts) | |
598 | { | |
599 | struct btrfs_timespec bts; | |
600 | read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); | |
601 | return tlv_put(sctx, attr, &bts, sizeof(bts)); | |
602 | } | |
603 | ||
604 | ||
895a72be | 605 | #define TLV_PUT(sctx, attrtype, data, attrlen) \ |
31db9f7c | 606 | do { \ |
895a72be | 607 | ret = tlv_put(sctx, attrtype, data, attrlen); \ |
31db9f7c AB |
608 | if (ret < 0) \ |
609 | goto tlv_put_failure; \ | |
610 | } while (0) | |
611 | ||
612 | #define TLV_PUT_INT(sctx, attrtype, bits, value) \ | |
613 | do { \ | |
614 | ret = tlv_put_u##bits(sctx, attrtype, value); \ | |
615 | if (ret < 0) \ | |
616 | goto tlv_put_failure; \ | |
617 | } while (0) | |
618 | ||
619 | #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) | |
620 | #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) | |
621 | #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) | |
622 | #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) | |
623 | #define TLV_PUT_STRING(sctx, attrtype, str, len) \ | |
624 | do { \ | |
625 | ret = tlv_put_string(sctx, attrtype, str, len); \ | |
626 | if (ret < 0) \ | |
627 | goto tlv_put_failure; \ | |
628 | } while (0) | |
629 | #define TLV_PUT_PATH(sctx, attrtype, p) \ | |
630 | do { \ | |
631 | ret = tlv_put_string(sctx, attrtype, p->start, \ | |
632 | p->end - p->start); \ | |
633 | if (ret < 0) \ | |
634 | goto tlv_put_failure; \ | |
635 | } while(0) | |
636 | #define TLV_PUT_UUID(sctx, attrtype, uuid) \ | |
637 | do { \ | |
638 | ret = tlv_put_uuid(sctx, attrtype, uuid); \ | |
639 | if (ret < 0) \ | |
640 | goto tlv_put_failure; \ | |
641 | } while (0) | |
31db9f7c AB |
642 | #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ |
643 | do { \ | |
644 | ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ | |
645 | if (ret < 0) \ | |
646 | goto tlv_put_failure; \ | |
647 | } while (0) | |
648 | ||
649 | static int send_header(struct send_ctx *sctx) | |
650 | { | |
651 | struct btrfs_stream_header hdr; | |
652 | ||
653 | strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); | |
654 | hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); | |
655 | ||
1bcea355 AJ |
656 | return write_buf(sctx->send_filp, &hdr, sizeof(hdr), |
657 | &sctx->send_off); | |
31db9f7c AB |
658 | } |
659 | ||
660 | /* | |
661 | * For each command/item we want to send to userspace, we call this function. | |
662 | */ | |
663 | static int begin_cmd(struct send_ctx *sctx, int cmd) | |
664 | { | |
665 | struct btrfs_cmd_header *hdr; | |
666 | ||
fae7f21c | 667 | if (WARN_ON(!sctx->send_buf)) |
31db9f7c | 668 | return -EINVAL; |
31db9f7c AB |
669 | |
670 | BUG_ON(sctx->send_size); | |
671 | ||
672 | sctx->send_size += sizeof(*hdr); | |
673 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
674 | hdr->cmd = cpu_to_le16(cmd); | |
675 | ||
676 | return 0; | |
677 | } | |
678 | ||
679 | static int send_cmd(struct send_ctx *sctx) | |
680 | { | |
681 | int ret; | |
682 | struct btrfs_cmd_header *hdr; | |
683 | u32 crc; | |
684 | ||
685 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
686 | hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr)); | |
687 | hdr->crc = 0; | |
688 | ||
9678c543 | 689 | crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); |
31db9f7c AB |
690 | hdr->crc = cpu_to_le32(crc); |
691 | ||
1bcea355 AJ |
692 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, |
693 | &sctx->send_off); | |
31db9f7c AB |
694 | |
695 | sctx->total_send_size += sctx->send_size; | |
696 | sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size; | |
697 | sctx->send_size = 0; | |
698 | ||
699 | return ret; | |
700 | } | |
701 | ||
702 | /* | |
703 | * Sends a move instruction to user space | |
704 | */ | |
705 | static int send_rename(struct send_ctx *sctx, | |
706 | struct fs_path *from, struct fs_path *to) | |
707 | { | |
04ab956e | 708 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
709 | int ret; |
710 | ||
04ab956e | 711 | btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start); |
31db9f7c AB |
712 | |
713 | ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); | |
714 | if (ret < 0) | |
715 | goto out; | |
716 | ||
717 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); | |
718 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); | |
719 | ||
720 | ret = send_cmd(sctx); | |
721 | ||
722 | tlv_put_failure: | |
723 | out: | |
724 | return ret; | |
725 | } | |
726 | ||
727 | /* | |
728 | * Sends a link instruction to user space | |
729 | */ | |
730 | static int send_link(struct send_ctx *sctx, | |
731 | struct fs_path *path, struct fs_path *lnk) | |
732 | { | |
04ab956e | 733 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
734 | int ret; |
735 | ||
04ab956e | 736 | btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start); |
31db9f7c AB |
737 | |
738 | ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); | |
739 | if (ret < 0) | |
740 | goto out; | |
741 | ||
742 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
743 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); | |
744 | ||
745 | ret = send_cmd(sctx); | |
746 | ||
747 | tlv_put_failure: | |
748 | out: | |
749 | return ret; | |
750 | } | |
751 | ||
752 | /* | |
753 | * Sends an unlink instruction to user space | |
754 | */ | |
755 | static int send_unlink(struct send_ctx *sctx, struct fs_path *path) | |
756 | { | |
04ab956e | 757 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
758 | int ret; |
759 | ||
04ab956e | 760 | btrfs_debug(fs_info, "send_unlink %s", path->start); |
31db9f7c AB |
761 | |
762 | ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); | |
763 | if (ret < 0) | |
764 | goto out; | |
765 | ||
766 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
767 | ||
768 | ret = send_cmd(sctx); | |
769 | ||
770 | tlv_put_failure: | |
771 | out: | |
772 | return ret; | |
773 | } | |
774 | ||
775 | /* | |
776 | * Sends a rmdir instruction to user space | |
777 | */ | |
778 | static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) | |
779 | { | |
04ab956e | 780 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
781 | int ret; |
782 | ||
04ab956e | 783 | btrfs_debug(fs_info, "send_rmdir %s", path->start); |
31db9f7c AB |
784 | |
785 | ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); | |
786 | if (ret < 0) | |
787 | goto out; | |
788 | ||
789 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
790 | ||
791 | ret = send_cmd(sctx); | |
792 | ||
793 | tlv_put_failure: | |
794 | out: | |
795 | return ret; | |
796 | } | |
797 | ||
798 | /* | |
799 | * Helper function to retrieve some fields from an inode item. | |
800 | */ | |
3f8a18cc JB |
801 | static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, |
802 | u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, | |
803 | u64 *gid, u64 *rdev) | |
31db9f7c AB |
804 | { |
805 | int ret; | |
806 | struct btrfs_inode_item *ii; | |
807 | struct btrfs_key key; | |
31db9f7c AB |
808 | |
809 | key.objectid = ino; | |
810 | key.type = BTRFS_INODE_ITEM_KEY; | |
811 | key.offset = 0; | |
812 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
31db9f7c | 813 | if (ret) { |
3f8a18cc JB |
814 | if (ret > 0) |
815 | ret = -ENOENT; | |
816 | return ret; | |
31db9f7c AB |
817 | } |
818 | ||
819 | ii = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
820 | struct btrfs_inode_item); | |
821 | if (size) | |
822 | *size = btrfs_inode_size(path->nodes[0], ii); | |
823 | if (gen) | |
824 | *gen = btrfs_inode_generation(path->nodes[0], ii); | |
825 | if (mode) | |
826 | *mode = btrfs_inode_mode(path->nodes[0], ii); | |
827 | if (uid) | |
828 | *uid = btrfs_inode_uid(path->nodes[0], ii); | |
829 | if (gid) | |
830 | *gid = btrfs_inode_gid(path->nodes[0], ii); | |
85a7b33b AB |
831 | if (rdev) |
832 | *rdev = btrfs_inode_rdev(path->nodes[0], ii); | |
31db9f7c | 833 | |
3f8a18cc JB |
834 | return ret; |
835 | } | |
836 | ||
837 | static int get_inode_info(struct btrfs_root *root, | |
838 | u64 ino, u64 *size, u64 *gen, | |
839 | u64 *mode, u64 *uid, u64 *gid, | |
840 | u64 *rdev) | |
841 | { | |
842 | struct btrfs_path *path; | |
843 | int ret; | |
844 | ||
845 | path = alloc_path_for_send(); | |
846 | if (!path) | |
847 | return -ENOMEM; | |
848 | ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, | |
849 | rdev); | |
31db9f7c AB |
850 | btrfs_free_path(path); |
851 | return ret; | |
852 | } | |
853 | ||
854 | typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, | |
855 | struct fs_path *p, | |
856 | void *ctx); | |
857 | ||
858 | /* | |
96b5bd77 JS |
859 | * Helper function to iterate the entries in ONE btrfs_inode_ref or |
860 | * btrfs_inode_extref. | |
31db9f7c AB |
861 | * The iterate callback may return a non zero value to stop iteration. This can |
862 | * be a negative value for error codes or 1 to simply stop it. | |
863 | * | |
96b5bd77 | 864 | * path must point to the INODE_REF or INODE_EXTREF when called. |
31db9f7c | 865 | */ |
924794c9 | 866 | static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
867 | struct btrfs_key *found_key, int resolve, |
868 | iterate_inode_ref_t iterate, void *ctx) | |
869 | { | |
96b5bd77 | 870 | struct extent_buffer *eb = path->nodes[0]; |
31db9f7c AB |
871 | struct btrfs_item *item; |
872 | struct btrfs_inode_ref *iref; | |
96b5bd77 | 873 | struct btrfs_inode_extref *extref; |
31db9f7c AB |
874 | struct btrfs_path *tmp_path; |
875 | struct fs_path *p; | |
96b5bd77 | 876 | u32 cur = 0; |
31db9f7c | 877 | u32 total; |
96b5bd77 | 878 | int slot = path->slots[0]; |
31db9f7c AB |
879 | u32 name_len; |
880 | char *start; | |
881 | int ret = 0; | |
96b5bd77 | 882 | int num = 0; |
31db9f7c | 883 | int index; |
96b5bd77 JS |
884 | u64 dir; |
885 | unsigned long name_off; | |
886 | unsigned long elem_size; | |
887 | unsigned long ptr; | |
31db9f7c | 888 | |
924794c9 | 889 | p = fs_path_alloc_reversed(); |
31db9f7c AB |
890 | if (!p) |
891 | return -ENOMEM; | |
892 | ||
893 | tmp_path = alloc_path_for_send(); | |
894 | if (!tmp_path) { | |
924794c9 | 895 | fs_path_free(p); |
31db9f7c AB |
896 | return -ENOMEM; |
897 | } | |
898 | ||
31db9f7c | 899 | |
96b5bd77 JS |
900 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
901 | ptr = (unsigned long)btrfs_item_ptr(eb, slot, | |
902 | struct btrfs_inode_ref); | |
dd3cc16b | 903 | item = btrfs_item_nr(slot); |
96b5bd77 JS |
904 | total = btrfs_item_size(eb, item); |
905 | elem_size = sizeof(*iref); | |
906 | } else { | |
907 | ptr = btrfs_item_ptr_offset(eb, slot); | |
908 | total = btrfs_item_size_nr(eb, slot); | |
909 | elem_size = sizeof(*extref); | |
910 | } | |
911 | ||
31db9f7c AB |
912 | while (cur < total) { |
913 | fs_path_reset(p); | |
914 | ||
96b5bd77 JS |
915 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
916 | iref = (struct btrfs_inode_ref *)(ptr + cur); | |
917 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
918 | name_off = (unsigned long)(iref + 1); | |
919 | index = btrfs_inode_ref_index(eb, iref); | |
920 | dir = found_key->offset; | |
921 | } else { | |
922 | extref = (struct btrfs_inode_extref *)(ptr + cur); | |
923 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
924 | name_off = (unsigned long)&extref->name; | |
925 | index = btrfs_inode_extref_index(eb, extref); | |
926 | dir = btrfs_inode_extref_parent(eb, extref); | |
927 | } | |
928 | ||
31db9f7c | 929 | if (resolve) { |
96b5bd77 JS |
930 | start = btrfs_ref_to_path(root, tmp_path, name_len, |
931 | name_off, eb, dir, | |
932 | p->buf, p->buf_len); | |
31db9f7c AB |
933 | if (IS_ERR(start)) { |
934 | ret = PTR_ERR(start); | |
935 | goto out; | |
936 | } | |
937 | if (start < p->buf) { | |
938 | /* overflow , try again with larger buffer */ | |
939 | ret = fs_path_ensure_buf(p, | |
940 | p->buf_len + p->buf - start); | |
941 | if (ret < 0) | |
942 | goto out; | |
96b5bd77 JS |
943 | start = btrfs_ref_to_path(root, tmp_path, |
944 | name_len, name_off, | |
945 | eb, dir, | |
946 | p->buf, p->buf_len); | |
31db9f7c AB |
947 | if (IS_ERR(start)) { |
948 | ret = PTR_ERR(start); | |
949 | goto out; | |
950 | } | |
951 | BUG_ON(start < p->buf); | |
952 | } | |
953 | p->start = start; | |
954 | } else { | |
96b5bd77 JS |
955 | ret = fs_path_add_from_extent_buffer(p, eb, name_off, |
956 | name_len); | |
31db9f7c AB |
957 | if (ret < 0) |
958 | goto out; | |
959 | } | |
960 | ||
96b5bd77 JS |
961 | cur += elem_size + name_len; |
962 | ret = iterate(num, dir, index, p, ctx); | |
31db9f7c AB |
963 | if (ret) |
964 | goto out; | |
31db9f7c AB |
965 | num++; |
966 | } | |
967 | ||
968 | out: | |
969 | btrfs_free_path(tmp_path); | |
924794c9 | 970 | fs_path_free(p); |
31db9f7c AB |
971 | return ret; |
972 | } | |
973 | ||
974 | typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, | |
975 | const char *name, int name_len, | |
976 | const char *data, int data_len, | |
977 | u8 type, void *ctx); | |
978 | ||
979 | /* | |
980 | * Helper function to iterate the entries in ONE btrfs_dir_item. | |
981 | * The iterate callback may return a non zero value to stop iteration. This can | |
982 | * be a negative value for error codes or 1 to simply stop it. | |
983 | * | |
984 | * path must point to the dir item when called. | |
985 | */ | |
924794c9 | 986 | static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
987 | iterate_dir_item_t iterate, void *ctx) |
988 | { | |
989 | int ret = 0; | |
990 | struct extent_buffer *eb; | |
991 | struct btrfs_item *item; | |
992 | struct btrfs_dir_item *di; | |
31db9f7c AB |
993 | struct btrfs_key di_key; |
994 | char *buf = NULL; | |
7e3ae33e | 995 | int buf_len; |
31db9f7c AB |
996 | u32 name_len; |
997 | u32 data_len; | |
998 | u32 cur; | |
999 | u32 len; | |
1000 | u32 total; | |
1001 | int slot; | |
1002 | int num; | |
1003 | u8 type; | |
1004 | ||
4395e0c4 FM |
1005 | /* |
1006 | * Start with a small buffer (1 page). If later we end up needing more | |
1007 | * space, which can happen for xattrs on a fs with a leaf size greater | |
1008 | * then the page size, attempt to increase the buffer. Typically xattr | |
1009 | * values are small. | |
1010 | */ | |
1011 | buf_len = PATH_MAX; | |
e780b0d1 | 1012 | buf = kmalloc(buf_len, GFP_KERNEL); |
31db9f7c AB |
1013 | if (!buf) { |
1014 | ret = -ENOMEM; | |
1015 | goto out; | |
1016 | } | |
1017 | ||
31db9f7c AB |
1018 | eb = path->nodes[0]; |
1019 | slot = path->slots[0]; | |
dd3cc16b | 1020 | item = btrfs_item_nr(slot); |
31db9f7c AB |
1021 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
1022 | cur = 0; | |
1023 | len = 0; | |
1024 | total = btrfs_item_size(eb, item); | |
1025 | ||
1026 | num = 0; | |
1027 | while (cur < total) { | |
1028 | name_len = btrfs_dir_name_len(eb, di); | |
1029 | data_len = btrfs_dir_data_len(eb, di); | |
1030 | type = btrfs_dir_type(eb, di); | |
1031 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1032 | ||
7e3ae33e FM |
1033 | if (type == BTRFS_FT_XATTR) { |
1034 | if (name_len > XATTR_NAME_MAX) { | |
1035 | ret = -ENAMETOOLONG; | |
1036 | goto out; | |
1037 | } | |
da17066c JM |
1038 | if (name_len + data_len > |
1039 | BTRFS_MAX_XATTR_SIZE(root->fs_info)) { | |
7e3ae33e FM |
1040 | ret = -E2BIG; |
1041 | goto out; | |
1042 | } | |
1043 | } else { | |
1044 | /* | |
1045 | * Path too long | |
1046 | */ | |
4395e0c4 | 1047 | if (name_len + data_len > PATH_MAX) { |
7e3ae33e FM |
1048 | ret = -ENAMETOOLONG; |
1049 | goto out; | |
1050 | } | |
31db9f7c AB |
1051 | } |
1052 | ||
4395e0c4 FM |
1053 | if (name_len + data_len > buf_len) { |
1054 | buf_len = name_len + data_len; | |
1055 | if (is_vmalloc_addr(buf)) { | |
1056 | vfree(buf); | |
1057 | buf = NULL; | |
1058 | } else { | |
1059 | char *tmp = krealloc(buf, buf_len, | |
e780b0d1 | 1060 | GFP_KERNEL | __GFP_NOWARN); |
4395e0c4 FM |
1061 | |
1062 | if (!tmp) | |
1063 | kfree(buf); | |
1064 | buf = tmp; | |
1065 | } | |
1066 | if (!buf) { | |
f11f7441 | 1067 | buf = kvmalloc(buf_len, GFP_KERNEL); |
4395e0c4 FM |
1068 | if (!buf) { |
1069 | ret = -ENOMEM; | |
1070 | goto out; | |
1071 | } | |
1072 | } | |
1073 | } | |
1074 | ||
31db9f7c AB |
1075 | read_extent_buffer(eb, buf, (unsigned long)(di + 1), |
1076 | name_len + data_len); | |
1077 | ||
1078 | len = sizeof(*di) + name_len + data_len; | |
1079 | di = (struct btrfs_dir_item *)((char *)di + len); | |
1080 | cur += len; | |
1081 | ||
1082 | ret = iterate(num, &di_key, buf, name_len, buf + name_len, | |
1083 | data_len, type, ctx); | |
1084 | if (ret < 0) | |
1085 | goto out; | |
1086 | if (ret) { | |
1087 | ret = 0; | |
1088 | goto out; | |
1089 | } | |
1090 | ||
1091 | num++; | |
1092 | } | |
1093 | ||
1094 | out: | |
4395e0c4 | 1095 | kvfree(buf); |
31db9f7c AB |
1096 | return ret; |
1097 | } | |
1098 | ||
1099 | static int __copy_first_ref(int num, u64 dir, int index, | |
1100 | struct fs_path *p, void *ctx) | |
1101 | { | |
1102 | int ret; | |
1103 | struct fs_path *pt = ctx; | |
1104 | ||
1105 | ret = fs_path_copy(pt, p); | |
1106 | if (ret < 0) | |
1107 | return ret; | |
1108 | ||
1109 | /* we want the first only */ | |
1110 | return 1; | |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * Retrieve the first path of an inode. If an inode has more then one | |
1115 | * ref/hardlink, this is ignored. | |
1116 | */ | |
924794c9 | 1117 | static int get_inode_path(struct btrfs_root *root, |
31db9f7c AB |
1118 | u64 ino, struct fs_path *path) |
1119 | { | |
1120 | int ret; | |
1121 | struct btrfs_key key, found_key; | |
1122 | struct btrfs_path *p; | |
1123 | ||
1124 | p = alloc_path_for_send(); | |
1125 | if (!p) | |
1126 | return -ENOMEM; | |
1127 | ||
1128 | fs_path_reset(path); | |
1129 | ||
1130 | key.objectid = ino; | |
1131 | key.type = BTRFS_INODE_REF_KEY; | |
1132 | key.offset = 0; | |
1133 | ||
1134 | ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); | |
1135 | if (ret < 0) | |
1136 | goto out; | |
1137 | if (ret) { | |
1138 | ret = 1; | |
1139 | goto out; | |
1140 | } | |
1141 | btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); | |
1142 | if (found_key.objectid != ino || | |
96b5bd77 JS |
1143 | (found_key.type != BTRFS_INODE_REF_KEY && |
1144 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1145 | ret = -ENOENT; |
1146 | goto out; | |
1147 | } | |
1148 | ||
924794c9 TI |
1149 | ret = iterate_inode_ref(root, p, &found_key, 1, |
1150 | __copy_first_ref, path); | |
31db9f7c AB |
1151 | if (ret < 0) |
1152 | goto out; | |
1153 | ret = 0; | |
1154 | ||
1155 | out: | |
1156 | btrfs_free_path(p); | |
1157 | return ret; | |
1158 | } | |
1159 | ||
1160 | struct backref_ctx { | |
1161 | struct send_ctx *sctx; | |
1162 | ||
3f8a18cc | 1163 | struct btrfs_path *path; |
31db9f7c AB |
1164 | /* number of total found references */ |
1165 | u64 found; | |
1166 | ||
1167 | /* | |
1168 | * used for clones found in send_root. clones found behind cur_objectid | |
1169 | * and cur_offset are not considered as allowed clones. | |
1170 | */ | |
1171 | u64 cur_objectid; | |
1172 | u64 cur_offset; | |
1173 | ||
1174 | /* may be truncated in case it's the last extent in a file */ | |
1175 | u64 extent_len; | |
1176 | ||
619d8c4e FM |
1177 | /* data offset in the file extent item */ |
1178 | u64 data_offset; | |
1179 | ||
31db9f7c | 1180 | /* Just to check for bugs in backref resolving */ |
ee849c04 | 1181 | int found_itself; |
31db9f7c AB |
1182 | }; |
1183 | ||
1184 | static int __clone_root_cmp_bsearch(const void *key, const void *elt) | |
1185 | { | |
995e01b7 | 1186 | u64 root = (u64)(uintptr_t)key; |
31db9f7c AB |
1187 | struct clone_root *cr = (struct clone_root *)elt; |
1188 | ||
1189 | if (root < cr->root->objectid) | |
1190 | return -1; | |
1191 | if (root > cr->root->objectid) | |
1192 | return 1; | |
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | static int __clone_root_cmp_sort(const void *e1, const void *e2) | |
1197 | { | |
1198 | struct clone_root *cr1 = (struct clone_root *)e1; | |
1199 | struct clone_root *cr2 = (struct clone_root *)e2; | |
1200 | ||
1201 | if (cr1->root->objectid < cr2->root->objectid) | |
1202 | return -1; | |
1203 | if (cr1->root->objectid > cr2->root->objectid) | |
1204 | return 1; | |
1205 | return 0; | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * Called for every backref that is found for the current extent. | |
766702ef | 1210 | * Results are collected in sctx->clone_roots->ino/offset/found_refs |
31db9f7c AB |
1211 | */ |
1212 | static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) | |
1213 | { | |
1214 | struct backref_ctx *bctx = ctx_; | |
1215 | struct clone_root *found; | |
1216 | int ret; | |
1217 | u64 i_size; | |
1218 | ||
1219 | /* First check if the root is in the list of accepted clone sources */ | |
995e01b7 | 1220 | found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, |
31db9f7c AB |
1221 | bctx->sctx->clone_roots_cnt, |
1222 | sizeof(struct clone_root), | |
1223 | __clone_root_cmp_bsearch); | |
1224 | if (!found) | |
1225 | return 0; | |
1226 | ||
1227 | if (found->root == bctx->sctx->send_root && | |
1228 | ino == bctx->cur_objectid && | |
1229 | offset == bctx->cur_offset) { | |
ee849c04 | 1230 | bctx->found_itself = 1; |
31db9f7c AB |
1231 | } |
1232 | ||
1233 | /* | |
766702ef | 1234 | * There are inodes that have extents that lie behind its i_size. Don't |
31db9f7c AB |
1235 | * accept clones from these extents. |
1236 | */ | |
3f8a18cc JB |
1237 | ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL, |
1238 | NULL, NULL, NULL); | |
1239 | btrfs_release_path(bctx->path); | |
31db9f7c AB |
1240 | if (ret < 0) |
1241 | return ret; | |
1242 | ||
619d8c4e | 1243 | if (offset + bctx->data_offset + bctx->extent_len > i_size) |
31db9f7c AB |
1244 | return 0; |
1245 | ||
1246 | /* | |
1247 | * Make sure we don't consider clones from send_root that are | |
1248 | * behind the current inode/offset. | |
1249 | */ | |
1250 | if (found->root == bctx->sctx->send_root) { | |
1251 | /* | |
1252 | * TODO for the moment we don't accept clones from the inode | |
1253 | * that is currently send. We may change this when | |
1254 | * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same | |
1255 | * file. | |
1256 | */ | |
1257 | if (ino >= bctx->cur_objectid) | |
1258 | return 0; | |
31db9f7c AB |
1259 | } |
1260 | ||
1261 | bctx->found++; | |
1262 | found->found_refs++; | |
1263 | if (ino < found->ino) { | |
1264 | found->ino = ino; | |
1265 | found->offset = offset; | |
1266 | } else if (found->ino == ino) { | |
1267 | /* | |
1268 | * same extent found more then once in the same file. | |
1269 | */ | |
1270 | if (found->offset > offset + bctx->extent_len) | |
1271 | found->offset = offset; | |
1272 | } | |
1273 | ||
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | /* | |
766702ef AB |
1278 | * Given an inode, offset and extent item, it finds a good clone for a clone |
1279 | * instruction. Returns -ENOENT when none could be found. The function makes | |
1280 | * sure that the returned clone is usable at the point where sending is at the | |
1281 | * moment. This means, that no clones are accepted which lie behind the current | |
1282 | * inode+offset. | |
1283 | * | |
31db9f7c AB |
1284 | * path must point to the extent item when called. |
1285 | */ | |
1286 | static int find_extent_clone(struct send_ctx *sctx, | |
1287 | struct btrfs_path *path, | |
1288 | u64 ino, u64 data_offset, | |
1289 | u64 ino_size, | |
1290 | struct clone_root **found) | |
1291 | { | |
04ab956e | 1292 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
1293 | int ret; |
1294 | int extent_type; | |
1295 | u64 logical; | |
74dd17fb | 1296 | u64 disk_byte; |
31db9f7c AB |
1297 | u64 num_bytes; |
1298 | u64 extent_item_pos; | |
69917e43 | 1299 | u64 flags = 0; |
31db9f7c AB |
1300 | struct btrfs_file_extent_item *fi; |
1301 | struct extent_buffer *eb = path->nodes[0]; | |
35075bb0 | 1302 | struct backref_ctx *backref_ctx = NULL; |
31db9f7c AB |
1303 | struct clone_root *cur_clone_root; |
1304 | struct btrfs_key found_key; | |
1305 | struct btrfs_path *tmp_path; | |
74dd17fb | 1306 | int compressed; |
31db9f7c AB |
1307 | u32 i; |
1308 | ||
1309 | tmp_path = alloc_path_for_send(); | |
1310 | if (!tmp_path) | |
1311 | return -ENOMEM; | |
1312 | ||
3f8a18cc JB |
1313 | /* We only use this path under the commit sem */ |
1314 | tmp_path->need_commit_sem = 0; | |
1315 | ||
e780b0d1 | 1316 | backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL); |
35075bb0 AB |
1317 | if (!backref_ctx) { |
1318 | ret = -ENOMEM; | |
1319 | goto out; | |
1320 | } | |
1321 | ||
3f8a18cc JB |
1322 | backref_ctx->path = tmp_path; |
1323 | ||
31db9f7c AB |
1324 | if (data_offset >= ino_size) { |
1325 | /* | |
1326 | * There may be extents that lie behind the file's size. | |
1327 | * I at least had this in combination with snapshotting while | |
1328 | * writing large files. | |
1329 | */ | |
1330 | ret = 0; | |
1331 | goto out; | |
1332 | } | |
1333 | ||
1334 | fi = btrfs_item_ptr(eb, path->slots[0], | |
1335 | struct btrfs_file_extent_item); | |
1336 | extent_type = btrfs_file_extent_type(eb, fi); | |
1337 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1338 | ret = -ENOENT; | |
1339 | goto out; | |
1340 | } | |
74dd17fb | 1341 | compressed = btrfs_file_extent_compression(eb, fi); |
31db9f7c AB |
1342 | |
1343 | num_bytes = btrfs_file_extent_num_bytes(eb, fi); | |
74dd17fb CM |
1344 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
1345 | if (disk_byte == 0) { | |
31db9f7c AB |
1346 | ret = -ENOENT; |
1347 | goto out; | |
1348 | } | |
74dd17fb | 1349 | logical = disk_byte + btrfs_file_extent_offset(eb, fi); |
31db9f7c | 1350 | |
04ab956e JM |
1351 | down_read(&fs_info->commit_root_sem); |
1352 | ret = extent_from_logical(fs_info, disk_byte, tmp_path, | |
69917e43 | 1353 | &found_key, &flags); |
04ab956e | 1354 | up_read(&fs_info->commit_root_sem); |
31db9f7c AB |
1355 | btrfs_release_path(tmp_path); |
1356 | ||
1357 | if (ret < 0) | |
1358 | goto out; | |
69917e43 | 1359 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
31db9f7c AB |
1360 | ret = -EIO; |
1361 | goto out; | |
1362 | } | |
1363 | ||
1364 | /* | |
1365 | * Setup the clone roots. | |
1366 | */ | |
1367 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1368 | cur_clone_root = sctx->clone_roots + i; | |
1369 | cur_clone_root->ino = (u64)-1; | |
1370 | cur_clone_root->offset = 0; | |
1371 | cur_clone_root->found_refs = 0; | |
1372 | } | |
1373 | ||
35075bb0 AB |
1374 | backref_ctx->sctx = sctx; |
1375 | backref_ctx->found = 0; | |
1376 | backref_ctx->cur_objectid = ino; | |
1377 | backref_ctx->cur_offset = data_offset; | |
1378 | backref_ctx->found_itself = 0; | |
1379 | backref_ctx->extent_len = num_bytes; | |
619d8c4e FM |
1380 | /* |
1381 | * For non-compressed extents iterate_extent_inodes() gives us extent | |
1382 | * offsets that already take into account the data offset, but not for | |
1383 | * compressed extents, since the offset is logical and not relative to | |
1384 | * the physical extent locations. We must take this into account to | |
1385 | * avoid sending clone offsets that go beyond the source file's size, | |
1386 | * which would result in the clone ioctl failing with -EINVAL on the | |
1387 | * receiving end. | |
1388 | */ | |
1389 | if (compressed == BTRFS_COMPRESS_NONE) | |
1390 | backref_ctx->data_offset = 0; | |
1391 | else | |
1392 | backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi); | |
31db9f7c AB |
1393 | |
1394 | /* | |
1395 | * The last extent of a file may be too large due to page alignment. | |
1396 | * We need to adjust extent_len in this case so that the checks in | |
1397 | * __iterate_backrefs work. | |
1398 | */ | |
1399 | if (data_offset + num_bytes >= ino_size) | |
35075bb0 | 1400 | backref_ctx->extent_len = ino_size - data_offset; |
31db9f7c AB |
1401 | |
1402 | /* | |
1403 | * Now collect all backrefs. | |
