]>
Commit | Line | Data |
---|---|---|
31db9f7c AB |
1 | /* |
2 | * Copyright (C) 2012 Alexander Block. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/bsearch.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/file.h> | |
22 | #include <linux/sort.h> | |
23 | #include <linux/mount.h> | |
24 | #include <linux/xattr.h> | |
25 | #include <linux/posix_acl_xattr.h> | |
26 | #include <linux/radix-tree.h> | |
a1857ebe | 27 | #include <linux/vmalloc.h> |
ed84885d | 28 | #include <linux/string.h> |
31db9f7c AB |
29 | |
30 | #include "send.h" | |
31 | #include "backref.h" | |
0b947aff | 32 | #include "hash.h" |
31db9f7c AB |
33 | #include "locking.h" |
34 | #include "disk-io.h" | |
35 | #include "btrfs_inode.h" | |
36 | #include "transaction.h" | |
ebb8765b | 37 | #include "compression.h" |
31db9f7c AB |
38 | |
39 | static int g_verbose = 0; | |
40 | ||
41 | #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__) | |
42 | ||
43 | /* | |
44 | * A fs_path is a helper to dynamically build path names with unknown size. | |
45 | * It reallocates the internal buffer on demand. | |
46 | * It allows fast adding of path elements on the right side (normal path) and | |
47 | * fast adding to the left side (reversed path). A reversed path can also be | |
48 | * unreversed if needed. | |
49 | */ | |
50 | struct fs_path { | |
51 | union { | |
52 | struct { | |
53 | char *start; | |
54 | char *end; | |
31db9f7c AB |
55 | |
56 | char *buf; | |
1f5a7ff9 DS |
57 | unsigned short buf_len:15; |
58 | unsigned short reversed:1; | |
31db9f7c AB |
59 | char inline_buf[]; |
60 | }; | |
ace01050 DS |
61 | /* |
62 | * Average path length does not exceed 200 bytes, we'll have | |
63 | * better packing in the slab and higher chance to satisfy | |
64 | * a allocation later during send. | |
65 | */ | |
66 | char pad[256]; | |
31db9f7c AB |
67 | }; |
68 | }; | |
69 | #define FS_PATH_INLINE_SIZE \ | |
70 | (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) | |
71 | ||
72 | ||
73 | /* reused for each extent */ | |
74 | struct clone_root { | |
75 | struct btrfs_root *root; | |
76 | u64 ino; | |
77 | u64 offset; | |
78 | ||
79 | u64 found_refs; | |
80 | }; | |
81 | ||
82 | #define SEND_CTX_MAX_NAME_CACHE_SIZE 128 | |
83 | #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2) | |
84 | ||
85 | struct send_ctx { | |
86 | struct file *send_filp; | |
87 | loff_t send_off; | |
88 | char *send_buf; | |
89 | u32 send_size; | |
90 | u32 send_max_size; | |
91 | u64 total_send_size; | |
92 | u64 cmd_send_size[BTRFS_SEND_C_MAX + 1]; | |
cb95e7bf | 93 | u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ |
31db9f7c | 94 | |
31db9f7c AB |
95 | struct btrfs_root *send_root; |
96 | struct btrfs_root *parent_root; | |
97 | struct clone_root *clone_roots; | |
98 | int clone_roots_cnt; | |
99 | ||
100 | /* current state of the compare_tree call */ | |
101 | struct btrfs_path *left_path; | |
102 | struct btrfs_path *right_path; | |
103 | struct btrfs_key *cmp_key; | |
104 | ||
105 | /* | |
106 | * infos of the currently processed inode. In case of deleted inodes, | |
107 | * these are the values from the deleted inode. | |
108 | */ | |
109 | u64 cur_ino; | |
110 | u64 cur_inode_gen; | |
111 | int cur_inode_new; | |
112 | int cur_inode_new_gen; | |
113 | int cur_inode_deleted; | |
31db9f7c AB |
114 | u64 cur_inode_size; |
115 | u64 cur_inode_mode; | |
644d1940 | 116 | u64 cur_inode_rdev; |
16e7549f | 117 | u64 cur_inode_last_extent; |
31db9f7c AB |
118 | |
119 | u64 send_progress; | |
120 | ||
121 | struct list_head new_refs; | |
122 | struct list_head deleted_refs; | |
123 | ||
124 | struct radix_tree_root name_cache; | |
125 | struct list_head name_cache_list; | |
126 | int name_cache_size; | |
127 | ||
2131bcd3 LB |
128 | struct file_ra_state ra; |
129 | ||
31db9f7c | 130 | char *read_buf; |
9f03740a FDBM |
131 | |
132 | /* | |
133 | * We process inodes by their increasing order, so if before an | |
134 | * incremental send we reverse the parent/child relationship of | |
135 | * directories such that a directory with a lower inode number was | |
136 | * the parent of a directory with a higher inode number, and the one | |
137 | * becoming the new parent got renamed too, we can't rename/move the | |
138 | * directory with lower inode number when we finish processing it - we | |
139 | * must process the directory with higher inode number first, then | |
140 | * rename/move it and then rename/move the directory with lower inode | |
141 | * number. Example follows. | |
142 | * | |
143 | * Tree state when the first send was performed: | |
144 | * | |
145 | * . | |
146 | * |-- a (ino 257) | |
147 | * |-- b (ino 258) | |
148 | * | | |
149 | * | | |
150 | * |-- c (ino 259) | |
151 | * | |-- d (ino 260) | |
152 | * | | |
153 | * |-- c2 (ino 261) | |
154 | * | |
155 | * Tree state when the second (incremental) send is performed: | |
156 | * | |
157 | * . | |
158 | * |-- a (ino 257) | |
159 | * |-- b (ino 258) | |
160 | * |-- c2 (ino 261) | |
161 | * |-- d2 (ino 260) | |
162 | * |-- cc (ino 259) | |
163 | * | |
164 | * The sequence of steps that lead to the second state was: | |
165 | * | |
166 | * mv /a/b/c/d /a/b/c2/d2 | |
167 | * mv /a/b/c /a/b/c2/d2/cc | |
168 | * | |
169 | * "c" has lower inode number, but we can't move it (2nd mv operation) | |
170 | * before we move "d", which has higher inode number. | |
171 | * | |
172 | * So we just memorize which move/rename operations must be performed | |
173 | * later when their respective parent is processed and moved/renamed. | |
174 | */ | |
175 | ||
176 | /* Indexed by parent directory inode number. */ | |
177 | struct rb_root pending_dir_moves; | |
178 | ||
179 | /* | |
180 | * Reverse index, indexed by the inode number of a directory that | |
181 | * is waiting for the move/rename of its immediate parent before its | |
182 | * own move/rename can be performed. | |
183 | */ | |
184 | struct rb_root waiting_dir_moves; | |
9dc44214 FM |
185 | |
186 | /* | |
187 | * A directory that is going to be rm'ed might have a child directory | |
188 | * which is in the pending directory moves index above. In this case, | |
189 | * the directory can only be removed after the move/rename of its child | |
190 | * is performed. Example: | |
191 | * | |
192 | * Parent snapshot: | |
193 | * | |
194 | * . (ino 256) | |
195 | * |-- a/ (ino 257) | |
196 | * |-- b/ (ino 258) | |
197 | * |-- c/ (ino 259) | |
198 | * | |-- x/ (ino 260) | |
199 | * | | |
200 | * |-- y/ (ino 261) | |
201 | * | |
202 | * Send snapshot: | |
203 | * | |
204 | * . (ino 256) | |
205 | * |-- a/ (ino 257) | |
206 | * |-- b/ (ino 258) | |
207 | * |-- YY/ (ino 261) | |
208 | * |-- x/ (ino 260) | |
209 | * | |
210 | * Sequence of steps that lead to the send snapshot: | |
211 | * rm -f /a/b/c/foo.txt | |
212 | * mv /a/b/y /a/b/YY | |
213 | * mv /a/b/c/x /a/b/YY | |
214 | * rmdir /a/b/c | |
215 | * | |
216 | * When the child is processed, its move/rename is delayed until its | |
217 | * parent is processed (as explained above), but all other operations | |
218 | * like update utimes, chown, chgrp, etc, are performed and the paths | |
219 | * that it uses for those operations must use the orphanized name of | |
220 | * its parent (the directory we're going to rm later), so we need to | |
221 | * memorize that name. | |
222 | * | |
223 | * Indexed by the inode number of the directory to be deleted. | |
224 | */ | |
225 | struct rb_root orphan_dirs; | |
9f03740a FDBM |
226 | }; |
227 | ||
228 | struct pending_dir_move { | |
229 | struct rb_node node; | |
230 | struct list_head list; | |
231 | u64 parent_ino; | |
232 | u64 ino; | |
233 | u64 gen; | |
234 | struct list_head update_refs; | |
235 | }; | |
236 | ||
237 | struct waiting_dir_move { | |
238 | struct rb_node node; | |
239 | u64 ino; | |
9dc44214 FM |
240 | /* |
241 | * There might be some directory that could not be removed because it | |
242 | * was waiting for this directory inode to be moved first. Therefore | |
243 | * after this directory is moved, we can try to rmdir the ino rmdir_ino. | |
244 | */ | |
245 | u64 rmdir_ino; | |
8b191a68 | 246 | bool orphanized; |
9dc44214 FM |
247 | }; |
248 | ||
249 | struct orphan_dir_info { | |
250 | struct rb_node node; | |
251 | u64 ino; | |
252 | u64 gen; | |
31db9f7c AB |
253 | }; |
254 | ||
255 | struct name_cache_entry { | |
256 | struct list_head list; | |
7e0926fe AB |
257 | /* |
258 | * radix_tree has only 32bit entries but we need to handle 64bit inums. | |
259 | * We use the lower 32bit of the 64bit inum to store it in the tree. If | |
260 | * more then one inum would fall into the same entry, we use radix_list | |
261 | * to store the additional entries. radix_list is also used to store | |
262 | * entries where two entries have the same inum but different | |
263 | * generations. | |
264 | */ | |
265 | struct list_head radix_list; | |
31db9f7c AB |
266 | u64 ino; |
267 | u64 gen; | |
268 | u64 parent_ino; | |
269 | u64 parent_gen; | |
270 | int ret; | |
271 | int need_later_update; | |
272 | int name_len; | |
273 | char name[]; | |
274 | }; | |
275 | ||
9f03740a FDBM |
276 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); |
277 | ||
9dc44214 FM |
278 | static struct waiting_dir_move * |
279 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino); | |
280 | ||
281 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino); | |
282 | ||
16e7549f JB |
283 | static int need_send_hole(struct send_ctx *sctx) |
284 | { | |
285 | return (sctx->parent_root && !sctx->cur_inode_new && | |
286 | !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && | |
287 | S_ISREG(sctx->cur_inode_mode)); | |
288 | } | |
289 | ||
31db9f7c AB |
290 | static void fs_path_reset(struct fs_path *p) |
291 | { | |
292 | if (p->reversed) { | |
293 | p->start = p->buf + p->buf_len - 1; | |
294 | p->end = p->start; | |
295 | *p->start = 0; | |
296 | } else { | |
297 | p->start = p->buf; | |
298 | p->end = p->start; | |
299 | *p->start = 0; | |
300 | } | |
301 | } | |
302 | ||
924794c9 | 303 | static struct fs_path *fs_path_alloc(void) |
31db9f7c AB |
304 | { |
305 | struct fs_path *p; | |
306 | ||
e780b0d1 | 307 | p = kmalloc(sizeof(*p), GFP_KERNEL); |
31db9f7c AB |
308 | if (!p) |
309 | return NULL; | |
310 | p->reversed = 0; | |
31db9f7c AB |
311 | p->buf = p->inline_buf; |
312 | p->buf_len = FS_PATH_INLINE_SIZE; | |
313 | fs_path_reset(p); | |
314 | return p; | |
315 | } | |
316 | ||
924794c9 | 317 | static struct fs_path *fs_path_alloc_reversed(void) |
31db9f7c AB |
318 | { |
319 | struct fs_path *p; | |
320 | ||
924794c9 | 321 | p = fs_path_alloc(); |
31db9f7c AB |
322 | if (!p) |
323 | return NULL; | |
324 | p->reversed = 1; | |
325 | fs_path_reset(p); | |
326 | return p; | |
327 | } | |
328 | ||
924794c9 | 329 | static void fs_path_free(struct fs_path *p) |
31db9f7c AB |
330 | { |
331 | if (!p) | |
332 | return; | |
ace01050 DS |
333 | if (p->buf != p->inline_buf) |
334 | kfree(p->buf); | |
31db9f7c AB |
335 | kfree(p); |
336 | } | |
337 | ||
338 | static int fs_path_len(struct fs_path *p) | |
339 | { | |
340 | return p->end - p->start; | |
341 | } | |
342 | ||
343 | static int fs_path_ensure_buf(struct fs_path *p, int len) | |
344 | { | |
345 | char *tmp_buf; | |
346 | int path_len; | |
347 | int old_buf_len; | |
348 | ||
349 | len++; | |
350 | ||
351 | if (p->buf_len >= len) | |
352 | return 0; | |
353 | ||
cfd4a535 CM |
354 | if (len > PATH_MAX) { |
355 | WARN_ON(1); | |
356 | return -ENOMEM; | |
357 | } | |
358 | ||
1b2782c8 DS |
359 | path_len = p->end - p->start; |
360 | old_buf_len = p->buf_len; | |
361 | ||
ace01050 DS |
362 | /* |
363 | * First time the inline_buf does not suffice | |
364 | */ | |
01a9a8a9 | 365 | if (p->buf == p->inline_buf) { |
e780b0d1 | 366 | tmp_buf = kmalloc(len, GFP_KERNEL); |
01a9a8a9 FM |
367 | if (tmp_buf) |
368 | memcpy(tmp_buf, p->buf, old_buf_len); | |
369 | } else { | |
e780b0d1 | 370 | tmp_buf = krealloc(p->buf, len, GFP_KERNEL); |
01a9a8a9 | 371 | } |
9c9ca00b DS |
372 | if (!tmp_buf) |
373 | return -ENOMEM; | |
374 | p->buf = tmp_buf; | |
375 | /* | |
376 | * The real size of the buffer is bigger, this will let the fast path | |
377 | * happen most of the time | |
378 | */ | |
379 | p->buf_len = ksize(p->buf); | |
ace01050 | 380 | |
31db9f7c AB |
381 | if (p->reversed) { |
382 | tmp_buf = p->buf + old_buf_len - path_len - 1; | |
383 | p->end = p->buf + p->buf_len - 1; | |
384 | p->start = p->end - path_len; | |
385 | memmove(p->start, tmp_buf, path_len + 1); | |
386 | } else { | |
387 | p->start = p->buf; | |
388 | p->end = p->start + path_len; | |
389 | } | |
390 | return 0; | |
391 | } | |
392 | ||
b23ab57d DS |
393 | static int fs_path_prepare_for_add(struct fs_path *p, int name_len, |
394 | char **prepared) | |
31db9f7c AB |
395 | { |
396 | int ret; | |
397 | int new_len; | |
398 | ||
399 | new_len = p->end - p->start + name_len; | |
400 | if (p->start != p->end) | |
401 | new_len++; | |
402 | ret = fs_path_ensure_buf(p, new_len); | |
403 | if (ret < 0) | |
404 | goto out; | |
405 | ||
406 | if (p->reversed) { | |
407 | if (p->start != p->end) | |
408 | *--p->start = '/'; | |
409 | p->start -= name_len; | |
b23ab57d | 410 | *prepared = p->start; |
31db9f7c AB |
411 | } else { |
412 | if (p->start != p->end) | |
413 | *p->end++ = '/'; | |
b23ab57d | 414 | *prepared = p->end; |
31db9f7c AB |
415 | p->end += name_len; |
416 | *p->end = 0; | |
417 | } | |
418 | ||
419 | out: | |
420 | return ret; | |
421 | } | |
422 | ||
423 | static int fs_path_add(struct fs_path *p, const char *name, int name_len) | |
424 | { | |
425 | int ret; | |
b23ab57d | 426 | char *prepared; |
31db9f7c | 427 | |
b23ab57d | 428 | ret = fs_path_prepare_for_add(p, name_len, &prepared); |
31db9f7c AB |
429 | if (ret < 0) |
430 | goto out; | |
b23ab57d | 431 | memcpy(prepared, name, name_len); |
31db9f7c AB |
432 | |
433 | out: | |
434 | return ret; | |
435 | } | |
436 | ||
437 | static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) | |
438 | { | |
439 | int ret; | |
b23ab57d | 440 | char *prepared; |
31db9f7c | 441 | |
b23ab57d | 442 | ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); |
31db9f7c AB |
443 | if (ret < 0) |
444 | goto out; | |
b23ab57d | 445 | memcpy(prepared, p2->start, p2->end - p2->start); |
31db9f7c AB |
446 | |
447 | out: | |
448 | return ret; | |
449 | } | |
450 | ||
451 | static int fs_path_add_from_extent_buffer(struct fs_path *p, | |
452 | struct extent_buffer *eb, | |
453 | unsigned long off, int len) | |
454 | { | |
455 | int ret; | |
b23ab57d | 456 | char *prepared; |
31db9f7c | 457 | |
b23ab57d | 458 | ret = fs_path_prepare_for_add(p, len, &prepared); |
31db9f7c AB |
459 | if (ret < 0) |
460 | goto out; | |
461 | ||
b23ab57d | 462 | read_extent_buffer(eb, prepared, off, len); |
31db9f7c AB |
463 | |
464 | out: | |
465 | return ret; | |
466 | } | |
467 | ||
31db9f7c AB |
468 | static int fs_path_copy(struct fs_path *p, struct fs_path *from) |
469 | { | |
470 | int ret; | |
471 | ||
472 | p->reversed = from->reversed; | |
473 | fs_path_reset(p); | |
474 | ||
475 | ret = fs_path_add_path(p, from); | |
476 | ||
477 | return ret; | |
478 | } | |
479 | ||
480 | ||
481 | static void fs_path_unreverse(struct fs_path *p) | |
482 | { | |
483 | char *tmp; | |
484 | int len; | |
485 | ||
486 | if (!p->reversed) | |
487 | return; | |
488 | ||
489 | tmp = p->start; | |
490 | len = p->end - p->start; | |
491 | p->start = p->buf; | |
492 | p->end = p->start + len; | |
493 | memmove(p->start, tmp, len + 1); | |
494 | p->reversed = 0; | |
495 | } | |
496 | ||
497 | static struct btrfs_path *alloc_path_for_send(void) | |
498 | { | |
499 | struct btrfs_path *path; | |
500 | ||
501 | path = btrfs_alloc_path(); | |
502 | if (!path) | |
503 | return NULL; | |
504 | path->search_commit_root = 1; | |
505 | path->skip_locking = 1; | |
3f8a18cc | 506 | path->need_commit_sem = 1; |
31db9f7c AB |
507 | return path; |
508 | } | |
509 | ||
48a3b636 | 510 | static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) |
31db9f7c AB |
511 | { |
512 | int ret; | |
513 | mm_segment_t old_fs; | |
514 | u32 pos = 0; | |
515 | ||
516 | old_fs = get_fs(); | |
517 | set_fs(KERNEL_DS); | |
518 | ||
519 | while (pos < len) { | |
d447d0da FF |
520 | ret = vfs_write(filp, (__force const char __user *)buf + pos, |
521 | len - pos, off); | |
31db9f7c AB |
522 | /* TODO handle that correctly */ |
523 | /*if (ret == -ERESTARTSYS) { | |
524 | continue; | |
525 | }*/ | |
526 | if (ret < 0) | |
527 | goto out; | |
528 | if (ret == 0) { | |
529 | ret = -EIO; | |
530 | goto out; | |
531 | } | |
532 | pos += ret; | |
533 | } | |
534 | ||
535 | ret = 0; | |
536 | ||
537 | out: | |
538 | set_fs(old_fs); | |
539 | return ret; | |
540 | } | |
541 | ||
542 | static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) | |
543 | { | |
544 | struct btrfs_tlv_header *hdr; | |
545 | int total_len = sizeof(*hdr) + len; | |
546 | int left = sctx->send_max_size - sctx->send_size; | |
547 | ||
548 | if (unlikely(left < total_len)) | |
549 | return -EOVERFLOW; | |
550 | ||
551 | hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); | |
552 | hdr->tlv_type = cpu_to_le16(attr); | |
553 | hdr->tlv_len = cpu_to_le16(len); | |
554 | memcpy(hdr + 1, data, len); | |
555 | sctx->send_size += total_len; | |
556 | ||
557 | return 0; | |
558 | } | |
559 | ||
95bc79d5 DS |
560 | #define TLV_PUT_DEFINE_INT(bits) \ |
561 | static int tlv_put_u##bits(struct send_ctx *sctx, \ | |
562 | u##bits attr, u##bits value) \ | |
563 | { \ | |
564 | __le##bits __tmp = cpu_to_le##bits(value); \ | |
565 | return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ | |
566 | } | |
31db9f7c | 567 | |
95bc79d5 | 568 | TLV_PUT_DEFINE_INT(64) |
31db9f7c AB |
569 | |
570 | static int tlv_put_string(struct send_ctx *sctx, u16 attr, | |
571 | const char *str, int len) | |
572 | { | |
573 | if (len == -1) | |
574 | len = strlen(str); | |
575 | return tlv_put(sctx, attr, str, len); | |
576 | } | |
577 | ||
578 | static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, | |
579 | const u8 *uuid) | |
580 | { | |
581 | return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); | |
582 | } | |
583 | ||
31db9f7c AB |
584 | static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, |
585 | struct extent_buffer *eb, | |
586 | struct btrfs_timespec *ts) | |
587 | { | |
588 | struct btrfs_timespec bts; | |
589 | read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); | |
590 | return tlv_put(sctx, attr, &bts, sizeof(bts)); | |
591 | } | |
592 | ||
593 | ||
594 | #define TLV_PUT(sctx, attrtype, attrlen, data) \ | |
595 | do { \ | |
596 | ret = tlv_put(sctx, attrtype, attrlen, data); \ | |
597 | if (ret < 0) \ | |
598 | goto tlv_put_failure; \ | |
599 | } while (0) | |
600 | ||
601 | #define TLV_PUT_INT(sctx, attrtype, bits, value) \ | |
602 | do { \ | |
603 | ret = tlv_put_u##bits(sctx, attrtype, value); \ | |
604 | if (ret < 0) \ | |
605 | goto tlv_put_failure; \ | |
606 | } while (0) | |
607 | ||
608 | #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) | |
609 | #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) | |
610 | #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) | |
611 | #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) | |
612 | #define TLV_PUT_STRING(sctx, attrtype, str, len) \ | |
613 | do { \ | |
614 | ret = tlv_put_string(sctx, attrtype, str, len); \ | |
615 | if (ret < 0) \ | |
616 | goto tlv_put_failure; \ | |
617 | } while (0) | |
618 | #define TLV_PUT_PATH(sctx, attrtype, p) \ | |
619 | do { \ | |
620 | ret = tlv_put_string(sctx, attrtype, p->start, \ | |
621 | p->end - p->start); \ | |
622 | if (ret < 0) \ | |
623 | goto tlv_put_failure; \ | |
624 | } while(0) | |
625 | #define TLV_PUT_UUID(sctx, attrtype, uuid) \ | |
626 | do { \ | |
627 | ret = tlv_put_uuid(sctx, attrtype, uuid); \ | |
628 | if (ret < 0) \ | |
629 | goto tlv_put_failure; \ | |
630 | } while (0) | |
31db9f7c AB |
631 | #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ |
632 | do { \ | |
633 | ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ | |
634 | if (ret < 0) \ | |
635 | goto tlv_put_failure; \ | |
636 | } while (0) | |
637 | ||
638 | static int send_header(struct send_ctx *sctx) | |
639 | { | |
640 | struct btrfs_stream_header hdr; | |
641 | ||
642 | strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); | |
643 | hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); | |
644 | ||
1bcea355 AJ |
645 | return write_buf(sctx->send_filp, &hdr, sizeof(hdr), |
646 | &sctx->send_off); | |
31db9f7c AB |
647 | } |
648 | ||
649 | /* | |
650 | * For each command/item we want to send to userspace, we call this function. | |
651 | */ | |
652 | static int begin_cmd(struct send_ctx *sctx, int cmd) | |
653 | { | |
654 | struct btrfs_cmd_header *hdr; | |
655 | ||
fae7f21c | 656 | if (WARN_ON(!sctx->send_buf)) |
31db9f7c | 657 | return -EINVAL; |
31db9f7c AB |
658 | |
659 | BUG_ON(sctx->send_size); | |
660 | ||
661 | sctx->send_size += sizeof(*hdr); | |
662 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
663 | hdr->cmd = cpu_to_le16(cmd); | |
664 | ||
665 | return 0; | |
666 | } | |
667 | ||
668 | static int send_cmd(struct send_ctx *sctx) | |
669 | { | |
670 | int ret; | |
671 | struct btrfs_cmd_header *hdr; | |
672 | u32 crc; | |
673 | ||
674 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
675 | hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr)); | |
676 | hdr->crc = 0; | |
677 | ||
0b947aff | 678 | crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); |
31db9f7c AB |
679 | hdr->crc = cpu_to_le32(crc); |
680 | ||
1bcea355 AJ |
681 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, |
682 | &sctx->send_off); | |
31db9f7c AB |
683 | |
684 | sctx->total_send_size += sctx->send_size; | |
685 | sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size; | |
686 | sctx->send_size = 0; | |
687 | ||
688 | return ret; | |
689 | } | |
690 | ||
691 | /* | |
692 | * Sends a move instruction to user space | |
693 | */ | |
694 | static int send_rename(struct send_ctx *sctx, | |
695 | struct fs_path *from, struct fs_path *to) | |
696 | { | |
697 | int ret; | |
698 | ||
699 | verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start); | |
700 | ||
701 | ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); | |
702 | if (ret < 0) | |
703 | goto out; | |
704 | ||
705 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); | |
706 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); | |
707 | ||
708 | ret = send_cmd(sctx); | |
709 | ||
710 | tlv_put_failure: | |
711 | out: | |
712 | return ret; | |
713 | } | |
714 | ||
715 | /* | |
716 | * Sends a link instruction to user space | |
717 | */ | |
718 | static int send_link(struct send_ctx *sctx, | |
719 | struct fs_path *path, struct fs_path *lnk) | |
720 | { | |
721 | int ret; | |
722 | ||
723 | verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start); | |
724 | ||
725 | ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); | |
726 | if (ret < 0) | |
727 | goto out; | |
728 | ||
729 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
730 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); | |
731 | ||
732 | ret = send_cmd(sctx); | |
733 | ||
734 | tlv_put_failure: | |
735 | out: | |
736 | return ret; | |
737 | } | |
738 | ||
739 | /* | |
740 | * Sends an unlink instruction to user space | |
741 | */ | |
742 | static int send_unlink(struct send_ctx *sctx, struct fs_path *path) | |
743 | { | |
744 | int ret; | |
745 | ||
746 | verbose_printk("btrfs: send_unlink %s\n", path->start); | |
747 | ||
748 | ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); | |
749 | if (ret < 0) | |
750 | goto out; | |
751 | ||
752 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
753 | ||
754 | ret = send_cmd(sctx); | |
755 | ||
756 | tlv_put_failure: | |
757 | out: | |
758 | return ret; | |
759 | } | |
760 | ||
761 | /* | |
762 | * Sends a rmdir instruction to user space | |
763 | */ | |
764 | static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) | |
765 | { | |
766 | int ret; | |
767 | ||
768 | verbose_printk("btrfs: send_rmdir %s\n", path->start); | |
769 | ||
770 | ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); | |
771 | if (ret < 0) | |
772 | goto out; | |
773 | ||
774 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
775 | ||
776 | ret = send_cmd(sctx); | |
777 | ||
778 | tlv_put_failure: | |
779 | out: | |
780 | return ret; | |
781 | } | |
782 | ||
783 | /* | |
784 | * Helper function to retrieve some fields from an inode item. | |
785 | */ | |
3f8a18cc JB |
786 | static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, |
787 | u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, | |
788 | u64 *gid, u64 *rdev) | |
31db9f7c AB |
789 | { |
790 | int ret; | |
791 | struct btrfs_inode_item *ii; | |
792 | struct btrfs_key key; | |
31db9f7c AB |
793 | |
794 | key.objectid = ino; | |
795 | key.type = BTRFS_INODE_ITEM_KEY; | |
796 | key.offset = 0; | |
797 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
31db9f7c | 798 | if (ret) { |
3f8a18cc JB |
799 | if (ret > 0) |
800 | ret = -ENOENT; | |
801 | return ret; | |
31db9f7c AB |
802 | } |
803 | ||
804 | ii = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
805 | struct btrfs_inode_item); | |
806 | if (size) | |
807 | *size = btrfs_inode_size(path->nodes[0], ii); | |
808 | if (gen) | |
809 | *gen = btrfs_inode_generation(path->nodes[0], ii); | |
810 | if (mode) | |
811 | *mode = btrfs_inode_mode(path->nodes[0], ii); | |
812 | if (uid) | |
813 | *uid = btrfs_inode_uid(path->nodes[0], ii); | |
814 | if (gid) | |
815 | *gid = btrfs_inode_gid(path->nodes[0], ii); | |
85a7b33b AB |
816 | if (rdev) |
817 | *rdev = btrfs_inode_rdev(path->nodes[0], ii); | |
31db9f7c | 818 | |
3f8a18cc JB |
819 | return ret; |
820 | } | |
821 | ||
822 | static int get_inode_info(struct btrfs_root *root, | |
823 | u64 ino, u64 *size, u64 *gen, | |
824 | u64 *mode, u64 *uid, u64 *gid, | |
825 | u64 *rdev) | |
826 | { | |
827 | struct btrfs_path *path; | |
828 | int ret; | |
829 | ||
830 | path = alloc_path_for_send(); | |
831 | if (!path) | |
832 | return -ENOMEM; | |
833 | ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, | |
834 | rdev); | |
31db9f7c AB |
835 | btrfs_free_path(path); |
836 | return ret; | |
837 | } | |
838 | ||
839 | typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, | |
840 | struct fs_path *p, | |
841 | void *ctx); | |
842 | ||
843 | /* | |
96b5bd77 JS |
844 | * Helper function to iterate the entries in ONE btrfs_inode_ref or |
845 | * btrfs_inode_extref. | |
31db9f7c AB |
846 | * The iterate callback may return a non zero value to stop iteration. This can |
847 | * be a negative value for error codes or 1 to simply stop it. | |
848 | * | |
96b5bd77 | 849 | * path must point to the INODE_REF or INODE_EXTREF when called. |
31db9f7c | 850 | */ |
924794c9 | 851 | static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
852 | struct btrfs_key *found_key, int resolve, |
853 | iterate_inode_ref_t iterate, void *ctx) | |
854 | { | |
96b5bd77 | 855 | struct extent_buffer *eb = path->nodes[0]; |
31db9f7c AB |
856 | struct btrfs_item *item; |
857 | struct btrfs_inode_ref *iref; | |
96b5bd77 | 858 | struct btrfs_inode_extref *extref; |
31db9f7c AB |
859 | struct btrfs_path *tmp_path; |
860 | struct fs_path *p; | |
96b5bd77 | 861 | u32 cur = 0; |
31db9f7c | 862 | u32 total; |
