]>
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" | |
37 | ||
38 | static int g_verbose = 0; | |
39 | ||
40 | #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__) | |
41 | ||
42 | /* | |
43 | * A fs_path is a helper to dynamically build path names with unknown size. | |
44 | * It reallocates the internal buffer on demand. | |
45 | * It allows fast adding of path elements on the right side (normal path) and | |
46 | * fast adding to the left side (reversed path). A reversed path can also be | |
47 | * unreversed if needed. | |
48 | */ | |
49 | struct fs_path { | |
50 | union { | |
51 | struct { | |
52 | char *start; | |
53 | char *end; | |
31db9f7c AB |
54 | |
55 | char *buf; | |
1f5a7ff9 DS |
56 | unsigned short buf_len:15; |
57 | unsigned short reversed:1; | |
31db9f7c AB |
58 | char inline_buf[]; |
59 | }; | |
ace01050 DS |
60 | /* |
61 | * Average path length does not exceed 200 bytes, we'll have | |
62 | * better packing in the slab and higher chance to satisfy | |
63 | * a allocation later during send. | |
64 | */ | |
65 | char pad[256]; | |
31db9f7c AB |
66 | }; |
67 | }; | |
68 | #define FS_PATH_INLINE_SIZE \ | |
69 | (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) | |
70 | ||
71 | ||
72 | /* reused for each extent */ | |
73 | struct clone_root { | |
74 | struct btrfs_root *root; | |
75 | u64 ino; | |
76 | u64 offset; | |
77 | ||
78 | u64 found_refs; | |
79 | }; | |
80 | ||
81 | #define SEND_CTX_MAX_NAME_CACHE_SIZE 128 | |
82 | #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2) | |
83 | ||
84 | struct send_ctx { | |
85 | struct file *send_filp; | |
86 | loff_t send_off; | |
87 | char *send_buf; | |
88 | u32 send_size; | |
89 | u32 send_max_size; | |
90 | u64 total_send_size; | |
91 | u64 cmd_send_size[BTRFS_SEND_C_MAX + 1]; | |
cb95e7bf | 92 | u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ |
31db9f7c | 93 | |
31db9f7c AB |
94 | struct btrfs_root *send_root; |
95 | struct btrfs_root *parent_root; | |
96 | struct clone_root *clone_roots; | |
97 | int clone_roots_cnt; | |
98 | ||
99 | /* current state of the compare_tree call */ | |
100 | struct btrfs_path *left_path; | |
101 | struct btrfs_path *right_path; | |
102 | struct btrfs_key *cmp_key; | |
103 | ||
104 | /* | |
105 | * infos of the currently processed inode. In case of deleted inodes, | |
106 | * these are the values from the deleted inode. | |
107 | */ | |
108 | u64 cur_ino; | |
109 | u64 cur_inode_gen; | |
110 | int cur_inode_new; | |
111 | int cur_inode_new_gen; | |
112 | int cur_inode_deleted; | |
31db9f7c AB |
113 | u64 cur_inode_size; |
114 | u64 cur_inode_mode; | |
644d1940 | 115 | u64 cur_inode_rdev; |
16e7549f | 116 | u64 cur_inode_last_extent; |
31db9f7c AB |
117 | |
118 | u64 send_progress; | |
119 | ||
120 | struct list_head new_refs; | |
121 | struct list_head deleted_refs; | |
122 | ||
123 | struct radix_tree_root name_cache; | |
124 | struct list_head name_cache_list; | |
125 | int name_cache_size; | |
126 | ||
2131bcd3 LB |
127 | struct file_ra_state ra; |
128 | ||
31db9f7c | 129 | char *read_buf; |
9f03740a FDBM |
130 | |
131 | /* | |
132 | * We process inodes by their increasing order, so if before an | |
133 | * incremental send we reverse the parent/child relationship of | |
134 | * directories such that a directory with a lower inode number was | |
135 | * the parent of a directory with a higher inode number, and the one | |
136 | * becoming the new parent got renamed too, we can't rename/move the | |
137 | * directory with lower inode number when we finish processing it - we | |
138 | * must process the directory with higher inode number first, then | |
139 | * rename/move it and then rename/move the directory with lower inode | |
140 | * number. Example follows. | |
141 | * | |
142 | * Tree state when the first send was performed: | |
143 | * | |
144 | * . | |
145 | * |-- a (ino 257) | |
146 | * |-- b (ino 258) | |
147 | * | | |
148 | * | | |
149 | * |-- c (ino 259) | |
150 | * | |-- d (ino 260) | |
151 | * | | |
152 | * |-- c2 (ino 261) | |
153 | * | |
154 | * Tree state when the second (incremental) send is performed: | |
155 | * | |
156 | * . | |
157 | * |-- a (ino 257) | |
158 | * |-- b (ino 258) | |
159 | * |-- c2 (ino 261) | |
160 | * |-- d2 (ino 260) | |
161 | * |-- cc (ino 259) | |
162 | * | |
163 | * The sequence of steps that lead to the second state was: | |
164 | * | |
165 | * mv /a/b/c/d /a/b/c2/d2 | |
166 | * mv /a/b/c /a/b/c2/d2/cc | |
167 | * | |
168 | * "c" has lower inode number, but we can't move it (2nd mv operation) | |
169 | * before we move "d", which has higher inode number. | |
170 | * | |
171 | * So we just memorize which move/rename operations must be performed | |
172 | * later when their respective parent is processed and moved/renamed. | |
173 | */ | |
174 | ||
175 | /* Indexed by parent directory inode number. */ | |
176 | struct rb_root pending_dir_moves; | |
177 | ||
178 | /* | |
179 | * Reverse index, indexed by the inode number of a directory that | |
180 | * is waiting for the move/rename of its immediate parent before its | |
181 | * own move/rename can be performed. | |
182 | */ | |
183 | struct rb_root waiting_dir_moves; | |
9dc44214 FM |
184 | |
185 | /* | |
186 | * A directory that is going to be rm'ed might have a child directory | |
187 | * which is in the pending directory moves index above. In this case, | |
188 | * the directory can only be removed after the move/rename of its child | |
189 | * is performed. Example: | |
190 | * | |
191 | * Parent snapshot: | |
192 | * | |
193 | * . (ino 256) | |
194 | * |-- a/ (ino 257) | |
195 | * |-- b/ (ino 258) | |
196 | * |-- c/ (ino 259) | |
197 | * | |-- x/ (ino 260) | |
198 | * | | |
199 | * |-- y/ (ino 261) | |
200 | * | |
201 | * Send snapshot: | |
202 | * | |
203 | * . (ino 256) | |
204 | * |-- a/ (ino 257) | |
205 | * |-- b/ (ino 258) | |
206 | * |-- YY/ (ino 261) | |
207 | * |-- x/ (ino 260) | |
208 | * | |
209 | * Sequence of steps that lead to the send snapshot: | |
210 | * rm -f /a/b/c/foo.txt | |
211 | * mv /a/b/y /a/b/YY | |
212 | * mv /a/b/c/x /a/b/YY | |
213 | * rmdir /a/b/c | |
214 | * | |
215 | * When the child is processed, its move/rename is delayed until its | |
216 | * parent is processed (as explained above), but all other operations | |
217 | * like update utimes, chown, chgrp, etc, are performed and the paths | |
218 | * that it uses for those operations must use the orphanized name of | |
219 | * its parent (the directory we're going to rm later), so we need to | |
220 | * memorize that name. | |
221 | * | |
222 | * Indexed by the inode number of the directory to be deleted. | |
223 | */ | |
224 | struct rb_root orphan_dirs; | |
9f03740a FDBM |
225 | }; |
226 | ||
227 | struct pending_dir_move { | |
228 | struct rb_node node; | |
229 | struct list_head list; | |
230 | u64 parent_ino; | |
231 | u64 ino; | |
232 | u64 gen; | |
233 | struct list_head update_refs; | |
234 | }; | |
235 | ||
236 | struct waiting_dir_move { | |
237 | struct rb_node node; | |
238 | u64 ino; | |
9dc44214 FM |
239 | /* |
240 | * There might be some directory that could not be removed because it | |
241 | * was waiting for this directory inode to be moved first. Therefore | |
242 | * after this directory is moved, we can try to rmdir the ino rmdir_ino. | |
243 | */ | |
244 | u64 rmdir_ino; | |
245 | }; | |
246 | ||
247 | struct orphan_dir_info { | |
248 | struct rb_node node; | |
249 | u64 ino; | |
250 | u64 gen; | |
31db9f7c AB |
251 | }; |
252 | ||
253 | struct name_cache_entry { | |
254 | struct list_head list; | |
7e0926fe AB |
255 | /* |
256 | * radix_tree has only 32bit entries but we need to handle 64bit inums. | |
257 | * We use the lower 32bit of the 64bit inum to store it in the tree. If | |
258 | * more then one inum would fall into the same entry, we use radix_list | |
259 | * to store the additional entries. radix_list is also used to store | |
260 | * entries where two entries have the same inum but different | |
261 | * generations. | |
262 | */ | |
263 | struct list_head radix_list; | |
31db9f7c AB |
264 | u64 ino; |
265 | u64 gen; | |
266 | u64 parent_ino; | |
267 | u64 parent_gen; | |
268 | int ret; | |
269 | int need_later_update; | |
270 | int name_len; | |
271 | char name[]; | |
272 | }; | |
273 | ||
9f03740a FDBM |
274 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); |
275 | ||
9dc44214 FM |
276 | static struct waiting_dir_move * |
277 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino); | |
278 | ||
279 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino); | |
280 | ||
16e7549f JB |
281 | static int need_send_hole(struct send_ctx *sctx) |
282 | { | |
283 | return (sctx->parent_root && !sctx->cur_inode_new && | |
284 | !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && | |
285 | S_ISREG(sctx->cur_inode_mode)); | |
286 | } | |
287 | ||
31db9f7c AB |
288 | static void fs_path_reset(struct fs_path *p) |
289 | { | |
290 | if (p->reversed) { | |
291 | p->start = p->buf + p->buf_len - 1; | |
292 | p->end = p->start; | |
293 | *p->start = 0; | |
294 | } else { | |
295 | p->start = p->buf; | |
296 | p->end = p->start; | |
297 | *p->start = 0; | |
298 | } | |
299 | } | |
300 | ||
924794c9 | 301 | static struct fs_path *fs_path_alloc(void) |
31db9f7c AB |
302 | { |
303 | struct fs_path *p; | |
304 | ||
305 | p = kmalloc(sizeof(*p), GFP_NOFS); | |
306 | if (!p) | |
307 | return NULL; | |
308 | p->reversed = 0; | |
31db9f7c AB |
309 | p->buf = p->inline_buf; |
310 | p->buf_len = FS_PATH_INLINE_SIZE; | |
311 | fs_path_reset(p); | |
312 | return p; | |
313 | } | |
314 | ||
924794c9 | 315 | static struct fs_path *fs_path_alloc_reversed(void) |
31db9f7c AB |
316 | { |
317 | struct fs_path *p; | |
318 | ||
924794c9 | 319 | p = fs_path_alloc(); |
31db9f7c AB |
320 | if (!p) |
321 | return NULL; | |
322 | p->reversed = 1; | |
323 | fs_path_reset(p); | |
324 | return p; | |
325 | } | |
326 | ||
924794c9 | 327 | static void fs_path_free(struct fs_path *p) |
31db9f7c AB |
328 | { |
329 | if (!p) | |
330 | return; | |
ace01050 DS |
331 | if (p->buf != p->inline_buf) |
332 | kfree(p->buf); | |
31db9f7c AB |
333 | kfree(p); |
334 | } | |
335 | ||
336 | static int fs_path_len(struct fs_path *p) | |
337 | { | |
338 | return p->end - p->start; | |
339 | } | |
340 | ||
341 | static int fs_path_ensure_buf(struct fs_path *p, int len) | |
342 | { | |
343 | char *tmp_buf; | |
344 | int path_len; | |
345 | int old_buf_len; | |
346 | ||
347 | len++; | |
348 | ||
349 | if (p->buf_len >= len) | |
350 | return 0; | |
351 | ||
1b2782c8 DS |
352 | path_len = p->end - p->start; |
353 | old_buf_len = p->buf_len; | |
354 | ||
ace01050 DS |
355 | /* |
356 | * First time the inline_buf does not suffice | |
357 | */ | |
9c9ca00b DS |
358 | if (p->buf == p->inline_buf) |
359 | tmp_buf = kmalloc(len, GFP_NOFS); | |
360 | else | |
361 | tmp_buf = krealloc(p->buf, len, GFP_NOFS); | |
362 | if (!tmp_buf) | |
363 | return -ENOMEM; | |
364 | p->buf = tmp_buf; | |
365 | /* | |
366 | * The real size of the buffer is bigger, this will let the fast path | |
367 | * happen most of the time | |
368 | */ | |
369 | p->buf_len = ksize(p->buf); | |
ace01050 | 370 | |
31db9f7c AB |
371 | if (p->reversed) { |
372 | tmp_buf = p->buf + old_buf_len - path_len - 1; | |
373 | p->end = p->buf + p->buf_len - 1; | |
374 | p->start = p->end - path_len; | |
375 | memmove(p->start, tmp_buf, path_len + 1); | |
376 | } else { | |
377 | p->start = p->buf; | |
378 | p->end = p->start + path_len; | |
379 | } | |
380 | return 0; | |
381 | } | |
382 | ||
b23ab57d DS |
383 | static int fs_path_prepare_for_add(struct fs_path *p, int name_len, |
384 | char **prepared) | |
31db9f7c AB |
385 | { |
386 | int ret; | |
387 | int new_len; | |
388 | ||
389 | new_len = p->end - p->start + name_len; | |
390 | if (p->start != p->end) | |
391 | new_len++; | |
392 | ret = fs_path_ensure_buf(p, new_len); | |
393 | if (ret < 0) | |
394 | goto out; | |
395 | ||
396 | if (p->reversed) { | |
397 | if (p->start != p->end) | |
398 | *--p->start = '/'; | |
399 | p->start -= name_len; | |
b23ab57d | 400 | *prepared = p->start; |
31db9f7c AB |
401 | } else { |
402 | if (p->start != p->end) | |
403 | *p->end++ = '/'; | |
b23ab57d | 404 | *prepared = p->end; |
31db9f7c AB |
405 | p->end += name_len; |
406 | *p->end = 0; | |
407 | } | |
408 | ||
409 | out: | |
410 | return ret; | |
411 | } | |
412 | ||
413 | static int fs_path_add(struct fs_path *p, const char *name, int name_len) | |
414 | { | |
415 | int ret; | |
b23ab57d | 416 | char *prepared; |
31db9f7c | 417 | |
b23ab57d | 418 | ret = fs_path_prepare_for_add(p, name_len, &prepared); |
31db9f7c AB |
419 | if (ret < 0) |
420 | goto out; | |
b23ab57d | 421 | memcpy(prepared, name, name_len); |
31db9f7c AB |
422 | |
423 | out: | |
424 | return ret; | |
425 | } | |
426 | ||
427 | static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) | |
428 | { | |
429 | int ret; | |
b23ab57d | 430 | char *prepared; |
31db9f7c | 431 | |
b23ab57d | 432 | ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); |
31db9f7c AB |
433 | if (ret < 0) |
434 | goto out; | |
b23ab57d | 435 | memcpy(prepared, p2->start, p2->end - p2->start); |
31db9f7c AB |
436 | |
437 | out: | |
438 | return ret; | |
439 | } | |
440 | ||
441 | static int fs_path_add_from_extent_buffer(struct fs_path *p, | |
442 | struct extent_buffer *eb, | |
443 | unsigned long off, int len) | |
444 | { | |
445 | int ret; | |
b23ab57d | 446 | char *prepared; |
31db9f7c | 447 | |
b23ab57d | 448 | ret = fs_path_prepare_for_add(p, len, &prepared); |
31db9f7c AB |
449 | if (ret < 0) |
450 | goto out; | |
451 | ||
b23ab57d | 452 | read_extent_buffer(eb, prepared, off, len); |
31db9f7c AB |
453 | |
454 | out: | |
455 | return ret; | |
456 | } | |
457 | ||
31db9f7c AB |
458 | static int fs_path_copy(struct fs_path *p, struct fs_path *from) |
459 | { | |
460 | int ret; | |
461 | ||
462 | p->reversed = from->reversed; | |
463 | fs_path_reset(p); | |
464 | ||
465 | ret = fs_path_add_path(p, from); | |
466 | ||
467 | return ret; | |
468 | } | |
469 | ||
470 | ||
471 | static void fs_path_unreverse(struct fs_path *p) | |
472 | { | |
473 | char *tmp; | |
474 | int len; | |
475 | ||
476 | if (!p->reversed) | |
477 | return; | |
478 | ||
479 | tmp = p->start; | |
480 | len = p->end - p->start; | |
481 | p->start = p->buf; | |
482 | p->end = p->start + len; | |
483 | memmove(p->start, tmp, len + 1); | |
484 | p->reversed = 0; | |
485 | } | |
486 | ||
487 | static struct btrfs_path *alloc_path_for_send(void) | |
488 | { | |
489 | struct btrfs_path *path; | |
490 | ||
491 | path = btrfs_alloc_path(); | |
492 | if (!path) | |
493 | return NULL; | |
494 | path->search_commit_root = 1; | |
495 | path->skip_locking = 1; | |
3f8a18cc | 496 | path->need_commit_sem = 1; |
31db9f7c AB |
497 | return path; |
498 | } | |
499 | ||
48a3b636 | 500 | static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) |
31db9f7c AB |
501 | { |
502 | int ret; | |
503 | mm_segment_t old_fs; | |
504 | u32 pos = 0; | |
505 | ||
506 | old_fs = get_fs(); | |
507 | set_fs(KERNEL_DS); | |
508 | ||
509 | while (pos < len) { | |
1bcea355 | 510 | ret = vfs_write(filp, (char *)buf + pos, len - pos, off); |
31db9f7c AB |
511 | /* TODO handle that correctly */ |
512 | /*if (ret == -ERESTARTSYS) { | |
513 | continue; | |
514 | }*/ | |
515 | if (ret < 0) | |
516 | goto out; | |
517 | if (ret == 0) { | |
518 | ret = -EIO; | |
519 | goto out; | |
520 | } | |
521 | pos += ret; | |
522 | } | |
523 | ||
524 | ret = 0; | |
525 | ||
526 | out: | |
527 | set_fs(old_fs); | |
528 | return ret; | |
529 | } | |
530 | ||
531 | static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) | |
532 | { | |
533 | struct btrfs_tlv_header *hdr; | |
534 | int total_len = sizeof(*hdr) + len; | |
535 | int left = sctx->send_max_size - sctx->send_size; | |
536 | ||
537 | if (unlikely(left < total_len)) | |
538 | return -EOVERFLOW; | |
539 | ||
540 | hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); | |
541 | hdr->tlv_type = cpu_to_le16(attr); | |
542 | hdr->tlv_len = cpu_to_le16(len); | |
543 | memcpy(hdr + 1, data, len); | |
544 | sctx->send_size += total_len; | |
545 | ||
546 | return 0; | |
547 | } | |
548 | ||
95bc79d5 DS |
549 | #define TLV_PUT_DEFINE_INT(bits) \ |
550 | static int tlv_put_u##bits(struct send_ctx *sctx, \ | |
551 | u##bits attr, u##bits value) \ | |
552 | { \ | |
553 | __le##bits __tmp = cpu_to_le##bits(value); \ | |
554 | return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ | |
555 | } | |
31db9f7c | 556 | |
95bc79d5 | 557 | TLV_PUT_DEFINE_INT(64) |
31db9f7c AB |
558 | |
559 | static int tlv_put_string(struct send_ctx *sctx, u16 attr, | |
560 | const char *str, int len) | |
561 | { | |
562 | if (len == -1) | |
563 | len = strlen(str); | |
564 | return tlv_put(sctx, attr, str, len); | |
565 | } | |
566 | ||
567 | static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, | |
568 | const u8 *uuid) | |
569 | { | |
570 | return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); | |
571 | } | |
572 | ||
31db9f7c AB |
573 | static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, |
574 | struct extent_buffer *eb, | |
575 | struct btrfs_timespec *ts) | |
576 | { | |
577 | struct btrfs_timespec bts; | |
578 | read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); | |
579 | return tlv_put(sctx, attr, &bts, sizeof(bts)); | |
580 | } | |
581 | ||
582 | ||
583 | #define TLV_PUT(sctx, attrtype, attrlen, data) \ | |
584 | do { \ | |
585 | ret = tlv_put(sctx, attrtype, attrlen, data); \ | |
586 | if (ret < 0) \ | |
587 | goto tlv_put_failure; \ | |
588 | } while (0) | |
589 | ||
590 | #define TLV_PUT_INT(sctx, attrtype, bits, value) \ | |
591 | do { \ | |
592 | ret = tlv_put_u##bits(sctx, attrtype, value); \ | |
593 | if (ret < 0) \ | |
594 | goto tlv_put_failure; \ | |
595 | } while (0) | |
596 | ||
597 | #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) | |
598 | #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) | |
599 | #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) | |
600 | #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) | |
601 | #define TLV_PUT_STRING(sctx, attrtype, str, len) \ | |
602 | do { \ | |
603 | ret = tlv_put_string(sctx, attrtype, str, len); \ | |
604 | if (ret < 0) \ | |
605 | goto tlv_put_failure; \ | |
606 | } while (0) | |
607 | #define TLV_PUT_PATH(sctx, attrtype, p) \ | |
608 | do { \ | |
609 | ret = tlv_put_string(sctx, attrtype, p->start, \ | |
610 | p->end - p->start); \ | |
611 | if (ret < 0) \ | |
612 | goto tlv_put_failure; \ | |
613 | } while(0) | |
614 | #define TLV_PUT_UUID(sctx, attrtype, uuid) \ | |
615 | do { \ | |
616 | ret = tlv_put_uuid(sctx, attrtype, uuid); \ | |
617 | if (ret < 0) \ | |
618 | goto tlv_put_failure; \ | |
619 | } while (0) | |
31db9f7c AB |
620 | #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ |
621 | do { \ | |
622 | ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ | |
623 | if (ret < 0) \ | |
624 | goto tlv_put_failure; \ | |
625 | } while (0) | |
626 | ||
627 | static int send_header(struct send_ctx *sctx) | |
628 | { | |
629 | struct btrfs_stream_header hdr; | |
630 | ||
631 | strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); | |
632 | hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); | |
633 | ||
1bcea355 AJ |
634 | return write_buf(sctx->send_filp, &hdr, sizeof(hdr), |
635 | &sctx->send_off); | |
31db9f7c AB |
636 | } |
637 | ||
638 | /* | |
639 | * For each command/item we want to send to userspace, we call this function. | |
640 | */ | |