1404 | */ | |
74dd17fb CM |
1405 | if (compressed == BTRFS_COMPRESS_NONE) |
1406 | extent_item_pos = logical - found_key.objectid; | |
1407 | else | |
1408 | extent_item_pos = 0; | |
0b246afa JM |
1409 | ret = iterate_extent_inodes(fs_info, found_key.objectid, |
1410 | extent_item_pos, 1, __iterate_backrefs, | |
c995ab3c | 1411 | backref_ctx, false); |
74dd17fb | 1412 | |
31db9f7c AB |
1413 | if (ret < 0) |
1414 | goto out; | |
1415 | ||
35075bb0 | 1416 | if (!backref_ctx->found_itself) { |
31db9f7c AB |
1417 | /* found a bug in backref code? */ |
1418 | ret = -EIO; | |
04ab956e | 1419 | btrfs_err(fs_info, |
5d163e0e | 1420 | "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu", |
04ab956e | 1421 | ino, data_offset, disk_byte, found_key.objectid); |
31db9f7c AB |
1422 | goto out; |
1423 | } | |
1424 | ||
04ab956e JM |
1425 | btrfs_debug(fs_info, |
1426 | "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu", | |
1427 | data_offset, ino, num_bytes, logical); | |
31db9f7c | 1428 | |
35075bb0 | 1429 | if (!backref_ctx->found) |
04ab956e | 1430 | btrfs_debug(fs_info, "no clones found"); |
31db9f7c AB |
1431 | |
1432 | cur_clone_root = NULL; | |
1433 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1434 | if (sctx->clone_roots[i].found_refs) { | |
1435 | if (!cur_clone_root) | |
1436 | cur_clone_root = sctx->clone_roots + i; | |
1437 | else if (sctx->clone_roots[i].root == sctx->send_root) | |
1438 | /* prefer clones from send_root over others */ | |
1439 | cur_clone_root = sctx->clone_roots + i; | |
31db9f7c AB |
1440 | } |
1441 | ||
1442 | } | |
1443 | ||
1444 | if (cur_clone_root) { | |
1445 | *found = cur_clone_root; | |
1446 | ret = 0; | |
1447 | } else { | |
1448 | ret = -ENOENT; | |
1449 | } | |
1450 | ||
1451 | out: | |
1452 | btrfs_free_path(tmp_path); | |
35075bb0 | 1453 | kfree(backref_ctx); |
31db9f7c AB |
1454 | return ret; |
1455 | } | |
1456 | ||
924794c9 | 1457 | static int read_symlink(struct btrfs_root *root, |
31db9f7c AB |
1458 | u64 ino, |
1459 | struct fs_path *dest) | |
1460 | { | |
1461 | int ret; | |
1462 | struct btrfs_path *path; | |
1463 | struct btrfs_key key; | |
1464 | struct btrfs_file_extent_item *ei; | |
1465 | u8 type; | |
1466 | u8 compression; | |
1467 | unsigned long off; | |
1468 | int len; | |
1469 | ||
1470 | path = alloc_path_for_send(); | |
1471 | if (!path) | |
1472 | return -ENOMEM; | |
1473 | ||
1474 | key.objectid = ino; | |
1475 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1476 | key.offset = 0; | |
1477 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1478 | if (ret < 0) | |
1479 | goto out; | |
a879719b FM |
1480 | if (ret) { |
1481 | /* | |
1482 | * An empty symlink inode. Can happen in rare error paths when | |
1483 | * creating a symlink (transaction committed before the inode | |
1484 | * eviction handler removed the symlink inode items and a crash | |
1485 | * happened in between or the subvol was snapshoted in between). | |
1486 | * Print an informative message to dmesg/syslog so that the user | |
1487 | * can delete the symlink. | |
1488 | */ | |
1489 | btrfs_err(root->fs_info, | |
1490 | "Found empty symlink inode %llu at root %llu", | |
1491 | ino, root->root_key.objectid); | |
1492 | ret = -EIO; | |
1493 | goto out; | |
1494 | } | |
31db9f7c AB |
1495 | |
1496 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1497 | struct btrfs_file_extent_item); | |
1498 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
1499 | compression = btrfs_file_extent_compression(path->nodes[0], ei); | |
1500 | BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); | |
1501 | BUG_ON(compression); | |
1502 | ||
1503 | off = btrfs_file_extent_inline_start(ei); | |
e41ca589 | 1504 | len = btrfs_file_extent_ram_bytes(path->nodes[0], ei); |
31db9f7c AB |
1505 | |
1506 | ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); | |
31db9f7c AB |
1507 | |
1508 | out: | |
1509 | btrfs_free_path(path); | |
1510 | return ret; | |
1511 | } | |
1512 | ||
1513 | /* | |
1514 | * Helper function to generate a file name that is unique in the root of | |
1515 | * send_root and parent_root. This is used to generate names for orphan inodes. | |
1516 | */ | |
1517 | static int gen_unique_name(struct send_ctx *sctx, | |
1518 | u64 ino, u64 gen, | |
1519 | struct fs_path *dest) | |
1520 | { | |
1521 | int ret = 0; | |
1522 | struct btrfs_path *path; | |
1523 | struct btrfs_dir_item *di; | |
1524 | char tmp[64]; | |
1525 | int len; | |
1526 | u64 idx = 0; | |
1527 | ||
1528 | path = alloc_path_for_send(); | |
1529 | if (!path) | |
1530 | return -ENOMEM; | |
1531 | ||
1532 | while (1) { | |
f74b86d8 | 1533 | len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", |
31db9f7c | 1534 | ino, gen, idx); |
64792f25 | 1535 | ASSERT(len < sizeof(tmp)); |
31db9f7c AB |
1536 | |
1537 | di = btrfs_lookup_dir_item(NULL, sctx->send_root, | |
1538 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1539 | tmp, strlen(tmp), 0); | |
1540 | btrfs_release_path(path); | |
1541 | if (IS_ERR(di)) { | |
1542 | ret = PTR_ERR(di); | |
1543 | goto out; | |
1544 | } | |
1545 | if (di) { | |
1546 | /* not unique, try again */ | |
1547 | idx++; | |
1548 | continue; | |
1549 | } | |
1550 | ||
1551 | if (!sctx->parent_root) { | |
1552 | /* unique */ | |
1553 | ret = 0; | |
1554 | break; | |
1555 | } | |
1556 | ||
1557 | di = btrfs_lookup_dir_item(NULL, sctx->parent_root, | |
1558 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1559 | tmp, strlen(tmp), 0); | |
1560 | btrfs_release_path(path); | |
1561 | if (IS_ERR(di)) { | |
1562 | ret = PTR_ERR(di); | |
1563 | goto out; | |
1564 | } | |
1565 | if (di) { | |
1566 | /* not unique, try again */ | |
1567 | idx++; | |
1568 | continue; | |
1569 | } | |
1570 | /* unique */ | |
1571 | break; | |
1572 | } | |
1573 | ||
1574 | ret = fs_path_add(dest, tmp, strlen(tmp)); | |
1575 | ||
1576 | out: | |
1577 | btrfs_free_path(path); | |
1578 | return ret; | |
1579 | } | |
1580 | ||
1581 | enum inode_state { | |
1582 | inode_state_no_change, | |
1583 | inode_state_will_create, | |
1584 | inode_state_did_create, | |
1585 | inode_state_will_delete, | |
1586 | inode_state_did_delete, | |
1587 | }; | |
1588 | ||
1589 | static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) | |
1590 | { | |
1591 | int ret; | |
1592 | int left_ret; | |
1593 | int right_ret; | |
1594 | u64 left_gen; | |
1595 | u64 right_gen; | |
1596 | ||
1597 | ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, | |
85a7b33b | 1598 | NULL, NULL); |
31db9f7c AB |
1599 | if (ret < 0 && ret != -ENOENT) |
1600 | goto out; | |
1601 | left_ret = ret; | |
1602 | ||
1603 | if (!sctx->parent_root) { | |
1604 | right_ret = -ENOENT; | |
1605 | } else { | |
1606 | ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, | |
85a7b33b | 1607 | NULL, NULL, NULL, NULL); |
31db9f7c AB |
1608 | if (ret < 0 && ret != -ENOENT) |
1609 | goto out; | |
1610 | right_ret = ret; | |
1611 | } | |
1612 | ||
1613 | if (!left_ret && !right_ret) { | |
e938c8ad | 1614 | if (left_gen == gen && right_gen == gen) { |
31db9f7c | 1615 | ret = inode_state_no_change; |
e938c8ad | 1616 | } else if (left_gen == gen) { |
31db9f7c AB |
1617 | if (ino < sctx->send_progress) |
1618 | ret = inode_state_did_create; | |
1619 | else | |
1620 | ret = inode_state_will_create; | |
1621 | } else if (right_gen == gen) { | |
1622 | if (ino < sctx->send_progress) | |
1623 | ret = inode_state_did_delete; | |
1624 | else | |
1625 | ret = inode_state_will_delete; | |
1626 | } else { | |
1627 | ret = -ENOENT; | |
1628 | } | |
1629 | } else if (!left_ret) { | |
1630 | if (left_gen == gen) { | |
1631 | if (ino < sctx->send_progress) | |
1632 | ret = inode_state_did_create; | |
1633 | else | |
1634 | ret = inode_state_will_create; | |
1635 | } else { | |
1636 | ret = -ENOENT; | |
1637 | } | |
1638 | } else if (!right_ret) { | |
1639 | if (right_gen == gen) { | |
1640 | if (ino < sctx->send_progress) | |
1641 | ret = inode_state_did_delete; | |
1642 | else | |
1643 | ret = inode_state_will_delete; | |
1644 | } else { | |
1645 | ret = -ENOENT; | |
1646 | } | |
1647 | } else { | |
1648 | ret = -ENOENT; | |
1649 | } | |
1650 | ||
1651 | out: | |
1652 | return ret; | |
1653 | } | |
1654 | ||
1655 | static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) | |
1656 | { | |
1657 | int ret; | |
1658 | ||
4dd9920d RK |
1659 | if (ino == BTRFS_FIRST_FREE_OBJECTID) |
1660 | return 1; | |
1661 | ||
31db9f7c AB |
1662 | ret = get_cur_inode_state(sctx, ino, gen); |
1663 | if (ret < 0) | |
1664 | goto out; | |
1665 | ||
1666 | if (ret == inode_state_no_change || | |
1667 | ret == inode_state_did_create || | |
1668 | ret == inode_state_will_delete) | |
1669 | ret = 1; | |
1670 | else | |
1671 | ret = 0; | |
1672 | ||
1673 | out: | |
1674 | return ret; | |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * Helper function to lookup a dir item in a dir. | |
1679 | */ | |
1680 | static int lookup_dir_item_inode(struct btrfs_root *root, | |
1681 | u64 dir, const char *name, int name_len, | |
1682 | u64 *found_inode, | |
1683 | u8 *found_type) | |
1684 | { | |
1685 | int ret = 0; | |
1686 | struct btrfs_dir_item *di; | |
1687 | struct btrfs_key key; | |
1688 | struct btrfs_path *path; | |
1689 | ||
1690 | path = alloc_path_for_send(); | |
1691 | if (!path) | |
1692 | return -ENOMEM; | |
1693 | ||
1694 | di = btrfs_lookup_dir_item(NULL, root, path, | |
1695 | dir, name, name_len, 0); | |
1696 | if (!di) { | |
1697 | ret = -ENOENT; | |
1698 | goto out; | |
1699 | } | |
1700 | if (IS_ERR(di)) { | |
1701 | ret = PTR_ERR(di); | |
1702 | goto out; | |
1703 | } | |
1704 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
1af56070 FM |
1705 | if (key.type == BTRFS_ROOT_ITEM_KEY) { |
1706 | ret = -ENOENT; | |
1707 | goto out; | |
1708 | } | |
31db9f7c AB |
1709 | *found_inode = key.objectid; |
1710 | *found_type = btrfs_dir_type(path->nodes[0], di); | |
1711 | ||
1712 | out: | |
1713 | btrfs_free_path(path); | |
1714 | return ret; | |
1715 | } | |
1716 | ||
766702ef AB |
1717 | /* |
1718 | * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, | |
1719 | * generation of the parent dir and the name of the dir entry. | |
1720 | */ | |
924794c9 | 1721 | static int get_first_ref(struct btrfs_root *root, u64 ino, |
31db9f7c AB |
1722 | u64 *dir, u64 *dir_gen, struct fs_path *name) |
1723 | { | |
1724 | int ret; | |
1725 | struct btrfs_key key; | |
1726 | struct btrfs_key found_key; | |
1727 | struct btrfs_path *path; | |
31db9f7c | 1728 | int len; |
96b5bd77 | 1729 | u64 parent_dir; |
31db9f7c AB |
1730 | |
1731 | path = alloc_path_for_send(); | |
1732 | if (!path) | |
1733 | return -ENOMEM; | |
1734 | ||
1735 | key.objectid = ino; | |
1736 | key.type = BTRFS_INODE_REF_KEY; | |
1737 | key.offset = 0; | |
1738 | ||
1739 | ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); | |
1740 | if (ret < 0) | |
1741 | goto out; | |
1742 | if (!ret) | |
1743 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1744 | path->slots[0]); | |
96b5bd77 JS |
1745 | if (ret || found_key.objectid != ino || |
1746 | (found_key.type != BTRFS_INODE_REF_KEY && | |
1747 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1748 | ret = -ENOENT; |
1749 | goto out; | |
1750 | } | |
1751 | ||
51a60253 | 1752 | if (found_key.type == BTRFS_INODE_REF_KEY) { |
96b5bd77 JS |
1753 | struct btrfs_inode_ref *iref; |
1754 | iref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1755 | struct btrfs_inode_ref); | |
1756 | len = btrfs_inode_ref_name_len(path->nodes[0], iref); | |
1757 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1758 | (unsigned long)(iref + 1), | |
1759 | len); | |
1760 | parent_dir = found_key.offset; | |
1761 | } else { | |
1762 | struct btrfs_inode_extref *extref; | |
1763 | extref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1764 | struct btrfs_inode_extref); | |
1765 | len = btrfs_inode_extref_name_len(path->nodes[0], extref); | |
1766 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1767 | (unsigned long)&extref->name, len); | |
1768 | parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); | |
1769 | } | |
31db9f7c AB |
1770 | if (ret < 0) |
1771 | goto out; | |
1772 | btrfs_release_path(path); | |
1773 | ||
b46ab97b FM |
1774 | if (dir_gen) { |
1775 | ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, | |
1776 | NULL, NULL, NULL); | |
1777 | if (ret < 0) | |
1778 | goto out; | |
1779 | } | |
31db9f7c | 1780 | |
96b5bd77 | 1781 | *dir = parent_dir; |
31db9f7c AB |
1782 | |
1783 | out: | |
1784 | btrfs_free_path(path); | |
1785 | return ret; | |
1786 | } | |
1787 | ||
924794c9 | 1788 | static int is_first_ref(struct btrfs_root *root, |
31db9f7c AB |
1789 | u64 ino, u64 dir, |
1790 | const char *name, int name_len) | |
1791 | { | |
1792 | int ret; | |
1793 | struct fs_path *tmp_name; | |
1794 | u64 tmp_dir; | |
31db9f7c | 1795 | |
924794c9 | 1796 | tmp_name = fs_path_alloc(); |
31db9f7c AB |
1797 | if (!tmp_name) |
1798 | return -ENOMEM; | |
1799 | ||
b46ab97b | 1800 | ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name); |
31db9f7c AB |
1801 | if (ret < 0) |
1802 | goto out; | |
1803 | ||
b9291aff | 1804 | if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { |
31db9f7c AB |
1805 | ret = 0; |
1806 | goto out; | |
1807 | } | |
1808 | ||
e938c8ad | 1809 | ret = !memcmp(tmp_name->start, name, name_len); |
31db9f7c AB |
1810 | |
1811 | out: | |
924794c9 | 1812 | fs_path_free(tmp_name); |
31db9f7c AB |
1813 | return ret; |
1814 | } | |
1815 | ||
766702ef AB |
1816 | /* |
1817 | * Used by process_recorded_refs to determine if a new ref would overwrite an | |
1818 | * already existing ref. In case it detects an overwrite, it returns the | |
1819 | * inode/gen in who_ino/who_gen. | |
1820 | * When an overwrite is detected, process_recorded_refs does proper orphanizing | |
1821 | * to make sure later references to the overwritten inode are possible. | |
1822 | * Orphanizing is however only required for the first ref of an inode. | |
1823 | * process_recorded_refs does an additional is_first_ref check to see if | |
1824 | * orphanizing is really required. | |
1825 | */ | |
31db9f7c AB |
1826 | static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
1827 | const char *name, int name_len, | |
f5962781 | 1828 | u64 *who_ino, u64 *who_gen, u64 *who_mode) |
31db9f7c AB |
1829 | { |
1830 | int ret = 0; | |
ebdad913 | 1831 | u64 gen; |
31db9f7c AB |
1832 | u64 other_inode = 0; |
1833 | u8 other_type = 0; | |
1834 | ||
1835 | if (!sctx->parent_root) | |
1836 | goto out; | |
1837 | ||
1838 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1839 | if (ret <= 0) | |
1840 | goto out; | |
1841 | ||
ebdad913 JB |
1842 | /* |
1843 | * If we have a parent root we need to verify that the parent dir was | |
01327610 | 1844 | * not deleted and then re-created, if it was then we have no overwrite |
ebdad913 JB |
1845 | * and we can just unlink this entry. |
1846 | */ | |
4dd9920d | 1847 | if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) { |
ebdad913 JB |
1848 | ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, |
1849 | NULL, NULL, NULL); | |
1850 | if (ret < 0 && ret != -ENOENT) | |
1851 | goto out; | |
1852 | if (ret) { | |
1853 | ret = 0; | |
1854 | goto out; | |
1855 | } | |
1856 | if (gen != dir_gen) | |
1857 | goto out; | |
1858 | } | |
1859 | ||
31db9f7c AB |
1860 | ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, |
1861 | &other_inode, &other_type); | |
1862 | if (ret < 0 && ret != -ENOENT) | |
1863 | goto out; | |
1864 | if (ret) { | |
1865 | ret = 0; | |
1866 | goto out; | |
1867 | } | |
1868 | ||
766702ef AB |
1869 | /* |
1870 | * Check if the overwritten ref was already processed. If yes, the ref | |
1871 | * was already unlinked/moved, so we can safely assume that we will not | |
1872 | * overwrite anything at this point in time. | |
1873 | */ | |
801bec36 RK |
1874 | if (other_inode > sctx->send_progress || |
1875 | is_waiting_for_move(sctx, other_inode)) { | |
31db9f7c | 1876 | ret = get_inode_info(sctx->parent_root, other_inode, NULL, |
f5962781 | 1877 | who_gen, who_mode, NULL, NULL, NULL); |
31db9f7c AB |
1878 | if (ret < 0) |
1879 | goto out; | |
1880 | ||
1881 | ret = 1; | |
1882 | *who_ino = other_inode; | |
1883 | } else { | |
1884 | ret = 0; | |
1885 | } | |
1886 | ||
1887 | out: | |
1888 | return ret; | |
1889 | } | |
1890 | ||
766702ef AB |
1891 | /* |
1892 | * Checks if the ref was overwritten by an already processed inode. This is | |
1893 | * used by __get_cur_name_and_parent to find out if the ref was orphanized and | |
1894 | * thus the orphan name needs be used. | |
1895 | * process_recorded_refs also uses it to avoid unlinking of refs that were | |
1896 | * overwritten. | |
1897 | */ | |
31db9f7c AB |
1898 | static int did_overwrite_ref(struct send_ctx *sctx, |
1899 | u64 dir, u64 dir_gen, | |
1900 | u64 ino, u64 ino_gen, | |
1901 | const char *name, int name_len) | |
1902 | { | |
1903 | int ret = 0; | |
1904 | u64 gen; | |
1905 | u64 ow_inode; | |
1906 | u8 other_type; | |
1907 | ||
1908 | if (!sctx->parent_root) | |
1909 | goto out; | |
1910 | ||
1911 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1912 | if (ret <= 0) | |
1913 | goto out; | |
1914 | ||
01914101 RK |
1915 | if (dir != BTRFS_FIRST_FREE_OBJECTID) { |
1916 | ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, | |
1917 | NULL, NULL, NULL); | |
1918 | if (ret < 0 && ret != -ENOENT) | |
1919 | goto out; | |
1920 | if (ret) { | |
1921 | ret = 0; | |
1922 | goto out; | |
1923 | } | |
1924 | if (gen != dir_gen) | |
1925 | goto out; | |
1926 | } | |
1927 | ||
31db9f7c AB |
1928 | /* check if the ref was overwritten by another ref */ |
1929 | ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, | |
1930 | &ow_inode, &other_type); | |
1931 | if (ret < 0 && ret != -ENOENT) | |
1932 | goto out; | |
1933 | if (ret) { | |
1934 | /* was never and will never be overwritten */ | |
1935 | ret = 0; | |
1936 | goto out; | |
1937 | } | |
1938 | ||
1939 | ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, | |
85a7b33b | 1940 | NULL, NULL); |
31db9f7c AB |
1941 | if (ret < 0) |
1942 | goto out; | |
1943 | ||
1944 | if (ow_inode == ino && gen == ino_gen) { | |
1945 | ret = 0; | |
1946 | goto out; | |
1947 | } | |
1948 | ||
8b191a68 FM |
1949 | /* |
1950 | * We know that it is or will be overwritten. Check this now. | |
1951 | * The current inode being processed might have been the one that caused | |
b786f16a FM |
1952 | * inode 'ino' to be orphanized, therefore check if ow_inode matches |
1953 | * the current inode being processed. | |
8b191a68 | 1954 | */ |
b786f16a FM |
1955 | if ((ow_inode < sctx->send_progress) || |
1956 | (ino != sctx->cur_ino && ow_inode == sctx->cur_ino && | |
1957 | gen == sctx->cur_inode_gen)) | |
31db9f7c AB |
1958 | ret = 1; |
1959 | else | |
1960 | ret = 0; | |
1961 | ||
1962 | out: | |
1963 | return ret; | |
1964 | } | |
1965 | ||
766702ef AB |
1966 | /* |
1967 | * Same as did_overwrite_ref, but also checks if it is the first ref of an inode | |
1968 | * that got overwritten. This is used by process_recorded_refs to determine | |
1969 | * if it has to use the path as returned by get_cur_path or the orphan name. | |
1970 | */ | |
31db9f7c AB |
1971 | static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) |
1972 | { | |
1973 | int ret = 0; | |
1974 | struct fs_path *name = NULL; | |
1975 | u64 dir; | |
1976 | u64 dir_gen; | |
1977 | ||
1978 | if (!sctx->parent_root) | |
1979 | goto out; | |
1980 | ||
924794c9 | 1981 | name = fs_path_alloc(); |
31db9f7c AB |
1982 | if (!name) |
1983 | return -ENOMEM; | |
1984 | ||
924794c9 | 1985 | ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); |
31db9f7c AB |
1986 | if (ret < 0) |
1987 | goto out; | |
1988 | ||
1989 | ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, | |
1990 | name->start, fs_path_len(name)); | |
31db9f7c AB |
1991 | |
1992 | out: | |
924794c9 | 1993 | fs_path_free(name); |
31db9f7c AB |
1994 | return ret; |
1995 | } | |
1996 | ||
766702ef AB |
1997 | /* |
1998 | * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, | |
1999 | * so we need to do some special handling in case we have clashes. This function | |
2000 | * takes care of this with the help of name_cache_entry::radix_list. | |
5dc67d0b | 2001 | * In case of error, nce is kfreed. |
766702ef | 2002 | */ |
31db9f7c AB |
2003 | static int name_cache_insert(struct send_ctx *sctx, |
2004 | struct name_cache_entry *nce) | |
2005 | { | |
2006 | int ret = 0; | |
7e0926fe AB |
2007 | struct list_head *nce_head; |
2008 | ||
2009 | nce_head = radix_tree_lookup(&sctx->name_cache, | |
2010 | (unsigned long)nce->ino); | |
2011 | if (!nce_head) { | |
e780b0d1 | 2012 | nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL); |
cfa7a9cc TI |
2013 | if (!nce_head) { |
2014 | kfree(nce); | |
31db9f7c | 2015 | return -ENOMEM; |
cfa7a9cc | 2016 | } |
7e0926fe | 2017 | INIT_LIST_HEAD(nce_head); |
31db9f7c | 2018 | |
7e0926fe | 2019 | ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); |
5dc67d0b AB |
2020 | if (ret < 0) { |
2021 | kfree(nce_head); | |
2022 | kfree(nce); | |
31db9f7c | 2023 | return ret; |
5dc67d0b | 2024 | } |
31db9f7c | 2025 | } |
7e0926fe | 2026 | list_add_tail(&nce->radix_list, nce_head); |
31db9f7c AB |
2027 | list_add_tail(&nce->list, &sctx->name_cache_list); |
2028 | sctx->name_cache_size++; | |
2029 | ||
2030 | return ret; | |
2031 | } | |
2032 | ||
2033 | static void name_cache_delete(struct send_ctx *sctx, | |
2034 | struct name_cache_entry *nce) | |
2035 | { | |
7e0926fe | 2036 | struct list_head *nce_head; |
31db9f7c | 2037 | |
7e0926fe AB |
2038 | nce_head = radix_tree_lookup(&sctx->name_cache, |
2039 | (unsigned long)nce->ino); | |
57fb8910 DS |
2040 | if (!nce_head) { |
2041 | btrfs_err(sctx->send_root->fs_info, | |
2042 | "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", | |
2043 | nce->ino, sctx->name_cache_size); | |
2044 | } | |
31db9f7c | 2045 | |
7e0926fe | 2046 | list_del(&nce->radix_list); |
31db9f7c | 2047 | list_del(&nce->list); |
31db9f7c | 2048 | sctx->name_cache_size--; |
7e0926fe | 2049 | |
57fb8910 DS |
2050 | /* |
2051 | * We may not get to the final release of nce_head if the lookup fails | |
2052 | */ | |
2053 | if (nce_head && list_empty(nce_head)) { | |
7e0926fe AB |
2054 | radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); |
2055 | kfree(nce_head); | |
2056 | } | |
31db9f7c AB |
2057 | } |
2058 | ||
2059 | static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, | |
2060 | u64 ino, u64 gen) | |
2061 | { | |
7e0926fe AB |
2062 | struct list_head *nce_head; |
2063 | struct name_cache_entry *cur; | |
31db9f7c | 2064 | |
7e0926fe AB |
2065 | nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); |
2066 | if (!nce_head) | |
31db9f7c AB |
2067 | return NULL; |
2068 | ||
7e0926fe AB |
2069 | list_for_each_entry(cur, nce_head, radix_list) { |
2070 | if (cur->ino == ino && cur->gen == gen) | |
2071 | return cur; | |
2072 | } | |
31db9f7c AB |
2073 | return NULL; |
2074 | } | |
2075 | ||
766702ef AB |
2076 | /* |
2077 | * Removes the entry from the list and adds it back to the end. This marks the | |
2078 | * entry as recently used so that name_cache_clean_unused does not remove it. | |
2079 | */ | |
31db9f7c AB |
2080 | static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) |
2081 | { | |
2082 | list_del(&nce->list); | |
2083 | list_add_tail(&nce->list, &sctx->name_cache_list); | |
2084 | } | |
2085 | ||
766702ef AB |
2086 | /* |
2087 | * Remove some entries from the beginning of name_cache_list. | |
2088 | */ | |
31db9f7c AB |
2089 | static void name_cache_clean_unused(struct send_ctx *sctx) |
2090 | { | |
2091 | struct name_cache_entry *nce; | |
2092 | ||
2093 | if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) | |
2094 | return; | |
2095 | ||
2096 | while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { | |
2097 | nce = list_entry(sctx->name_cache_list.next, | |
2098 | struct name_cache_entry, list); | |
2099 | name_cache_delete(sctx, nce); | |
2100 | kfree(nce); | |
2101 | } | |
2102 | } | |
2103 | ||
2104 | static void name_cache_free(struct send_ctx *sctx) | |
2105 | { | |
2106 | struct name_cache_entry *nce; | |
31db9f7c | 2107 | |
e938c8ad AB |
2108 | while (!list_empty(&sctx->name_cache_list)) { |
2109 | nce = list_entry(sctx->name_cache_list.next, | |
2110 | struct name_cache_entry, list); | |
31db9f7c | 2111 | name_cache_delete(sctx, nce); |
17589bd9 | 2112 | kfree(nce); |
31db9f7c AB |
2113 | } |
2114 | } | |
2115 | ||
766702ef AB |
2116 | /* |
2117 | * Used by get_cur_path for each ref up to the root. | |
2118 | * Returns 0 if it succeeded. | |
2119 | * Returns 1 if the inode is not existent or got overwritten. In that case, the | |
2120 | * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 | |
2121 | * is returned, parent_ino/parent_gen are not guaranteed to be valid. | |
2122 | * Returns <0 in case of error. | |
2123 | */ | |
31db9f7c AB |
2124 | static int __get_cur_name_and_parent(struct send_ctx *sctx, |
2125 | u64 ino, u64 gen, | |
2126 | u64 *parent_ino, | |
2127 | u64 *parent_gen, | |
2128 | struct fs_path *dest) | |
2129 | { | |
2130 | int ret; | |
2131 | int nce_ret; | |
31db9f7c AB |
2132 | struct name_cache_entry *nce = NULL; |
2133 | ||
766702ef AB |
2134 | /* |
2135 | * First check if we already did a call to this function with the same | |
2136 | * ino/gen. If yes, check if the cache entry is still up-to-date. If yes | |
2137 | * return the cached result. | |
2138 | */ | |
31db9f7c AB |
2139 | nce = name_cache_search(sctx, ino, gen); |
2140 | if (nce) { | |
2141 | if (ino < sctx->send_progress && nce->need_later_update) { | |
2142 | name_cache_delete(sctx, nce); | |
2143 | kfree(nce); | |
2144 | nce = NULL; | |
2145 | } else { | |
2146 | name_cache_used(sctx, nce); | |
2147 | *parent_ino = nce->parent_ino; | |
2148 | *parent_gen = nce->parent_gen; | |
2149 | ret = fs_path_add(dest, nce->name, nce->name_len); | |
2150 | if (ret < 0) | |
2151 | goto out; | |
2152 | ret = nce->ret; | |
2153 | goto out; | |
2154 | } | |
2155 | } | |
2156 | ||
766702ef AB |
2157 | /* |
2158 | * If the inode is not existent yet, add the orphan name and return 1. | |
2159 | * This should only happen for the parent dir that we determine in | |
2160 | * __record_new_ref | |
2161 | */ | |
31db9f7c AB |
2162 | ret = is_inode_existent(sctx, ino, gen); |
2163 | if (ret < 0) | |
2164 | goto out; | |
2165 | ||
2166 | if (!ret) { | |
2167 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2168 | if (ret < 0) | |
2169 | goto out; | |
2170 | ret = 1; | |
2171 | goto out_cache; | |
2172 | } | |
2173 | ||
766702ef AB |
2174 | /* |
2175 | * Depending on whether the inode was already processed or not, use | |
2176 | * send_root or parent_root for ref lookup. | |
2177 | */ | |
bf0d1f44 | 2178 | if (ino < sctx->send_progress) |
924794c9 TI |
2179 | ret = get_first_ref(sctx->send_root, ino, |
2180 | parent_ino, parent_gen, dest); | |
31db9f7c | 2181 | else |
924794c9 TI |
2182 | ret = get_first_ref(sctx->parent_root, ino, |
2183 | parent_ino, parent_gen, dest); | |
31db9f7c AB |
2184 | if (ret < 0) |
2185 | goto out; | |
2186 | ||
766702ef AB |
2187 | /* |
2188 | * Check if the ref was overwritten by an inode's ref that was processed | |
2189 | * earlier. If yes, treat as orphan and return 1. | |
2190 | */ | |
31db9f7c AB |
2191 | ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, |
2192 | dest->start, dest->end - dest->start); | |
2193 | if (ret < 0) | |
2194 | goto out; | |
2195 | if (ret) { | |
2196 | fs_path_reset(dest); | |
2197 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2198 | if (ret < 0) | |
2199 | goto out; | |
2200 | ret = 1; | |
2201 | } | |
2202 | ||
2203 | out_cache: | |