96b5bd77 | 863 | int slot = path->slots[0]; |
31db9f7c AB |
864 | u32 name_len; |
865 | char *start; | |
866 | int ret = 0; | |
96b5bd77 | 867 | int num = 0; |
31db9f7c | 868 | int index; |
96b5bd77 JS |
869 | u64 dir; |
870 | unsigned long name_off; | |
871 | unsigned long elem_size; | |
872 | unsigned long ptr; | |
31db9f7c | 873 | |
924794c9 | 874 | p = fs_path_alloc_reversed(); |
31db9f7c AB |
875 | if (!p) |
876 | return -ENOMEM; | |
877 | ||
878 | tmp_path = alloc_path_for_send(); | |
879 | if (!tmp_path) { | |
924794c9 | 880 | fs_path_free(p); |
31db9f7c AB |
881 | return -ENOMEM; |
882 | } | |
883 | ||
31db9f7c | 884 | |
96b5bd77 JS |
885 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
886 | ptr = (unsigned long)btrfs_item_ptr(eb, slot, | |
887 | struct btrfs_inode_ref); | |
dd3cc16b | 888 | item = btrfs_item_nr(slot); |
96b5bd77 JS |
889 | total = btrfs_item_size(eb, item); |
890 | elem_size = sizeof(*iref); | |
891 | } else { | |
892 | ptr = btrfs_item_ptr_offset(eb, slot); | |
893 | total = btrfs_item_size_nr(eb, slot); | |
894 | elem_size = sizeof(*extref); | |
895 | } | |
896 | ||
31db9f7c AB |
897 | while (cur < total) { |
898 | fs_path_reset(p); | |
899 | ||
96b5bd77 JS |
900 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
901 | iref = (struct btrfs_inode_ref *)(ptr + cur); | |
902 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
903 | name_off = (unsigned long)(iref + 1); | |
904 | index = btrfs_inode_ref_index(eb, iref); | |
905 | dir = found_key->offset; | |
906 | } else { | |
907 | extref = (struct btrfs_inode_extref *)(ptr + cur); | |
908 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
909 | name_off = (unsigned long)&extref->name; | |
910 | index = btrfs_inode_extref_index(eb, extref); | |
911 | dir = btrfs_inode_extref_parent(eb, extref); | |
912 | } | |
913 | ||
31db9f7c | 914 | if (resolve) { |
96b5bd77 JS |
915 | start = btrfs_ref_to_path(root, tmp_path, name_len, |
916 | name_off, eb, dir, | |
917 | p->buf, p->buf_len); | |
31db9f7c AB |
918 | if (IS_ERR(start)) { |
919 | ret = PTR_ERR(start); | |
920 | goto out; | |
921 | } | |
922 | if (start < p->buf) { | |
923 | /* overflow , try again with larger buffer */ | |
924 | ret = fs_path_ensure_buf(p, | |
925 | p->buf_len + p->buf - start); | |
926 | if (ret < 0) | |
927 | goto out; | |
96b5bd77 JS |
928 | start = btrfs_ref_to_path(root, tmp_path, |
929 | name_len, name_off, | |
930 | eb, dir, | |
931 | p->buf, p->buf_len); | |
31db9f7c AB |
932 | if (IS_ERR(start)) { |
933 | ret = PTR_ERR(start); | |
934 | goto out; | |
935 | } | |
936 | BUG_ON(start < p->buf); | |
937 | } | |
938 | p->start = start; | |
939 | } else { | |
96b5bd77 JS |
940 | ret = fs_path_add_from_extent_buffer(p, eb, name_off, |
941 | name_len); | |
31db9f7c AB |
942 | if (ret < 0) |
943 | goto out; | |
944 | } | |
945 | ||
96b5bd77 JS |
946 | cur += elem_size + name_len; |
947 | ret = iterate(num, dir, index, p, ctx); | |
31db9f7c AB |
948 | if (ret) |
949 | goto out; | |
31db9f7c AB |
950 | num++; |
951 | } | |
952 | ||
953 | out: | |
954 | btrfs_free_path(tmp_path); | |
924794c9 | 955 | fs_path_free(p); |
31db9f7c AB |
956 | return ret; |
957 | } | |
958 | ||
959 | typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, | |
960 | const char *name, int name_len, | |
961 | const char *data, int data_len, | |
962 | u8 type, void *ctx); | |
963 | ||
964 | /* | |
965 | * Helper function to iterate the entries in ONE btrfs_dir_item. | |
966 | * The iterate callback may return a non zero value to stop iteration. This can | |
967 | * be a negative value for error codes or 1 to simply stop it. | |
968 | * | |
969 | * path must point to the dir item when called. | |
970 | */ | |
924794c9 | 971 | static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
972 | struct btrfs_key *found_key, |
973 | iterate_dir_item_t iterate, void *ctx) | |
974 | { | |
975 | int ret = 0; | |
976 | struct extent_buffer *eb; | |
977 | struct btrfs_item *item; | |
978 | struct btrfs_dir_item *di; | |
31db9f7c AB |
979 | struct btrfs_key di_key; |
980 | char *buf = NULL; | |
7e3ae33e | 981 | int buf_len; |
31db9f7c AB |
982 | u32 name_len; |
983 | u32 data_len; | |
984 | u32 cur; | |
985 | u32 len; | |
986 | u32 total; | |
987 | int slot; | |
988 | int num; | |
989 | u8 type; | |
990 | ||
4395e0c4 FM |
991 | /* |
992 | * Start with a small buffer (1 page). If later we end up needing more | |
993 | * space, which can happen for xattrs on a fs with a leaf size greater | |
994 | * then the page size, attempt to increase the buffer. Typically xattr | |
995 | * values are small. | |
996 | */ | |
997 | buf_len = PATH_MAX; | |
e780b0d1 | 998 | buf = kmalloc(buf_len, GFP_KERNEL); |
31db9f7c AB |
999 | if (!buf) { |
1000 | ret = -ENOMEM; | |
1001 | goto out; | |
1002 | } | |
1003 | ||
31db9f7c AB |
1004 | eb = path->nodes[0]; |
1005 | slot = path->slots[0]; | |
dd3cc16b | 1006 | item = btrfs_item_nr(slot); |
31db9f7c AB |
1007 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
1008 | cur = 0; | |
1009 | len = 0; | |
1010 | total = btrfs_item_size(eb, item); | |
1011 | ||
1012 | num = 0; | |
1013 | while (cur < total) { | |
1014 | name_len = btrfs_dir_name_len(eb, di); | |
1015 | data_len = btrfs_dir_data_len(eb, di); | |
1016 | type = btrfs_dir_type(eb, di); | |
1017 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1018 | ||
7e3ae33e FM |
1019 | if (type == BTRFS_FT_XATTR) { |
1020 | if (name_len > XATTR_NAME_MAX) { | |
1021 | ret = -ENAMETOOLONG; | |
1022 | goto out; | |
1023 | } | |
4395e0c4 | 1024 | if (name_len + data_len > BTRFS_MAX_XATTR_SIZE(root)) { |
7e3ae33e FM |
1025 | ret = -E2BIG; |
1026 | goto out; | |
1027 | } | |
1028 | } else { | |
1029 | /* | |
1030 | * Path too long | |
1031 | */ | |
4395e0c4 | 1032 | if (name_len + data_len > PATH_MAX) { |
7e3ae33e FM |
1033 | ret = -ENAMETOOLONG; |
1034 | goto out; | |
1035 | } | |
31db9f7c AB |
1036 | } |
1037 | ||
4395e0c4 FM |
1038 | if (name_len + data_len > buf_len) { |
1039 | buf_len = name_len + data_len; | |
1040 | if (is_vmalloc_addr(buf)) { | |
1041 | vfree(buf); | |
1042 | buf = NULL; | |
1043 | } else { | |
1044 | char *tmp = krealloc(buf, buf_len, | |
e780b0d1 | 1045 | GFP_KERNEL | __GFP_NOWARN); |
4395e0c4 FM |
1046 | |
1047 | if (!tmp) | |
1048 | kfree(buf); | |
1049 | buf = tmp; | |
1050 | } | |
1051 | if (!buf) { | |
1052 | buf = vmalloc(buf_len); | |
1053 | if (!buf) { | |
1054 | ret = -ENOMEM; | |
1055 | goto out; | |
1056 | } | |
1057 | } | |
1058 | } | |
1059 | ||
31db9f7c AB |
1060 | read_extent_buffer(eb, buf, (unsigned long)(di + 1), |
1061 | name_len + data_len); | |
1062 | ||
1063 | len = sizeof(*di) + name_len + data_len; | |
1064 | di = (struct btrfs_dir_item *)((char *)di + len); | |
1065 | cur += len; | |
1066 | ||
1067 | ret = iterate(num, &di_key, buf, name_len, buf + name_len, | |
1068 | data_len, type, ctx); | |
1069 | if (ret < 0) | |
1070 | goto out; | |
1071 | if (ret) { | |
1072 | ret = 0; | |
1073 | goto out; | |
1074 | } | |
1075 | ||
1076 | num++; | |
1077 | } | |
1078 | ||
1079 | out: | |
4395e0c4 | 1080 | kvfree(buf); |
31db9f7c AB |
1081 | return ret; |
1082 | } | |
1083 | ||
1084 | static int __copy_first_ref(int num, u64 dir, int index, | |
1085 | struct fs_path *p, void *ctx) | |
1086 | { | |
1087 | int ret; | |
1088 | struct fs_path *pt = ctx; | |
1089 | ||
1090 | ret = fs_path_copy(pt, p); | |
1091 | if (ret < 0) | |
1092 | return ret; | |
1093 | ||
1094 | /* we want the first only */ | |
1095 | return 1; | |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * Retrieve the first path of an inode. If an inode has more then one | |
1100 | * ref/hardlink, this is ignored. | |
1101 | */ | |
924794c9 | 1102 | static int get_inode_path(struct btrfs_root *root, |
31db9f7c AB |
1103 | u64 ino, struct fs_path *path) |
1104 | { | |
1105 | int ret; | |
1106 | struct btrfs_key key, found_key; | |
1107 | struct btrfs_path *p; | |
1108 | ||
1109 | p = alloc_path_for_send(); | |
1110 | if (!p) | |
1111 | return -ENOMEM; | |
1112 | ||
1113 | fs_path_reset(path); | |
1114 | ||
1115 | key.objectid = ino; | |
1116 | key.type = BTRFS_INODE_REF_KEY; | |
1117 | key.offset = 0; | |
1118 | ||
1119 | ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); | |
1120 | if (ret < 0) | |
1121 | goto out; | |
1122 | if (ret) { | |
1123 | ret = 1; | |
1124 | goto out; | |
1125 | } | |
1126 | btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); | |
1127 | if (found_key.objectid != ino || | |
96b5bd77 JS |
1128 | (found_key.type != BTRFS_INODE_REF_KEY && |
1129 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1130 | ret = -ENOENT; |
1131 | goto out; | |
1132 | } | |
1133 | ||
924794c9 TI |
1134 | ret = iterate_inode_ref(root, p, &found_key, 1, |
1135 | __copy_first_ref, path); | |
31db9f7c AB |
1136 | if (ret < 0) |
1137 | goto out; | |
1138 | ret = 0; | |
1139 | ||
1140 | out: | |
1141 | btrfs_free_path(p); | |
1142 | return ret; | |
1143 | } | |
1144 | ||
1145 | struct backref_ctx { | |
1146 | struct send_ctx *sctx; | |
1147 | ||
3f8a18cc | 1148 | struct btrfs_path *path; |
31db9f7c AB |
1149 | /* number of total found references */ |
1150 | u64 found; | |
1151 | ||
1152 | /* | |
1153 | * used for clones found in send_root. clones found behind cur_objectid | |
1154 | * and cur_offset are not considered as allowed clones. | |
1155 | */ | |
1156 | u64 cur_objectid; | |
1157 | u64 cur_offset; | |
1158 | ||
1159 | /* may be truncated in case it's the last extent in a file */ | |
1160 | u64 extent_len; | |
1161 | ||
619d8c4e FM |
1162 | /* data offset in the file extent item */ |
1163 | u64 data_offset; | |
1164 | ||
31db9f7c | 1165 | /* Just to check for bugs in backref resolving */ |
ee849c04 | 1166 | int found_itself; |
31db9f7c AB |
1167 | }; |
1168 | ||
1169 | static int __clone_root_cmp_bsearch(const void *key, const void *elt) | |
1170 | { | |
995e01b7 | 1171 | u64 root = (u64)(uintptr_t)key; |
31db9f7c AB |
1172 | struct clone_root *cr = (struct clone_root *)elt; |
1173 | ||
1174 | if (root < cr->root->objectid) | |
1175 | return -1; | |
1176 | if (root > cr->root->objectid) | |
1177 | return 1; | |
1178 | return 0; | |
1179 | } | |
1180 | ||
1181 | static int __clone_root_cmp_sort(const void *e1, const void *e2) | |
1182 | { | |
1183 | struct clone_root *cr1 = (struct clone_root *)e1; | |
1184 | struct clone_root *cr2 = (struct clone_root *)e2; | |
1185 | ||
1186 | if (cr1->root->objectid < cr2->root->objectid) | |
1187 | return -1; | |
1188 | if (cr1->root->objectid > cr2->root->objectid) | |
1189 | return 1; | |
1190 | return 0; | |
1191 | } | |
1192 | ||
1193 | /* | |
1194 | * Called for every backref that is found for the current extent. | |
766702ef | 1195 | * Results are collected in sctx->clone_roots->ino/offset/found_refs |
31db9f7c AB |
1196 | */ |
1197 | static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) | |
1198 | { | |
1199 | struct backref_ctx *bctx = ctx_; | |
1200 | struct clone_root *found; | |
1201 | int ret; | |
1202 | u64 i_size; | |
1203 | ||
1204 | /* First check if the root is in the list of accepted clone sources */ | |
995e01b7 | 1205 | found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, |
31db9f7c AB |
1206 | bctx->sctx->clone_roots_cnt, |
1207 | sizeof(struct clone_root), | |
1208 | __clone_root_cmp_bsearch); | |
1209 | if (!found) | |
1210 | return 0; | |
1211 | ||
1212 | if (found->root == bctx->sctx->send_root && | |
1213 | ino == bctx->cur_objectid && | |
1214 | offset == bctx->cur_offset) { | |
ee849c04 | 1215 | bctx->found_itself = 1; |
31db9f7c AB |
1216 | } |
1217 | ||
1218 | /* | |
766702ef | 1219 | * There are inodes that have extents that lie behind its i_size. Don't |
31db9f7c AB |
1220 | * accept clones from these extents. |
1221 | */ | |
3f8a18cc JB |
1222 | ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL, |
1223 | NULL, NULL, NULL); | |
1224 | btrfs_release_path(bctx->path); | |
31db9f7c AB |
1225 | if (ret < 0) |
1226 | return ret; | |
1227 | ||
619d8c4e | 1228 | if (offset + bctx->data_offset + bctx->extent_len > i_size) |
31db9f7c AB |
1229 | return 0; |
1230 | ||
1231 | /* | |
1232 | * Make sure we don't consider clones from send_root that are | |
1233 | * behind the current inode/offset. | |
1234 | */ | |
1235 | if (found->root == bctx->sctx->send_root) { | |
1236 | /* | |
1237 | * TODO for the moment we don't accept clones from the inode | |
1238 | * that is currently send. We may change this when | |
1239 | * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same | |
1240 | * file. | |
1241 | */ | |
1242 | if (ino >= bctx->cur_objectid) | |
1243 | return 0; | |
e938c8ad AB |
1244 | #if 0 |
1245 | if (ino > bctx->cur_objectid) | |
1246 | return 0; | |
1247 | if (offset + bctx->extent_len > bctx->cur_offset) | |
31db9f7c | 1248 | return 0; |
e938c8ad | 1249 | #endif |
31db9f7c AB |
1250 | } |
1251 | ||
1252 | bctx->found++; | |
1253 | found->found_refs++; | |
1254 | if (ino < found->ino) { | |
1255 | found->ino = ino; | |
1256 | found->offset = offset; | |
1257 | } else if (found->ino == ino) { | |
1258 | /* | |
1259 | * same extent found more then once in the same file. | |
1260 | */ | |
1261 | if (found->offset > offset + bctx->extent_len) | |
1262 | found->offset = offset; | |
1263 | } | |
1264 | ||
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | /* | |
766702ef AB |
1269 | * Given an inode, offset and extent item, it finds a good clone for a clone |
1270 | * instruction. Returns -ENOENT when none could be found. The function makes | |
1271 | * sure that the returned clone is usable at the point where sending is at the | |
1272 | * moment. This means, that no clones are accepted which lie behind the current | |
1273 | * inode+offset. | |
1274 | * | |
31db9f7c AB |
1275 | * path must point to the extent item when called. |
1276 | */ | |
1277 | static int find_extent_clone(struct send_ctx *sctx, | |
1278 | struct btrfs_path *path, | |
1279 | u64 ino, u64 data_offset, | |
1280 | u64 ino_size, | |
1281 | struct clone_root **found) | |
1282 | { | |
1283 | int ret; | |
1284 | int extent_type; | |
1285 | u64 logical; | |
74dd17fb | 1286 | u64 disk_byte; |
31db9f7c AB |
1287 | u64 num_bytes; |
1288 | u64 extent_item_pos; | |
69917e43 | 1289 | u64 flags = 0; |
31db9f7c AB |
1290 | struct btrfs_file_extent_item *fi; |
1291 | struct extent_buffer *eb = path->nodes[0]; | |
35075bb0 | 1292 | struct backref_ctx *backref_ctx = NULL; |
31db9f7c AB |
1293 | struct clone_root *cur_clone_root; |
1294 | struct btrfs_key found_key; | |
1295 | struct btrfs_path *tmp_path; | |
74dd17fb | 1296 | int compressed; |
31db9f7c AB |
1297 | u32 i; |
1298 | ||
1299 | tmp_path = alloc_path_for_send(); | |
1300 | if (!tmp_path) | |
1301 | return -ENOMEM; | |
1302 | ||
3f8a18cc JB |
1303 | /* We only use this path under the commit sem */ |
1304 | tmp_path->need_commit_sem = 0; | |
1305 | ||
e780b0d1 | 1306 | backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL); |
35075bb0 AB |
1307 | if (!backref_ctx) { |
1308 | ret = -ENOMEM; | |
1309 | goto out; | |
1310 | } | |
1311 | ||
3f8a18cc JB |
1312 | backref_ctx->path = tmp_path; |
1313 | ||
31db9f7c AB |
1314 | if (data_offset >= ino_size) { |
1315 | /* | |
1316 | * There may be extents that lie behind the file's size. | |
1317 | * I at least had this in combination with snapshotting while | |
1318 | * writing large files. | |
1319 | */ | |
1320 | ret = 0; | |
1321 | goto out; | |
1322 | } | |
1323 | ||
1324 | fi = btrfs_item_ptr(eb, path->slots[0], | |
1325 | struct btrfs_file_extent_item); | |
1326 | extent_type = btrfs_file_extent_type(eb, fi); | |
1327 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1328 | ret = -ENOENT; | |
1329 | goto out; | |
1330 | } | |
74dd17fb | 1331 | compressed = btrfs_file_extent_compression(eb, fi); |
31db9f7c AB |
1332 | |
1333 | num_bytes = btrfs_file_extent_num_bytes(eb, fi); | |
74dd17fb CM |
1334 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
1335 | if (disk_byte == 0) { | |
31db9f7c AB |
1336 | ret = -ENOENT; |
1337 | goto out; | |
1338 | } | |
74dd17fb | 1339 | logical = disk_byte + btrfs_file_extent_offset(eb, fi); |
31db9f7c | 1340 | |
9e351cc8 | 1341 | down_read(&sctx->send_root->fs_info->commit_root_sem); |
69917e43 LB |
1342 | ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path, |
1343 | &found_key, &flags); | |
9e351cc8 | 1344 | up_read(&sctx->send_root->fs_info->commit_root_sem); |
31db9f7c AB |
1345 | btrfs_release_path(tmp_path); |
1346 | ||
1347 | if (ret < 0) | |
1348 | goto out; | |
69917e43 | 1349 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
31db9f7c AB |
1350 | ret = -EIO; |
1351 | goto out; | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * Setup the clone roots. | |
1356 | */ | |
1357 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1358 | cur_clone_root = sctx->clone_roots + i; | |
1359 | cur_clone_root->ino = (u64)-1; | |
1360 | cur_clone_root->offset = 0; | |
1361 | cur_clone_root->found_refs = 0; | |
1362 | } | |
1363 | ||
35075bb0 AB |
1364 | backref_ctx->sctx = sctx; |
1365 | backref_ctx->found = 0; | |
1366 | backref_ctx->cur_objectid = ino; | |
1367 | backref_ctx->cur_offset = data_offset; | |
1368 | backref_ctx->found_itself = 0; | |
1369 | backref_ctx->extent_len = num_bytes; | |
619d8c4e FM |
1370 | /* |
1371 | * For non-compressed extents iterate_extent_inodes() gives us extent | |
1372 | * offsets that already take into account the data offset, but not for | |
1373 | * compressed extents, since the offset is logical and not relative to | |
1374 | * the physical extent locations. We must take this into account to | |
1375 | * avoid sending clone offsets that go beyond the source file's size, | |
1376 | * which would result in the clone ioctl failing with -EINVAL on the | |
1377 | * receiving end. | |
1378 | */ | |
1379 | if (compressed == BTRFS_COMPRESS_NONE) | |
1380 | backref_ctx->data_offset = 0; | |
1381 | else | |
1382 | backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi); | |
31db9f7c AB |
1383 | |
1384 | /* | |
1385 | * The last extent of a file may be too large due to page alignment. | |
1386 | * We need to adjust extent_len in this case so that the checks in | |
1387 | * __iterate_backrefs work. | |
1388 | */ | |
1389 | if (data_offset + num_bytes >= ino_size) | |
35075bb0 | 1390 | backref_ctx->extent_len = ino_size - data_offset; |
31db9f7c AB |
1391 | |
1392 | /* | |
1393 | * Now collect all backrefs. | |
1394 | */ | |
74dd17fb CM |
1395 | if (compressed == BTRFS_COMPRESS_NONE) |
1396 | extent_item_pos = logical - found_key.objectid; | |
1397 | else | |
1398 | extent_item_pos = 0; | |
31db9f7c AB |
1399 | ret = iterate_extent_inodes(sctx->send_root->fs_info, |
1400 | found_key.objectid, extent_item_pos, 1, | |
35075bb0 | 1401 | __iterate_backrefs, backref_ctx); |
74dd17fb | 1402 | |
31db9f7c AB |
1403 | if (ret < 0) |
1404 | goto out; | |
1405 | ||
35075bb0 | 1406 | if (!backref_ctx->found_itself) { |
31db9f7c AB |
1407 | /* found a bug in backref code? */ |
1408 | ret = -EIO; | |
efe120a0 | 1409 | btrfs_err(sctx->send_root->fs_info, "did not find backref in " |
31db9f7c | 1410 | "send_root. inode=%llu, offset=%llu, " |
351fd353 | 1411 | "disk_byte=%llu found extent=%llu", |
74dd17fb | 1412 | ino, data_offset, disk_byte, found_key.objectid); |
31db9f7c AB |
1413 | goto out; |
1414 | } | |
1415 | ||
1416 | verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, " | |
1417 | "ino=%llu, " | |
1418 | "num_bytes=%llu, logical=%llu\n", | |
1419 | data_offset, ino, num_bytes, logical); | |
1420 | ||
35075bb0 | 1421 | if (!backref_ctx->found) |
31db9f7c AB |
1422 | verbose_printk("btrfs: no clones found\n"); |
1423 | ||
1424 | cur_clone_root = NULL; | |
1425 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1426 | if (sctx->clone_roots[i].found_refs) { | |
1427 | if (!cur_clone_root) | |
1428 | cur_clone_root = sctx->clone_roots + i; | |
1429 | else if (sctx->clone_roots[i].root == sctx->send_root) | |
1430 | /* prefer clones from send_root over others */ | |
1431 | cur_clone_root = sctx->clone_roots + i; | |
31db9f7c AB |
1432 | } |
1433 | ||
1434 | } | |
1435 | ||
1436 | if (cur_clone_root) { | |
1437 | *found = cur_clone_root; | |
1438 | ret = 0; | |
1439 | } else { | |
1440 | ret = -ENOENT; | |
1441 | } | |
1442 | ||
1443 | out: | |
1444 | btrfs_free_path(tmp_path); | |
35075bb0 | 1445 | kfree(backref_ctx); |
31db9f7c AB |
1446 | return ret; |
1447 | } | |
1448 | ||
924794c9 | 1449 | static int read_symlink(struct btrfs_root *root, |
31db9f7c AB |
1450 | u64 ino, |
1451 | struct fs_path *dest) | |
1452 | { | |
1453 | int ret; | |
1454 | struct btrfs_path *path; | |
1455 | struct btrfs_key key; | |
1456 | struct btrfs_file_extent_item *ei; | |
1457 | u8 type; | |
1458 | u8 compression; | |
1459 | unsigned long off; | |
1460 | int len; | |
1461 | ||
1462 | path = alloc_path_for_send(); | |
1463 | if (!path) | |
1464 | return -ENOMEM; | |
1465 | ||
1466 | key.objectid = ino; | |
1467 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1468 | key.offset = 0; | |
1469 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1470 | if (ret < 0) | |
1471 | goto out; | |
a879719b FM |
1472 | if (ret) { |
1473 | /* | |
1474 | * An empty symlink inode. Can happen in rare error paths when | |
1475 | * creating a symlink (transaction committed before the inode | |
1476 | * eviction handler removed the symlink inode items and a crash | |
1477 | * happened in between or the subvol was snapshoted in between). | |
1478 | * Print an informative message to dmesg/syslog so that the user | |
1479 | * can delete the symlink. | |
1480 | */ | |
1481 | btrfs_err(root->fs_info, | |
1482 | "Found empty symlink inode %llu at root %llu", | |
1483 | ino, root->root_key.objectid); | |
1484 | ret = -EIO; | |
1485 | goto out; | |
1486 | } | |
31db9f7c AB |
1487 | |
1488 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1489 | struct btrfs_file_extent_item); | |
1490 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
1491 | compression = btrfs_file_extent_compression(path->nodes[0], ei); | |
1492 | BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); | |
1493 | BUG_ON(compression); | |
1494 | ||
1495 | off = btrfs_file_extent_inline_start(ei); | |
514ac8ad | 1496 | len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei); |
31db9f7c AB |
1497 | |
1498 | ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); | |
31db9f7c AB |
1499 | |
1500 | out: | |
1501 | btrfs_free_path(path); | |
1502 | return ret; | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Helper function to generate a file name that is unique in the root of | |
1507 | * send_root and parent_root. This is used to generate names for orphan inodes. | |
1508 | */ | |
1509 | static int gen_unique_name(struct send_ctx *sctx, | |
1510 | u64 ino, u64 gen, | |
1511 | struct fs_path *dest) | |
1512 | { | |
1513 | int ret = 0; | |
1514 | struct btrfs_path *path; | |
1515 | struct btrfs_dir_item *di; | |
1516 | char tmp[64]; | |
1517 | int len; | |
1518 | u64 idx = 0; | |
1519 | ||
1520 | path = alloc_path_for_send(); | |
1521 | if (!path) | |
1522 | return -ENOMEM; | |
1523 | ||
1524 | while (1) { | |
f74b86d8 | 1525 | len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", |
31db9f7c | 1526 | ino, gen, idx); |
64792f25 | 1527 | ASSERT(len < sizeof(tmp)); |
31db9f7c AB |
1528 | |
1529 | di = btrfs_lookup_dir_item(NULL, sctx->send_root, | |
1530 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1531 | tmp, strlen(tmp), 0); | |
1532 | btrfs_release_path(path); | |
1533 | if (IS_ERR(di)) { | |
1534 | ret = PTR_ERR(di); | |
1535 | goto out; | |
1536 | } | |
1537 | if (di) { | |
1538 | /* not unique, try again */ | |
1539 | idx++; | |
1540 | continue; | |
1541 | } | |
1542 | ||
1543 | if (!sctx->parent_root) { | |
1544 | /* unique */ | |
1545 | ret = 0; | |
1546 | break; | |
1547 | } | |
1548 | ||
1549 | di = btrfs_lookup_dir_item(NULL, sctx->parent_root, | |
1550 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1551 | tmp, strlen(tmp), 0); | |
1552 | btrfs_release_path(path); | |
1553 | if (IS_ERR(di)) { | |
1554 | ret = PTR_ERR(di); | |
1555 | goto out; | |
1556 | } | |
1557 | if (di) { | |
1558 | /* not unique, try again */ | |
1559 | idx++; | |
1560 | continue; | |
1561 | } | |
1562 | /* unique */ | |
1563 | break; | |
1564 | } | |
1565 | ||
1566 | ret = fs_path_add(dest, tmp, strlen(tmp)); | |
1567 | ||
1568 | out: | |
1569 | btrfs_free_path(path); | |
1570 | return ret; | |
1571 | } | |
1572 | ||
1573 | enum inode_state { | |
1574 | inode_state_no_change, | |
1575 | inode_state_will_create, | |
1576 | inode_state_did_create, | |
1577 | inode_state_will_delete, | |
1578 | inode_state_did_delete, | |
1579 | }; | |
1580 | ||
1581 | static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) | |
1582 | { | |
1583 | int ret; | |
1584 | int left_ret; | |
1585 | int right_ret; | |
1586 | u64 left_gen; | |
1587 | u64 right_gen; | |
1588 | ||
1589 | ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, | |
85a7b33b | 1590 | NULL, NULL); |
31db9f7c AB |
1591 | if (ret < 0 && ret != -ENOENT) |
1592 | goto out; | |
1593 | left_ret = ret; | |
1594 | ||
1595 | if (!sctx->parent_root) { | |
1596 | right_ret = -ENOENT; | |
1597 | } else { | |
1598 | ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, | |
85a7b33b | 1599 | NULL, NULL, NULL, NULL); |
31db9f7c AB |
1600 | if (ret < 0 && ret != -ENOENT) |
1601 | goto out; | |
1602 | right_ret = ret; | |
1603 | } | |
1604 | ||
1605 | if (!left_ret && !right_ret) { | |
e938c8ad | 1606 | if (left_gen == gen && right_gen == gen) { |
31db9f7c | 1607 | ret = inode_state_no_change; |
e938c8ad | 1608 | } else if (left_gen == gen) { |
31db9f7c AB |
1609 | if (ino < sctx->send_progress) |
1610 | ret = inode_state_did_create; | |
1611 | else | |
1612 | ret = inode_state_will_create; | |
1613 | } else if (right_gen == gen) { | |
1614 | if (ino < sctx->send_progress) | |
1615 | ret = inode_state_did_delete; | |
1616 | else | |
1617 | ret = inode_state_will_delete; | |
1618 | } else { | |
1619 | ret = -ENOENT; | |
1620 | } | |
1621 | } else if (!left_ret) { | |
1622 | if (left_gen == gen) { | |
1623 | if (ino < sctx->send_progress) | |
1624 | ret = inode_state_did_create; | |
1625 | else | |
1626 | ret = inode_state_will_create; | |
1627 | } else { | |
1628 | ret = -ENOENT; | |
1629 | } | |
1630 | } else if (!right_ret) { | |
1631 | if (right_gen == gen) { | |
1632 | if (ino < sctx->send_progress) | |
1633 | ret = inode_state_did_delete; | |
1634 | else | |
1635 | ret = inode_state_will_delete; | |
1636 | } else { | |
1637 | ret = -ENOENT; | |
1638 | } | |
1639 | } else { | |
1640 | ret = -ENOENT; | |
1641 | } | |
1642 | ||
1643 | out: | |
1644 | return ret; | |
1645 | } | |
1646 | ||
1647 | static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) | |
1648 | { | |
1649 | int ret; | |
1650 | ||
1651 | ret = get_cur_inode_state(sctx, ino, gen); | |
1652 | if (ret < 0) | |
1653 | goto out; | |
1654 | ||
1655 | if (ret == inode_state_no_change || | |
1656 | ret == inode_state_did_create || | |
1657 | ret == inode_state_will_delete) | |
1658 | ret = 1; | |
1659 | else | |
1660 | ret = 0; | |
1661 | ||
1662 | out: | |
1663 | return ret; | |
1664 | } | |
1665 | ||
1666 | /* | |
1667 | * Helper function to lookup a dir item in a dir. | |
1668 | */ | |
1669 | static int lookup_dir_item_inode(struct btrfs_root *root, | |