641 | static int begin_cmd(struct send_ctx *sctx, int cmd) | |
642 | { | |
643 | struct btrfs_cmd_header *hdr; | |
644 | ||
fae7f21c | 645 | if (WARN_ON(!sctx->send_buf)) |
31db9f7c | 646 | return -EINVAL; |
31db9f7c AB |
647 | |
648 | BUG_ON(sctx->send_size); | |
649 | ||
650 | sctx->send_size += sizeof(*hdr); | |
651 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
652 | hdr->cmd = cpu_to_le16(cmd); | |
653 | ||
654 | return 0; | |
655 | } | |
656 | ||
657 | static int send_cmd(struct send_ctx *sctx) | |
658 | { | |
659 | int ret; | |
660 | struct btrfs_cmd_header *hdr; | |
661 | u32 crc; | |
662 | ||
663 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
664 | hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr)); | |
665 | hdr->crc = 0; | |
666 | ||
0b947aff | 667 | crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); |
31db9f7c AB |
668 | hdr->crc = cpu_to_le32(crc); |
669 | ||
1bcea355 AJ |
670 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, |
671 | &sctx->send_off); | |
31db9f7c AB |
672 | |
673 | sctx->total_send_size += sctx->send_size; | |
674 | sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size; | |
675 | sctx->send_size = 0; | |
676 | ||
677 | return ret; | |
678 | } | |
679 | ||
680 | /* | |
681 | * Sends a move instruction to user space | |
682 | */ | |
683 | static int send_rename(struct send_ctx *sctx, | |
684 | struct fs_path *from, struct fs_path *to) | |
685 | { | |
686 | int ret; | |
687 | ||
688 | verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start); | |
689 | ||
690 | ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); | |
691 | if (ret < 0) | |
692 | goto out; | |
693 | ||
694 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); | |
695 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); | |
696 | ||
697 | ret = send_cmd(sctx); | |
698 | ||
699 | tlv_put_failure: | |
700 | out: | |
701 | return ret; | |
702 | } | |
703 | ||
704 | /* | |
705 | * Sends a link instruction to user space | |
706 | */ | |
707 | static int send_link(struct send_ctx *sctx, | |
708 | struct fs_path *path, struct fs_path *lnk) | |
709 | { | |
710 | int ret; | |
711 | ||
712 | verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start); | |
713 | ||
714 | ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); | |
715 | if (ret < 0) | |
716 | goto out; | |
717 | ||
718 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
719 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); | |
720 | ||
721 | ret = send_cmd(sctx); | |
722 | ||
723 | tlv_put_failure: | |
724 | out: | |
725 | return ret; | |
726 | } | |
727 | ||
728 | /* | |
729 | * Sends an unlink instruction to user space | |
730 | */ | |
731 | static int send_unlink(struct send_ctx *sctx, struct fs_path *path) | |
732 | { | |
733 | int ret; | |
734 | ||
735 | verbose_printk("btrfs: send_unlink %s\n", path->start); | |
736 | ||
737 | ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); | |
738 | if (ret < 0) | |
739 | goto out; | |
740 | ||
741 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
742 | ||
743 | ret = send_cmd(sctx); | |
744 | ||
745 | tlv_put_failure: | |
746 | out: | |
747 | return ret; | |
748 | } | |
749 | ||
750 | /* | |
751 | * Sends a rmdir instruction to user space | |
752 | */ | |
753 | static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) | |
754 | { | |
755 | int ret; | |
756 | ||
757 | verbose_printk("btrfs: send_rmdir %s\n", path->start); | |
758 | ||
759 | ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); | |
760 | if (ret < 0) | |
761 | goto out; | |
762 | ||
763 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
764 | ||
765 | ret = send_cmd(sctx); | |
766 | ||
767 | tlv_put_failure: | |
768 | out: | |
769 | return ret; | |
770 | } | |
771 | ||
772 | /* | |
773 | * Helper function to retrieve some fields from an inode item. | |
774 | */ | |
3f8a18cc JB |
775 | static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, |
776 | u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, | |
777 | u64 *gid, u64 *rdev) | |
31db9f7c AB |
778 | { |
779 | int ret; | |
780 | struct btrfs_inode_item *ii; | |
781 | struct btrfs_key key; | |
31db9f7c AB |
782 | |
783 | key.objectid = ino; | |
784 | key.type = BTRFS_INODE_ITEM_KEY; | |
785 | key.offset = 0; | |
786 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
31db9f7c | 787 | if (ret) { |
3f8a18cc JB |
788 | if (ret > 0) |
789 | ret = -ENOENT; | |
790 | return ret; | |
31db9f7c AB |
791 | } |
792 | ||
793 | ii = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
794 | struct btrfs_inode_item); | |
795 | if (size) | |
796 | *size = btrfs_inode_size(path->nodes[0], ii); | |
797 | if (gen) | |
798 | *gen = btrfs_inode_generation(path->nodes[0], ii); | |
799 | if (mode) | |
800 | *mode = btrfs_inode_mode(path->nodes[0], ii); | |
801 | if (uid) | |
802 | *uid = btrfs_inode_uid(path->nodes[0], ii); | |
803 | if (gid) | |
804 | *gid = btrfs_inode_gid(path->nodes[0], ii); | |
85a7b33b AB |
805 | if (rdev) |
806 | *rdev = btrfs_inode_rdev(path->nodes[0], ii); | |
31db9f7c | 807 | |
3f8a18cc JB |
808 | return ret; |
809 | } | |
810 | ||
811 | static int get_inode_info(struct btrfs_root *root, | |
812 | u64 ino, u64 *size, u64 *gen, | |
813 | u64 *mode, u64 *uid, u64 *gid, | |
814 | u64 *rdev) | |
815 | { | |
816 | struct btrfs_path *path; | |
817 | int ret; | |
818 | ||
819 | path = alloc_path_for_send(); | |
820 | if (!path) | |
821 | return -ENOMEM; | |
822 | ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, | |
823 | rdev); | |
31db9f7c AB |
824 | btrfs_free_path(path); |
825 | return ret; | |
826 | } | |
827 | ||
828 | typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, | |
829 | struct fs_path *p, | |
830 | void *ctx); | |
831 | ||
832 | /* | |
96b5bd77 JS |
833 | * Helper function to iterate the entries in ONE btrfs_inode_ref or |
834 | * btrfs_inode_extref. | |
31db9f7c AB |
835 | * The iterate callback may return a non zero value to stop iteration. This can |
836 | * be a negative value for error codes or 1 to simply stop it. | |
837 | * | |
96b5bd77 | 838 | * path must point to the INODE_REF or INODE_EXTREF when called. |
31db9f7c | 839 | */ |
924794c9 | 840 | static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
841 | struct btrfs_key *found_key, int resolve, |
842 | iterate_inode_ref_t iterate, void *ctx) | |
843 | { | |
96b5bd77 | 844 | struct extent_buffer *eb = path->nodes[0]; |
31db9f7c AB |
845 | struct btrfs_item *item; |
846 | struct btrfs_inode_ref *iref; | |
96b5bd77 | 847 | struct btrfs_inode_extref *extref; |
31db9f7c AB |
848 | struct btrfs_path *tmp_path; |
849 | struct fs_path *p; | |
96b5bd77 | 850 | u32 cur = 0; |
31db9f7c | 851 | u32 total; |
96b5bd77 | 852 | int slot = path->slots[0]; |
31db9f7c AB |
853 | u32 name_len; |
854 | char *start; | |
855 | int ret = 0; | |
96b5bd77 | 856 | int num = 0; |
31db9f7c | 857 | int index; |
96b5bd77 JS |
858 | u64 dir; |
859 | unsigned long name_off; | |
860 | unsigned long elem_size; | |
861 | unsigned long ptr; | |
31db9f7c | 862 | |
924794c9 | 863 | p = fs_path_alloc_reversed(); |
31db9f7c AB |
864 | if (!p) |
865 | return -ENOMEM; | |
866 | ||
867 | tmp_path = alloc_path_for_send(); | |
868 | if (!tmp_path) { | |
924794c9 | 869 | fs_path_free(p); |
31db9f7c AB |
870 | return -ENOMEM; |
871 | } | |
872 | ||
31db9f7c | 873 | |
96b5bd77 JS |
874 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
875 | ptr = (unsigned long)btrfs_item_ptr(eb, slot, | |
876 | struct btrfs_inode_ref); | |
dd3cc16b | 877 | item = btrfs_item_nr(slot); |
96b5bd77 JS |
878 | total = btrfs_item_size(eb, item); |
879 | elem_size = sizeof(*iref); | |
880 | } else { | |
881 | ptr = btrfs_item_ptr_offset(eb, slot); | |
882 | total = btrfs_item_size_nr(eb, slot); | |
883 | elem_size = sizeof(*extref); | |
884 | } | |
885 | ||
31db9f7c AB |
886 | while (cur < total) { |
887 | fs_path_reset(p); | |
888 | ||
96b5bd77 JS |
889 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
890 | iref = (struct btrfs_inode_ref *)(ptr + cur); | |
891 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
892 | name_off = (unsigned long)(iref + 1); | |
893 | index = btrfs_inode_ref_index(eb, iref); | |
894 | dir = found_key->offset; | |
895 | } else { | |
896 | extref = (struct btrfs_inode_extref *)(ptr + cur); | |
897 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
898 | name_off = (unsigned long)&extref->name; | |
899 | index = btrfs_inode_extref_index(eb, extref); | |
900 | dir = btrfs_inode_extref_parent(eb, extref); | |
901 | } | |
902 | ||
31db9f7c | 903 | if (resolve) { |
96b5bd77 JS |
904 | start = btrfs_ref_to_path(root, tmp_path, name_len, |
905 | name_off, eb, dir, | |
906 | p->buf, p->buf_len); | |
31db9f7c AB |
907 | if (IS_ERR(start)) { |
908 | ret = PTR_ERR(start); | |
909 | goto out; | |
910 | } | |
911 | if (start < p->buf) { | |
912 | /* overflow , try again with larger buffer */ | |
913 | ret = fs_path_ensure_buf(p, | |
914 | p->buf_len + p->buf - start); | |
915 | if (ret < 0) | |
916 | goto out; | |
96b5bd77 JS |
917 | start = btrfs_ref_to_path(root, tmp_path, |
918 | name_len, name_off, | |
919 | eb, dir, | |
920 | p->buf, p->buf_len); | |
31db9f7c AB |
921 | if (IS_ERR(start)) { |
922 | ret = PTR_ERR(start); | |
923 | goto out; | |
924 | } | |
925 | BUG_ON(start < p->buf); | |
926 | } | |
927 | p->start = start; | |
928 | } else { | |
96b5bd77 JS |
929 | ret = fs_path_add_from_extent_buffer(p, eb, name_off, |
930 | name_len); | |
31db9f7c AB |
931 | if (ret < 0) |
932 | goto out; | |
933 | } | |
934 | ||
96b5bd77 JS |
935 | cur += elem_size + name_len; |
936 | ret = iterate(num, dir, index, p, ctx); | |
31db9f7c AB |
937 | if (ret) |
938 | goto out; | |
31db9f7c AB |
939 | num++; |
940 | } | |
941 | ||
942 | out: | |
943 | btrfs_free_path(tmp_path); | |
924794c9 | 944 | fs_path_free(p); |
31db9f7c AB |
945 | return ret; |
946 | } | |
947 | ||
948 | typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, | |
949 | const char *name, int name_len, | |
950 | const char *data, int data_len, | |
951 | u8 type, void *ctx); | |
952 | ||
953 | /* | |
954 | * Helper function to iterate the entries in ONE btrfs_dir_item. | |
955 | * The iterate callback may return a non zero value to stop iteration. This can | |
956 | * be a negative value for error codes or 1 to simply stop it. | |
957 | * | |
958 | * path must point to the dir item when called. | |
959 | */ | |
924794c9 | 960 | static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
961 | struct btrfs_key *found_key, |
962 | iterate_dir_item_t iterate, void *ctx) | |
963 | { | |
964 | int ret = 0; | |
965 | struct extent_buffer *eb; | |
966 | struct btrfs_item *item; | |
967 | struct btrfs_dir_item *di; | |
31db9f7c AB |
968 | struct btrfs_key di_key; |
969 | char *buf = NULL; | |
ace01050 | 970 | const int buf_len = PATH_MAX; |
31db9f7c AB |
971 | u32 name_len; |
972 | u32 data_len; | |
973 | u32 cur; | |
974 | u32 len; | |
975 | u32 total; | |
976 | int slot; | |
977 | int num; | |
978 | u8 type; | |
979 | ||
31db9f7c AB |
980 | buf = kmalloc(buf_len, GFP_NOFS); |
981 | if (!buf) { | |
982 | ret = -ENOMEM; | |
983 | goto out; | |
984 | } | |
985 | ||
31db9f7c AB |
986 | eb = path->nodes[0]; |
987 | slot = path->slots[0]; | |
dd3cc16b | 988 | item = btrfs_item_nr(slot); |
31db9f7c AB |
989 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
990 | cur = 0; | |
991 | len = 0; | |
992 | total = btrfs_item_size(eb, item); | |
993 | ||
994 | num = 0; | |
995 | while (cur < total) { | |
996 | name_len = btrfs_dir_name_len(eb, di); | |
997 | data_len = btrfs_dir_data_len(eb, di); | |
998 | type = btrfs_dir_type(eb, di); | |
999 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
1000 | ||
ace01050 DS |
1001 | /* |
1002 | * Path too long | |
1003 | */ | |
31db9f7c | 1004 | if (name_len + data_len > buf_len) { |
ace01050 DS |
1005 | ret = -ENAMETOOLONG; |
1006 | goto out; | |
31db9f7c AB |
1007 | } |
1008 | ||
1009 | read_extent_buffer(eb, buf, (unsigned long)(di + 1), | |
1010 | name_len + data_len); | |
1011 | ||
1012 | len = sizeof(*di) + name_len + data_len; | |
1013 | di = (struct btrfs_dir_item *)((char *)di + len); | |
1014 | cur += len; | |
1015 | ||
1016 | ret = iterate(num, &di_key, buf, name_len, buf + name_len, | |
1017 | data_len, type, ctx); | |
1018 | if (ret < 0) | |
1019 | goto out; | |
1020 | if (ret) { | |
1021 | ret = 0; | |
1022 | goto out; | |
1023 | } | |
1024 | ||
1025 | num++; | |
1026 | } | |
1027 | ||
1028 | out: | |
ace01050 | 1029 | kfree(buf); |
31db9f7c AB |
1030 | return ret; |
1031 | } | |
1032 | ||
1033 | static int __copy_first_ref(int num, u64 dir, int index, | |
1034 | struct fs_path *p, void *ctx) | |
1035 | { | |
1036 | int ret; | |
1037 | struct fs_path *pt = ctx; | |
1038 | ||
1039 | ret = fs_path_copy(pt, p); | |
1040 | if (ret < 0) | |
1041 | return ret; | |
1042 | ||
1043 | /* we want the first only */ | |
1044 | return 1; | |
1045 | } | |
1046 | ||
1047 | /* | |
1048 | * Retrieve the first path of an inode. If an inode has more then one | |
1049 | * ref/hardlink, this is ignored. | |
1050 | */ | |
924794c9 | 1051 | static int get_inode_path(struct btrfs_root *root, |
31db9f7c AB |
1052 | u64 ino, struct fs_path *path) |
1053 | { | |
1054 | int ret; | |
1055 | struct btrfs_key key, found_key; | |
1056 | struct btrfs_path *p; | |
1057 | ||
1058 | p = alloc_path_for_send(); | |
1059 | if (!p) | |
1060 | return -ENOMEM; | |
1061 | ||
1062 | fs_path_reset(path); | |
1063 | ||
1064 | key.objectid = ino; | |
1065 | key.type = BTRFS_INODE_REF_KEY; | |
1066 | key.offset = 0; | |
1067 | ||
1068 | ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); | |
1069 | if (ret < 0) | |
1070 | goto out; | |
1071 | if (ret) { | |
1072 | ret = 1; | |
1073 | goto out; | |
1074 | } | |
1075 | btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); | |
1076 | if (found_key.objectid != ino || | |
96b5bd77 JS |
1077 | (found_key.type != BTRFS_INODE_REF_KEY && |
1078 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1079 | ret = -ENOENT; |
1080 | goto out; | |
1081 | } | |
1082 | ||
924794c9 TI |
1083 | ret = iterate_inode_ref(root, p, &found_key, 1, |
1084 | __copy_first_ref, path); | |
31db9f7c AB |
1085 | if (ret < 0) |
1086 | goto out; | |
1087 | ret = 0; | |
1088 | ||
1089 | out: | |
1090 | btrfs_free_path(p); | |
1091 | return ret; | |
1092 | } | |
1093 | ||
1094 | struct backref_ctx { | |
1095 | struct send_ctx *sctx; | |
1096 | ||
3f8a18cc | 1097 | struct btrfs_path *path; |
31db9f7c AB |
1098 | /* number of total found references */ |
1099 | u64 found; | |
1100 | ||
1101 | /* | |
1102 | * used for clones found in send_root. clones found behind cur_objectid | |
1103 | * and cur_offset are not considered as allowed clones. | |
1104 | */ | |
1105 | u64 cur_objectid; | |
1106 | u64 cur_offset; | |
1107 | ||
1108 | /* may be truncated in case it's the last extent in a file */ | |
1109 | u64 extent_len; | |
1110 | ||
1111 | /* Just to check for bugs in backref resolving */ | |
ee849c04 | 1112 | int found_itself; |
31db9f7c AB |
1113 | }; |
1114 | ||
1115 | static int __clone_root_cmp_bsearch(const void *key, const void *elt) | |
1116 | { | |
995e01b7 | 1117 | u64 root = (u64)(uintptr_t)key; |
31db9f7c AB |
1118 | struct clone_root *cr = (struct clone_root *)elt; |
1119 | ||
1120 | if (root < cr->root->objectid) | |
1121 | return -1; | |
1122 | if (root > cr->root->objectid) | |
1123 | return 1; | |
1124 | return 0; | |
1125 | } | |
1126 | ||
1127 | static int __clone_root_cmp_sort(const void *e1, const void *e2) | |
1128 | { | |
1129 | struct clone_root *cr1 = (struct clone_root *)e1; | |
1130 | struct clone_root *cr2 = (struct clone_root *)e2; | |
1131 | ||
1132 | if (cr1->root->objectid < cr2->root->objectid) | |
1133 | return -1; | |
1134 | if (cr1->root->objectid > cr2->root->objectid) | |
1135 | return 1; | |
1136 | return 0; | |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * Called for every backref that is found for the current extent. | |
766702ef | 1141 | * Results are collected in sctx->clone_roots->ino/offset/found_refs |
31db9f7c AB |
1142 | */ |
1143 | static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) | |
1144 | { | |
1145 | struct backref_ctx *bctx = ctx_; | |
1146 | struct clone_root *found; | |
1147 | int ret; | |
1148 | u64 i_size; | |
1149 | ||
1150 | /* First check if the root is in the list of accepted clone sources */ | |
995e01b7 | 1151 | found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, |
31db9f7c AB |
1152 | bctx->sctx->clone_roots_cnt, |
1153 | sizeof(struct clone_root), | |
1154 | __clone_root_cmp_bsearch); | |
1155 | if (!found) | |
1156 | return 0; | |
1157 | ||
1158 | if (found->root == bctx->sctx->send_root && | |
1159 | ino == bctx->cur_objectid && | |
1160 | offset == bctx->cur_offset) { | |
ee849c04 | 1161 | bctx->found_itself = 1; |
31db9f7c AB |
1162 | } |
1163 | ||
1164 | /* | |
766702ef | 1165 | * There are inodes that have extents that lie behind its i_size. Don't |
31db9f7c AB |
1166 | * accept clones from these extents. |
1167 | */ | |
3f8a18cc JB |
1168 | ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL, |
1169 | NULL, NULL, NULL); | |
1170 | btrfs_release_path(bctx->path); | |
31db9f7c AB |
1171 | if (ret < 0) |
1172 | return ret; | |
1173 | ||
1174 | if (offset + bctx->extent_len > i_size) | |
1175 | return 0; | |
1176 | ||
1177 | /* | |
1178 | * Make sure we don't consider clones from send_root that are | |
1179 | * behind the current inode/offset. | |
1180 | */ | |
1181 | if (found->root == bctx->sctx->send_root) { | |
1182 | /* | |
1183 | * TODO for the moment we don't accept clones from the inode | |
1184 | * that is currently send. We may change this when | |
1185 | * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same | |
1186 | * file. | |
1187 | */ | |
1188 | if (ino >= bctx->cur_objectid) | |
1189 | return 0; | |
e938c8ad AB |
1190 | #if 0 |
1191 | if (ino > bctx->cur_objectid) | |
1192 | return 0; | |
1193 | if (offset + bctx->extent_len > bctx->cur_offset) | |
31db9f7c | 1194 | return 0; |
e938c8ad | 1195 | #endif |
31db9f7c AB |
1196 | } |
1197 | ||
1198 | bctx->found++; | |
1199 | found->found_refs++; | |
1200 | if (ino < found->ino) { | |
1201 | found->ino = ino; | |
1202 | found->offset = offset; | |
1203 | } else if (found->ino == ino) { | |
1204 | /* | |
1205 | * same extent found more then once in the same file. | |
1206 | */ | |
1207 | if (found->offset > offset + bctx->extent_len) | |
1208 | found->offset = offset; | |
1209 | } | |
1210 | ||
1211 | return 0; | |
1212 | } | |
1213 | ||
1214 | /* | |
766702ef AB |
1215 | * Given an inode, offset and extent item, it finds a good clone for a clone |
1216 | * instruction. Returns -ENOENT when none could be found. The function makes | |
1217 | * sure that the returned clone is usable at the point where sending is at the | |
1218 | * moment. This means, that no clones are accepted which lie behind the current | |
1219 | * inode+offset. | |
1220 | * | |
31db9f7c AB |
1221 | * path must point to the extent item when called. |
1222 | */ | |
1223 | static int find_extent_clone(struct send_ctx *sctx, | |
1224 | struct btrfs_path *path, | |
1225 | u64 ino, u64 data_offset, | |
1226 | u64 ino_size, | |
1227 | struct clone_root **found) | |
1228 | { | |
1229 | int ret; | |
1230 | int extent_type; | |
1231 | u64 logical; | |
74dd17fb | 1232 | u64 disk_byte; |
31db9f7c AB |
1233 | u64 num_bytes; |
1234 | u64 extent_item_pos; | |
69917e43 | 1235 | u64 flags = 0; |
31db9f7c AB |
1236 | struct btrfs_file_extent_item *fi; |