766702ef AB |
2204 | /* |
2205 | * Store the result of the lookup in the name cache. | |
2206 | */ | |
e780b0d1 | 2207 | nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); |
31db9f7c AB |
2208 | if (!nce) { |
2209 | ret = -ENOMEM; | |
2210 | goto out; | |
2211 | } | |
2212 | ||
2213 | nce->ino = ino; | |
2214 | nce->gen = gen; | |
2215 | nce->parent_ino = *parent_ino; | |
2216 | nce->parent_gen = *parent_gen; | |
2217 | nce->name_len = fs_path_len(dest); | |
2218 | nce->ret = ret; | |
2219 | strcpy(nce->name, dest->start); | |
31db9f7c AB |
2220 | |
2221 | if (ino < sctx->send_progress) | |
2222 | nce->need_later_update = 0; | |
2223 | else | |
2224 | nce->need_later_update = 1; | |
2225 | ||
2226 | nce_ret = name_cache_insert(sctx, nce); | |
2227 | if (nce_ret < 0) | |
2228 | ret = nce_ret; | |
2229 | name_cache_clean_unused(sctx); | |
2230 | ||
2231 | out: | |
31db9f7c AB |
2232 | return ret; |
2233 | } | |
2234 | ||
2235 | /* | |
2236 | * Magic happens here. This function returns the first ref to an inode as it | |
2237 | * would look like while receiving the stream at this point in time. | |
2238 | * We walk the path up to the root. For every inode in between, we check if it | |
2239 | * was already processed/sent. If yes, we continue with the parent as found | |
2240 | * in send_root. If not, we continue with the parent as found in parent_root. | |
2241 | * If we encounter an inode that was deleted at this point in time, we use the | |
2242 | * inodes "orphan" name instead of the real name and stop. Same with new inodes | |
2243 | * that were not created yet and overwritten inodes/refs. | |
2244 | * | |
2245 | * When do we have have orphan inodes: | |
2246 | * 1. When an inode is freshly created and thus no valid refs are available yet | |
2247 | * 2. When a directory lost all it's refs (deleted) but still has dir items | |
2248 | * inside which were not processed yet (pending for move/delete). If anyone | |
2249 | * tried to get the path to the dir items, it would get a path inside that | |
2250 | * orphan directory. | |
2251 | * 3. When an inode is moved around or gets new links, it may overwrite the ref | |
2252 | * of an unprocessed inode. If in that case the first ref would be | |
2253 | * overwritten, the overwritten inode gets "orphanized". Later when we | |
2254 | * process this overwritten inode, it is restored at a new place by moving | |
2255 | * the orphan inode. | |
2256 | * | |
2257 | * sctx->send_progress tells this function at which point in time receiving | |
2258 | * would be. | |
2259 | */ | |
2260 | static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, | |
2261 | struct fs_path *dest) | |
2262 | { | |
2263 | int ret = 0; | |
2264 | struct fs_path *name = NULL; | |
2265 | u64 parent_inode = 0; | |
2266 | u64 parent_gen = 0; | |
2267 | int stop = 0; | |
2268 | ||
924794c9 | 2269 | name = fs_path_alloc(); |
31db9f7c AB |
2270 | if (!name) { |
2271 | ret = -ENOMEM; | |
2272 | goto out; | |
2273 | } | |
2274 | ||
2275 | dest->reversed = 1; | |
2276 | fs_path_reset(dest); | |
2277 | ||
2278 | while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { | |
8b191a68 FM |
2279 | struct waiting_dir_move *wdm; |
2280 | ||
31db9f7c AB |
2281 | fs_path_reset(name); |
2282 | ||
9dc44214 FM |
2283 | if (is_waiting_for_rm(sctx, ino)) { |
2284 | ret = gen_unique_name(sctx, ino, gen, name); | |
2285 | if (ret < 0) | |
2286 | goto out; | |
2287 | ret = fs_path_add_path(dest, name); | |
2288 | break; | |
2289 | } | |
2290 | ||
8b191a68 FM |
2291 | wdm = get_waiting_dir_move(sctx, ino); |
2292 | if (wdm && wdm->orphanized) { | |
2293 | ret = gen_unique_name(sctx, ino, gen, name); | |
2294 | stop = 1; | |
2295 | } else if (wdm) { | |
bf0d1f44 FM |
2296 | ret = get_first_ref(sctx->parent_root, ino, |
2297 | &parent_inode, &parent_gen, name); | |
2298 | } else { | |
2299 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
2300 | &parent_inode, | |
2301 | &parent_gen, name); | |
2302 | if (ret) | |
2303 | stop = 1; | |
2304 | } | |
2305 | ||
31db9f7c AB |
2306 | if (ret < 0) |
2307 | goto out; | |
9f03740a | 2308 | |
31db9f7c AB |
2309 | ret = fs_path_add_path(dest, name); |
2310 | if (ret < 0) | |
2311 | goto out; | |
2312 | ||
2313 | ino = parent_inode; | |
2314 | gen = parent_gen; | |
2315 | } | |
2316 | ||
2317 | out: | |
924794c9 | 2318 | fs_path_free(name); |
31db9f7c AB |
2319 | if (!ret) |
2320 | fs_path_unreverse(dest); | |
2321 | return ret; | |
2322 | } | |
2323 | ||
31db9f7c AB |
2324 | /* |
2325 | * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace | |
2326 | */ | |
2327 | static int send_subvol_begin(struct send_ctx *sctx) | |
2328 | { | |
2329 | int ret; | |
2330 | struct btrfs_root *send_root = sctx->send_root; | |
2331 | struct btrfs_root *parent_root = sctx->parent_root; | |
2332 | struct btrfs_path *path; | |
2333 | struct btrfs_key key; | |
2334 | struct btrfs_root_ref *ref; | |
2335 | struct extent_buffer *leaf; | |
2336 | char *name = NULL; | |
2337 | int namelen; | |
2338 | ||
ffcfaf81 | 2339 | path = btrfs_alloc_path(); |
31db9f7c AB |
2340 | if (!path) |
2341 | return -ENOMEM; | |
2342 | ||
e780b0d1 | 2343 | name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); |
31db9f7c AB |
2344 | if (!name) { |
2345 | btrfs_free_path(path); | |
2346 | return -ENOMEM; | |
2347 | } | |
2348 | ||
2349 | key.objectid = send_root->objectid; | |
2350 | key.type = BTRFS_ROOT_BACKREF_KEY; | |
2351 | key.offset = 0; | |
2352 | ||
2353 | ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, | |
2354 | &key, path, 1, 0); | |
2355 | if (ret < 0) | |
2356 | goto out; | |
2357 | if (ret) { | |
2358 | ret = -ENOENT; | |
2359 | goto out; | |
2360 | } | |
2361 | ||
2362 | leaf = path->nodes[0]; | |
2363 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2364 | if (key.type != BTRFS_ROOT_BACKREF_KEY || | |
2365 | key.objectid != send_root->objectid) { | |
2366 | ret = -ENOENT; | |
2367 | goto out; | |
2368 | } | |
2369 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
2370 | namelen = btrfs_root_ref_name_len(leaf, ref); | |
2371 | read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); | |
2372 | btrfs_release_path(path); | |
2373 | ||
31db9f7c AB |
2374 | if (parent_root) { |
2375 | ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); | |
2376 | if (ret < 0) | |
2377 | goto out; | |
2378 | } else { | |
2379 | ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); | |
2380 | if (ret < 0) | |
2381 | goto out; | |
2382 | } | |
2383 | ||
2384 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); | |
b96b1db0 RR |
2385 | |
2386 | if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid)) | |
2387 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2388 | sctx->send_root->root_item.received_uuid); | |
2389 | else | |
2390 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2391 | sctx->send_root->root_item.uuid); | |
2392 | ||
31db9f7c | 2393 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, |
5a0f4e2c | 2394 | le64_to_cpu(sctx->send_root->root_item.ctransid)); |
31db9f7c | 2395 | if (parent_root) { |
37b8d27d JB |
2396 | if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) |
2397 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2398 | parent_root->root_item.received_uuid); | |
2399 | else | |
2400 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2401 | parent_root->root_item.uuid); | |
31db9f7c | 2402 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 2403 | le64_to_cpu(sctx->parent_root->root_item.ctransid)); |
31db9f7c AB |
2404 | } |
2405 | ||
2406 | ret = send_cmd(sctx); | |
2407 | ||
2408 | tlv_put_failure: | |
2409 | out: | |
2410 | btrfs_free_path(path); | |
2411 | kfree(name); | |
2412 | return ret; | |
2413 | } | |
2414 | ||
2415 | static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) | |
2416 | { | |
04ab956e | 2417 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2418 | int ret = 0; |
2419 | struct fs_path *p; | |
2420 | ||
04ab956e | 2421 | btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size); |
31db9f7c | 2422 | |
924794c9 | 2423 | p = fs_path_alloc(); |
31db9f7c AB |
2424 | if (!p) |
2425 | return -ENOMEM; | |
2426 | ||
2427 | ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); | |
2428 | if (ret < 0) | |
2429 | goto out; | |
2430 | ||
2431 | ret = get_cur_path(sctx, ino, gen, p); | |
2432 | if (ret < 0) | |
2433 | goto out; | |
2434 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2435 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); | |
2436 | ||
2437 | ret = send_cmd(sctx); | |
2438 | ||
2439 | tlv_put_failure: | |
2440 | out: | |
924794c9 | 2441 | fs_path_free(p); |
31db9f7c AB |
2442 | return ret; |
2443 | } | |
2444 | ||
2445 | static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) | |
2446 | { | |
04ab956e | 2447 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2448 | int ret = 0; |
2449 | struct fs_path *p; | |
2450 | ||
04ab956e | 2451 | btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode); |
31db9f7c | 2452 | |
924794c9 | 2453 | p = fs_path_alloc(); |
31db9f7c AB |
2454 | if (!p) |
2455 | return -ENOMEM; | |
2456 | ||
2457 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); | |
2458 | if (ret < 0) | |
2459 | goto out; | |
2460 | ||
2461 | ret = get_cur_path(sctx, ino, gen, p); | |
2462 | if (ret < 0) | |
2463 | goto out; | |
2464 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2465 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); | |
2466 | ||
2467 | ret = send_cmd(sctx); | |
2468 | ||
2469 | tlv_put_failure: | |
2470 | out: | |
924794c9 | 2471 | fs_path_free(p); |
31db9f7c AB |
2472 | return ret; |
2473 | } | |
2474 | ||
2475 | static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) | |
2476 | { | |
04ab956e | 2477 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2478 | int ret = 0; |
2479 | struct fs_path *p; | |
2480 | ||
04ab956e JM |
2481 | btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu", |
2482 | ino, uid, gid); | |
31db9f7c | 2483 | |
924794c9 | 2484 | p = fs_path_alloc(); |
31db9f7c AB |
2485 | if (!p) |
2486 | return -ENOMEM; | |
2487 | ||
2488 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); | |
2489 | if (ret < 0) | |
2490 | goto out; | |
2491 | ||
2492 | ret = get_cur_path(sctx, ino, gen, p); | |
2493 | if (ret < 0) | |
2494 | goto out; | |
2495 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2496 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); | |
2497 | TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); | |
2498 | ||
2499 | ret = send_cmd(sctx); | |
2500 | ||
2501 | tlv_put_failure: | |
2502 | out: | |
924794c9 | 2503 | fs_path_free(p); |
31db9f7c AB |
2504 | return ret; |
2505 | } | |
2506 | ||
2507 | static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) | |
2508 | { | |
04ab956e | 2509 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2510 | int ret = 0; |
2511 | struct fs_path *p = NULL; | |
2512 | struct btrfs_inode_item *ii; | |
2513 | struct btrfs_path *path = NULL; | |
2514 | struct extent_buffer *eb; | |
2515 | struct btrfs_key key; | |
2516 | int slot; | |
2517 | ||
04ab956e | 2518 | btrfs_debug(fs_info, "send_utimes %llu", ino); |
31db9f7c | 2519 | |
924794c9 | 2520 | p = fs_path_alloc(); |
31db9f7c AB |
2521 | if (!p) |
2522 | return -ENOMEM; | |
2523 | ||
2524 | path = alloc_path_for_send(); | |
2525 | if (!path) { | |
2526 | ret = -ENOMEM; | |
2527 | goto out; | |
2528 | } | |
2529 | ||
2530 | key.objectid = ino; | |
2531 | key.type = BTRFS_INODE_ITEM_KEY; | |
2532 | key.offset = 0; | |
2533 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); | |
15b253ea FM |
2534 | if (ret > 0) |
2535 | ret = -ENOENT; | |
31db9f7c AB |
2536 | if (ret < 0) |
2537 | goto out; | |
2538 | ||
2539 | eb = path->nodes[0]; | |
2540 | slot = path->slots[0]; | |
2541 | ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
2542 | ||
2543 | ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); | |
2544 | if (ret < 0) | |
2545 | goto out; | |
2546 | ||
2547 | ret = get_cur_path(sctx, ino, gen, p); | |
2548 | if (ret < 0) | |
2549 | goto out; | |
2550 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
a937b979 DS |
2551 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); |
2552 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); | |
2553 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); | |
766702ef | 2554 | /* TODO Add otime support when the otime patches get into upstream */ |
31db9f7c AB |
2555 | |
2556 | ret = send_cmd(sctx); | |
2557 | ||
2558 | tlv_put_failure: | |
2559 | out: | |
924794c9 | 2560 | fs_path_free(p); |
31db9f7c AB |
2561 | btrfs_free_path(path); |
2562 | return ret; | |
2563 | } | |
2564 | ||
2565 | /* | |
2566 | * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have | |
2567 | * a valid path yet because we did not process the refs yet. So, the inode | |
2568 | * is created as orphan. | |
2569 | */ | |
1f4692da | 2570 | static int send_create_inode(struct send_ctx *sctx, u64 ino) |
31db9f7c | 2571 | { |
04ab956e | 2572 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c | 2573 | int ret = 0; |
31db9f7c | 2574 | struct fs_path *p; |
31db9f7c | 2575 | int cmd; |
1f4692da | 2576 | u64 gen; |
31db9f7c | 2577 | u64 mode; |
1f4692da | 2578 | u64 rdev; |
31db9f7c | 2579 | |
04ab956e | 2580 | btrfs_debug(fs_info, "send_create_inode %llu", ino); |
31db9f7c | 2581 | |
924794c9 | 2582 | p = fs_path_alloc(); |
31db9f7c AB |
2583 | if (!p) |
2584 | return -ENOMEM; | |
2585 | ||
644d1940 LB |
2586 | if (ino != sctx->cur_ino) { |
2587 | ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, | |
2588 | NULL, NULL, &rdev); | |
2589 | if (ret < 0) | |
2590 | goto out; | |
2591 | } else { | |
2592 | gen = sctx->cur_inode_gen; | |
2593 | mode = sctx->cur_inode_mode; | |
2594 | rdev = sctx->cur_inode_rdev; | |
2595 | } | |
31db9f7c | 2596 | |
e938c8ad | 2597 | if (S_ISREG(mode)) { |
31db9f7c | 2598 | cmd = BTRFS_SEND_C_MKFILE; |
e938c8ad | 2599 | } else if (S_ISDIR(mode)) { |
31db9f7c | 2600 | cmd = BTRFS_SEND_C_MKDIR; |
e938c8ad | 2601 | } else if (S_ISLNK(mode)) { |
31db9f7c | 2602 | cmd = BTRFS_SEND_C_SYMLINK; |
e938c8ad | 2603 | } else if (S_ISCHR(mode) || S_ISBLK(mode)) { |
31db9f7c | 2604 | cmd = BTRFS_SEND_C_MKNOD; |
e938c8ad | 2605 | } else if (S_ISFIFO(mode)) { |
31db9f7c | 2606 | cmd = BTRFS_SEND_C_MKFIFO; |
e938c8ad | 2607 | } else if (S_ISSOCK(mode)) { |
31db9f7c | 2608 | cmd = BTRFS_SEND_C_MKSOCK; |
e938c8ad | 2609 | } else { |
f14d104d | 2610 | btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o", |
31db9f7c | 2611 | (int)(mode & S_IFMT)); |
ca6842bf | 2612 | ret = -EOPNOTSUPP; |
31db9f7c AB |
2613 | goto out; |
2614 | } | |
2615 | ||
2616 | ret = begin_cmd(sctx, cmd); | |
2617 | if (ret < 0) | |
2618 | goto out; | |
2619 | ||
1f4692da | 2620 | ret = gen_unique_name(sctx, ino, gen, p); |
31db9f7c AB |
2621 | if (ret < 0) |
2622 | goto out; | |
2623 | ||
2624 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
1f4692da | 2625 | TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); |
31db9f7c AB |
2626 | |
2627 | if (S_ISLNK(mode)) { | |
2628 | fs_path_reset(p); | |
924794c9 | 2629 | ret = read_symlink(sctx->send_root, ino, p); |
31db9f7c AB |
2630 | if (ret < 0) |
2631 | goto out; | |
2632 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); | |
2633 | } else if (S_ISCHR(mode) || S_ISBLK(mode) || | |
2634 | S_ISFIFO(mode) || S_ISSOCK(mode)) { | |
d79e5043 AJ |
2635 | TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); |
2636 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); | |
31db9f7c AB |
2637 | } |
2638 | ||
2639 | ret = send_cmd(sctx); | |
2640 | if (ret < 0) | |
2641 | goto out; | |
2642 | ||
2643 | ||
2644 | tlv_put_failure: | |
2645 | out: | |
924794c9 | 2646 | fs_path_free(p); |
31db9f7c AB |
2647 | return ret; |
2648 | } | |
2649 | ||
1f4692da AB |
2650 | /* |
2651 | * We need some special handling for inodes that get processed before the parent | |
2652 | * directory got created. See process_recorded_refs for details. | |
2653 | * This function does the check if we already created the dir out of order. | |
2654 | */ | |
2655 | static int did_create_dir(struct send_ctx *sctx, u64 dir) | |
2656 | { | |
2657 | int ret = 0; | |
2658 | struct btrfs_path *path = NULL; | |
2659 | struct btrfs_key key; | |
2660 | struct btrfs_key found_key; | |
2661 | struct btrfs_key di_key; | |
2662 | struct extent_buffer *eb; | |
2663 | struct btrfs_dir_item *di; | |
2664 | int slot; | |
2665 | ||
2666 | path = alloc_path_for_send(); | |
2667 | if (!path) { | |
2668 | ret = -ENOMEM; | |
2669 | goto out; | |
2670 | } | |
2671 | ||
2672 | key.objectid = dir; | |
2673 | key.type = BTRFS_DIR_INDEX_KEY; | |
2674 | key.offset = 0; | |
dff6d0ad FDBM |
2675 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); |
2676 | if (ret < 0) | |
2677 | goto out; | |
2678 | ||
1f4692da | 2679 | while (1) { |
dff6d0ad FDBM |
2680 | eb = path->nodes[0]; |
2681 | slot = path->slots[0]; | |
2682 | if (slot >= btrfs_header_nritems(eb)) { | |
2683 | ret = btrfs_next_leaf(sctx->send_root, path); | |
2684 | if (ret < 0) { | |
2685 | goto out; | |
2686 | } else if (ret > 0) { | |
2687 | ret = 0; | |
2688 | break; | |
2689 | } | |
2690 | continue; | |
1f4692da | 2691 | } |
dff6d0ad FDBM |
2692 | |
2693 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
2694 | if (found_key.objectid != key.objectid || | |
1f4692da AB |
2695 | found_key.type != key.type) { |
2696 | ret = 0; | |
2697 | goto out; | |
2698 | } | |
2699 | ||
2700 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); | |
2701 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2702 | ||
a0525414 JB |
2703 | if (di_key.type != BTRFS_ROOT_ITEM_KEY && |
2704 | di_key.objectid < sctx->send_progress) { | |
1f4692da AB |
2705 | ret = 1; |
2706 | goto out; | |
2707 | } | |
2708 | ||
dff6d0ad | 2709 | path->slots[0]++; |
1f4692da AB |
2710 | } |
2711 | ||
2712 | out: | |
2713 | btrfs_free_path(path); | |
2714 | return ret; | |
2715 | } | |
2716 | ||
2717 | /* | |
2718 | * Only creates the inode if it is: | |
2719 | * 1. Not a directory | |
2720 | * 2. Or a directory which was not created already due to out of order | |
2721 | * directories. See did_create_dir and process_recorded_refs for details. | |
2722 | */ | |
2723 | static int send_create_inode_if_needed(struct send_ctx *sctx) | |
2724 | { | |
2725 | int ret; | |
2726 | ||
2727 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
2728 | ret = did_create_dir(sctx, sctx->cur_ino); | |
2729 | if (ret < 0) | |
2730 | goto out; | |
2731 | if (ret) { | |
2732 | ret = 0; | |
2733 | goto out; | |
2734 | } | |
2735 | } | |
2736 | ||
2737 | ret = send_create_inode(sctx, sctx->cur_ino); | |
2738 | if (ret < 0) | |
2739 | goto out; | |
2740 | ||
2741 | out: | |
2742 | return ret; | |
2743 | } | |
2744 | ||
31db9f7c AB |
2745 | struct recorded_ref { |
2746 | struct list_head list; | |
31db9f7c AB |
2747 | char *name; |
2748 | struct fs_path *full_path; | |
2749 | u64 dir; | |
2750 | u64 dir_gen; | |
31db9f7c AB |
2751 | int name_len; |
2752 | }; | |
2753 | ||
fdb13889 FM |
2754 | static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) |
2755 | { | |
2756 | ref->full_path = path; | |
2757 | ref->name = (char *)kbasename(ref->full_path->start); | |
2758 | ref->name_len = ref->full_path->end - ref->name; | |
2759 | } | |
2760 | ||
31db9f7c AB |
2761 | /* |
2762 | * We need to process new refs before deleted refs, but compare_tree gives us | |
2763 | * everything mixed. So we first record all refs and later process them. | |
2764 | * This function is a helper to record one ref. | |
2765 | */ | |
a4d96d62 | 2766 | static int __record_ref(struct list_head *head, u64 dir, |
31db9f7c AB |
2767 | u64 dir_gen, struct fs_path *path) |
2768 | { | |
2769 | struct recorded_ref *ref; | |
31db9f7c | 2770 | |
e780b0d1 | 2771 | ref = kmalloc(sizeof(*ref), GFP_KERNEL); |
31db9f7c AB |
2772 | if (!ref) |
2773 | return -ENOMEM; | |
2774 | ||
2775 | ref->dir = dir; | |
2776 | ref->dir_gen = dir_gen; | |
fdb13889 | 2777 | set_ref_path(ref, path); |
31db9f7c AB |
2778 | list_add_tail(&ref->list, head); |
2779 | return 0; | |
2780 | } | |
2781 | ||
ba5e8f2e JB |
2782 | static int dup_ref(struct recorded_ref *ref, struct list_head *list) |
2783 | { | |
2784 | struct recorded_ref *new; | |
2785 | ||
e780b0d1 | 2786 | new = kmalloc(sizeof(*ref), GFP_KERNEL); |
ba5e8f2e JB |
2787 | if (!new) |
2788 | return -ENOMEM; | |
2789 | ||
2790 | new->dir = ref->dir; | |
2791 | new->dir_gen = ref->dir_gen; | |
2792 | new->full_path = NULL; | |
2793 | INIT_LIST_HEAD(&new->list); | |
2794 | list_add_tail(&new->list, list); | |
2795 | return 0; | |
2796 | } | |
2797 | ||
924794c9 | 2798 | static void __free_recorded_refs(struct list_head *head) |
31db9f7c AB |
2799 | { |
2800 | struct recorded_ref *cur; | |
31db9f7c | 2801 | |
e938c8ad AB |
2802 | while (!list_empty(head)) { |
2803 | cur = list_entry(head->next, struct recorded_ref, list); | |
924794c9 | 2804 | fs_path_free(cur->full_path); |
e938c8ad | 2805 | list_del(&cur->list); |
31db9f7c AB |
2806 | kfree(cur); |
2807 | } | |
31db9f7c AB |
2808 | } |
2809 | ||
2810 | static void free_recorded_refs(struct send_ctx *sctx) | |
2811 | { | |
924794c9 TI |
2812 | __free_recorded_refs(&sctx->new_refs); |
2813 | __free_recorded_refs(&sctx->deleted_refs); | |
31db9f7c AB |
2814 | } |
2815 | ||
2816 | /* | |
766702ef | 2817 | * Renames/moves a file/dir to its orphan name. Used when the first |
31db9f7c AB |
2818 | * ref of an unprocessed inode gets overwritten and for all non empty |
2819 | * directories. | |
2820 | */ | |
2821 | static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, | |
2822 | struct fs_path *path) | |
2823 | { | |
2824 | int ret; | |
2825 | struct fs_path *orphan; | |
2826 | ||
924794c9 | 2827 | orphan = fs_path_alloc(); |
31db9f7c AB |
2828 | if (!orphan) |
2829 | return -ENOMEM; | |
2830 | ||
2831 | ret = gen_unique_name(sctx, ino, gen, orphan); | |
2832 | if (ret < 0) | |
2833 | goto out; | |
2834 | ||
2835 | ret = send_rename(sctx, path, orphan); | |
2836 | ||
2837 | out: | |
924794c9 | 2838 | fs_path_free(orphan); |
31db9f7c AB |
2839 | return ret; |
2840 | } | |
2841 | ||
9dc44214 FM |
2842 | static struct orphan_dir_info * |
2843 | add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2844 | { | |
2845 | struct rb_node **p = &sctx->orphan_dirs.rb_node; | |
2846 | struct rb_node *parent = NULL; | |
2847 | struct orphan_dir_info *entry, *odi; | |
2848 | ||
9dc44214 FM |
2849 | while (*p) { |
2850 | parent = *p; | |
2851 | entry = rb_entry(parent, struct orphan_dir_info, node); | |
2852 | if (dir_ino < entry->ino) { | |
2853 | p = &(*p)->rb_left; | |
2854 | } else if (dir_ino > entry->ino) { | |
2855 | p = &(*p)->rb_right; | |
2856 | } else { | |
9dc44214 FM |
2857 | return entry; |
2858 | } | |
2859 | } | |
2860 | ||
35c8eda1 RK |
2861 | odi = kmalloc(sizeof(*odi), GFP_KERNEL); |
2862 | if (!odi) | |
2863 | return ERR_PTR(-ENOMEM); | |
2864 | odi->ino = dir_ino; | |
2865 | odi->gen = 0; | |
0f96f517 | 2866 | odi->last_dir_index_offset = 0; |
35c8eda1 | 2867 | |
9dc44214 FM |
2868 | rb_link_node(&odi->node, parent, p); |
2869 | rb_insert_color(&odi->node, &sctx->orphan_dirs); | |
2870 | return odi; | |
2871 | } | |
2872 | ||
2873 | static struct orphan_dir_info * | |
2874 | get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2875 | { | |
2876 | struct rb_node *n = sctx->orphan_dirs.rb_node; | |
2877 | struct orphan_dir_info *entry; | |
2878 | ||
2879 | while (n) { | |
2880 | entry = rb_entry(n, struct orphan_dir_info, node); | |
2881 | if (dir_ino < entry->ino) | |
2882 | n = n->rb_left; | |
2883 | else if (dir_ino > entry->ino) | |
2884 | n = n->rb_right; | |
2885 | else | |
2886 | return entry; | |
2887 | } | |
2888 | return NULL; | |
2889 | } | |
2890 | ||
2891 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino) | |
2892 | { | |
2893 | struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino); | |
2894 | ||
2895 | return odi != NULL; | |
2896 | } | |
2897 | ||
2898 | static void free_orphan_dir_info(struct send_ctx *sctx, | |
2899 | struct orphan_dir_info *odi) | |
2900 | { | |
2901 | if (!odi) | |
2902 | return; | |
2903 | rb_erase(&odi->node, &sctx->orphan_dirs); | |
2904 | kfree(odi); | |
2905 | } | |
2906 | ||
31db9f7c AB |
2907 | /* |
2908 | * Returns 1 if a directory can be removed at this point in time. | |
2909 | * We check this by iterating all dir items and checking if the inode behind | |
2910 | * the dir item was already processed. | |
2911 | */ | |
9dc44214 FM |
2912 | static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
2913 | u64 send_progress) | |
31db9f7c AB |
2914 | { |
2915 | int ret = 0; | |
2916 | struct btrfs_root *root = sctx->parent_root; | |
2917 | struct btrfs_path *path; | |
2918 | struct btrfs_key key; | |
2919 | struct btrfs_key found_key; | |
2920 | struct btrfs_key loc; | |
2921 | struct btrfs_dir_item *di; | |
0f96f517 | 2922 | struct orphan_dir_info *odi = NULL; |
31db9f7c | 2923 | |
6d85ed05 AB |
2924 | /* |
2925 | * Don't try to rmdir the top/root subvolume dir. | |
2926 | */ | |
2927 | if (dir == BTRFS_FIRST_FREE_OBJECTID) | |
2928 | return 0; | |
2929 | ||
31db9f7c AB |
2930 | path = alloc_path_for_send(); |
2931 | if (!path) | |
2932 | return -ENOMEM; | |
2933 | ||
2934 | key.objectid = dir; | |
2935 | key.type = BTRFS_DIR_INDEX_KEY; | |
2936 | key.offset = 0; | |
0f96f517 RK |
2937 | |
2938 | odi = get_orphan_dir_info(sctx, dir); | |
2939 | if (odi) | |
2940 | key.offset = odi->last_dir_index_offset; | |
2941 | ||
dff6d0ad FDBM |
2942 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2943 | if (ret < 0) | |
2944 | goto out; | |
31db9f7c AB |
2945 | |
2946 | while (1) { | |
9dc44214 FM |
2947 | struct waiting_dir_move *dm; |
2948 | ||
dff6d0ad FDBM |
2949 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
2950 | ret = btrfs_next_leaf(root, path); | |
2951 | if (ret < 0) | |
2952 | goto out; | |
2953 | else if (ret > 0) | |
2954 | break; | |
2955 | continue; | |
31db9f7c | 2956 | } |
dff6d0ad FDBM |
2957 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2958 | path->slots[0]); | |
2959 | if (found_key.objectid != key.objectid || | |
2960 | found_key.type != key.type) | |
31db9f7c | 2961 | break; |
31db9f7c AB |
2962 | |
2963 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2964 | struct btrfs_dir_item); | |
2965 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
2966 | ||
9dc44214 FM |
2967 | dm = get_waiting_dir_move(sctx, loc.objectid); |
2968 | if (dm) { | |
9dc44214 FM |
2969 | odi = add_orphan_dir_info(sctx, dir); |
2970 | if (IS_ERR(odi)) { | |
2971 | ret = PTR_ERR(odi); | |
2972 | goto out; | |
2973 | } | |
2974 | odi->gen = dir_gen; | |
0f96f517 | 2975 | odi->last_dir_index_offset = found_key.offset; |
9dc44214 FM |
2976 | dm->rmdir_ino = dir; |
2977 | ret = 0; | |
2978 | goto out; | |
2979 | } | |
2980 | ||
31db9f7c | 2981 | if (loc.objectid > send_progress) { |
0f96f517 RK |
2982 | odi = add_orphan_dir_info(sctx, dir); |
2983 | if (IS_ERR(odi)) { | |
2984 | ret = PTR_ERR(odi); | |
2985 | goto out; | |
2986 | } | |
2987 | odi->gen = dir_gen; | |
2988 | odi->last_dir_index_offset = found_key.offset; | |
31db9f7c AB |
2989 | ret = 0; |
2990 | goto out; | |
2991 | } | |
2992 | ||
dff6d0ad | 2993 | path->slots[0]++; |
31db9f7c | 2994 | } |
0f96f517 | 2995 | free_orphan_dir_info(sctx, odi); |
31db9f7c AB |
2996 | |
2997 | ret = 1; | |
2998 | ||
2999 | out: | |
3000 | btrfs_free_path(path); | |
3001 | return ret; | |
3002 | } | |
3003 | ||
9f03740a FDBM |
3004 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) |
3005 | { | |
9dc44214 | 3006 | struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); |
9f03740a | 3007 | |