1670 | u64 dir, const char *name, int name_len, | |
1671 | u64 *found_inode, | |
1672 | u8 *found_type) | |
1673 | { | |
1674 | int ret = 0; | |
1675 | struct btrfs_dir_item *di; | |
1676 | struct btrfs_key key; | |
1677 | struct btrfs_path *path; | |
1678 | ||
1679 | path = alloc_path_for_send(); | |
1680 | if (!path) | |
1681 | return -ENOMEM; | |
1682 | ||
1683 | di = btrfs_lookup_dir_item(NULL, root, path, | |
1684 | dir, name, name_len, 0); | |
1685 | if (!di) { | |
1686 | ret = -ENOENT; | |
1687 | goto out; | |
1688 | } | |
1689 | if (IS_ERR(di)) { | |
1690 | ret = PTR_ERR(di); | |
1691 | goto out; | |
1692 | } | |
1693 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
1af56070 FM |
1694 | if (key.type == BTRFS_ROOT_ITEM_KEY) { |
1695 | ret = -ENOENT; | |
1696 | goto out; | |
1697 | } | |
31db9f7c AB |
1698 | *found_inode = key.objectid; |
1699 | *found_type = btrfs_dir_type(path->nodes[0], di); | |
1700 | ||
1701 | out: | |
1702 | btrfs_free_path(path); | |
1703 | return ret; | |
1704 | } | |
1705 | ||
766702ef AB |
1706 | /* |
1707 | * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, | |
1708 | * generation of the parent dir and the name of the dir entry. | |
1709 | */ | |
924794c9 | 1710 | static int get_first_ref(struct btrfs_root *root, u64 ino, |
31db9f7c AB |
1711 | u64 *dir, u64 *dir_gen, struct fs_path *name) |
1712 | { | |
1713 | int ret; | |
1714 | struct btrfs_key key; | |
1715 | struct btrfs_key found_key; | |
1716 | struct btrfs_path *path; | |
31db9f7c | 1717 | int len; |
96b5bd77 | 1718 | u64 parent_dir; |
31db9f7c AB |
1719 | |
1720 | path = alloc_path_for_send(); | |
1721 | if (!path) | |
1722 | return -ENOMEM; | |
1723 | ||
1724 | key.objectid = ino; | |
1725 | key.type = BTRFS_INODE_REF_KEY; | |
1726 | key.offset = 0; | |
1727 | ||
1728 | ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); | |
1729 | if (ret < 0) | |
1730 | goto out; | |
1731 | if (!ret) | |
1732 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1733 | path->slots[0]); | |
96b5bd77 JS |
1734 | if (ret || found_key.objectid != ino || |
1735 | (found_key.type != BTRFS_INODE_REF_KEY && | |
1736 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1737 | ret = -ENOENT; |
1738 | goto out; | |
1739 | } | |
1740 | ||
51a60253 | 1741 | if (found_key.type == BTRFS_INODE_REF_KEY) { |
96b5bd77 JS |
1742 | struct btrfs_inode_ref *iref; |
1743 | iref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1744 | struct btrfs_inode_ref); | |
1745 | len = btrfs_inode_ref_name_len(path->nodes[0], iref); | |
1746 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1747 | (unsigned long)(iref + 1), | |
1748 | len); | |
1749 | parent_dir = found_key.offset; | |
1750 | } else { | |
1751 | struct btrfs_inode_extref *extref; | |
1752 | extref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1753 | struct btrfs_inode_extref); | |
1754 | len = btrfs_inode_extref_name_len(path->nodes[0], extref); | |
1755 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1756 | (unsigned long)&extref->name, len); | |
1757 | parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); | |
1758 | } | |
31db9f7c AB |
1759 | if (ret < 0) |
1760 | goto out; | |
1761 | btrfs_release_path(path); | |
1762 | ||
b46ab97b FM |
1763 | if (dir_gen) { |
1764 | ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, | |
1765 | NULL, NULL, NULL); | |
1766 | if (ret < 0) | |
1767 | goto out; | |
1768 | } | |
31db9f7c | 1769 | |
96b5bd77 | 1770 | *dir = parent_dir; |
31db9f7c AB |
1771 | |
1772 | out: | |
1773 | btrfs_free_path(path); | |
1774 | return ret; | |
1775 | } | |
1776 | ||
924794c9 | 1777 | static int is_first_ref(struct btrfs_root *root, |
31db9f7c AB |
1778 | u64 ino, u64 dir, |
1779 | const char *name, int name_len) | |
1780 | { | |
1781 | int ret; | |
1782 | struct fs_path *tmp_name; | |
1783 | u64 tmp_dir; | |
31db9f7c | 1784 | |
924794c9 | 1785 | tmp_name = fs_path_alloc(); |
31db9f7c AB |
1786 | if (!tmp_name) |
1787 | return -ENOMEM; | |
1788 | ||
b46ab97b | 1789 | ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name); |
31db9f7c AB |
1790 | if (ret < 0) |
1791 | goto out; | |
1792 | ||
b9291aff | 1793 | if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { |
31db9f7c AB |
1794 | ret = 0; |
1795 | goto out; | |
1796 | } | |
1797 | ||
e938c8ad | 1798 | ret = !memcmp(tmp_name->start, name, name_len); |
31db9f7c AB |
1799 | |
1800 | out: | |
924794c9 | 1801 | fs_path_free(tmp_name); |
31db9f7c AB |
1802 | return ret; |
1803 | } | |
1804 | ||
766702ef AB |
1805 | /* |
1806 | * Used by process_recorded_refs to determine if a new ref would overwrite an | |
1807 | * already existing ref. In case it detects an overwrite, it returns the | |
1808 | * inode/gen in who_ino/who_gen. | |
1809 | * When an overwrite is detected, process_recorded_refs does proper orphanizing | |
1810 | * to make sure later references to the overwritten inode are possible. | |
1811 | * Orphanizing is however only required for the first ref of an inode. | |
1812 | * process_recorded_refs does an additional is_first_ref check to see if | |
1813 | * orphanizing is really required. | |
1814 | */ | |
31db9f7c AB |
1815 | static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
1816 | const char *name, int name_len, | |
1817 | u64 *who_ino, u64 *who_gen) | |
1818 | { | |
1819 | int ret = 0; | |
ebdad913 | 1820 | u64 gen; |
31db9f7c AB |
1821 | u64 other_inode = 0; |
1822 | u8 other_type = 0; | |
1823 | ||
1824 | if (!sctx->parent_root) | |
1825 | goto out; | |
1826 | ||
1827 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1828 | if (ret <= 0) | |
1829 | goto out; | |
1830 | ||
ebdad913 JB |
1831 | /* |
1832 | * If we have a parent root we need to verify that the parent dir was | |
01327610 | 1833 | * not deleted and then re-created, if it was then we have no overwrite |
ebdad913 JB |
1834 | * and we can just unlink this entry. |
1835 | */ | |
1836 | if (sctx->parent_root) { | |
1837 | ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, | |
1838 | NULL, NULL, NULL); | |
1839 | if (ret < 0 && ret != -ENOENT) | |
1840 | goto out; | |
1841 | if (ret) { | |
1842 | ret = 0; | |
1843 | goto out; | |
1844 | } | |
1845 | if (gen != dir_gen) | |
1846 | goto out; | |
1847 | } | |
1848 | ||
31db9f7c AB |
1849 | ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, |
1850 | &other_inode, &other_type); | |
1851 | if (ret < 0 && ret != -ENOENT) | |
1852 | goto out; | |
1853 | if (ret) { | |
1854 | ret = 0; | |
1855 | goto out; | |
1856 | } | |
1857 | ||
766702ef AB |
1858 | /* |
1859 | * Check if the overwritten ref was already processed. If yes, the ref | |
1860 | * was already unlinked/moved, so we can safely assume that we will not | |
1861 | * overwrite anything at this point in time. | |
1862 | */ | |
801bec36 RK |
1863 | if (other_inode > sctx->send_progress || |
1864 | is_waiting_for_move(sctx, other_inode)) { | |
31db9f7c | 1865 | ret = get_inode_info(sctx->parent_root, other_inode, NULL, |
85a7b33b | 1866 | who_gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
1867 | if (ret < 0) |
1868 | goto out; | |
1869 | ||
1870 | ret = 1; | |
1871 | *who_ino = other_inode; | |
1872 | } else { | |
1873 | ret = 0; | |
1874 | } | |
1875 | ||
1876 | out: | |
1877 | return ret; | |
1878 | } | |
1879 | ||
766702ef AB |
1880 | /* |
1881 | * Checks if the ref was overwritten by an already processed inode. This is | |
1882 | * used by __get_cur_name_and_parent to find out if the ref was orphanized and | |
1883 | * thus the orphan name needs be used. | |
1884 | * process_recorded_refs also uses it to avoid unlinking of refs that were | |
1885 | * overwritten. | |
1886 | */ | |
31db9f7c AB |
1887 | static int did_overwrite_ref(struct send_ctx *sctx, |
1888 | u64 dir, u64 dir_gen, | |
1889 | u64 ino, u64 ino_gen, | |
1890 | const char *name, int name_len) | |
1891 | { | |
1892 | int ret = 0; | |
1893 | u64 gen; | |
1894 | u64 ow_inode; | |
1895 | u8 other_type; | |
1896 | ||
1897 | if (!sctx->parent_root) | |
1898 | goto out; | |
1899 | ||
1900 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1901 | if (ret <= 0) | |
1902 | goto out; | |
1903 | ||
1904 | /* check if the ref was overwritten by another ref */ | |
1905 | ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, | |
1906 | &ow_inode, &other_type); | |
1907 | if (ret < 0 && ret != -ENOENT) | |
1908 | goto out; | |
1909 | if (ret) { | |
1910 | /* was never and will never be overwritten */ | |
1911 | ret = 0; | |
1912 | goto out; | |
1913 | } | |
1914 | ||
1915 | ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, | |
85a7b33b | 1916 | NULL, NULL); |
31db9f7c AB |
1917 | if (ret < 0) |
1918 | goto out; | |
1919 | ||
1920 | if (ow_inode == ino && gen == ino_gen) { | |
1921 | ret = 0; | |
1922 | goto out; | |
1923 | } | |
1924 | ||
8b191a68 FM |
1925 | /* |
1926 | * We know that it is or will be overwritten. Check this now. | |
1927 | * The current inode being processed might have been the one that caused | |
b786f16a FM |
1928 | * inode 'ino' to be orphanized, therefore check if ow_inode matches |
1929 | * the current inode being processed. | |
8b191a68 | 1930 | */ |
b786f16a FM |
1931 | if ((ow_inode < sctx->send_progress) || |
1932 | (ino != sctx->cur_ino && ow_inode == sctx->cur_ino && | |
1933 | gen == sctx->cur_inode_gen)) | |
31db9f7c AB |
1934 | ret = 1; |
1935 | else | |
1936 | ret = 0; | |
1937 | ||
1938 | out: | |
1939 | return ret; | |
1940 | } | |
1941 | ||
766702ef AB |
1942 | /* |
1943 | * Same as did_overwrite_ref, but also checks if it is the first ref of an inode | |
1944 | * that got overwritten. This is used by process_recorded_refs to determine | |
1945 | * if it has to use the path as returned by get_cur_path or the orphan name. | |
1946 | */ | |
31db9f7c AB |
1947 | static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) |
1948 | { | |
1949 | int ret = 0; | |
1950 | struct fs_path *name = NULL; | |
1951 | u64 dir; | |
1952 | u64 dir_gen; | |
1953 | ||
1954 | if (!sctx->parent_root) | |
1955 | goto out; | |
1956 | ||
924794c9 | 1957 | name = fs_path_alloc(); |
31db9f7c AB |
1958 | if (!name) |
1959 | return -ENOMEM; | |
1960 | ||
924794c9 | 1961 | ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); |
31db9f7c AB |
1962 | if (ret < 0) |
1963 | goto out; | |
1964 | ||
1965 | ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, | |
1966 | name->start, fs_path_len(name)); | |
31db9f7c AB |
1967 | |
1968 | out: | |
924794c9 | 1969 | fs_path_free(name); |
31db9f7c AB |
1970 | return ret; |
1971 | } | |
1972 | ||
766702ef AB |
1973 | /* |
1974 | * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, | |
1975 | * so we need to do some special handling in case we have clashes. This function | |
1976 | * takes care of this with the help of name_cache_entry::radix_list. | |
5dc67d0b | 1977 | * In case of error, nce is kfreed. |
766702ef | 1978 | */ |
31db9f7c AB |
1979 | static int name_cache_insert(struct send_ctx *sctx, |
1980 | struct name_cache_entry *nce) | |
1981 | { | |
1982 | int ret = 0; | |
7e0926fe AB |
1983 | struct list_head *nce_head; |
1984 | ||
1985 | nce_head = radix_tree_lookup(&sctx->name_cache, | |
1986 | (unsigned long)nce->ino); | |
1987 | if (!nce_head) { | |
e780b0d1 | 1988 | nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL); |
cfa7a9cc TI |
1989 | if (!nce_head) { |
1990 | kfree(nce); | |
31db9f7c | 1991 | return -ENOMEM; |
cfa7a9cc | 1992 | } |
7e0926fe | 1993 | INIT_LIST_HEAD(nce_head); |
31db9f7c | 1994 | |
7e0926fe | 1995 | ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); |
5dc67d0b AB |
1996 | if (ret < 0) { |
1997 | kfree(nce_head); | |
1998 | kfree(nce); | |
31db9f7c | 1999 | return ret; |
5dc67d0b | 2000 | } |
31db9f7c | 2001 | } |
7e0926fe | 2002 | list_add_tail(&nce->radix_list, nce_head); |
31db9f7c AB |
2003 | list_add_tail(&nce->list, &sctx->name_cache_list); |
2004 | sctx->name_cache_size++; | |
2005 | ||
2006 | return ret; | |
2007 | } | |
2008 | ||
2009 | static void name_cache_delete(struct send_ctx *sctx, | |
2010 | struct name_cache_entry *nce) | |
2011 | { | |
7e0926fe | 2012 | struct list_head *nce_head; |
31db9f7c | 2013 | |
7e0926fe AB |
2014 | nce_head = radix_tree_lookup(&sctx->name_cache, |
2015 | (unsigned long)nce->ino); | |
57fb8910 DS |
2016 | if (!nce_head) { |
2017 | btrfs_err(sctx->send_root->fs_info, | |
2018 | "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", | |
2019 | nce->ino, sctx->name_cache_size); | |
2020 | } | |
31db9f7c | 2021 | |
7e0926fe | 2022 | list_del(&nce->radix_list); |
31db9f7c | 2023 | list_del(&nce->list); |
31db9f7c | 2024 | sctx->name_cache_size--; |
7e0926fe | 2025 | |
57fb8910 DS |
2026 | /* |
2027 | * We may not get to the final release of nce_head if the lookup fails | |
2028 | */ | |
2029 | if (nce_head && list_empty(nce_head)) { | |
7e0926fe AB |
2030 | radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); |
2031 | kfree(nce_head); | |
2032 | } | |
31db9f7c AB |
2033 | } |
2034 | ||
2035 | static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, | |
2036 | u64 ino, u64 gen) | |
2037 | { | |
7e0926fe AB |
2038 | struct list_head *nce_head; |
2039 | struct name_cache_entry *cur; | |
31db9f7c | 2040 | |
7e0926fe AB |
2041 | nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); |
2042 | if (!nce_head) | |
31db9f7c AB |
2043 | return NULL; |
2044 | ||
7e0926fe AB |
2045 | list_for_each_entry(cur, nce_head, radix_list) { |
2046 | if (cur->ino == ino && cur->gen == gen) | |
2047 | return cur; | |
2048 | } | |
31db9f7c AB |
2049 | return NULL; |
2050 | } | |
2051 | ||
766702ef AB |
2052 | /* |
2053 | * Removes the entry from the list and adds it back to the end. This marks the | |
2054 | * entry as recently used so that name_cache_clean_unused does not remove it. | |
2055 | */ | |
31db9f7c AB |
2056 | static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) |
2057 | { | |
2058 | list_del(&nce->list); | |
2059 | list_add_tail(&nce->list, &sctx->name_cache_list); | |
2060 | } | |
2061 | ||
766702ef AB |
2062 | /* |
2063 | * Remove some entries from the beginning of name_cache_list. | |
2064 | */ | |
31db9f7c AB |
2065 | static void name_cache_clean_unused(struct send_ctx *sctx) |
2066 | { | |
2067 | struct name_cache_entry *nce; | |
2068 | ||
2069 | if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) | |
2070 | return; | |
2071 | ||
2072 | while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { | |
2073 | nce = list_entry(sctx->name_cache_list.next, | |
2074 | struct name_cache_entry, list); | |
2075 | name_cache_delete(sctx, nce); | |
2076 | kfree(nce); | |
2077 | } | |
2078 | } | |
2079 | ||
2080 | static void name_cache_free(struct send_ctx *sctx) | |
2081 | { | |
2082 | struct name_cache_entry *nce; | |
31db9f7c | 2083 | |
e938c8ad AB |
2084 | while (!list_empty(&sctx->name_cache_list)) { |
2085 | nce = list_entry(sctx->name_cache_list.next, | |
2086 | struct name_cache_entry, list); | |
31db9f7c | 2087 | name_cache_delete(sctx, nce); |
17589bd9 | 2088 | kfree(nce); |
31db9f7c AB |
2089 | } |
2090 | } | |
2091 | ||
766702ef AB |
2092 | /* |
2093 | * Used by get_cur_path for each ref up to the root. | |
2094 | * Returns 0 if it succeeded. | |
2095 | * Returns 1 if the inode is not existent or got overwritten. In that case, the | |
2096 | * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 | |
2097 | * is returned, parent_ino/parent_gen are not guaranteed to be valid. | |
2098 | * Returns <0 in case of error. | |
2099 | */ | |
31db9f7c AB |
2100 | static int __get_cur_name_and_parent(struct send_ctx *sctx, |
2101 | u64 ino, u64 gen, | |
2102 | u64 *parent_ino, | |
2103 | u64 *parent_gen, | |
2104 | struct fs_path *dest) | |
2105 | { | |
2106 | int ret; | |
2107 | int nce_ret; | |
31db9f7c AB |
2108 | struct name_cache_entry *nce = NULL; |
2109 | ||
766702ef AB |
2110 | /* |
2111 | * First check if we already did a call to this function with the same | |
2112 | * ino/gen. If yes, check if the cache entry is still up-to-date. If yes | |
2113 | * return the cached result. | |
2114 | */ | |
31db9f7c AB |
2115 | nce = name_cache_search(sctx, ino, gen); |
2116 | if (nce) { | |
2117 | if (ino < sctx->send_progress && nce->need_later_update) { | |
2118 | name_cache_delete(sctx, nce); | |
2119 | kfree(nce); | |
2120 | nce = NULL; | |
2121 | } else { | |
2122 | name_cache_used(sctx, nce); | |
2123 | *parent_ino = nce->parent_ino; | |
2124 | *parent_gen = nce->parent_gen; | |
2125 | ret = fs_path_add(dest, nce->name, nce->name_len); | |
2126 | if (ret < 0) | |
2127 | goto out; | |
2128 | ret = nce->ret; | |
2129 | goto out; | |
2130 | } | |
2131 | } | |
2132 | ||
766702ef AB |
2133 | /* |
2134 | * If the inode is not existent yet, add the orphan name and return 1. | |
2135 | * This should only happen for the parent dir that we determine in | |
2136 | * __record_new_ref | |
2137 | */ | |
31db9f7c AB |
2138 | ret = is_inode_existent(sctx, ino, gen); |
2139 | if (ret < 0) | |
2140 | goto out; | |
2141 | ||
2142 | if (!ret) { | |
2143 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2144 | if (ret < 0) | |
2145 | goto out; | |
2146 | ret = 1; | |
2147 | goto out_cache; | |
2148 | } | |
2149 | ||
766702ef AB |
2150 | /* |
2151 | * Depending on whether the inode was already processed or not, use | |
2152 | * send_root or parent_root for ref lookup. | |
2153 | */ | |
bf0d1f44 | 2154 | if (ino < sctx->send_progress) |
924794c9 TI |
2155 | ret = get_first_ref(sctx->send_root, ino, |
2156 | parent_ino, parent_gen, dest); | |
31db9f7c | 2157 | else |
924794c9 TI |
2158 | ret = get_first_ref(sctx->parent_root, ino, |
2159 | parent_ino, parent_gen, dest); | |
31db9f7c AB |
2160 | if (ret < 0) |
2161 | goto out; | |
2162 | ||
766702ef AB |
2163 | /* |
2164 | * Check if the ref was overwritten by an inode's ref that was processed | |
2165 | * earlier. If yes, treat as orphan and return 1. | |
2166 | */ | |
31db9f7c AB |
2167 | ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, |
2168 | dest->start, dest->end - dest->start); | |
2169 | if (ret < 0) | |
2170 | goto out; | |
2171 | if (ret) { | |
2172 | fs_path_reset(dest); | |
2173 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2174 | if (ret < 0) | |
2175 | goto out; | |
2176 | ret = 1; | |
2177 | } | |
2178 | ||
2179 | out_cache: | |
766702ef AB |
2180 | /* |
2181 | * Store the result of the lookup in the name cache. | |
2182 | */ | |
e780b0d1 | 2183 | nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); |
31db9f7c AB |
2184 | if (!nce) { |
2185 | ret = -ENOMEM; | |
2186 | goto out; | |
2187 | } | |
2188 | ||
2189 | nce->ino = ino; | |
2190 | nce->gen = gen; | |
2191 | nce->parent_ino = *parent_ino; | |
2192 | nce->parent_gen = *parent_gen; | |
2193 | nce->name_len = fs_path_len(dest); | |
2194 | nce->ret = ret; | |
2195 | strcpy(nce->name, dest->start); | |
31db9f7c AB |
2196 | |
2197 | if (ino < sctx->send_progress) | |
2198 | nce->need_later_update = 0; | |
2199 | else | |
2200 | nce->need_later_update = 1; | |
2201 | ||
2202 | nce_ret = name_cache_insert(sctx, nce); | |
2203 | if (nce_ret < 0) | |
2204 | ret = nce_ret; | |
2205 | name_cache_clean_unused(sctx); | |
2206 | ||
2207 | out: | |
31db9f7c AB |
2208 | return ret; |
2209 | } | |
2210 | ||
2211 | /* | |
2212 | * Magic happens here. This function returns the first ref to an inode as it | |
2213 | * would look like while receiving the stream at this point in time. | |
2214 | * We walk the path up to the root. For every inode in between, we check if it | |
2215 | * was already processed/sent. If yes, we continue with the parent as found | |
2216 | * in send_root. If not, we continue with the parent as found in parent_root. | |
2217 | * If we encounter an inode that was deleted at this point in time, we use the | |
2218 | * inodes "orphan" name instead of the real name and stop. Same with new inodes | |
2219 | * that were not created yet and overwritten inodes/refs. | |
2220 | * | |
2221 | * When do we have have orphan inodes: | |
2222 | * 1. When an inode is freshly created and thus no valid refs are available yet | |
2223 | * 2. When a directory lost all it's refs (deleted) but still has dir items | |
2224 | * inside which were not processed yet (pending for move/delete). If anyone | |
2225 | * tried to get the path to the dir items, it would get a path inside that | |
2226 | * orphan directory. | |
2227 | * 3. When an inode is moved around or gets new links, it may overwrite the ref | |
2228 | * of an unprocessed inode. If in that case the first ref would be | |
2229 | * overwritten, the overwritten inode gets "orphanized". Later when we | |
2230 | * process this overwritten inode, it is restored at a new place by moving | |
2231 | * the orphan inode. | |
2232 | * | |
2233 | * sctx->send_progress tells this function at which point in time receiving | |
2234 | * would be. | |
2235 | */ | |
2236 | static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, | |
2237 | struct fs_path *dest) | |
2238 | { | |
2239 | int ret = 0; | |
2240 | struct fs_path *name = NULL; | |
2241 | u64 parent_inode = 0; | |
2242 | u64 parent_gen = 0; | |
2243 | int stop = 0; | |
2244 | ||
924794c9 | 2245 | name = fs_path_alloc(); |
31db9f7c AB |
2246 | if (!name) { |
2247 | ret = -ENOMEM; | |
2248 | goto out; | |
2249 | } | |
2250 | ||
2251 | dest->reversed = 1; | |
2252 | fs_path_reset(dest); | |
2253 | ||
2254 | while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { | |
8b191a68 FM |
2255 | struct waiting_dir_move *wdm; |
2256 | ||
31db9f7c AB |
2257 | fs_path_reset(name); |
2258 | ||
9dc44214 FM |
2259 | if (is_waiting_for_rm(sctx, ino)) { |
2260 | ret = gen_unique_name(sctx, ino, gen, name); | |
2261 | if (ret < 0) | |
2262 | goto out; | |
2263 | ret = fs_path_add_path(dest, name); | |
2264 | break; | |
2265 | } | |
2266 | ||
8b191a68 FM |
2267 | wdm = get_waiting_dir_move(sctx, ino); |
2268 | if (wdm && wdm->orphanized) { | |
2269 | ret = gen_unique_name(sctx, ino, gen, name); | |
2270 | stop = 1; | |
2271 | } else if (wdm) { | |
bf0d1f44 FM |
2272 | ret = get_first_ref(sctx->parent_root, ino, |
2273 | &parent_inode, &parent_gen, name); | |
2274 | } else { | |
2275 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
2276 | &parent_inode, | |
2277 | &parent_gen, name); | |
2278 | if (ret) | |
2279 | stop = 1; | |
2280 | } | |
2281 | ||
31db9f7c AB |
2282 | if (ret < 0) |
2283 | goto out; | |
9f03740a | 2284 | |
31db9f7c AB |
2285 | ret = fs_path_add_path(dest, name); |
2286 | if (ret < 0) | |
2287 | goto out; | |
2288 | ||
2289 | ino = parent_inode; | |
2290 | gen = parent_gen; | |
2291 | } | |
2292 | ||
2293 | out: | |
924794c9 | 2294 | fs_path_free(name); |
31db9f7c AB |
2295 | if (!ret) |
2296 | fs_path_unreverse(dest); | |
2297 | return ret; | |
2298 | } | |
2299 | ||
31db9f7c AB |
2300 | /* |
2301 | * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace | |
2302 | */ | |
2303 | static int send_subvol_begin(struct send_ctx *sctx) | |
2304 | { | |
2305 | int ret; | |
2306 | struct btrfs_root *send_root = sctx->send_root; | |
2307 | struct btrfs_root *parent_root = sctx->parent_root; | |
2308 | struct btrfs_path *path; | |
2309 | struct btrfs_key key; | |
2310 | struct btrfs_root_ref *ref; | |
2311 | struct extent_buffer *leaf; | |
2312 | char *name = NULL; | |
2313 | int namelen; | |
2314 | ||
ffcfaf81 | 2315 | path = btrfs_alloc_path(); |
31db9f7c AB |
2316 | if (!path) |
2317 | return -ENOMEM; | |
2318 | ||
e780b0d1 | 2319 | name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); |
31db9f7c AB |
2320 | if (!name) { |
2321 | btrfs_free_path(path); | |
2322 | return -ENOMEM; | |
2323 | } | |
2324 | ||
2325 | key.objectid = send_root->objectid; | |
2326 | key.type = BTRFS_ROOT_BACKREF_KEY; | |
2327 | key.offset = 0; | |
2328 | ||
2329 | ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, | |
2330 | &key, path, 1, 0); | |
2331 | if (ret < 0) | |
2332 | goto out; | |
2333 | if (ret) { | |
2334 | ret = -ENOENT; | |
2335 | goto out; | |
2336 | } | |
2337 | ||
2338 | leaf = path->nodes[0]; | |
2339 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2340 | if (key.type != BTRFS_ROOT_BACKREF_KEY || | |
2341 | key.objectid != send_root->objectid) { | |
2342 | ret = -ENOENT; | |
2343 | goto out; | |
2344 | } | |
2345 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
2346 | namelen = btrfs_root_ref_name_len(leaf, ref); | |
2347 | read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); | |
2348 | btrfs_release_path(path); | |
2349 | ||
31db9f7c AB |
2350 | if (parent_root) { |
2351 | ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); | |
2352 | if (ret < 0) | |
2353 | goto out; | |
2354 | } else { | |
2355 | ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); | |
2356 | if (ret < 0) | |
2357 | goto out; | |
2358 | } | |
2359 | ||
2360 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); | |
b96b1db0 RR |
2361 | |
2362 | if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid)) | |
2363 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2364 | sctx->send_root->root_item.received_uuid); | |
2365 | else | |
2366 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2367 | sctx->send_root->root_item.uuid); | |
2368 | ||
31db9f7c | 2369 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, |
5a0f4e2c | 2370 | le64_to_cpu(sctx->send_root->root_item.ctransid)); |
31db9f7c | 2371 | if (parent_root) { |
37b8d27d JB |
2372 | if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) |
2373 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2374 | parent_root->root_item.received_uuid); | |
2375 | else | |
2376 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2377 | parent_root->root_item.uuid); | |
31db9f7c | 2378 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 2379 | le64_to_cpu(sctx->parent_root->root_item.ctransid)); |
31db9f7c AB |
2380 | } |
2381 | ||
2382 | ret = send_cmd(sctx); | |
2383 | ||
2384 | tlv_put_failure: | |
2385 | out: | |
2386 | btrfs_free_path(path); | |
2387 | kfree(name); | |
2388 | return ret; | |
2389 | } | |
2390 | ||
2391 | static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) | |
2392 | { | |
2393 | int ret = 0; | |
2394 | struct fs_path *p; | |
2395 | ||
2396 | verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size); | |
2397 | ||
924794c9 | 2398 | p = fs_path_alloc(); |
31db9f7c AB |
2399 | if (!p) |
2400 | return -ENOMEM; | |
2401 | ||
2402 | ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); | |
2403 | if (ret < 0) | |
2404 | goto out; | |
2405 | ||
2406 | ret = get_cur_path(sctx, ino, gen, p); | |
2407 | if (ret < 0) | |
2408 | goto out; | |
2409 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2410 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); | |
2411 | ||
2412 | ret = send_cmd(sctx); | |
2413 | ||
2414 | tlv_put_failure: | |
2415 | out: | |
924794c9 | 2416 | fs_path_free(p); |
31db9f7c AB |
2417 | return ret; |
2418 | } | |
2419 | ||
2420 | static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) | |
2421 | { | |
2422 | int ret = 0; | |
2423 | struct fs_path *p; | |
2424 | ||
2425 | verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode); | |
2426 | ||
924794c9 | 2427 | p = fs_path_alloc(); |
31db9f7c AB |
2428 | if (!p) |
2429 | return -ENOMEM; | |
2430 | ||
2431 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); | |
2432 | if (ret < 0) | |
2433 | goto out; | |
2434 | ||