1237 | struct extent_buffer *eb = path->nodes[0]; | |
35075bb0 | 1238 | struct backref_ctx *backref_ctx = NULL; |
31db9f7c AB |
1239 | struct clone_root *cur_clone_root; |
1240 | struct btrfs_key found_key; | |
1241 | struct btrfs_path *tmp_path; | |
74dd17fb | 1242 | int compressed; |
31db9f7c AB |
1243 | u32 i; |
1244 | ||
1245 | tmp_path = alloc_path_for_send(); | |
1246 | if (!tmp_path) | |
1247 | return -ENOMEM; | |
1248 | ||
3f8a18cc JB |
1249 | /* We only use this path under the commit sem */ |
1250 | tmp_path->need_commit_sem = 0; | |
1251 | ||
35075bb0 AB |
1252 | backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS); |
1253 | if (!backref_ctx) { | |
1254 | ret = -ENOMEM; | |
1255 | goto out; | |
1256 | } | |
1257 | ||
3f8a18cc JB |
1258 | backref_ctx->path = tmp_path; |
1259 | ||
31db9f7c AB |
1260 | if (data_offset >= ino_size) { |
1261 | /* | |
1262 | * There may be extents that lie behind the file's size. | |
1263 | * I at least had this in combination with snapshotting while | |
1264 | * writing large files. | |
1265 | */ | |
1266 | ret = 0; | |
1267 | goto out; | |
1268 | } | |
1269 | ||
1270 | fi = btrfs_item_ptr(eb, path->slots[0], | |
1271 | struct btrfs_file_extent_item); | |
1272 | extent_type = btrfs_file_extent_type(eb, fi); | |
1273 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1274 | ret = -ENOENT; | |
1275 | goto out; | |
1276 | } | |
74dd17fb | 1277 | compressed = btrfs_file_extent_compression(eb, fi); |
31db9f7c AB |
1278 | |
1279 | num_bytes = btrfs_file_extent_num_bytes(eb, fi); | |
74dd17fb CM |
1280 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
1281 | if (disk_byte == 0) { | |
31db9f7c AB |
1282 | ret = -ENOENT; |
1283 | goto out; | |
1284 | } | |
74dd17fb | 1285 | logical = disk_byte + btrfs_file_extent_offset(eb, fi); |
31db9f7c | 1286 | |
9e351cc8 | 1287 | down_read(&sctx->send_root->fs_info->commit_root_sem); |
69917e43 LB |
1288 | ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path, |
1289 | &found_key, &flags); | |
9e351cc8 | 1290 | up_read(&sctx->send_root->fs_info->commit_root_sem); |
31db9f7c AB |
1291 | btrfs_release_path(tmp_path); |
1292 | ||
1293 | if (ret < 0) | |
1294 | goto out; | |
69917e43 | 1295 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { |
31db9f7c AB |
1296 | ret = -EIO; |
1297 | goto out; | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Setup the clone roots. | |
1302 | */ | |
1303 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1304 | cur_clone_root = sctx->clone_roots + i; | |
1305 | cur_clone_root->ino = (u64)-1; | |
1306 | cur_clone_root->offset = 0; | |
1307 | cur_clone_root->found_refs = 0; | |
1308 | } | |
1309 | ||
35075bb0 AB |
1310 | backref_ctx->sctx = sctx; |
1311 | backref_ctx->found = 0; | |
1312 | backref_ctx->cur_objectid = ino; | |
1313 | backref_ctx->cur_offset = data_offset; | |
1314 | backref_ctx->found_itself = 0; | |
1315 | backref_ctx->extent_len = num_bytes; | |
31db9f7c AB |
1316 | |
1317 | /* | |
1318 | * The last extent of a file may be too large due to page alignment. | |
1319 | * We need to adjust extent_len in this case so that the checks in | |
1320 | * __iterate_backrefs work. | |
1321 | */ | |
1322 | if (data_offset + num_bytes >= ino_size) | |
35075bb0 | 1323 | backref_ctx->extent_len = ino_size - data_offset; |
31db9f7c AB |
1324 | |
1325 | /* | |
1326 | * Now collect all backrefs. | |
1327 | */ | |
74dd17fb CM |
1328 | if (compressed == BTRFS_COMPRESS_NONE) |
1329 | extent_item_pos = logical - found_key.objectid; | |
1330 | else | |
1331 | extent_item_pos = 0; | |
31db9f7c AB |
1332 | ret = iterate_extent_inodes(sctx->send_root->fs_info, |
1333 | found_key.objectid, extent_item_pos, 1, | |
35075bb0 | 1334 | __iterate_backrefs, backref_ctx); |
74dd17fb | 1335 | |
31db9f7c AB |
1336 | if (ret < 0) |
1337 | goto out; | |
1338 | ||
35075bb0 | 1339 | if (!backref_ctx->found_itself) { |
31db9f7c AB |
1340 | /* found a bug in backref code? */ |
1341 | ret = -EIO; | |
efe120a0 | 1342 | btrfs_err(sctx->send_root->fs_info, "did not find backref in " |
31db9f7c | 1343 | "send_root. inode=%llu, offset=%llu, " |
74dd17fb CM |
1344 | "disk_byte=%llu found extent=%llu\n", |
1345 | ino, data_offset, disk_byte, found_key.objectid); | |
31db9f7c AB |
1346 | goto out; |
1347 | } | |
1348 | ||
1349 | verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, " | |
1350 | "ino=%llu, " | |
1351 | "num_bytes=%llu, logical=%llu\n", | |
1352 | data_offset, ino, num_bytes, logical); | |
1353 | ||
35075bb0 | 1354 | if (!backref_ctx->found) |
31db9f7c AB |
1355 | verbose_printk("btrfs: no clones found\n"); |
1356 | ||
1357 | cur_clone_root = NULL; | |
1358 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1359 | if (sctx->clone_roots[i].found_refs) { | |
1360 | if (!cur_clone_root) | |
1361 | cur_clone_root = sctx->clone_roots + i; | |
1362 | else if (sctx->clone_roots[i].root == sctx->send_root) | |
1363 | /* prefer clones from send_root over others */ | |
1364 | cur_clone_root = sctx->clone_roots + i; | |
31db9f7c AB |
1365 | } |
1366 | ||
1367 | } | |
1368 | ||
1369 | if (cur_clone_root) { | |
93de4ba8 FDBM |
1370 | if (compressed != BTRFS_COMPRESS_NONE) { |
1371 | /* | |
1372 | * Offsets given by iterate_extent_inodes() are relative | |
1373 | * to the start of the extent, we need to add logical | |
1374 | * offset from the file extent item. | |
1375 | * (See why at backref.c:check_extent_in_eb()) | |
1376 | */ | |
1377 | cur_clone_root->offset += btrfs_file_extent_offset(eb, | |
1378 | fi); | |
1379 | } | |
31db9f7c AB |
1380 | *found = cur_clone_root; |
1381 | ret = 0; | |
1382 | } else { | |
1383 | ret = -ENOENT; | |
1384 | } | |
1385 | ||
1386 | out: | |
1387 | btrfs_free_path(tmp_path); | |
35075bb0 | 1388 | kfree(backref_ctx); |
31db9f7c AB |
1389 | return ret; |
1390 | } | |
1391 | ||
924794c9 | 1392 | static int read_symlink(struct btrfs_root *root, |
31db9f7c AB |
1393 | u64 ino, |
1394 | struct fs_path *dest) | |
1395 | { | |
1396 | int ret; | |
1397 | struct btrfs_path *path; | |
1398 | struct btrfs_key key; | |
1399 | struct btrfs_file_extent_item *ei; | |
1400 | u8 type; | |
1401 | u8 compression; | |
1402 | unsigned long off; | |
1403 | int len; | |
1404 | ||
1405 | path = alloc_path_for_send(); | |
1406 | if (!path) | |
1407 | return -ENOMEM; | |
1408 | ||
1409 | key.objectid = ino; | |
1410 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1411 | key.offset = 0; | |
1412 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1413 | if (ret < 0) | |
1414 | goto out; | |
1415 | BUG_ON(ret); | |
1416 | ||
1417 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1418 | struct btrfs_file_extent_item); | |
1419 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
1420 | compression = btrfs_file_extent_compression(path->nodes[0], ei); | |
1421 | BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); | |
1422 | BUG_ON(compression); | |
1423 | ||
1424 | off = btrfs_file_extent_inline_start(ei); | |
514ac8ad | 1425 | len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei); |
31db9f7c AB |
1426 | |
1427 | ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); | |
31db9f7c AB |
1428 | |
1429 | out: | |
1430 | btrfs_free_path(path); | |
1431 | return ret; | |
1432 | } | |
1433 | ||
1434 | /* | |
1435 | * Helper function to generate a file name that is unique in the root of | |
1436 | * send_root and parent_root. This is used to generate names for orphan inodes. | |
1437 | */ | |
1438 | static int gen_unique_name(struct send_ctx *sctx, | |
1439 | u64 ino, u64 gen, | |
1440 | struct fs_path *dest) | |
1441 | { | |
1442 | int ret = 0; | |
1443 | struct btrfs_path *path; | |
1444 | struct btrfs_dir_item *di; | |
1445 | char tmp[64]; | |
1446 | int len; | |
1447 | u64 idx = 0; | |
1448 | ||
1449 | path = alloc_path_for_send(); | |
1450 | if (!path) | |
1451 | return -ENOMEM; | |
1452 | ||
1453 | while (1) { | |
f74b86d8 | 1454 | len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", |
31db9f7c | 1455 | ino, gen, idx); |
64792f25 | 1456 | ASSERT(len < sizeof(tmp)); |
31db9f7c AB |
1457 | |
1458 | di = btrfs_lookup_dir_item(NULL, sctx->send_root, | |
1459 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1460 | tmp, strlen(tmp), 0); | |
1461 | btrfs_release_path(path); | |
1462 | if (IS_ERR(di)) { | |
1463 | ret = PTR_ERR(di); | |
1464 | goto out; | |
1465 | } | |
1466 | if (di) { | |
1467 | /* not unique, try again */ | |
1468 | idx++; | |
1469 | continue; | |
1470 | } | |
1471 | ||
1472 | if (!sctx->parent_root) { | |
1473 | /* unique */ | |
1474 | ret = 0; | |
1475 | break; | |
1476 | } | |
1477 | ||
1478 | di = btrfs_lookup_dir_item(NULL, sctx->parent_root, | |
1479 | path, BTRFS_FIRST_FREE_OBJECTID, | |
1480 | tmp, strlen(tmp), 0); | |
1481 | btrfs_release_path(path); | |
1482 | if (IS_ERR(di)) { | |
1483 | ret = PTR_ERR(di); | |
1484 | goto out; | |
1485 | } | |
1486 | if (di) { | |
1487 | /* not unique, try again */ | |
1488 | idx++; | |
1489 | continue; | |
1490 | } | |
1491 | /* unique */ | |
1492 | break; | |
1493 | } | |
1494 | ||
1495 | ret = fs_path_add(dest, tmp, strlen(tmp)); | |
1496 | ||
1497 | out: | |
1498 | btrfs_free_path(path); | |
1499 | return ret; | |
1500 | } | |
1501 | ||
1502 | enum inode_state { | |
1503 | inode_state_no_change, | |
1504 | inode_state_will_create, | |
1505 | inode_state_did_create, | |
1506 | inode_state_will_delete, | |
1507 | inode_state_did_delete, | |
1508 | }; | |
1509 | ||
1510 | static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) | |
1511 | { | |
1512 | int ret; | |
1513 | int left_ret; | |
1514 | int right_ret; | |
1515 | u64 left_gen; | |
1516 | u64 right_gen; | |
1517 | ||
1518 | ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, | |
85a7b33b | 1519 | NULL, NULL); |
31db9f7c AB |
1520 | if (ret < 0 && ret != -ENOENT) |
1521 | goto out; | |
1522 | left_ret = ret; | |
1523 | ||
1524 | if (!sctx->parent_root) { | |
1525 | right_ret = -ENOENT; | |
1526 | } else { | |
1527 | ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, | |
85a7b33b | 1528 | NULL, NULL, NULL, NULL); |
31db9f7c AB |
1529 | if (ret < 0 && ret != -ENOENT) |
1530 | goto out; | |
1531 | right_ret = ret; | |
1532 | } | |
1533 | ||
1534 | if (!left_ret && !right_ret) { | |
e938c8ad | 1535 | if (left_gen == gen && right_gen == gen) { |
31db9f7c | 1536 | ret = inode_state_no_change; |
e938c8ad | 1537 | } else if (left_gen == gen) { |
31db9f7c AB |
1538 | if (ino < sctx->send_progress) |
1539 | ret = inode_state_did_create; | |
1540 | else | |
1541 | ret = inode_state_will_create; | |
1542 | } else if (right_gen == gen) { | |
1543 | if (ino < sctx->send_progress) | |
1544 | ret = inode_state_did_delete; | |
1545 | else | |
1546 | ret = inode_state_will_delete; | |
1547 | } else { | |
1548 | ret = -ENOENT; | |
1549 | } | |
1550 | } else if (!left_ret) { | |
1551 | if (left_gen == gen) { | |
1552 | if (ino < sctx->send_progress) | |
1553 | ret = inode_state_did_create; | |
1554 | else | |
1555 | ret = inode_state_will_create; | |
1556 | } else { | |
1557 | ret = -ENOENT; | |
1558 | } | |
1559 | } else if (!right_ret) { | |
1560 | if (right_gen == gen) { | |
1561 | if (ino < sctx->send_progress) | |
1562 | ret = inode_state_did_delete; | |
1563 | else | |
1564 | ret = inode_state_will_delete; | |
1565 | } else { | |
1566 | ret = -ENOENT; | |
1567 | } | |
1568 | } else { | |
1569 | ret = -ENOENT; | |
1570 | } | |
1571 | ||
1572 | out: | |
1573 | return ret; | |
1574 | } | |
1575 | ||
1576 | static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) | |
1577 | { | |
1578 | int ret; | |
1579 | ||
1580 | ret = get_cur_inode_state(sctx, ino, gen); | |
1581 | if (ret < 0) | |
1582 | goto out; | |
1583 | ||
1584 | if (ret == inode_state_no_change || | |
1585 | ret == inode_state_did_create || | |
1586 | ret == inode_state_will_delete) | |
1587 | ret = 1; | |
1588 | else | |
1589 | ret = 0; | |
1590 | ||
1591 | out: | |
1592 | return ret; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * Helper function to lookup a dir item in a dir. | |
1597 | */ | |
1598 | static int lookup_dir_item_inode(struct btrfs_root *root, | |
1599 | u64 dir, const char *name, int name_len, | |
1600 | u64 *found_inode, | |
1601 | u8 *found_type) | |
1602 | { | |
1603 | int ret = 0; | |
1604 | struct btrfs_dir_item *di; | |
1605 | struct btrfs_key key; | |
1606 | struct btrfs_path *path; | |
1607 | ||
1608 | path = alloc_path_for_send(); | |
1609 | if (!path) | |
1610 | return -ENOMEM; | |
1611 | ||
1612 | di = btrfs_lookup_dir_item(NULL, root, path, | |
1613 | dir, name, name_len, 0); | |
1614 | if (!di) { | |
1615 | ret = -ENOENT; | |
1616 | goto out; | |
1617 | } | |
1618 | if (IS_ERR(di)) { | |
1619 | ret = PTR_ERR(di); | |
1620 | goto out; | |
1621 | } | |
1622 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
1623 | *found_inode = key.objectid; | |
1624 | *found_type = btrfs_dir_type(path->nodes[0], di); | |
1625 | ||
1626 | out: | |
1627 | btrfs_free_path(path); | |
1628 | return ret; | |
1629 | } | |
1630 | ||
766702ef AB |
1631 | /* |
1632 | * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, | |
1633 | * generation of the parent dir and the name of the dir entry. | |
1634 | */ | |
924794c9 | 1635 | static int get_first_ref(struct btrfs_root *root, u64 ino, |
31db9f7c AB |
1636 | u64 *dir, u64 *dir_gen, struct fs_path *name) |
1637 | { | |
1638 | int ret; | |
1639 | struct btrfs_key key; | |
1640 | struct btrfs_key found_key; | |
1641 | struct btrfs_path *path; | |
31db9f7c | 1642 | int len; |
96b5bd77 | 1643 | u64 parent_dir; |
31db9f7c AB |
1644 | |
1645 | path = alloc_path_for_send(); | |
1646 | if (!path) | |
1647 | return -ENOMEM; | |
1648 | ||
1649 | key.objectid = ino; | |
1650 | key.type = BTRFS_INODE_REF_KEY; | |
1651 | key.offset = 0; | |
1652 | ||
1653 | ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); | |
1654 | if (ret < 0) | |
1655 | goto out; | |
1656 | if (!ret) | |
1657 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1658 | path->slots[0]); | |
96b5bd77 JS |
1659 | if (ret || found_key.objectid != ino || |
1660 | (found_key.type != BTRFS_INODE_REF_KEY && | |
1661 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1662 | ret = -ENOENT; |
1663 | goto out; | |
1664 | } | |
1665 | ||
96b5bd77 JS |
1666 | if (key.type == BTRFS_INODE_REF_KEY) { |
1667 | struct btrfs_inode_ref *iref; | |
1668 | iref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1669 | struct btrfs_inode_ref); | |
1670 | len = btrfs_inode_ref_name_len(path->nodes[0], iref); | |
1671 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1672 | (unsigned long)(iref + 1), | |
1673 | len); | |
1674 | parent_dir = found_key.offset; | |
1675 | } else { | |
1676 | struct btrfs_inode_extref *extref; | |
1677 | extref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1678 | struct btrfs_inode_extref); | |
1679 | len = btrfs_inode_extref_name_len(path->nodes[0], extref); | |
1680 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
1681 | (unsigned long)&extref->name, len); | |
1682 | parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); | |
1683 | } | |
31db9f7c AB |
1684 | if (ret < 0) |
1685 | goto out; | |
1686 | btrfs_release_path(path); | |
1687 | ||
96b5bd77 | 1688 | ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL, |
85a7b33b | 1689 | NULL, NULL); |
31db9f7c AB |
1690 | if (ret < 0) |
1691 | goto out; | |
1692 | ||
96b5bd77 | 1693 | *dir = parent_dir; |
31db9f7c AB |
1694 | |
1695 | out: | |
1696 | btrfs_free_path(path); | |
1697 | return ret; | |
1698 | } | |
1699 | ||
924794c9 | 1700 | static int is_first_ref(struct btrfs_root *root, |
31db9f7c AB |
1701 | u64 ino, u64 dir, |
1702 | const char *name, int name_len) | |
1703 | { | |
1704 | int ret; | |
1705 | struct fs_path *tmp_name; | |
1706 | u64 tmp_dir; | |
1707 | u64 tmp_dir_gen; | |
1708 | ||
924794c9 | 1709 | tmp_name = fs_path_alloc(); |
31db9f7c AB |
1710 | if (!tmp_name) |
1711 | return -ENOMEM; | |
1712 | ||
924794c9 | 1713 | ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name); |
31db9f7c AB |
1714 | if (ret < 0) |
1715 | goto out; | |
1716 | ||
b9291aff | 1717 | if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { |
31db9f7c AB |
1718 | ret = 0; |
1719 | goto out; | |
1720 | } | |
1721 | ||
e938c8ad | 1722 | ret = !memcmp(tmp_name->start, name, name_len); |
31db9f7c AB |
1723 | |
1724 | out: | |
924794c9 | 1725 | fs_path_free(tmp_name); |
31db9f7c AB |
1726 | return ret; |
1727 | } | |
1728 | ||
766702ef AB |
1729 | /* |
1730 | * Used by process_recorded_refs to determine if a new ref would overwrite an | |
1731 | * already existing ref. In case it detects an overwrite, it returns the | |
1732 | * inode/gen in who_ino/who_gen. | |
1733 | * When an overwrite is detected, process_recorded_refs does proper orphanizing | |
1734 | * to make sure later references to the overwritten inode are possible. | |
1735 | * Orphanizing is however only required for the first ref of an inode. | |
1736 | * process_recorded_refs does an additional is_first_ref check to see if | |
1737 | * orphanizing is really required. | |
1738 | */ | |
31db9f7c AB |
1739 | static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
1740 | const char *name, int name_len, | |
1741 | u64 *who_ino, u64 *who_gen) | |
1742 | { | |
1743 | int ret = 0; | |
ebdad913 | 1744 | u64 gen; |
31db9f7c AB |
1745 | u64 other_inode = 0; |
1746 | u8 other_type = 0; | |
1747 | ||
1748 | if (!sctx->parent_root) | |
1749 | goto out; | |
1750 | ||
1751 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1752 | if (ret <= 0) | |
1753 | goto out; | |
1754 | ||
ebdad913 JB |
1755 | /* |
1756 | * If we have a parent root we need to verify that the parent dir was | |
1757 | * not delted and then re-created, if it was then we have no overwrite | |
1758 | * and we can just unlink this entry. | |
1759 | */ | |
1760 | if (sctx->parent_root) { | |
1761 | ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, | |
1762 | NULL, NULL, NULL); | |
1763 | if (ret < 0 && ret != -ENOENT) | |
1764 | goto out; | |
1765 | if (ret) { | |
1766 | ret = 0; | |
1767 | goto out; | |
1768 | } | |
1769 | if (gen != dir_gen) | |
1770 | goto out; | |
1771 | } | |
1772 | ||
31db9f7c AB |
1773 | ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, |
1774 | &other_inode, &other_type); | |
1775 | if (ret < 0 && ret != -ENOENT) | |
1776 | goto out; | |
1777 | if (ret) { | |
1778 | ret = 0; | |
1779 | goto out; | |
1780 | } | |
1781 | ||
766702ef AB |
1782 | /* |
1783 | * Check if the overwritten ref was already processed. If yes, the ref | |
1784 | * was already unlinked/moved, so we can safely assume that we will not | |
1785 | * overwrite anything at this point in time. | |
1786 | */ | |
31db9f7c AB |
1787 | if (other_inode > sctx->send_progress) { |
1788 | ret = get_inode_info(sctx->parent_root, other_inode, NULL, | |
85a7b33b | 1789 | who_gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
1790 | if (ret < 0) |
1791 | goto out; | |
1792 | ||
1793 | ret = 1; | |
1794 | *who_ino = other_inode; | |
1795 | } else { | |
1796 | ret = 0; | |
1797 | } | |
1798 | ||
1799 | out: | |
1800 | return ret; | |
1801 | } | |
1802 | ||
766702ef AB |
1803 | /* |
1804 | * Checks if the ref was overwritten by an already processed inode. This is | |
1805 | * used by __get_cur_name_and_parent to find out if the ref was orphanized and | |
1806 | * thus the orphan name needs be used. | |
1807 | * process_recorded_refs also uses it to avoid unlinking of refs that were | |
1808 | * overwritten. | |
1809 | */ | |
31db9f7c AB |