9dc44214 | 3008 | return entry != NULL; |
9f03740a FDBM |
3009 | } |
3010 | ||
8b191a68 | 3011 | static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) |
9f03740a FDBM |
3012 | { |
3013 | struct rb_node **p = &sctx->waiting_dir_moves.rb_node; | |
3014 | struct rb_node *parent = NULL; | |
3015 | struct waiting_dir_move *entry, *dm; | |
3016 | ||
e780b0d1 | 3017 | dm = kmalloc(sizeof(*dm), GFP_KERNEL); |
9f03740a FDBM |
3018 | if (!dm) |
3019 | return -ENOMEM; | |
3020 | dm->ino = ino; | |
9dc44214 | 3021 | dm->rmdir_ino = 0; |
8b191a68 | 3022 | dm->orphanized = orphanized; |
9f03740a FDBM |
3023 | |
3024 | while (*p) { | |
3025 | parent = *p; | |
3026 | entry = rb_entry(parent, struct waiting_dir_move, node); | |
3027 | if (ino < entry->ino) { | |
3028 | p = &(*p)->rb_left; | |
3029 | } else if (ino > entry->ino) { | |
3030 | p = &(*p)->rb_right; | |
3031 | } else { | |
3032 | kfree(dm); | |
3033 | return -EEXIST; | |
3034 | } | |
3035 | } | |
3036 | ||
3037 | rb_link_node(&dm->node, parent, p); | |
3038 | rb_insert_color(&dm->node, &sctx->waiting_dir_moves); | |
3039 | return 0; | |
3040 | } | |
3041 | ||
9dc44214 FM |
3042 | static struct waiting_dir_move * |
3043 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
9f03740a FDBM |
3044 | { |
3045 | struct rb_node *n = sctx->waiting_dir_moves.rb_node; | |
3046 | struct waiting_dir_move *entry; | |
3047 | ||
3048 | while (n) { | |
3049 | entry = rb_entry(n, struct waiting_dir_move, node); | |
9dc44214 | 3050 | if (ino < entry->ino) |
9f03740a | 3051 | n = n->rb_left; |
9dc44214 | 3052 | else if (ino > entry->ino) |
9f03740a | 3053 | n = n->rb_right; |
9dc44214 FM |
3054 | else |
3055 | return entry; | |
9f03740a | 3056 | } |
9dc44214 FM |
3057 | return NULL; |
3058 | } | |
3059 | ||
3060 | static void free_waiting_dir_move(struct send_ctx *sctx, | |
3061 | struct waiting_dir_move *dm) | |
3062 | { | |
3063 | if (!dm) | |
3064 | return; | |
3065 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
3066 | kfree(dm); | |
9f03740a FDBM |
3067 | } |
3068 | ||
bfa7e1f8 FM |
3069 | static int add_pending_dir_move(struct send_ctx *sctx, |
3070 | u64 ino, | |
3071 | u64 ino_gen, | |
f959492f FM |
3072 | u64 parent_ino, |
3073 | struct list_head *new_refs, | |
84471e24 FM |
3074 | struct list_head *deleted_refs, |
3075 | const bool is_orphan) | |
9f03740a FDBM |
3076 | { |
3077 | struct rb_node **p = &sctx->pending_dir_moves.rb_node; | |
3078 | struct rb_node *parent = NULL; | |
73b802f4 | 3079 | struct pending_dir_move *entry = NULL, *pm; |
9f03740a FDBM |
3080 | struct recorded_ref *cur; |
3081 | int exists = 0; | |
3082 | int ret; | |
3083 | ||
e780b0d1 | 3084 | pm = kmalloc(sizeof(*pm), GFP_KERNEL); |
9f03740a FDBM |
3085 | if (!pm) |
3086 | return -ENOMEM; | |
3087 | pm->parent_ino = parent_ino; | |
bfa7e1f8 FM |
3088 | pm->ino = ino; |
3089 | pm->gen = ino_gen; | |
9f03740a FDBM |
3090 | INIT_LIST_HEAD(&pm->list); |
3091 | INIT_LIST_HEAD(&pm->update_refs); | |
3092 | RB_CLEAR_NODE(&pm->node); | |
3093 | ||
3094 | while (*p) { | |
3095 | parent = *p; | |
3096 | entry = rb_entry(parent, struct pending_dir_move, node); | |
3097 | if (parent_ino < entry->parent_ino) { | |
3098 | p = &(*p)->rb_left; | |
3099 | } else if (parent_ino > entry->parent_ino) { | |
3100 | p = &(*p)->rb_right; | |
3101 | } else { | |
3102 | exists = 1; | |
3103 | break; | |
3104 | } | |
3105 | } | |
3106 | ||
f959492f | 3107 | list_for_each_entry(cur, deleted_refs, list) { |
9f03740a FDBM |
3108 | ret = dup_ref(cur, &pm->update_refs); |
3109 | if (ret < 0) | |
3110 | goto out; | |
3111 | } | |
f959492f | 3112 | list_for_each_entry(cur, new_refs, list) { |
9f03740a FDBM |
3113 | ret = dup_ref(cur, &pm->update_refs); |
3114 | if (ret < 0) | |
3115 | goto out; | |
3116 | } | |
3117 | ||
8b191a68 | 3118 | ret = add_waiting_dir_move(sctx, pm->ino, is_orphan); |
9f03740a FDBM |
3119 | if (ret) |
3120 | goto out; | |
3121 | ||
3122 | if (exists) { | |
3123 | list_add_tail(&pm->list, &entry->list); | |
3124 | } else { | |
3125 | rb_link_node(&pm->node, parent, p); | |
3126 | rb_insert_color(&pm->node, &sctx->pending_dir_moves); | |
3127 | } | |
3128 | ret = 0; | |
3129 | out: | |
3130 | if (ret) { | |
3131 | __free_recorded_refs(&pm->update_refs); | |
3132 | kfree(pm); | |
3133 | } | |
3134 | return ret; | |
3135 | } | |
3136 | ||
3137 | static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, | |
3138 | u64 parent_ino) | |
3139 | { | |
3140 | struct rb_node *n = sctx->pending_dir_moves.rb_node; | |
3141 | struct pending_dir_move *entry; | |
3142 | ||
3143 | while (n) { | |
3144 | entry = rb_entry(n, struct pending_dir_move, node); | |
3145 | if (parent_ino < entry->parent_ino) | |
3146 | n = n->rb_left; | |
3147 | else if (parent_ino > entry->parent_ino) | |
3148 | n = n->rb_right; | |
3149 | else | |
3150 | return entry; | |
3151 | } | |
3152 | return NULL; | |
3153 | } | |
3154 | ||
801bec36 RK |
3155 | static int path_loop(struct send_ctx *sctx, struct fs_path *name, |
3156 | u64 ino, u64 gen, u64 *ancestor_ino) | |
3157 | { | |
3158 | int ret = 0; | |
3159 | u64 parent_inode = 0; | |
3160 | u64 parent_gen = 0; | |
3161 | u64 start_ino = ino; | |
3162 | ||
3163 | *ancestor_ino = 0; | |
3164 | while (ino != BTRFS_FIRST_FREE_OBJECTID) { | |
3165 | fs_path_reset(name); | |
3166 | ||
3167 | if (is_waiting_for_rm(sctx, ino)) | |
3168 | break; | |
3169 | if (is_waiting_for_move(sctx, ino)) { | |
3170 | if (*ancestor_ino == 0) | |
3171 | *ancestor_ino = ino; | |
3172 | ret = get_first_ref(sctx->parent_root, ino, | |
3173 | &parent_inode, &parent_gen, name); | |
3174 | } else { | |
3175 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
3176 | &parent_inode, | |
3177 | &parent_gen, name); | |
3178 | if (ret > 0) { | |
3179 | ret = 0; | |
3180 | break; | |
3181 | } | |
3182 | } | |
3183 | if (ret < 0) | |
3184 | break; | |
3185 | if (parent_inode == start_ino) { | |
3186 | ret = 1; | |
3187 | if (*ancestor_ino == 0) | |
3188 | *ancestor_ino = ino; | |
3189 | break; | |
3190 | } | |
3191 | ino = parent_inode; | |
3192 | gen = parent_gen; | |
3193 | } | |
3194 | return ret; | |
3195 | } | |
3196 | ||
9f03740a FDBM |
3197 | static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) |
3198 | { | |
3199 | struct fs_path *from_path = NULL; | |
3200 | struct fs_path *to_path = NULL; | |
2b863a13 | 3201 | struct fs_path *name = NULL; |
9f03740a FDBM |
3202 | u64 orig_progress = sctx->send_progress; |
3203 | struct recorded_ref *cur; | |
2b863a13 | 3204 | u64 parent_ino, parent_gen; |
9dc44214 FM |
3205 | struct waiting_dir_move *dm = NULL; |
3206 | u64 rmdir_ino = 0; | |
801bec36 RK |
3207 | u64 ancestor; |
3208 | bool is_orphan; | |
9f03740a FDBM |
3209 | int ret; |
3210 | ||
2b863a13 | 3211 | name = fs_path_alloc(); |
9f03740a | 3212 | from_path = fs_path_alloc(); |
2b863a13 FM |
3213 | if (!name || !from_path) { |
3214 | ret = -ENOMEM; | |
3215 | goto out; | |
3216 | } | |
9f03740a | 3217 | |
9dc44214 FM |
3218 | dm = get_waiting_dir_move(sctx, pm->ino); |
3219 | ASSERT(dm); | |
3220 | rmdir_ino = dm->rmdir_ino; | |
801bec36 | 3221 | is_orphan = dm->orphanized; |
9dc44214 | 3222 | free_waiting_dir_move(sctx, dm); |
2b863a13 | 3223 | |
801bec36 | 3224 | if (is_orphan) { |
84471e24 FM |
3225 | ret = gen_unique_name(sctx, pm->ino, |
3226 | pm->gen, from_path); | |
3227 | } else { | |
3228 | ret = get_first_ref(sctx->parent_root, pm->ino, | |
3229 | &parent_ino, &parent_gen, name); | |
3230 | if (ret < 0) | |
3231 | goto out; | |
3232 | ret = get_cur_path(sctx, parent_ino, parent_gen, | |
3233 | from_path); | |
3234 | if (ret < 0) | |
3235 | goto out; | |
3236 | ret = fs_path_add_path(from_path, name); | |
3237 | } | |
c992ec94 FM |
3238 | if (ret < 0) |
3239 | goto out; | |
2b863a13 | 3240 | |
f959492f | 3241 | sctx->send_progress = sctx->cur_ino + 1; |
801bec36 | 3242 | ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor); |
7969e77a FM |
3243 | if (ret < 0) |
3244 | goto out; | |
801bec36 RK |
3245 | if (ret) { |
3246 | LIST_HEAD(deleted_refs); | |
3247 | ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID); | |
3248 | ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor, | |
3249 | &pm->update_refs, &deleted_refs, | |
3250 | is_orphan); | |
3251 | if (ret < 0) | |
3252 | goto out; | |
3253 | if (rmdir_ino) { | |
3254 | dm = get_waiting_dir_move(sctx, pm->ino); | |
3255 | ASSERT(dm); | |
3256 | dm->rmdir_ino = rmdir_ino; | |
3257 | } | |
3258 | goto out; | |
3259 | } | |
c992ec94 FM |
3260 | fs_path_reset(name); |
3261 | to_path = name; | |
2b863a13 | 3262 | name = NULL; |
9f03740a FDBM |
3263 | ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); |
3264 | if (ret < 0) | |
3265 | goto out; | |
3266 | ||
3267 | ret = send_rename(sctx, from_path, to_path); | |
3268 | if (ret < 0) | |
3269 | goto out; | |
3270 | ||
9dc44214 FM |
3271 | if (rmdir_ino) { |
3272 | struct orphan_dir_info *odi; | |
0f96f517 | 3273 | u64 gen; |
9dc44214 FM |
3274 | |
3275 | odi = get_orphan_dir_info(sctx, rmdir_ino); | |
3276 | if (!odi) { | |
3277 | /* already deleted */ | |
3278 | goto finish; | |
3279 | } | |
0f96f517 RK |
3280 | gen = odi->gen; |
3281 | ||
3282 | ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino); | |
9dc44214 FM |
3283 | if (ret < 0) |
3284 | goto out; | |
3285 | if (!ret) | |
3286 | goto finish; | |
3287 | ||
3288 | name = fs_path_alloc(); | |
3289 | if (!name) { | |
3290 | ret = -ENOMEM; | |
3291 | goto out; | |
3292 | } | |
0f96f517 | 3293 | ret = get_cur_path(sctx, rmdir_ino, gen, name); |
9dc44214 FM |
3294 | if (ret < 0) |
3295 | goto out; | |
3296 | ret = send_rmdir(sctx, name); | |
3297 | if (ret < 0) | |
3298 | goto out; | |
9dc44214 FM |
3299 | } |
3300 | ||
3301 | finish: | |
9f03740a FDBM |
3302 | ret = send_utimes(sctx, pm->ino, pm->gen); |
3303 | if (ret < 0) | |
3304 | goto out; | |
3305 | ||
3306 | /* | |
3307 | * After rename/move, need to update the utimes of both new parent(s) | |
3308 | * and old parent(s). | |
3309 | */ | |
3310 | list_for_each_entry(cur, &pm->update_refs, list) { | |
764433a1 RK |
3311 | /* |
3312 | * The parent inode might have been deleted in the send snapshot | |
3313 | */ | |
3314 | ret = get_inode_info(sctx->send_root, cur->dir, NULL, | |
3315 | NULL, NULL, NULL, NULL, NULL); | |
3316 | if (ret == -ENOENT) { | |
3317 | ret = 0; | |
9dc44214 | 3318 | continue; |
764433a1 RK |
3319 | } |
3320 | if (ret < 0) | |
3321 | goto out; | |
3322 | ||
9f03740a FDBM |
3323 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
3324 | if (ret < 0) | |
3325 | goto out; | |
3326 | } | |
3327 | ||
3328 | out: | |
2b863a13 | 3329 | fs_path_free(name); |
9f03740a FDBM |
3330 | fs_path_free(from_path); |
3331 | fs_path_free(to_path); | |
3332 | sctx->send_progress = orig_progress; | |
3333 | ||
3334 | return ret; | |
3335 | } | |
3336 | ||
3337 | static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) | |
3338 | { | |
3339 | if (!list_empty(&m->list)) | |
3340 | list_del(&m->list); | |
3341 | if (!RB_EMPTY_NODE(&m->node)) | |
3342 | rb_erase(&m->node, &sctx->pending_dir_moves); | |
3343 | __free_recorded_refs(&m->update_refs); | |
3344 | kfree(m); | |
3345 | } | |
3346 | ||
3347 | static void tail_append_pending_moves(struct pending_dir_move *moves, | |
3348 | struct list_head *stack) | |
3349 | { | |
3350 | if (list_empty(&moves->list)) { | |
3351 | list_add_tail(&moves->list, stack); | |
3352 | } else { | |
3353 | LIST_HEAD(list); | |
3354 | list_splice_init(&moves->list, &list); | |
3355 | list_add_tail(&moves->list, stack); | |
3356 | list_splice_tail(&list, stack); | |
3357 | } | |
3358 | } | |
3359 | ||
3360 | static int apply_children_dir_moves(struct send_ctx *sctx) | |
3361 | { | |
3362 | struct pending_dir_move *pm; | |
3363 | struct list_head stack; | |
3364 | u64 parent_ino = sctx->cur_ino; | |
3365 | int ret = 0; | |
3366 | ||
3367 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3368 | if (!pm) | |
3369 | return 0; | |
3370 | ||
3371 | INIT_LIST_HEAD(&stack); | |
3372 | tail_append_pending_moves(pm, &stack); | |
3373 | ||
3374 | while (!list_empty(&stack)) { | |
3375 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3376 | parent_ino = pm->ino; | |
3377 | ret = apply_dir_move(sctx, pm); | |
3378 | free_pending_move(sctx, pm); | |
3379 | if (ret) | |
3380 | goto out; | |
3381 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3382 | if (pm) | |
3383 | tail_append_pending_moves(pm, &stack); | |
3384 | } | |
3385 | return 0; | |
3386 | ||
3387 | out: | |
3388 | while (!list_empty(&stack)) { | |
3389 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3390 | free_pending_move(sctx, pm); | |
3391 | } | |
3392 | return ret; | |
3393 | } | |
3394 | ||
84471e24 FM |
3395 | /* |
3396 | * We might need to delay a directory rename even when no ancestor directory | |
3397 | * (in the send root) with a higher inode number than ours (sctx->cur_ino) was | |
3398 | * renamed. This happens when we rename a directory to the old name (the name | |
3399 | * in the parent root) of some other unrelated directory that got its rename | |
3400 | * delayed due to some ancestor with higher number that got renamed. | |
3401 | * | |
3402 | * Example: | |
3403 | * | |
3404 | * Parent snapshot: | |
3405 | * . (ino 256) | |
3406 | * |---- a/ (ino 257) | |
3407 | * | |---- file (ino 260) | |
3408 | * | | |
3409 | * |---- b/ (ino 258) | |
3410 | * |---- c/ (ino 259) | |
3411 | * | |
3412 | * Send snapshot: | |
3413 | * . (ino 256) | |
3414 | * |---- a/ (ino 258) | |
3415 | * |---- x/ (ino 259) | |
3416 | * |---- y/ (ino 257) | |
3417 | * |----- file (ino 260) | |
3418 | * | |
3419 | * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257 | |
3420 | * from 'a' to 'x/y' happening first, which in turn depends on the rename of | |
3421 | * inode 259 from 'c' to 'x'. So the order of rename commands the send stream | |
3422 | * must issue is: | |
3423 | * | |
3424 | * 1 - rename 259 from 'c' to 'x' | |
3425 | * 2 - rename 257 from 'a' to 'x/y' | |
3426 | * 3 - rename 258 from 'b' to 'a' | |
3427 | * | |
3428 | * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can | |
3429 | * be done right away and < 0 on error. | |
3430 | */ | |
3431 | static int wait_for_dest_dir_move(struct send_ctx *sctx, | |
3432 | struct recorded_ref *parent_ref, | |
3433 | const bool is_orphan) | |
3434 | { | |
2ff7e61e | 3435 | struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info; |
84471e24 FM |
3436 | struct btrfs_path *path; |
3437 | struct btrfs_key key; | |
3438 | struct btrfs_key di_key; | |
3439 | struct btrfs_dir_item *di; | |
3440 | u64 left_gen; | |
3441 | u64 right_gen; | |
3442 | int ret = 0; | |
801bec36 | 3443 | struct waiting_dir_move *wdm; |
84471e24 FM |
3444 | |
3445 | if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) | |
3446 | return 0; | |
3447 | ||
3448 | path = alloc_path_for_send(); | |
3449 | if (!path) | |
3450 | return -ENOMEM; | |
3451 | ||
3452 | key.objectid = parent_ref->dir; | |
3453 | key.type = BTRFS_DIR_ITEM_KEY; | |
3454 | key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); | |
3455 | ||
3456 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
3457 | if (ret < 0) { | |
3458 | goto out; | |
3459 | } else if (ret > 0) { | |
3460 | ret = 0; | |
3461 | goto out; | |
3462 | } | |
3463 | ||
2ff7e61e JM |
3464 | di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name, |
3465 | parent_ref->name_len); | |
84471e24 FM |
3466 | if (!di) { |
3467 | ret = 0; | |
3468 | goto out; | |
3469 | } | |
3470 | /* | |
3471 | * di_key.objectid has the number of the inode that has a dentry in the | |
3472 | * parent directory with the same name that sctx->cur_ino is being | |
3473 | * renamed to. We need to check if that inode is in the send root as | |
3474 | * well and if it is currently marked as an inode with a pending rename, | |
3475 | * if it is, we need to delay the rename of sctx->cur_ino as well, so | |
3476 | * that it happens after that other inode is renamed. | |
3477 | */ | |
3478 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); | |
3479 | if (di_key.type != BTRFS_INODE_ITEM_KEY) { | |
3480 | ret = 0; | |
3481 | goto out; | |
3482 | } | |
3483 | ||
3484 | ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL, | |
3485 | &left_gen, NULL, NULL, NULL, NULL); | |
3486 | if (ret < 0) | |
3487 | goto out; | |
3488 | ret = get_inode_info(sctx->send_root, di_key.objectid, NULL, | |
3489 | &right_gen, NULL, NULL, NULL, NULL); | |
3490 | if (ret < 0) { | |
3491 | if (ret == -ENOENT) | |
3492 | ret = 0; | |
3493 | goto out; | |
3494 | } | |
3495 | ||
3496 | /* Different inode, no need to delay the rename of sctx->cur_ino */ | |
3497 | if (right_gen != left_gen) { | |
3498 | ret = 0; | |
3499 | goto out; | |
3500 | } | |
3501 | ||
801bec36 RK |
3502 | wdm = get_waiting_dir_move(sctx, di_key.objectid); |
3503 | if (wdm && !wdm->orphanized) { | |
84471e24 FM |
3504 | ret = add_pending_dir_move(sctx, |
3505 | sctx->cur_ino, | |
3506 | sctx->cur_inode_gen, | |
3507 | di_key.objectid, | |
3508 | &sctx->new_refs, | |
3509 | &sctx->deleted_refs, | |
3510 | is_orphan); | |
3511 | if (!ret) | |
3512 | ret = 1; | |
3513 | } | |
3514 | out: | |
3515 | btrfs_free_path(path); | |
3516 | return ret; | |
3517 | } | |
3518 | ||
80aa6027 | 3519 | /* |
ea37d599 FM |
3520 | * Check if inode ino2, or any of its ancestors, is inode ino1. |
3521 | * Return 1 if true, 0 if false and < 0 on error. | |
3522 | */ | |
3523 | static int check_ino_in_path(struct btrfs_root *root, | |
3524 | const u64 ino1, | |
3525 | const u64 ino1_gen, | |
3526 | const u64 ino2, | |
3527 | const u64 ino2_gen, | |
3528 | struct fs_path *fs_path) | |
3529 | { | |
3530 | u64 ino = ino2; | |
3531 | ||
3532 | if (ino1 == ino2) | |
3533 | return ino1_gen == ino2_gen; | |
3534 | ||
3535 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { | |
3536 | u64 parent; | |
3537 | u64 parent_gen; | |
3538 | int ret; | |
3539 | ||
3540 | fs_path_reset(fs_path); | |
3541 | ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path); | |
3542 | if (ret < 0) | |
3543 | return ret; | |
3544 | if (parent == ino1) | |
3545 | return parent_gen == ino1_gen; | |
3546 | ino = parent; | |
3547 | } | |
3548 | return 0; | |
3549 | } | |
3550 | ||
3551 | /* | |
3552 | * Check if ino ino1 is an ancestor of inode ino2 in the given root for any | |
3553 | * possible path (in case ino2 is not a directory and has multiple hard links). | |
80aa6027 FM |
3554 | * Return 1 if true, 0 if false and < 0 on error. |
3555 | */ | |
3556 | static int is_ancestor(struct btrfs_root *root, | |
3557 | const u64 ino1, | |
3558 | const u64 ino1_gen, | |
3559 | const u64 ino2, | |
3560 | struct fs_path *fs_path) | |
3561 | { | |
ea37d599 | 3562 | bool free_fs_path = false; |
72c3668f | 3563 | int ret = 0; |
ea37d599 FM |
3564 | struct btrfs_path *path = NULL; |
3565 | struct btrfs_key key; | |
72c3668f FM |
3566 | |
3567 | if (!fs_path) { | |
3568 | fs_path = fs_path_alloc(); | |
3569 | if (!fs_path) | |
3570 | return -ENOMEM; | |
ea37d599 | 3571 | free_fs_path = true; |
72c3668f | 3572 | } |
80aa6027 | 3573 | |
ea37d599 FM |
3574 | path = alloc_path_for_send(); |
3575 | if (!path) { | |
3576 | ret = -ENOMEM; | |
3577 | goto out; | |
3578 | } | |
80aa6027 | 3579 | |
ea37d599 FM |
3580 | key.objectid = ino2; |
3581 | key.type = BTRFS_INODE_REF_KEY; | |
3582 | key.offset = 0; | |
3583 | ||
3584 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3585 | if (ret < 0) | |
3586 | goto out; | |
3587 | ||
3588 | while (true) { | |
3589 | struct extent_buffer *leaf = path->nodes[0]; | |
3590 | int slot = path->slots[0]; | |
3591 | u32 cur_offset = 0; | |
3592 | u32 item_size; | |
3593 | ||
3594 | if (slot >= btrfs_header_nritems(leaf)) { | |
3595 | ret = btrfs_next_leaf(root, path); | |
3596 | if (ret < 0) | |
3597 | goto out; | |
3598 | if (ret > 0) | |
3599 | break; | |
3600 | continue; | |
72c3668f | 3601 | } |
ea37d599 FM |
3602 | |
3603 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
3604 | if (key.objectid != ino2) | |
3605 | break; | |
3606 | if (key.type != BTRFS_INODE_REF_KEY && | |
3607 | key.type != BTRFS_INODE_EXTREF_KEY) | |
3608 | break; | |
3609 | ||
3610 | item_size = btrfs_item_size_nr(leaf, slot); | |
3611 | while (cur_offset < item_size) { | |
3612 | u64 parent; | |
3613 | u64 parent_gen; | |
3614 | ||
3615 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
3616 | unsigned long ptr; | |
3617 | struct btrfs_inode_extref *extref; | |
3618 | ||
3619 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3620 | extref = (struct btrfs_inode_extref *) | |
3621 | (ptr + cur_offset); | |
3622 | parent = btrfs_inode_extref_parent(leaf, | |
3623 | extref); | |
3624 | cur_offset += sizeof(*extref); | |
3625 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
3626 | extref); | |
3627 | } else { | |
3628 | parent = key.offset; | |
3629 | cur_offset = item_size; | |
3630 | } | |
3631 | ||
3632 | ret = get_inode_info(root, parent, NULL, &parent_gen, | |
3633 | NULL, NULL, NULL, NULL); | |
3634 | if (ret < 0) | |
3635 | goto out; | |
3636 | ret = check_ino_in_path(root, ino1, ino1_gen, | |
3637 | parent, parent_gen, fs_path); | |
3638 | if (ret) | |
3639 | goto out; | |
80aa6027 | 3640 | } |
ea37d599 | 3641 | path->slots[0]++; |
80aa6027 | 3642 | } |
ea37d599 | 3643 | ret = 0; |
72c3668f | 3644 | out: |
ea37d599 FM |
3645 | btrfs_free_path(path); |
3646 | if (free_fs_path) | |
72c3668f FM |
3647 | fs_path_free(fs_path); |
3648 | return ret; | |
80aa6027 FM |
3649 | } |
3650 | ||
9f03740a | 3651 | static int wait_for_parent_move(struct send_ctx *sctx, |
8b191a68 FM |
3652 | struct recorded_ref *parent_ref, |
3653 | const bool is_orphan) | |
9f03740a | 3654 | { |
f959492f | 3655 | int ret = 0; |
9f03740a | 3656 | u64 ino = parent_ref->dir; |
fe9c798d | 3657 | u64 ino_gen = parent_ref->dir_gen; |
9f03740a | 3658 | u64 parent_ino_before, parent_ino_after; |
9f03740a FDBM |
3659 | struct fs_path *path_before = NULL; |
3660 | struct fs_path *path_after = NULL; | |
3661 | int len1, len2; | |
9f03740a FDBM |
3662 | |
3663 | path_after = fs_path_alloc(); | |
f959492f FM |
3664 | path_before = fs_path_alloc(); |
3665 | if (!path_after || !path_before) { | |
9f03740a FDBM |
3666 | ret = -ENOMEM; |
3667 | goto out; | |
3668 | } | |
3669 | ||
bfa7e1f8 | 3670 | /* |
f959492f FM |
3671 | * Our current directory inode may not yet be renamed/moved because some |
3672 | * ancestor (immediate or not) has to be renamed/moved first. So find if | |
3673 | * such ancestor exists and make sure our own rename/move happens after | |
80aa6027 FM |
3674 | * that ancestor is processed to avoid path build infinite loops (done |
3675 | * at get_cur_path()). | |
bfa7e1f8 | 3676 | */ |
f959492f | 3677 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { |
fe9c798d FM |
3678 | u64 parent_ino_after_gen; |
3679 | ||
f959492f | 3680 | if (is_waiting_for_move(sctx, ino)) { |
80aa6027 FM |
3681 | /* |
3682 | * If the current inode is an ancestor of ino in the | |
3683 | * parent root, we need to delay the rename of the | |
3684 | * current inode, otherwise don't delayed the rename | |
3685 | * because we can end up with a circular dependency | |
3686 | * of renames, resulting in some directories never | |
3687 | * getting the respective rename operations issued in | |
3688 | * the send stream or getting into infinite path build | |
3689 | * loops. | |
3690 | */ | |
3691 | ret = is_ancestor(sctx->parent_root, | |
3692 | sctx->cur_ino, sctx->cur_inode_gen, | |
3693 | ino, path_before); | |
4122ea64 FM |
3694 | if (ret) |
3695 | break; | |
f959492f | 3696 | } |
bfa7e1f8 FM |
3697 | |
3698 | fs_path_reset(path_before); | |
3699 | fs_path_reset(path_after); | |
3700 | ||
3701 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
fe9c798d | 3702 | &parent_ino_after_gen, path_after); |
bfa7e1f8 FM |
3703 | if (ret < 0) |
3704 | goto out; | |
3705 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
3706 | NULL, path_before); | |
f959492f | 3707 | if (ret < 0 && ret != -ENOENT) { |
bfa7e1f8 | 3708 | goto out; |
f959492f | 3709 | } else if (ret == -ENOENT) { |
bf8e8ca6 | 3710 | ret = 0; |
f959492f | 3711 | break; |
bfa7e1f8 FM |
3712 | } |
3713 | ||
3714 | len1 = fs_path_len(path_before); | |
3715 | len2 = fs_path_len(path_after); | |
f959492f FM |
3716 | if (ino > sctx->cur_ino && |
3717 | (parent_ino_before != parent_ino_after || len1 != len2 || | |
3718 | memcmp(path_before->start, path_after->start, len1))) { | |
fe9c798d FM |
3719 | u64 parent_ino_gen; |
3720 | ||
3721 | ret = get_inode_info(sctx->parent_root, ino, NULL, | |
3722 | &parent_ino_gen, NULL, NULL, NULL, | |
3723 | NULL); | |
3724 | if (ret < 0) | |
3725 | goto out; | |
3726 | if (ino_gen == parent_ino_gen) { | |
3727 | ret = 1; | |
3728 | break; | |
3729 | } | |
bfa7e1f8 | 3730 | } |
bfa7e1f8 | 3731 | ino = parent_ino_after; |
fe9c798d | 3732 | ino_gen = parent_ino_after_gen; |
bfa7e1f8 FM |
3733 | } |
3734 | ||
9f03740a FDBM |
3735 | out: |
3736 | fs_path_free(path_before); | |
3737 | fs_path_free(path_after); | |
3738 | ||
f959492f FM |
3739 | if (ret == 1) { |
3740 | ret = add_pending_dir_move(sctx, | |
3741 | sctx->cur_ino, | |
3742 | sctx->cur_inode_gen, | |
3743 | ino, | |
3744 | &sctx->new_refs, | |
84471e24 | 3745 | &sctx->deleted_refs, |
8b191a68 | 3746 | is_orphan); |
f959492f FM |
3747 | if (!ret) |
3748 | ret = 1; | |
3749 | } | |
3750 | ||
9f03740a FDBM |
3751 | return ret; |
3752 | } | |
3753 | ||
f5962781 FM |
3754 | static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) |
3755 | { | |
3756 | int ret; | |
3757 | struct fs_path *new_path; | |
3758 | ||
3759 | /* | |
3760 | * Our reference's name member points to its full_path member string, so | |
3761 | * we use here a new path. | |
3762 | */ | |
3763 | new_path = fs_path_alloc(); | |
3764 | if (!new_path) | |
3765 | return -ENOMEM; | |
3766 | ||
3767 | ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path); | |
3768 | if (ret < 0) { | |
3769 | fs_path_free(new_path); | |
3770 | return ret; | |
3771 | } | |
3772 | ret = fs_path_add(new_path, ref->name, ref->name_len); | |
3773 | if (ret < 0) { | |
3774 | fs_path_free(new_path); | |
3775 | return ret; | |
3776 | } | |
3777 | ||
3778 | fs_path_free(ref->full_path); | |
3779 | set_ref_path(ref, new_path); | |
3780 | ||
3781 | return 0; | |
3782 | } | |
3783 | ||
31db9f7c AB |
3784 | /* |
3785 | * This does all the move/link/unlink/rmdir magic. | |
3786 | */ | |
9f03740a | 3787 | static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) |
31db9f7c | 3788 | { |
04ab956e | 3789 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
3790 | int ret = 0; |
3791 | struct recorded_ref *cur; | |
1f4692da | 3792 | struct recorded_ref *cur2; |
ba5e8f2e | 3793 | struct list_head check_dirs; |
31db9f7c | 3794 | struct fs_path *valid_path = NULL; |
b24baf69 | 3795 | u64 ow_inode = 0; |
31db9f7c | 3796 | u64 ow_gen; |
f5962781 | 3797 | u64 ow_mode; |
31db9f7c AB |
3798 | int did_overwrite = 0; |
3799 | int is_orphan = 0; | |
29d6d30f | 3800 | u64 last_dir_ino_rm = 0; |
84471e24 | 3801 | bool can_rename = true; |
f5962781 | 3802 | bool orphanized_dir = false; |
fdb13889 | 3803 | bool orphanized_ancestor = false; |
31db9f7c | 3804 | |
04ab956e | 3805 | btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino); |
31db9f7c | 3806 | |
6d85ed05 AB |
3807 | /* |
3808 | * This should never happen as the root dir always has the same ref | |
3809 | * which is always '..' | |
3810 | */ | |
3811 | BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); | |
ba5e8f2e | 3812 | INIT_LIST_HEAD(&check_dirs); |
6d85ed05 | 3813 | |
924794c9 | 3814 | valid_path = fs_path_alloc(); |
31db9f7c AB |
3815 | if (!valid_path) { |
3816 | ret = -ENOMEM; | |