2435 | ret = get_cur_path(sctx, ino, gen, p); | |
2436 | if (ret < 0) | |
2437 | goto out; | |
2438 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2439 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); | |
2440 | ||
2441 | ret = send_cmd(sctx); | |
2442 | ||
2443 | tlv_put_failure: | |
2444 | out: | |
924794c9 | 2445 | fs_path_free(p); |
31db9f7c AB |
2446 | return ret; |
2447 | } | |
2448 | ||
2449 | static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) | |
2450 | { | |
2451 | int ret = 0; | |
2452 | struct fs_path *p; | |
2453 | ||
2454 | verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid); | |
2455 | ||
924794c9 | 2456 | p = fs_path_alloc(); |
31db9f7c AB |
2457 | if (!p) |
2458 | return -ENOMEM; | |
2459 | ||
2460 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); | |
2461 | if (ret < 0) | |
2462 | goto out; | |
2463 | ||
2464 | ret = get_cur_path(sctx, ino, gen, p); | |
2465 | if (ret < 0) | |
2466 | goto out; | |
2467 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2468 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); | |
2469 | TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); | |
2470 | ||
2471 | ret = send_cmd(sctx); | |
2472 | ||
2473 | tlv_put_failure: | |
2474 | out: | |
924794c9 | 2475 | fs_path_free(p); |
31db9f7c AB |
2476 | return ret; |
2477 | } | |
2478 | ||
2479 | static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) | |
2480 | { | |
2481 | int ret = 0; | |
2482 | struct fs_path *p = NULL; | |
2483 | struct btrfs_inode_item *ii; | |
2484 | struct btrfs_path *path = NULL; | |
2485 | struct extent_buffer *eb; | |
2486 | struct btrfs_key key; | |
2487 | int slot; | |
2488 | ||
2489 | verbose_printk("btrfs: send_utimes %llu\n", ino); | |
2490 | ||
924794c9 | 2491 | p = fs_path_alloc(); |
31db9f7c AB |
2492 | if (!p) |
2493 | return -ENOMEM; | |
2494 | ||
2495 | path = alloc_path_for_send(); | |
2496 | if (!path) { | |
2497 | ret = -ENOMEM; | |
2498 | goto out; | |
2499 | } | |
2500 | ||
2501 | key.objectid = ino; | |
2502 | key.type = BTRFS_INODE_ITEM_KEY; | |
2503 | key.offset = 0; | |
2504 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); | |
15b253ea FM |
2505 | if (ret > 0) |
2506 | ret = -ENOENT; | |
31db9f7c AB |
2507 | if (ret < 0) |
2508 | goto out; | |
2509 | ||
2510 | eb = path->nodes[0]; | |
2511 | slot = path->slots[0]; | |
2512 | ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
2513 | ||
2514 | ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); | |
2515 | if (ret < 0) | |
2516 | goto out; | |
2517 | ||
2518 | ret = get_cur_path(sctx, ino, gen, p); | |
2519 | if (ret < 0) | |
2520 | goto out; | |
2521 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
a937b979 DS |
2522 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); |
2523 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); | |
2524 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); | |
766702ef | 2525 | /* TODO Add otime support when the otime patches get into upstream */ |
31db9f7c AB |
2526 | |
2527 | ret = send_cmd(sctx); | |
2528 | ||
2529 | tlv_put_failure: | |
2530 | out: | |
924794c9 | 2531 | fs_path_free(p); |
31db9f7c AB |
2532 | btrfs_free_path(path); |
2533 | return ret; | |
2534 | } | |
2535 | ||
2536 | /* | |
2537 | * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have | |
2538 | * a valid path yet because we did not process the refs yet. So, the inode | |
2539 | * is created as orphan. | |
2540 | */ | |
1f4692da | 2541 | static int send_create_inode(struct send_ctx *sctx, u64 ino) |
31db9f7c AB |
2542 | { |
2543 | int ret = 0; | |
31db9f7c | 2544 | struct fs_path *p; |
31db9f7c | 2545 | int cmd; |
1f4692da | 2546 | u64 gen; |
31db9f7c | 2547 | u64 mode; |
1f4692da | 2548 | u64 rdev; |
31db9f7c | 2549 | |
1f4692da | 2550 | verbose_printk("btrfs: send_create_inode %llu\n", ino); |
31db9f7c | 2551 | |
924794c9 | 2552 | p = fs_path_alloc(); |
31db9f7c AB |
2553 | if (!p) |
2554 | return -ENOMEM; | |
2555 | ||
644d1940 LB |
2556 | if (ino != sctx->cur_ino) { |
2557 | ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, | |
2558 | NULL, NULL, &rdev); | |
2559 | if (ret < 0) | |
2560 | goto out; | |
2561 | } else { | |
2562 | gen = sctx->cur_inode_gen; | |
2563 | mode = sctx->cur_inode_mode; | |
2564 | rdev = sctx->cur_inode_rdev; | |
2565 | } | |
31db9f7c | 2566 | |
e938c8ad | 2567 | if (S_ISREG(mode)) { |
31db9f7c | 2568 | cmd = BTRFS_SEND_C_MKFILE; |
e938c8ad | 2569 | } else if (S_ISDIR(mode)) { |
31db9f7c | 2570 | cmd = BTRFS_SEND_C_MKDIR; |
e938c8ad | 2571 | } else if (S_ISLNK(mode)) { |
31db9f7c | 2572 | cmd = BTRFS_SEND_C_SYMLINK; |
e938c8ad | 2573 | } else if (S_ISCHR(mode) || S_ISBLK(mode)) { |
31db9f7c | 2574 | cmd = BTRFS_SEND_C_MKNOD; |
e938c8ad | 2575 | } else if (S_ISFIFO(mode)) { |
31db9f7c | 2576 | cmd = BTRFS_SEND_C_MKFIFO; |
e938c8ad | 2577 | } else if (S_ISSOCK(mode)) { |
31db9f7c | 2578 | cmd = BTRFS_SEND_C_MKSOCK; |
e938c8ad | 2579 | } else { |
f14d104d | 2580 | btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o", |
31db9f7c AB |
2581 | (int)(mode & S_IFMT)); |
2582 | ret = -ENOTSUPP; | |
2583 | goto out; | |
2584 | } | |
2585 | ||
2586 | ret = begin_cmd(sctx, cmd); | |
2587 | if (ret < 0) | |
2588 | goto out; | |
2589 | ||
1f4692da | 2590 | ret = gen_unique_name(sctx, ino, gen, p); |
31db9f7c AB |
2591 | if (ret < 0) |
2592 | goto out; | |
2593 | ||
2594 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
1f4692da | 2595 | TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); |
31db9f7c AB |
2596 | |
2597 | if (S_ISLNK(mode)) { | |
2598 | fs_path_reset(p); | |
924794c9 | 2599 | ret = read_symlink(sctx->send_root, ino, p); |
31db9f7c AB |
2600 | if (ret < 0) |
2601 | goto out; | |
2602 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); | |
2603 | } else if (S_ISCHR(mode) || S_ISBLK(mode) || | |
2604 | S_ISFIFO(mode) || S_ISSOCK(mode)) { | |
d79e5043 AJ |
2605 | TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); |
2606 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); | |
31db9f7c AB |
2607 | } |
2608 | ||
2609 | ret = send_cmd(sctx); | |
2610 | if (ret < 0) | |
2611 | goto out; | |
2612 | ||
2613 | ||
2614 | tlv_put_failure: | |
2615 | out: | |
924794c9 | 2616 | fs_path_free(p); |
31db9f7c AB |
2617 | return ret; |
2618 | } | |
2619 | ||
1f4692da AB |
2620 | /* |
2621 | * We need some special handling for inodes that get processed before the parent | |
2622 | * directory got created. See process_recorded_refs for details. | |
2623 | * This function does the check if we already created the dir out of order. | |
2624 | */ | |
2625 | static int did_create_dir(struct send_ctx *sctx, u64 dir) | |
2626 | { | |
2627 | int ret = 0; | |
2628 | struct btrfs_path *path = NULL; | |
2629 | struct btrfs_key key; | |
2630 | struct btrfs_key found_key; | |
2631 | struct btrfs_key di_key; | |
2632 | struct extent_buffer *eb; | |
2633 | struct btrfs_dir_item *di; | |
2634 | int slot; | |
2635 | ||
2636 | path = alloc_path_for_send(); | |
2637 | if (!path) { | |
2638 | ret = -ENOMEM; | |
2639 | goto out; | |
2640 | } | |
2641 | ||
2642 | key.objectid = dir; | |
2643 | key.type = BTRFS_DIR_INDEX_KEY; | |
2644 | key.offset = 0; | |
dff6d0ad FDBM |
2645 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); |
2646 | if (ret < 0) | |
2647 | goto out; | |
2648 | ||
1f4692da | 2649 | while (1) { |
dff6d0ad FDBM |
2650 | eb = path->nodes[0]; |
2651 | slot = path->slots[0]; | |
2652 | if (slot >= btrfs_header_nritems(eb)) { | |
2653 | ret = btrfs_next_leaf(sctx->send_root, path); | |
2654 | if (ret < 0) { | |
2655 | goto out; | |
2656 | } else if (ret > 0) { | |
2657 | ret = 0; | |
2658 | break; | |
2659 | } | |
2660 | continue; | |
1f4692da | 2661 | } |
dff6d0ad FDBM |
2662 | |
2663 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
2664 | if (found_key.objectid != key.objectid || | |
1f4692da AB |
2665 | found_key.type != key.type) { |
2666 | ret = 0; | |
2667 | goto out; | |
2668 | } | |
2669 | ||
2670 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); | |
2671 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2672 | ||
a0525414 JB |
2673 | if (di_key.type != BTRFS_ROOT_ITEM_KEY && |
2674 | di_key.objectid < sctx->send_progress) { | |
1f4692da AB |
2675 | ret = 1; |
2676 | goto out; | |
2677 | } | |
2678 | ||
dff6d0ad | 2679 | path->slots[0]++; |
1f4692da AB |
2680 | } |
2681 | ||
2682 | out: | |
2683 | btrfs_free_path(path); | |
2684 | return ret; | |
2685 | } | |
2686 | ||
2687 | /* | |
2688 | * Only creates the inode if it is: | |
2689 | * 1. Not a directory | |
2690 | * 2. Or a directory which was not created already due to out of order | |
2691 | * directories. See did_create_dir and process_recorded_refs for details. | |
2692 | */ | |
2693 | static int send_create_inode_if_needed(struct send_ctx *sctx) | |
2694 | { | |
2695 | int ret; | |
2696 | ||
2697 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
2698 | ret = did_create_dir(sctx, sctx->cur_ino); | |
2699 | if (ret < 0) | |
2700 | goto out; | |
2701 | if (ret) { | |
2702 | ret = 0; | |
2703 | goto out; | |
2704 | } | |
2705 | } | |
2706 | ||
2707 | ret = send_create_inode(sctx, sctx->cur_ino); | |
2708 | if (ret < 0) | |
2709 | goto out; | |
2710 | ||
2711 | out: | |
2712 | return ret; | |
2713 | } | |
2714 | ||
31db9f7c AB |
2715 | struct recorded_ref { |
2716 | struct list_head list; | |
2717 | char *dir_path; | |
2718 | char *name; | |
2719 | struct fs_path *full_path; | |
2720 | u64 dir; | |
2721 | u64 dir_gen; | |
2722 | int dir_path_len; | |
2723 | int name_len; | |
2724 | }; | |
2725 | ||
2726 | /* | |
2727 | * We need to process new refs before deleted refs, but compare_tree gives us | |
2728 | * everything mixed. So we first record all refs and later process them. | |
2729 | * This function is a helper to record one ref. | |
2730 | */ | |
a4d96d62 | 2731 | static int __record_ref(struct list_head *head, u64 dir, |
31db9f7c AB |
2732 | u64 dir_gen, struct fs_path *path) |
2733 | { | |
2734 | struct recorded_ref *ref; | |
31db9f7c | 2735 | |
e780b0d1 | 2736 | ref = kmalloc(sizeof(*ref), GFP_KERNEL); |
31db9f7c AB |
2737 | if (!ref) |
2738 | return -ENOMEM; | |
2739 | ||
2740 | ref->dir = dir; | |
2741 | ref->dir_gen = dir_gen; | |
2742 | ref->full_path = path; | |
2743 | ||
ed84885d AS |
2744 | ref->name = (char *)kbasename(ref->full_path->start); |
2745 | ref->name_len = ref->full_path->end - ref->name; | |
2746 | ref->dir_path = ref->full_path->start; | |
2747 | if (ref->name == ref->full_path->start) | |
31db9f7c | 2748 | ref->dir_path_len = 0; |
ed84885d | 2749 | else |
31db9f7c AB |
2750 | ref->dir_path_len = ref->full_path->end - |
2751 | ref->full_path->start - 1 - ref->name_len; | |
31db9f7c AB |
2752 | |
2753 | list_add_tail(&ref->list, head); | |
2754 | return 0; | |
2755 | } | |
2756 | ||
ba5e8f2e JB |
2757 | static int dup_ref(struct recorded_ref *ref, struct list_head *list) |
2758 | { | |
2759 | struct recorded_ref *new; | |
2760 | ||
e780b0d1 | 2761 | new = kmalloc(sizeof(*ref), GFP_KERNEL); |
ba5e8f2e JB |
2762 | if (!new) |
2763 | return -ENOMEM; | |
2764 | ||
2765 | new->dir = ref->dir; | |
2766 | new->dir_gen = ref->dir_gen; | |
2767 | new->full_path = NULL; | |
2768 | INIT_LIST_HEAD(&new->list); | |
2769 | list_add_tail(&new->list, list); | |
2770 | return 0; | |
2771 | } | |
2772 | ||
924794c9 | 2773 | static void __free_recorded_refs(struct list_head *head) |
31db9f7c AB |
2774 | { |
2775 | struct recorded_ref *cur; | |
31db9f7c | 2776 | |
e938c8ad AB |
2777 | while (!list_empty(head)) { |
2778 | cur = list_entry(head->next, struct recorded_ref, list); | |
924794c9 | 2779 | fs_path_free(cur->full_path); |
e938c8ad | 2780 | list_del(&cur->list); |
31db9f7c AB |
2781 | kfree(cur); |
2782 | } | |
31db9f7c AB |
2783 | } |
2784 | ||
2785 | static void free_recorded_refs(struct send_ctx *sctx) | |
2786 | { | |
924794c9 TI |
2787 | __free_recorded_refs(&sctx->new_refs); |
2788 | __free_recorded_refs(&sctx->deleted_refs); | |
31db9f7c AB |
2789 | } |
2790 | ||
2791 | /* | |
766702ef | 2792 | * Renames/moves a file/dir to its orphan name. Used when the first |
31db9f7c AB |
2793 | * ref of an unprocessed inode gets overwritten and for all non empty |
2794 | * directories. | |
2795 | */ | |
2796 | static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, | |
2797 | struct fs_path *path) | |
2798 | { | |
2799 | int ret; | |
2800 | struct fs_path *orphan; | |
2801 | ||
924794c9 | 2802 | orphan = fs_path_alloc(); |
31db9f7c AB |
2803 | if (!orphan) |
2804 | return -ENOMEM; | |
2805 | ||
2806 | ret = gen_unique_name(sctx, ino, gen, orphan); | |
2807 | if (ret < 0) | |
2808 | goto out; | |
2809 | ||
2810 | ret = send_rename(sctx, path, orphan); | |
2811 | ||
2812 | out: | |
924794c9 | 2813 | fs_path_free(orphan); |
31db9f7c AB |
2814 | return ret; |
2815 | } | |
2816 | ||
9dc44214 FM |
2817 | static struct orphan_dir_info * |
2818 | add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2819 | { | |
2820 | struct rb_node **p = &sctx->orphan_dirs.rb_node; | |
2821 | struct rb_node *parent = NULL; | |
2822 | struct orphan_dir_info *entry, *odi; | |
2823 | ||
e780b0d1 | 2824 | odi = kmalloc(sizeof(*odi), GFP_KERNEL); |
9dc44214 FM |
2825 | if (!odi) |
2826 | return ERR_PTR(-ENOMEM); | |
2827 | odi->ino = dir_ino; | |
2828 | odi->gen = 0; | |
2829 | ||
2830 | while (*p) { | |
2831 | parent = *p; | |
2832 | entry = rb_entry(parent, struct orphan_dir_info, node); | |
2833 | if (dir_ino < entry->ino) { | |
2834 | p = &(*p)->rb_left; | |
2835 | } else if (dir_ino > entry->ino) { | |
2836 | p = &(*p)->rb_right; | |
2837 | } else { | |
2838 | kfree(odi); | |
2839 | return entry; | |
2840 | } | |
2841 | } | |
2842 | ||
2843 | rb_link_node(&odi->node, parent, p); | |
2844 | rb_insert_color(&odi->node, &sctx->orphan_dirs); | |
2845 | return odi; | |
2846 | } | |
2847 | ||
2848 | static struct orphan_dir_info * | |
2849 | get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2850 | { | |
2851 | struct rb_node *n = sctx->orphan_dirs.rb_node; | |
2852 | struct orphan_dir_info *entry; | |
2853 | ||
2854 | while (n) { | |
2855 | entry = rb_entry(n, struct orphan_dir_info, node); | |
2856 | if (dir_ino < entry->ino) | |
2857 | n = n->rb_left; | |
2858 | else if (dir_ino > entry->ino) | |
2859 | n = n->rb_right; | |
2860 | else | |
2861 | return entry; | |
2862 | } | |
2863 | return NULL; | |
2864 | } | |
2865 | ||
2866 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino) | |
2867 | { | |
2868 | struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino); | |
2869 | ||
2870 | return odi != NULL; | |
2871 | } | |
2872 | ||
2873 | static void free_orphan_dir_info(struct send_ctx *sctx, | |
2874 | struct orphan_dir_info *odi) | |
2875 | { | |
2876 | if (!odi) | |
2877 | return; | |
2878 | rb_erase(&odi->node, &sctx->orphan_dirs); | |
2879 | kfree(odi); | |
2880 | } | |
2881 | ||
31db9f7c AB |
2882 | /* |
2883 | * Returns 1 if a directory can be removed at this point in time. | |
2884 | * We check this by iterating all dir items and checking if the inode behind | |
2885 | * the dir item was already processed. | |
2886 | */ | |
9dc44214 FM |
2887 | static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
2888 | u64 send_progress) | |
31db9f7c AB |
2889 | { |
2890 | int ret = 0; | |
2891 | struct btrfs_root *root = sctx->parent_root; | |
2892 | struct btrfs_path *path; | |
2893 | struct btrfs_key key; | |
2894 | struct btrfs_key found_key; | |
2895 | struct btrfs_key loc; | |
2896 | struct btrfs_dir_item *di; | |
2897 | ||
6d85ed05 AB |
2898 | /* |
2899 | * Don't try to rmdir the top/root subvolume dir. | |
2900 | */ | |
2901 | if (dir == BTRFS_FIRST_FREE_OBJECTID) | |
2902 | return 0; | |
2903 | ||
31db9f7c AB |
2904 | path = alloc_path_for_send(); |
2905 | if (!path) | |
2906 | return -ENOMEM; | |
2907 | ||
2908 | key.objectid = dir; | |
2909 | key.type = BTRFS_DIR_INDEX_KEY; | |
2910 | key.offset = 0; | |
dff6d0ad FDBM |
2911 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2912 | if (ret < 0) | |
2913 | goto out; | |
31db9f7c AB |
2914 | |
2915 | while (1) { | |
9dc44214 FM |
2916 | struct waiting_dir_move *dm; |
2917 | ||
dff6d0ad FDBM |
2918 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
2919 | ret = btrfs_next_leaf(root, path); | |
2920 | if (ret < 0) | |
2921 | goto out; | |
2922 | else if (ret > 0) | |
2923 | break; | |
2924 | continue; | |
31db9f7c | 2925 | } |
dff6d0ad FDBM |
2926 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2927 | path->slots[0]); | |
2928 | if (found_key.objectid != key.objectid || | |
2929 | found_key.type != key.type) | |
31db9f7c | 2930 | break; |
31db9f7c AB |
2931 | |
2932 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2933 | struct btrfs_dir_item); | |
2934 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
2935 | ||
9dc44214 FM |
2936 | dm = get_waiting_dir_move(sctx, loc.objectid); |
2937 | if (dm) { | |
2938 | struct orphan_dir_info *odi; | |
2939 | ||
2940 | odi = add_orphan_dir_info(sctx, dir); | |
2941 | if (IS_ERR(odi)) { | |
2942 | ret = PTR_ERR(odi); | |
2943 | goto out; | |
2944 | } | |
2945 | odi->gen = dir_gen; | |
2946 | dm->rmdir_ino = dir; | |
2947 | ret = 0; | |
2948 | goto out; | |
2949 | } | |
2950 | ||
31db9f7c | 2951 | if (loc.objectid > send_progress) { |
443f9d26 RK |
2952 | struct orphan_dir_info *odi; |
2953 | ||
2954 | odi = get_orphan_dir_info(sctx, dir); | |
2955 | free_orphan_dir_info(sctx, odi); | |
31db9f7c AB |
2956 | ret = 0; |
2957 | goto out; | |
2958 | } | |
2959 | ||
dff6d0ad | 2960 | path->slots[0]++; |
31db9f7c AB |
2961 | } |
2962 | ||
2963 | ret = 1; | |
2964 | ||
2965 | out: | |
2966 | btrfs_free_path(path); | |
2967 | return ret; | |
2968 | } | |
2969 | ||
9f03740a FDBM |
2970 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) |
2971 | { | |
9dc44214 | 2972 | struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); |
9f03740a | 2973 | |
9dc44214 | 2974 | return entry != NULL; |
9f03740a FDBM |
2975 | } |
2976 | ||
8b191a68 | 2977 | static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) |
9f03740a FDBM |
2978 | { |
2979 | struct rb_node **p = &sctx->waiting_dir_moves.rb_node; | |
2980 | struct rb_node *parent = NULL; | |
2981 | struct waiting_dir_move *entry, *dm; | |
2982 | ||
e780b0d1 | 2983 | dm = kmalloc(sizeof(*dm), GFP_KERNEL); |
9f03740a FDBM |
2984 | if (!dm) |
2985 | return -ENOMEM; | |
2986 | dm->ino = ino; | |
9dc44214 | 2987 | dm->rmdir_ino = 0; |
8b191a68 | 2988 | dm->orphanized = orphanized; |
9f03740a FDBM |
2989 | |
2990 | while (*p) { | |
2991 | parent = *p; | |
2992 | entry = rb_entry(parent, struct waiting_dir_move, node); | |
2993 | if (ino < entry->ino) { | |
2994 | p = &(*p)->rb_left; | |
2995 | } else if (ino > entry->ino) { | |
2996 | p = &(*p)->rb_right; | |
2997 | } else { | |
2998 | kfree(dm); | |
2999 | return -EEXIST; | |
3000 | } | |
3001 | } | |
3002 | ||
3003 | rb_link_node(&dm->node, parent, p); | |
3004 | rb_insert_color(&dm->node, &sctx->waiting_dir_moves); | |
3005 | return 0; | |
3006 | } | |
3007 | ||
9dc44214 FM |
3008 | static struct waiting_dir_move * |
3009 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
9f03740a FDBM |
3010 | { |
3011 | struct rb_node *n = sctx->waiting_dir_moves.rb_node; | |
3012 | struct waiting_dir_move *entry; | |
3013 | ||
3014 | while (n) { | |
3015 | entry = rb_entry(n, struct waiting_dir_move, node); | |
9dc44214 | 3016 | if (ino < entry->ino) |
9f03740a | 3017 | n = n->rb_left; |
9dc44214 | 3018 | else if (ino > entry->ino) |
9f03740a | 3019 | n = n->rb_right; |
9dc44214 FM |
3020 | else |
3021 | return entry; | |
9f03740a | 3022 | } |
9dc44214 FM |
3023 | return NULL; |
3024 | } | |
3025 | ||
3026 | static void free_waiting_dir_move(struct send_ctx *sctx, | |
3027 | struct waiting_dir_move *dm) | |
3028 | { | |
3029 | if (!dm) | |
3030 | return; | |
3031 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
3032 | kfree(dm); | |
9f03740a FDBM |
3033 | } |
3034 | ||
bfa7e1f8 FM |
3035 | static int add_pending_dir_move(struct send_ctx *sctx, |
3036 | u64 ino, | |
3037 | u64 ino_gen, | |
f959492f FM |
3038 | u64 parent_ino, |
3039 | struct list_head *new_refs, | |
84471e24 FM |
3040 | struct list_head *deleted_refs, |
3041 | const bool is_orphan) | |
9f03740a FDBM |
3042 | { |
3043 | struct rb_node **p = &sctx->pending_dir_moves.rb_node; | |
3044 | struct rb_node *parent = NULL; | |
73b802f4 | 3045 | struct pending_dir_move *entry = NULL, *pm; |
9f03740a FDBM |
3046 | struct recorded_ref *cur; |
3047 | int exists = 0; | |
3048 | int ret; | |
3049 | ||
e780b0d1 | 3050 | pm = kmalloc(sizeof(*pm), GFP_KERNEL); |
9f03740a FDBM |
3051 | if (!pm) |
3052 | return -ENOMEM; | |
3053 | pm->parent_ino = parent_ino; | |
bfa7e1f8 FM |
3054 | pm->ino = ino; |
3055 | pm->gen = ino_gen; | |
9f03740a FDBM |
3056 | INIT_LIST_HEAD(&pm->list); |
3057 | INIT_LIST_HEAD(&pm->update_refs); | |
3058 | RB_CLEAR_NODE(&pm->node); | |
3059 | ||
3060 | while (*p) { | |
3061 | parent = *p; | |
3062 | entry = rb_entry(parent, struct pending_dir_move, node); | |
3063 | if (parent_ino < entry->parent_ino) { | |
3064 | p = &(*p)->rb_left; | |
3065 | } else if (parent_ino > entry->parent_ino) { | |
3066 | p = &(*p)->rb_right; | |
3067 | } else { | |
3068 | exists = 1; | |
3069 | break; | |
3070 | } | |
3071 | } | |
3072 | ||
f959492f | 3073 | list_for_each_entry(cur, deleted_refs, list) { |
9f03740a FDBM |
3074 | ret = dup_ref(cur, &pm->update_refs); |
3075 | if (ret < 0) | |
3076 | goto out; | |
3077 | } | |
f959492f | 3078 | list_for_each_entry(cur, new_refs, list) { |
9f03740a FDBM |
3079 | ret = dup_ref(cur, &pm->update_refs); |
3080 | if (ret < 0) | |
3081 | goto out; | |
3082 | } | |
3083 | ||
8b191a68 | 3084 | ret = add_waiting_dir_move(sctx, pm->ino, is_orphan); |
9f03740a FDBM |
3085 | if (ret) |
3086 | goto out; | |
3087 | ||
3088 | if (exists) { | |
3089 | list_add_tail(&pm->list, &entry->list); | |
3090 | } else { | |
3091 | rb_link_node(&pm->node, parent, p); | |
3092 | rb_insert_color(&pm->node, &sctx->pending_dir_moves); | |
3093 | } | |
3094 | ret = 0; | |
3095 | out: | |
3096 | if (ret) { | |
3097 | __free_recorded_refs(&pm->update_refs); | |
3098 | kfree(pm); | |
3099 | } | |
3100 | return ret; | |
3101 | } | |
3102 | ||
3103 | static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, | |
3104 | u64 parent_ino) | |
3105 | { | |
3106 | struct rb_node *n = sctx->pending_dir_moves.rb_node; | |
3107 | struct pending_dir_move *entry; | |
3108 | ||
3109 | while (n) { | |
3110 | entry = rb_entry(n, struct pending_dir_move, node); | |
3111 | if (parent_ino < entry->parent_ino) | |
3112 | n = n->rb_left; | |
3113 | else if (parent_ino > entry->parent_ino) | |
3114 | n = n->rb_right; | |
3115 | else | |
3116 | return entry; | |
3117 | } | |
3118 | return NULL; | |
3119 | } | |
3120 | ||
801bec36 RK |
3121 | static int path_loop(struct send_ctx *sctx, struct fs_path *name, |
3122 | u64 ino, u64 gen, u64 *ancestor_ino) | |
3123 | { | |
3124 | int ret = 0; | |
3125 | u64 parent_inode = 0; | |
3126 | u64 parent_gen = 0; | |
3127 | u64 start_ino = ino; | |
3128 | ||
3129 | *ancestor_ino = 0; | |
3130 | while (ino != BTRFS_FIRST_FREE_OBJECTID) { | |
3131 | fs_path_reset(name); | |
3132 | ||
3133 | if (is_waiting_for_rm(sctx, ino)) | |
3134 | break; | |
3135 | if (is_waiting_for_move(sctx, ino)) { | |
3136 | if (*ancestor_ino == 0) | |
3137 | *ancestor_ino = ino; | |
3138 | ret = get_first_ref(sctx->parent_root, ino, | |
3139 | &parent_inode, &parent_gen, name); | |
3140 | } else { | |
3141 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
3142 | &parent_inode, | |
3143 | &parent_gen, name); | |
3144 | if (ret > 0) { | |
3145 | ret = 0; | |
3146 | break; | |
3147 | } | |
3148 | } | |
3149 | if (ret < 0) | |
3150 | break; | |
3151 | if (parent_inode == start_ino) { | |
3152 | ret = 1; | |
3153 | if (*ancestor_ino == 0) | |
3154 | *ancestor_ino = ino; | |
3155 | break; | |
3156 | } | |
3157 | ino = parent_inode; | |
3158 | gen = parent_gen; | |
3159 | } | |
3160 | return ret; | |
3161 | } | |
3162 | ||
9f03740a FDBM |
3163 | static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) |
3164 | { | |
3165 | struct fs_path *from_path = NULL; | |
3166 | struct fs_path *to_path = NULL; | |
2b863a13 | 3167 | struct fs_path *name = NULL; |
9f03740a FDBM |
3168 | u64 orig_progress = sctx->send_progress; |
3169 | struct recorded_ref *cur; | |
2b863a13 | 3170 | u64 parent_ino, parent_gen; |
9dc44214 FM |
3171 | struct waiting_dir_move *dm = NULL; |
3172 | u64 rmdir_ino = 0; | |
801bec36 RK |
3173 | u64 ancestor; |
3174 | bool is_orphan; | |
9f03740a FDBM |
3175 | int ret; |
3176 | ||
2b863a13 | 3177 | name = fs_path_alloc(); |
9f03740a | 3178 | from_path = fs_path_alloc(); |
2b863a13 FM |
3179 | if (!name || !from_path) { |
3180 | ret = -ENOMEM; | |
3181 | goto out; | |
3182 | } | |
9f03740a | 3183 | |
9dc44214 FM |
3184 | dm = get_waiting_dir_move(sctx, pm->ino); |
3185 | ASSERT(dm); | |
3186 | rmdir_ino = dm->rmdir_ino; | |
801bec36 | 3187 | is_orphan = dm->orphanized; |
9dc44214 | 3188 | free_waiting_dir_move(sctx, dm); |
2b863a13 | 3189 | |
801bec36 | 3190 | if (is_orphan) { |
84471e24 FM |
3191 | ret = gen_unique_name(sctx, pm->ino, |
3192 | pm->gen, from_path); | |
3193 | } else { | |
3194 | ret = get_first_ref(sctx->parent_root, pm->ino, | |
3195 | &parent_ino, &parent_gen, name); | |
3196 | if (ret < 0) | |
3197 | goto out; | |
3198 | ret = get_cur_path(sctx, parent_ino, parent_gen, | |
3199 | from_path); | |
3200 | if (ret < 0) | |
3201 | goto out; | |
3202 | ret = fs_path_add_path(from_path, name); | |
3203 | } | |
c992ec94 FM |
3204 | if (ret < 0) |
3205 | goto out; | |
2b863a13 | 3206 | |
f959492f | 3207 | sctx->send_progress = sctx->cur_ino + 1; |
801bec36 | 3208 | ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor); |
7969e77a FM |
3209 | if (ret < 0) |
3210 | goto out; | |
801bec36 RK |
3211 | if (ret) { |
3212 | LIST_HEAD(deleted_refs); | |
3213 | ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID); | |
3214 | ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor, | |
3215 | &pm->update_refs, &deleted_refs, | |
3216 | is_orphan); | |
3217 | if (ret < 0) | |
3218 | goto out; | |
3219 | if (rmdir_ino) { | |
3220 | dm = get_waiting_dir_move(sctx, pm->ino); | |
3221 | ASSERT(dm); | |
3222 | dm->rmdir_ino = rmdir_ino; | |
3223 | } | |
3224 | goto out; | |
3225 | } | |
c992ec94 FM |
3226 | fs_path_reset(name); |
3227 | to_path = name; | |
2b863a13 | 3228 | name = NULL; |
9f03740a FDBM |
3229 | ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); |
3230 | if (ret < 0) | |
3231 | goto out; | |
3232 | ||
3233 | ret = send_rename(sctx, from_path, to_path); | |
3234 | if (ret < 0) | |
3235 | goto out; | |
3236 | ||
9dc44214 FM |
3237 | if (rmdir_ino) { |
3238 | struct orphan_dir_info *odi; | |
3239 | ||
3240 | odi = get_orphan_dir_info(sctx, rmdir_ino); | |
3241 | if (!odi) { | |
3242 | /* already deleted */ | |
3243 | goto finish; | |
3244 | } | |
99ea42dd | 3245 | ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino); |
9dc44214 FM |
3246 | if (ret < 0) |
3247 | goto out; | |
3248 | if (!ret) | |
3249 | goto finish; | |
3250 | ||
3251 | name = fs_path_alloc(); | |
3252 | if (!name) { | |
3253 | ret = -ENOMEM; | |
3254 | goto out; | |
3255 | } | |
3256 | ret = get_cur_path(sctx, rmdir_ino, odi->gen, name); | |
3257 | if (ret < 0) | |
3258 | goto out; | |
3259 | ret = send_rmdir(sctx, name); | |