1810 | static int did_overwrite_ref(struct send_ctx *sctx, |
1811 | u64 dir, u64 dir_gen, | |
1812 | u64 ino, u64 ino_gen, | |
1813 | const char *name, int name_len) | |
1814 | { | |
1815 | int ret = 0; | |
1816 | u64 gen; | |
1817 | u64 ow_inode; | |
1818 | u8 other_type; | |
1819 | ||
1820 | if (!sctx->parent_root) | |
1821 | goto out; | |
1822 | ||
1823 | ret = is_inode_existent(sctx, dir, dir_gen); | |
1824 | if (ret <= 0) | |
1825 | goto out; | |
1826 | ||
1827 | /* check if the ref was overwritten by another ref */ | |
1828 | ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, | |
1829 | &ow_inode, &other_type); | |
1830 | if (ret < 0 && ret != -ENOENT) | |
1831 | goto out; | |
1832 | if (ret) { | |
1833 | /* was never and will never be overwritten */ | |
1834 | ret = 0; | |
1835 | goto out; | |
1836 | } | |
1837 | ||
1838 | ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, | |
85a7b33b | 1839 | NULL, NULL); |
31db9f7c AB |
1840 | if (ret < 0) |
1841 | goto out; | |
1842 | ||
1843 | if (ow_inode == ino && gen == ino_gen) { | |
1844 | ret = 0; | |
1845 | goto out; | |
1846 | } | |
1847 | ||
1848 | /* we know that it is or will be overwritten. check this now */ | |
1849 | if (ow_inode < sctx->send_progress) | |
1850 | ret = 1; | |
1851 | else | |
1852 | ret = 0; | |
1853 | ||
1854 | out: | |
1855 | return ret; | |
1856 | } | |
1857 | ||
766702ef AB |
1858 | /* |
1859 | * Same as did_overwrite_ref, but also checks if it is the first ref of an inode | |
1860 | * that got overwritten. This is used by process_recorded_refs to determine | |
1861 | * if it has to use the path as returned by get_cur_path or the orphan name. | |
1862 | */ | |
31db9f7c AB |
1863 | static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) |
1864 | { | |
1865 | int ret = 0; | |
1866 | struct fs_path *name = NULL; | |
1867 | u64 dir; | |
1868 | u64 dir_gen; | |
1869 | ||
1870 | if (!sctx->parent_root) | |
1871 | goto out; | |
1872 | ||
924794c9 | 1873 | name = fs_path_alloc(); |
31db9f7c AB |
1874 | if (!name) |
1875 | return -ENOMEM; | |
1876 | ||
924794c9 | 1877 | ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); |
31db9f7c AB |
1878 | if (ret < 0) |
1879 | goto out; | |
1880 | ||
1881 | ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, | |
1882 | name->start, fs_path_len(name)); | |
31db9f7c AB |
1883 | |
1884 | out: | |
924794c9 | 1885 | fs_path_free(name); |
31db9f7c AB |
1886 | return ret; |
1887 | } | |
1888 | ||
766702ef AB |
1889 | /* |
1890 | * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, | |
1891 | * so we need to do some special handling in case we have clashes. This function | |
1892 | * takes care of this with the help of name_cache_entry::radix_list. | |
5dc67d0b | 1893 | * In case of error, nce is kfreed. |
766702ef | 1894 | */ |
31db9f7c AB |
1895 | static int name_cache_insert(struct send_ctx *sctx, |
1896 | struct name_cache_entry *nce) | |
1897 | { | |
1898 | int ret = 0; | |
7e0926fe AB |
1899 | struct list_head *nce_head; |
1900 | ||
1901 | nce_head = radix_tree_lookup(&sctx->name_cache, | |
1902 | (unsigned long)nce->ino); | |
1903 | if (!nce_head) { | |
1904 | nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS); | |
cfa7a9cc TI |
1905 | if (!nce_head) { |
1906 | kfree(nce); | |
31db9f7c | 1907 | return -ENOMEM; |
cfa7a9cc | 1908 | } |
7e0926fe | 1909 | INIT_LIST_HEAD(nce_head); |
31db9f7c | 1910 | |
7e0926fe | 1911 | ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); |
5dc67d0b AB |
1912 | if (ret < 0) { |
1913 | kfree(nce_head); | |
1914 | kfree(nce); | |
31db9f7c | 1915 | return ret; |
5dc67d0b | 1916 | } |
31db9f7c | 1917 | } |
7e0926fe | 1918 | list_add_tail(&nce->radix_list, nce_head); |
31db9f7c AB |
1919 | list_add_tail(&nce->list, &sctx->name_cache_list); |
1920 | sctx->name_cache_size++; | |
1921 | ||
1922 | return ret; | |
1923 | } | |
1924 | ||
1925 | static void name_cache_delete(struct send_ctx *sctx, | |
1926 | struct name_cache_entry *nce) | |
1927 | { | |
7e0926fe | 1928 | struct list_head *nce_head; |
31db9f7c | 1929 | |
7e0926fe AB |
1930 | nce_head = radix_tree_lookup(&sctx->name_cache, |
1931 | (unsigned long)nce->ino); | |
57fb8910 DS |
1932 | if (!nce_head) { |
1933 | btrfs_err(sctx->send_root->fs_info, | |
1934 | "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", | |
1935 | nce->ino, sctx->name_cache_size); | |
1936 | } | |
31db9f7c | 1937 | |
7e0926fe | 1938 | list_del(&nce->radix_list); |
31db9f7c | 1939 | list_del(&nce->list); |
31db9f7c | 1940 | sctx->name_cache_size--; |
7e0926fe | 1941 | |
57fb8910 DS |
1942 | /* |
1943 | * We may not get to the final release of nce_head if the lookup fails | |
1944 | */ | |
1945 | if (nce_head && list_empty(nce_head)) { | |
7e0926fe AB |
1946 | radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); |
1947 | kfree(nce_head); | |
1948 | } | |
31db9f7c AB |
1949 | } |
1950 | ||
1951 | static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, | |
1952 | u64 ino, u64 gen) | |
1953 | { | |
7e0926fe AB |
1954 | struct list_head *nce_head; |
1955 | struct name_cache_entry *cur; | |
31db9f7c | 1956 | |
7e0926fe AB |
1957 | nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); |
1958 | if (!nce_head) | |
31db9f7c AB |
1959 | return NULL; |
1960 | ||
7e0926fe AB |
1961 | list_for_each_entry(cur, nce_head, radix_list) { |
1962 | if (cur->ino == ino && cur->gen == gen) | |
1963 | return cur; | |
1964 | } | |
31db9f7c AB |
1965 | return NULL; |
1966 | } | |
1967 | ||
766702ef AB |
1968 | /* |
1969 | * Removes the entry from the list and adds it back to the end. This marks the | |
1970 | * entry as recently used so that name_cache_clean_unused does not remove it. | |
1971 | */ | |
31db9f7c AB |
1972 | static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) |
1973 | { | |
1974 | list_del(&nce->list); | |
1975 | list_add_tail(&nce->list, &sctx->name_cache_list); | |
1976 | } | |
1977 | ||
766702ef AB |
1978 | /* |
1979 | * Remove some entries from the beginning of name_cache_list. | |
1980 | */ | |
31db9f7c AB |
1981 | static void name_cache_clean_unused(struct send_ctx *sctx) |
1982 | { | |
1983 | struct name_cache_entry *nce; | |
1984 | ||
1985 | if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) | |
1986 | return; | |
1987 | ||
1988 | while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { | |
1989 | nce = list_entry(sctx->name_cache_list.next, | |
1990 | struct name_cache_entry, list); | |
1991 | name_cache_delete(sctx, nce); | |
1992 | kfree(nce); | |
1993 | } | |
1994 | } | |
1995 | ||
1996 | static void name_cache_free(struct send_ctx *sctx) | |
1997 | { | |
1998 | struct name_cache_entry *nce; | |
31db9f7c | 1999 | |
e938c8ad AB |
2000 | while (!list_empty(&sctx->name_cache_list)) { |
2001 | nce = list_entry(sctx->name_cache_list.next, | |
2002 | struct name_cache_entry, list); | |
31db9f7c | 2003 | name_cache_delete(sctx, nce); |
17589bd9 | 2004 | kfree(nce); |
31db9f7c AB |
2005 | } |
2006 | } | |
2007 | ||
766702ef AB |
2008 | /* |
2009 | * Used by get_cur_path for each ref up to the root. | |
2010 | * Returns 0 if it succeeded. | |
2011 | * Returns 1 if the inode is not existent or got overwritten. In that case, the | |
2012 | * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 | |
2013 | * is returned, parent_ino/parent_gen are not guaranteed to be valid. | |
2014 | * Returns <0 in case of error. | |
2015 | */ | |
31db9f7c AB |
2016 | static int __get_cur_name_and_parent(struct send_ctx *sctx, |
2017 | u64 ino, u64 gen, | |
2018 | u64 *parent_ino, | |
2019 | u64 *parent_gen, | |
2020 | struct fs_path *dest) | |
2021 | { | |
2022 | int ret; | |
2023 | int nce_ret; | |
2024 | struct btrfs_path *path = NULL; | |
2025 | struct name_cache_entry *nce = NULL; | |
2026 | ||
766702ef AB |
2027 | /* |
2028 | * First check if we already did a call to this function with the same | |
2029 | * ino/gen. If yes, check if the cache entry is still up-to-date. If yes | |
2030 | * return the cached result. | |
2031 | */ | |
31db9f7c AB |
2032 | nce = name_cache_search(sctx, ino, gen); |
2033 | if (nce) { | |
2034 | if (ino < sctx->send_progress && nce->need_later_update) { | |
2035 | name_cache_delete(sctx, nce); | |
2036 | kfree(nce); | |
2037 | nce = NULL; | |
2038 | } else { | |
2039 | name_cache_used(sctx, nce); | |
2040 | *parent_ino = nce->parent_ino; | |
2041 | *parent_gen = nce->parent_gen; | |
2042 | ret = fs_path_add(dest, nce->name, nce->name_len); | |
2043 | if (ret < 0) | |
2044 | goto out; | |
2045 | ret = nce->ret; | |
2046 | goto out; | |
2047 | } | |
2048 | } | |
2049 | ||
2050 | path = alloc_path_for_send(); | |
2051 | if (!path) | |
2052 | return -ENOMEM; | |
2053 | ||
766702ef AB |
2054 | /* |
2055 | * If the inode is not existent yet, add the orphan name and return 1. | |
2056 | * This should only happen for the parent dir that we determine in | |
2057 | * __record_new_ref | |
2058 | */ | |
31db9f7c AB |
2059 | ret = is_inode_existent(sctx, ino, gen); |
2060 | if (ret < 0) | |
2061 | goto out; | |
2062 | ||
2063 | if (!ret) { | |
2064 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2065 | if (ret < 0) | |
2066 | goto out; | |
2067 | ret = 1; | |
2068 | goto out_cache; | |
2069 | } | |
2070 | ||
766702ef AB |
2071 | /* |
2072 | * Depending on whether the inode was already processed or not, use | |
2073 | * send_root or parent_root for ref lookup. | |
2074 | */ | |
bf0d1f44 | 2075 | if (ino < sctx->send_progress) |
924794c9 TI |
2076 | ret = get_first_ref(sctx->send_root, ino, |
2077 | parent_ino, parent_gen, dest); | |
31db9f7c | 2078 | else |
924794c9 TI |
2079 | ret = get_first_ref(sctx->parent_root, ino, |
2080 | parent_ino, parent_gen, dest); | |
31db9f7c AB |
2081 | if (ret < 0) |
2082 | goto out; | |
2083 | ||
766702ef AB |
2084 | /* |
2085 | * Check if the ref was overwritten by an inode's ref that was processed | |
2086 | * earlier. If yes, treat as orphan and return 1. | |
2087 | */ | |
31db9f7c AB |
2088 | ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, |
2089 | dest->start, dest->end - dest->start); | |
2090 | if (ret < 0) | |
2091 | goto out; | |
2092 | if (ret) { | |
2093 | fs_path_reset(dest); | |
2094 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2095 | if (ret < 0) | |
2096 | goto out; | |
2097 | ret = 1; | |
2098 | } | |
2099 | ||
2100 | out_cache: | |
766702ef AB |
2101 | /* |
2102 | * Store the result of the lookup in the name cache. | |
2103 | */ | |
31db9f7c AB |
2104 | nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS); |
2105 | if (!nce) { | |
2106 | ret = -ENOMEM; | |
2107 | goto out; | |
2108 | } | |
2109 | ||
2110 | nce->ino = ino; | |
2111 | nce->gen = gen; | |
2112 | nce->parent_ino = *parent_ino; | |
2113 | nce->parent_gen = *parent_gen; | |
2114 | nce->name_len = fs_path_len(dest); | |
2115 | nce->ret = ret; | |
2116 | strcpy(nce->name, dest->start); | |
31db9f7c AB |
2117 | |
2118 | if (ino < sctx->send_progress) | |
2119 | nce->need_later_update = 0; | |
2120 | else | |
2121 | nce->need_later_update = 1; | |
2122 | ||
2123 | nce_ret = name_cache_insert(sctx, nce); | |
2124 | if (nce_ret < 0) | |
2125 | ret = nce_ret; | |
2126 | name_cache_clean_unused(sctx); | |
2127 | ||
2128 | out: | |
2129 | btrfs_free_path(path); | |
2130 | return ret; | |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * Magic happens here. This function returns the first ref to an inode as it | |
2135 | * would look like while receiving the stream at this point in time. | |
2136 | * We walk the path up to the root. For every inode in between, we check if it | |
2137 | * was already processed/sent. If yes, we continue with the parent as found | |
2138 | * in send_root. If not, we continue with the parent as found in parent_root. | |
2139 | * If we encounter an inode that was deleted at this point in time, we use the | |
2140 | * inodes "orphan" name instead of the real name and stop. Same with new inodes | |
2141 | * that were not created yet and overwritten inodes/refs. | |
2142 | * | |
2143 | * When do we have have orphan inodes: | |
2144 | * 1. When an inode is freshly created and thus no valid refs are available yet | |
2145 | * 2. When a directory lost all it's refs (deleted) but still has dir items | |
2146 | * inside which were not processed yet (pending for move/delete). If anyone | |
2147 | * tried to get the path to the dir items, it would get a path inside that | |
2148 | * orphan directory. | |
2149 | * 3. When an inode is moved around or gets new links, it may overwrite the ref | |
2150 | * of an unprocessed inode. If in that case the first ref would be | |
2151 | * overwritten, the overwritten inode gets "orphanized". Later when we | |
2152 | * process this overwritten inode, it is restored at a new place by moving | |
2153 | * the orphan inode. | |
2154 | * | |
2155 | * sctx->send_progress tells this function at which point in time receiving | |
2156 | * would be. | |
2157 | */ | |
2158 | static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, | |
2159 | struct fs_path *dest) | |
2160 | { | |
2161 | int ret = 0; | |
2162 | struct fs_path *name = NULL; | |
2163 | u64 parent_inode = 0; | |
2164 | u64 parent_gen = 0; | |
2165 | int stop = 0; | |
2166 | ||
924794c9 | 2167 | name = fs_path_alloc(); |
31db9f7c AB |
2168 | if (!name) { |
2169 | ret = -ENOMEM; | |
2170 | goto out; | |
2171 | } | |
2172 | ||
2173 | dest->reversed = 1; | |
2174 | fs_path_reset(dest); | |
2175 | ||
2176 | while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { | |
2177 | fs_path_reset(name); | |
2178 | ||
9dc44214 FM |
2179 | if (is_waiting_for_rm(sctx, ino)) { |
2180 | ret = gen_unique_name(sctx, ino, gen, name); | |
2181 | if (ret < 0) | |
2182 | goto out; | |
2183 | ret = fs_path_add_path(dest, name); | |
2184 | break; | |
2185 | } | |
2186 | ||
bf0d1f44 FM |
2187 | if (is_waiting_for_move(sctx, ino)) { |
2188 | ret = get_first_ref(sctx->parent_root, ino, | |
2189 | &parent_inode, &parent_gen, name); | |
2190 | } else { | |
2191 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
2192 | &parent_inode, | |
2193 | &parent_gen, name); | |
2194 | if (ret) | |
2195 | stop = 1; | |
2196 | } | |
2197 | ||
31db9f7c AB |
2198 | if (ret < 0) |
2199 | goto out; | |
9f03740a | 2200 | |
31db9f7c AB |
2201 | ret = fs_path_add_path(dest, name); |
2202 | if (ret < 0) | |
2203 | goto out; | |
2204 | ||
2205 | ino = parent_inode; | |
2206 | gen = parent_gen; | |
2207 | } | |
2208 | ||
2209 | out: | |
924794c9 | 2210 | fs_path_free(name); |
31db9f7c AB |
2211 | if (!ret) |
2212 | fs_path_unreverse(dest); | |
2213 | return ret; | |
2214 | } | |
2215 | ||
31db9f7c AB |
2216 | /* |
2217 | * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace | |
2218 | */ | |
2219 | static int send_subvol_begin(struct send_ctx *sctx) | |
2220 | { | |
2221 | int ret; | |
2222 | struct btrfs_root *send_root = sctx->send_root; | |
2223 | struct btrfs_root *parent_root = sctx->parent_root; | |
2224 | struct btrfs_path *path; | |
2225 | struct btrfs_key key; | |
2226 | struct btrfs_root_ref *ref; | |
2227 | struct extent_buffer *leaf; | |
2228 | char *name = NULL; | |
2229 | int namelen; | |
2230 | ||
ffcfaf81 | 2231 | path = btrfs_alloc_path(); |
31db9f7c AB |
2232 | if (!path) |
2233 | return -ENOMEM; | |
2234 | ||
2235 | name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS); | |
2236 | if (!name) { | |
2237 | btrfs_free_path(path); | |
2238 | return -ENOMEM; | |
2239 | } | |
2240 | ||
2241 | key.objectid = send_root->objectid; | |
2242 | key.type = BTRFS_ROOT_BACKREF_KEY; | |
2243 | key.offset = 0; | |
2244 | ||
2245 | ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, | |
2246 | &key, path, 1, 0); | |
2247 | if (ret < 0) | |
2248 | goto out; | |
2249 | if (ret) { | |
2250 | ret = -ENOENT; | |
2251 | goto out; | |
2252 | } | |
2253 | ||
2254 | leaf = path->nodes[0]; | |
2255 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2256 | if (key.type != BTRFS_ROOT_BACKREF_KEY || | |
2257 | key.objectid != send_root->objectid) { | |
2258 | ret = -ENOENT; | |
2259 | goto out; | |
2260 | } | |
2261 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
2262 | namelen = btrfs_root_ref_name_len(leaf, ref); | |
2263 | read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); | |
2264 | btrfs_release_path(path); | |
2265 | ||
31db9f7c AB |
2266 | if (parent_root) { |
2267 | ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); | |
2268 | if (ret < 0) | |
2269 | goto out; | |
2270 | } else { | |
2271 | ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); | |
2272 | if (ret < 0) | |
2273 | goto out; | |
2274 | } | |
2275 | ||
2276 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); | |
2277 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2278 | sctx->send_root->root_item.uuid); | |
2279 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, | |
5a0f4e2c | 2280 | le64_to_cpu(sctx->send_root->root_item.ctransid)); |
31db9f7c AB |
2281 | if (parent_root) { |
2282 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2283 | sctx->parent_root->root_item.uuid); | |
2284 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, | |
5a0f4e2c | 2285 | le64_to_cpu(sctx->parent_root->root_item.ctransid)); |
31db9f7c AB |
2286 | } |
2287 | ||
2288 | ret = send_cmd(sctx); | |
2289 | ||
2290 | tlv_put_failure: | |
2291 | out: | |
2292 | btrfs_free_path(path); | |
2293 | kfree(name); | |
2294 | return ret; | |
2295 | } | |
2296 | ||
2297 | static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) | |
2298 | { | |
2299 | int ret = 0; | |
2300 | struct fs_path *p; | |
2301 | ||
2302 | verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size); | |
2303 | ||
924794c9 | 2304 | p = fs_path_alloc(); |
31db9f7c AB |
2305 | if (!p) |
2306 | return -ENOMEM; | |
2307 | ||
2308 | ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); | |
2309 | if (ret < 0) | |
2310 | goto out; | |
2311 | ||
2312 | ret = get_cur_path(sctx, ino, gen, p); | |
2313 | if (ret < 0) | |
2314 | goto out; | |
2315 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2316 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); | |
2317 | ||
2318 | ret = send_cmd(sctx); | |
2319 | ||
2320 | tlv_put_failure: | |
2321 | out: | |
924794c9 | 2322 | fs_path_free(p); |
31db9f7c AB |
2323 | return ret; |
2324 | } | |
2325 | ||
2326 | static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) | |
2327 | { | |
2328 | int ret = 0; | |
2329 | struct fs_path *p; | |
2330 | ||
2331 | verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode); | |
2332 | ||
924794c9 | 2333 | p = fs_path_alloc(); |
31db9f7c AB |
2334 | if (!p) |
2335 | return -ENOMEM; | |
2336 | ||
2337 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); | |
2338 | if (ret < 0) | |
2339 | goto out; | |
2340 | ||
2341 | ret = get_cur_path(sctx, ino, gen, p); | |
2342 | if (ret < 0) | |
2343 | goto out; | |
2344 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2345 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); | |
2346 | ||
2347 | ret = send_cmd(sctx); | |
2348 | ||
2349 | tlv_put_failure: | |
2350 | out: | |
924794c9 | 2351 | fs_path_free(p); |
31db9f7c AB |
2352 | return ret; |
2353 | } | |
2354 | ||
2355 | static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) | |
2356 | { | |
2357 | int ret = 0; | |
2358 | struct fs_path *p; | |
2359 | ||
2360 | verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid); | |
2361 | ||
924794c9 | 2362 | p = fs_path_alloc(); |
31db9f7c AB |
2363 | if (!p) |
2364 | return -ENOMEM; | |
2365 | ||
2366 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); | |
2367 | if (ret < 0) | |