3817 | goto out; | |
3818 | } | |
3819 | ||
31db9f7c AB |
3820 | /* |
3821 | * First, check if the first ref of the current inode was overwritten | |
3822 | * before. If yes, we know that the current inode was already orphanized | |
3823 | * and thus use the orphan name. If not, we can use get_cur_path to | |
3824 | * get the path of the first ref as it would like while receiving at | |
3825 | * this point in time. | |
3826 | * New inodes are always orphan at the beginning, so force to use the | |
3827 | * orphan name in this case. | |
3828 | * The first ref is stored in valid_path and will be updated if it | |
3829 | * gets moved around. | |
3830 | */ | |
3831 | if (!sctx->cur_inode_new) { | |
3832 | ret = did_overwrite_first_ref(sctx, sctx->cur_ino, | |
3833 | sctx->cur_inode_gen); | |
3834 | if (ret < 0) | |
3835 | goto out; | |
3836 | if (ret) | |
3837 | did_overwrite = 1; | |
3838 | } | |
3839 | if (sctx->cur_inode_new || did_overwrite) { | |
3840 | ret = gen_unique_name(sctx, sctx->cur_ino, | |
3841 | sctx->cur_inode_gen, valid_path); | |
3842 | if (ret < 0) | |
3843 | goto out; | |
3844 | is_orphan = 1; | |
3845 | } else { | |
3846 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
3847 | valid_path); | |
3848 | if (ret < 0) | |
3849 | goto out; | |
3850 | } | |
3851 | ||
3852 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
1f4692da AB |
3853 | /* |
3854 | * We may have refs where the parent directory does not exist | |
3855 | * yet. This happens if the parent directories inum is higher | |
3856 | * the the current inum. To handle this case, we create the | |
3857 | * parent directory out of order. But we need to check if this | |
3858 | * did already happen before due to other refs in the same dir. | |
3859 | */ | |
3860 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); | |
3861 | if (ret < 0) | |
3862 | goto out; | |
3863 | if (ret == inode_state_will_create) { | |
3864 | ret = 0; | |
3865 | /* | |
3866 | * First check if any of the current inodes refs did | |
3867 | * already create the dir. | |
3868 | */ | |
3869 | list_for_each_entry(cur2, &sctx->new_refs, list) { | |
3870 | if (cur == cur2) | |
3871 | break; | |
3872 | if (cur2->dir == cur->dir) { | |
3873 | ret = 1; | |
3874 | break; | |
3875 | } | |
3876 | } | |
3877 | ||
3878 | /* | |
3879 | * If that did not happen, check if a previous inode | |
3880 | * did already create the dir. | |
3881 | */ | |
3882 | if (!ret) | |
3883 | ret = did_create_dir(sctx, cur->dir); | |
3884 | if (ret < 0) | |
3885 | goto out; | |
3886 | if (!ret) { | |
3887 | ret = send_create_inode(sctx, cur->dir); | |
3888 | if (ret < 0) | |
3889 | goto out; | |
3890 | } | |
3891 | } | |
3892 | ||
31db9f7c AB |
3893 | /* |
3894 | * Check if this new ref would overwrite the first ref of | |
3895 | * another unprocessed inode. If yes, orphanize the | |
3896 | * overwritten inode. If we find an overwritten ref that is | |
3897 | * not the first ref, simply unlink it. | |
3898 | */ | |
3899 | ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
3900 | cur->name, cur->name_len, | |
f5962781 | 3901 | &ow_inode, &ow_gen, &ow_mode); |
31db9f7c AB |
3902 | if (ret < 0) |
3903 | goto out; | |
3904 | if (ret) { | |
924794c9 TI |
3905 | ret = is_first_ref(sctx->parent_root, |
3906 | ow_inode, cur->dir, cur->name, | |
3907 | cur->name_len); | |
31db9f7c AB |
3908 | if (ret < 0) |
3909 | goto out; | |
3910 | if (ret) { | |
8996a48c | 3911 | struct name_cache_entry *nce; |
801bec36 | 3912 | struct waiting_dir_move *wdm; |
8996a48c | 3913 | |
31db9f7c AB |
3914 | ret = orphanize_inode(sctx, ow_inode, ow_gen, |
3915 | cur->full_path); | |
3916 | if (ret < 0) | |
3917 | goto out; | |
f5962781 FM |
3918 | if (S_ISDIR(ow_mode)) |
3919 | orphanized_dir = true; | |
801bec36 RK |
3920 | |
3921 | /* | |
3922 | * If ow_inode has its rename operation delayed | |
3923 | * make sure that its orphanized name is used in | |
3924 | * the source path when performing its rename | |
3925 | * operation. | |
3926 | */ | |
3927 | if (is_waiting_for_move(sctx, ow_inode)) { | |
3928 | wdm = get_waiting_dir_move(sctx, | |
3929 | ow_inode); | |
3930 | ASSERT(wdm); | |
3931 | wdm->orphanized = true; | |
3932 | } | |
3933 | ||
8996a48c FM |
3934 | /* |
3935 | * Make sure we clear our orphanized inode's | |
3936 | * name from the name cache. This is because the | |
3937 | * inode ow_inode might be an ancestor of some | |
3938 | * other inode that will be orphanized as well | |
3939 | * later and has an inode number greater than | |
3940 | * sctx->send_progress. We need to prevent | |
3941 | * future name lookups from using the old name | |
3942 | * and get instead the orphan name. | |
3943 | */ | |
3944 | nce = name_cache_search(sctx, ow_inode, ow_gen); | |
3945 | if (nce) { | |
3946 | name_cache_delete(sctx, nce); | |
3947 | kfree(nce); | |
3948 | } | |
801bec36 RK |
3949 | |
3950 | /* | |
3951 | * ow_inode might currently be an ancestor of | |
3952 | * cur_ino, therefore compute valid_path (the | |
3953 | * current path of cur_ino) again because it | |
3954 | * might contain the pre-orphanization name of | |
3955 | * ow_inode, which is no longer valid. | |
3956 | */ | |
72c3668f FM |
3957 | ret = is_ancestor(sctx->parent_root, |
3958 | ow_inode, ow_gen, | |
3959 | sctx->cur_ino, NULL); | |
3960 | if (ret > 0) { | |
fdb13889 | 3961 | orphanized_ancestor = true; |
72c3668f FM |
3962 | fs_path_reset(valid_path); |
3963 | ret = get_cur_path(sctx, sctx->cur_ino, | |
3964 | sctx->cur_inode_gen, | |
3965 | valid_path); | |
3966 | } | |
801bec36 RK |
3967 | if (ret < 0) |
3968 | goto out; | |
31db9f7c AB |
3969 | } else { |
3970 | ret = send_unlink(sctx, cur->full_path); | |
3971 | if (ret < 0) | |
3972 | goto out; | |
3973 | } | |
3974 | } | |
3975 | ||
84471e24 FM |
3976 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { |
3977 | ret = wait_for_dest_dir_move(sctx, cur, is_orphan); | |
3978 | if (ret < 0) | |
3979 | goto out; | |
3980 | if (ret == 1) { | |
3981 | can_rename = false; | |
3982 | *pending_move = 1; | |
3983 | } | |
3984 | } | |
3985 | ||
8b191a68 FM |
3986 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root && |
3987 | can_rename) { | |
3988 | ret = wait_for_parent_move(sctx, cur, is_orphan); | |
3989 | if (ret < 0) | |
3990 | goto out; | |
3991 | if (ret == 1) { | |
3992 | can_rename = false; | |
3993 | *pending_move = 1; | |
3994 | } | |
3995 | } | |
3996 | ||
31db9f7c AB |
3997 | /* |
3998 | * link/move the ref to the new place. If we have an orphan | |
3999 | * inode, move it and update valid_path. If not, link or move | |
4000 | * it depending on the inode mode. | |
4001 | */ | |
84471e24 | 4002 | if (is_orphan && can_rename) { |
31db9f7c AB |
4003 | ret = send_rename(sctx, valid_path, cur->full_path); |
4004 | if (ret < 0) | |
4005 | goto out; | |
4006 | is_orphan = 0; | |
4007 | ret = fs_path_copy(valid_path, cur->full_path); | |
4008 | if (ret < 0) | |
4009 | goto out; | |
84471e24 | 4010 | } else if (can_rename) { |
31db9f7c AB |
4011 | if (S_ISDIR(sctx->cur_inode_mode)) { |
4012 | /* | |
4013 | * Dirs can't be linked, so move it. For moved | |
4014 | * dirs, we always have one new and one deleted | |
4015 | * ref. The deleted ref is ignored later. | |
4016 | */ | |
8b191a68 FM |
4017 | ret = send_rename(sctx, valid_path, |
4018 | cur->full_path); | |
4019 | if (!ret) | |
4020 | ret = fs_path_copy(valid_path, | |
4021 | cur->full_path); | |
31db9f7c AB |
4022 | if (ret < 0) |
4023 | goto out; | |
4024 | } else { | |
f5962781 FM |
4025 | /* |
4026 | * We might have previously orphanized an inode | |
4027 | * which is an ancestor of our current inode, | |
4028 | * so our reference's full path, which was | |
4029 | * computed before any such orphanizations, must | |
4030 | * be updated. | |
4031 | */ | |
4032 | if (orphanized_dir) { | |
4033 | ret = update_ref_path(sctx, cur); | |
4034 | if (ret < 0) | |
4035 | goto out; | |
4036 | } | |
31db9f7c AB |
4037 | ret = send_link(sctx, cur->full_path, |
4038 | valid_path); | |
4039 | if (ret < 0) | |
4040 | goto out; | |
4041 | } | |
4042 | } | |
ba5e8f2e | 4043 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4044 | if (ret < 0) |
4045 | goto out; | |
4046 | } | |
4047 | ||
4048 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { | |
4049 | /* | |
4050 | * Check if we can already rmdir the directory. If not, | |
4051 | * orphanize it. For every dir item inside that gets deleted | |
4052 | * later, we do this check again and rmdir it then if possible. | |
4053 | * See the use of check_dirs for more details. | |
4054 | */ | |
9dc44214 FM |
4055 | ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
4056 | sctx->cur_ino); | |
31db9f7c AB |
4057 | if (ret < 0) |
4058 | goto out; | |
4059 | if (ret) { | |
4060 | ret = send_rmdir(sctx, valid_path); | |
4061 | if (ret < 0) | |
4062 | goto out; | |
4063 | } else if (!is_orphan) { | |
4064 | ret = orphanize_inode(sctx, sctx->cur_ino, | |
4065 | sctx->cur_inode_gen, valid_path); | |
4066 | if (ret < 0) | |
4067 | goto out; | |
4068 | is_orphan = 1; | |
4069 | } | |
4070 | ||
4071 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
ba5e8f2e | 4072 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4073 | if (ret < 0) |
4074 | goto out; | |
4075 | } | |
ccf1626b AB |
4076 | } else if (S_ISDIR(sctx->cur_inode_mode) && |
4077 | !list_empty(&sctx->deleted_refs)) { | |
4078 | /* | |
4079 | * We have a moved dir. Add the old parent to check_dirs | |
4080 | */ | |
4081 | cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, | |
4082 | list); | |
ba5e8f2e | 4083 | ret = dup_ref(cur, &check_dirs); |
ccf1626b AB |
4084 | if (ret < 0) |
4085 | goto out; | |
31db9f7c AB |
4086 | } else if (!S_ISDIR(sctx->cur_inode_mode)) { |
4087 | /* | |
4088 | * We have a non dir inode. Go through all deleted refs and | |
4089 | * unlink them if they were not already overwritten by other | |
4090 | * inodes. | |
4091 | */ | |
4092 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
4093 | ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
4094 | sctx->cur_ino, sctx->cur_inode_gen, | |
4095 | cur->name, cur->name_len); | |
4096 | if (ret < 0) | |
4097 | goto out; | |
4098 | if (!ret) { | |
fdb13889 FM |
4099 | /* |
4100 | * If we orphanized any ancestor before, we need | |
4101 | * to recompute the full path for deleted names, | |
4102 | * since any such path was computed before we | |
4103 | * processed any references and orphanized any | |
4104 | * ancestor inode. | |
4105 | */ | |
4106 | if (orphanized_ancestor) { | |
f5962781 FM |
4107 | ret = update_ref_path(sctx, cur); |
4108 | if (ret < 0) | |
fdb13889 | 4109 | goto out; |
fdb13889 | 4110 | } |
1f4692da AB |
4111 | ret = send_unlink(sctx, cur->full_path); |
4112 | if (ret < 0) | |
4113 | goto out; | |
31db9f7c | 4114 | } |
ba5e8f2e | 4115 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4116 | if (ret < 0) |
4117 | goto out; | |
4118 | } | |
31db9f7c AB |
4119 | /* |
4120 | * If the inode is still orphan, unlink the orphan. This may | |
4121 | * happen when a previous inode did overwrite the first ref | |
4122 | * of this inode and no new refs were added for the current | |
766702ef AB |
4123 | * inode. Unlinking does not mean that the inode is deleted in |
4124 | * all cases. There may still be links to this inode in other | |
4125 | * places. | |
31db9f7c | 4126 | */ |
1f4692da | 4127 | if (is_orphan) { |
31db9f7c AB |
4128 | ret = send_unlink(sctx, valid_path); |
4129 | if (ret < 0) | |
4130 | goto out; | |
4131 | } | |
4132 | } | |
4133 | ||
4134 | /* | |
4135 | * We did collect all parent dirs where cur_inode was once located. We | |
4136 | * now go through all these dirs and check if they are pending for | |
4137 | * deletion and if it's finally possible to perform the rmdir now. | |
4138 | * We also update the inode stats of the parent dirs here. | |
4139 | */ | |
ba5e8f2e | 4140 | list_for_each_entry(cur, &check_dirs, list) { |
766702ef AB |
4141 | /* |
4142 | * In case we had refs into dirs that were not processed yet, | |
4143 | * we don't need to do the utime and rmdir logic for these dirs. | |
4144 | * The dir will be processed later. | |
4145 | */ | |
ba5e8f2e | 4146 | if (cur->dir > sctx->cur_ino) |
31db9f7c AB |
4147 | continue; |
4148 | ||
ba5e8f2e | 4149 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4150 | if (ret < 0) |
4151 | goto out; | |
4152 | ||
4153 | if (ret == inode_state_did_create || | |
4154 | ret == inode_state_no_change) { | |
4155 | /* TODO delayed utimes */ | |
ba5e8f2e | 4156 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4157 | if (ret < 0) |
4158 | goto out; | |
29d6d30f FM |
4159 | } else if (ret == inode_state_did_delete && |
4160 | cur->dir != last_dir_ino_rm) { | |
9dc44214 FM |
4161 | ret = can_rmdir(sctx, cur->dir, cur->dir_gen, |
4162 | sctx->cur_ino); | |
31db9f7c AB |
4163 | if (ret < 0) |
4164 | goto out; | |
4165 | if (ret) { | |
ba5e8f2e JB |
4166 | ret = get_cur_path(sctx, cur->dir, |
4167 | cur->dir_gen, valid_path); | |
31db9f7c AB |
4168 | if (ret < 0) |
4169 | goto out; | |
4170 | ret = send_rmdir(sctx, valid_path); | |
4171 | if (ret < 0) | |
4172 | goto out; | |
29d6d30f | 4173 | last_dir_ino_rm = cur->dir; |
31db9f7c AB |
4174 | } |
4175 | } | |
4176 | } | |
4177 | ||
31db9f7c AB |
4178 | ret = 0; |
4179 | ||
4180 | out: | |
ba5e8f2e | 4181 | __free_recorded_refs(&check_dirs); |
31db9f7c | 4182 | free_recorded_refs(sctx); |
924794c9 | 4183 | fs_path_free(valid_path); |
31db9f7c AB |
4184 | return ret; |
4185 | } | |
4186 | ||
a0357511 NB |
4187 | static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name, |
4188 | void *ctx, struct list_head *refs) | |
31db9f7c AB |
4189 | { |
4190 | int ret = 0; | |
4191 | struct send_ctx *sctx = ctx; | |
4192 | struct fs_path *p; | |
4193 | u64 gen; | |
4194 | ||
924794c9 | 4195 | p = fs_path_alloc(); |
31db9f7c AB |
4196 | if (!p) |
4197 | return -ENOMEM; | |
4198 | ||
a4d96d62 | 4199 | ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, |
85a7b33b | 4200 | NULL, NULL); |
31db9f7c AB |
4201 | if (ret < 0) |
4202 | goto out; | |
4203 | ||
31db9f7c AB |
4204 | ret = get_cur_path(sctx, dir, gen, p); |
4205 | if (ret < 0) | |
4206 | goto out; | |
4207 | ret = fs_path_add_path(p, name); | |
4208 | if (ret < 0) | |
4209 | goto out; | |
4210 | ||
a4d96d62 | 4211 | ret = __record_ref(refs, dir, gen, p); |
31db9f7c AB |
4212 | |
4213 | out: | |
4214 | if (ret) | |
924794c9 | 4215 | fs_path_free(p); |
31db9f7c AB |
4216 | return ret; |
4217 | } | |
4218 | ||
a4d96d62 LB |
4219 | static int __record_new_ref(int num, u64 dir, int index, |
4220 | struct fs_path *name, | |
4221 | void *ctx) | |
4222 | { | |
4223 | struct send_ctx *sctx = ctx; | |
a0357511 | 4224 | return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs); |
a4d96d62 LB |
4225 | } |
4226 | ||
4227 | ||
31db9f7c AB |
4228 | static int __record_deleted_ref(int num, u64 dir, int index, |
4229 | struct fs_path *name, | |
4230 | void *ctx) | |
4231 | { | |
31db9f7c | 4232 | struct send_ctx *sctx = ctx; |
a0357511 NB |
4233 | return record_ref(sctx->parent_root, dir, name, ctx, |
4234 | &sctx->deleted_refs); | |
31db9f7c AB |
4235 | } |
4236 | ||
4237 | static int record_new_ref(struct send_ctx *sctx) | |
4238 | { | |
4239 | int ret; | |
4240 | ||
924794c9 TI |
4241 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
4242 | sctx->cmp_key, 0, __record_new_ref, sctx); | |
31db9f7c AB |
4243 | if (ret < 0) |
4244 | goto out; | |
4245 | ret = 0; | |
4246 | ||
4247 | out: | |
4248 | return ret; | |
4249 | } | |
4250 | ||
4251 | static int record_deleted_ref(struct send_ctx *sctx) | |
4252 | { | |
4253 | int ret; | |
4254 | ||
924794c9 TI |
4255 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
4256 | sctx->cmp_key, 0, __record_deleted_ref, sctx); | |
31db9f7c AB |
4257 | if (ret < 0) |
4258 | goto out; | |
4259 | ret = 0; | |
4260 | ||
4261 | out: | |
4262 | return ret; | |
4263 | } | |
4264 | ||
4265 | struct find_ref_ctx { | |
4266 | u64 dir; | |
ba5e8f2e JB |
4267 | u64 dir_gen; |
4268 | struct btrfs_root *root; | |
31db9f7c AB |
4269 | struct fs_path *name; |
4270 | int found_idx; | |
4271 | }; | |
4272 | ||
4273 | static int __find_iref(int num, u64 dir, int index, | |
4274 | struct fs_path *name, | |
4275 | void *ctx_) | |
4276 | { | |
4277 | struct find_ref_ctx *ctx = ctx_; | |
ba5e8f2e JB |
4278 | u64 dir_gen; |
4279 | int ret; | |
31db9f7c AB |
4280 | |
4281 | if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && | |
4282 | strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { | |
ba5e8f2e JB |
4283 | /* |
4284 | * To avoid doing extra lookups we'll only do this if everything | |
4285 | * else matches. | |
4286 | */ | |
4287 | ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, | |
4288 | NULL, NULL, NULL); | |
4289 | if (ret) | |
4290 | return ret; | |
4291 | if (dir_gen != ctx->dir_gen) | |
4292 | return 0; | |
31db9f7c AB |
4293 | ctx->found_idx = num; |
4294 | return 1; | |
4295 | } | |
4296 | return 0; | |
4297 | } | |
4298 | ||
924794c9 | 4299 | static int find_iref(struct btrfs_root *root, |
31db9f7c AB |
4300 | struct btrfs_path *path, |
4301 | struct btrfs_key *key, | |
ba5e8f2e | 4302 | u64 dir, u64 dir_gen, struct fs_path *name) |
31db9f7c AB |
4303 | { |
4304 | int ret; | |
4305 | struct find_ref_ctx ctx; | |
4306 | ||
4307 | ctx.dir = dir; | |
4308 | ctx.name = name; | |
ba5e8f2e | 4309 | ctx.dir_gen = dir_gen; |
31db9f7c | 4310 | ctx.found_idx = -1; |
ba5e8f2e | 4311 | ctx.root = root; |
31db9f7c | 4312 | |
924794c9 | 4313 | ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); |
31db9f7c AB |
4314 | if (ret < 0) |
4315 | return ret; | |
4316 | ||
4317 | if (ctx.found_idx == -1) | |
4318 | return -ENOENT; | |
4319 | ||
4320 | return ctx.found_idx; | |
4321 | } | |
4322 | ||
4323 | static int __record_changed_new_ref(int num, u64 dir, int index, | |
4324 | struct fs_path *name, | |
4325 | void *ctx) | |
4326 | { | |
ba5e8f2e | 4327 | u64 dir_gen; |
31db9f7c AB |
4328 | int ret; |
4329 | struct send_ctx *sctx = ctx; | |
4330 | ||
ba5e8f2e JB |
4331 | ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, |
4332 | NULL, NULL, NULL); | |
4333 | if (ret) | |
4334 | return ret; | |
4335 | ||
924794c9 | 4336 | ret = find_iref(sctx->parent_root, sctx->right_path, |
ba5e8f2e | 4337 | sctx->cmp_key, dir, dir_gen, name); |
31db9f7c AB |
4338 | if (ret == -ENOENT) |
4339 | ret = __record_new_ref(num, dir, index, name, sctx); | |
4340 | else if (ret > 0) | |
4341 | ret = 0; | |
4342 | ||
4343 | return ret; | |
4344 | } | |
4345 | ||
4346 | static int __record_changed_deleted_ref(int num, u64 dir, int index, | |
4347 | struct fs_path *name, | |
4348 | void *ctx) | |
4349 | { | |
ba5e8f2e | 4350 | u64 dir_gen; |
31db9f7c AB |
4351 | int ret; |
4352 | struct send_ctx *sctx = ctx; | |
4353 | ||
ba5e8f2e JB |
4354 | ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, |
4355 | NULL, NULL, NULL); | |
4356 | if (ret) | |
4357 | return ret; | |
4358 | ||
924794c9 | 4359 | ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, |
ba5e8f2e | 4360 | dir, dir_gen, name); |
31db9f7c AB |
4361 | if (ret == -ENOENT) |
4362 | ret = __record_deleted_ref(num, dir, index, name, sctx); | |
4363 | else if (ret > 0) | |
4364 | ret = 0; | |
4365 | ||
4366 | return ret; | |
4367 | } | |
4368 | ||
4369 | static int record_changed_ref(struct send_ctx *sctx) | |
4370 | { | |
4371 | int ret = 0; | |
4372 | ||
924794c9 | 4373 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
31db9f7c AB |
4374 | sctx->cmp_key, 0, __record_changed_new_ref, sctx); |
4375 | if (ret < 0) | |
4376 | goto out; | |
924794c9 | 4377 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
4378 | sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); |
4379 | if (ret < 0) | |
4380 | goto out; | |
4381 | ret = 0; | |
4382 | ||
4383 | out: | |
4384 | return ret; | |
4385 | } | |
4386 | ||
4387 | /* | |
4388 | * Record and process all refs at once. Needed when an inode changes the | |
4389 | * generation number, which means that it was deleted and recreated. | |
4390 | */ | |
4391 | static int process_all_refs(struct send_ctx *sctx, | |
4392 | enum btrfs_compare_tree_result cmd) | |
4393 | { | |
4394 | int ret; | |
4395 | struct btrfs_root *root; | |
4396 | struct btrfs_path *path; | |
4397 | struct btrfs_key key; | |
4398 | struct btrfs_key found_key; | |
4399 | struct extent_buffer *eb; | |
4400 | int slot; | |
4401 | iterate_inode_ref_t cb; | |
9f03740a | 4402 | int pending_move = 0; |
31db9f7c AB |
4403 | |
4404 | path = alloc_path_for_send(); | |
4405 | if (!path) | |
4406 | return -ENOMEM; | |
4407 | ||
4408 | if (cmd == BTRFS_COMPARE_TREE_NEW) { | |
4409 | root = sctx->send_root; | |
4410 | cb = __record_new_ref; | |
4411 | } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { | |
4412 | root = sctx->parent_root; | |
4413 | cb = __record_deleted_ref; | |
4414 | } else { | |
4d1a63b2 DS |
4415 | btrfs_err(sctx->send_root->fs_info, |
4416 | "Wrong command %d in process_all_refs", cmd); | |
4417 | ret = -EINVAL; | |
4418 | goto out; | |
31db9f7c AB |
4419 | } |
4420 | ||
4421 | key.objectid = sctx->cmp_key->objectid; | |
4422 | key.type = BTRFS_INODE_REF_KEY; | |
4423 | key.offset = 0; | |
dff6d0ad FDBM |
4424 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4425 | if (ret < 0) | |
4426 | goto out; | |
31db9f7c | 4427 | |
dff6d0ad | 4428 | while (1) { |
31db9f7c AB |
4429 | eb = path->nodes[0]; |
4430 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4431 | if (slot >= btrfs_header_nritems(eb)) { |
4432 | ret = btrfs_next_leaf(root, path); | |
4433 | if (ret < 0) | |
4434 | goto out; | |
4435 | else if (ret > 0) | |
4436 | break; | |
4437 | continue; | |
4438 | } | |
4439 | ||
31db9f7c AB |
4440 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
4441 | ||
4442 | if (found_key.objectid != key.objectid || | |
96b5bd77 JS |
4443 | (found_key.type != BTRFS_INODE_REF_KEY && |
4444 | found_key.type != BTRFS_INODE_EXTREF_KEY)) | |
31db9f7c | 4445 | break; |
31db9f7c | 4446 | |
924794c9 | 4447 | ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); |
31db9f7c AB |
4448 | if (ret < 0) |
4449 | goto out; | |
4450 | ||
dff6d0ad | 4451 | path->slots[0]++; |
31db9f7c | 4452 | } |
e938c8ad | 4453 | btrfs_release_path(path); |
31db9f7c | 4454 | |
3dc09ec8 JB |
4455 | /* |
4456 | * We don't actually care about pending_move as we are simply | |
4457 | * re-creating this inode and will be rename'ing it into place once we | |
4458 | * rename the parent directory. | |
4459 | */ | |
9f03740a | 4460 | ret = process_recorded_refs(sctx, &pending_move); |
31db9f7c AB |
4461 | out: |
4462 | btrfs_free_path(path); | |
4463 | return ret; | |
4464 | } | |
4465 | ||
4466 | static int send_set_xattr(struct send_ctx *sctx, | |
4467 | struct fs_path *path, | |
4468 | const char *name, int name_len, | |
4469 | const char *data, int data_len) | |
4470 | { | |
4471 | int ret = 0; | |
4472 | ||
4473 | ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); | |
4474 | if (ret < 0) | |
4475 | goto out; | |
4476 | ||
4477 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4478 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4479 | TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); | |
4480 | ||
4481 | ret = send_cmd(sctx); | |
4482 | ||
4483 | tlv_put_failure: | |
4484 | out: | |
4485 | return ret; | |
4486 | } | |
4487 | ||
4488 | static int send_remove_xattr(struct send_ctx *sctx, | |
4489 | struct fs_path *path, | |
4490 | const char *name, int name_len) | |
4491 | { | |
4492 | int ret = 0; | |
4493 | ||
4494 | ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); | |
4495 | if (ret < 0) | |
4496 | goto out; | |
4497 | ||
4498 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4499 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4500 | ||
4501 | ret = send_cmd(sctx); | |
4502 | ||
4503 | tlv_put_failure: | |
4504 | out: | |
4505 | return ret; | |
4506 | } | |
4507 | ||
4508 | static int __process_new_xattr(int num, struct btrfs_key *di_key, | |
4509 | const char *name, int name_len, | |
4510 | const char *data, int data_len, | |
4511 | u8 type, void *ctx) | |
4512 | { | |
4513 | int ret; | |
4514 | struct send_ctx *sctx = ctx; | |
4515 | struct fs_path *p; | |
2211d5ba | 4516 | struct posix_acl_xattr_header dummy_acl; |
31db9f7c | 4517 | |
924794c9 | 4518 | p = fs_path_alloc(); |
31db9f7c AB |
4519 | if (!p) |
4520 | return -ENOMEM; | |
4521 | ||
4522 | /* | |
01327610 | 4523 | * This hack is needed because empty acls are stored as zero byte |
31db9f7c | 4524 | * data in xattrs. Problem with that is, that receiving these zero byte |
01327610 | 4525 | * acls will fail later. To fix this, we send a dummy acl list that |
31db9f7c AB |
4526 | * only contains the version number and no entries. |
4527 | */ | |
4528 | if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || | |
4529 | !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { | |
4530 | if (data_len == 0) { | |
4531 | dummy_acl.a_version = | |
4532 | cpu_to_le32(POSIX_ACL_XATTR_VERSION); | |
4533 | data = (char *)&dummy_acl; | |
4534 | data_len = sizeof(dummy_acl); | |
4535 | } | |
4536 | } | |
4537 | ||
4538 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4539 | if (ret < 0) | |
4540 | goto out; | |
4541 | ||
4542 | ret = send_set_xattr(sctx, p, name, name_len, data, data_len); | |
4543 | ||
4544 | out: | |
924794c9 | 4545 | fs_path_free(p); |
31db9f7c AB |
4546 | return ret; |
4547 | } | |
4548 | ||
4549 | static int __process_deleted_xattr(int num, struct btrfs_key *di_key, | |
4550 | const char *name, int name_len, | |
4551 | const char *data, int data_len, | |
4552 | u8 type, void *ctx) | |
4553 | { | |
4554 | int ret; | |
4555 | struct send_ctx *sctx = ctx; | |
4556 | struct fs_path *p; | |
4557 | ||
924794c9 | 4558 | p = fs_path_alloc(); |
31db9f7c AB |
4559 | if (!p) |
4560 | return -ENOMEM; | |
4561 | ||
4562 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4563 | if (ret < 0) | |
4564 | goto out; | |
4565 | ||
4566 | ret = send_remove_xattr(sctx, p, name, name_len); | |
4567 | ||
4568 | out: | |
924794c9 | 4569 | fs_path_free(p); |
31db9f7c AB |
4570 | return ret; |
4571 | } | |
4572 | ||
4573 | static int process_new_xattr(struct send_ctx *sctx) | |
4574 | { | |
4575 | int ret = 0; | |
4576 | ||
924794c9 | 4577 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4578 | __process_new_xattr, sctx); |
31db9f7c AB |
4579 | |
4580 | return ret; | |
4581 | } | |
4582 | ||
4583 | static int process_deleted_xattr(struct send_ctx *sctx) | |
4584 | { | |
e2c89907 | 4585 | return iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4586 | __process_deleted_xattr, sctx); |
31db9f7c AB |
4587 | } |
4588 | ||
4589 | struct find_xattr_ctx { | |
4590 | const char *name; | |
4591 | int name_len; | |
4592 | int found_idx; | |
4593 | char *found_data; | |
4594 | int found_data_len; | |
4595 | }; | |
4596 | ||
4597 | static int __find_xattr(int num, struct btrfs_key *di_key, | |
4598 | const char *name, int name_len, | |
4599 | const char *data, int data_len, | |
4600 | u8 type, void *vctx) | |
4601 | { | |
4602 | struct find_xattr_ctx *ctx = vctx; | |
4603 | ||
4604 | if (name_len == ctx->name_len && | |
4605 | strncmp(name, ctx->name, name_len) == 0) { | |
4606 | ctx->found_idx = num; | |
4607 | ctx->found_data_len = data_len; | |
e780b0d1 | 4608 | ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); |
31db9f7c AB |
4609 | if (!ctx->found_data) |
4610 | return -ENOMEM; | |
31db9f7c AB |
4611 | return 1; |
4612 | } | |
4613 | return 0; | |
4614 | } | |
4615 | ||
924794c9 | 4616 | static int find_xattr(struct btrfs_root *root, |
31db9f7c AB |
4617 | struct btrfs_path *path, |
4618 | struct btrfs_key *key, | |