3260 | if (ret < 0) | |
3261 | goto out; | |
3262 | free_orphan_dir_info(sctx, odi); | |
3263 | } | |
3264 | ||
3265 | finish: | |
9f03740a FDBM |
3266 | ret = send_utimes(sctx, pm->ino, pm->gen); |
3267 | if (ret < 0) | |
3268 | goto out; | |
3269 | ||
3270 | /* | |
3271 | * After rename/move, need to update the utimes of both new parent(s) | |
3272 | * and old parent(s). | |
3273 | */ | |
3274 | list_for_each_entry(cur, &pm->update_refs, list) { | |
764433a1 RK |
3275 | /* |
3276 | * The parent inode might have been deleted in the send snapshot | |
3277 | */ | |
3278 | ret = get_inode_info(sctx->send_root, cur->dir, NULL, | |
3279 | NULL, NULL, NULL, NULL, NULL); | |
3280 | if (ret == -ENOENT) { | |
3281 | ret = 0; | |
9dc44214 | 3282 | continue; |
764433a1 RK |
3283 | } |
3284 | if (ret < 0) | |
3285 | goto out; | |
3286 | ||
9f03740a FDBM |
3287 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
3288 | if (ret < 0) | |
3289 | goto out; | |
3290 | } | |
3291 | ||
3292 | out: | |
2b863a13 | 3293 | fs_path_free(name); |
9f03740a FDBM |
3294 | fs_path_free(from_path); |
3295 | fs_path_free(to_path); | |
3296 | sctx->send_progress = orig_progress; | |
3297 | ||
3298 | return ret; | |
3299 | } | |
3300 | ||
3301 | static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) | |
3302 | { | |
3303 | if (!list_empty(&m->list)) | |
3304 | list_del(&m->list); | |
3305 | if (!RB_EMPTY_NODE(&m->node)) | |
3306 | rb_erase(&m->node, &sctx->pending_dir_moves); | |
3307 | __free_recorded_refs(&m->update_refs); | |
3308 | kfree(m); | |
3309 | } | |
3310 | ||
3311 | static void tail_append_pending_moves(struct pending_dir_move *moves, | |
3312 | struct list_head *stack) | |
3313 | { | |
3314 | if (list_empty(&moves->list)) { | |
3315 | list_add_tail(&moves->list, stack); | |
3316 | } else { | |
3317 | LIST_HEAD(list); | |
3318 | list_splice_init(&moves->list, &list); | |
3319 | list_add_tail(&moves->list, stack); | |
3320 | list_splice_tail(&list, stack); | |
3321 | } | |
3322 | } | |
3323 | ||
3324 | static int apply_children_dir_moves(struct send_ctx *sctx) | |
3325 | { | |
3326 | struct pending_dir_move *pm; | |
3327 | struct list_head stack; | |
3328 | u64 parent_ino = sctx->cur_ino; | |
3329 | int ret = 0; | |
3330 | ||
3331 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3332 | if (!pm) | |
3333 | return 0; | |
3334 | ||
3335 | INIT_LIST_HEAD(&stack); | |
3336 | tail_append_pending_moves(pm, &stack); | |
3337 | ||
3338 | while (!list_empty(&stack)) { | |
3339 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3340 | parent_ino = pm->ino; | |
3341 | ret = apply_dir_move(sctx, pm); | |
3342 | free_pending_move(sctx, pm); | |
3343 | if (ret) | |
3344 | goto out; | |
3345 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3346 | if (pm) | |
3347 | tail_append_pending_moves(pm, &stack); | |
3348 | } | |
3349 | return 0; | |
3350 | ||
3351 | out: | |
3352 | while (!list_empty(&stack)) { | |
3353 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3354 | free_pending_move(sctx, pm); | |
3355 | } | |
3356 | return ret; | |
3357 | } | |
3358 | ||
84471e24 FM |
3359 | /* |
3360 | * We might need to delay a directory rename even when no ancestor directory | |
3361 | * (in the send root) with a higher inode number than ours (sctx->cur_ino) was | |
3362 | * renamed. This happens when we rename a directory to the old name (the name | |
3363 | * in the parent root) of some other unrelated directory that got its rename | |
3364 | * delayed due to some ancestor with higher number that got renamed. | |
3365 | * | |
3366 | * Example: | |
3367 | * | |
3368 | * Parent snapshot: | |
3369 | * . (ino 256) | |
3370 | * |---- a/ (ino 257) | |
3371 | * | |---- file (ino 260) | |
3372 | * | | |
3373 | * |---- b/ (ino 258) | |
3374 | * |---- c/ (ino 259) | |
3375 | * | |
3376 | * Send snapshot: | |
3377 | * . (ino 256) | |
3378 | * |---- a/ (ino 258) | |
3379 | * |---- x/ (ino 259) | |
3380 | * |---- y/ (ino 257) | |
3381 | * |----- file (ino 260) | |
3382 | * | |
3383 | * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257 | |
3384 | * from 'a' to 'x/y' happening first, which in turn depends on the rename of | |
3385 | * inode 259 from 'c' to 'x'. So the order of rename commands the send stream | |
3386 | * must issue is: | |
3387 | * | |
3388 | * 1 - rename 259 from 'c' to 'x' | |
3389 | * 2 - rename 257 from 'a' to 'x/y' | |
3390 | * 3 - rename 258 from 'b' to 'a' | |
3391 | * | |
3392 | * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can | |
3393 | * be done right away and < 0 on error. | |
3394 | */ | |
3395 | static int wait_for_dest_dir_move(struct send_ctx *sctx, | |
3396 | struct recorded_ref *parent_ref, | |
3397 | const bool is_orphan) | |
3398 | { | |
3399 | struct btrfs_path *path; | |
3400 | struct btrfs_key key; | |
3401 | struct btrfs_key di_key; | |
3402 | struct btrfs_dir_item *di; | |
3403 | u64 left_gen; | |
3404 | u64 right_gen; | |
3405 | int ret = 0; | |
801bec36 | 3406 | struct waiting_dir_move *wdm; |
84471e24 FM |
3407 | |
3408 | if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) | |
3409 | return 0; | |
3410 | ||
3411 | path = alloc_path_for_send(); | |
3412 | if (!path) | |
3413 | return -ENOMEM; | |
3414 | ||
3415 | key.objectid = parent_ref->dir; | |
3416 | key.type = BTRFS_DIR_ITEM_KEY; | |
3417 | key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); | |
3418 | ||
3419 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
3420 | if (ret < 0) { | |
3421 | goto out; | |
3422 | } else if (ret > 0) { | |
3423 | ret = 0; | |
3424 | goto out; | |
3425 | } | |
3426 | ||
3427 | di = btrfs_match_dir_item_name(sctx->parent_root, path, | |
3428 | parent_ref->name, parent_ref->name_len); | |
3429 | if (!di) { | |
3430 | ret = 0; | |
3431 | goto out; | |
3432 | } | |
3433 | /* | |
3434 | * di_key.objectid has the number of the inode that has a dentry in the | |
3435 | * parent directory with the same name that sctx->cur_ino is being | |
3436 | * renamed to. We need to check if that inode is in the send root as | |
3437 | * well and if it is currently marked as an inode with a pending rename, | |
3438 | * if it is, we need to delay the rename of sctx->cur_ino as well, so | |
3439 | * that it happens after that other inode is renamed. | |
3440 | */ | |
3441 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); | |
3442 | if (di_key.type != BTRFS_INODE_ITEM_KEY) { | |
3443 | ret = 0; | |
3444 | goto out; | |
3445 | } | |
3446 | ||
3447 | ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL, | |
3448 | &left_gen, NULL, NULL, NULL, NULL); | |
3449 | if (ret < 0) | |
3450 | goto out; | |
3451 | ret = get_inode_info(sctx->send_root, di_key.objectid, NULL, | |
3452 | &right_gen, NULL, NULL, NULL, NULL); | |
3453 | if (ret < 0) { | |
3454 | if (ret == -ENOENT) | |
3455 | ret = 0; | |
3456 | goto out; | |
3457 | } | |
3458 | ||
3459 | /* Different inode, no need to delay the rename of sctx->cur_ino */ | |
3460 | if (right_gen != left_gen) { | |
3461 | ret = 0; | |
3462 | goto out; | |
3463 | } | |
3464 | ||
801bec36 RK |
3465 | wdm = get_waiting_dir_move(sctx, di_key.objectid); |
3466 | if (wdm && !wdm->orphanized) { | |
84471e24 FM |
3467 | ret = add_pending_dir_move(sctx, |
3468 | sctx->cur_ino, | |
3469 | sctx->cur_inode_gen, | |
3470 | di_key.objectid, | |
3471 | &sctx->new_refs, | |
3472 | &sctx->deleted_refs, | |
3473 | is_orphan); | |
3474 | if (!ret) | |
3475 | ret = 1; | |
3476 | } | |
3477 | out: | |
3478 | btrfs_free_path(path); | |
3479 | return ret; | |
3480 | } | |
3481 | ||
80aa6027 FM |
3482 | /* |
3483 | * Check if ino ino1 is an ancestor of inode ino2 in the given root. | |
3484 | * Return 1 if true, 0 if false and < 0 on error. | |
3485 | */ | |
3486 | static int is_ancestor(struct btrfs_root *root, | |
3487 | const u64 ino1, | |
3488 | const u64 ino1_gen, | |
3489 | const u64 ino2, | |
3490 | struct fs_path *fs_path) | |
3491 | { | |
3492 | u64 ino = ino2; | |
3493 | ||
3494 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { | |
3495 | int ret; | |
3496 | u64 parent; | |
3497 | u64 parent_gen; | |
3498 | ||
3499 | fs_path_reset(fs_path); | |
3500 | ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path); | |
3501 | if (ret < 0) { | |
3502 | if (ret == -ENOENT && ino == ino2) | |
3503 | ret = 0; | |
3504 | return ret; | |
3505 | } | |
3506 | if (parent == ino1) | |
3507 | return parent_gen == ino1_gen ? 1 : 0; | |
3508 | ino = parent; | |
3509 | } | |
3510 | return 0; | |
3511 | } | |
3512 | ||
9f03740a | 3513 | static int wait_for_parent_move(struct send_ctx *sctx, |
8b191a68 FM |
3514 | struct recorded_ref *parent_ref, |
3515 | const bool is_orphan) | |
9f03740a | 3516 | { |
f959492f | 3517 | int ret = 0; |
9f03740a FDBM |
3518 | u64 ino = parent_ref->dir; |
3519 | u64 parent_ino_before, parent_ino_after; | |
9f03740a FDBM |
3520 | struct fs_path *path_before = NULL; |
3521 | struct fs_path *path_after = NULL; | |
3522 | int len1, len2; | |
9f03740a FDBM |
3523 | |
3524 | path_after = fs_path_alloc(); | |
f959492f FM |
3525 | path_before = fs_path_alloc(); |
3526 | if (!path_after || !path_before) { | |
9f03740a FDBM |
3527 | ret = -ENOMEM; |
3528 | goto out; | |
3529 | } | |
3530 | ||
bfa7e1f8 | 3531 | /* |
f959492f FM |
3532 | * Our current directory inode may not yet be renamed/moved because some |
3533 | * ancestor (immediate or not) has to be renamed/moved first. So find if | |
3534 | * such ancestor exists and make sure our own rename/move happens after | |
80aa6027 FM |
3535 | * that ancestor is processed to avoid path build infinite loops (done |
3536 | * at get_cur_path()). | |
bfa7e1f8 | 3537 | */ |
f959492f FM |
3538 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { |
3539 | if (is_waiting_for_move(sctx, ino)) { | |
80aa6027 FM |
3540 | /* |
3541 | * If the current inode is an ancestor of ino in the | |
3542 | * parent root, we need to delay the rename of the | |
3543 | * current inode, otherwise don't delayed the rename | |
3544 | * because we can end up with a circular dependency | |
3545 | * of renames, resulting in some directories never | |
3546 | * getting the respective rename operations issued in | |
3547 | * the send stream or getting into infinite path build | |
3548 | * loops. | |
3549 | */ | |
3550 | ret = is_ancestor(sctx->parent_root, | |
3551 | sctx->cur_ino, sctx->cur_inode_gen, | |
3552 | ino, path_before); | |
4122ea64 FM |
3553 | if (ret) |
3554 | break; | |
f959492f | 3555 | } |
bfa7e1f8 FM |
3556 | |
3557 | fs_path_reset(path_before); | |
3558 | fs_path_reset(path_after); | |
3559 | ||
3560 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
f959492f | 3561 | NULL, path_after); |
bfa7e1f8 FM |
3562 | if (ret < 0) |
3563 | goto out; | |
3564 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
3565 | NULL, path_before); | |
f959492f | 3566 | if (ret < 0 && ret != -ENOENT) { |
bfa7e1f8 | 3567 | goto out; |
f959492f | 3568 | } else if (ret == -ENOENT) { |
bf8e8ca6 | 3569 | ret = 0; |
f959492f | 3570 | break; |
bfa7e1f8 FM |
3571 | } |
3572 | ||
3573 | len1 = fs_path_len(path_before); | |
3574 | len2 = fs_path_len(path_after); | |
f959492f FM |
3575 | if (ino > sctx->cur_ino && |
3576 | (parent_ino_before != parent_ino_after || len1 != len2 || | |
3577 | memcmp(path_before->start, path_after->start, len1))) { | |
bfa7e1f8 | 3578 | ret = 1; |
f959492f | 3579 | break; |
bfa7e1f8 | 3580 | } |
bfa7e1f8 | 3581 | ino = parent_ino_after; |
bfa7e1f8 FM |
3582 | } |
3583 | ||
9f03740a FDBM |
3584 | out: |
3585 | fs_path_free(path_before); | |
3586 | fs_path_free(path_after); | |
3587 | ||
f959492f FM |
3588 | if (ret == 1) { |
3589 | ret = add_pending_dir_move(sctx, | |
3590 | sctx->cur_ino, | |
3591 | sctx->cur_inode_gen, | |
3592 | ino, | |
3593 | &sctx->new_refs, | |
84471e24 | 3594 | &sctx->deleted_refs, |
8b191a68 | 3595 | is_orphan); |
f959492f FM |
3596 | if (!ret) |
3597 | ret = 1; | |
3598 | } | |
3599 | ||
9f03740a FDBM |
3600 | return ret; |
3601 | } | |
3602 | ||
31db9f7c AB |
3603 | /* |
3604 | * This does all the move/link/unlink/rmdir magic. | |
3605 | */ | |
9f03740a | 3606 | static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) |
31db9f7c AB |
3607 | { |
3608 | int ret = 0; | |
3609 | struct recorded_ref *cur; | |
1f4692da | 3610 | struct recorded_ref *cur2; |
ba5e8f2e | 3611 | struct list_head check_dirs; |
31db9f7c | 3612 | struct fs_path *valid_path = NULL; |
b24baf69 | 3613 | u64 ow_inode = 0; |
31db9f7c AB |
3614 | u64 ow_gen; |
3615 | int did_overwrite = 0; | |
3616 | int is_orphan = 0; | |
29d6d30f | 3617 | u64 last_dir_ino_rm = 0; |
84471e24 | 3618 | bool can_rename = true; |
31db9f7c AB |
3619 | |
3620 | verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino); | |
3621 | ||
6d85ed05 AB |
3622 | /* |
3623 | * This should never happen as the root dir always has the same ref | |
3624 | * which is always '..' | |
3625 | */ | |
3626 | BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); | |
ba5e8f2e | 3627 | INIT_LIST_HEAD(&check_dirs); |
6d85ed05 | 3628 | |
924794c9 | 3629 | valid_path = fs_path_alloc(); |
31db9f7c AB |
3630 | if (!valid_path) { |
3631 | ret = -ENOMEM; | |
3632 | goto out; | |
3633 | } | |
3634 | ||
31db9f7c AB |
3635 | /* |
3636 | * First, check if the first ref of the current inode was overwritten | |
3637 | * before. If yes, we know that the current inode was already orphanized | |
3638 | * and thus use the orphan name. If not, we can use get_cur_path to | |
3639 | * get the path of the first ref as it would like while receiving at | |
3640 | * this point in time. | |
3641 | * New inodes are always orphan at the beginning, so force to use the | |
3642 | * orphan name in this case. | |
3643 | * The first ref is stored in valid_path and will be updated if it | |
3644 | * gets moved around. | |
3645 | */ | |
3646 | if (!sctx->cur_inode_new) { | |
3647 | ret = did_overwrite_first_ref(sctx, sctx->cur_ino, | |
3648 | sctx->cur_inode_gen); | |
3649 | if (ret < 0) | |
3650 | goto out; | |
3651 | if (ret) | |
3652 | did_overwrite = 1; | |
3653 | } | |
3654 | if (sctx->cur_inode_new || did_overwrite) { | |
3655 | ret = gen_unique_name(sctx, sctx->cur_ino, | |
3656 | sctx->cur_inode_gen, valid_path); | |
3657 | if (ret < 0) | |
3658 | goto out; | |
3659 | is_orphan = 1; | |
3660 | } else { | |
3661 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
3662 | valid_path); | |
3663 | if (ret < 0) | |
3664 | goto out; | |
3665 | } | |
3666 | ||
3667 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
1f4692da AB |
3668 | /* |
3669 | * We may have refs where the parent directory does not exist | |
3670 | * yet. This happens if the parent directories inum is higher | |
3671 | * the the current inum. To handle this case, we create the | |
3672 | * parent directory out of order. But we need to check if this | |
3673 | * did already happen before due to other refs in the same dir. | |
3674 | */ | |
3675 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); | |
3676 | if (ret < 0) | |
3677 | goto out; | |
3678 | if (ret == inode_state_will_create) { | |
3679 | ret = 0; | |
3680 | /* | |
3681 | * First check if any of the current inodes refs did | |
3682 | * already create the dir. | |
3683 | */ | |
3684 | list_for_each_entry(cur2, &sctx->new_refs, list) { | |
3685 | if (cur == cur2) | |
3686 | break; | |
3687 | if (cur2->dir == cur->dir) { | |
3688 | ret = 1; | |
3689 | break; | |
3690 | } | |
3691 | } | |
3692 | ||
3693 | /* | |
3694 | * If that did not happen, check if a previous inode | |
3695 | * did already create the dir. | |
3696 | */ | |
3697 | if (!ret) | |
3698 | ret = did_create_dir(sctx, cur->dir); | |
3699 | if (ret < 0) | |
3700 | goto out; | |
3701 | if (!ret) { | |
3702 | ret = send_create_inode(sctx, cur->dir); | |
3703 | if (ret < 0) | |
3704 | goto out; | |
3705 | } | |
3706 | } | |
3707 | ||
31db9f7c AB |
3708 | /* |
3709 | * Check if this new ref would overwrite the first ref of | |
3710 | * another unprocessed inode. If yes, orphanize the | |
3711 | * overwritten inode. If we find an overwritten ref that is | |
3712 | * not the first ref, simply unlink it. | |
3713 | */ | |
3714 | ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
3715 | cur->name, cur->name_len, | |
3716 | &ow_inode, &ow_gen); | |
3717 | if (ret < 0) | |
3718 | goto out; | |
3719 | if (ret) { | |
924794c9 TI |
3720 | ret = is_first_ref(sctx->parent_root, |
3721 | ow_inode, cur->dir, cur->name, | |
3722 | cur->name_len); | |
31db9f7c AB |
3723 | if (ret < 0) |
3724 | goto out; | |
3725 | if (ret) { | |
8996a48c | 3726 | struct name_cache_entry *nce; |
801bec36 | 3727 | struct waiting_dir_move *wdm; |
8996a48c | 3728 | |
31db9f7c AB |
3729 | ret = orphanize_inode(sctx, ow_inode, ow_gen, |
3730 | cur->full_path); | |
3731 | if (ret < 0) | |
3732 | goto out; | |
801bec36 RK |
3733 | |
3734 | /* | |
3735 | * If ow_inode has its rename operation delayed | |
3736 | * make sure that its orphanized name is used in | |
3737 | * the source path when performing its rename | |
3738 | * operation. | |
3739 | */ | |
3740 | if (is_waiting_for_move(sctx, ow_inode)) { | |
3741 | wdm = get_waiting_dir_move(sctx, | |
3742 | ow_inode); | |
3743 | ASSERT(wdm); | |
3744 | wdm->orphanized = true; | |
3745 | } | |
3746 | ||
8996a48c FM |
3747 | /* |
3748 | * Make sure we clear our orphanized inode's | |
3749 | * name from the name cache. This is because the | |
3750 | * inode ow_inode might be an ancestor of some | |
3751 | * other inode that will be orphanized as well | |
3752 | * later and has an inode number greater than | |
3753 | * sctx->send_progress. We need to prevent | |
3754 | * future name lookups from using the old name | |
3755 | * and get instead the orphan name. | |
3756 | */ | |
3757 | nce = name_cache_search(sctx, ow_inode, ow_gen); | |
3758 | if (nce) { | |
3759 | name_cache_delete(sctx, nce); | |
3760 | kfree(nce); | |
3761 | } | |
801bec36 RK |
3762 | |
3763 | /* | |
3764 | * ow_inode might currently be an ancestor of | |
3765 | * cur_ino, therefore compute valid_path (the | |
3766 | * current path of cur_ino) again because it | |
3767 | * might contain the pre-orphanization name of | |
3768 | * ow_inode, which is no longer valid. | |
3769 | */ | |
3770 | fs_path_reset(valid_path); | |
3771 | ret = get_cur_path(sctx, sctx->cur_ino, | |
3772 | sctx->cur_inode_gen, valid_path); | |
3773 | if (ret < 0) | |
3774 | goto out; | |
31db9f7c AB |
3775 | } else { |
3776 | ret = send_unlink(sctx, cur->full_path); | |
3777 | if (ret < 0) | |
3778 | goto out; | |
3779 | } | |
3780 | } | |
3781 | ||
84471e24 FM |
3782 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { |
3783 | ret = wait_for_dest_dir_move(sctx, cur, is_orphan); | |
3784 | if (ret < 0) | |
3785 | goto out; | |
3786 | if (ret == 1) { | |
3787 | can_rename = false; | |
3788 | *pending_move = 1; | |
3789 | } | |
3790 | } | |
3791 | ||
8b191a68 FM |
3792 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root && |
3793 | can_rename) { | |
3794 | ret = wait_for_parent_move(sctx, cur, is_orphan); | |
3795 | if (ret < 0) | |
3796 | goto out; | |
3797 | if (ret == 1) { | |
3798 | can_rename = false; | |
3799 | *pending_move = 1; | |
3800 | } | |
3801 | } | |
3802 | ||
31db9f7c AB |
3803 | /* |
3804 | * link/move the ref to the new place. If we have an orphan | |
3805 | * inode, move it and update valid_path. If not, link or move | |
3806 | * it depending on the inode mode. | |
3807 | */ | |
84471e24 | 3808 | if (is_orphan && can_rename) { |
31db9f7c AB |
3809 | ret = send_rename(sctx, valid_path, cur->full_path); |
3810 | if (ret < 0) | |
3811 | goto out; | |
3812 | is_orphan = 0; | |
3813 | ret = fs_path_copy(valid_path, cur->full_path); | |
3814 | if (ret < 0) | |
3815 | goto out; | |
84471e24 | 3816 | } else if (can_rename) { |
31db9f7c AB |
3817 | if (S_ISDIR(sctx->cur_inode_mode)) { |
3818 | /* | |
3819 | * Dirs can't be linked, so move it. For moved | |
3820 | * dirs, we always have one new and one deleted | |
3821 | * ref. The deleted ref is ignored later. | |
3822 | */ | |
8b191a68 FM |
3823 | ret = send_rename(sctx, valid_path, |
3824 | cur->full_path); | |
3825 | if (!ret) | |
3826 | ret = fs_path_copy(valid_path, | |
3827 | cur->full_path); | |
31db9f7c AB |
3828 | if (ret < 0) |
3829 | goto out; | |
3830 | } else { | |
3831 | ret = send_link(sctx, cur->full_path, | |
3832 | valid_path); | |
3833 | if (ret < 0) | |
3834 | goto out; | |
3835 | } | |
3836 | } | |
ba5e8f2e | 3837 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
3838 | if (ret < 0) |
3839 | goto out; | |
3840 | } | |
3841 | ||
3842 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { | |
3843 | /* | |
3844 | * Check if we can already rmdir the directory. If not, | |
3845 | * orphanize it. For every dir item inside that gets deleted | |
3846 | * later, we do this check again and rmdir it then if possible. | |
3847 | * See the use of check_dirs for more details. | |
3848 | */ | |
9dc44214 FM |
3849 | ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
3850 | sctx->cur_ino); | |
31db9f7c AB |
3851 | if (ret < 0) |
3852 | goto out; | |
3853 | if (ret) { | |
3854 | ret = send_rmdir(sctx, valid_path); | |
3855 | if (ret < 0) | |
3856 | goto out; | |
3857 | } else if (!is_orphan) { | |
3858 | ret = orphanize_inode(sctx, sctx->cur_ino, | |
3859 | sctx->cur_inode_gen, valid_path); | |
3860 | if (ret < 0) | |
3861 | goto out; | |
3862 | is_orphan = 1; | |
3863 | } | |
3864 | ||
3865 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
ba5e8f2e | 3866 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
3867 | if (ret < 0) |
3868 | goto out; | |
3869 | } | |
ccf1626b AB |
3870 | } else if (S_ISDIR(sctx->cur_inode_mode) && |
3871 | !list_empty(&sctx->deleted_refs)) { | |
3872 | /* | |
3873 | * We have a moved dir. Add the old parent to check_dirs | |
3874 | */ | |
3875 | cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, | |
3876 | list); | |
ba5e8f2e | 3877 | ret = dup_ref(cur, &check_dirs); |
ccf1626b AB |
3878 | if (ret < 0) |
3879 | goto out; | |
31db9f7c AB |
3880 | } else if (!S_ISDIR(sctx->cur_inode_mode)) { |
3881 | /* | |
3882 | * We have a non dir inode. Go through all deleted refs and | |
3883 | * unlink them if they were not already overwritten by other | |
3884 | * inodes. | |
3885 | */ | |
3886 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
3887 | ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
3888 | sctx->cur_ino, sctx->cur_inode_gen, | |
3889 | cur->name, cur->name_len); | |
3890 | if (ret < 0) | |
3891 | goto out; | |
3892 | if (!ret) { | |
1f4692da AB |
3893 | ret = send_unlink(sctx, cur->full_path); |
3894 | if (ret < 0) | |
3895 | goto out; | |
31db9f7c | 3896 | } |
ba5e8f2e | 3897 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
3898 | if (ret < 0) |
3899 | goto out; | |
3900 | } | |
31db9f7c AB |
3901 | /* |
3902 | * If the inode is still orphan, unlink the orphan. This may | |
3903 | * happen when a previous inode did overwrite the first ref | |
3904 | * of this inode and no new refs were added for the current | |
766702ef AB |
3905 | * inode. Unlinking does not mean that the inode is deleted in |
3906 | * all cases. There may still be links to this inode in other | |
3907 | * places. | |
31db9f7c | 3908 | */ |
1f4692da | 3909 | if (is_orphan) { |
31db9f7c AB |
3910 | ret = send_unlink(sctx, valid_path); |
3911 | if (ret < 0) | |
3912 | goto out; | |
3913 | } | |
3914 | } | |
3915 | ||
3916 | /* | |
3917 | * We did collect all parent dirs where cur_inode was once located. We | |
3918 | * now go through all these dirs and check if they are pending for | |
3919 | * deletion and if it's finally possible to perform the rmdir now. | |
3920 | * We also update the inode stats of the parent dirs here. | |
3921 | */ | |
ba5e8f2e | 3922 | list_for_each_entry(cur, &check_dirs, list) { |
766702ef AB |
3923 | /* |
3924 | * In case we had refs into dirs that were not processed yet, | |
3925 | * we don't need to do the utime and rmdir logic for these dirs. | |
3926 | * The dir will be processed later. | |
3927 | */ | |
ba5e8f2e | 3928 | if (cur->dir > sctx->cur_ino) |
31db9f7c AB |
3929 | continue; |
3930 | ||
ba5e8f2e | 3931 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
3932 | if (ret < 0) |
3933 | goto out; | |
3934 | ||
3935 | if (ret == inode_state_did_create || | |
3936 | ret == inode_state_no_change) { | |
3937 | /* TODO delayed utimes */ | |
ba5e8f2e | 3938 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
3939 | if (ret < 0) |
3940 | goto out; | |
29d6d30f FM |
3941 | } else if (ret == inode_state_did_delete && |
3942 | cur->dir != last_dir_ino_rm) { | |
9dc44214 FM |
3943 | ret = can_rmdir(sctx, cur->dir, cur->dir_gen, |
3944 | sctx->cur_ino); | |
31db9f7c AB |
3945 | if (ret < 0) |
3946 | goto out; | |
3947 | if (ret) { | |
ba5e8f2e JB |
3948 | ret = get_cur_path(sctx, cur->dir, |
3949 | cur->dir_gen, valid_path); | |
31db9f7c AB |
3950 | if (ret < 0) |
3951 | goto out; | |
3952 | ret = send_rmdir(sctx, valid_path); | |
3953 | if (ret < 0) | |
3954 | goto out; | |
29d6d30f | 3955 | last_dir_ino_rm = cur->dir; |
31db9f7c AB |
3956 | } |
3957 | } | |
3958 | } | |
3959 | ||
31db9f7c AB |
3960 | ret = 0; |
3961 | ||
3962 | out: | |
ba5e8f2e | 3963 | __free_recorded_refs(&check_dirs); |
31db9f7c | 3964 | free_recorded_refs(sctx); |
924794c9 | 3965 | fs_path_free(valid_path); |
31db9f7c AB |
3966 | return ret; |
3967 | } | |
3968 | ||
a4d96d62 LB |
3969 | static int record_ref(struct btrfs_root *root, int num, u64 dir, int index, |
3970 | struct fs_path *name, void *ctx, struct list_head *refs) | |
31db9f7c AB |
3971 | { |
3972 | int ret = 0; | |
3973 | struct send_ctx *sctx = ctx; | |
3974 | struct fs_path *p; | |
3975 | u64 gen; | |
3976 | ||
924794c9 | 3977 | p = fs_path_alloc(); |
31db9f7c AB |
3978 | if (!p) |
3979 | return -ENOMEM; | |
3980 | ||
a4d96d62 | 3981 | ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, |
85a7b33b | 3982 | NULL, NULL); |
31db9f7c AB |
3983 | if (ret < 0) |
3984 | goto out; | |
3985 | ||
31db9f7c AB |
3986 | ret = get_cur_path(sctx, dir, gen, p); |
3987 | if (ret < 0) | |
3988 | goto out; | |
3989 | ret = fs_path_add_path(p, name); | |
3990 | if (ret < 0) | |
3991 | goto out; | |
3992 | ||
a4d96d62 | 3993 | ret = __record_ref(refs, dir, gen, p); |
31db9f7c AB |
3994 | |
3995 | out: | |
3996 | if (ret) | |
924794c9 | 3997 | fs_path_free(p); |
31db9f7c AB |
3998 | return ret; |
3999 | } | |
4000 | ||
a4d96d62 LB |
4001 | static int __record_new_ref(int num, u64 dir, int index, |
4002 | struct fs_path *name, | |
4003 | void *ctx) | |
4004 | { | |
4005 | struct send_ctx *sctx = ctx; | |
4006 | return record_ref(sctx->send_root, num, dir, index, name, | |
4007 | ctx, &sctx->new_refs); | |
4008 | } | |
4009 | ||
4010 | ||
31db9f7c AB |
4011 | static int __record_deleted_ref(int num, u64 dir, int index, |
4012 | struct fs_path *name, | |
4013 | void *ctx) | |
4014 | { | |
31db9f7c | 4015 | struct send_ctx *sctx = ctx; |
a4d96d62 LB |
4016 | return record_ref(sctx->parent_root, num, dir, index, name, |
4017 | ctx, &sctx->deleted_refs); | |
31db9f7c AB |
4018 | } |
4019 | ||
4020 | static int record_new_ref(struct send_ctx *sctx) | |
4021 | { | |
4022 | int ret; | |
4023 | ||
924794c9 TI |
4024 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
4025 | sctx->cmp_key, 0, __record_new_ref, sctx); | |
31db9f7c AB |