2368 | goto out; | |
2369 | ||
2370 | ret = get_cur_path(sctx, ino, gen, p); | |
2371 | if (ret < 0) | |
2372 | goto out; | |
2373 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2374 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); | |
2375 | TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); | |
2376 | ||
2377 | ret = send_cmd(sctx); | |
2378 | ||
2379 | tlv_put_failure: | |
2380 | out: | |
924794c9 | 2381 | fs_path_free(p); |
31db9f7c AB |
2382 | return ret; |
2383 | } | |
2384 | ||
2385 | static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) | |
2386 | { | |
2387 | int ret = 0; | |
2388 | struct fs_path *p = NULL; | |
2389 | struct btrfs_inode_item *ii; | |
2390 | struct btrfs_path *path = NULL; | |
2391 | struct extent_buffer *eb; | |
2392 | struct btrfs_key key; | |
2393 | int slot; | |
2394 | ||
2395 | verbose_printk("btrfs: send_utimes %llu\n", ino); | |
2396 | ||
924794c9 | 2397 | p = fs_path_alloc(); |
31db9f7c AB |
2398 | if (!p) |
2399 | return -ENOMEM; | |
2400 | ||
2401 | path = alloc_path_for_send(); | |
2402 | if (!path) { | |
2403 | ret = -ENOMEM; | |
2404 | goto out; | |
2405 | } | |
2406 | ||
2407 | key.objectid = ino; | |
2408 | key.type = BTRFS_INODE_ITEM_KEY; | |
2409 | key.offset = 0; | |
2410 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); | |
2411 | if (ret < 0) | |
2412 | goto out; | |
2413 | ||
2414 | eb = path->nodes[0]; | |
2415 | slot = path->slots[0]; | |
2416 | ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
2417 | ||
2418 | ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); | |
2419 | if (ret < 0) | |
2420 | goto out; | |
2421 | ||
2422 | ret = get_cur_path(sctx, ino, gen, p); | |
2423 | if (ret < 0) | |
2424 | goto out; | |
2425 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2426 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, | |
2427 | btrfs_inode_atime(ii)); | |
2428 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, | |
2429 | btrfs_inode_mtime(ii)); | |
2430 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, | |
2431 | btrfs_inode_ctime(ii)); | |
766702ef | 2432 | /* TODO Add otime support when the otime patches get into upstream */ |
31db9f7c AB |
2433 | |
2434 | ret = send_cmd(sctx); | |
2435 | ||
2436 | tlv_put_failure: | |
2437 | out: | |
924794c9 | 2438 | fs_path_free(p); |
31db9f7c AB |
2439 | btrfs_free_path(path); |
2440 | return ret; | |
2441 | } | |
2442 | ||
2443 | /* | |
2444 | * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have | |
2445 | * a valid path yet because we did not process the refs yet. So, the inode | |
2446 | * is created as orphan. | |
2447 | */ | |
1f4692da | 2448 | static int send_create_inode(struct send_ctx *sctx, u64 ino) |
31db9f7c AB |
2449 | { |
2450 | int ret = 0; | |
31db9f7c | 2451 | struct fs_path *p; |
31db9f7c | 2452 | int cmd; |
1f4692da | 2453 | u64 gen; |
31db9f7c | 2454 | u64 mode; |
1f4692da | 2455 | u64 rdev; |
31db9f7c | 2456 | |
1f4692da | 2457 | verbose_printk("btrfs: send_create_inode %llu\n", ino); |
31db9f7c | 2458 | |
924794c9 | 2459 | p = fs_path_alloc(); |
31db9f7c AB |
2460 | if (!p) |
2461 | return -ENOMEM; | |
2462 | ||
644d1940 LB |
2463 | if (ino != sctx->cur_ino) { |
2464 | ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, | |
2465 | NULL, NULL, &rdev); | |
2466 | if (ret < 0) | |
2467 | goto out; | |
2468 | } else { | |
2469 | gen = sctx->cur_inode_gen; | |
2470 | mode = sctx->cur_inode_mode; | |
2471 | rdev = sctx->cur_inode_rdev; | |
2472 | } | |
31db9f7c | 2473 | |
e938c8ad | 2474 | if (S_ISREG(mode)) { |
31db9f7c | 2475 | cmd = BTRFS_SEND_C_MKFILE; |
e938c8ad | 2476 | } else if (S_ISDIR(mode)) { |
31db9f7c | 2477 | cmd = BTRFS_SEND_C_MKDIR; |
e938c8ad | 2478 | } else if (S_ISLNK(mode)) { |
31db9f7c | 2479 | cmd = BTRFS_SEND_C_SYMLINK; |
e938c8ad | 2480 | } else if (S_ISCHR(mode) || S_ISBLK(mode)) { |
31db9f7c | 2481 | cmd = BTRFS_SEND_C_MKNOD; |
e938c8ad | 2482 | } else if (S_ISFIFO(mode)) { |
31db9f7c | 2483 | cmd = BTRFS_SEND_C_MKFIFO; |
e938c8ad | 2484 | } else if (S_ISSOCK(mode)) { |
31db9f7c | 2485 | cmd = BTRFS_SEND_C_MKSOCK; |
e938c8ad | 2486 | } else { |
31db9f7c AB |
2487 | printk(KERN_WARNING "btrfs: unexpected inode type %o", |
2488 | (int)(mode & S_IFMT)); | |
2489 | ret = -ENOTSUPP; | |
2490 | goto out; | |
2491 | } | |
2492 | ||
2493 | ret = begin_cmd(sctx, cmd); | |
2494 | if (ret < 0) | |
2495 | goto out; | |
2496 | ||
1f4692da | 2497 | ret = gen_unique_name(sctx, ino, gen, p); |
31db9f7c AB |
2498 | if (ret < 0) |
2499 | goto out; | |
2500 | ||
2501 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
1f4692da | 2502 | TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); |
31db9f7c AB |
2503 | |
2504 | if (S_ISLNK(mode)) { | |
2505 | fs_path_reset(p); | |
924794c9 | 2506 | ret = read_symlink(sctx->send_root, ino, p); |
31db9f7c AB |
2507 | if (ret < 0) |
2508 | goto out; | |
2509 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); | |
2510 | } else if (S_ISCHR(mode) || S_ISBLK(mode) || | |
2511 | S_ISFIFO(mode) || S_ISSOCK(mode)) { | |
d79e5043 AJ |
2512 | TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); |
2513 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); | |
31db9f7c AB |
2514 | } |
2515 | ||
2516 | ret = send_cmd(sctx); | |
2517 | if (ret < 0) | |
2518 | goto out; | |
2519 | ||
2520 | ||
2521 | tlv_put_failure: | |
2522 | out: | |
924794c9 | 2523 | fs_path_free(p); |
31db9f7c AB |
2524 | return ret; |
2525 | } | |
2526 | ||
1f4692da AB |
2527 | /* |
2528 | * We need some special handling for inodes that get processed before the parent | |
2529 | * directory got created. See process_recorded_refs for details. | |
2530 | * This function does the check if we already created the dir out of order. | |
2531 | */ | |
2532 | static int did_create_dir(struct send_ctx *sctx, u64 dir) | |
2533 | { | |
2534 | int ret = 0; | |
2535 | struct btrfs_path *path = NULL; | |
2536 | struct btrfs_key key; | |
2537 | struct btrfs_key found_key; | |
2538 | struct btrfs_key di_key; | |
2539 | struct extent_buffer *eb; | |
2540 | struct btrfs_dir_item *di; | |
2541 | int slot; | |
2542 | ||
2543 | path = alloc_path_for_send(); | |
2544 | if (!path) { | |
2545 | ret = -ENOMEM; | |
2546 | goto out; | |
2547 | } | |
2548 | ||
2549 | key.objectid = dir; | |
2550 | key.type = BTRFS_DIR_INDEX_KEY; | |
2551 | key.offset = 0; | |
dff6d0ad FDBM |
2552 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); |
2553 | if (ret < 0) | |
2554 | goto out; | |
2555 | ||
1f4692da | 2556 | while (1) { |
dff6d0ad FDBM |
2557 | eb = path->nodes[0]; |
2558 | slot = path->slots[0]; | |
2559 | if (slot >= btrfs_header_nritems(eb)) { | |
2560 | ret = btrfs_next_leaf(sctx->send_root, path); | |
2561 | if (ret < 0) { | |
2562 | goto out; | |
2563 | } else if (ret > 0) { | |
2564 | ret = 0; | |
2565 | break; | |
2566 | } | |
2567 | continue; | |
1f4692da | 2568 | } |
dff6d0ad FDBM |
2569 | |
2570 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
2571 | if (found_key.objectid != key.objectid || | |
1f4692da AB |
2572 | found_key.type != key.type) { |
2573 | ret = 0; | |
2574 | goto out; | |
2575 | } | |
2576 | ||
2577 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); | |
2578 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); | |
2579 | ||
a0525414 JB |
2580 | if (di_key.type != BTRFS_ROOT_ITEM_KEY && |
2581 | di_key.objectid < sctx->send_progress) { | |
1f4692da AB |
2582 | ret = 1; |
2583 | goto out; | |
2584 | } | |
2585 | ||
dff6d0ad | 2586 | path->slots[0]++; |
1f4692da AB |
2587 | } |
2588 | ||
2589 | out: | |
2590 | btrfs_free_path(path); | |
2591 | return ret; | |
2592 | } | |
2593 | ||
2594 | /* | |
2595 | * Only creates the inode if it is: | |
2596 | * 1. Not a directory | |
2597 | * 2. Or a directory which was not created already due to out of order | |
2598 | * directories. See did_create_dir and process_recorded_refs for details. | |
2599 | */ | |
2600 | static int send_create_inode_if_needed(struct send_ctx *sctx) | |
2601 | { | |
2602 | int ret; | |
2603 | ||
2604 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
2605 | ret = did_create_dir(sctx, sctx->cur_ino); | |
2606 | if (ret < 0) | |
2607 | goto out; | |
2608 | if (ret) { | |
2609 | ret = 0; | |
2610 | goto out; | |
2611 | } | |
2612 | } | |
2613 | ||
2614 | ret = send_create_inode(sctx, sctx->cur_ino); | |
2615 | if (ret < 0) | |
2616 | goto out; | |
2617 | ||
2618 | out: | |
2619 | return ret; | |
2620 | } | |
2621 | ||
31db9f7c AB |
2622 | struct recorded_ref { |
2623 | struct list_head list; | |
2624 | char *dir_path; | |
2625 | char *name; | |
2626 | struct fs_path *full_path; | |
2627 | u64 dir; | |
2628 | u64 dir_gen; | |
2629 | int dir_path_len; | |
2630 | int name_len; | |
2631 | }; | |
2632 | ||
2633 | /* | |
2634 | * We need to process new refs before deleted refs, but compare_tree gives us | |
2635 | * everything mixed. So we first record all refs and later process them. | |
2636 | * This function is a helper to record one ref. | |
2637 | */ | |
a4d96d62 | 2638 | static int __record_ref(struct list_head *head, u64 dir, |
31db9f7c AB |
2639 | u64 dir_gen, struct fs_path *path) |
2640 | { | |
2641 | struct recorded_ref *ref; | |
31db9f7c AB |
2642 | |
2643 | ref = kmalloc(sizeof(*ref), GFP_NOFS); | |
2644 | if (!ref) | |
2645 | return -ENOMEM; | |
2646 | ||
2647 | ref->dir = dir; | |
2648 | ref->dir_gen = dir_gen; | |
2649 | ref->full_path = path; | |
2650 | ||
ed84885d AS |
2651 | ref->name = (char *)kbasename(ref->full_path->start); |
2652 | ref->name_len = ref->full_path->end - ref->name; | |
2653 | ref->dir_path = ref->full_path->start; | |
2654 | if (ref->name == ref->full_path->start) | |
31db9f7c | 2655 | ref->dir_path_len = 0; |
ed84885d | 2656 | else |
31db9f7c AB |
2657 | ref->dir_path_len = ref->full_path->end - |
2658 | ref->full_path->start - 1 - ref->name_len; | |
31db9f7c AB |
2659 | |
2660 | list_add_tail(&ref->list, head); | |
2661 | return 0; | |
2662 | } | |
2663 | ||
ba5e8f2e JB |
2664 | static int dup_ref(struct recorded_ref *ref, struct list_head *list) |
2665 | { | |
2666 | struct recorded_ref *new; | |
2667 | ||
2668 | new = kmalloc(sizeof(*ref), GFP_NOFS); | |
2669 | if (!new) | |
2670 | return -ENOMEM; | |
2671 | ||
2672 | new->dir = ref->dir; | |
2673 | new->dir_gen = ref->dir_gen; | |
2674 | new->full_path = NULL; | |
2675 | INIT_LIST_HEAD(&new->list); | |
2676 | list_add_tail(&new->list, list); | |
2677 | return 0; | |
2678 | } | |
2679 | ||
924794c9 | 2680 | static void __free_recorded_refs(struct list_head *head) |
31db9f7c AB |
2681 | { |
2682 | struct recorded_ref *cur; | |
31db9f7c | 2683 | |
e938c8ad AB |
2684 | while (!list_empty(head)) { |
2685 | cur = list_entry(head->next, struct recorded_ref, list); | |
924794c9 | 2686 | fs_path_free(cur->full_path); |
e938c8ad | 2687 | list_del(&cur->list); |
31db9f7c AB |
2688 | kfree(cur); |
2689 | } | |
31db9f7c AB |
2690 | } |
2691 | ||
2692 | static void free_recorded_refs(struct send_ctx *sctx) | |
2693 | { | |
924794c9 TI |
2694 | __free_recorded_refs(&sctx->new_refs); |
2695 | __free_recorded_refs(&sctx->deleted_refs); | |
31db9f7c AB |
2696 | } |
2697 | ||
2698 | /* | |
766702ef | 2699 | * Renames/moves a file/dir to its orphan name. Used when the first |
31db9f7c AB |
2700 | * ref of an unprocessed inode gets overwritten and for all non empty |
2701 | * directories. | |
2702 | */ | |
2703 | static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, | |
2704 | struct fs_path *path) | |
2705 | { | |
2706 | int ret; | |
2707 | struct fs_path *orphan; | |
2708 | ||
924794c9 | 2709 | orphan = fs_path_alloc(); |
31db9f7c AB |
2710 | if (!orphan) |
2711 | return -ENOMEM; | |
2712 | ||
2713 | ret = gen_unique_name(sctx, ino, gen, orphan); | |
2714 | if (ret < 0) | |
2715 | goto out; | |
2716 | ||
2717 | ret = send_rename(sctx, path, orphan); | |
2718 | ||
2719 | out: | |
924794c9 | 2720 | fs_path_free(orphan); |
31db9f7c AB |
2721 | return ret; |
2722 | } | |
2723 | ||
9dc44214 FM |
2724 | static struct orphan_dir_info * |
2725 | add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2726 | { | |
2727 | struct rb_node **p = &sctx->orphan_dirs.rb_node; | |
2728 | struct rb_node *parent = NULL; | |
2729 | struct orphan_dir_info *entry, *odi; | |
2730 | ||
2731 | odi = kmalloc(sizeof(*odi), GFP_NOFS); | |
2732 | if (!odi) | |
2733 | return ERR_PTR(-ENOMEM); | |
2734 | odi->ino = dir_ino; | |
2735 | odi->gen = 0; | |
2736 | ||
2737 | while (*p) { | |
2738 | parent = *p; | |
2739 | entry = rb_entry(parent, struct orphan_dir_info, node); | |
2740 | if (dir_ino < entry->ino) { | |
2741 | p = &(*p)->rb_left; | |
2742 | } else if (dir_ino > entry->ino) { | |
2743 | p = &(*p)->rb_right; | |
2744 | } else { | |
2745 | kfree(odi); | |
2746 | return entry; | |
2747 | } | |
2748 | } | |
2749 | ||
2750 | rb_link_node(&odi->node, parent, p); | |
2751 | rb_insert_color(&odi->node, &sctx->orphan_dirs); | |
2752 | return odi; | |
2753 | } | |
2754 | ||
2755 | static struct orphan_dir_info * | |
2756 | get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) | |
2757 | { | |
2758 | struct rb_node *n = sctx->orphan_dirs.rb_node; | |
2759 | struct orphan_dir_info *entry; | |
2760 | ||
2761 | while (n) { | |
2762 | entry = rb_entry(n, struct orphan_dir_info, node); | |
2763 | if (dir_ino < entry->ino) | |
2764 | n = n->rb_left; | |
2765 | else if (dir_ino > entry->ino) | |
2766 | n = n->rb_right; | |
2767 | else | |
2768 | return entry; | |
2769 | } | |
2770 | return NULL; | |
2771 | } | |
2772 | ||
2773 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino) | |
2774 | { | |
2775 | struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino); | |
2776 | ||
2777 | return odi != NULL; | |
2778 | } | |
2779 | ||
2780 | static void free_orphan_dir_info(struct send_ctx *sctx, | |
2781 | struct orphan_dir_info *odi) | |
2782 | { | |
2783 | if (!odi) | |
2784 | return; | |
2785 | rb_erase(&odi->node, &sctx->orphan_dirs); | |
2786 | kfree(odi); | |
2787 | } | |
2788 | ||
31db9f7c AB |
2789 | /* |
2790 | * Returns 1 if a directory can be removed at this point in time. | |
2791 | * We check this by iterating all dir items and checking if the inode behind | |
2792 | * the dir item was already processed. | |
2793 | */ | |
9dc44214 FM |
2794 | static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
2795 | u64 send_progress) | |
31db9f7c AB |
2796 | { |
2797 | int ret = 0; | |
2798 | struct btrfs_root *root = sctx->parent_root; | |
2799 | struct btrfs_path *path; | |
2800 | struct btrfs_key key; | |
2801 | struct btrfs_key found_key; | |
2802 | struct btrfs_key loc; | |
2803 | struct btrfs_dir_item *di; | |
2804 | ||
6d85ed05 AB |
2805 | /* |
2806 | * Don't try to rmdir the top/root subvolume dir. | |
2807 | */ | |
2808 | if (dir == BTRFS_FIRST_FREE_OBJECTID) | |
2809 | return 0; | |
2810 | ||
31db9f7c AB |
2811 | path = alloc_path_for_send(); |
2812 | if (!path) | |
2813 | return -ENOMEM; | |
2814 | ||
2815 | key.objectid = dir; | |
2816 | key.type = BTRFS_DIR_INDEX_KEY; | |
2817 | key.offset = 0; | |
dff6d0ad FDBM |
2818 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2819 | if (ret < 0) | |
2820 | goto out; | |
31db9f7c AB |
2821 | |
2822 | while (1) { | |
9dc44214 FM |
2823 | struct waiting_dir_move *dm; |
2824 | ||
dff6d0ad FDBM |
2825 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
2826 | ret = btrfs_next_leaf(root, path); | |
2827 | if (ret < 0) | |
2828 | goto out; | |
2829 | else if (ret > 0) | |
2830 | break; | |
2831 | continue; | |
31db9f7c | 2832 | } |
dff6d0ad FDBM |
2833 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2834 | path->slots[0]); | |
2835 | if (found_key.objectid != key.objectid || | |
2836 | found_key.type != key.type) | |
31db9f7c | 2837 | break; |
31db9f7c AB |
2838 | |
2839 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2840 | struct btrfs_dir_item); | |
2841 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
2842 | ||
9dc44214 FM |
2843 | dm = get_waiting_dir_move(sctx, loc.objectid); |
2844 | if (dm) { | |
2845 | struct orphan_dir_info *odi; | |
2846 | ||
2847 | odi = add_orphan_dir_info(sctx, dir); | |
2848 | if (IS_ERR(odi)) { | |
2849 | ret = PTR_ERR(odi); | |
2850 | goto out; | |
2851 | } | |
2852 | odi->gen = dir_gen; | |
2853 | dm->rmdir_ino = dir; | |
2854 | ret = 0; | |
2855 | goto out; | |
2856 | } | |
2857 | ||
31db9f7c AB |
2858 | if (loc.objectid > send_progress) { |
2859 | ret = 0; | |
2860 | goto out; | |
2861 | } | |
2862 | ||
dff6d0ad | 2863 | path->slots[0]++; |
31db9f7c AB |
2864 | } |
2865 | ||
2866 | ret = 1; | |
2867 | ||
2868 | out: | |
2869 | btrfs_free_path(path); | |
2870 | return ret; | |
2871 | } | |
2872 | ||
9f03740a FDBM |
2873 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) |
2874 | { | |
9dc44214 | 2875 | struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); |
9f03740a | 2876 | |
9dc44214 | 2877 | return entry != NULL; |
9f03740a FDBM |
2878 | } |
2879 | ||
2880 | static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
2881 | { | |
2882 | struct rb_node **p = &sctx->waiting_dir_moves.rb_node; | |
2883 | struct rb_node *parent = NULL; | |
2884 | struct waiting_dir_move *entry, *dm; | |
2885 | ||
2886 | dm = kmalloc(sizeof(*dm), GFP_NOFS); | |
2887 | if (!dm) | |
2888 | return -ENOMEM; | |
2889 | dm->ino = ino; | |
9dc44214 | 2890 | dm->rmdir_ino = 0; |
9f03740a FDBM |
2891 | |
2892 | while (*p) { | |
2893 | parent = *p; | |
2894 | entry = rb_entry(parent, struct waiting_dir_move, node); | |
2895 | if (ino < entry->ino) { | |
2896 | p = &(*p)->rb_left; | |
2897 | } else if (ino > entry->ino) { | |
2898 | p = &(*p)->rb_right; | |
2899 | } else { | |
2900 | kfree(dm); | |
2901 | return -EEXIST; | |
2902 | } | |
2903 | } | |
2904 | ||
2905 | rb_link_node(&dm->node, parent, p); | |
2906 | rb_insert_color(&dm->node, &sctx->waiting_dir_moves); | |
2907 | return 0; | |
2908 | } | |
2909 | ||
9dc44214 FM |
2910 | static struct waiting_dir_move * |
2911 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
9f03740a FDBM |
2912 | { |
2913 | struct rb_node *n = sctx->waiting_dir_moves.rb_node; | |
2914 | struct waiting_dir_move *entry; | |
2915 | ||
2916 | while (n) { | |
2917 | entry = rb_entry(n, struct waiting_dir_move, node); | |
9dc44214 | 2918 | if (ino < entry->ino) |
9f03740a | 2919 | n = n->rb_left; |
9dc44214 | 2920 | else if (ino > entry->ino) |
9f03740a | 2921 | n = n->rb_right; |
9dc44214 FM |
2922 | else |
2923 | return entry; | |
9f03740a | 2924 | } |
9dc44214 FM |
2925 | return NULL; |
2926 | } | |
2927 | ||
2928 | static void free_waiting_dir_move(struct send_ctx *sctx, | |
2929 | struct waiting_dir_move *dm) | |
2930 | { | |
2931 | if (!dm) | |
2932 | return; | |
2933 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
2934 | kfree(dm); | |
9f03740a FDBM |
2935 | } |
2936 | ||
bfa7e1f8 FM |
2937 | static int add_pending_dir_move(struct send_ctx *sctx, |
2938 | u64 ino, | |
2939 | u64 ino_gen, | |
2940 | u64 parent_ino) | |
9f03740a FDBM |
2941 | { |
2942 | struct rb_node **p = &sctx->pending_dir_moves.rb_node; | |
2943 | struct rb_node *parent = NULL; | |
73b802f4 | 2944 | struct pending_dir_move *entry = NULL, *pm; |
9f03740a FDBM |
2945 | struct recorded_ref *cur; |
2946 | int exists = 0; | |
2947 | int ret; | |
2948 | ||
2949 | pm = kmalloc(sizeof(*pm), GFP_NOFS); | |
2950 | if (!pm) | |
2951 | return -ENOMEM; | |
2952 | pm->parent_ino = parent_ino; | |
bfa7e1f8 FM |
2953 | pm->ino = ino; |
2954 | pm->gen = ino_gen; | |
9f03740a FDBM |
2955 | INIT_LIST_HEAD(&pm->list); |
2956 | INIT_LIST_HEAD(&pm->update_refs); | |
2957 | RB_CLEAR_NODE(&pm->node); | |