4619 | const char *name, int name_len, | |
4620 | char **data, int *data_len) | |
4621 | { | |
4622 | int ret; | |
4623 | struct find_xattr_ctx ctx; | |
4624 | ||
4625 | ctx.name = name; | |
4626 | ctx.name_len = name_len; | |
4627 | ctx.found_idx = -1; | |
4628 | ctx.found_data = NULL; | |
4629 | ctx.found_data_len = 0; | |
4630 | ||
a0357511 | 4631 | ret = iterate_dir_item(root, path, __find_xattr, &ctx); |
31db9f7c AB |
4632 | if (ret < 0) |
4633 | return ret; | |
4634 | ||
4635 | if (ctx.found_idx == -1) | |
4636 | return -ENOENT; | |
4637 | if (data) { | |
4638 | *data = ctx.found_data; | |
4639 | *data_len = ctx.found_data_len; | |
4640 | } else { | |
4641 | kfree(ctx.found_data); | |
4642 | } | |
4643 | return ctx.found_idx; | |
4644 | } | |
4645 | ||
4646 | ||
4647 | static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, | |
4648 | const char *name, int name_len, | |
4649 | const char *data, int data_len, | |
4650 | u8 type, void *ctx) | |
4651 | { | |
4652 | int ret; | |
4653 | struct send_ctx *sctx = ctx; | |
4654 | char *found_data = NULL; | |
4655 | int found_data_len = 0; | |
31db9f7c | 4656 | |
924794c9 TI |
4657 | ret = find_xattr(sctx->parent_root, sctx->right_path, |
4658 | sctx->cmp_key, name, name_len, &found_data, | |
4659 | &found_data_len); | |
31db9f7c AB |
4660 | if (ret == -ENOENT) { |
4661 | ret = __process_new_xattr(num, di_key, name, name_len, data, | |
4662 | data_len, type, ctx); | |
4663 | } else if (ret >= 0) { | |
4664 | if (data_len != found_data_len || | |
4665 | memcmp(data, found_data, data_len)) { | |
4666 | ret = __process_new_xattr(num, di_key, name, name_len, | |
4667 | data, data_len, type, ctx); | |
4668 | } else { | |
4669 | ret = 0; | |
4670 | } | |
4671 | } | |
4672 | ||
4673 | kfree(found_data); | |
31db9f7c AB |
4674 | return ret; |
4675 | } | |
4676 | ||
4677 | static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, | |
4678 | const char *name, int name_len, | |
4679 | const char *data, int data_len, | |
4680 | u8 type, void *ctx) | |
4681 | { | |
4682 | int ret; | |
4683 | struct send_ctx *sctx = ctx; | |
4684 | ||
924794c9 TI |
4685 | ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, |
4686 | name, name_len, NULL, NULL); | |
31db9f7c AB |
4687 | if (ret == -ENOENT) |
4688 | ret = __process_deleted_xattr(num, di_key, name, name_len, data, | |
4689 | data_len, type, ctx); | |
4690 | else if (ret >= 0) | |
4691 | ret = 0; | |
4692 | ||
4693 | return ret; | |
4694 | } | |
4695 | ||
4696 | static int process_changed_xattr(struct send_ctx *sctx) | |
4697 | { | |
4698 | int ret = 0; | |
4699 | ||
924794c9 | 4700 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4701 | __process_changed_new_xattr, sctx); |
31db9f7c AB |
4702 | if (ret < 0) |
4703 | goto out; | |
924794c9 | 4704 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4705 | __process_changed_deleted_xattr, sctx); |
31db9f7c AB |
4706 | |
4707 | out: | |
4708 | return ret; | |
4709 | } | |
4710 | ||
4711 | static int process_all_new_xattrs(struct send_ctx *sctx) | |
4712 | { | |
4713 | int ret; | |
4714 | struct btrfs_root *root; | |
4715 | struct btrfs_path *path; | |
4716 | struct btrfs_key key; | |
4717 | struct btrfs_key found_key; | |
4718 | struct extent_buffer *eb; | |
4719 | int slot; | |
4720 | ||
4721 | path = alloc_path_for_send(); | |
4722 | if (!path) | |
4723 | return -ENOMEM; | |
4724 | ||
4725 | root = sctx->send_root; | |
4726 | ||
4727 | key.objectid = sctx->cmp_key->objectid; | |
4728 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4729 | key.offset = 0; | |
dff6d0ad FDBM |
4730 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4731 | if (ret < 0) | |
4732 | goto out; | |
31db9f7c | 4733 | |
dff6d0ad | 4734 | while (1) { |
31db9f7c AB |
4735 | eb = path->nodes[0]; |
4736 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4737 | if (slot >= btrfs_header_nritems(eb)) { |
4738 | ret = btrfs_next_leaf(root, path); | |
4739 | if (ret < 0) { | |
4740 | goto out; | |
4741 | } else if (ret > 0) { | |
4742 | ret = 0; | |
4743 | break; | |
4744 | } | |
4745 | continue; | |
4746 | } | |
31db9f7c | 4747 | |
dff6d0ad | 4748 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
31db9f7c AB |
4749 | if (found_key.objectid != key.objectid || |
4750 | found_key.type != key.type) { | |
4751 | ret = 0; | |
4752 | goto out; | |
4753 | } | |
4754 | ||
a0357511 | 4755 | ret = iterate_dir_item(root, path, __process_new_xattr, sctx); |
31db9f7c AB |
4756 | if (ret < 0) |
4757 | goto out; | |
4758 | ||
dff6d0ad | 4759 | path->slots[0]++; |
31db9f7c AB |
4760 | } |
4761 | ||
4762 | out: | |
4763 | btrfs_free_path(path); | |
4764 | return ret; | |
4765 | } | |
4766 | ||
ed259095 JB |
4767 | static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len) |
4768 | { | |
4769 | struct btrfs_root *root = sctx->send_root; | |
4770 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4771 | struct inode *inode; | |
4772 | struct page *page; | |
4773 | char *addr; | |
4774 | struct btrfs_key key; | |
09cbfeaf | 4775 | pgoff_t index = offset >> PAGE_SHIFT; |
ed259095 | 4776 | pgoff_t last_index; |
09cbfeaf | 4777 | unsigned pg_offset = offset & ~PAGE_MASK; |
ed259095 JB |
4778 | ssize_t ret = 0; |
4779 | ||
4780 | key.objectid = sctx->cur_ino; | |
4781 | key.type = BTRFS_INODE_ITEM_KEY; | |
4782 | key.offset = 0; | |
4783 | ||
4784 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
4785 | if (IS_ERR(inode)) | |
4786 | return PTR_ERR(inode); | |
4787 | ||
4788 | if (offset + len > i_size_read(inode)) { | |
4789 | if (offset > i_size_read(inode)) | |
4790 | len = 0; | |
4791 | else | |
4792 | len = offset - i_size_read(inode); | |
4793 | } | |
4794 | if (len == 0) | |
4795 | goto out; | |
4796 | ||
09cbfeaf | 4797 | last_index = (offset + len - 1) >> PAGE_SHIFT; |
2131bcd3 LB |
4798 | |
4799 | /* initial readahead */ | |
4800 | memset(&sctx->ra, 0, sizeof(struct file_ra_state)); | |
4801 | file_ra_state_init(&sctx->ra, inode->i_mapping); | |
2131bcd3 | 4802 | |
ed259095 JB |
4803 | while (index <= last_index) { |
4804 | unsigned cur_len = min_t(unsigned, len, | |
09cbfeaf | 4805 | PAGE_SIZE - pg_offset); |
eef16ba2 KH |
4806 | |
4807 | page = find_lock_page(inode->i_mapping, index); | |
ed259095 | 4808 | if (!page) { |
eef16ba2 KH |
4809 | page_cache_sync_readahead(inode->i_mapping, &sctx->ra, |
4810 | NULL, index, last_index + 1 - index); | |
4811 | ||
4812 | page = find_or_create_page(inode->i_mapping, index, | |
4813 | GFP_KERNEL); | |
4814 | if (!page) { | |
4815 | ret = -ENOMEM; | |
4816 | break; | |
4817 | } | |
4818 | } | |
4819 | ||
4820 | if (PageReadahead(page)) { | |
4821 | page_cache_async_readahead(inode->i_mapping, &sctx->ra, | |
4822 | NULL, page, index, last_index + 1 - index); | |
ed259095 JB |
4823 | } |
4824 | ||
4825 | if (!PageUptodate(page)) { | |
4826 | btrfs_readpage(NULL, page); | |
4827 | lock_page(page); | |
4828 | if (!PageUptodate(page)) { | |
4829 | unlock_page(page); | |
09cbfeaf | 4830 | put_page(page); |
ed259095 JB |
4831 | ret = -EIO; |
4832 | break; | |
4833 | } | |
4834 | } | |
4835 | ||
4836 | addr = kmap(page); | |
4837 | memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len); | |
4838 | kunmap(page); | |
4839 | unlock_page(page); | |
09cbfeaf | 4840 | put_page(page); |
ed259095 JB |
4841 | index++; |
4842 | pg_offset = 0; | |
4843 | len -= cur_len; | |
4844 | ret += cur_len; | |
4845 | } | |
4846 | out: | |
4847 | iput(inode); | |
4848 | return ret; | |
4849 | } | |
4850 | ||
31db9f7c AB |
4851 | /* |
4852 | * Read some bytes from the current inode/file and send a write command to | |
4853 | * user space. | |
4854 | */ | |
4855 | static int send_write(struct send_ctx *sctx, u64 offset, u32 len) | |
4856 | { | |
04ab956e | 4857 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
4858 | int ret = 0; |
4859 | struct fs_path *p; | |
ed259095 | 4860 | ssize_t num_read = 0; |
31db9f7c | 4861 | |
924794c9 | 4862 | p = fs_path_alloc(); |
31db9f7c AB |
4863 | if (!p) |
4864 | return -ENOMEM; | |
4865 | ||
04ab956e | 4866 | btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len); |
31db9f7c | 4867 | |
ed259095 JB |
4868 | num_read = fill_read_buf(sctx, offset, len); |
4869 | if (num_read <= 0) { | |
4870 | if (num_read < 0) | |
4871 | ret = num_read; | |
31db9f7c | 4872 | goto out; |
ed259095 | 4873 | } |
31db9f7c AB |
4874 | |
4875 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
4876 | if (ret < 0) | |
4877 | goto out; | |
4878 | ||
4879 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4880 | if (ret < 0) | |
4881 | goto out; | |
4882 | ||
4883 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4884 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
e938c8ad | 4885 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read); |
31db9f7c AB |
4886 | |
4887 | ret = send_cmd(sctx); | |
4888 | ||
4889 | tlv_put_failure: | |
4890 | out: | |
924794c9 | 4891 | fs_path_free(p); |
31db9f7c AB |
4892 | if (ret < 0) |
4893 | return ret; | |
e938c8ad | 4894 | return num_read; |
31db9f7c AB |
4895 | } |
4896 | ||
4897 | /* | |
4898 | * Send a clone command to user space. | |
4899 | */ | |
4900 | static int send_clone(struct send_ctx *sctx, | |
4901 | u64 offset, u32 len, | |
4902 | struct clone_root *clone_root) | |
4903 | { | |
4904 | int ret = 0; | |
31db9f7c AB |
4905 | struct fs_path *p; |
4906 | u64 gen; | |
4907 | ||
04ab956e JM |
4908 | btrfs_debug(sctx->send_root->fs_info, |
4909 | "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu", | |
4910 | offset, len, clone_root->root->objectid, clone_root->ino, | |
4911 | clone_root->offset); | |
31db9f7c | 4912 | |
924794c9 | 4913 | p = fs_path_alloc(); |
31db9f7c AB |
4914 | if (!p) |
4915 | return -ENOMEM; | |
4916 | ||
4917 | ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); | |
4918 | if (ret < 0) | |
4919 | goto out; | |
4920 | ||
4921 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4922 | if (ret < 0) | |
4923 | goto out; | |
4924 | ||
4925 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4926 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); | |
4927 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4928 | ||
e938c8ad | 4929 | if (clone_root->root == sctx->send_root) { |
31db9f7c | 4930 | ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, |
85a7b33b | 4931 | &gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
4932 | if (ret < 0) |
4933 | goto out; | |
4934 | ret = get_cur_path(sctx, clone_root->ino, gen, p); | |
4935 | } else { | |
924794c9 | 4936 | ret = get_inode_path(clone_root->root, clone_root->ino, p); |
31db9f7c AB |
4937 | } |
4938 | if (ret < 0) | |
4939 | goto out; | |
4940 | ||
37b8d27d JB |
4941 | /* |
4942 | * If the parent we're using has a received_uuid set then use that as | |
4943 | * our clone source as that is what we will look for when doing a | |
4944 | * receive. | |
4945 | * | |
4946 | * This covers the case that we create a snapshot off of a received | |
4947 | * subvolume and then use that as the parent and try to receive on a | |
4948 | * different host. | |
4949 | */ | |
4950 | if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid)) | |
4951 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4952 | clone_root->root->root_item.received_uuid); | |
4953 | else | |
4954 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4955 | clone_root->root->root_item.uuid); | |
31db9f7c | 4956 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 4957 | le64_to_cpu(clone_root->root->root_item.ctransid)); |
31db9f7c AB |
4958 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); |
4959 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, | |
4960 | clone_root->offset); | |
4961 | ||
4962 | ret = send_cmd(sctx); | |
4963 | ||
4964 | tlv_put_failure: | |
4965 | out: | |
924794c9 | 4966 | fs_path_free(p); |
31db9f7c AB |
4967 | return ret; |
4968 | } | |
4969 | ||
cb95e7bf MF |
4970 | /* |
4971 | * Send an update extent command to user space. | |
4972 | */ | |
4973 | static int send_update_extent(struct send_ctx *sctx, | |
4974 | u64 offset, u32 len) | |
4975 | { | |
4976 | int ret = 0; | |
4977 | struct fs_path *p; | |
4978 | ||
924794c9 | 4979 | p = fs_path_alloc(); |
cb95e7bf MF |
4980 | if (!p) |
4981 | return -ENOMEM; | |
4982 | ||
4983 | ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); | |
4984 | if (ret < 0) | |
4985 | goto out; | |
4986 | ||
4987 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4988 | if (ret < 0) | |
4989 | goto out; | |
4990 | ||
4991 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4992 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4993 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); | |
4994 | ||
4995 | ret = send_cmd(sctx); | |
4996 | ||
4997 | tlv_put_failure: | |
4998 | out: | |
924794c9 | 4999 | fs_path_free(p); |
cb95e7bf MF |
5000 | return ret; |
5001 | } | |
5002 | ||
16e7549f JB |
5003 | static int send_hole(struct send_ctx *sctx, u64 end) |
5004 | { | |
5005 | struct fs_path *p = NULL; | |
5006 | u64 offset = sctx->cur_inode_last_extent; | |
5007 | u64 len; | |
5008 | int ret = 0; | |
5009 | ||
22d3151c FM |
5010 | /* |
5011 | * A hole that starts at EOF or beyond it. Since we do not yet support | |
5012 | * fallocate (for extent preallocation and hole punching), sending a | |
5013 | * write of zeroes starting at EOF or beyond would later require issuing | |
5014 | * a truncate operation which would undo the write and achieve nothing. | |
5015 | */ | |
5016 | if (offset >= sctx->cur_inode_size) | |
5017 | return 0; | |
5018 | ||
d4dfc0f4 FM |
5019 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) |
5020 | return send_update_extent(sctx, offset, end - offset); | |
5021 | ||
16e7549f JB |
5022 | p = fs_path_alloc(); |
5023 | if (!p) | |
5024 | return -ENOMEM; | |
c715e155 FM |
5025 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); |
5026 | if (ret < 0) | |
5027 | goto tlv_put_failure; | |
16e7549f JB |
5028 | memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE); |
5029 | while (offset < end) { | |
5030 | len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE); | |
5031 | ||
5032 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
16e7549f JB |
5033 | if (ret < 0) |
5034 | break; | |
5035 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5036 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5037 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len); | |
5038 | ret = send_cmd(sctx); | |
5039 | if (ret < 0) | |
5040 | break; | |
5041 | offset += len; | |
5042 | } | |
ffa7c429 | 5043 | sctx->cur_inode_next_write_offset = offset; |
16e7549f JB |
5044 | tlv_put_failure: |
5045 | fs_path_free(p); | |
5046 | return ret; | |
5047 | } | |
5048 | ||
d906d49f FM |
5049 | static int send_extent_data(struct send_ctx *sctx, |
5050 | const u64 offset, | |
5051 | const u64 len) | |
5052 | { | |
5053 | u64 sent = 0; | |
5054 | ||
5055 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) | |
5056 | return send_update_extent(sctx, offset, len); | |
5057 | ||
5058 | while (sent < len) { | |
5059 | u64 size = len - sent; | |
5060 | int ret; | |
5061 | ||
5062 | if (size > BTRFS_SEND_READ_SIZE) | |
5063 | size = BTRFS_SEND_READ_SIZE; | |
5064 | ret = send_write(sctx, offset + sent, size); | |
5065 | if (ret < 0) | |
5066 | return ret; | |
5067 | if (!ret) | |
5068 | break; | |
5069 | sent += ret; | |
5070 | } | |
5071 | return 0; | |
5072 | } | |
5073 | ||
5074 | static int clone_range(struct send_ctx *sctx, | |
5075 | struct clone_root *clone_root, | |
5076 | const u64 disk_byte, | |
5077 | u64 data_offset, | |
5078 | u64 offset, | |
5079 | u64 len) | |
5080 | { | |
5081 | struct btrfs_path *path; | |
5082 | struct btrfs_key key; | |
5083 | int ret; | |
5084 | ||
72610b1b FM |
5085 | /* |
5086 | * Prevent cloning from a zero offset with a length matching the sector | |
5087 | * size because in some scenarios this will make the receiver fail. | |
5088 | * | |
5089 | * For example, if in the source filesystem the extent at offset 0 | |
5090 | * has a length of sectorsize and it was written using direct IO, then | |
5091 | * it can never be an inline extent (even if compression is enabled). | |
5092 | * Then this extent can be cloned in the original filesystem to a non | |
5093 | * zero file offset, but it may not be possible to clone in the | |
5094 | * destination filesystem because it can be inlined due to compression | |
5095 | * on the destination filesystem (as the receiver's write operations are | |
5096 | * always done using buffered IO). The same happens when the original | |
5097 | * filesystem does not have compression enabled but the destination | |
5098 | * filesystem has. | |
5099 | */ | |
5100 | if (clone_root->offset == 0 && | |
5101 | len == sctx->send_root->fs_info->sectorsize) | |
5102 | return send_extent_data(sctx, offset, len); | |
5103 | ||
d906d49f FM |
5104 | path = alloc_path_for_send(); |
5105 | if (!path) | |
5106 | return -ENOMEM; | |
5107 | ||
5108 | /* | |
5109 | * We can't send a clone operation for the entire range if we find | |
5110 | * extent items in the respective range in the source file that | |
5111 | * refer to different extents or if we find holes. | |
5112 | * So check for that and do a mix of clone and regular write/copy | |
5113 | * operations if needed. | |
5114 | * | |
5115 | * Example: | |
5116 | * | |
5117 | * mkfs.btrfs -f /dev/sda | |
5118 | * mount /dev/sda /mnt | |
5119 | * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo | |
5120 | * cp --reflink=always /mnt/foo /mnt/bar | |
5121 | * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo | |
5122 | * btrfs subvolume snapshot -r /mnt /mnt/snap | |
5123 | * | |
5124 | * If when we send the snapshot and we are processing file bar (which | |
5125 | * has a higher inode number than foo) we blindly send a clone operation | |
5126 | * for the [0, 100K[ range from foo to bar, the receiver ends up getting | |
5127 | * a file bar that matches the content of file foo - iow, doesn't match | |
5128 | * the content from bar in the original filesystem. | |
5129 | */ | |
5130 | key.objectid = clone_root->ino; | |
5131 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5132 | key.offset = clone_root->offset; | |
5133 | ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); | |
5134 | if (ret < 0) | |
5135 | goto out; | |
5136 | if (ret > 0 && path->slots[0] > 0) { | |
5137 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
5138 | if (key.objectid == clone_root->ino && | |
5139 | key.type == BTRFS_EXTENT_DATA_KEY) | |
5140 | path->slots[0]--; | |
5141 | } | |
5142 | ||
5143 | while (true) { | |
5144 | struct extent_buffer *leaf = path->nodes[0]; | |
5145 | int slot = path->slots[0]; | |
5146 | struct btrfs_file_extent_item *ei; | |
5147 | u8 type; | |
5148 | u64 ext_len; | |
5149 | u64 clone_len; | |
5150 | ||
5151 | if (slot >= btrfs_header_nritems(leaf)) { | |
5152 | ret = btrfs_next_leaf(clone_root->root, path); | |
5153 | if (ret < 0) | |
5154 | goto out; | |
5155 | else if (ret > 0) | |
5156 | break; | |
5157 | continue; | |
5158 | } | |
5159 | ||
5160 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5161 | ||
5162 | /* | |
5163 | * We might have an implicit trailing hole (NO_HOLES feature | |
5164 | * enabled). We deal with it after leaving this loop. | |
5165 | */ | |
5166 | if (key.objectid != clone_root->ino || | |
5167 | key.type != BTRFS_EXTENT_DATA_KEY) | |
5168 | break; | |
5169 | ||
5170 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5171 | type = btrfs_file_extent_type(leaf, ei); | |
5172 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 5173 | ext_len = btrfs_file_extent_ram_bytes(leaf, ei); |
09cbfeaf | 5174 | ext_len = PAGE_ALIGN(ext_len); |
d906d49f FM |
5175 | } else { |
5176 | ext_len = btrfs_file_extent_num_bytes(leaf, ei); | |
5177 | } | |
5178 | ||
5179 | if (key.offset + ext_len <= clone_root->offset) | |
5180 | goto next; | |
5181 | ||
5182 | if (key.offset > clone_root->offset) { | |
5183 | /* Implicit hole, NO_HOLES feature enabled. */ | |
5184 | u64 hole_len = key.offset - clone_root->offset; | |
5185 | ||
5186 | if (hole_len > len) | |
5187 | hole_len = len; | |
5188 | ret = send_extent_data(sctx, offset, hole_len); | |
5189 | if (ret < 0) | |
5190 | goto out; | |
5191 | ||
5192 | len -= hole_len; | |
5193 | if (len == 0) | |
5194 | break; | |
5195 | offset += hole_len; | |
5196 | clone_root->offset += hole_len; | |
5197 | data_offset += hole_len; | |
5198 | } | |
5199 | ||
5200 | if (key.offset >= clone_root->offset + len) | |
5201 | break; | |
5202 | ||
5203 | clone_len = min_t(u64, ext_len, len); | |
5204 | ||
5205 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && | |
5206 | btrfs_file_extent_offset(leaf, ei) == data_offset) | |
5207 | ret = send_clone(sctx, offset, clone_len, clone_root); | |
5208 | else | |
5209 | ret = send_extent_data(sctx, offset, clone_len); | |
5210 | ||
5211 | if (ret < 0) | |
5212 | goto out; | |
5213 | ||
5214 | len -= clone_len; | |
5215 | if (len == 0) | |
5216 | break; | |
5217 | offset += clone_len; | |
5218 | clone_root->offset += clone_len; | |
5219 | data_offset += clone_len; | |
5220 | next: | |
5221 | path->slots[0]++; | |
5222 | } | |
5223 | ||
5224 | if (len > 0) | |
5225 | ret = send_extent_data(sctx, offset, len); | |
5226 | else | |
5227 | ret = 0; | |
5228 | out: | |
5229 | btrfs_free_path(path); | |
5230 | return ret; | |
5231 | } | |
5232 | ||
31db9f7c AB |
5233 | static int send_write_or_clone(struct send_ctx *sctx, |
5234 | struct btrfs_path *path, | |
5235 | struct btrfs_key *key, | |
5236 | struct clone_root *clone_root) | |
5237 | { | |
5238 | int ret = 0; | |
5239 | struct btrfs_file_extent_item *ei; | |
5240 | u64 offset = key->offset; | |
31db9f7c | 5241 | u64 len; |
31db9f7c | 5242 | u8 type; |
28e5dd8f | 5243 | u64 bs = sctx->send_root->fs_info->sb->s_blocksize; |
31db9f7c AB |
5244 | |
5245 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5246 | struct btrfs_file_extent_item); | |
5247 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
74dd17fb | 5248 | if (type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 | 5249 | len = btrfs_file_extent_ram_bytes(path->nodes[0], ei); |
74dd17fb CM |
5250 | /* |
5251 | * it is possible the inline item won't cover the whole page, | |
5252 | * but there may be items after this page. Make | |
5253 | * sure to send the whole thing | |
5254 | */ | |
09cbfeaf | 5255 | len = PAGE_ALIGN(len); |
74dd17fb | 5256 | } else { |
31db9f7c | 5257 | len = btrfs_file_extent_num_bytes(path->nodes[0], ei); |
74dd17fb | 5258 | } |
31db9f7c | 5259 | |
a6aa10c7 FM |
5260 | if (offset >= sctx->cur_inode_size) { |
5261 | ret = 0; | |
5262 | goto out; | |
5263 | } | |
31db9f7c AB |
5264 | if (offset + len > sctx->cur_inode_size) |
5265 | len = sctx->cur_inode_size - offset; | |
5266 | if (len == 0) { | |
5267 | ret = 0; | |
5268 | goto out; | |
5269 | } | |
5270 | ||
28e5dd8f | 5271 | if (clone_root && IS_ALIGNED(offset + len, bs)) { |
d906d49f FM |
5272 | u64 disk_byte; |
5273 | u64 data_offset; | |
5274 | ||
5275 | disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); | |
5276 | data_offset = btrfs_file_extent_offset(path->nodes[0], ei); | |
5277 | ret = clone_range(sctx, clone_root, disk_byte, data_offset, | |
5278 | offset, len); | |
cb95e7bf | 5279 | } else { |
d906d49f | 5280 | ret = send_extent_data(sctx, offset, len); |
31db9f7c | 5281 | } |
ffa7c429 | 5282 | sctx->cur_inode_next_write_offset = offset + len; |
31db9f7c AB |
5283 | out: |
5284 | return ret; | |
5285 | } | |
5286 | ||
5287 | static int is_extent_unchanged(struct send_ctx *sctx, | |
5288 | struct btrfs_path *left_path, | |
5289 | struct btrfs_key *ekey) | |
5290 | { | |
5291 | int ret = 0; | |
5292 | struct btrfs_key key; | |
5293 | struct btrfs_path *path = NULL; | |
5294 | struct extent_buffer *eb; | |
5295 | int slot; | |
5296 | struct btrfs_key found_key; | |
5297 | struct btrfs_file_extent_item *ei; | |
5298 | u64 left_disknr; | |
5299 | u64 right_disknr; | |
5300 | u64 left_offset; | |
5301 | u64 right_offset; | |
5302 | u64 left_offset_fixed; | |
5303 | u64 left_len; | |
5304 | u64 right_len; | |
74dd17fb CM |
5305 | u64 left_gen; |
5306 | u64 right_gen; | |
31db9f7c AB |
5307 | u8 left_type; |
5308 | u8 right_type; | |
5309 | ||
5310 | path = alloc_path_for_send(); | |
5311 | if (!path) | |
5312 | return -ENOMEM; | |
5313 | ||
5314 | eb = left_path->nodes[0]; | |
5315 | slot = left_path->slots[0]; | |
31db9f7c AB |
5316 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
5317 | left_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
5318 | |
5319 | if (left_type != BTRFS_FILE_EXTENT_REG) { | |
5320 | ret = 0; | |
5321 | goto out; | |
5322 | } | |
74dd17fb CM |
5323 | left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5324 | left_len = btrfs_file_extent_num_bytes(eb, ei); | |
5325 | left_offset = btrfs_file_extent_offset(eb, ei); | |
5326 | left_gen = btrfs_file_extent_generation(eb, ei); | |
31db9f7c AB |
5327 | |
5328 | /* | |
5329 | * Following comments will refer to these graphics. L is the left | |
5330 | * extents which we are checking at the moment. 1-8 are the right | |
5331 | * extents that we iterate. | |
5332 | * | |
5333 | * |-----L-----| | |
5334 | * |-1-|-2a-|-3-|-4-|-5-|-6-| | |
5335 | * | |
5336 | * |-----L-----| | |
5337 | * |--1--|-2b-|...(same as above) | |
5338 | * | |
5339 | * Alternative situation. Happens on files where extents got split. | |
5340 | * |-----L-----| | |
5341 | * |-----------7-----------|-6-| | |
5342 | * | |
5343 | * Alternative situation. Happens on files which got larger. | |
5344 | * |-----L-----| | |
5345 | * |-8-| | |
5346 | * Nothing follows after 8. | |
5347 | */ | |
5348 | ||
5349 | key.objectid = ekey->objectid; | |
5350 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5351 | key.offset = ekey->offset; | |
5352 | ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); | |
5353 | if (ret < 0) | |
5354 | goto out; | |
5355 | if (ret) { | |
5356 | ret = 0; | |
5357 | goto out; | |
5358 | } | |
5359 | ||
5360 | /* | |
5361 | * Handle special case where the right side has no extents at all. | |
5362 | */ | |
5363 | eb = path->nodes[0]; | |
5364 | slot = path->slots[0]; | |
5365 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5366 | if (found_key.objectid != key.objectid || | |
5367 | found_key.type != key.type) { | |
57cfd462 JB |
5368 | /* If we're a hole then just pretend nothing changed */ |
5369 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5370 | goto out; |
5371 | } | |
5372 | ||
5373 | /* | |
5374 | * We're now on 2a, 2b or 7. | |
5375 | */ | |
5376 | key = found_key; | |
5377 | while (key.offset < ekey->offset + left_len) { | |
5378 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
5379 | right_type = btrfs_file_extent_type(eb, ei); | |
e1cbfd7b FM |
5380 | if (right_type != BTRFS_FILE_EXTENT_REG && |
5381 | right_type != BTRFS_FILE_EXTENT_INLINE) { | |
31db9f7c AB |
5382 | ret = 0; |
5383 | goto out; | |
5384 | } | |
5385 | ||
e1cbfd7b | 5386 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 | 5387 | right_len = btrfs_file_extent_ram_bytes(eb, ei); |
e1cbfd7b FM |
5388 | right_len = PAGE_ALIGN(right_len); |
5389 | } else { | |
5390 | right_len = btrfs_file_extent_num_bytes(eb, ei); | |
5391 | } | |
007d31f7 | 5392 | |
31db9f7c AB |
5393 | /* |
5394 | * Are we at extent 8? If yes, we know the extent is changed. | |
5395 | * This may only happen on the first iteration. | |
5396 | */ | |
d8347fa4 | 5397 | if (found_key.offset + right_len <= ekey->offset) { |
57cfd462 JB |
5398 | /* If we're a hole just pretend nothing changed */ |
5399 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5400 | goto out; |
5401 | } | |