4026 | if (ret < 0) |
4027 | goto out; | |
4028 | ret = 0; | |
4029 | ||
4030 | out: | |
4031 | return ret; | |
4032 | } | |
4033 | ||
4034 | static int record_deleted_ref(struct send_ctx *sctx) | |
4035 | { | |
4036 | int ret; | |
4037 | ||
924794c9 TI |
4038 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
4039 | sctx->cmp_key, 0, __record_deleted_ref, sctx); | |
31db9f7c AB |
4040 | if (ret < 0) |
4041 | goto out; | |
4042 | ret = 0; | |
4043 | ||
4044 | out: | |
4045 | return ret; | |
4046 | } | |
4047 | ||
4048 | struct find_ref_ctx { | |
4049 | u64 dir; | |
ba5e8f2e JB |
4050 | u64 dir_gen; |
4051 | struct btrfs_root *root; | |
31db9f7c AB |
4052 | struct fs_path *name; |
4053 | int found_idx; | |
4054 | }; | |
4055 | ||
4056 | static int __find_iref(int num, u64 dir, int index, | |
4057 | struct fs_path *name, | |
4058 | void *ctx_) | |
4059 | { | |
4060 | struct find_ref_ctx *ctx = ctx_; | |
ba5e8f2e JB |
4061 | u64 dir_gen; |
4062 | int ret; | |
31db9f7c AB |
4063 | |
4064 | if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && | |
4065 | strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { | |
ba5e8f2e JB |
4066 | /* |
4067 | * To avoid doing extra lookups we'll only do this if everything | |
4068 | * else matches. | |
4069 | */ | |
4070 | ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, | |
4071 | NULL, NULL, NULL); | |
4072 | if (ret) | |
4073 | return ret; | |
4074 | if (dir_gen != ctx->dir_gen) | |
4075 | return 0; | |
31db9f7c AB |
4076 | ctx->found_idx = num; |
4077 | return 1; | |
4078 | } | |
4079 | return 0; | |
4080 | } | |
4081 | ||
924794c9 | 4082 | static int find_iref(struct btrfs_root *root, |
31db9f7c AB |
4083 | struct btrfs_path *path, |
4084 | struct btrfs_key *key, | |
ba5e8f2e | 4085 | u64 dir, u64 dir_gen, struct fs_path *name) |
31db9f7c AB |
4086 | { |
4087 | int ret; | |
4088 | struct find_ref_ctx ctx; | |
4089 | ||
4090 | ctx.dir = dir; | |
4091 | ctx.name = name; | |
ba5e8f2e | 4092 | ctx.dir_gen = dir_gen; |
31db9f7c | 4093 | ctx.found_idx = -1; |
ba5e8f2e | 4094 | ctx.root = root; |
31db9f7c | 4095 | |
924794c9 | 4096 | ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); |
31db9f7c AB |
4097 | if (ret < 0) |
4098 | return ret; | |
4099 | ||
4100 | if (ctx.found_idx == -1) | |
4101 | return -ENOENT; | |
4102 | ||
4103 | return ctx.found_idx; | |
4104 | } | |
4105 | ||
4106 | static int __record_changed_new_ref(int num, u64 dir, int index, | |
4107 | struct fs_path *name, | |
4108 | void *ctx) | |
4109 | { | |
ba5e8f2e | 4110 | u64 dir_gen; |
31db9f7c AB |
4111 | int ret; |
4112 | struct send_ctx *sctx = ctx; | |
4113 | ||
ba5e8f2e JB |
4114 | ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, |
4115 | NULL, NULL, NULL); | |
4116 | if (ret) | |
4117 | return ret; | |
4118 | ||
924794c9 | 4119 | ret = find_iref(sctx->parent_root, sctx->right_path, |
ba5e8f2e | 4120 | sctx->cmp_key, dir, dir_gen, name); |
31db9f7c AB |
4121 | if (ret == -ENOENT) |
4122 | ret = __record_new_ref(num, dir, index, name, sctx); | |
4123 | else if (ret > 0) | |
4124 | ret = 0; | |
4125 | ||
4126 | return ret; | |
4127 | } | |
4128 | ||
4129 | static int __record_changed_deleted_ref(int num, u64 dir, int index, | |
4130 | struct fs_path *name, | |
4131 | void *ctx) | |
4132 | { | |
ba5e8f2e | 4133 | u64 dir_gen; |
31db9f7c AB |
4134 | int ret; |
4135 | struct send_ctx *sctx = ctx; | |
4136 | ||
ba5e8f2e JB |
4137 | ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, |
4138 | NULL, NULL, NULL); | |
4139 | if (ret) | |
4140 | return ret; | |
4141 | ||
924794c9 | 4142 | ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, |
ba5e8f2e | 4143 | dir, dir_gen, name); |
31db9f7c AB |
4144 | if (ret == -ENOENT) |
4145 | ret = __record_deleted_ref(num, dir, index, name, sctx); | |
4146 | else if (ret > 0) | |
4147 | ret = 0; | |
4148 | ||
4149 | return ret; | |
4150 | } | |
4151 | ||
4152 | static int record_changed_ref(struct send_ctx *sctx) | |
4153 | { | |
4154 | int ret = 0; | |
4155 | ||
924794c9 | 4156 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
31db9f7c AB |
4157 | sctx->cmp_key, 0, __record_changed_new_ref, sctx); |
4158 | if (ret < 0) | |
4159 | goto out; | |
924794c9 | 4160 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
4161 | sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); |
4162 | if (ret < 0) | |
4163 | goto out; | |
4164 | ret = 0; | |
4165 | ||
4166 | out: | |
4167 | return ret; | |
4168 | } | |
4169 | ||
4170 | /* | |
4171 | * Record and process all refs at once. Needed when an inode changes the | |
4172 | * generation number, which means that it was deleted and recreated. | |
4173 | */ | |
4174 | static int process_all_refs(struct send_ctx *sctx, | |
4175 | enum btrfs_compare_tree_result cmd) | |
4176 | { | |
4177 | int ret; | |
4178 | struct btrfs_root *root; | |
4179 | struct btrfs_path *path; | |
4180 | struct btrfs_key key; | |
4181 | struct btrfs_key found_key; | |
4182 | struct extent_buffer *eb; | |
4183 | int slot; | |
4184 | iterate_inode_ref_t cb; | |
9f03740a | 4185 | int pending_move = 0; |
31db9f7c AB |
4186 | |
4187 | path = alloc_path_for_send(); | |
4188 | if (!path) | |
4189 | return -ENOMEM; | |
4190 | ||
4191 | if (cmd == BTRFS_COMPARE_TREE_NEW) { | |
4192 | root = sctx->send_root; | |
4193 | cb = __record_new_ref; | |
4194 | } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { | |
4195 | root = sctx->parent_root; | |
4196 | cb = __record_deleted_ref; | |
4197 | } else { | |
4d1a63b2 DS |
4198 | btrfs_err(sctx->send_root->fs_info, |
4199 | "Wrong command %d in process_all_refs", cmd); | |
4200 | ret = -EINVAL; | |
4201 | goto out; | |
31db9f7c AB |
4202 | } |
4203 | ||
4204 | key.objectid = sctx->cmp_key->objectid; | |
4205 | key.type = BTRFS_INODE_REF_KEY; | |
4206 | key.offset = 0; | |
dff6d0ad FDBM |
4207 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4208 | if (ret < 0) | |
4209 | goto out; | |
31db9f7c | 4210 | |
dff6d0ad | 4211 | while (1) { |
31db9f7c AB |
4212 | eb = path->nodes[0]; |
4213 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4214 | if (slot >= btrfs_header_nritems(eb)) { |
4215 | ret = btrfs_next_leaf(root, path); | |
4216 | if (ret < 0) | |
4217 | goto out; | |
4218 | else if (ret > 0) | |
4219 | break; | |
4220 | continue; | |
4221 | } | |
4222 | ||
31db9f7c AB |
4223 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
4224 | ||
4225 | if (found_key.objectid != key.objectid || | |
96b5bd77 JS |
4226 | (found_key.type != BTRFS_INODE_REF_KEY && |
4227 | found_key.type != BTRFS_INODE_EXTREF_KEY)) | |
31db9f7c | 4228 | break; |
31db9f7c | 4229 | |
924794c9 | 4230 | ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); |
31db9f7c AB |
4231 | if (ret < 0) |
4232 | goto out; | |
4233 | ||
dff6d0ad | 4234 | path->slots[0]++; |
31db9f7c | 4235 | } |
e938c8ad | 4236 | btrfs_release_path(path); |
31db9f7c | 4237 | |
9f03740a FDBM |
4238 | ret = process_recorded_refs(sctx, &pending_move); |
4239 | /* Only applicable to an incremental send. */ | |
4240 | ASSERT(pending_move == 0); | |
31db9f7c AB |
4241 | |
4242 | out: | |
4243 | btrfs_free_path(path); | |
4244 | return ret; | |
4245 | } | |
4246 | ||
4247 | static int send_set_xattr(struct send_ctx *sctx, | |
4248 | struct fs_path *path, | |
4249 | const char *name, int name_len, | |
4250 | const char *data, int data_len) | |
4251 | { | |
4252 | int ret = 0; | |
4253 | ||
4254 | ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); | |
4255 | if (ret < 0) | |
4256 | goto out; | |
4257 | ||
4258 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4259 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4260 | TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); | |
4261 | ||
4262 | ret = send_cmd(sctx); | |
4263 | ||
4264 | tlv_put_failure: | |
4265 | out: | |
4266 | return ret; | |
4267 | } | |
4268 | ||
4269 | static int send_remove_xattr(struct send_ctx *sctx, | |
4270 | struct fs_path *path, | |
4271 | const char *name, int name_len) | |
4272 | { | |
4273 | int ret = 0; | |
4274 | ||
4275 | ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); | |
4276 | if (ret < 0) | |
4277 | goto out; | |
4278 | ||
4279 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4280 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4281 | ||
4282 | ret = send_cmd(sctx); | |
4283 | ||
4284 | tlv_put_failure: | |
4285 | out: | |
4286 | return ret; | |
4287 | } | |
4288 | ||
4289 | static int __process_new_xattr(int num, struct btrfs_key *di_key, | |
4290 | const char *name, int name_len, | |
4291 | const char *data, int data_len, | |
4292 | u8 type, void *ctx) | |
4293 | { | |
4294 | int ret; | |
4295 | struct send_ctx *sctx = ctx; | |
4296 | struct fs_path *p; | |
4297 | posix_acl_xattr_header dummy_acl; | |
4298 | ||
924794c9 | 4299 | p = fs_path_alloc(); |
31db9f7c AB |
4300 | if (!p) |
4301 | return -ENOMEM; | |
4302 | ||
4303 | /* | |
01327610 | 4304 | * This hack is needed because empty acls are stored as zero byte |
31db9f7c | 4305 | * data in xattrs. Problem with that is, that receiving these zero byte |
01327610 | 4306 | * acls will fail later. To fix this, we send a dummy acl list that |
31db9f7c AB |
4307 | * only contains the version number and no entries. |
4308 | */ | |
4309 | if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || | |
4310 | !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { | |
4311 | if (data_len == 0) { | |
4312 | dummy_acl.a_version = | |
4313 | cpu_to_le32(POSIX_ACL_XATTR_VERSION); | |
4314 | data = (char *)&dummy_acl; | |
4315 | data_len = sizeof(dummy_acl); | |
4316 | } | |
4317 | } | |
4318 | ||
4319 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4320 | if (ret < 0) | |
4321 | goto out; | |
4322 | ||
4323 | ret = send_set_xattr(sctx, p, name, name_len, data, data_len); | |
4324 | ||
4325 | out: | |
924794c9 | 4326 | fs_path_free(p); |
31db9f7c AB |
4327 | return ret; |
4328 | } | |
4329 | ||
4330 | static int __process_deleted_xattr(int num, struct btrfs_key *di_key, | |
4331 | const char *name, int name_len, | |
4332 | const char *data, int data_len, | |
4333 | u8 type, void *ctx) | |
4334 | { | |
4335 | int ret; | |
4336 | struct send_ctx *sctx = ctx; | |
4337 | struct fs_path *p; | |
4338 | ||
924794c9 | 4339 | p = fs_path_alloc(); |
31db9f7c AB |
4340 | if (!p) |
4341 | return -ENOMEM; | |
4342 | ||
4343 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4344 | if (ret < 0) | |
4345 | goto out; | |
4346 | ||
4347 | ret = send_remove_xattr(sctx, p, name, name_len); | |
4348 | ||
4349 | out: | |
924794c9 | 4350 | fs_path_free(p); |
31db9f7c AB |
4351 | return ret; |
4352 | } | |
4353 | ||
4354 | static int process_new_xattr(struct send_ctx *sctx) | |
4355 | { | |
4356 | int ret = 0; | |
4357 | ||
924794c9 TI |
4358 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
4359 | sctx->cmp_key, __process_new_xattr, sctx); | |
31db9f7c AB |
4360 | |
4361 | return ret; | |
4362 | } | |
4363 | ||
4364 | static int process_deleted_xattr(struct send_ctx *sctx) | |
4365 | { | |
4366 | int ret; | |
4367 | ||
924794c9 TI |
4368 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
4369 | sctx->cmp_key, __process_deleted_xattr, sctx); | |
31db9f7c AB |
4370 | |
4371 | return ret; | |
4372 | } | |
4373 | ||
4374 | struct find_xattr_ctx { | |
4375 | const char *name; | |
4376 | int name_len; | |
4377 | int found_idx; | |
4378 | char *found_data; | |
4379 | int found_data_len; | |
4380 | }; | |
4381 | ||
4382 | static int __find_xattr(int num, struct btrfs_key *di_key, | |
4383 | const char *name, int name_len, | |
4384 | const char *data, int data_len, | |
4385 | u8 type, void *vctx) | |
4386 | { | |
4387 | struct find_xattr_ctx *ctx = vctx; | |
4388 | ||
4389 | if (name_len == ctx->name_len && | |
4390 | strncmp(name, ctx->name, name_len) == 0) { | |
4391 | ctx->found_idx = num; | |
4392 | ctx->found_data_len = data_len; | |
e780b0d1 | 4393 | ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); |
31db9f7c AB |
4394 | if (!ctx->found_data) |
4395 | return -ENOMEM; | |
31db9f7c AB |
4396 | return 1; |
4397 | } | |
4398 | return 0; | |
4399 | } | |
4400 | ||
924794c9 | 4401 | static int find_xattr(struct btrfs_root *root, |
31db9f7c AB |
4402 | struct btrfs_path *path, |
4403 | struct btrfs_key *key, | |
4404 | const char *name, int name_len, | |
4405 | char **data, int *data_len) | |
4406 | { | |
4407 | int ret; | |
4408 | struct find_xattr_ctx ctx; | |
4409 | ||
4410 | ctx.name = name; | |
4411 | ctx.name_len = name_len; | |
4412 | ctx.found_idx = -1; | |
4413 | ctx.found_data = NULL; | |
4414 | ctx.found_data_len = 0; | |
4415 | ||
924794c9 | 4416 | ret = iterate_dir_item(root, path, key, __find_xattr, &ctx); |
31db9f7c AB |
4417 | if (ret < 0) |
4418 | return ret; | |
4419 | ||
4420 | if (ctx.found_idx == -1) | |
4421 | return -ENOENT; | |
4422 | if (data) { | |
4423 | *data = ctx.found_data; | |
4424 | *data_len = ctx.found_data_len; | |
4425 | } else { | |
4426 | kfree(ctx.found_data); | |
4427 | } | |
4428 | return ctx.found_idx; | |
4429 | } | |
4430 | ||
4431 | ||
4432 | static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, | |
4433 | const char *name, int name_len, | |
4434 | const char *data, int data_len, | |
4435 | u8 type, void *ctx) | |
4436 | { | |
4437 | int ret; | |
4438 | struct send_ctx *sctx = ctx; | |
4439 | char *found_data = NULL; | |
4440 | int found_data_len = 0; | |
31db9f7c | 4441 | |
924794c9 TI |
4442 | ret = find_xattr(sctx->parent_root, sctx->right_path, |
4443 | sctx->cmp_key, name, name_len, &found_data, | |
4444 | &found_data_len); | |
31db9f7c AB |
4445 | if (ret == -ENOENT) { |
4446 | ret = __process_new_xattr(num, di_key, name, name_len, data, | |
4447 | data_len, type, ctx); | |
4448 | } else if (ret >= 0) { | |
4449 | if (data_len != found_data_len || | |
4450 | memcmp(data, found_data, data_len)) { | |
4451 | ret = __process_new_xattr(num, di_key, name, name_len, | |
4452 | data, data_len, type, ctx); | |
4453 | } else { | |
4454 | ret = 0; | |
4455 | } | |
4456 | } | |
4457 | ||
4458 | kfree(found_data); | |
31db9f7c AB |
4459 | return ret; |
4460 | } | |
4461 | ||
4462 | static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, | |
4463 | const char *name, int name_len, | |
4464 | const char *data, int data_len, | |
4465 | u8 type, void *ctx) | |
4466 | { | |
4467 | int ret; | |
4468 | struct send_ctx *sctx = ctx; | |
4469 | ||
924794c9 TI |
4470 | ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, |
4471 | name, name_len, NULL, NULL); | |
31db9f7c AB |
4472 | if (ret == -ENOENT) |
4473 | ret = __process_deleted_xattr(num, di_key, name, name_len, data, | |
4474 | data_len, type, ctx); | |
4475 | else if (ret >= 0) | |
4476 | ret = 0; | |
4477 | ||
4478 | return ret; | |
4479 | } | |
4480 | ||
4481 | static int process_changed_xattr(struct send_ctx *sctx) | |
4482 | { | |
4483 | int ret = 0; | |
4484 | ||
924794c9 | 4485 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
31db9f7c AB |
4486 | sctx->cmp_key, __process_changed_new_xattr, sctx); |
4487 | if (ret < 0) | |
4488 | goto out; | |
924794c9 | 4489 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
4490 | sctx->cmp_key, __process_changed_deleted_xattr, sctx); |
4491 | ||
4492 | out: | |
4493 | return ret; | |
4494 | } | |
4495 | ||
4496 | static int process_all_new_xattrs(struct send_ctx *sctx) | |
4497 | { | |
4498 | int ret; | |
4499 | struct btrfs_root *root; | |
4500 | struct btrfs_path *path; | |
4501 | struct btrfs_key key; | |
4502 | struct btrfs_key found_key; | |
4503 | struct extent_buffer *eb; | |
4504 | int slot; | |
4505 | ||
4506 | path = alloc_path_for_send(); | |
4507 | if (!path) | |
4508 | return -ENOMEM; | |
4509 | ||
4510 | root = sctx->send_root; | |
4511 | ||
4512 | key.objectid = sctx->cmp_key->objectid; | |
4513 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4514 | key.offset = 0; | |
dff6d0ad FDBM |
4515 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4516 | if (ret < 0) | |
4517 | goto out; | |
31db9f7c | 4518 | |
dff6d0ad | 4519 | while (1) { |
31db9f7c AB |
4520 | eb = path->nodes[0]; |
4521 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4522 | if (slot >= btrfs_header_nritems(eb)) { |
4523 | ret = btrfs_next_leaf(root, path); | |
4524 | if (ret < 0) { | |
4525 | goto out; | |
4526 | } else if (ret > 0) { | |
4527 | ret = 0; | |
4528 | break; | |
4529 | } | |
4530 | continue; | |
4531 | } | |
31db9f7c | 4532 | |
dff6d0ad | 4533 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
31db9f7c AB |
4534 | if (found_key.objectid != key.objectid || |
4535 | found_key.type != key.type) { | |
4536 | ret = 0; | |
4537 | goto out; | |
4538 | } | |
4539 | ||
924794c9 TI |
4540 | ret = iterate_dir_item(root, path, &found_key, |
4541 | __process_new_xattr, sctx); | |
31db9f7c AB |
4542 | if (ret < 0) |
4543 | goto out; | |
4544 | ||
dff6d0ad | 4545 | path->slots[0]++; |
31db9f7c AB |
4546 | } |
4547 | ||
4548 | out: | |
4549 | btrfs_free_path(path); | |
4550 | return ret; | |
4551 | } | |
4552 | ||
ed259095 JB |
4553 | static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len) |
4554 | { | |
4555 | struct btrfs_root *root = sctx->send_root; | |
4556 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4557 | struct inode *inode; | |
4558 | struct page *page; | |
4559 | char *addr; | |
4560 | struct btrfs_key key; | |
09cbfeaf | 4561 | pgoff_t index = offset >> PAGE_SHIFT; |
ed259095 | 4562 | pgoff_t last_index; |
09cbfeaf | 4563 | unsigned pg_offset = offset & ~PAGE_MASK; |
ed259095 JB |
4564 | ssize_t ret = 0; |
4565 | ||
4566 | key.objectid = sctx->cur_ino; | |
4567 | key.type = BTRFS_INODE_ITEM_KEY; | |
4568 | key.offset = 0; | |
4569 | ||
4570 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
4571 | if (IS_ERR(inode)) | |
4572 | return PTR_ERR(inode); | |
4573 | ||
4574 | if (offset + len > i_size_read(inode)) { | |
4575 | if (offset > i_size_read(inode)) | |
4576 | len = 0; | |
4577 | else | |
4578 | len = offset - i_size_read(inode); | |
4579 | } | |
4580 | if (len == 0) | |
4581 | goto out; | |
4582 | ||
09cbfeaf | 4583 | last_index = (offset + len - 1) >> PAGE_SHIFT; |
2131bcd3 LB |
4584 | |
4585 | /* initial readahead */ | |
4586 | memset(&sctx->ra, 0, sizeof(struct file_ra_state)); | |
4587 | file_ra_state_init(&sctx->ra, inode->i_mapping); | |
4588 | btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index, | |
4589 | last_index - index + 1); | |
4590 | ||
ed259095 JB |
4591 | while (index <= last_index) { |
4592 | unsigned cur_len = min_t(unsigned, len, | |
09cbfeaf | 4593 | PAGE_SIZE - pg_offset); |
e780b0d1 | 4594 | page = find_or_create_page(inode->i_mapping, index, GFP_KERNEL); |
ed259095 JB |
4595 | if (!page) { |
4596 | ret = -ENOMEM; | |
4597 | break; | |
4598 | } | |
4599 | ||
4600 | if (!PageUptodate(page)) { | |
4601 | btrfs_readpage(NULL, page); | |
4602 | lock_page(page); | |
4603 | if (!PageUptodate(page)) { | |
4604 | unlock_page(page); | |
09cbfeaf | 4605 | put_page(page); |
ed259095 JB |
4606 | ret = -EIO; |
4607 | break; | |
4608 | } | |
4609 | } | |
4610 | ||
4611 | addr = kmap(page); | |
4612 | memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len); | |
4613 | kunmap(page); | |
4614 | unlock_page(page); | |
09cbfeaf | 4615 | put_page(page); |
ed259095 JB |
4616 | index++; |
4617 | pg_offset = 0; | |
4618 | len -= cur_len; | |
4619 | ret += cur_len; | |
4620 | } | |
4621 | out: | |
4622 | iput(inode); | |
4623 | return ret; | |
4624 | } | |
4625 | ||
31db9f7c AB |
4626 | /* |
4627 | * Read some bytes from the current inode/file and send a write command to | |
4628 | * user space. | |
4629 | */ | |
4630 | static int send_write(struct send_ctx *sctx, u64 offset, u32 len) | |
4631 | { | |
4632 | int ret = 0; | |
4633 | struct fs_path *p; | |
ed259095 | 4634 | ssize_t num_read = 0; |
31db9f7c | 4635 | |
924794c9 | 4636 | p = fs_path_alloc(); |
31db9f7c AB |
4637 | if (!p) |
4638 | return -ENOMEM; | |
4639 | ||
31db9f7c AB |
4640 | verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len); |
4641 | ||
ed259095 JB |
4642 | num_read = fill_read_buf(sctx, offset, len); |
4643 | if (num_read <= 0) { | |
4644 | if (num_read < 0) | |
4645 | ret = num_read; | |
31db9f7c | 4646 | goto out; |
ed259095 | 4647 | } |
31db9f7c AB |
4648 | |
4649 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
4650 | if (ret < 0) | |
4651 | goto out; | |
4652 | ||
4653 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4654 | if (ret < 0) | |
4655 | goto out; | |
4656 | ||
4657 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4658 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
e938c8ad | 4659 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read); |
31db9f7c AB |
4660 | |
4661 | ret = send_cmd(sctx); | |
4662 | ||
4663 | tlv_put_failure: | |
4664 | out: | |
924794c9 | 4665 | fs_path_free(p); |
31db9f7c AB |
4666 | if (ret < 0) |
4667 | return ret; | |
e938c8ad | 4668 | return num_read; |
31db9f7c AB |
4669 | } |
4670 | ||
4671 | /* | |
4672 | * Send a clone command to user space. | |
4673 | */ | |
4674 | static int send_clone(struct send_ctx *sctx, | |
4675 | u64 offset, u32 len, | |
4676 | struct clone_root *clone_root) | |
4677 | { | |
4678 | int ret = 0; | |
31db9f7c AB |
4679 | struct fs_path *p; |
4680 | u64 gen; | |
4681 | ||
4682 | verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, " | |
4683 | "clone_inode=%llu, clone_offset=%llu\n", offset, len, | |
4684 | clone_root->root->objectid, clone_root->ino, | |
4685 | clone_root->offset); | |
4686 | ||
924794c9 | 4687 | p = fs_path_alloc(); |
31db9f7c AB |
4688 | if (!p) |
4689 | return -ENOMEM; | |
4690 | ||
4691 | ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); | |
4692 | if (ret < 0) | |
4693 | goto out; | |
4694 | ||
4695 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4696 | if (ret < 0) | |
4697 | goto out; | |
4698 | ||
4699 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4700 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); | |
4701 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4702 | ||
e938c8ad | 4703 | if (clone_root->root == sctx->send_root) { |
31db9f7c | 4704 | ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, |
85a7b33b | 4705 | &gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
4706 | if (ret < 0) |
4707 | goto out; | |
4708 | ret = get_cur_path(sctx, clone_root->ino, gen, p); | |
4709 | } else { | |
924794c9 | 4710 | ret = get_inode_path(clone_root->root, clone_root->ino, p); |
31db9f7c AB |
4711 | } |
4712 | if (ret < 0) | |
4713 | goto out; | |
4714 | ||
37b8d27d JB |
4715 | /* |
4716 | * If the parent we're using has a received_uuid set then use that as | |
4717 | * our clone source as that is what we will look for when doing a | |
4718 | * receive. | |
4719 | * | |
4720 | * This covers the case that we create a snapshot off of a received | |
4721 | * subvolume and then use that as the parent and try to receive on a | |
4722 | * different host. | |
4723 | */ | |
4724 | if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid)) | |
4725 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4726 | clone_root->root->root_item.received_uuid); | |
4727 | else | |
4728 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
4729 | clone_root->root->root_item.uuid); | |
31db9f7c | 4730 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 4731 | le64_to_cpu(clone_root->root->root_item.ctransid)); |
31db9f7c AB |
4732 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); |
4733 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, | |
4734 | clone_root->offset); | |
4735 | ||
4736 | ret = send_cmd(sctx); | |
4737 | ||
4738 | tlv_put_failure: | |
4739 | out: | |
924794c9 | 4740 | fs_path_free(p); |
31db9f7c AB |
4741 | return ret; |
4742 | } | |
4743 | ||
cb95e7bf MF |
4744 | /* |
4745 | * Send an update extent command to user space. | |
4746 | */ | |
4747 | static int send_update_extent(struct send_ctx *sctx, | |
4748 | u64 offset, u32 len) | |
4749 | { | |
4750 | int ret = 0; | |
4751 | struct fs_path *p; | |
4752 | ||
924794c9 | 4753 | p = fs_path_alloc(); |
cb95e7bf MF |
4754 | if (!p) |
4755 | return -ENOMEM; | |
4756 | ||
4757 | ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); | |
4758 | if (ret < 0) | |
4759 | goto out; | |
4760 | ||
4761 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4762 | if (ret < 0) | |
4763 | goto out; | |
4764 | ||
4765 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4766 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4767 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); | |
4768 | ||
4769 | ret = send_cmd(sctx); | |
4770 | ||
4771 | tlv_put_failure: | |
4772 | out: | |
924794c9 | 4773 | fs_path_free(p); |
cb95e7bf MF |
4774 | return ret; |
4775 | } | |
4776 | ||
16e7549f JB |
4777 | static int send_hole(struct send_ctx *sctx, u64 end) |
4778 | { | |
4779 | struct fs_path *p = NULL; | |
4780 | u64 offset = sctx->cur_inode_last_extent; | |
4781 | u64 len; | |
4782 | int ret = 0; | |
4783 | ||
4784 | p = fs_path_alloc(); | |
4785 | if (!p) | |
4786 | return -ENOMEM; | |
c715e155 FM |
4787 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); |
4788 | if (ret < 0) | |
4789 | goto tlv_put_failure; | |
16e7549f JB |
4790 | memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE); |
4791 | while (offset < end) { | |
4792 | len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE); | |
4793 | ||
4794 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
16e7549f JB |
4795 | if (ret < 0) |
4796 | break; | |
4797 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4798 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4799 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len); | |
4800 | ret = send_cmd(sctx); | |
4801 | if (ret < 0) | |
4802 | break; | |
4803 | offset += len; | |
4804 | } | |
4805 | tlv_put_failure: | |
4806 | fs_path_free(p); | |
4807 | return ret; | |
4808 | } | |
4809 | ||
d906d49f FM |
4810 | static int send_extent_data(struct send_ctx *sctx, |
4811 | const u64 offset, | |
4812 | const u64 len) | |
4813 | { | |
4814 | u64 sent = 0; | |
4815 | ||
4816 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) | |
4817 | return send_update_extent(sctx, offset, len); | |
4818 | ||
4819 | while (sent < len) { | |
4820 | u64 size = len - sent; | |
4821 | int ret; | |
4822 | ||
4823 | if (size > BTRFS_SEND_READ_SIZE) | |
4824 | size = BTRFS_SEND_READ_SIZE; | |
4825 | ret = send_write(sctx, offset + sent, size); | |
4826 | if (ret < 0) | |
4827 | return ret; | |
4828 | if (!ret) | |
4829 | break; | |
4830 | sent += ret; | |
4831 | } | |
4832 | return 0; | |
4833 | } | |
4834 | ||
4835 | static int clone_range(struct send_ctx *sctx, | |
4836 | struct clone_root *clone_root, | |
4837 | const u64 disk_byte, | |
4838 | u64 data_offset, | |
4839 | u64 offset, | |
4840 | u64 len) | |
4841 | { | |
4842 | struct btrfs_path *path; | |
4843 | struct btrfs_key key; | |
4844 | int ret; | |
4845 | ||
4846 | path = alloc_path_for_send(); | |
4847 | if (!path) | |
4848 | return -ENOMEM; | |
4849 | ||
4850 | /* | |
4851 | * We can't send a clone operation for the entire range if we find | |
4852 | * extent items in the respective range in the source file that | |
4853 | * refer to different extents or if we find holes. | |
4854 | * So check for that and do a mix of clone and regular write/copy | |
4855 | * operations if needed. | |
4856 | * | |
4857 | * Example: | |