2958 | ||
2959 | while (*p) { | |
2960 | parent = *p; | |
2961 | entry = rb_entry(parent, struct pending_dir_move, node); | |
2962 | if (parent_ino < entry->parent_ino) { | |
2963 | p = &(*p)->rb_left; | |
2964 | } else if (parent_ino > entry->parent_ino) { | |
2965 | p = &(*p)->rb_right; | |
2966 | } else { | |
2967 | exists = 1; | |
2968 | break; | |
2969 | } | |
2970 | } | |
2971 | ||
2972 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
2973 | ret = dup_ref(cur, &pm->update_refs); | |
2974 | if (ret < 0) | |
2975 | goto out; | |
2976 | } | |
2977 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
2978 | ret = dup_ref(cur, &pm->update_refs); | |
2979 | if (ret < 0) | |
2980 | goto out; | |
2981 | } | |
2982 | ||
2983 | ret = add_waiting_dir_move(sctx, pm->ino); | |
2984 | if (ret) | |
2985 | goto out; | |
2986 | ||
2987 | if (exists) { | |
2988 | list_add_tail(&pm->list, &entry->list); | |
2989 | } else { | |
2990 | rb_link_node(&pm->node, parent, p); | |
2991 | rb_insert_color(&pm->node, &sctx->pending_dir_moves); | |
2992 | } | |
2993 | ret = 0; | |
2994 | out: | |
2995 | if (ret) { | |
2996 | __free_recorded_refs(&pm->update_refs); | |
2997 | kfree(pm); | |
2998 | } | |
2999 | return ret; | |
3000 | } | |
3001 | ||
3002 | static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, | |
3003 | u64 parent_ino) | |
3004 | { | |
3005 | struct rb_node *n = sctx->pending_dir_moves.rb_node; | |
3006 | struct pending_dir_move *entry; | |
3007 | ||
3008 | while (n) { | |
3009 | entry = rb_entry(n, struct pending_dir_move, node); | |
3010 | if (parent_ino < entry->parent_ino) | |
3011 | n = n->rb_left; | |
3012 | else if (parent_ino > entry->parent_ino) | |
3013 | n = n->rb_right; | |
3014 | else | |
3015 | return entry; | |
3016 | } | |
3017 | return NULL; | |
3018 | } | |
3019 | ||
3020 | static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) | |
3021 | { | |
3022 | struct fs_path *from_path = NULL; | |
3023 | struct fs_path *to_path = NULL; | |
2b863a13 | 3024 | struct fs_path *name = NULL; |
9f03740a FDBM |
3025 | u64 orig_progress = sctx->send_progress; |
3026 | struct recorded_ref *cur; | |
2b863a13 | 3027 | u64 parent_ino, parent_gen; |
9dc44214 FM |
3028 | struct waiting_dir_move *dm = NULL; |
3029 | u64 rmdir_ino = 0; | |
9f03740a FDBM |
3030 | int ret; |
3031 | ||
2b863a13 | 3032 | name = fs_path_alloc(); |
9f03740a | 3033 | from_path = fs_path_alloc(); |
2b863a13 FM |
3034 | if (!name || !from_path) { |
3035 | ret = -ENOMEM; | |
3036 | goto out; | |
3037 | } | |
9f03740a | 3038 | |
9dc44214 FM |
3039 | dm = get_waiting_dir_move(sctx, pm->ino); |
3040 | ASSERT(dm); | |
3041 | rmdir_ino = dm->rmdir_ino; | |
3042 | free_waiting_dir_move(sctx, dm); | |
2b863a13 FM |
3043 | |
3044 | ret = get_first_ref(sctx->parent_root, pm->ino, | |
3045 | &parent_ino, &parent_gen, name); | |
9f03740a FDBM |
3046 | if (ret < 0) |
3047 | goto out; | |
3048 | ||
2b863a13 FM |
3049 | if (parent_ino == sctx->cur_ino) { |
3050 | /* child only renamed, not moved */ | |
3051 | ASSERT(parent_gen == sctx->cur_inode_gen); | |
3052 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
3053 | from_path); | |
3054 | if (ret < 0) | |
3055 | goto out; | |
3056 | ret = fs_path_add_path(from_path, name); | |
3057 | if (ret < 0) | |
3058 | goto out; | |
3059 | } else { | |
3060 | /* child moved and maybe renamed too */ | |
3061 | sctx->send_progress = pm->ino; | |
3062 | ret = get_cur_path(sctx, pm->ino, pm->gen, from_path); | |
3063 | if (ret < 0) | |
3064 | goto out; | |
3065 | } | |
3066 | ||
3067 | fs_path_free(name); | |
3068 | name = NULL; | |
3069 | ||
9f03740a FDBM |
3070 | to_path = fs_path_alloc(); |
3071 | if (!to_path) { | |
3072 | ret = -ENOMEM; | |
3073 | goto out; | |
3074 | } | |
3075 | ||
3076 | sctx->send_progress = sctx->cur_ino + 1; | |
9f03740a FDBM |
3077 | ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); |
3078 | if (ret < 0) | |
3079 | goto out; | |
3080 | ||
3081 | ret = send_rename(sctx, from_path, to_path); | |
3082 | if (ret < 0) | |
3083 | goto out; | |
3084 | ||
9dc44214 FM |
3085 | if (rmdir_ino) { |
3086 | struct orphan_dir_info *odi; | |
3087 | ||
3088 | odi = get_orphan_dir_info(sctx, rmdir_ino); | |
3089 | if (!odi) { | |
3090 | /* already deleted */ | |
3091 | goto finish; | |
3092 | } | |
3093 | ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1); | |
3094 | if (ret < 0) | |
3095 | goto out; | |
3096 | if (!ret) | |
3097 | goto finish; | |
3098 | ||
3099 | name = fs_path_alloc(); | |
3100 | if (!name) { | |
3101 | ret = -ENOMEM; | |
3102 | goto out; | |
3103 | } | |
3104 | ret = get_cur_path(sctx, rmdir_ino, odi->gen, name); | |
3105 | if (ret < 0) | |
3106 | goto out; | |
3107 | ret = send_rmdir(sctx, name); | |
3108 | if (ret < 0) | |
3109 | goto out; | |
3110 | free_orphan_dir_info(sctx, odi); | |
3111 | } | |
3112 | ||
3113 | finish: | |
9f03740a FDBM |
3114 | ret = send_utimes(sctx, pm->ino, pm->gen); |
3115 | if (ret < 0) | |
3116 | goto out; | |
3117 | ||
3118 | /* | |
3119 | * After rename/move, need to update the utimes of both new parent(s) | |
3120 | * and old parent(s). | |
3121 | */ | |
3122 | list_for_each_entry(cur, &pm->update_refs, list) { | |
9dc44214 FM |
3123 | if (cur->dir == rmdir_ino) |
3124 | continue; | |
9f03740a FDBM |
3125 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
3126 | if (ret < 0) | |
3127 | goto out; | |
3128 | } | |
3129 | ||
3130 | out: | |
2b863a13 | 3131 | fs_path_free(name); |
9f03740a FDBM |
3132 | fs_path_free(from_path); |
3133 | fs_path_free(to_path); | |
3134 | sctx->send_progress = orig_progress; | |
3135 | ||
3136 | return ret; | |
3137 | } | |
3138 | ||
3139 | static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) | |
3140 | { | |
3141 | if (!list_empty(&m->list)) | |
3142 | list_del(&m->list); | |
3143 | if (!RB_EMPTY_NODE(&m->node)) | |
3144 | rb_erase(&m->node, &sctx->pending_dir_moves); | |
3145 | __free_recorded_refs(&m->update_refs); | |
3146 | kfree(m); | |
3147 | } | |
3148 | ||
3149 | static void tail_append_pending_moves(struct pending_dir_move *moves, | |
3150 | struct list_head *stack) | |
3151 | { | |
3152 | if (list_empty(&moves->list)) { | |
3153 | list_add_tail(&moves->list, stack); | |
3154 | } else { | |
3155 | LIST_HEAD(list); | |
3156 | list_splice_init(&moves->list, &list); | |
3157 | list_add_tail(&moves->list, stack); | |
3158 | list_splice_tail(&list, stack); | |
3159 | } | |
3160 | } | |
3161 | ||
3162 | static int apply_children_dir_moves(struct send_ctx *sctx) | |
3163 | { | |
3164 | struct pending_dir_move *pm; | |
3165 | struct list_head stack; | |
3166 | u64 parent_ino = sctx->cur_ino; | |
3167 | int ret = 0; | |
3168 | ||
3169 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3170 | if (!pm) | |
3171 | return 0; | |
3172 | ||
3173 | INIT_LIST_HEAD(&stack); | |
3174 | tail_append_pending_moves(pm, &stack); | |
3175 | ||
3176 | while (!list_empty(&stack)) { | |
3177 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3178 | parent_ino = pm->ino; | |
3179 | ret = apply_dir_move(sctx, pm); | |
3180 | free_pending_move(sctx, pm); | |
3181 | if (ret) | |
3182 | goto out; | |
3183 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3184 | if (pm) | |
3185 | tail_append_pending_moves(pm, &stack); | |
3186 | } | |
3187 | return 0; | |
3188 | ||
3189 | out: | |
3190 | while (!list_empty(&stack)) { | |
3191 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3192 | free_pending_move(sctx, pm); | |
3193 | } | |
3194 | return ret; | |
3195 | } | |
3196 | ||
3197 | static int wait_for_parent_move(struct send_ctx *sctx, | |
3198 | struct recorded_ref *parent_ref) | |
3199 | { | |
3200 | int ret; | |
3201 | u64 ino = parent_ref->dir; | |
3202 | u64 parent_ino_before, parent_ino_after; | |
425b5daf | 3203 | u64 old_gen; |
9f03740a FDBM |
3204 | struct fs_path *path_before = NULL; |
3205 | struct fs_path *path_after = NULL; | |
3206 | int len1, len2; | |
bfa7e1f8 FM |
3207 | int register_upper_dirs; |
3208 | u64 gen; | |
9f03740a | 3209 | |
9f03740a FDBM |
3210 | if (is_waiting_for_move(sctx, ino)) |
3211 | return 1; | |
3212 | ||
7b119a8b FM |
3213 | if (parent_ref->dir <= sctx->cur_ino) |
3214 | return 0; | |
3215 | ||
9f03740a FDBM |
3216 | ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen, |
3217 | NULL, NULL, NULL, NULL); | |
3218 | if (ret == -ENOENT) | |
3219 | return 0; | |
3220 | else if (ret < 0) | |
3221 | return ret; | |
3222 | ||
425b5daf | 3223 | if (parent_ref->dir_gen != old_gen) |
9f03740a FDBM |
3224 | return 0; |
3225 | ||
3226 | path_before = fs_path_alloc(); | |
3227 | if (!path_before) | |
3228 | return -ENOMEM; | |
3229 | ||
3230 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
3231 | NULL, path_before); | |
3232 | if (ret == -ENOENT) { | |
3233 | ret = 0; | |
3234 | goto out; | |
3235 | } else if (ret < 0) { | |
3236 | goto out; | |
3237 | } | |
3238 | ||
3239 | path_after = fs_path_alloc(); | |
3240 | if (!path_after) { | |
3241 | ret = -ENOMEM; | |
3242 | goto out; | |
3243 | } | |
3244 | ||
3245 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
bfa7e1f8 | 3246 | &gen, path_after); |
9f03740a FDBM |
3247 | if (ret == -ENOENT) { |
3248 | ret = 0; | |
3249 | goto out; | |
3250 | } else if (ret < 0) { | |
3251 | goto out; | |
3252 | } | |
3253 | ||
3254 | len1 = fs_path_len(path_before); | |
3255 | len2 = fs_path_len(path_after); | |
5ed7f9ff FDBM |
3256 | if (parent_ino_before != parent_ino_after || len1 != len2 || |
3257 | memcmp(path_before->start, path_after->start, len1)) { | |
9f03740a FDBM |
3258 | ret = 1; |
3259 | goto out; | |
3260 | } | |
3261 | ret = 0; | |
3262 | ||
bfa7e1f8 FM |
3263 | /* |
3264 | * Ok, our new most direct ancestor has a higher inode number but | |
3265 | * wasn't moved/renamed. So maybe some of the new ancestors higher in | |
3266 | * the hierarchy have an higher inode number too *and* were renamed | |
3267 | * or moved - in this case we need to wait for the ancestor's rename | |
3268 | * or move operation before we can do the move/rename for the current | |
3269 | * inode. | |
3270 | */ | |
3271 | register_upper_dirs = 0; | |
3272 | ino = parent_ino_after; | |
3273 | again: | |
3274 | while ((ret == 0 || register_upper_dirs) && ino > sctx->cur_ino) { | |
3275 | u64 parent_gen; | |
3276 | ||
3277 | fs_path_reset(path_before); | |
3278 | fs_path_reset(path_after); | |
3279 | ||
3280 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
3281 | &parent_gen, path_after); | |
3282 | if (ret < 0) | |
3283 | goto out; | |
3284 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
3285 | NULL, path_before); | |
3286 | if (ret == -ENOENT) { | |
3287 | ret = 0; | |
3288 | break; | |
3289 | } else if (ret < 0) { | |
3290 | goto out; | |
3291 | } | |
3292 | ||
3293 | len1 = fs_path_len(path_before); | |
3294 | len2 = fs_path_len(path_after); | |
3295 | if (parent_ino_before != parent_ino_after || len1 != len2 || | |
3296 | memcmp(path_before->start, path_after->start, len1)) { | |
3297 | ret = 1; | |
3298 | if (register_upper_dirs) { | |
3299 | break; | |
3300 | } else { | |
3301 | register_upper_dirs = 1; | |
3302 | ino = parent_ref->dir; | |
3303 | gen = parent_ref->dir_gen; | |
3304 | goto again; | |
3305 | } | |
3306 | } else if (register_upper_dirs) { | |
3307 | ret = add_pending_dir_move(sctx, ino, gen, | |
3308 | parent_ino_after); | |
3309 | if (ret < 0 && ret != -EEXIST) | |
3310 | goto out; | |
3311 | } | |
3312 | ||
3313 | ino = parent_ino_after; | |
3314 | gen = parent_gen; | |
3315 | } | |
3316 | ||
9f03740a FDBM |
3317 | out: |
3318 | fs_path_free(path_before); | |
3319 | fs_path_free(path_after); | |
3320 | ||
3321 | return ret; | |
3322 | } | |
3323 | ||
31db9f7c AB |
3324 | /* |
3325 | * This does all the move/link/unlink/rmdir magic. | |
3326 | */ | |
9f03740a | 3327 | static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) |
31db9f7c AB |
3328 | { |
3329 | int ret = 0; | |
3330 | struct recorded_ref *cur; | |
1f4692da | 3331 | struct recorded_ref *cur2; |
ba5e8f2e | 3332 | struct list_head check_dirs; |
31db9f7c | 3333 | struct fs_path *valid_path = NULL; |
b24baf69 | 3334 | u64 ow_inode = 0; |
31db9f7c AB |
3335 | u64 ow_gen; |
3336 | int did_overwrite = 0; | |
3337 | int is_orphan = 0; | |
29d6d30f | 3338 | u64 last_dir_ino_rm = 0; |
31db9f7c AB |
3339 | |
3340 | verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino); | |
3341 | ||
6d85ed05 AB |
3342 | /* |
3343 | * This should never happen as the root dir always has the same ref | |
3344 | * which is always '..' | |
3345 | */ | |
3346 | BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); | |
ba5e8f2e | 3347 | INIT_LIST_HEAD(&check_dirs); |
6d85ed05 | 3348 | |
924794c9 | 3349 | valid_path = fs_path_alloc(); |
31db9f7c AB |
3350 | if (!valid_path) { |
3351 | ret = -ENOMEM; | |
3352 | goto out; | |
3353 | } | |
3354 | ||
31db9f7c AB |
3355 | /* |
3356 | * First, check if the first ref of the current inode was overwritten | |
3357 | * before. If yes, we know that the current inode was already orphanized | |
3358 | * and thus use the orphan name. If not, we can use get_cur_path to | |
3359 | * get the path of the first ref as it would like while receiving at | |
3360 | * this point in time. | |
3361 | * New inodes are always orphan at the beginning, so force to use the | |
3362 | * orphan name in this case. | |
3363 | * The first ref is stored in valid_path and will be updated if it | |
3364 | * gets moved around. | |
3365 | */ | |
3366 | if (!sctx->cur_inode_new) { | |
3367 | ret = did_overwrite_first_ref(sctx, sctx->cur_ino, | |
3368 | sctx->cur_inode_gen); | |
3369 | if (ret < 0) | |
3370 | goto out; | |
3371 | if (ret) | |
3372 | did_overwrite = 1; | |
3373 | } | |
3374 | if (sctx->cur_inode_new || did_overwrite) { | |
3375 | ret = gen_unique_name(sctx, sctx->cur_ino, | |
3376 | sctx->cur_inode_gen, valid_path); | |
3377 | if (ret < 0) | |
3378 | goto out; | |
3379 | is_orphan = 1; | |
3380 | } else { | |
3381 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
3382 | valid_path); | |
3383 | if (ret < 0) | |
3384 | goto out; | |
3385 | } | |
3386 | ||
3387 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
1f4692da AB |
3388 | /* |
3389 | * We may have refs where the parent directory does not exist | |
3390 | * yet. This happens if the parent directories inum is higher | |
3391 | * the the current inum. To handle this case, we create the | |
3392 | * parent directory out of order. But we need to check if this | |
3393 | * did already happen before due to other refs in the same dir. | |
3394 | */ | |
3395 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); | |
3396 | if (ret < 0) | |
3397 | goto out; | |
3398 | if (ret == inode_state_will_create) { | |
3399 | ret = 0; | |
3400 | /* | |
3401 | * First check if any of the current inodes refs did | |
3402 | * already create the dir. | |
3403 | */ | |
3404 | list_for_each_entry(cur2, &sctx->new_refs, list) { | |
3405 | if (cur == cur2) | |
3406 | break; | |
3407 | if (cur2->dir == cur->dir) { | |
3408 | ret = 1; | |
3409 | break; | |
3410 | } | |
3411 | } | |
3412 | ||
3413 | /* | |
3414 | * If that did not happen, check if a previous inode | |
3415 | * did already create the dir. | |
3416 | */ | |
3417 | if (!ret) | |
3418 | ret = did_create_dir(sctx, cur->dir); | |
3419 | if (ret < 0) | |
3420 | goto out; | |
3421 | if (!ret) { | |
3422 | ret = send_create_inode(sctx, cur->dir); | |
3423 | if (ret < 0) | |
3424 | goto out; | |
3425 | } | |
3426 | } | |
3427 | ||
31db9f7c AB |
3428 | /* |
3429 | * Check if this new ref would overwrite the first ref of | |
3430 | * another unprocessed inode. If yes, orphanize the | |
3431 | * overwritten inode. If we find an overwritten ref that is | |
3432 | * not the first ref, simply unlink it. | |
3433 | */ | |
3434 | ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
3435 | cur->name, cur->name_len, | |
3436 | &ow_inode, &ow_gen); | |
3437 | if (ret < 0) | |
3438 | goto out; | |
3439 | if (ret) { | |
924794c9 TI |
3440 | ret = is_first_ref(sctx->parent_root, |
3441 | ow_inode, cur->dir, cur->name, | |
3442 | cur->name_len); | |
31db9f7c AB |
3443 | if (ret < 0) |
3444 | goto out; | |
3445 | if (ret) { | |
3446 | ret = orphanize_inode(sctx, ow_inode, ow_gen, | |
3447 | cur->full_path); | |
3448 | if (ret < 0) | |
3449 | goto out; | |
3450 | } else { | |
3451 | ret = send_unlink(sctx, cur->full_path); | |
3452 | if (ret < 0) | |
3453 | goto out; | |
3454 | } | |
3455 | } | |
3456 | ||
3457 | /* | |
3458 | * link/move the ref to the new place. If we have an orphan | |
3459 | * inode, move it and update valid_path. If not, link or move | |
3460 | * it depending on the inode mode. | |
3461 | */ | |
1f4692da | 3462 | if (is_orphan) { |
31db9f7c AB |
3463 | ret = send_rename(sctx, valid_path, cur->full_path); |
3464 | if (ret < 0) | |
3465 | goto out; | |
3466 | is_orphan = 0; | |
3467 | ret = fs_path_copy(valid_path, cur->full_path); | |
3468 | if (ret < 0) | |
3469 | goto out; | |
3470 | } else { | |
3471 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
3472 | /* | |
3473 | * Dirs can't be linked, so move it. For moved | |
3474 | * dirs, we always have one new and one deleted | |
3475 | * ref. The deleted ref is ignored later. | |
3476 | */ | |
d86477b3 FDBM |
3477 | ret = wait_for_parent_move(sctx, cur); |
3478 | if (ret < 0) | |
3479 | goto out; | |
3480 | if (ret) { | |
9f03740a | 3481 | ret = add_pending_dir_move(sctx, |
bfa7e1f8 FM |
3482 | sctx->cur_ino, |
3483 | sctx->cur_inode_gen, | |
3484 | cur->dir); | |
9f03740a FDBM |
3485 | *pending_move = 1; |
3486 | } else { | |
3487 | ret = send_rename(sctx, valid_path, | |
3488 | cur->full_path); | |
3489 | if (!ret) | |
3490 | ret = fs_path_copy(valid_path, | |
3491 | cur->full_path); | |
3492 | } | |
31db9f7c AB |
3493 | if (ret < 0) |
3494 | goto out; | |
3495 | } else { | |
3496 | ret = send_link(sctx, cur->full_path, | |
3497 | valid_path); | |
3498 | if (ret < 0) | |
3499 | goto out; | |
3500 | } | |
3501 | } | |
ba5e8f2e | 3502 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
3503 | if (ret < 0) |
3504 | goto out; | |
3505 | } | |
3506 | ||
3507 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { | |
3508 | /* | |
3509 | * Check if we can already rmdir the directory. If not, | |
3510 | * orphanize it. For every dir item inside that gets deleted | |
3511 | * later, we do this check again and rmdir it then if possible. | |
3512 | * See the use of check_dirs for more details. | |
3513 | */ | |
9dc44214 FM |
3514 | ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
3515 | sctx->cur_ino); | |
31db9f7c AB |
3516 | if (ret < 0) |
3517 | goto out; | |
3518 | if (ret) { | |
3519 | ret = send_rmdir(sctx, valid_path); | |
3520 | if (ret < 0) | |
3521 | goto out; | |
3522 | } else if (!is_orphan) { | |
3523 | ret = orphanize_inode(sctx, sctx->cur_ino, | |
3524 | sctx->cur_inode_gen, valid_path); | |
3525 | if (ret < 0) | |
3526 | goto out; | |
3527 | is_orphan = 1; | |
3528 | } | |
3529 | ||
3530 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
ba5e8f2e | 3531 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
3532 | if (ret < 0) |
3533 | goto out; | |
3534 | } | |
ccf1626b AB |
3535 | } else if (S_ISDIR(sctx->cur_inode_mode) && |
3536 | !list_empty(&sctx->deleted_refs)) { | |
3537 | /* | |
3538 | * We have a moved dir. Add the old parent to check_dirs | |
3539 | */ | |
3540 | cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, | |
3541 | list); | |
ba5e8f2e | 3542 | ret = dup_ref(cur, &check_dirs); |
ccf1626b AB |
3543 | if (ret < 0) |