5402 | ||
e1cbfd7b FM |
5403 | /* |
5404 | * We just wanted to see if when we have an inline extent, what | |
5405 | * follows it is a regular extent (wanted to check the above | |
5406 | * condition for inline extents too). This should normally not | |
5407 | * happen but it's possible for example when we have an inline | |
5408 | * compressed extent representing data with a size matching | |
5409 | * the page size (currently the same as sector size). | |
5410 | */ | |
5411 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { | |
5412 | ret = 0; | |
5413 | goto out; | |
5414 | } | |
5415 | ||
24e52b11 FM |
5416 | right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5417 | right_offset = btrfs_file_extent_offset(eb, ei); | |
5418 | right_gen = btrfs_file_extent_generation(eb, ei); | |
5419 | ||
31db9f7c AB |
5420 | left_offset_fixed = left_offset; |
5421 | if (key.offset < ekey->offset) { | |
5422 | /* Fix the right offset for 2a and 7. */ | |
5423 | right_offset += ekey->offset - key.offset; | |
5424 | } else { | |
5425 | /* Fix the left offset for all behind 2a and 2b */ | |
5426 | left_offset_fixed += key.offset - ekey->offset; | |
5427 | } | |
5428 | ||
5429 | /* | |
5430 | * Check if we have the same extent. | |
5431 | */ | |
3954096d | 5432 | if (left_disknr != right_disknr || |
74dd17fb CM |
5433 | left_offset_fixed != right_offset || |
5434 | left_gen != right_gen) { | |
31db9f7c AB |
5435 | ret = 0; |
5436 | goto out; | |
5437 | } | |
5438 | ||
5439 | /* | |
5440 | * Go to the next extent. | |
5441 | */ | |
5442 | ret = btrfs_next_item(sctx->parent_root, path); | |
5443 | if (ret < 0) | |
5444 | goto out; | |
5445 | if (!ret) { | |
5446 | eb = path->nodes[0]; | |
5447 | slot = path->slots[0]; | |
5448 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5449 | } | |
5450 | if (ret || found_key.objectid != key.objectid || | |
5451 | found_key.type != key.type) { | |
5452 | key.offset += right_len; | |
5453 | break; | |
adaa4b8e JS |
5454 | } |
5455 | if (found_key.offset != key.offset + right_len) { | |
5456 | ret = 0; | |
5457 | goto out; | |
31db9f7c AB |
5458 | } |
5459 | key = found_key; | |
5460 | } | |
5461 | ||
5462 | /* | |
5463 | * We're now behind the left extent (treat as unchanged) or at the end | |
5464 | * of the right side (treat as changed). | |
5465 | */ | |
5466 | if (key.offset >= ekey->offset + left_len) | |
5467 | ret = 1; | |
5468 | else | |
5469 | ret = 0; | |
5470 | ||
5471 | ||
5472 | out: | |
5473 | btrfs_free_path(path); | |
5474 | return ret; | |
5475 | } | |
5476 | ||
16e7549f JB |
5477 | static int get_last_extent(struct send_ctx *sctx, u64 offset) |
5478 | { | |
5479 | struct btrfs_path *path; | |
5480 | struct btrfs_root *root = sctx->send_root; | |
5481 | struct btrfs_file_extent_item *fi; | |
5482 | struct btrfs_key key; | |
5483 | u64 extent_end; | |
5484 | u8 type; | |
5485 | int ret; | |
5486 | ||
5487 | path = alloc_path_for_send(); | |
5488 | if (!path) | |
5489 | return -ENOMEM; | |
5490 | ||
5491 | sctx->cur_inode_last_extent = 0; | |
5492 | ||
5493 | key.objectid = sctx->cur_ino; | |
5494 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5495 | key.offset = offset; | |
5496 | ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); | |
5497 | if (ret < 0) | |
5498 | goto out; | |
5499 | ret = 0; | |
5500 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
5501 | if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
5502 | goto out; | |
5503 | ||
5504 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5505 | struct btrfs_file_extent_item); | |
5506 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5507 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 5508 | u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi); |
16e7549f | 5509 | extent_end = ALIGN(key.offset + size, |
da17066c | 5510 | sctx->send_root->fs_info->sectorsize); |
16e7549f JB |
5511 | } else { |
5512 | extent_end = key.offset + | |
5513 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5514 | } | |
5515 | sctx->cur_inode_last_extent = extent_end; | |
5516 | out: | |
5517 | btrfs_free_path(path); | |
5518 | return ret; | |
5519 | } | |
5520 | ||
82bfb2e7 FM |
5521 | static int range_is_hole_in_parent(struct send_ctx *sctx, |
5522 | const u64 start, | |
5523 | const u64 end) | |
5524 | { | |
5525 | struct btrfs_path *path; | |
5526 | struct btrfs_key key; | |
5527 | struct btrfs_root *root = sctx->parent_root; | |
5528 | u64 search_start = start; | |
5529 | int ret; | |
5530 | ||
5531 | path = alloc_path_for_send(); | |
5532 | if (!path) | |
5533 | return -ENOMEM; | |
5534 | ||
5535 | key.objectid = sctx->cur_ino; | |
5536 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5537 | key.offset = search_start; | |
5538 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5539 | if (ret < 0) | |
5540 | goto out; | |
5541 | if (ret > 0 && path->slots[0] > 0) | |
5542 | path->slots[0]--; | |
5543 | ||
5544 | while (search_start < end) { | |
5545 | struct extent_buffer *leaf = path->nodes[0]; | |
5546 | int slot = path->slots[0]; | |
5547 | struct btrfs_file_extent_item *fi; | |
5548 | u64 extent_end; | |
5549 | ||
5550 | if (slot >= btrfs_header_nritems(leaf)) { | |
5551 | ret = btrfs_next_leaf(root, path); | |
5552 | if (ret < 0) | |
5553 | goto out; | |
5554 | else if (ret > 0) | |
5555 | break; | |
5556 | continue; | |
5557 | } | |
5558 | ||
5559 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5560 | if (key.objectid < sctx->cur_ino || | |
5561 | key.type < BTRFS_EXTENT_DATA_KEY) | |
5562 | goto next; | |
5563 | if (key.objectid > sctx->cur_ino || | |
5564 | key.type > BTRFS_EXTENT_DATA_KEY || | |
5565 | key.offset >= end) | |
5566 | break; | |
5567 | ||
5568 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5569 | if (btrfs_file_extent_type(leaf, fi) == | |
5570 | BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 5571 | u64 size = btrfs_file_extent_ram_bytes(leaf, fi); |
82bfb2e7 FM |
5572 | |
5573 | extent_end = ALIGN(key.offset + size, | |
5574 | root->fs_info->sectorsize); | |
5575 | } else { | |
5576 | extent_end = key.offset + | |
5577 | btrfs_file_extent_num_bytes(leaf, fi); | |
5578 | } | |
5579 | if (extent_end <= start) | |
5580 | goto next; | |
5581 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) { | |
5582 | search_start = extent_end; | |
5583 | goto next; | |
5584 | } | |
5585 | ret = 0; | |
5586 | goto out; | |
5587 | next: | |
5588 | path->slots[0]++; | |
5589 | } | |
5590 | ret = 1; | |
5591 | out: | |
5592 | btrfs_free_path(path); | |
5593 | return ret; | |
5594 | } | |
5595 | ||
16e7549f JB |
5596 | static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, |
5597 | struct btrfs_key *key) | |
5598 | { | |
5599 | struct btrfs_file_extent_item *fi; | |
5600 | u64 extent_end; | |
5601 | u8 type; | |
5602 | int ret = 0; | |
5603 | ||
5604 | if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) | |
5605 | return 0; | |
5606 | ||
5607 | if (sctx->cur_inode_last_extent == (u64)-1) { | |
5608 | ret = get_last_extent(sctx, key->offset - 1); | |
5609 | if (ret) | |
5610 | return ret; | |
5611 | } | |
5612 | ||
5613 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5614 | struct btrfs_file_extent_item); | |
5615 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5616 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 5617 | u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi); |
16e7549f | 5618 | extent_end = ALIGN(key->offset + size, |
da17066c | 5619 | sctx->send_root->fs_info->sectorsize); |
16e7549f JB |
5620 | } else { |
5621 | extent_end = key->offset + | |
5622 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5623 | } | |
bf54f412 FDBM |
5624 | |
5625 | if (path->slots[0] == 0 && | |
5626 | sctx->cur_inode_last_extent < key->offset) { | |
5627 | /* | |
5628 | * We might have skipped entire leafs that contained only | |
5629 | * file extent items for our current inode. These leafs have | |
5630 | * a generation number smaller (older) than the one in the | |
5631 | * current leaf and the leaf our last extent came from, and | |
5632 | * are located between these 2 leafs. | |
5633 | */ | |
5634 | ret = get_last_extent(sctx, key->offset - 1); | |
5635 | if (ret) | |
5636 | return ret; | |
5637 | } | |
5638 | ||
82bfb2e7 FM |
5639 | if (sctx->cur_inode_last_extent < key->offset) { |
5640 | ret = range_is_hole_in_parent(sctx, | |
5641 | sctx->cur_inode_last_extent, | |
5642 | key->offset); | |
5643 | if (ret < 0) | |
5644 | return ret; | |
5645 | else if (ret == 0) | |
5646 | ret = send_hole(sctx, key->offset); | |
5647 | else | |
5648 | ret = 0; | |
5649 | } | |
16e7549f JB |
5650 | sctx->cur_inode_last_extent = extent_end; |
5651 | return ret; | |
5652 | } | |
5653 | ||
31db9f7c AB |
5654 | static int process_extent(struct send_ctx *sctx, |
5655 | struct btrfs_path *path, | |
5656 | struct btrfs_key *key) | |
5657 | { | |
31db9f7c | 5658 | struct clone_root *found_clone = NULL; |
57cfd462 | 5659 | int ret = 0; |
31db9f7c AB |
5660 | |
5661 | if (S_ISLNK(sctx->cur_inode_mode)) | |
5662 | return 0; | |
5663 | ||
5664 | if (sctx->parent_root && !sctx->cur_inode_new) { | |
5665 | ret = is_extent_unchanged(sctx, path, key); | |
5666 | if (ret < 0) | |
5667 | goto out; | |
5668 | if (ret) { | |
5669 | ret = 0; | |
16e7549f | 5670 | goto out_hole; |
31db9f7c | 5671 | } |
57cfd462 JB |
5672 | } else { |
5673 | struct btrfs_file_extent_item *ei; | |
5674 | u8 type; | |
5675 | ||
5676 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5677 | struct btrfs_file_extent_item); | |
5678 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
5679 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
5680 | type == BTRFS_FILE_EXTENT_REG) { | |
5681 | /* | |
5682 | * The send spec does not have a prealloc command yet, | |
5683 | * so just leave a hole for prealloc'ed extents until | |
5684 | * we have enough commands queued up to justify rev'ing | |
5685 | * the send spec. | |
5686 | */ | |
5687 | if (type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5688 | ret = 0; | |
5689 | goto out; | |
5690 | } | |
5691 | ||
5692 | /* Have a hole, just skip it. */ | |
5693 | if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { | |
5694 | ret = 0; | |
5695 | goto out; | |
5696 | } | |
5697 | } | |
31db9f7c AB |
5698 | } |
5699 | ||
5700 | ret = find_extent_clone(sctx, path, key->objectid, key->offset, | |
5701 | sctx->cur_inode_size, &found_clone); | |
5702 | if (ret != -ENOENT && ret < 0) | |
5703 | goto out; | |
5704 | ||
5705 | ret = send_write_or_clone(sctx, path, key, found_clone); | |
16e7549f JB |
5706 | if (ret) |
5707 | goto out; | |
5708 | out_hole: | |
5709 | ret = maybe_send_hole(sctx, path, key); | |
31db9f7c AB |
5710 | out: |
5711 | return ret; | |
5712 | } | |
5713 | ||
5714 | static int process_all_extents(struct send_ctx *sctx) | |
5715 | { | |
5716 | int ret; | |
5717 | struct btrfs_root *root; | |
5718 | struct btrfs_path *path; | |
5719 | struct btrfs_key key; | |
5720 | struct btrfs_key found_key; | |
5721 | struct extent_buffer *eb; | |
5722 | int slot; | |
5723 | ||
5724 | root = sctx->send_root; | |
5725 | path = alloc_path_for_send(); | |
5726 | if (!path) | |
5727 | return -ENOMEM; | |
5728 | ||
5729 | key.objectid = sctx->cmp_key->objectid; | |
5730 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5731 | key.offset = 0; | |
7fdd29d0 FDBM |
5732 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5733 | if (ret < 0) | |
5734 | goto out; | |
31db9f7c | 5735 | |
7fdd29d0 | 5736 | while (1) { |
31db9f7c AB |
5737 | eb = path->nodes[0]; |
5738 | slot = path->slots[0]; | |
7fdd29d0 FDBM |
5739 | |
5740 | if (slot >= btrfs_header_nritems(eb)) { | |
5741 | ret = btrfs_next_leaf(root, path); | |
5742 | if (ret < 0) { | |
5743 | goto out; | |
5744 | } else if (ret > 0) { | |
5745 | ret = 0; | |
5746 | break; | |
5747 | } | |
5748 | continue; | |
5749 | } | |
5750 | ||
31db9f7c AB |
5751 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
5752 | ||
5753 | if (found_key.objectid != key.objectid || | |
5754 | found_key.type != key.type) { | |
5755 | ret = 0; | |
5756 | goto out; | |
5757 | } | |
5758 | ||
5759 | ret = process_extent(sctx, path, &found_key); | |
5760 | if (ret < 0) | |
5761 | goto out; | |
5762 | ||
7fdd29d0 | 5763 | path->slots[0]++; |
31db9f7c AB |
5764 | } |
5765 | ||
5766 | out: | |
5767 | btrfs_free_path(path); | |
5768 | return ret; | |
5769 | } | |
5770 | ||
9f03740a FDBM |
5771 | static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, |
5772 | int *pending_move, | |
5773 | int *refs_processed) | |
31db9f7c AB |
5774 | { |
5775 | int ret = 0; | |
5776 | ||
5777 | if (sctx->cur_ino == 0) | |
5778 | goto out; | |
5779 | if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && | |
96b5bd77 | 5780 | sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) |
31db9f7c AB |
5781 | goto out; |
5782 | if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) | |
5783 | goto out; | |
5784 | ||
9f03740a | 5785 | ret = process_recorded_refs(sctx, pending_move); |
e479d9bb AB |
5786 | if (ret < 0) |
5787 | goto out; | |
5788 | ||
9f03740a | 5789 | *refs_processed = 1; |
31db9f7c AB |
5790 | out: |
5791 | return ret; | |
5792 | } | |
5793 | ||
5794 | static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) | |
5795 | { | |
5796 | int ret = 0; | |
5797 | u64 left_mode; | |
5798 | u64 left_uid; | |
5799 | u64 left_gid; | |
5800 | u64 right_mode; | |
5801 | u64 right_uid; | |
5802 | u64 right_gid; | |
5803 | int need_chmod = 0; | |
5804 | int need_chown = 0; | |
ffa7c429 | 5805 | int need_truncate = 1; |
9f03740a FDBM |
5806 | int pending_move = 0; |
5807 | int refs_processed = 0; | |
31db9f7c | 5808 | |
46b2f459 FM |
5809 | if (sctx->ignore_cur_inode) |
5810 | return 0; | |
5811 | ||
9f03740a FDBM |
5812 | ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, |
5813 | &refs_processed); | |
31db9f7c AB |
5814 | if (ret < 0) |
5815 | goto out; | |
5816 | ||
9f03740a FDBM |
5817 | /* |
5818 | * We have processed the refs and thus need to advance send_progress. | |
5819 | * Now, calls to get_cur_xxx will take the updated refs of the current | |
5820 | * inode into account. | |
5821 | * | |
5822 | * On the other hand, if our current inode is a directory and couldn't | |
5823 | * be moved/renamed because its parent was renamed/moved too and it has | |
5824 | * a higher inode number, we can only move/rename our current inode | |
5825 | * after we moved/renamed its parent. Therefore in this case operate on | |
5826 | * the old path (pre move/rename) of our current inode, and the | |
5827 | * move/rename will be performed later. | |
5828 | */ | |
5829 | if (refs_processed && !pending_move) | |
5830 | sctx->send_progress = sctx->cur_ino + 1; | |
5831 | ||
31db9f7c AB |
5832 | if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) |
5833 | goto out; | |
5834 | if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) | |
5835 | goto out; | |
5836 | ||
5837 | ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, | |
85a7b33b | 5838 | &left_mode, &left_uid, &left_gid, NULL); |
31db9f7c AB |
5839 | if (ret < 0) |
5840 | goto out; | |
5841 | ||
e2d044fe AL |
5842 | if (!sctx->parent_root || sctx->cur_inode_new) { |
5843 | need_chown = 1; | |
5844 | if (!S_ISLNK(sctx->cur_inode_mode)) | |
31db9f7c | 5845 | need_chmod = 1; |
ffa7c429 FM |
5846 | if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size) |
5847 | need_truncate = 0; | |
e2d044fe | 5848 | } else { |
ffa7c429 FM |
5849 | u64 old_size; |
5850 | ||
e2d044fe | 5851 | ret = get_inode_info(sctx->parent_root, sctx->cur_ino, |
ffa7c429 | 5852 | &old_size, NULL, &right_mode, &right_uid, |
e2d044fe AL |
5853 | &right_gid, NULL); |
5854 | if (ret < 0) | |
5855 | goto out; | |
31db9f7c | 5856 | |
e2d044fe AL |
5857 | if (left_uid != right_uid || left_gid != right_gid) |
5858 | need_chown = 1; | |
5859 | if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) | |
5860 | need_chmod = 1; | |
ffa7c429 FM |
5861 | if ((old_size == sctx->cur_inode_size) || |
5862 | (sctx->cur_inode_size > old_size && | |
5863 | sctx->cur_inode_next_write_offset == sctx->cur_inode_size)) | |
5864 | need_truncate = 0; | |
31db9f7c AB |
5865 | } |
5866 | ||
5867 | if (S_ISREG(sctx->cur_inode_mode)) { | |
16e7549f | 5868 | if (need_send_hole(sctx)) { |
766b5e5a FM |
5869 | if (sctx->cur_inode_last_extent == (u64)-1 || |
5870 | sctx->cur_inode_last_extent < | |
5871 | sctx->cur_inode_size) { | |
16e7549f JB |
5872 | ret = get_last_extent(sctx, (u64)-1); |
5873 | if (ret) | |
5874 | goto out; | |
5875 | } | |
5876 | if (sctx->cur_inode_last_extent < | |
5877 | sctx->cur_inode_size) { | |
5878 | ret = send_hole(sctx, sctx->cur_inode_size); | |
5879 | if (ret) | |
5880 | goto out; | |
5881 | } | |
5882 | } | |
ffa7c429 FM |
5883 | if (need_truncate) { |
5884 | ret = send_truncate(sctx, sctx->cur_ino, | |
5885 | sctx->cur_inode_gen, | |
5886 | sctx->cur_inode_size); | |
5887 | if (ret < 0) | |
5888 | goto out; | |
5889 | } | |
31db9f7c AB |
5890 | } |
5891 | ||
5892 | if (need_chown) { | |
5893 | ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5894 | left_uid, left_gid); | |
5895 | if (ret < 0) | |
5896 | goto out; | |
5897 | } | |
5898 | if (need_chmod) { | |
5899 | ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5900 | left_mode); | |
5901 | if (ret < 0) | |
5902 | goto out; | |
5903 | } | |
5904 | ||
5905 | /* | |
9f03740a FDBM |
5906 | * If other directory inodes depended on our current directory |
5907 | * inode's move/rename, now do their move/rename operations. | |
31db9f7c | 5908 | */ |
9f03740a FDBM |
5909 | if (!is_waiting_for_move(sctx, sctx->cur_ino)) { |
5910 | ret = apply_children_dir_moves(sctx); | |
5911 | if (ret) | |
5912 | goto out; | |
fcbd2154 FM |
5913 | /* |
5914 | * Need to send that every time, no matter if it actually | |
5915 | * changed between the two trees as we have done changes to | |
5916 | * the inode before. If our inode is a directory and it's | |
5917 | * waiting to be moved/renamed, we will send its utimes when | |
5918 | * it's moved/renamed, therefore we don't need to do it here. | |
5919 | */ | |
5920 | sctx->send_progress = sctx->cur_ino + 1; | |
5921 | ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
5922 | if (ret < 0) | |
5923 | goto out; | |
9f03740a FDBM |
5924 | } |
5925 | ||
31db9f7c AB |
5926 | out: |
5927 | return ret; | |
5928 | } | |
5929 | ||
46b2f459 FM |
5930 | struct parent_paths_ctx { |
5931 | struct list_head *refs; | |
5932 | struct send_ctx *sctx; | |
5933 | }; | |
5934 | ||
5935 | static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name, | |
5936 | void *ctx) | |
5937 | { | |
5938 | struct parent_paths_ctx *ppctx = ctx; | |
5939 | ||
5940 | return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx, | |
5941 | ppctx->refs); | |
5942 | } | |
5943 | ||
5944 | /* | |
5945 | * Issue unlink operations for all paths of the current inode found in the | |
5946 | * parent snapshot. | |
5947 | */ | |
5948 | static int btrfs_unlink_all_paths(struct send_ctx *sctx) | |
5949 | { | |
5950 | LIST_HEAD(deleted_refs); | |
5951 | struct btrfs_path *path; | |
5952 | struct btrfs_key key; | |
5953 | struct parent_paths_ctx ctx; | |
5954 | int ret; | |
5955 | ||
5956 | path = alloc_path_for_send(); | |
5957 | if (!path) | |
5958 | return -ENOMEM; | |
5959 | ||
5960 | key.objectid = sctx->cur_ino; | |
5961 | key.type = BTRFS_INODE_REF_KEY; | |
5962 | key.offset = 0; | |
5963 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
5964 | if (ret < 0) | |
5965 | goto out; | |
5966 | ||
5967 | ctx.refs = &deleted_refs; | |
5968 | ctx.sctx = sctx; | |
5969 | ||
5970 | while (true) { | |
5971 | struct extent_buffer *eb = path->nodes[0]; | |
5972 | int slot = path->slots[0]; | |
5973 | ||
5974 | if (slot >= btrfs_header_nritems(eb)) { | |
5975 | ret = btrfs_next_leaf(sctx->parent_root, path); | |
5976 | if (ret < 0) | |
5977 | goto out; | |
5978 | else if (ret > 0) | |
5979 | break; | |
5980 | continue; | |
5981 | } | |
5982 | ||
5983 | btrfs_item_key_to_cpu(eb, &key, slot); | |
5984 | if (key.objectid != sctx->cur_ino) | |
5985 | break; | |
5986 | if (key.type != BTRFS_INODE_REF_KEY && | |
5987 | key.type != BTRFS_INODE_EXTREF_KEY) | |
5988 | break; | |
5989 | ||
5990 | ret = iterate_inode_ref(sctx->parent_root, path, &key, 1, | |
5991 | record_parent_ref, &ctx); | |
5992 | if (ret < 0) | |
5993 | goto out; | |
5994 | ||
5995 | path->slots[0]++; | |
5996 | } | |
5997 | ||
5998 | while (!list_empty(&deleted_refs)) { | |
5999 | struct recorded_ref *ref; | |
6000 | ||
6001 | ref = list_first_entry(&deleted_refs, struct recorded_ref, list); | |
6002 | ret = send_unlink(sctx, ref->full_path); | |
6003 | if (ret < 0) | |
6004 | goto out; | |
6005 | fs_path_free(ref->full_path); | |
6006 | list_del(&ref->list); | |
6007 | kfree(ref); | |
6008 | } | |
6009 | ret = 0; | |
6010 | out: | |
6011 | btrfs_free_path(path); | |
6012 | if (ret) | |
6013 | __free_recorded_refs(&deleted_refs); | |
6014 | return ret; | |
6015 | } | |
6016 | ||
31db9f7c AB |
6017 | static int changed_inode(struct send_ctx *sctx, |
6018 | enum btrfs_compare_tree_result result) | |
6019 | { | |
6020 | int ret = 0; | |
6021 | struct btrfs_key *key = sctx->cmp_key; | |
6022 | struct btrfs_inode_item *left_ii = NULL; | |
6023 | struct btrfs_inode_item *right_ii = NULL; | |
6024 | u64 left_gen = 0; | |
6025 | u64 right_gen = 0; | |
6026 | ||
31db9f7c AB |
6027 | sctx->cur_ino = key->objectid; |
6028 | sctx->cur_inode_new_gen = 0; | |
16e7549f | 6029 | sctx->cur_inode_last_extent = (u64)-1; |
ffa7c429 | 6030 | sctx->cur_inode_next_write_offset = 0; |
46b2f459 | 6031 | sctx->ignore_cur_inode = false; |
e479d9bb AB |
6032 | |
6033 | /* | |
6034 | * Set send_progress to current inode. This will tell all get_cur_xxx | |
6035 | * functions that the current inode's refs are not updated yet. Later, | |
6036 | * when process_recorded_refs is finished, it is set to cur_ino + 1. | |
6037 | */ | |
31db9f7c AB |
6038 | sctx->send_progress = sctx->cur_ino; |
6039 | ||
6040 | if (result == BTRFS_COMPARE_TREE_NEW || | |
6041 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
6042 | left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], | |
6043 | sctx->left_path->slots[0], | |
6044 | struct btrfs_inode_item); | |
6045 | left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], | |
6046 | left_ii); | |
6047 | } else { | |
6048 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6049 | sctx->right_path->slots[0], | |
6050 | struct btrfs_inode_item); | |
6051 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6052 | right_ii); | |
6053 | } | |
6054 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
6055 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6056 | sctx->right_path->slots[0], | |
6057 | struct btrfs_inode_item); | |
6058 | ||
6059 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6060 | right_ii); | |
6d85ed05 AB |
6061 | |
6062 | /* | |
6063 | * The cur_ino = root dir case is special here. We can't treat | |
6064 | * the inode as deleted+reused because it would generate a | |
6065 | * stream that tries to delete/mkdir the root dir. | |
6066 | */ | |
6067 | if (left_gen != right_gen && | |
6068 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) | |
31db9f7c AB |
6069 | sctx->cur_inode_new_gen = 1; |
6070 | } | |
6071 | ||
46b2f459 FM |
6072 | /* |
6073 | * Normally we do not find inodes with a link count of zero (orphans) | |
6074 | * because the most common case is to create a snapshot and use it | |
6075 | * for a send operation. However other less common use cases involve | |
6076 | * using a subvolume and send it after turning it to RO mode just | |
6077 | * after deleting all hard links of a file while holding an open | |
6078 | * file descriptor against it or turning a RO snapshot into RW mode, | |
6079 | * keep an open file descriptor against a file, delete it and then | |
6080 | * turn the snapshot back to RO mode before using it for a send | |
6081 | * operation. So if we find such cases, ignore the inode and all its | |
6082 | * items completely if it's a new inode, or if it's a changed inode | |
6083 | * make sure all its previous paths (from the parent snapshot) are all | |
6084 | * unlinked and all other the inode items are ignored. | |
6085 | */ | |
6086 | if (result == BTRFS_COMPARE_TREE_NEW || | |
6087 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
6088 | u32 nlinks; | |
6089 | ||
6090 | nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii); | |
6091 | if (nlinks == 0) { | |
6092 | sctx->ignore_cur_inode = true; | |
6093 | if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6094 | ret = btrfs_unlink_all_paths(sctx); | |
6095 | goto out; | |
6096 | } | |
6097 | } | |
6098 | ||
31db9f7c AB |
6099 | if (result == BTRFS_COMPARE_TREE_NEW) { |
6100 | sctx->cur_inode_gen = left_gen; | |
6101 | sctx->cur_inode_new = 1; | |
6102 | sctx->cur_inode_deleted = 0; | |
6103 | sctx->cur_inode_size = btrfs_inode_size( | |
6104 | sctx->left_path->nodes[0], left_ii); | |
6105 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6106 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6107 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6108 | sctx->left_path->nodes[0], left_ii); | |
31db9f7c | 6109 | if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) |
1f4692da | 6110 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6111 | } else if (result == BTRFS_COMPARE_TREE_DELETED) { |
6112 | sctx->cur_inode_gen = right_gen; | |
6113 | sctx->cur_inode_new = 0; | |
6114 | sctx->cur_inode_deleted = 1; | |
6115 | sctx->cur_inode_size = btrfs_inode_size( | |
6116 | sctx->right_path->nodes[0], right_ii); | |
6117 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6118 | sctx->right_path->nodes[0], right_ii); | |
6119 | } else if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
766702ef AB |
6120 | /* |
6121 | * We need to do some special handling in case the inode was | |
6122 | * reported as changed with a changed generation number. This | |
6123 | * means that the original inode was deleted and new inode | |
6124 | * reused the same inum. So we have to treat the old inode as | |
6125 | * deleted and the new one as new. | |
6126 | */ | |
31db9f7c | 6127 | if (sctx->cur_inode_new_gen) { |
766702ef AB |
6128 | /* |
6129 | * First, process the inode as if it was deleted. | |
6130 | */ | |
31db9f7c AB |
6131 | sctx->cur_inode_gen = right_gen; |
6132 | sctx->cur_inode_new = 0; | |
6133 | sctx->cur_inode_deleted = 1; | |
6134 | sctx->cur_inode_size = btrfs_inode_size( | |
6135 | sctx->right_path->nodes[0], right_ii); | |
6136 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6137 | sctx->right_path->nodes[0], right_ii); | |
6138 | ret = process_all_refs(sctx, | |
6139 | BTRFS_COMPARE_TREE_DELETED); | |
6140 | if (ret < 0) | |
6141 | goto out; | |
6142 | ||
766702ef AB |
6143 | /* |
6144 | * Now process the inode as if it was new. | |
6145 | */ | |
31db9f7c AB |
6146 | sctx->cur_inode_gen = left_gen; |
6147 | sctx->cur_inode_new = 1; | |
6148 | sctx->cur_inode_deleted = 0; | |
6149 | sctx->cur_inode_size = btrfs_inode_size( | |
6150 | sctx->left_path->nodes[0], left_ii); | |
6151 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6152 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6153 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6154 | sctx->left_path->nodes[0], left_ii); | |
1f4692da | 6155 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6156 | if (ret < 0) |
6157 | goto out; | |
6158 | ||
6159 | ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); | |
6160 | if (ret < 0) | |
6161 | goto out; | |
e479d9bb AB |
6162 | /* |
6163 | * Advance send_progress now as we did not get into | |