4858 | * | |
4859 | * mkfs.btrfs -f /dev/sda | |
4860 | * mount /dev/sda /mnt | |
4861 | * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo | |
4862 | * cp --reflink=always /mnt/foo /mnt/bar | |
4863 | * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo | |
4864 | * btrfs subvolume snapshot -r /mnt /mnt/snap | |
4865 | * | |
4866 | * If when we send the snapshot and we are processing file bar (which | |
4867 | * has a higher inode number than foo) we blindly send a clone operation | |
4868 | * for the [0, 100K[ range from foo to bar, the receiver ends up getting | |
4869 | * a file bar that matches the content of file foo - iow, doesn't match | |
4870 | * the content from bar in the original filesystem. | |
4871 | */ | |
4872 | key.objectid = clone_root->ino; | |
4873 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4874 | key.offset = clone_root->offset; | |
4875 | ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); | |
4876 | if (ret < 0) | |
4877 | goto out; | |
4878 | if (ret > 0 && path->slots[0] > 0) { | |
4879 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
4880 | if (key.objectid == clone_root->ino && | |
4881 | key.type == BTRFS_EXTENT_DATA_KEY) | |
4882 | path->slots[0]--; | |
4883 | } | |
4884 | ||
4885 | while (true) { | |
4886 | struct extent_buffer *leaf = path->nodes[0]; | |
4887 | int slot = path->slots[0]; | |
4888 | struct btrfs_file_extent_item *ei; | |
4889 | u8 type; | |
4890 | u64 ext_len; | |
4891 | u64 clone_len; | |
4892 | ||
4893 | if (slot >= btrfs_header_nritems(leaf)) { | |
4894 | ret = btrfs_next_leaf(clone_root->root, path); | |
4895 | if (ret < 0) | |
4896 | goto out; | |
4897 | else if (ret > 0) | |
4898 | break; | |
4899 | continue; | |
4900 | } | |
4901 | ||
4902 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4903 | ||
4904 | /* | |
4905 | * We might have an implicit trailing hole (NO_HOLES feature | |
4906 | * enabled). We deal with it after leaving this loop. | |
4907 | */ | |
4908 | if (key.objectid != clone_root->ino || | |
4909 | key.type != BTRFS_EXTENT_DATA_KEY) | |
4910 | break; | |
4911 | ||
4912 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4913 | type = btrfs_file_extent_type(leaf, ei); | |
4914 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
4915 | ext_len = btrfs_file_extent_inline_len(leaf, slot, ei); | |
09cbfeaf | 4916 | ext_len = PAGE_ALIGN(ext_len); |
d906d49f FM |
4917 | } else { |
4918 | ext_len = btrfs_file_extent_num_bytes(leaf, ei); | |
4919 | } | |
4920 | ||
4921 | if (key.offset + ext_len <= clone_root->offset) | |
4922 | goto next; | |
4923 | ||
4924 | if (key.offset > clone_root->offset) { | |
4925 | /* Implicit hole, NO_HOLES feature enabled. */ | |
4926 | u64 hole_len = key.offset - clone_root->offset; | |
4927 | ||
4928 | if (hole_len > len) | |
4929 | hole_len = len; | |
4930 | ret = send_extent_data(sctx, offset, hole_len); | |
4931 | if (ret < 0) | |
4932 | goto out; | |
4933 | ||
4934 | len -= hole_len; | |
4935 | if (len == 0) | |
4936 | break; | |
4937 | offset += hole_len; | |
4938 | clone_root->offset += hole_len; | |
4939 | data_offset += hole_len; | |
4940 | } | |
4941 | ||
4942 | if (key.offset >= clone_root->offset + len) | |
4943 | break; | |
4944 | ||
4945 | clone_len = min_t(u64, ext_len, len); | |
4946 | ||
4947 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && | |
4948 | btrfs_file_extent_offset(leaf, ei) == data_offset) | |
4949 | ret = send_clone(sctx, offset, clone_len, clone_root); | |
4950 | else | |
4951 | ret = send_extent_data(sctx, offset, clone_len); | |
4952 | ||
4953 | if (ret < 0) | |
4954 | goto out; | |
4955 | ||
4956 | len -= clone_len; | |
4957 | if (len == 0) | |
4958 | break; | |
4959 | offset += clone_len; | |
4960 | clone_root->offset += clone_len; | |
4961 | data_offset += clone_len; | |
4962 | next: | |
4963 | path->slots[0]++; | |
4964 | } | |
4965 | ||
4966 | if (len > 0) | |
4967 | ret = send_extent_data(sctx, offset, len); | |
4968 | else | |
4969 | ret = 0; | |
4970 | out: | |
4971 | btrfs_free_path(path); | |
4972 | return ret; | |
4973 | } | |
4974 | ||
31db9f7c AB |
4975 | static int send_write_or_clone(struct send_ctx *sctx, |
4976 | struct btrfs_path *path, | |
4977 | struct btrfs_key *key, | |
4978 | struct clone_root *clone_root) | |
4979 | { | |
4980 | int ret = 0; | |
4981 | struct btrfs_file_extent_item *ei; | |
4982 | u64 offset = key->offset; | |
31db9f7c | 4983 | u64 len; |
31db9f7c | 4984 | u8 type; |
28e5dd8f | 4985 | u64 bs = sctx->send_root->fs_info->sb->s_blocksize; |
31db9f7c AB |
4986 | |
4987 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4988 | struct btrfs_file_extent_item); | |
4989 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
74dd17fb | 4990 | if (type == BTRFS_FILE_EXTENT_INLINE) { |
514ac8ad CM |
4991 | len = btrfs_file_extent_inline_len(path->nodes[0], |
4992 | path->slots[0], ei); | |
74dd17fb CM |
4993 | /* |
4994 | * it is possible the inline item won't cover the whole page, | |
4995 | * but there may be items after this page. Make | |
4996 | * sure to send the whole thing | |
4997 | */ | |
09cbfeaf | 4998 | len = PAGE_ALIGN(len); |
74dd17fb | 4999 | } else { |
31db9f7c | 5000 | len = btrfs_file_extent_num_bytes(path->nodes[0], ei); |
74dd17fb | 5001 | } |
31db9f7c AB |
5002 | |
5003 | if (offset + len > sctx->cur_inode_size) | |
5004 | len = sctx->cur_inode_size - offset; | |
5005 | if (len == 0) { | |
5006 | ret = 0; | |
5007 | goto out; | |
5008 | } | |
5009 | ||
28e5dd8f | 5010 | if (clone_root && IS_ALIGNED(offset + len, bs)) { |
d906d49f FM |
5011 | u64 disk_byte; |
5012 | u64 data_offset; | |
5013 | ||
5014 | disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); | |
5015 | data_offset = btrfs_file_extent_offset(path->nodes[0], ei); | |
5016 | ret = clone_range(sctx, clone_root, disk_byte, data_offset, | |
5017 | offset, len); | |
cb95e7bf | 5018 | } else { |
d906d49f | 5019 | ret = send_extent_data(sctx, offset, len); |
31db9f7c | 5020 | } |
31db9f7c AB |
5021 | out: |
5022 | return ret; | |
5023 | } | |
5024 | ||
5025 | static int is_extent_unchanged(struct send_ctx *sctx, | |
5026 | struct btrfs_path *left_path, | |
5027 | struct btrfs_key *ekey) | |
5028 | { | |
5029 | int ret = 0; | |
5030 | struct btrfs_key key; | |
5031 | struct btrfs_path *path = NULL; | |
5032 | struct extent_buffer *eb; | |
5033 | int slot; | |
5034 | struct btrfs_key found_key; | |
5035 | struct btrfs_file_extent_item *ei; | |
5036 | u64 left_disknr; | |
5037 | u64 right_disknr; | |
5038 | u64 left_offset; | |
5039 | u64 right_offset; | |
5040 | u64 left_offset_fixed; | |
5041 | u64 left_len; | |
5042 | u64 right_len; | |
74dd17fb CM |
5043 | u64 left_gen; |
5044 | u64 right_gen; | |
31db9f7c AB |
5045 | u8 left_type; |
5046 | u8 right_type; | |
5047 | ||
5048 | path = alloc_path_for_send(); | |
5049 | if (!path) | |
5050 | return -ENOMEM; | |
5051 | ||
5052 | eb = left_path->nodes[0]; | |
5053 | slot = left_path->slots[0]; | |
31db9f7c AB |
5054 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
5055 | left_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
5056 | |
5057 | if (left_type != BTRFS_FILE_EXTENT_REG) { | |
5058 | ret = 0; | |
5059 | goto out; | |
5060 | } | |
74dd17fb CM |
5061 | left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5062 | left_len = btrfs_file_extent_num_bytes(eb, ei); | |
5063 | left_offset = btrfs_file_extent_offset(eb, ei); | |
5064 | left_gen = btrfs_file_extent_generation(eb, ei); | |
31db9f7c AB |
5065 | |
5066 | /* | |
5067 | * Following comments will refer to these graphics. L is the left | |
5068 | * extents which we are checking at the moment. 1-8 are the right | |
5069 | * extents that we iterate. | |
5070 | * | |
5071 | * |-----L-----| | |
5072 | * |-1-|-2a-|-3-|-4-|-5-|-6-| | |
5073 | * | |
5074 | * |-----L-----| | |
5075 | * |--1--|-2b-|...(same as above) | |
5076 | * | |
5077 | * Alternative situation. Happens on files where extents got split. | |
5078 | * |-----L-----| | |
5079 | * |-----------7-----------|-6-| | |
5080 | * | |
5081 | * Alternative situation. Happens on files which got larger. | |
5082 | * |-----L-----| | |
5083 | * |-8-| | |
5084 | * Nothing follows after 8. | |
5085 | */ | |
5086 | ||
5087 | key.objectid = ekey->objectid; | |
5088 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5089 | key.offset = ekey->offset; | |
5090 | ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); | |
5091 | if (ret < 0) | |
5092 | goto out; | |
5093 | if (ret) { | |
5094 | ret = 0; | |
5095 | goto out; | |
5096 | } | |
5097 | ||
5098 | /* | |
5099 | * Handle special case where the right side has no extents at all. | |
5100 | */ | |
5101 | eb = path->nodes[0]; | |
5102 | slot = path->slots[0]; | |
5103 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5104 | if (found_key.objectid != key.objectid || | |
5105 | found_key.type != key.type) { | |
57cfd462 JB |
5106 | /* If we're a hole then just pretend nothing changed */ |
5107 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5108 | goto out; |
5109 | } | |
5110 | ||
5111 | /* | |
5112 | * We're now on 2a, 2b or 7. | |
5113 | */ | |
5114 | key = found_key; | |
5115 | while (key.offset < ekey->offset + left_len) { | |
5116 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
5117 | right_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
5118 | if (right_type != BTRFS_FILE_EXTENT_REG) { |
5119 | ret = 0; | |
5120 | goto out; | |
5121 | } | |
5122 | ||
007d31f7 JB |
5123 | right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
5124 | right_len = btrfs_file_extent_num_bytes(eb, ei); | |
5125 | right_offset = btrfs_file_extent_offset(eb, ei); | |
5126 | right_gen = btrfs_file_extent_generation(eb, ei); | |
5127 | ||
31db9f7c AB |
5128 | /* |
5129 | * Are we at extent 8? If yes, we know the extent is changed. | |
5130 | * This may only happen on the first iteration. | |
5131 | */ | |
d8347fa4 | 5132 | if (found_key.offset + right_len <= ekey->offset) { |
57cfd462 JB |
5133 | /* If we're a hole just pretend nothing changed */ |
5134 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
5135 | goto out; |
5136 | } | |
5137 | ||
5138 | left_offset_fixed = left_offset; | |
5139 | if (key.offset < ekey->offset) { | |
5140 | /* Fix the right offset for 2a and 7. */ | |
5141 | right_offset += ekey->offset - key.offset; | |
5142 | } else { | |
5143 | /* Fix the left offset for all behind 2a and 2b */ | |
5144 | left_offset_fixed += key.offset - ekey->offset; | |
5145 | } | |
5146 | ||
5147 | /* | |
5148 | * Check if we have the same extent. | |
5149 | */ | |
3954096d | 5150 | if (left_disknr != right_disknr || |
74dd17fb CM |
5151 | left_offset_fixed != right_offset || |
5152 | left_gen != right_gen) { | |
31db9f7c AB |
5153 | ret = 0; |
5154 | goto out; | |
5155 | } | |
5156 | ||
5157 | /* | |
5158 | * Go to the next extent. | |
5159 | */ | |
5160 | ret = btrfs_next_item(sctx->parent_root, path); | |
5161 | if (ret < 0) | |
5162 | goto out; | |
5163 | if (!ret) { | |
5164 | eb = path->nodes[0]; | |
5165 | slot = path->slots[0]; | |
5166 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5167 | } | |
5168 | if (ret || found_key.objectid != key.objectid || | |
5169 | found_key.type != key.type) { | |
5170 | key.offset += right_len; | |
5171 | break; | |
adaa4b8e JS |
5172 | } |
5173 | if (found_key.offset != key.offset + right_len) { | |
5174 | ret = 0; | |
5175 | goto out; | |
31db9f7c AB |
5176 | } |
5177 | key = found_key; | |
5178 | } | |
5179 | ||
5180 | /* | |
5181 | * We're now behind the left extent (treat as unchanged) or at the end | |
5182 | * of the right side (treat as changed). | |
5183 | */ | |
5184 | if (key.offset >= ekey->offset + left_len) | |
5185 | ret = 1; | |
5186 | else | |
5187 | ret = 0; | |
5188 | ||
5189 | ||
5190 | out: | |
5191 | btrfs_free_path(path); | |
5192 | return ret; | |
5193 | } | |
5194 | ||
16e7549f JB |
5195 | static int get_last_extent(struct send_ctx *sctx, u64 offset) |
5196 | { | |
5197 | struct btrfs_path *path; | |
5198 | struct btrfs_root *root = sctx->send_root; | |
5199 | struct btrfs_file_extent_item *fi; | |
5200 | struct btrfs_key key; | |
5201 | u64 extent_end; | |
5202 | u8 type; | |
5203 | int ret; | |
5204 | ||
5205 | path = alloc_path_for_send(); | |
5206 | if (!path) | |
5207 | return -ENOMEM; | |
5208 | ||
5209 | sctx->cur_inode_last_extent = 0; | |
5210 | ||
5211 | key.objectid = sctx->cur_ino; | |
5212 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5213 | key.offset = offset; | |
5214 | ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); | |
5215 | if (ret < 0) | |
5216 | goto out; | |
5217 | ret = 0; | |
5218 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
5219 | if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
5220 | goto out; | |
5221 | ||
5222 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5223 | struct btrfs_file_extent_item); | |
5224 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5225 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
5226 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
5227 | path->slots[0], fi); | |
16e7549f JB |
5228 | extent_end = ALIGN(key.offset + size, |
5229 | sctx->send_root->sectorsize); | |
5230 | } else { | |
5231 | extent_end = key.offset + | |
5232 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5233 | } | |
5234 | sctx->cur_inode_last_extent = extent_end; | |
5235 | out: | |
5236 | btrfs_free_path(path); | |
5237 | return ret; | |
5238 | } | |
5239 | ||
5240 | static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, | |
5241 | struct btrfs_key *key) | |
5242 | { | |
5243 | struct btrfs_file_extent_item *fi; | |
5244 | u64 extent_end; | |
5245 | u8 type; | |
5246 | int ret = 0; | |
5247 | ||
5248 | if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) | |
5249 | return 0; | |
5250 | ||
5251 | if (sctx->cur_inode_last_extent == (u64)-1) { | |
5252 | ret = get_last_extent(sctx, key->offset - 1); | |
5253 | if (ret) | |
5254 | return ret; | |
5255 | } | |
5256 | ||
5257 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5258 | struct btrfs_file_extent_item); | |
5259 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
5260 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
5261 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
5262 | path->slots[0], fi); | |
16e7549f JB |
5263 | extent_end = ALIGN(key->offset + size, |
5264 | sctx->send_root->sectorsize); | |
5265 | } else { | |
5266 | extent_end = key->offset + | |
5267 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
5268 | } | |
bf54f412 FDBM |
5269 | |
5270 | if (path->slots[0] == 0 && | |
5271 | sctx->cur_inode_last_extent < key->offset) { | |
5272 | /* | |
5273 | * We might have skipped entire leafs that contained only | |
5274 | * file extent items for our current inode. These leafs have | |
5275 | * a generation number smaller (older) than the one in the | |
5276 | * current leaf and the leaf our last extent came from, and | |
5277 | * are located between these 2 leafs. | |
5278 | */ | |
5279 | ret = get_last_extent(sctx, key->offset - 1); | |
5280 | if (ret) | |
5281 | return ret; | |
5282 | } | |
5283 | ||
16e7549f JB |
5284 | if (sctx->cur_inode_last_extent < key->offset) |
5285 | ret = send_hole(sctx, key->offset); | |
5286 | sctx->cur_inode_last_extent = extent_end; | |
5287 | return ret; | |
5288 | } | |
5289 | ||
31db9f7c AB |
5290 | static int process_extent(struct send_ctx *sctx, |
5291 | struct btrfs_path *path, | |
5292 | struct btrfs_key *key) | |
5293 | { | |
31db9f7c | 5294 | struct clone_root *found_clone = NULL; |
57cfd462 | 5295 | int ret = 0; |
31db9f7c AB |
5296 | |
5297 | if (S_ISLNK(sctx->cur_inode_mode)) | |
5298 | return 0; | |
5299 | ||
5300 | if (sctx->parent_root && !sctx->cur_inode_new) { | |
5301 | ret = is_extent_unchanged(sctx, path, key); | |
5302 | if (ret < 0) | |
5303 | goto out; | |
5304 | if (ret) { | |
5305 | ret = 0; | |
16e7549f | 5306 | goto out_hole; |
31db9f7c | 5307 | } |
57cfd462 JB |
5308 | } else { |
5309 | struct btrfs_file_extent_item *ei; | |
5310 | u8 type; | |
5311 | ||
5312 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
5313 | struct btrfs_file_extent_item); | |
5314 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
5315 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
5316 | type == BTRFS_FILE_EXTENT_REG) { | |
5317 | /* | |
5318 | * The send spec does not have a prealloc command yet, | |
5319 | * so just leave a hole for prealloc'ed extents until | |
5320 | * we have enough commands queued up to justify rev'ing | |
5321 | * the send spec. | |
5322 | */ | |
5323 | if (type == BTRFS_FILE_EXTENT_PREALLOC) { | |
5324 | ret = 0; | |
5325 | goto out; | |
5326 | } | |
5327 | ||
5328 | /* Have a hole, just skip it. */ | |
5329 | if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { | |
5330 | ret = 0; | |
5331 | goto out; | |
5332 | } | |
5333 | } | |
31db9f7c AB |
5334 | } |
5335 | ||
5336 | ret = find_extent_clone(sctx, path, key->objectid, key->offset, | |
5337 | sctx->cur_inode_size, &found_clone); | |
5338 | if (ret != -ENOENT && ret < 0) | |
5339 | goto out; | |
5340 | ||
5341 | ret = send_write_or_clone(sctx, path, key, found_clone); | |
16e7549f JB |
5342 | if (ret) |
5343 | goto out; | |
5344 | out_hole: | |
5345 | ret = maybe_send_hole(sctx, path, key); | |
31db9f7c AB |
5346 | out: |
5347 | return ret; | |
5348 | } | |
5349 | ||
5350 | static int process_all_extents(struct send_ctx *sctx) | |
5351 | { | |
5352 | int ret; | |
5353 | struct btrfs_root *root; | |
5354 | struct btrfs_path *path; | |
5355 | struct btrfs_key key; | |
5356 | struct btrfs_key found_key; | |
5357 | struct extent_buffer *eb; | |
5358 | int slot; | |
5359 | ||
5360 | root = sctx->send_root; | |
5361 | path = alloc_path_for_send(); | |
5362 | if (!path) | |
5363 | return -ENOMEM; | |
5364 | ||
5365 | key.objectid = sctx->cmp_key->objectid; | |
5366 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5367 | key.offset = 0; | |
7fdd29d0 FDBM |
5368 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5369 | if (ret < 0) | |
5370 | goto out; | |
31db9f7c | 5371 | |
7fdd29d0 | 5372 | while (1) { |
31db9f7c AB |
5373 | eb = path->nodes[0]; |
5374 | slot = path->slots[0]; | |
7fdd29d0 FDBM |
5375 | |
5376 | if (slot >= btrfs_header_nritems(eb)) { | |
5377 | ret = btrfs_next_leaf(root, path); | |
5378 | if (ret < 0) { | |
5379 | goto out; | |
5380 | } else if (ret > 0) { | |
5381 | ret = 0; | |
5382 | break; | |
5383 | } | |
5384 | continue; | |
5385 | } | |
5386 | ||
31db9f7c AB |
5387 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
5388 | ||
5389 | if (found_key.objectid != key.objectid || | |
5390 | found_key.type != key.type) { | |
5391 | ret = 0; | |
5392 | goto out; | |
5393 | } | |
5394 | ||
5395 | ret = process_extent(sctx, path, &found_key); | |
5396 | if (ret < 0) | |
5397 | goto out; | |
5398 | ||
7fdd29d0 | 5399 | path->slots[0]++; |
31db9f7c AB |
5400 | } |
5401 | ||
5402 | out: | |
5403 | btrfs_free_path(path); | |
5404 | return ret; | |
5405 | } | |
5406 | ||
9f03740a FDBM |
5407 | static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, |
5408 | int *pending_move, | |
5409 | int *refs_processed) | |
31db9f7c AB |
5410 | { |
5411 | int ret = 0; | |
5412 | ||
5413 | if (sctx->cur_ino == 0) | |
5414 | goto out; | |
5415 | if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && | |
96b5bd77 | 5416 | sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) |
31db9f7c AB |
5417 | goto out; |
5418 | if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) | |
5419 | goto out; | |
5420 | ||
9f03740a | 5421 | ret = process_recorded_refs(sctx, pending_move); |
e479d9bb AB |
5422 | if (ret < 0) |
5423 | goto out; | |
5424 | ||
9f03740a | 5425 | *refs_processed = 1; |
31db9f7c AB |
5426 | out: |
5427 | return ret; | |
5428 | } | |
5429 | ||
5430 | static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) | |
5431 | { | |
5432 | int ret = 0; | |
5433 | u64 left_mode; | |
5434 | u64 left_uid; | |
5435 | u64 left_gid; | |
5436 | u64 right_mode; | |
5437 | u64 right_uid; | |
5438 | u64 right_gid; | |
5439 | int need_chmod = 0; | |
5440 | int need_chown = 0; | |
9f03740a FDBM |
5441 | int pending_move = 0; |
5442 | int refs_processed = 0; | |
31db9f7c | 5443 | |
9f03740a FDBM |
5444 | ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, |
5445 | &refs_processed); | |
31db9f7c AB |
5446 | if (ret < 0) |
5447 | goto out; | |
5448 | ||
9f03740a FDBM |
5449 | /* |
5450 | * We have processed the refs and thus need to advance send_progress. | |
5451 | * Now, calls to get_cur_xxx will take the updated refs of the current | |
5452 | * inode into account. | |
5453 | * | |
5454 | * On the other hand, if our current inode is a directory and couldn't | |
5455 | * be moved/renamed because its parent was renamed/moved too and it has | |
5456 | * a higher inode number, we can only move/rename our current inode | |
5457 | * after we moved/renamed its parent. Therefore in this case operate on | |
5458 | * the old path (pre move/rename) of our current inode, and the | |
5459 | * move/rename will be performed later. | |
5460 | */ | |
5461 | if (refs_processed && !pending_move) | |
5462 | sctx->send_progress = sctx->cur_ino + 1; | |
5463 | ||
31db9f7c AB |
5464 | if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) |
5465 | goto out; | |
5466 | if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) | |
5467 | goto out; | |
5468 | ||
5469 | ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, | |
85a7b33b | 5470 | &left_mode, &left_uid, &left_gid, NULL); |
31db9f7c AB |
5471 | if (ret < 0) |
5472 | goto out; | |
5473 | ||
e2d044fe AL |
5474 | if (!sctx->parent_root || sctx->cur_inode_new) { |
5475 | need_chown = 1; | |
5476 | if (!S_ISLNK(sctx->cur_inode_mode)) | |
31db9f7c | 5477 | need_chmod = 1; |
e2d044fe AL |
5478 | } else { |
5479 | ret = get_inode_info(sctx->parent_root, sctx->cur_ino, | |
5480 | NULL, NULL, &right_mode, &right_uid, | |
5481 | &right_gid, NULL); | |
5482 | if (ret < 0) | |
5483 | goto out; | |
31db9f7c | 5484 | |
e2d044fe AL |
5485 | if (left_uid != right_uid || left_gid != right_gid) |
5486 | need_chown = 1; | |
5487 | if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) | |
5488 | need_chmod = 1; | |
31db9f7c AB |
5489 | } |
5490 | ||
5491 | if (S_ISREG(sctx->cur_inode_mode)) { | |
16e7549f | 5492 | if (need_send_hole(sctx)) { |
766b5e5a FM |
5493 | if (sctx->cur_inode_last_extent == (u64)-1 || |
5494 | sctx->cur_inode_last_extent < | |
5495 | sctx->cur_inode_size) { | |
16e7549f JB |
5496 | ret = get_last_extent(sctx, (u64)-1); |
5497 | if (ret) | |
5498 | goto out; | |
5499 | } | |
5500 | if (sctx->cur_inode_last_extent < | |
5501 | sctx->cur_inode_size) { | |
5502 | ret = send_hole(sctx, sctx->cur_inode_size); | |
5503 | if (ret) | |
5504 | goto out; | |
5505 | } | |
5506 | } | |
31db9f7c AB |
5507 | ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
5508 | sctx->cur_inode_size); | |
5509 | if (ret < 0) | |
5510 | goto out; | |
5511 | } | |
5512 | ||
5513 | if (need_chown) { | |
5514 | ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5515 | left_uid, left_gid); | |
5516 | if (ret < 0) | |
5517 | goto out; | |
5518 | } | |
5519 | if (need_chmod) { | |
5520 | ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5521 | left_mode); | |
5522 | if (ret < 0) | |
5523 | goto out; | |
5524 | } | |
5525 | ||
5526 | /* | |
9f03740a FDBM |
5527 | * If other directory inodes depended on our current directory |
5528 | * inode's move/rename, now do their move/rename operations. | |
31db9f7c | 5529 | */ |
9f03740a FDBM |
5530 | if (!is_waiting_for_move(sctx, sctx->cur_ino)) { |
5531 | ret = apply_children_dir_moves(sctx); | |
5532 | if (ret) | |
5533 | goto out; | |
fcbd2154 FM |
5534 | /* |
5535 | * Need to send that every time, no matter if it actually | |
5536 | * changed between the two trees as we have done changes to | |
5537 | * the inode before. If our inode is a directory and it's | |
5538 | * waiting to be moved/renamed, we will send its utimes when | |
5539 | * it's moved/renamed, therefore we don't need to do it here. | |
5540 | */ | |
5541 | sctx->send_progress = sctx->cur_ino + 1; | |
5542 | ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
5543 | if (ret < 0) | |
5544 | goto out; | |
9f03740a FDBM |
5545 | } |
5546 | ||
31db9f7c AB |
5547 | out: |
5548 | return ret; | |
5549 | } | |
5550 | ||
5551 | static int changed_inode(struct send_ctx *sctx, | |
5552 | enum btrfs_compare_tree_result result) | |
5553 | { | |
5554 | int ret = 0; | |
5555 | struct btrfs_key *key = sctx->cmp_key; | |
5556 | struct btrfs_inode_item *left_ii = NULL; | |
5557 | struct btrfs_inode_item *right_ii = NULL; | |
5558 | u64 left_gen = 0; | |
5559 | u64 right_gen = 0; | |
5560 | ||
31db9f7c AB |
5561 | sctx->cur_ino = key->objectid; |
5562 | sctx->cur_inode_new_gen = 0; | |
16e7549f | 5563 | sctx->cur_inode_last_extent = (u64)-1; |
e479d9bb AB |
5564 | |
5565 | /* | |
5566 | * Set send_progress to current inode. This will tell all get_cur_xxx | |
5567 | * functions that the current inode's refs are not updated yet. Later, | |
5568 | * when process_recorded_refs is finished, it is set to cur_ino + 1. | |
5569 | */ | |
31db9f7c AB |
5570 | sctx->send_progress = sctx->cur_ino; |
5571 | ||
5572 | if (result == BTRFS_COMPARE_TREE_NEW || | |
5573 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
5574 | left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], | |
5575 | sctx->left_path->slots[0], | |
5576 | struct btrfs_inode_item); | |
5577 | left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], | |
5578 | left_ii); | |
5579 | } else { | |
5580 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
5581 | sctx->right_path->slots[0], | |
5582 | struct btrfs_inode_item); | |
5583 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
5584 | right_ii); | |
5585 | } | |
5586 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
5587 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
5588 | sctx->right_path->slots[0], | |
5589 | struct btrfs_inode_item); | |
5590 | ||
5591 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
5592 | right_ii); | |
6d85ed05 AB |
5593 | |
5594 | /* | |
5595 | * The cur_ino = root dir case is special here. We can't treat | |
5596 | * the inode as deleted+reused because it would generate a | |
5597 | * stream that tries to delete/mkdir the root dir. | |
5598 | */ | |
5599 | if (left_gen != right_gen && | |
5600 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) | |
31db9f7c AB |
5601 | sctx->cur_inode_new_gen = 1; |
5602 | } | |
5603 | ||
5604 | if (result == BTRFS_COMPARE_TREE_NEW) { | |
5605 | sctx->cur_inode_gen = left_gen; | |
5606 | sctx->cur_inode_new = 1; | |
5607 | sctx->cur_inode_deleted = 0; | |
5608 | sctx->cur_inode_size = btrfs_inode_size( | |
5609 | sctx->left_path->nodes[0], left_ii); | |
5610 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5611 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
5612 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
5613 | sctx->left_path->nodes[0], left_ii); | |
31db9f7c | 5614 | if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) |
1f4692da | 5615 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
5616 | } else if (result == BTRFS_COMPARE_TREE_DELETED) { |
5617 | sctx->cur_inode_gen = right_gen; | |
5618 | sctx->cur_inode_new = 0; | |