3544 | goto out; | |
31db9f7c AB |
3545 | } else if (!S_ISDIR(sctx->cur_inode_mode)) { |
3546 | /* | |
3547 | * We have a non dir inode. Go through all deleted refs and | |
3548 | * unlink them if they were not already overwritten by other | |
3549 | * inodes. | |
3550 | */ | |
3551 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
3552 | ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
3553 | sctx->cur_ino, sctx->cur_inode_gen, | |
3554 | cur->name, cur->name_len); | |
3555 | if (ret < 0) | |
3556 | goto out; | |
3557 | if (!ret) { | |
1f4692da AB |
3558 | ret = send_unlink(sctx, cur->full_path); |
3559 | if (ret < 0) | |
3560 | goto out; | |
31db9f7c | 3561 | } |
ba5e8f2e | 3562 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
3563 | if (ret < 0) |
3564 | goto out; | |
3565 | } | |
31db9f7c AB |
3566 | /* |
3567 | * If the inode is still orphan, unlink the orphan. This may | |
3568 | * happen when a previous inode did overwrite the first ref | |
3569 | * of this inode and no new refs were added for the current | |
766702ef AB |
3570 | * inode. Unlinking does not mean that the inode is deleted in |
3571 | * all cases. There may still be links to this inode in other | |
3572 | * places. | |
31db9f7c | 3573 | */ |
1f4692da | 3574 | if (is_orphan) { |
31db9f7c AB |
3575 | ret = send_unlink(sctx, valid_path); |
3576 | if (ret < 0) | |
3577 | goto out; | |
3578 | } | |
3579 | } | |
3580 | ||
3581 | /* | |
3582 | * We did collect all parent dirs where cur_inode was once located. We | |
3583 | * now go through all these dirs and check if they are pending for | |
3584 | * deletion and if it's finally possible to perform the rmdir now. | |
3585 | * We also update the inode stats of the parent dirs here. | |
3586 | */ | |
ba5e8f2e | 3587 | list_for_each_entry(cur, &check_dirs, list) { |
766702ef AB |
3588 | /* |
3589 | * In case we had refs into dirs that were not processed yet, | |
3590 | * we don't need to do the utime and rmdir logic for these dirs. | |
3591 | * The dir will be processed later. | |
3592 | */ | |
ba5e8f2e | 3593 | if (cur->dir > sctx->cur_ino) |
31db9f7c AB |
3594 | continue; |
3595 | ||
ba5e8f2e | 3596 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
3597 | if (ret < 0) |
3598 | goto out; | |
3599 | ||
3600 | if (ret == inode_state_did_create || | |
3601 | ret == inode_state_no_change) { | |
3602 | /* TODO delayed utimes */ | |
ba5e8f2e | 3603 | ret = send_utimes(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
3604 | if (ret < 0) |
3605 | goto out; | |
29d6d30f FM |
3606 | } else if (ret == inode_state_did_delete && |
3607 | cur->dir != last_dir_ino_rm) { | |
9dc44214 FM |
3608 | ret = can_rmdir(sctx, cur->dir, cur->dir_gen, |
3609 | sctx->cur_ino); | |
31db9f7c AB |
3610 | if (ret < 0) |
3611 | goto out; | |
3612 | if (ret) { | |
ba5e8f2e JB |
3613 | ret = get_cur_path(sctx, cur->dir, |
3614 | cur->dir_gen, valid_path); | |
31db9f7c AB |
3615 | if (ret < 0) |
3616 | goto out; | |
3617 | ret = send_rmdir(sctx, valid_path); | |
3618 | if (ret < 0) | |
3619 | goto out; | |
29d6d30f | 3620 | last_dir_ino_rm = cur->dir; |
31db9f7c AB |
3621 | } |
3622 | } | |
3623 | } | |
3624 | ||
31db9f7c AB |
3625 | ret = 0; |
3626 | ||
3627 | out: | |
ba5e8f2e | 3628 | __free_recorded_refs(&check_dirs); |
31db9f7c | 3629 | free_recorded_refs(sctx); |
924794c9 | 3630 | fs_path_free(valid_path); |
31db9f7c AB |
3631 | return ret; |
3632 | } | |
3633 | ||
a4d96d62 LB |
3634 | static int record_ref(struct btrfs_root *root, int num, u64 dir, int index, |
3635 | struct fs_path *name, void *ctx, struct list_head *refs) | |
31db9f7c AB |
3636 | { |
3637 | int ret = 0; | |
3638 | struct send_ctx *sctx = ctx; | |
3639 | struct fs_path *p; | |
3640 | u64 gen; | |
3641 | ||
924794c9 | 3642 | p = fs_path_alloc(); |
31db9f7c AB |
3643 | if (!p) |
3644 | return -ENOMEM; | |
3645 | ||
a4d96d62 | 3646 | ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, |
85a7b33b | 3647 | NULL, NULL); |
31db9f7c AB |
3648 | if (ret < 0) |
3649 | goto out; | |
3650 | ||
31db9f7c AB |
3651 | ret = get_cur_path(sctx, dir, gen, p); |
3652 | if (ret < 0) | |
3653 | goto out; | |
3654 | ret = fs_path_add_path(p, name); | |
3655 | if (ret < 0) | |
3656 | goto out; | |
3657 | ||
a4d96d62 | 3658 | ret = __record_ref(refs, dir, gen, p); |
31db9f7c AB |
3659 | |
3660 | out: | |
3661 | if (ret) | |
924794c9 | 3662 | fs_path_free(p); |
31db9f7c AB |
3663 | return ret; |
3664 | } | |
3665 | ||
a4d96d62 LB |
3666 | static int __record_new_ref(int num, u64 dir, int index, |
3667 | struct fs_path *name, | |
3668 | void *ctx) | |
3669 | { | |
3670 | struct send_ctx *sctx = ctx; | |
3671 | return record_ref(sctx->send_root, num, dir, index, name, | |
3672 | ctx, &sctx->new_refs); | |
3673 | } | |
3674 | ||
3675 | ||
31db9f7c AB |
3676 | static int __record_deleted_ref(int num, u64 dir, int index, |
3677 | struct fs_path *name, | |
3678 | void *ctx) | |
3679 | { | |
31db9f7c | 3680 | struct send_ctx *sctx = ctx; |
a4d96d62 LB |
3681 | return record_ref(sctx->parent_root, num, dir, index, name, |
3682 | ctx, &sctx->deleted_refs); | |
31db9f7c AB |
3683 | } |
3684 | ||
3685 | static int record_new_ref(struct send_ctx *sctx) | |
3686 | { | |
3687 | int ret; | |
3688 | ||
924794c9 TI |
3689 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
3690 | sctx->cmp_key, 0, __record_new_ref, sctx); | |
31db9f7c AB |
3691 | if (ret < 0) |
3692 | goto out; | |
3693 | ret = 0; | |
3694 | ||
3695 | out: | |
3696 | return ret; | |
3697 | } | |
3698 | ||
3699 | static int record_deleted_ref(struct send_ctx *sctx) | |
3700 | { | |
3701 | int ret; | |
3702 | ||
924794c9 TI |
3703 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
3704 | sctx->cmp_key, 0, __record_deleted_ref, sctx); | |
31db9f7c AB |
3705 | if (ret < 0) |
3706 | goto out; | |
3707 | ret = 0; | |
3708 | ||
3709 | out: | |
3710 | return ret; | |
3711 | } | |
3712 | ||
3713 | struct find_ref_ctx { | |
3714 | u64 dir; | |
ba5e8f2e JB |
3715 | u64 dir_gen; |
3716 | struct btrfs_root *root; | |
31db9f7c AB |
3717 | struct fs_path *name; |
3718 | int found_idx; | |
3719 | }; | |
3720 | ||
3721 | static int __find_iref(int num, u64 dir, int index, | |
3722 | struct fs_path *name, | |
3723 | void *ctx_) | |
3724 | { | |
3725 | struct find_ref_ctx *ctx = ctx_; | |
ba5e8f2e JB |
3726 | u64 dir_gen; |
3727 | int ret; | |
31db9f7c AB |
3728 | |
3729 | if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && | |
3730 | strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { | |
ba5e8f2e JB |
3731 | /* |
3732 | * To avoid doing extra lookups we'll only do this if everything | |
3733 | * else matches. | |
3734 | */ | |
3735 | ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, | |
3736 | NULL, NULL, NULL); | |
3737 | if (ret) | |
3738 | return ret; | |
3739 | if (dir_gen != ctx->dir_gen) | |
3740 | return 0; | |
31db9f7c AB |
3741 | ctx->found_idx = num; |
3742 | return 1; | |
3743 | } | |
3744 | return 0; | |
3745 | } | |
3746 | ||
924794c9 | 3747 | static int find_iref(struct btrfs_root *root, |
31db9f7c AB |
3748 | struct btrfs_path *path, |
3749 | struct btrfs_key *key, | |
ba5e8f2e | 3750 | u64 dir, u64 dir_gen, struct fs_path *name) |
31db9f7c AB |
3751 | { |
3752 | int ret; | |
3753 | struct find_ref_ctx ctx; | |
3754 | ||
3755 | ctx.dir = dir; | |
3756 | ctx.name = name; | |
ba5e8f2e | 3757 | ctx.dir_gen = dir_gen; |
31db9f7c | 3758 | ctx.found_idx = -1; |
ba5e8f2e | 3759 | ctx.root = root; |
31db9f7c | 3760 | |
924794c9 | 3761 | ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); |
31db9f7c AB |
3762 | if (ret < 0) |
3763 | return ret; | |
3764 | ||
3765 | if (ctx.found_idx == -1) | |
3766 | return -ENOENT; | |
3767 | ||
3768 | return ctx.found_idx; | |
3769 | } | |
3770 | ||
3771 | static int __record_changed_new_ref(int num, u64 dir, int index, | |
3772 | struct fs_path *name, | |
3773 | void *ctx) | |
3774 | { | |
ba5e8f2e | 3775 | u64 dir_gen; |
31db9f7c AB |
3776 | int ret; |
3777 | struct send_ctx *sctx = ctx; | |
3778 | ||
ba5e8f2e JB |
3779 | ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, |
3780 | NULL, NULL, NULL); | |
3781 | if (ret) | |
3782 | return ret; | |
3783 | ||
924794c9 | 3784 | ret = find_iref(sctx->parent_root, sctx->right_path, |
ba5e8f2e | 3785 | sctx->cmp_key, dir, dir_gen, name); |
31db9f7c AB |
3786 | if (ret == -ENOENT) |
3787 | ret = __record_new_ref(num, dir, index, name, sctx); | |
3788 | else if (ret > 0) | |
3789 | ret = 0; | |
3790 | ||
3791 | return ret; | |
3792 | } | |
3793 | ||
3794 | static int __record_changed_deleted_ref(int num, u64 dir, int index, | |
3795 | struct fs_path *name, | |
3796 | void *ctx) | |
3797 | { | |
ba5e8f2e | 3798 | u64 dir_gen; |
31db9f7c AB |
3799 | int ret; |
3800 | struct send_ctx *sctx = ctx; | |
3801 | ||
ba5e8f2e JB |
3802 | ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, |
3803 | NULL, NULL, NULL); | |
3804 | if (ret) | |
3805 | return ret; | |
3806 | ||
924794c9 | 3807 | ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, |
ba5e8f2e | 3808 | dir, dir_gen, name); |
31db9f7c AB |
3809 | if (ret == -ENOENT) |
3810 | ret = __record_deleted_ref(num, dir, index, name, sctx); | |
3811 | else if (ret > 0) | |
3812 | ret = 0; | |
3813 | ||
3814 | return ret; | |
3815 | } | |
3816 | ||
3817 | static int record_changed_ref(struct send_ctx *sctx) | |
3818 | { | |
3819 | int ret = 0; | |
3820 | ||
924794c9 | 3821 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
31db9f7c AB |
3822 | sctx->cmp_key, 0, __record_changed_new_ref, sctx); |
3823 | if (ret < 0) | |
3824 | goto out; | |
924794c9 | 3825 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
3826 | sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); |
3827 | if (ret < 0) | |
3828 | goto out; | |
3829 | ret = 0; | |
3830 | ||
3831 | out: | |
3832 | return ret; | |
3833 | } | |
3834 | ||
3835 | /* | |
3836 | * Record and process all refs at once. Needed when an inode changes the | |
3837 | * generation number, which means that it was deleted and recreated. | |
3838 | */ | |
3839 | static int process_all_refs(struct send_ctx *sctx, | |
3840 | enum btrfs_compare_tree_result cmd) | |
3841 | { | |
3842 | int ret; | |
3843 | struct btrfs_root *root; | |
3844 | struct btrfs_path *path; | |
3845 | struct btrfs_key key; | |
3846 | struct btrfs_key found_key; | |
3847 | struct extent_buffer *eb; | |
3848 | int slot; | |
3849 | iterate_inode_ref_t cb; | |
9f03740a | 3850 | int pending_move = 0; |
31db9f7c AB |
3851 | |
3852 | path = alloc_path_for_send(); | |
3853 | if (!path) | |
3854 | return -ENOMEM; | |
3855 | ||
3856 | if (cmd == BTRFS_COMPARE_TREE_NEW) { | |
3857 | root = sctx->send_root; | |
3858 | cb = __record_new_ref; | |
3859 | } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { | |
3860 | root = sctx->parent_root; | |
3861 | cb = __record_deleted_ref; | |
3862 | } else { | |
4d1a63b2 DS |
3863 | btrfs_err(sctx->send_root->fs_info, |
3864 | "Wrong command %d in process_all_refs", cmd); | |
3865 | ret = -EINVAL; | |
3866 | goto out; | |
31db9f7c AB |
3867 | } |
3868 | ||
3869 | key.objectid = sctx->cmp_key->objectid; | |
3870 | key.type = BTRFS_INODE_REF_KEY; | |
3871 | key.offset = 0; | |
dff6d0ad FDBM |
3872 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3873 | if (ret < 0) | |
3874 | goto out; | |
31db9f7c | 3875 | |
dff6d0ad | 3876 | while (1) { |
31db9f7c AB |
3877 | eb = path->nodes[0]; |
3878 | slot = path->slots[0]; | |
dff6d0ad FDBM |
3879 | if (slot >= btrfs_header_nritems(eb)) { |
3880 | ret = btrfs_next_leaf(root, path); | |
3881 | if (ret < 0) | |
3882 | goto out; | |
3883 | else if (ret > 0) | |
3884 | break; | |
3885 | continue; | |
3886 | } | |
3887 | ||
31db9f7c AB |
3888 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
3889 | ||
3890 | if (found_key.objectid != key.objectid || | |
96b5bd77 JS |
3891 | (found_key.type != BTRFS_INODE_REF_KEY && |
3892 | found_key.type != BTRFS_INODE_EXTREF_KEY)) | |
31db9f7c | 3893 | break; |
31db9f7c | 3894 | |
924794c9 | 3895 | ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); |
31db9f7c AB |
3896 | if (ret < 0) |
3897 | goto out; | |
3898 | ||
dff6d0ad | 3899 | path->slots[0]++; |
31db9f7c | 3900 | } |
e938c8ad | 3901 | btrfs_release_path(path); |
31db9f7c | 3902 | |
9f03740a FDBM |
3903 | ret = process_recorded_refs(sctx, &pending_move); |
3904 | /* Only applicable to an incremental send. */ | |
3905 | ASSERT(pending_move == 0); | |
31db9f7c AB |
3906 | |
3907 | out: | |
3908 | btrfs_free_path(path); | |
3909 | return ret; | |
3910 | } | |
3911 | ||
3912 | static int send_set_xattr(struct send_ctx *sctx, | |
3913 | struct fs_path *path, | |
3914 | const char *name, int name_len, | |
3915 | const char *data, int data_len) | |
3916 | { | |
3917 | int ret = 0; | |
3918 | ||
3919 | ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); | |
3920 | if (ret < 0) | |
3921 | goto out; | |
3922 | ||
3923 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
3924 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
3925 | TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); | |
3926 | ||
3927 | ret = send_cmd(sctx); | |
3928 | ||
3929 | tlv_put_failure: | |
3930 | out: | |
3931 | return ret; | |
3932 | } | |
3933 | ||
3934 | static int send_remove_xattr(struct send_ctx *sctx, | |
3935 | struct fs_path *path, | |
3936 | const char *name, int name_len) | |
3937 | { | |
3938 | int ret = 0; | |
3939 | ||
3940 | ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); | |
3941 | if (ret < 0) | |
3942 | goto out; | |
3943 | ||
3944 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
3945 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
3946 | ||
3947 | ret = send_cmd(sctx); | |
3948 | ||
3949 | tlv_put_failure: | |
3950 | out: | |
3951 | return ret; | |
3952 | } | |
3953 | ||
3954 | static int __process_new_xattr(int num, struct btrfs_key *di_key, | |
3955 | const char *name, int name_len, | |
3956 | const char *data, int data_len, | |
3957 | u8 type, void *ctx) | |
3958 | { | |
3959 | int ret; | |
3960 | struct send_ctx *sctx = ctx; | |
3961 | struct fs_path *p; | |
3962 | posix_acl_xattr_header dummy_acl; | |
3963 | ||
924794c9 | 3964 | p = fs_path_alloc(); |
31db9f7c AB |
3965 | if (!p) |
3966 | return -ENOMEM; | |
3967 | ||
3968 | /* | |
3969 | * This hack is needed because empty acl's are stored as zero byte | |
3970 | * data in xattrs. Problem with that is, that receiving these zero byte | |
3971 | * acl's will fail later. To fix this, we send a dummy acl list that | |
3972 | * only contains the version number and no entries. | |
3973 | */ | |
3974 | if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || | |
3975 | !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { | |
3976 | if (data_len == 0) { | |
3977 | dummy_acl.a_version = | |
3978 | cpu_to_le32(POSIX_ACL_XATTR_VERSION); | |
3979 | data = (char *)&dummy_acl; | |
3980 | data_len = sizeof(dummy_acl); | |
3981 | } | |
3982 | } | |
3983 | ||
3984 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
3985 | if (ret < 0) | |
3986 | goto out; | |
3987 | ||
3988 | ret = send_set_xattr(sctx, p, name, name_len, data, data_len); | |
3989 | ||
3990 | out: | |
924794c9 | 3991 | fs_path_free(p); |
31db9f7c AB |
3992 | return ret; |
3993 | } | |
3994 | ||
3995 | static int __process_deleted_xattr(int num, struct btrfs_key *di_key, | |
3996 | const char *name, int name_len, | |
3997 | const char *data, int data_len, | |
3998 | u8 type, void *ctx) | |
3999 | { | |
4000 | int ret; | |
4001 | struct send_ctx *sctx = ctx; | |
4002 | struct fs_path *p; | |
4003 | ||
924794c9 | 4004 | p = fs_path_alloc(); |
31db9f7c AB |
4005 | if (!p) |
4006 | return -ENOMEM; | |
4007 | ||
4008 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4009 | if (ret < 0) | |
4010 | goto out; | |
4011 | ||
4012 | ret = send_remove_xattr(sctx, p, name, name_len); | |
4013 | ||
4014 | out: | |
924794c9 | 4015 | fs_path_free(p); |
31db9f7c AB |
4016 | return ret; |
4017 | } | |
4018 | ||
4019 | static int process_new_xattr(struct send_ctx *sctx) | |
4020 | { | |
4021 | int ret = 0; | |
4022 | ||
924794c9 TI |
4023 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
4024 | sctx->cmp_key, __process_new_xattr, sctx); | |
31db9f7c AB |
4025 | |
4026 | return ret; | |
4027 | } | |
4028 | ||
4029 | static int process_deleted_xattr(struct send_ctx *sctx) | |
4030 | { | |
4031 | int ret; | |
4032 | ||
924794c9 TI |
4033 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
4034 | sctx->cmp_key, __process_deleted_xattr, sctx); | |
31db9f7c AB |
4035 | |
4036 | return ret; | |
4037 | } | |
4038 | ||
4039 | struct find_xattr_ctx { | |
4040 | const char *name; | |
4041 | int name_len; | |
4042 | int found_idx; | |
4043 | char *found_data; | |
4044 | int found_data_len; | |
4045 | }; | |
4046 | ||
4047 | static int __find_xattr(int num, struct btrfs_key *di_key, | |
4048 | const char *name, int name_len, | |
4049 | const char *data, int data_len, | |
4050 | u8 type, void *vctx) | |
4051 | { | |
4052 | struct find_xattr_ctx *ctx = vctx; | |
4053 | ||
4054 | if (name_len == ctx->name_len && | |
4055 | strncmp(name, ctx->name, name_len) == 0) { | |
4056 | ctx->found_idx = num; | |
4057 | ctx->found_data_len = data_len; | |
a5959bc0 | 4058 | ctx->found_data = kmemdup(data, data_len, GFP_NOFS); |
31db9f7c AB |
4059 | if (!ctx->found_data) |
4060 | return -ENOMEM; | |
31db9f7c AB |
4061 | return 1; |
4062 | } | |
4063 | return 0; | |
4064 | } | |
4065 | ||
924794c9 | 4066 | static int find_xattr(struct btrfs_root *root, |
31db9f7c AB |
4067 | struct btrfs_path *path, |
4068 | struct btrfs_key *key, | |
4069 | const char *name, int name_len, | |
4070 | char **data, int *data_len) | |
4071 | { | |
4072 | int ret; | |
4073 | struct find_xattr_ctx ctx; | |
4074 | ||
4075 | ctx.name = name; | |
4076 | ctx.name_len = name_len; | |
4077 | ctx.found_idx = -1; | |
4078 | ctx.found_data = NULL; | |
4079 | ctx.found_data_len = 0; | |
4080 | ||
924794c9 | 4081 | ret = iterate_dir_item(root, path, key, __find_xattr, &ctx); |
31db9f7c AB |
4082 | if (ret < 0) |
4083 | return ret; | |
4084 | ||
4085 | if (ctx.found_idx == -1) | |
4086 | return -ENOENT; | |
4087 | if (data) { | |
4088 | *data = ctx.found_data; | |
4089 | *data_len = ctx.found_data_len; | |
4090 | } else { | |
4091 | kfree(ctx.found_data); | |
4092 | } | |
4093 | return ctx.found_idx; | |
4094 | } | |
4095 | ||
4096 | ||
4097 | static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, | |
4098 | const char *name, int name_len, | |
4099 | const char *data, int data_len, | |
4100 | u8 type, void *ctx) | |
4101 | { | |
4102 | int ret; | |
4103 | struct send_ctx *sctx = ctx; | |
4104 | char *found_data = NULL; | |
4105 | int found_data_len = 0; | |
31db9f7c | 4106 | |
924794c9 TI |
4107 | ret = find_xattr(sctx->parent_root, sctx->right_path, |
4108 | sctx->cmp_key, name, name_len, &found_data, | |
4109 | &found_data_len); | |
31db9f7c AB |
4110 | if (ret == -ENOENT) { |
4111 | ret = __process_new_xattr(num, di_key, name, name_len, data, | |
4112 | data_len, type, ctx); | |
4113 | } else if (ret >= 0) { | |
4114 | if (data_len != found_data_len || | |
4115 | memcmp(data, found_data, data_len)) { | |
4116 | ret = __process_new_xattr(num, di_key, name, name_len, | |
4117 | data, data_len, type, ctx); | |
4118 | } else { | |
4119 | ret = 0; | |
4120 | } | |
4121 | } | |
4122 | ||
4123 | kfree(found_data); | |
31db9f7c AB |
4124 | return ret; |
4125 | } | |
4126 | ||
4127 | static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, | |
4128 | const char *name, int name_len, | |