6164 | * process_recorded_refs_if_needed in the new_gen case. | |
6165 | */ | |
6166 | sctx->send_progress = sctx->cur_ino + 1; | |
766702ef AB |
6167 | |
6168 | /* | |
6169 | * Now process all extents and xattrs of the inode as if | |
6170 | * they were all new. | |
6171 | */ | |
31db9f7c AB |
6172 | ret = process_all_extents(sctx); |
6173 | if (ret < 0) | |
6174 | goto out; | |
6175 | ret = process_all_new_xattrs(sctx); | |
6176 | if (ret < 0) | |
6177 | goto out; | |
6178 | } else { | |
6179 | sctx->cur_inode_gen = left_gen; | |
6180 | sctx->cur_inode_new = 0; | |
6181 | sctx->cur_inode_new_gen = 0; | |
6182 | sctx->cur_inode_deleted = 0; | |
6183 | sctx->cur_inode_size = btrfs_inode_size( | |
6184 | sctx->left_path->nodes[0], left_ii); | |
6185 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6186 | sctx->left_path->nodes[0], left_ii); | |
6187 | } | |
6188 | } | |
6189 | ||
6190 | out: | |
6191 | return ret; | |
6192 | } | |
6193 | ||
766702ef AB |
6194 | /* |
6195 | * We have to process new refs before deleted refs, but compare_trees gives us | |
6196 | * the new and deleted refs mixed. To fix this, we record the new/deleted refs | |
6197 | * first and later process them in process_recorded_refs. | |
6198 | * For the cur_inode_new_gen case, we skip recording completely because | |
6199 | * changed_inode did already initiate processing of refs. The reason for this is | |
6200 | * that in this case, compare_tree actually compares the refs of 2 different | |
6201 | * inodes. To fix this, process_all_refs is used in changed_inode to handle all | |
6202 | * refs of the right tree as deleted and all refs of the left tree as new. | |
6203 | */ | |
31db9f7c AB |
6204 | static int changed_ref(struct send_ctx *sctx, |
6205 | enum btrfs_compare_tree_result result) | |
6206 | { | |
6207 | int ret = 0; | |
6208 | ||
95155585 FM |
6209 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6210 | inconsistent_snapshot_error(sctx, result, "reference"); | |
6211 | return -EIO; | |
6212 | } | |
31db9f7c AB |
6213 | |
6214 | if (!sctx->cur_inode_new_gen && | |
6215 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { | |
6216 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6217 | ret = record_new_ref(sctx); | |
6218 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6219 | ret = record_deleted_ref(sctx); | |
6220 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6221 | ret = record_changed_ref(sctx); | |
6222 | } | |
6223 | ||
6224 | return ret; | |
6225 | } | |
6226 | ||
766702ef AB |
6227 | /* |
6228 | * Process new/deleted/changed xattrs. We skip processing in the | |
6229 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6230 | * of xattrs. The reason is the same as in changed_ref | |
6231 | */ | |
31db9f7c AB |
6232 | static int changed_xattr(struct send_ctx *sctx, |
6233 | enum btrfs_compare_tree_result result) | |
6234 | { | |
6235 | int ret = 0; | |
6236 | ||
95155585 FM |
6237 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
6238 | inconsistent_snapshot_error(sctx, result, "xattr"); | |
6239 | return -EIO; | |
6240 | } | |
31db9f7c AB |
6241 | |
6242 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6243 | if (result == BTRFS_COMPARE_TREE_NEW) | |
6244 | ret = process_new_xattr(sctx); | |
6245 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
6246 | ret = process_deleted_xattr(sctx); | |
6247 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
6248 | ret = process_changed_xattr(sctx); | |
6249 | } | |
6250 | ||
6251 | return ret; | |
6252 | } | |
6253 | ||
766702ef AB |
6254 | /* |
6255 | * Process new/deleted/changed extents. We skip processing in the | |
6256 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
6257 | * of extents. The reason is the same as in changed_ref | |
6258 | */ | |
31db9f7c AB |
6259 | static int changed_extent(struct send_ctx *sctx, |
6260 | enum btrfs_compare_tree_result result) | |
6261 | { | |
6262 | int ret = 0; | |
6263 | ||
95155585 | 6264 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
d5e84fd8 FM |
6265 | |
6266 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
6267 | struct extent_buffer *leaf_l; | |
6268 | struct extent_buffer *leaf_r; | |
6269 | struct btrfs_file_extent_item *ei_l; | |
6270 | struct btrfs_file_extent_item *ei_r; | |
6271 | ||
6272 | leaf_l = sctx->left_path->nodes[0]; | |
6273 | leaf_r = sctx->right_path->nodes[0]; | |
6274 | ei_l = btrfs_item_ptr(leaf_l, | |
6275 | sctx->left_path->slots[0], | |
6276 | struct btrfs_file_extent_item); | |
6277 | ei_r = btrfs_item_ptr(leaf_r, | |
6278 | sctx->right_path->slots[0], | |
6279 | struct btrfs_file_extent_item); | |
6280 | ||
6281 | /* | |
6282 | * We may have found an extent item that has changed | |
6283 | * only its disk_bytenr field and the corresponding | |
6284 | * inode item was not updated. This case happens due to | |
6285 | * very specific timings during relocation when a leaf | |
6286 | * that contains file extent items is COWed while | |
6287 | * relocation is ongoing and its in the stage where it | |
6288 | * updates data pointers. So when this happens we can | |
6289 | * safely ignore it since we know it's the same extent, | |
6290 | * but just at different logical and physical locations | |
6291 | * (when an extent is fully replaced with a new one, we | |
6292 | * know the generation number must have changed too, | |
6293 | * since snapshot creation implies committing the current | |
6294 | * transaction, and the inode item must have been updated | |
6295 | * as well). | |
6296 | * This replacement of the disk_bytenr happens at | |
6297 | * relocation.c:replace_file_extents() through | |
6298 | * relocation.c:btrfs_reloc_cow_block(). | |
6299 | */ | |
6300 | if (btrfs_file_extent_generation(leaf_l, ei_l) == | |
6301 | btrfs_file_extent_generation(leaf_r, ei_r) && | |
6302 | btrfs_file_extent_ram_bytes(leaf_l, ei_l) == | |
6303 | btrfs_file_extent_ram_bytes(leaf_r, ei_r) && | |
6304 | btrfs_file_extent_compression(leaf_l, ei_l) == | |
6305 | btrfs_file_extent_compression(leaf_r, ei_r) && | |
6306 | btrfs_file_extent_encryption(leaf_l, ei_l) == | |
6307 | btrfs_file_extent_encryption(leaf_r, ei_r) && | |
6308 | btrfs_file_extent_other_encoding(leaf_l, ei_l) == | |
6309 | btrfs_file_extent_other_encoding(leaf_r, ei_r) && | |
6310 | btrfs_file_extent_type(leaf_l, ei_l) == | |
6311 | btrfs_file_extent_type(leaf_r, ei_r) && | |
6312 | btrfs_file_extent_disk_bytenr(leaf_l, ei_l) != | |
6313 | btrfs_file_extent_disk_bytenr(leaf_r, ei_r) && | |
6314 | btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) == | |
6315 | btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) && | |
6316 | btrfs_file_extent_offset(leaf_l, ei_l) == | |
6317 | btrfs_file_extent_offset(leaf_r, ei_r) && | |
6318 | btrfs_file_extent_num_bytes(leaf_l, ei_l) == | |
6319 | btrfs_file_extent_num_bytes(leaf_r, ei_r)) | |
6320 | return 0; | |
6321 | } | |
6322 | ||
95155585 FM |
6323 | inconsistent_snapshot_error(sctx, result, "extent"); |
6324 | return -EIO; | |
6325 | } | |
31db9f7c AB |
6326 | |
6327 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
6328 | if (result != BTRFS_COMPARE_TREE_DELETED) | |
6329 | ret = process_extent(sctx, sctx->left_path, | |
6330 | sctx->cmp_key); | |
6331 | } | |
6332 | ||
6333 | return ret; | |
6334 | } | |
6335 | ||
ba5e8f2e JB |
6336 | static int dir_changed(struct send_ctx *sctx, u64 dir) |
6337 | { | |
6338 | u64 orig_gen, new_gen; | |
6339 | int ret; | |
6340 | ||
6341 | ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, | |
6342 | NULL, NULL); | |
6343 | if (ret) | |
6344 | return ret; | |
6345 | ||
6346 | ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, | |
6347 | NULL, NULL, NULL); | |
6348 | if (ret) | |
6349 | return ret; | |
6350 | ||
6351 | return (orig_gen != new_gen) ? 1 : 0; | |
6352 | } | |
6353 | ||
6354 | static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, | |
6355 | struct btrfs_key *key) | |
6356 | { | |
6357 | struct btrfs_inode_extref *extref; | |
6358 | struct extent_buffer *leaf; | |
6359 | u64 dirid = 0, last_dirid = 0; | |
6360 | unsigned long ptr; | |
6361 | u32 item_size; | |
6362 | u32 cur_offset = 0; | |
6363 | int ref_name_len; | |
6364 | int ret = 0; | |
6365 | ||
6366 | /* Easy case, just check this one dirid */ | |
6367 | if (key->type == BTRFS_INODE_REF_KEY) { | |
6368 | dirid = key->offset; | |
6369 | ||
6370 | ret = dir_changed(sctx, dirid); | |
6371 | goto out; | |
6372 | } | |
6373 | ||
6374 | leaf = path->nodes[0]; | |
6375 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6376 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
6377 | while (cur_offset < item_size) { | |
6378 | extref = (struct btrfs_inode_extref *)(ptr + | |
6379 | cur_offset); | |
6380 | dirid = btrfs_inode_extref_parent(leaf, extref); | |
6381 | ref_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
6382 | cur_offset += ref_name_len + sizeof(*extref); | |
6383 | if (dirid == last_dirid) | |
6384 | continue; | |
6385 | ret = dir_changed(sctx, dirid); | |
6386 | if (ret) | |
6387 | break; | |
6388 | last_dirid = dirid; | |
6389 | } | |
6390 | out: | |
6391 | return ret; | |
6392 | } | |
6393 | ||
766702ef AB |
6394 | /* |
6395 | * Updates compare related fields in sctx and simply forwards to the actual | |
6396 | * changed_xxx functions. | |
6397 | */ | |
ee8c494f | 6398 | static int changed_cb(struct btrfs_path *left_path, |
31db9f7c AB |
6399 | struct btrfs_path *right_path, |
6400 | struct btrfs_key *key, | |
6401 | enum btrfs_compare_tree_result result, | |
6402 | void *ctx) | |
6403 | { | |
6404 | int ret = 0; | |
6405 | struct send_ctx *sctx = ctx; | |
6406 | ||
ba5e8f2e | 6407 | if (result == BTRFS_COMPARE_TREE_SAME) { |
16e7549f JB |
6408 | if (key->type == BTRFS_INODE_REF_KEY || |
6409 | key->type == BTRFS_INODE_EXTREF_KEY) { | |
6410 | ret = compare_refs(sctx, left_path, key); | |
6411 | if (!ret) | |
6412 | return 0; | |
6413 | if (ret < 0) | |
6414 | return ret; | |
6415 | } else if (key->type == BTRFS_EXTENT_DATA_KEY) { | |
6416 | return maybe_send_hole(sctx, left_path, key); | |
6417 | } else { | |
ba5e8f2e | 6418 | return 0; |
16e7549f | 6419 | } |
ba5e8f2e JB |
6420 | result = BTRFS_COMPARE_TREE_CHANGED; |
6421 | ret = 0; | |
6422 | } | |
6423 | ||
31db9f7c AB |
6424 | sctx->left_path = left_path; |
6425 | sctx->right_path = right_path; | |
6426 | sctx->cmp_key = key; | |
6427 | ||
6428 | ret = finish_inode_if_needed(sctx, 0); | |
6429 | if (ret < 0) | |
6430 | goto out; | |
6431 | ||
2981e225 AB |
6432 | /* Ignore non-FS objects */ |
6433 | if (key->objectid == BTRFS_FREE_INO_OBJECTID || | |
6434 | key->objectid == BTRFS_FREE_SPACE_OBJECTID) | |
6435 | goto out; | |
6436 | ||
46b2f459 | 6437 | if (key->type == BTRFS_INODE_ITEM_KEY) { |
31db9f7c | 6438 | ret = changed_inode(sctx, result); |
46b2f459 FM |
6439 | } else if (!sctx->ignore_cur_inode) { |
6440 | if (key->type == BTRFS_INODE_REF_KEY || | |
6441 | key->type == BTRFS_INODE_EXTREF_KEY) | |
6442 | ret = changed_ref(sctx, result); | |
6443 | else if (key->type == BTRFS_XATTR_ITEM_KEY) | |
6444 | ret = changed_xattr(sctx, result); | |
6445 | else if (key->type == BTRFS_EXTENT_DATA_KEY) | |
6446 | ret = changed_extent(sctx, result); | |
6447 | } | |
31db9f7c AB |
6448 | |
6449 | out: | |
6450 | return ret; | |
6451 | } | |
6452 | ||
6453 | static int full_send_tree(struct send_ctx *sctx) | |
6454 | { | |
6455 | int ret; | |
31db9f7c AB |
6456 | struct btrfs_root *send_root = sctx->send_root; |
6457 | struct btrfs_key key; | |
31db9f7c AB |
6458 | struct btrfs_path *path; |
6459 | struct extent_buffer *eb; | |
6460 | int slot; | |
31db9f7c AB |
6461 | |
6462 | path = alloc_path_for_send(); | |
6463 | if (!path) | |
6464 | return -ENOMEM; | |
6465 | ||
31db9f7c AB |
6466 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; |
6467 | key.type = BTRFS_INODE_ITEM_KEY; | |
6468 | key.offset = 0; | |
6469 | ||
31db9f7c AB |
6470 | ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); |
6471 | if (ret < 0) | |
6472 | goto out; | |
6473 | if (ret) | |
6474 | goto out_finish; | |
6475 | ||
6476 | while (1) { | |
31db9f7c AB |
6477 | eb = path->nodes[0]; |
6478 | slot = path->slots[0]; | |
ca5d2ba1 | 6479 | btrfs_item_key_to_cpu(eb, &key, slot); |
31db9f7c | 6480 | |
ca5d2ba1 | 6481 | ret = changed_cb(path, NULL, &key, |
ee8c494f | 6482 | BTRFS_COMPARE_TREE_NEW, sctx); |
31db9f7c AB |
6483 | if (ret < 0) |
6484 | goto out; | |
6485 | ||
31db9f7c AB |
6486 | ret = btrfs_next_item(send_root, path); |
6487 | if (ret < 0) | |
6488 | goto out; | |
6489 | if (ret) { | |
6490 | ret = 0; | |
6491 | break; | |
6492 | } | |
6493 | } | |
6494 | ||
6495 | out_finish: | |
6496 | ret = finish_inode_if_needed(sctx, 1); | |
6497 | ||
6498 | out: | |
6499 | btrfs_free_path(path); | |
31db9f7c AB |
6500 | return ret; |
6501 | } | |
6502 | ||
6503 | static int send_subvol(struct send_ctx *sctx) | |
6504 | { | |
6505 | int ret; | |
6506 | ||
c2c71324 SB |
6507 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { |
6508 | ret = send_header(sctx); | |
6509 | if (ret < 0) | |
6510 | goto out; | |
6511 | } | |
31db9f7c AB |
6512 | |
6513 | ret = send_subvol_begin(sctx); | |
6514 | if (ret < 0) | |
6515 | goto out; | |
6516 | ||
6517 | if (sctx->parent_root) { | |
6518 | ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, | |
6519 | changed_cb, sctx); | |
6520 | if (ret < 0) | |
6521 | goto out; | |
6522 | ret = finish_inode_if_needed(sctx, 1); | |
6523 | if (ret < 0) | |
6524 | goto out; | |
6525 | } else { | |
6526 | ret = full_send_tree(sctx); | |
6527 | if (ret < 0) | |
6528 | goto out; | |
6529 | } | |
6530 | ||
6531 | out: | |
31db9f7c AB |
6532 | free_recorded_refs(sctx); |
6533 | return ret; | |
6534 | } | |
6535 | ||
e5fa8f86 FM |
6536 | /* |
6537 | * If orphan cleanup did remove any orphans from a root, it means the tree | |
6538 | * was modified and therefore the commit root is not the same as the current | |
6539 | * root anymore. This is a problem, because send uses the commit root and | |
6540 | * therefore can see inode items that don't exist in the current root anymore, | |
6541 | * and for example make calls to btrfs_iget, which will do tree lookups based | |
6542 | * on the current root and not on the commit root. Those lookups will fail, | |
6543 | * returning a -ESTALE error, and making send fail with that error. So make | |
6544 | * sure a send does not see any orphans we have just removed, and that it will | |
6545 | * see the same inodes regardless of whether a transaction commit happened | |
6546 | * before it started (meaning that the commit root will be the same as the | |
6547 | * current root) or not. | |
6548 | */ | |
6549 | static int ensure_commit_roots_uptodate(struct send_ctx *sctx) | |
6550 | { | |
6551 | int i; | |
6552 | struct btrfs_trans_handle *trans = NULL; | |
6553 | ||
6554 | again: | |
6555 | if (sctx->parent_root && | |
6556 | sctx->parent_root->node != sctx->parent_root->commit_root) | |
6557 | goto commit_trans; | |
6558 | ||
6559 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6560 | if (sctx->clone_roots[i].root->node != | |
6561 | sctx->clone_roots[i].root->commit_root) | |
6562 | goto commit_trans; | |
6563 | ||
6564 | if (trans) | |
3a45bb20 | 6565 | return btrfs_end_transaction(trans); |
e5fa8f86 FM |
6566 | |
6567 | return 0; | |
6568 | ||
6569 | commit_trans: | |
6570 | /* Use any root, all fs roots will get their commit roots updated. */ | |
6571 | if (!trans) { | |
6572 | trans = btrfs_join_transaction(sctx->send_root); | |
6573 | if (IS_ERR(trans)) | |
6574 | return PTR_ERR(trans); | |
6575 | goto again; | |
6576 | } | |
6577 | ||
3a45bb20 | 6578 | return btrfs_commit_transaction(trans); |
e5fa8f86 FM |
6579 | } |
6580 | ||
66ef7d65 DS |
6581 | static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) |
6582 | { | |
6583 | spin_lock(&root->root_item_lock); | |
6584 | root->send_in_progress--; | |
6585 | /* | |
6586 | * Not much left to do, we don't know why it's unbalanced and | |
6587 | * can't blindly reset it to 0. | |
6588 | */ | |
6589 | if (root->send_in_progress < 0) | |
6590 | btrfs_err(root->fs_info, | |
f5686e3a | 6591 | "send_in_progress unbalanced %d root %llu", |
0b246afa | 6592 | root->send_in_progress, root->root_key.objectid); |
66ef7d65 DS |
6593 | spin_unlock(&root->root_item_lock); |
6594 | } | |
6595 | ||
2351f431 | 6596 | long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) |
31db9f7c AB |
6597 | { |
6598 | int ret = 0; | |
0b246afa JM |
6599 | struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root; |
6600 | struct btrfs_fs_info *fs_info = send_root->fs_info; | |
31db9f7c | 6601 | struct btrfs_root *clone_root; |
31db9f7c | 6602 | struct btrfs_key key; |
31db9f7c AB |
6603 | struct send_ctx *sctx = NULL; |
6604 | u32 i; | |
6605 | u64 *clone_sources_tmp = NULL; | |
2c686537 | 6606 | int clone_sources_to_rollback = 0; |
e55d1153 | 6607 | unsigned alloc_size; |
896c14f9 | 6608 | int sort_clone_roots = 0; |
18f687d5 | 6609 | int index; |
31db9f7c AB |
6610 | |
6611 | if (!capable(CAP_SYS_ADMIN)) | |
6612 | return -EPERM; | |
6613 | ||
2c686537 DS |
6614 | /* |
6615 | * The subvolume must remain read-only during send, protect against | |
521e0546 | 6616 | * making it RW. This also protects against deletion. |
2c686537 DS |
6617 | */ |
6618 | spin_lock(&send_root->root_item_lock); | |
6619 | send_root->send_in_progress++; | |
6620 | spin_unlock(&send_root->root_item_lock); | |
6621 | ||
139f807a JB |
6622 | /* |
6623 | * This is done when we lookup the root, it should already be complete | |
6624 | * by the time we get here. | |
6625 | */ | |
6626 | WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE); | |
6627 | ||
2c686537 DS |
6628 | /* |
6629 | * Userspace tools do the checks and warn the user if it's | |
6630 | * not RO. | |
6631 | */ | |
6632 | if (!btrfs_root_readonly(send_root)) { | |
6633 | ret = -EPERM; | |
6634 | goto out; | |
6635 | } | |
6636 | ||
457ae726 DC |
6637 | /* |
6638 | * Check that we don't overflow at later allocations, we request | |
6639 | * clone_sources_count + 1 items, and compare to unsigned long inside | |
6640 | * access_ok. | |
6641 | */ | |
f5ecec3c | 6642 | if (arg->clone_sources_count > |
457ae726 | 6643 | ULONG_MAX / sizeof(struct clone_root) - 1) { |
f5ecec3c DC |
6644 | ret = -EINVAL; |
6645 | goto out; | |
6646 | } | |
6647 | ||
31db9f7c | 6648 | if (!access_ok(VERIFY_READ, arg->clone_sources, |
700ff4f0 DC |
6649 | sizeof(*arg->clone_sources) * |
6650 | arg->clone_sources_count)) { | |
31db9f7c AB |
6651 | ret = -EFAULT; |
6652 | goto out; | |
6653 | } | |
6654 | ||
c2c71324 | 6655 | if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { |
cb95e7bf MF |
6656 | ret = -EINVAL; |
6657 | goto out; | |
6658 | } | |
6659 | ||
e780b0d1 | 6660 | sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL); |
31db9f7c AB |
6661 | if (!sctx) { |
6662 | ret = -ENOMEM; | |
6663 | goto out; | |
6664 | } | |
6665 | ||
6666 | INIT_LIST_HEAD(&sctx->new_refs); | |
6667 | INIT_LIST_HEAD(&sctx->deleted_refs); | |
e780b0d1 | 6668 | INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL); |
31db9f7c AB |
6669 | INIT_LIST_HEAD(&sctx->name_cache_list); |
6670 | ||
cb95e7bf MF |
6671 | sctx->flags = arg->flags; |
6672 | ||
31db9f7c | 6673 | sctx->send_filp = fget(arg->send_fd); |
ecc7ada7 TI |
6674 | if (!sctx->send_filp) { |
6675 | ret = -EBADF; | |
31db9f7c AB |
6676 | goto out; |
6677 | } | |
6678 | ||
31db9f7c | 6679 | sctx->send_root = send_root; |
521e0546 DS |
6680 | /* |
6681 | * Unlikely but possible, if the subvolume is marked for deletion but | |
6682 | * is slow to remove the directory entry, send can still be started | |
6683 | */ | |
6684 | if (btrfs_root_dead(sctx->send_root)) { | |
6685 | ret = -EPERM; | |
6686 | goto out; | |
6687 | } | |
6688 | ||
31db9f7c AB |
6689 | sctx->clone_roots_cnt = arg->clone_sources_count; |
6690 | ||
6691 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE; | |
752ade68 | 6692 | sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); |
31db9f7c | 6693 | if (!sctx->send_buf) { |
752ade68 MH |
6694 | ret = -ENOMEM; |
6695 | goto out; | |
31db9f7c AB |
6696 | } |
6697 | ||
752ade68 | 6698 | sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL); |
31db9f7c | 6699 | if (!sctx->read_buf) { |
752ade68 MH |
6700 | ret = -ENOMEM; |
6701 | goto out; | |
31db9f7c AB |
6702 | } |
6703 | ||
9f03740a FDBM |
6704 | sctx->pending_dir_moves = RB_ROOT; |
6705 | sctx->waiting_dir_moves = RB_ROOT; | |
9dc44214 | 6706 | sctx->orphan_dirs = RB_ROOT; |
9f03740a | 6707 | |
e55d1153 DS |
6708 | alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1); |
6709 | ||
818e010b | 6710 | sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 6711 | if (!sctx->clone_roots) { |
818e010b DS |
6712 | ret = -ENOMEM; |
6713 | goto out; | |
31db9f7c AB |
6714 | } |
6715 | ||
e55d1153 DS |
6716 | alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources); |
6717 | ||
31db9f7c | 6718 | if (arg->clone_sources_count) { |
752ade68 | 6719 | clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 6720 | if (!clone_sources_tmp) { |
752ade68 MH |
6721 | ret = -ENOMEM; |
6722 | goto out; | |
31db9f7c AB |
6723 | } |
6724 | ||
6725 | ret = copy_from_user(clone_sources_tmp, arg->clone_sources, | |
e55d1153 | 6726 | alloc_size); |
31db9f7c AB |
6727 | if (ret) { |
6728 | ret = -EFAULT; | |
6729 | goto out; | |
6730 | } | |
6731 | ||
6732 | for (i = 0; i < arg->clone_sources_count; i++) { | |
6733 | key.objectid = clone_sources_tmp[i]; | |
6734 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6735 | key.offset = (u64)-1; | |
18f687d5 WS |
6736 | |
6737 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6738 | ||
31db9f7c | 6739 | clone_root = btrfs_read_fs_root_no_name(fs_info, &key); |
31db9f7c | 6740 | if (IS_ERR(clone_root)) { |
18f687d5 | 6741 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
31db9f7c AB |
6742 | ret = PTR_ERR(clone_root); |
6743 | goto out; | |
6744 | } | |
2c686537 | 6745 | spin_lock(&clone_root->root_item_lock); |
5cc2b17e FM |
6746 | if (!btrfs_root_readonly(clone_root) || |
6747 | btrfs_root_dead(clone_root)) { | |
2c686537 | 6748 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 | 6749 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6750 | ret = -EPERM; |
6751 | goto out; | |
6752 | } | |
2f1f465a | 6753 | clone_root->send_in_progress++; |
2c686537 | 6754 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 WS |
6755 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
6756 | ||
31db9f7c | 6757 | sctx->clone_roots[i].root = clone_root; |
2f1f465a | 6758 | clone_sources_to_rollback = i + 1; |
31db9f7c | 6759 | } |
2f91306a | 6760 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6761 | clone_sources_tmp = NULL; |
6762 | } | |
6763 | ||
6764 | if (arg->parent_root) { | |
6765 | key.objectid = arg->parent_root; | |
6766 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6767 | key.offset = (u64)-1; | |
18f687d5 WS |
6768 | |
6769 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6770 | ||
31db9f7c | 6771 | sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key); |
b1b19596 | 6772 | if (IS_ERR(sctx->parent_root)) { |
18f687d5 | 6773 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
b1b19596 | 6774 | ret = PTR_ERR(sctx->parent_root); |
31db9f7c AB |
6775 | goto out; |
6776 | } | |
18f687d5 | 6777 | |
2c686537 DS |
6778 | spin_lock(&sctx->parent_root->root_item_lock); |
6779 | sctx->parent_root->send_in_progress++; | |
521e0546 DS |
6780 | if (!btrfs_root_readonly(sctx->parent_root) || |
6781 | btrfs_root_dead(sctx->parent_root)) { | |
2c686537 | 6782 | spin_unlock(&sctx->parent_root->root_item_lock); |
18f687d5 | 6783 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6784 | ret = -EPERM; |
6785 | goto out; | |
6786 | } | |
6787 | spin_unlock(&sctx->parent_root->root_item_lock); | |
18f687d5 WS |
6788 | |
6789 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
31db9f7c AB |
6790 | } |
6791 | ||
6792 | /* | |
6793 | * Clones from send_root are allowed, but only if the clone source | |
6794 | * is behind the current send position. This is checked while searching | |
6795 | * for possible clone sources. | |
6796 | */ | |
6797 | sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root; | |
6798 | ||
6799 | /* We do a bsearch later */ | |
6800 | sort(sctx->clone_roots, sctx->clone_roots_cnt, | |
6801 | sizeof(*sctx->clone_roots), __clone_root_cmp_sort, | |
6802 | NULL); | |
896c14f9 | 6803 | sort_clone_roots = 1; |
31db9f7c | 6804 | |
e5fa8f86 FM |
6805 | ret = ensure_commit_roots_uptodate(sctx); |
6806 | if (ret) | |
6807 | goto out; | |
6808 | ||
2755a0de | 6809 | current->journal_info = BTRFS_SEND_TRANS_STUB; |
31db9f7c | 6810 | ret = send_subvol(sctx); |
a26e8c9f | 6811 | current->journal_info = NULL; |
31db9f7c AB |
6812 | if (ret < 0) |
6813 | goto out; | |
6814 | ||
c2c71324 SB |
6815 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { |
6816 | ret = begin_cmd(sctx, BTRFS_SEND_C_END); | |
6817 | if (ret < 0) | |
6818 | goto out; | |
6819 | ret = send_cmd(sctx); | |
6820 | if (ret < 0) | |
6821 | goto out; | |
6822 | } | |
31db9f7c AB |
6823 | |
6824 | out: | |
9f03740a FDBM |
6825 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); |
6826 | while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { | |
6827 | struct rb_node *n; | |
6828 | struct pending_dir_move *pm; | |
6829 | ||
6830 | n = rb_first(&sctx->pending_dir_moves); | |
6831 | pm = rb_entry(n, struct pending_dir_move, node); | |
6832 | while (!list_empty(&pm->list)) { | |
6833 | struct pending_dir_move *pm2; | |
6834 | ||
6835 | pm2 = list_first_entry(&pm->list, | |
6836 | struct pending_dir_move, list); | |
6837 | free_pending_move(sctx, pm2); | |
6838 | } | |
6839 | free_pending_move(sctx, pm); | |
6840 | } | |
6841 | ||
6842 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); | |
6843 | while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { | |
6844 | struct rb_node *n; | |
6845 | struct waiting_dir_move *dm; | |
6846 | ||
6847 | n = rb_first(&sctx->waiting_dir_moves); | |
6848 | dm = rb_entry(n, struct waiting_dir_move, node); | |
6849 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
6850 | kfree(dm); | |
6851 | } | |
6852 | ||
9dc44214 FM |
6853 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); |
6854 | while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { | |
6855 | struct rb_node *n; | |
6856 | struct orphan_dir_info *odi; | |
6857 | ||
6858 | n = rb_first(&sctx->orphan_dirs); | |
6859 | odi = rb_entry(n, struct orphan_dir_info, node); | |
6860 | free_orphan_dir_info(sctx, odi); | |
6861 | } | |
6862 | ||
896c14f9 WS |
6863 | if (sort_clone_roots) { |
6864 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6865 | btrfs_root_dec_send_in_progress( | |
6866 | sctx->clone_roots[i].root); | |
6867 | } else { | |
6868 | for (i = 0; sctx && i < clone_sources_to_rollback; i++) | |
6869 | btrfs_root_dec_send_in_progress( | |
6870 | sctx->clone_roots[i].root); | |
6871 | ||
6872 | btrfs_root_dec_send_in_progress(send_root); | |
6873 | } | |
66ef7d65 DS |
6874 | if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) |
6875 | btrfs_root_dec_send_in_progress(sctx->parent_root); | |
2c686537 | 6876 | |
2f91306a | 6877 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6878 | |
6879 | if (sctx) { | |
6880 | if (sctx->send_filp) | |
6881 | fput(sctx->send_filp); | |
6882 | ||
c03d01f3 | 6883 | kvfree(sctx->clone_roots); |
6ff48ce0 | 6884 | kvfree(sctx->send_buf); |
eb5b75fe | 6885 | kvfree(sctx->read_buf); |
31db9f7c AB |
6886 | |
6887 | name_cache_free(sctx); | |
6888 | ||
6889 | kfree(sctx); | |
6890 | } | |
6891 | ||
6892 | return ret; | |
6893 | } |