5619 | sctx->cur_inode_deleted = 1; | |
5620 | sctx->cur_inode_size = btrfs_inode_size( | |
5621 | sctx->right_path->nodes[0], right_ii); | |
5622 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5623 | sctx->right_path->nodes[0], right_ii); | |
5624 | } else if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
766702ef AB |
5625 | /* |
5626 | * We need to do some special handling in case the inode was | |
5627 | * reported as changed with a changed generation number. This | |
5628 | * means that the original inode was deleted and new inode | |
5629 | * reused the same inum. So we have to treat the old inode as | |
5630 | * deleted and the new one as new. | |
5631 | */ | |
31db9f7c | 5632 | if (sctx->cur_inode_new_gen) { |
766702ef AB |
5633 | /* |
5634 | * First, process the inode as if it was deleted. | |
5635 | */ | |
31db9f7c AB |
5636 | sctx->cur_inode_gen = right_gen; |
5637 | sctx->cur_inode_new = 0; | |
5638 | sctx->cur_inode_deleted = 1; | |
5639 | sctx->cur_inode_size = btrfs_inode_size( | |
5640 | sctx->right_path->nodes[0], right_ii); | |
5641 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5642 | sctx->right_path->nodes[0], right_ii); | |
5643 | ret = process_all_refs(sctx, | |
5644 | BTRFS_COMPARE_TREE_DELETED); | |
5645 | if (ret < 0) | |
5646 | goto out; | |
5647 | ||
766702ef AB |
5648 | /* |
5649 | * Now process the inode as if it was new. | |
5650 | */ | |
31db9f7c AB |
5651 | sctx->cur_inode_gen = left_gen; |
5652 | sctx->cur_inode_new = 1; | |
5653 | sctx->cur_inode_deleted = 0; | |
5654 | sctx->cur_inode_size = btrfs_inode_size( | |
5655 | sctx->left_path->nodes[0], left_ii); | |
5656 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5657 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
5658 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
5659 | sctx->left_path->nodes[0], left_ii); | |
1f4692da | 5660 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
5661 | if (ret < 0) |
5662 | goto out; | |
5663 | ||
5664 | ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); | |
5665 | if (ret < 0) | |
5666 | goto out; | |
e479d9bb AB |
5667 | /* |
5668 | * Advance send_progress now as we did not get into | |
5669 | * process_recorded_refs_if_needed in the new_gen case. | |
5670 | */ | |
5671 | sctx->send_progress = sctx->cur_ino + 1; | |
766702ef AB |
5672 | |
5673 | /* | |
5674 | * Now process all extents and xattrs of the inode as if | |
5675 | * they were all new. | |
5676 | */ | |
31db9f7c AB |
5677 | ret = process_all_extents(sctx); |
5678 | if (ret < 0) | |
5679 | goto out; | |
5680 | ret = process_all_new_xattrs(sctx); | |
5681 | if (ret < 0) | |
5682 | goto out; | |
5683 | } else { | |
5684 | sctx->cur_inode_gen = left_gen; | |
5685 | sctx->cur_inode_new = 0; | |
5686 | sctx->cur_inode_new_gen = 0; | |
5687 | sctx->cur_inode_deleted = 0; | |
5688 | sctx->cur_inode_size = btrfs_inode_size( | |
5689 | sctx->left_path->nodes[0], left_ii); | |
5690 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5691 | sctx->left_path->nodes[0], left_ii); | |
5692 | } | |
5693 | } | |
5694 | ||
5695 | out: | |
5696 | return ret; | |
5697 | } | |
5698 | ||
766702ef AB |
5699 | /* |
5700 | * We have to process new refs before deleted refs, but compare_trees gives us | |
5701 | * the new and deleted refs mixed. To fix this, we record the new/deleted refs | |
5702 | * first and later process them in process_recorded_refs. | |
5703 | * For the cur_inode_new_gen case, we skip recording completely because | |
5704 | * changed_inode did already initiate processing of refs. The reason for this is | |
5705 | * that in this case, compare_tree actually compares the refs of 2 different | |
5706 | * inodes. To fix this, process_all_refs is used in changed_inode to handle all | |
5707 | * refs of the right tree as deleted and all refs of the left tree as new. | |
5708 | */ | |
31db9f7c AB |
5709 | static int changed_ref(struct send_ctx *sctx, |
5710 | enum btrfs_compare_tree_result result) | |
5711 | { | |
5712 | int ret = 0; | |
5713 | ||
5714 | BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid); | |
5715 | ||
5716 | if (!sctx->cur_inode_new_gen && | |
5717 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { | |
5718 | if (result == BTRFS_COMPARE_TREE_NEW) | |
5719 | ret = record_new_ref(sctx); | |
5720 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
5721 | ret = record_deleted_ref(sctx); | |
5722 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
5723 | ret = record_changed_ref(sctx); | |
5724 | } | |
5725 | ||
5726 | return ret; | |
5727 | } | |
5728 | ||
766702ef AB |
5729 | /* |
5730 | * Process new/deleted/changed xattrs. We skip processing in the | |
5731 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
5732 | * of xattrs. The reason is the same as in changed_ref | |
5733 | */ | |
31db9f7c AB |
5734 | static int changed_xattr(struct send_ctx *sctx, |
5735 | enum btrfs_compare_tree_result result) | |
5736 | { | |
5737 | int ret = 0; | |
5738 | ||
5739 | BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid); | |
5740 | ||
5741 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
5742 | if (result == BTRFS_COMPARE_TREE_NEW) | |
5743 | ret = process_new_xattr(sctx); | |
5744 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
5745 | ret = process_deleted_xattr(sctx); | |
5746 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
5747 | ret = process_changed_xattr(sctx); | |
5748 | } | |
5749 | ||
5750 | return ret; | |
5751 | } | |
5752 | ||
766702ef AB |
5753 | /* |
5754 | * Process new/deleted/changed extents. We skip processing in the | |
5755 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
5756 | * of extents. The reason is the same as in changed_ref | |
5757 | */ | |
31db9f7c AB |
5758 | static int changed_extent(struct send_ctx *sctx, |
5759 | enum btrfs_compare_tree_result result) | |
5760 | { | |
5761 | int ret = 0; | |
5762 | ||
5763 | BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid); | |
5764 | ||
5765 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
5766 | if (result != BTRFS_COMPARE_TREE_DELETED) | |
5767 | ret = process_extent(sctx, sctx->left_path, | |
5768 | sctx->cmp_key); | |
5769 | } | |
5770 | ||
5771 | return ret; | |
5772 | } | |
5773 | ||
ba5e8f2e JB |
5774 | static int dir_changed(struct send_ctx *sctx, u64 dir) |
5775 | { | |
5776 | u64 orig_gen, new_gen; | |
5777 | int ret; | |
5778 | ||
5779 | ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, | |
5780 | NULL, NULL); | |
5781 | if (ret) | |
5782 | return ret; | |
5783 | ||
5784 | ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, | |
5785 | NULL, NULL, NULL); | |
5786 | if (ret) | |
5787 | return ret; | |
5788 | ||
5789 | return (orig_gen != new_gen) ? 1 : 0; | |
5790 | } | |
5791 | ||
5792 | static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, | |
5793 | struct btrfs_key *key) | |
5794 | { | |
5795 | struct btrfs_inode_extref *extref; | |
5796 | struct extent_buffer *leaf; | |
5797 | u64 dirid = 0, last_dirid = 0; | |
5798 | unsigned long ptr; | |
5799 | u32 item_size; | |
5800 | u32 cur_offset = 0; | |
5801 | int ref_name_len; | |
5802 | int ret = 0; | |
5803 | ||
5804 | /* Easy case, just check this one dirid */ | |
5805 | if (key->type == BTRFS_INODE_REF_KEY) { | |
5806 | dirid = key->offset; | |
5807 | ||
5808 | ret = dir_changed(sctx, dirid); | |
5809 | goto out; | |
5810 | } | |
5811 | ||
5812 | leaf = path->nodes[0]; | |
5813 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5814 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
5815 | while (cur_offset < item_size) { | |
5816 | extref = (struct btrfs_inode_extref *)(ptr + | |
5817 | cur_offset); | |
5818 | dirid = btrfs_inode_extref_parent(leaf, extref); | |
5819 | ref_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
5820 | cur_offset += ref_name_len + sizeof(*extref); | |
5821 | if (dirid == last_dirid) | |
5822 | continue; | |
5823 | ret = dir_changed(sctx, dirid); | |
5824 | if (ret) | |
5825 | break; | |
5826 | last_dirid = dirid; | |
5827 | } | |
5828 | out: | |
5829 | return ret; | |
5830 | } | |
5831 | ||
766702ef AB |
5832 | /* |
5833 | * Updates compare related fields in sctx and simply forwards to the actual | |
5834 | * changed_xxx functions. | |
5835 | */ | |
31db9f7c AB |
5836 | static int changed_cb(struct btrfs_root *left_root, |
5837 | struct btrfs_root *right_root, | |
5838 | struct btrfs_path *left_path, | |
5839 | struct btrfs_path *right_path, | |
5840 | struct btrfs_key *key, | |
5841 | enum btrfs_compare_tree_result result, | |
5842 | void *ctx) | |
5843 | { | |
5844 | int ret = 0; | |
5845 | struct send_ctx *sctx = ctx; | |
5846 | ||
ba5e8f2e | 5847 | if (result == BTRFS_COMPARE_TREE_SAME) { |
16e7549f JB |
5848 | if (key->type == BTRFS_INODE_REF_KEY || |
5849 | key->type == BTRFS_INODE_EXTREF_KEY) { | |
5850 | ret = compare_refs(sctx, left_path, key); | |
5851 | if (!ret) | |
5852 | return 0; | |
5853 | if (ret < 0) | |
5854 | return ret; | |
5855 | } else if (key->type == BTRFS_EXTENT_DATA_KEY) { | |
5856 | return maybe_send_hole(sctx, left_path, key); | |
5857 | } else { | |
ba5e8f2e | 5858 | return 0; |
16e7549f | 5859 | } |
ba5e8f2e JB |
5860 | result = BTRFS_COMPARE_TREE_CHANGED; |
5861 | ret = 0; | |
5862 | } | |
5863 | ||
31db9f7c AB |
5864 | sctx->left_path = left_path; |
5865 | sctx->right_path = right_path; | |
5866 | sctx->cmp_key = key; | |
5867 | ||
5868 | ret = finish_inode_if_needed(sctx, 0); | |
5869 | if (ret < 0) | |
5870 | goto out; | |
5871 | ||
2981e225 AB |
5872 | /* Ignore non-FS objects */ |
5873 | if (key->objectid == BTRFS_FREE_INO_OBJECTID || | |
5874 | key->objectid == BTRFS_FREE_SPACE_OBJECTID) | |
5875 | goto out; | |
5876 | ||
31db9f7c AB |
5877 | if (key->type == BTRFS_INODE_ITEM_KEY) |
5878 | ret = changed_inode(sctx, result); | |
96b5bd77 JS |
5879 | else if (key->type == BTRFS_INODE_REF_KEY || |
5880 | key->type == BTRFS_INODE_EXTREF_KEY) | |
31db9f7c AB |
5881 | ret = changed_ref(sctx, result); |
5882 | else if (key->type == BTRFS_XATTR_ITEM_KEY) | |
5883 | ret = changed_xattr(sctx, result); | |
5884 | else if (key->type == BTRFS_EXTENT_DATA_KEY) | |
5885 | ret = changed_extent(sctx, result); | |
5886 | ||
5887 | out: | |
5888 | return ret; | |
5889 | } | |
5890 | ||
5891 | static int full_send_tree(struct send_ctx *sctx) | |
5892 | { | |
5893 | int ret; | |
31db9f7c AB |
5894 | struct btrfs_root *send_root = sctx->send_root; |
5895 | struct btrfs_key key; | |
5896 | struct btrfs_key found_key; | |
5897 | struct btrfs_path *path; | |
5898 | struct extent_buffer *eb; | |
5899 | int slot; | |
31db9f7c AB |
5900 | |
5901 | path = alloc_path_for_send(); | |
5902 | if (!path) | |
5903 | return -ENOMEM; | |
5904 | ||
31db9f7c AB |
5905 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; |
5906 | key.type = BTRFS_INODE_ITEM_KEY; | |
5907 | key.offset = 0; | |
5908 | ||
31db9f7c AB |
5909 | ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); |
5910 | if (ret < 0) | |
5911 | goto out; | |
5912 | if (ret) | |
5913 | goto out_finish; | |
5914 | ||
5915 | while (1) { | |
31db9f7c AB |
5916 | eb = path->nodes[0]; |
5917 | slot = path->slots[0]; | |
5918 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5919 | ||
5920 | ret = changed_cb(send_root, NULL, path, NULL, | |
5921 | &found_key, BTRFS_COMPARE_TREE_NEW, sctx); | |
5922 | if (ret < 0) | |
5923 | goto out; | |
5924 | ||
5925 | key.objectid = found_key.objectid; | |
5926 | key.type = found_key.type; | |
5927 | key.offset = found_key.offset + 1; | |
5928 | ||
5929 | ret = btrfs_next_item(send_root, path); | |
5930 | if (ret < 0) | |
5931 | goto out; | |
5932 | if (ret) { | |
5933 | ret = 0; | |
5934 | break; | |
5935 | } | |
5936 | } | |
5937 | ||
5938 | out_finish: | |
5939 | ret = finish_inode_if_needed(sctx, 1); | |
5940 | ||
5941 | out: | |
5942 | btrfs_free_path(path); | |
31db9f7c AB |
5943 | return ret; |
5944 | } | |
5945 | ||
5946 | static int send_subvol(struct send_ctx *sctx) | |
5947 | { | |
5948 | int ret; | |
5949 | ||
c2c71324 SB |
5950 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { |
5951 | ret = send_header(sctx); | |
5952 | if (ret < 0) | |
5953 | goto out; | |
5954 | } | |
31db9f7c AB |
5955 | |
5956 | ret = send_subvol_begin(sctx); | |
5957 | if (ret < 0) | |
5958 | goto out; | |
5959 | ||
5960 | if (sctx->parent_root) { | |
5961 | ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, | |
5962 | changed_cb, sctx); | |
5963 | if (ret < 0) | |
5964 | goto out; | |
5965 | ret = finish_inode_if_needed(sctx, 1); | |
5966 | if (ret < 0) | |
5967 | goto out; | |
5968 | } else { | |
5969 | ret = full_send_tree(sctx); | |
5970 | if (ret < 0) | |
5971 | goto out; | |
5972 | } | |
5973 | ||
5974 | out: | |
31db9f7c AB |
5975 | free_recorded_refs(sctx); |
5976 | return ret; | |
5977 | } | |
5978 | ||
e5fa8f86 FM |
5979 | /* |
5980 | * If orphan cleanup did remove any orphans from a root, it means the tree | |
5981 | * was modified and therefore the commit root is not the same as the current | |
5982 | * root anymore. This is a problem, because send uses the commit root and | |
5983 | * therefore can see inode items that don't exist in the current root anymore, | |
5984 | * and for example make calls to btrfs_iget, which will do tree lookups based | |
5985 | * on the current root and not on the commit root. Those lookups will fail, | |
5986 | * returning a -ESTALE error, and making send fail with that error. So make | |
5987 | * sure a send does not see any orphans we have just removed, and that it will | |
5988 | * see the same inodes regardless of whether a transaction commit happened | |
5989 | * before it started (meaning that the commit root will be the same as the | |
5990 | * current root) or not. | |
5991 | */ | |
5992 | static int ensure_commit_roots_uptodate(struct send_ctx *sctx) | |
5993 | { | |
5994 | int i; | |
5995 | struct btrfs_trans_handle *trans = NULL; | |
5996 | ||
5997 | again: | |
5998 | if (sctx->parent_root && | |
5999 | sctx->parent_root->node != sctx->parent_root->commit_root) | |
6000 | goto commit_trans; | |
6001 | ||
6002 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6003 | if (sctx->clone_roots[i].root->node != | |
6004 | sctx->clone_roots[i].root->commit_root) | |
6005 | goto commit_trans; | |
6006 | ||
6007 | if (trans) | |
6008 | return btrfs_end_transaction(trans, sctx->send_root); | |
6009 | ||
6010 | return 0; | |
6011 | ||
6012 | commit_trans: | |
6013 | /* Use any root, all fs roots will get their commit roots updated. */ | |
6014 | if (!trans) { | |
6015 | trans = btrfs_join_transaction(sctx->send_root); | |
6016 | if (IS_ERR(trans)) | |
6017 | return PTR_ERR(trans); | |
6018 | goto again; | |
6019 | } | |
6020 | ||
6021 | return btrfs_commit_transaction(trans, sctx->send_root); | |
6022 | } | |
6023 | ||
66ef7d65 DS |
6024 | static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) |
6025 | { | |
6026 | spin_lock(&root->root_item_lock); | |
6027 | root->send_in_progress--; | |
6028 | /* | |
6029 | * Not much left to do, we don't know why it's unbalanced and | |
6030 | * can't blindly reset it to 0. | |
6031 | */ | |
6032 | if (root->send_in_progress < 0) | |
6033 | btrfs_err(root->fs_info, | |
351fd353 | 6034 | "send_in_progres unbalanced %d root %llu", |
66ef7d65 DS |
6035 | root->send_in_progress, root->root_key.objectid); |
6036 | spin_unlock(&root->root_item_lock); | |
6037 | } | |
6038 | ||
31db9f7c AB |
6039 | long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_) |
6040 | { | |
6041 | int ret = 0; | |
6042 | struct btrfs_root *send_root; | |
6043 | struct btrfs_root *clone_root; | |
6044 | struct btrfs_fs_info *fs_info; | |
6045 | struct btrfs_ioctl_send_args *arg = NULL; | |
6046 | struct btrfs_key key; | |
31db9f7c AB |
6047 | struct send_ctx *sctx = NULL; |
6048 | u32 i; | |
6049 | u64 *clone_sources_tmp = NULL; | |
2c686537 | 6050 | int clone_sources_to_rollback = 0; |
e55d1153 | 6051 | unsigned alloc_size; |
896c14f9 | 6052 | int sort_clone_roots = 0; |
18f687d5 | 6053 | int index; |
31db9f7c AB |
6054 | |
6055 | if (!capable(CAP_SYS_ADMIN)) | |
6056 | return -EPERM; | |
6057 | ||
496ad9aa | 6058 | send_root = BTRFS_I(file_inode(mnt_file))->root; |
31db9f7c AB |
6059 | fs_info = send_root->fs_info; |
6060 | ||
2c686537 DS |
6061 | /* |
6062 | * The subvolume must remain read-only during send, protect against | |
521e0546 | 6063 | * making it RW. This also protects against deletion. |
2c686537 DS |
6064 | */ |
6065 | spin_lock(&send_root->root_item_lock); | |
6066 | send_root->send_in_progress++; | |
6067 | spin_unlock(&send_root->root_item_lock); | |
6068 | ||
139f807a JB |
6069 | /* |
6070 | * This is done when we lookup the root, it should already be complete | |
6071 | * by the time we get here. | |
6072 | */ | |
6073 | WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE); | |
6074 | ||
2c686537 DS |
6075 | /* |
6076 | * Userspace tools do the checks and warn the user if it's | |
6077 | * not RO. | |
6078 | */ | |
6079 | if (!btrfs_root_readonly(send_root)) { | |
6080 | ret = -EPERM; | |
6081 | goto out; | |
6082 | } | |
6083 | ||
31db9f7c AB |
6084 | arg = memdup_user(arg_, sizeof(*arg)); |
6085 | if (IS_ERR(arg)) { | |
6086 | ret = PTR_ERR(arg); | |
6087 | arg = NULL; | |
6088 | goto out; | |
6089 | } | |
6090 | ||
f5ecec3c DC |
6091 | if (arg->clone_sources_count > |
6092 | ULLONG_MAX / sizeof(*arg->clone_sources)) { | |
6093 | ret = -EINVAL; | |
6094 | goto out; | |
6095 | } | |
6096 | ||
31db9f7c | 6097 | if (!access_ok(VERIFY_READ, arg->clone_sources, |
700ff4f0 DC |
6098 | sizeof(*arg->clone_sources) * |
6099 | arg->clone_sources_count)) { | |
31db9f7c AB |
6100 | ret = -EFAULT; |
6101 | goto out; | |
6102 | } | |
6103 | ||
c2c71324 | 6104 | if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { |
cb95e7bf MF |
6105 | ret = -EINVAL; |
6106 | goto out; | |
6107 | } | |
6108 | ||
e780b0d1 | 6109 | sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL); |
31db9f7c AB |
6110 | if (!sctx) { |
6111 | ret = -ENOMEM; | |
6112 | goto out; | |
6113 | } | |
6114 | ||
6115 | INIT_LIST_HEAD(&sctx->new_refs); | |
6116 | INIT_LIST_HEAD(&sctx->deleted_refs); | |
e780b0d1 | 6117 | INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL); |
31db9f7c AB |
6118 | INIT_LIST_HEAD(&sctx->name_cache_list); |
6119 | ||
cb95e7bf MF |
6120 | sctx->flags = arg->flags; |
6121 | ||
31db9f7c | 6122 | sctx->send_filp = fget(arg->send_fd); |
ecc7ada7 TI |
6123 | if (!sctx->send_filp) { |
6124 | ret = -EBADF; | |
31db9f7c AB |
6125 | goto out; |
6126 | } | |
6127 | ||
31db9f7c | 6128 | sctx->send_root = send_root; |
521e0546 DS |
6129 | /* |
6130 | * Unlikely but possible, if the subvolume is marked for deletion but | |
6131 | * is slow to remove the directory entry, send can still be started | |
6132 | */ | |
6133 | if (btrfs_root_dead(sctx->send_root)) { | |
6134 | ret = -EPERM; | |
6135 | goto out; | |
6136 | } | |
6137 | ||
31db9f7c AB |
6138 | sctx->clone_roots_cnt = arg->clone_sources_count; |
6139 | ||
6140 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE; | |
6ff48ce0 | 6141 | sctx->send_buf = kmalloc(sctx->send_max_size, GFP_KERNEL | __GFP_NOWARN); |
31db9f7c | 6142 | if (!sctx->send_buf) { |
6ff48ce0 DS |
6143 | sctx->send_buf = vmalloc(sctx->send_max_size); |
6144 | if (!sctx->send_buf) { | |
6145 | ret = -ENOMEM; | |
6146 | goto out; | |
6147 | } | |
31db9f7c AB |
6148 | } |
6149 | ||
eb5b75fe | 6150 | sctx->read_buf = kmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL | __GFP_NOWARN); |
31db9f7c | 6151 | if (!sctx->read_buf) { |
eb5b75fe DS |
6152 | sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE); |
6153 | if (!sctx->read_buf) { | |
6154 | ret = -ENOMEM; | |
6155 | goto out; | |
6156 | } | |
31db9f7c AB |
6157 | } |
6158 | ||
9f03740a FDBM |
6159 | sctx->pending_dir_moves = RB_ROOT; |
6160 | sctx->waiting_dir_moves = RB_ROOT; | |
9dc44214 | 6161 | sctx->orphan_dirs = RB_ROOT; |
9f03740a | 6162 | |
e55d1153 DS |
6163 | alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1); |
6164 | ||
c03d01f3 | 6165 | sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN); |
31db9f7c | 6166 | if (!sctx->clone_roots) { |
c03d01f3 DS |
6167 | sctx->clone_roots = vzalloc(alloc_size); |
6168 | if (!sctx->clone_roots) { | |
6169 | ret = -ENOMEM; | |
6170 | goto out; | |
6171 | } | |
31db9f7c AB |
6172 | } |
6173 | ||
e55d1153 DS |
6174 | alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources); |
6175 | ||
31db9f7c | 6176 | if (arg->clone_sources_count) { |
2f91306a | 6177 | clone_sources_tmp = kmalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN); |
31db9f7c | 6178 | if (!clone_sources_tmp) { |
2f91306a DS |
6179 | clone_sources_tmp = vmalloc(alloc_size); |
6180 | if (!clone_sources_tmp) { | |
6181 | ret = -ENOMEM; | |
6182 | goto out; | |
6183 | } | |
31db9f7c AB |
6184 | } |
6185 | ||
6186 | ret = copy_from_user(clone_sources_tmp, arg->clone_sources, | |
e55d1153 | 6187 | alloc_size); |
31db9f7c AB |
6188 | if (ret) { |
6189 | ret = -EFAULT; | |
6190 | goto out; | |
6191 | } | |
6192 | ||
6193 | for (i = 0; i < arg->clone_sources_count; i++) { | |
6194 | key.objectid = clone_sources_tmp[i]; | |
6195 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6196 | key.offset = (u64)-1; | |
18f687d5 WS |
6197 | |
6198 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6199 | ||
31db9f7c | 6200 | clone_root = btrfs_read_fs_root_no_name(fs_info, &key); |
31db9f7c | 6201 | if (IS_ERR(clone_root)) { |
18f687d5 | 6202 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
31db9f7c AB |
6203 | ret = PTR_ERR(clone_root); |
6204 | goto out; | |
6205 | } | |
2c686537 | 6206 | spin_lock(&clone_root->root_item_lock); |
5cc2b17e FM |
6207 | if (!btrfs_root_readonly(clone_root) || |
6208 | btrfs_root_dead(clone_root)) { | |
2c686537 | 6209 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 | 6210 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6211 | ret = -EPERM; |
6212 | goto out; | |
6213 | } | |
2f1f465a | 6214 | clone_root->send_in_progress++; |
2c686537 | 6215 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 WS |
6216 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
6217 | ||
31db9f7c | 6218 | sctx->clone_roots[i].root = clone_root; |
2f1f465a | 6219 | clone_sources_to_rollback = i + 1; |
31db9f7c | 6220 | } |
2f91306a | 6221 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6222 | clone_sources_tmp = NULL; |
6223 | } | |
6224 | ||
6225 | if (arg->parent_root) { | |
6226 | key.objectid = arg->parent_root; | |
6227 | key.type = BTRFS_ROOT_ITEM_KEY; | |
6228 | key.offset = (u64)-1; | |
18f687d5 WS |
6229 | |
6230 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
6231 | ||
31db9f7c | 6232 | sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key); |
b1b19596 | 6233 | if (IS_ERR(sctx->parent_root)) { |
18f687d5 | 6234 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
b1b19596 | 6235 | ret = PTR_ERR(sctx->parent_root); |
31db9f7c AB |
6236 | goto out; |
6237 | } | |
18f687d5 | 6238 | |
2c686537 DS |
6239 | spin_lock(&sctx->parent_root->root_item_lock); |
6240 | sctx->parent_root->send_in_progress++; | |
521e0546 DS |
6241 | if (!btrfs_root_readonly(sctx->parent_root) || |
6242 | btrfs_root_dead(sctx->parent_root)) { | |
2c686537 | 6243 | spin_unlock(&sctx->parent_root->root_item_lock); |
18f687d5 | 6244 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
6245 | ret = -EPERM; |
6246 | goto out; | |
6247 | } | |
6248 | spin_unlock(&sctx->parent_root->root_item_lock); | |
18f687d5 WS |
6249 | |
6250 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
31db9f7c AB |
6251 | } |
6252 | ||
6253 | /* | |
6254 | * Clones from send_root are allowed, but only if the clone source | |
6255 | * is behind the current send position. This is checked while searching | |
6256 | * for possible clone sources. | |
6257 | */ | |
6258 | sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root; | |
6259 | ||
6260 | /* We do a bsearch later */ | |
6261 | sort(sctx->clone_roots, sctx->clone_roots_cnt, | |
6262 | sizeof(*sctx->clone_roots), __clone_root_cmp_sort, | |
6263 | NULL); | |
896c14f9 | 6264 | sort_clone_roots = 1; |
31db9f7c | 6265 | |
e5fa8f86 FM |
6266 | ret = ensure_commit_roots_uptodate(sctx); |
6267 | if (ret) | |
6268 | goto out; | |
6269 | ||
2755a0de | 6270 | current->journal_info = BTRFS_SEND_TRANS_STUB; |
31db9f7c | 6271 | ret = send_subvol(sctx); |
a26e8c9f | 6272 | current->journal_info = NULL; |
31db9f7c AB |
6273 | if (ret < 0) |
6274 | goto out; | |
6275 | ||
c2c71324 SB |
6276 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { |
6277 | ret = begin_cmd(sctx, BTRFS_SEND_C_END); | |
6278 | if (ret < 0) | |
6279 | goto out; | |
6280 | ret = send_cmd(sctx); | |
6281 | if (ret < 0) | |
6282 | goto out; | |
6283 | } | |
31db9f7c AB |
6284 | |
6285 | out: | |
9f03740a FDBM |
6286 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); |
6287 | while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { | |
6288 | struct rb_node *n; | |
6289 | struct pending_dir_move *pm; | |
6290 | ||
6291 | n = rb_first(&sctx->pending_dir_moves); | |
6292 | pm = rb_entry(n, struct pending_dir_move, node); | |
6293 | while (!list_empty(&pm->list)) { | |
6294 | struct pending_dir_move *pm2; | |
6295 | ||
6296 | pm2 = list_first_entry(&pm->list, | |
6297 | struct pending_dir_move, list); | |
6298 | free_pending_move(sctx, pm2); | |
6299 | } | |
6300 | free_pending_move(sctx, pm); | |
6301 | } | |
6302 | ||
6303 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); | |
6304 | while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { | |
6305 | struct rb_node *n; | |
6306 | struct waiting_dir_move *dm; | |
6307 | ||
6308 | n = rb_first(&sctx->waiting_dir_moves); | |
6309 | dm = rb_entry(n, struct waiting_dir_move, node); | |
6310 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
6311 | kfree(dm); | |
6312 | } | |
6313 | ||
9dc44214 FM |
6314 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); |
6315 | while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { | |
6316 | struct rb_node *n; | |
6317 | struct orphan_dir_info *odi; | |
6318 | ||
6319 | n = rb_first(&sctx->orphan_dirs); | |
6320 | odi = rb_entry(n, struct orphan_dir_info, node); | |
6321 | free_orphan_dir_info(sctx, odi); | |
6322 | } | |
6323 | ||
896c14f9 WS |
6324 | if (sort_clone_roots) { |
6325 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
6326 | btrfs_root_dec_send_in_progress( | |
6327 | sctx->clone_roots[i].root); | |
6328 | } else { | |
6329 | for (i = 0; sctx && i < clone_sources_to_rollback; i++) | |
6330 | btrfs_root_dec_send_in_progress( | |
6331 | sctx->clone_roots[i].root); | |
6332 | ||
6333 | btrfs_root_dec_send_in_progress(send_root); | |
6334 | } | |
66ef7d65 DS |
6335 | if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) |
6336 | btrfs_root_dec_send_in_progress(sctx->parent_root); | |
2c686537 | 6337 | |
31db9f7c | 6338 | kfree(arg); |
2f91306a | 6339 | kvfree(clone_sources_tmp); |
31db9f7c AB |
6340 | |
6341 | if (sctx) { | |
6342 | if (sctx->send_filp) | |
6343 | fput(sctx->send_filp); | |
6344 | ||
c03d01f3 | 6345 | kvfree(sctx->clone_roots); |
6ff48ce0 | 6346 | kvfree(sctx->send_buf); |
eb5b75fe | 6347 | kvfree(sctx->read_buf); |
31db9f7c AB |
6348 | |
6349 | name_cache_free(sctx); | |
6350 | ||
6351 | kfree(sctx); | |
6352 | } | |
6353 | ||
6354 | return ret; | |
6355 | } |