4129 | const char *data, int data_len, | |
4130 | u8 type, void *ctx) | |
4131 | { | |
4132 | int ret; | |
4133 | struct send_ctx *sctx = ctx; | |
4134 | ||
924794c9 TI |
4135 | ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, |
4136 | name, name_len, NULL, NULL); | |
31db9f7c AB |
4137 | if (ret == -ENOENT) |
4138 | ret = __process_deleted_xattr(num, di_key, name, name_len, data, | |
4139 | data_len, type, ctx); | |
4140 | else if (ret >= 0) | |
4141 | ret = 0; | |
4142 | ||
4143 | return ret; | |
4144 | } | |
4145 | ||
4146 | static int process_changed_xattr(struct send_ctx *sctx) | |
4147 | { | |
4148 | int ret = 0; | |
4149 | ||
924794c9 | 4150 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
31db9f7c AB |
4151 | sctx->cmp_key, __process_changed_new_xattr, sctx); |
4152 | if (ret < 0) | |
4153 | goto out; | |
924794c9 | 4154 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
31db9f7c AB |
4155 | sctx->cmp_key, __process_changed_deleted_xattr, sctx); |
4156 | ||
4157 | out: | |
4158 | return ret; | |
4159 | } | |
4160 | ||
4161 | static int process_all_new_xattrs(struct send_ctx *sctx) | |
4162 | { | |
4163 | int ret; | |
4164 | struct btrfs_root *root; | |
4165 | struct btrfs_path *path; | |
4166 | struct btrfs_key key; | |
4167 | struct btrfs_key found_key; | |
4168 | struct extent_buffer *eb; | |
4169 | int slot; | |
4170 | ||
4171 | path = alloc_path_for_send(); | |
4172 | if (!path) | |
4173 | return -ENOMEM; | |
4174 | ||
4175 | root = sctx->send_root; | |
4176 | ||
4177 | key.objectid = sctx->cmp_key->objectid; | |
4178 | key.type = BTRFS_XATTR_ITEM_KEY; | |
4179 | key.offset = 0; | |
dff6d0ad FDBM |
4180 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4181 | if (ret < 0) | |
4182 | goto out; | |
31db9f7c | 4183 | |
dff6d0ad | 4184 | while (1) { |
31db9f7c AB |
4185 | eb = path->nodes[0]; |
4186 | slot = path->slots[0]; | |
dff6d0ad FDBM |
4187 | if (slot >= btrfs_header_nritems(eb)) { |
4188 | ret = btrfs_next_leaf(root, path); | |
4189 | if (ret < 0) { | |
4190 | goto out; | |
4191 | } else if (ret > 0) { | |
4192 | ret = 0; | |
4193 | break; | |
4194 | } | |
4195 | continue; | |
4196 | } | |
31db9f7c | 4197 | |
dff6d0ad | 4198 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
31db9f7c AB |
4199 | if (found_key.objectid != key.objectid || |
4200 | found_key.type != key.type) { | |
4201 | ret = 0; | |
4202 | goto out; | |
4203 | } | |
4204 | ||
924794c9 TI |
4205 | ret = iterate_dir_item(root, path, &found_key, |
4206 | __process_new_xattr, sctx); | |
31db9f7c AB |
4207 | if (ret < 0) |
4208 | goto out; | |
4209 | ||
dff6d0ad | 4210 | path->slots[0]++; |
31db9f7c AB |
4211 | } |
4212 | ||
4213 | out: | |
4214 | btrfs_free_path(path); | |
4215 | return ret; | |
4216 | } | |
4217 | ||
ed259095 JB |
4218 | static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len) |
4219 | { | |
4220 | struct btrfs_root *root = sctx->send_root; | |
4221 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4222 | struct inode *inode; | |
4223 | struct page *page; | |
4224 | char *addr; | |
4225 | struct btrfs_key key; | |
4226 | pgoff_t index = offset >> PAGE_CACHE_SHIFT; | |
4227 | pgoff_t last_index; | |
4228 | unsigned pg_offset = offset & ~PAGE_CACHE_MASK; | |
4229 | ssize_t ret = 0; | |
4230 | ||
4231 | key.objectid = sctx->cur_ino; | |
4232 | key.type = BTRFS_INODE_ITEM_KEY; | |
4233 | key.offset = 0; | |
4234 | ||
4235 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
4236 | if (IS_ERR(inode)) | |
4237 | return PTR_ERR(inode); | |
4238 | ||
4239 | if (offset + len > i_size_read(inode)) { | |
4240 | if (offset > i_size_read(inode)) | |
4241 | len = 0; | |
4242 | else | |
4243 | len = offset - i_size_read(inode); | |
4244 | } | |
4245 | if (len == 0) | |
4246 | goto out; | |
4247 | ||
4248 | last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT; | |
2131bcd3 LB |
4249 | |
4250 | /* initial readahead */ | |
4251 | memset(&sctx->ra, 0, sizeof(struct file_ra_state)); | |
4252 | file_ra_state_init(&sctx->ra, inode->i_mapping); | |
4253 | btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index, | |
4254 | last_index - index + 1); | |
4255 | ||
ed259095 JB |
4256 | while (index <= last_index) { |
4257 | unsigned cur_len = min_t(unsigned, len, | |
4258 | PAGE_CACHE_SIZE - pg_offset); | |
4259 | page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); | |
4260 | if (!page) { | |
4261 | ret = -ENOMEM; | |
4262 | break; | |
4263 | } | |
4264 | ||
4265 | if (!PageUptodate(page)) { | |
4266 | btrfs_readpage(NULL, page); | |
4267 | lock_page(page); | |
4268 | if (!PageUptodate(page)) { | |
4269 | unlock_page(page); | |
4270 | page_cache_release(page); | |
4271 | ret = -EIO; | |
4272 | break; | |
4273 | } | |
4274 | } | |
4275 | ||
4276 | addr = kmap(page); | |
4277 | memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len); | |
4278 | kunmap(page); | |
4279 | unlock_page(page); | |
4280 | page_cache_release(page); | |
4281 | index++; | |
4282 | pg_offset = 0; | |
4283 | len -= cur_len; | |
4284 | ret += cur_len; | |
4285 | } | |
4286 | out: | |
4287 | iput(inode); | |
4288 | return ret; | |
4289 | } | |
4290 | ||
31db9f7c AB |
4291 | /* |
4292 | * Read some bytes from the current inode/file and send a write command to | |
4293 | * user space. | |
4294 | */ | |
4295 | static int send_write(struct send_ctx *sctx, u64 offset, u32 len) | |
4296 | { | |
4297 | int ret = 0; | |
4298 | struct fs_path *p; | |
ed259095 | 4299 | ssize_t num_read = 0; |
31db9f7c | 4300 | |
924794c9 | 4301 | p = fs_path_alloc(); |
31db9f7c AB |
4302 | if (!p) |
4303 | return -ENOMEM; | |
4304 | ||
31db9f7c AB |
4305 | verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len); |
4306 | ||
ed259095 JB |
4307 | num_read = fill_read_buf(sctx, offset, len); |
4308 | if (num_read <= 0) { | |
4309 | if (num_read < 0) | |
4310 | ret = num_read; | |
31db9f7c | 4311 | goto out; |
ed259095 | 4312 | } |
31db9f7c AB |
4313 | |
4314 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
4315 | if (ret < 0) | |
4316 | goto out; | |
4317 | ||
4318 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4319 | if (ret < 0) | |
4320 | goto out; | |
4321 | ||
4322 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4323 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
e938c8ad | 4324 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read); |
31db9f7c AB |
4325 | |
4326 | ret = send_cmd(sctx); | |
4327 | ||
4328 | tlv_put_failure: | |
4329 | out: | |
924794c9 | 4330 | fs_path_free(p); |
31db9f7c AB |
4331 | if (ret < 0) |
4332 | return ret; | |
e938c8ad | 4333 | return num_read; |
31db9f7c AB |
4334 | } |
4335 | ||
4336 | /* | |
4337 | * Send a clone command to user space. | |
4338 | */ | |
4339 | static int send_clone(struct send_ctx *sctx, | |
4340 | u64 offset, u32 len, | |
4341 | struct clone_root *clone_root) | |
4342 | { | |
4343 | int ret = 0; | |
31db9f7c AB |
4344 | struct fs_path *p; |
4345 | u64 gen; | |
4346 | ||
4347 | verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, " | |
4348 | "clone_inode=%llu, clone_offset=%llu\n", offset, len, | |
4349 | clone_root->root->objectid, clone_root->ino, | |
4350 | clone_root->offset); | |
4351 | ||
924794c9 | 4352 | p = fs_path_alloc(); |
31db9f7c AB |
4353 | if (!p) |
4354 | return -ENOMEM; | |
4355 | ||
4356 | ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); | |
4357 | if (ret < 0) | |
4358 | goto out; | |
4359 | ||
4360 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4361 | if (ret < 0) | |
4362 | goto out; | |
4363 | ||
4364 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4365 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); | |
4366 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4367 | ||
e938c8ad | 4368 | if (clone_root->root == sctx->send_root) { |
31db9f7c | 4369 | ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, |
85a7b33b | 4370 | &gen, NULL, NULL, NULL, NULL); |
31db9f7c AB |
4371 | if (ret < 0) |
4372 | goto out; | |
4373 | ret = get_cur_path(sctx, clone_root->ino, gen, p); | |
4374 | } else { | |
924794c9 | 4375 | ret = get_inode_path(clone_root->root, clone_root->ino, p); |
31db9f7c AB |
4376 | } |
4377 | if (ret < 0) | |
4378 | goto out; | |
4379 | ||
4380 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
e938c8ad | 4381 | clone_root->root->root_item.uuid); |
31db9f7c | 4382 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
5a0f4e2c | 4383 | le64_to_cpu(clone_root->root->root_item.ctransid)); |
31db9f7c AB |
4384 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); |
4385 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, | |
4386 | clone_root->offset); | |
4387 | ||
4388 | ret = send_cmd(sctx); | |
4389 | ||
4390 | tlv_put_failure: | |
4391 | out: | |
924794c9 | 4392 | fs_path_free(p); |
31db9f7c AB |
4393 | return ret; |
4394 | } | |
4395 | ||
cb95e7bf MF |
4396 | /* |
4397 | * Send an update extent command to user space. | |
4398 | */ | |
4399 | static int send_update_extent(struct send_ctx *sctx, | |
4400 | u64 offset, u32 len) | |
4401 | { | |
4402 | int ret = 0; | |
4403 | struct fs_path *p; | |
4404 | ||
924794c9 | 4405 | p = fs_path_alloc(); |
cb95e7bf MF |
4406 | if (!p) |
4407 | return -ENOMEM; | |
4408 | ||
4409 | ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); | |
4410 | if (ret < 0) | |
4411 | goto out; | |
4412 | ||
4413 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4414 | if (ret < 0) | |
4415 | goto out; | |
4416 | ||
4417 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4418 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4419 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); | |
4420 | ||
4421 | ret = send_cmd(sctx); | |
4422 | ||
4423 | tlv_put_failure: | |
4424 | out: | |
924794c9 | 4425 | fs_path_free(p); |
cb95e7bf MF |
4426 | return ret; |
4427 | } | |
4428 | ||
16e7549f JB |
4429 | static int send_hole(struct send_ctx *sctx, u64 end) |
4430 | { | |
4431 | struct fs_path *p = NULL; | |
4432 | u64 offset = sctx->cur_inode_last_extent; | |
4433 | u64 len; | |
4434 | int ret = 0; | |
4435 | ||
4436 | p = fs_path_alloc(); | |
4437 | if (!p) | |
4438 | return -ENOMEM; | |
c715e155 FM |
4439 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); |
4440 | if (ret < 0) | |
4441 | goto tlv_put_failure; | |
16e7549f JB |
4442 | memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE); |
4443 | while (offset < end) { | |
4444 | len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE); | |
4445 | ||
4446 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
16e7549f JB |
4447 | if (ret < 0) |
4448 | break; | |
4449 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
4450 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
4451 | TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len); | |
4452 | ret = send_cmd(sctx); | |
4453 | if (ret < 0) | |
4454 | break; | |
4455 | offset += len; | |
4456 | } | |
4457 | tlv_put_failure: | |
4458 | fs_path_free(p); | |
4459 | return ret; | |
4460 | } | |
4461 | ||
31db9f7c AB |
4462 | static int send_write_or_clone(struct send_ctx *sctx, |
4463 | struct btrfs_path *path, | |
4464 | struct btrfs_key *key, | |
4465 | struct clone_root *clone_root) | |
4466 | { | |
4467 | int ret = 0; | |
4468 | struct btrfs_file_extent_item *ei; | |
4469 | u64 offset = key->offset; | |
4470 | u64 pos = 0; | |
4471 | u64 len; | |
4472 | u32 l; | |
4473 | u8 type; | |
28e5dd8f | 4474 | u64 bs = sctx->send_root->fs_info->sb->s_blocksize; |
31db9f7c AB |
4475 | |
4476 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4477 | struct btrfs_file_extent_item); | |
4478 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
74dd17fb | 4479 | if (type == BTRFS_FILE_EXTENT_INLINE) { |
514ac8ad CM |
4480 | len = btrfs_file_extent_inline_len(path->nodes[0], |
4481 | path->slots[0], ei); | |
74dd17fb CM |
4482 | /* |
4483 | * it is possible the inline item won't cover the whole page, | |
4484 | * but there may be items after this page. Make | |
4485 | * sure to send the whole thing | |
4486 | */ | |
4487 | len = PAGE_CACHE_ALIGN(len); | |
4488 | } else { | |
31db9f7c | 4489 | len = btrfs_file_extent_num_bytes(path->nodes[0], ei); |
74dd17fb | 4490 | } |
31db9f7c AB |
4491 | |
4492 | if (offset + len > sctx->cur_inode_size) | |
4493 | len = sctx->cur_inode_size - offset; | |
4494 | if (len == 0) { | |
4495 | ret = 0; | |
4496 | goto out; | |
4497 | } | |
4498 | ||
28e5dd8f | 4499 | if (clone_root && IS_ALIGNED(offset + len, bs)) { |
cb95e7bf MF |
4500 | ret = send_clone(sctx, offset, len, clone_root); |
4501 | } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) { | |
4502 | ret = send_update_extent(sctx, offset, len); | |
4503 | } else { | |
31db9f7c AB |
4504 | while (pos < len) { |
4505 | l = len - pos; | |
4506 | if (l > BTRFS_SEND_READ_SIZE) | |
4507 | l = BTRFS_SEND_READ_SIZE; | |
4508 | ret = send_write(sctx, pos + offset, l); | |
4509 | if (ret < 0) | |
4510 | goto out; | |
4511 | if (!ret) | |
4512 | break; | |
4513 | pos += ret; | |
4514 | } | |
4515 | ret = 0; | |
31db9f7c | 4516 | } |
31db9f7c AB |
4517 | out: |
4518 | return ret; | |
4519 | } | |
4520 | ||
4521 | static int is_extent_unchanged(struct send_ctx *sctx, | |
4522 | struct btrfs_path *left_path, | |
4523 | struct btrfs_key *ekey) | |
4524 | { | |
4525 | int ret = 0; | |
4526 | struct btrfs_key key; | |
4527 | struct btrfs_path *path = NULL; | |
4528 | struct extent_buffer *eb; | |
4529 | int slot; | |
4530 | struct btrfs_key found_key; | |
4531 | struct btrfs_file_extent_item *ei; | |
4532 | u64 left_disknr; | |
4533 | u64 right_disknr; | |
4534 | u64 left_offset; | |
4535 | u64 right_offset; | |
4536 | u64 left_offset_fixed; | |
4537 | u64 left_len; | |
4538 | u64 right_len; | |
74dd17fb CM |
4539 | u64 left_gen; |
4540 | u64 right_gen; | |
31db9f7c AB |
4541 | u8 left_type; |
4542 | u8 right_type; | |
4543 | ||
4544 | path = alloc_path_for_send(); | |
4545 | if (!path) | |
4546 | return -ENOMEM; | |
4547 | ||
4548 | eb = left_path->nodes[0]; | |
4549 | slot = left_path->slots[0]; | |
31db9f7c AB |
4550 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
4551 | left_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
4552 | |
4553 | if (left_type != BTRFS_FILE_EXTENT_REG) { | |
4554 | ret = 0; | |
4555 | goto out; | |
4556 | } | |
74dd17fb CM |
4557 | left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
4558 | left_len = btrfs_file_extent_num_bytes(eb, ei); | |
4559 | left_offset = btrfs_file_extent_offset(eb, ei); | |
4560 | left_gen = btrfs_file_extent_generation(eb, ei); | |
31db9f7c AB |
4561 | |
4562 | /* | |
4563 | * Following comments will refer to these graphics. L is the left | |
4564 | * extents which we are checking at the moment. 1-8 are the right | |
4565 | * extents that we iterate. | |
4566 | * | |
4567 | * |-----L-----| | |
4568 | * |-1-|-2a-|-3-|-4-|-5-|-6-| | |
4569 | * | |
4570 | * |-----L-----| | |
4571 | * |--1--|-2b-|...(same as above) | |
4572 | * | |
4573 | * Alternative situation. Happens on files where extents got split. | |
4574 | * |-----L-----| | |
4575 | * |-----------7-----------|-6-| | |
4576 | * | |
4577 | * Alternative situation. Happens on files which got larger. | |
4578 | * |-----L-----| | |
4579 | * |-8-| | |
4580 | * Nothing follows after 8. | |
4581 | */ | |
4582 | ||
4583 | key.objectid = ekey->objectid; | |
4584 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4585 | key.offset = ekey->offset; | |
4586 | ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); | |
4587 | if (ret < 0) | |
4588 | goto out; | |
4589 | if (ret) { | |
4590 | ret = 0; | |
4591 | goto out; | |
4592 | } | |
4593 | ||
4594 | /* | |
4595 | * Handle special case where the right side has no extents at all. | |
4596 | */ | |
4597 | eb = path->nodes[0]; | |
4598 | slot = path->slots[0]; | |
4599 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
4600 | if (found_key.objectid != key.objectid || | |
4601 | found_key.type != key.type) { | |
57cfd462 JB |
4602 | /* If we're a hole then just pretend nothing changed */ |
4603 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
4604 | goto out; |
4605 | } | |
4606 | ||
4607 | /* | |
4608 | * We're now on 2a, 2b or 7. | |
4609 | */ | |
4610 | key = found_key; | |
4611 | while (key.offset < ekey->offset + left_len) { | |
4612 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
4613 | right_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
4614 | if (right_type != BTRFS_FILE_EXTENT_REG) { |
4615 | ret = 0; | |
4616 | goto out; | |
4617 | } | |
4618 | ||
007d31f7 JB |
4619 | right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
4620 | right_len = btrfs_file_extent_num_bytes(eb, ei); | |
4621 | right_offset = btrfs_file_extent_offset(eb, ei); | |
4622 | right_gen = btrfs_file_extent_generation(eb, ei); | |
4623 | ||
31db9f7c AB |
4624 | /* |
4625 | * Are we at extent 8? If yes, we know the extent is changed. | |
4626 | * This may only happen on the first iteration. | |
4627 | */ | |
d8347fa4 | 4628 | if (found_key.offset + right_len <= ekey->offset) { |
57cfd462 JB |
4629 | /* If we're a hole just pretend nothing changed */ |
4630 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
4631 | goto out; |
4632 | } | |
4633 | ||
4634 | left_offset_fixed = left_offset; | |
4635 | if (key.offset < ekey->offset) { | |
4636 | /* Fix the right offset for 2a and 7. */ | |
4637 | right_offset += ekey->offset - key.offset; | |
4638 | } else { | |
4639 | /* Fix the left offset for all behind 2a and 2b */ | |
4640 | left_offset_fixed += key.offset - ekey->offset; | |
4641 | } | |
4642 | ||
4643 | /* | |
4644 | * Check if we have the same extent. | |
4645 | */ | |
3954096d | 4646 | if (left_disknr != right_disknr || |
74dd17fb CM |
4647 | left_offset_fixed != right_offset || |
4648 | left_gen != right_gen) { | |
31db9f7c AB |
4649 | ret = 0; |
4650 | goto out; | |
4651 | } | |
4652 | ||
4653 | /* | |
4654 | * Go to the next extent. | |
4655 | */ | |
4656 | ret = btrfs_next_item(sctx->parent_root, path); | |
4657 | if (ret < 0) | |
4658 | goto out; | |
4659 | if (!ret) { | |
4660 | eb = path->nodes[0]; | |
4661 | slot = path->slots[0]; | |
4662 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
4663 | } | |
4664 | if (ret || found_key.objectid != key.objectid || | |
4665 | found_key.type != key.type) { | |
4666 | key.offset += right_len; | |
4667 | break; | |
adaa4b8e JS |
4668 | } |
4669 | if (found_key.offset != key.offset + right_len) { | |
4670 | ret = 0; | |
4671 | goto out; | |
31db9f7c AB |
4672 | } |
4673 | key = found_key; | |
4674 | } | |
4675 | ||
4676 | /* | |
4677 | * We're now behind the left extent (treat as unchanged) or at the end | |
4678 | * of the right side (treat as changed). | |
4679 | */ | |
4680 | if (key.offset >= ekey->offset + left_len) | |
4681 | ret = 1; | |
4682 | else | |
4683 | ret = 0; | |
4684 | ||
4685 | ||
4686 | out: | |
4687 | btrfs_free_path(path); | |
4688 | return ret; | |
4689 | } | |
4690 | ||
16e7549f JB |
4691 | static int get_last_extent(struct send_ctx *sctx, u64 offset) |
4692 | { | |
4693 | struct btrfs_path *path; | |
4694 | struct btrfs_root *root = sctx->send_root; | |
4695 | struct btrfs_file_extent_item *fi; | |
4696 | struct btrfs_key key; | |
4697 | u64 extent_end; | |
4698 | u8 type; | |
4699 | int ret; | |
4700 | ||
4701 | path = alloc_path_for_send(); | |
4702 | if (!path) | |
4703 | return -ENOMEM; | |
4704 | ||
4705 | sctx->cur_inode_last_extent = 0; | |
4706 | ||
4707 | key.objectid = sctx->cur_ino; | |
4708 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4709 | key.offset = offset; | |
4710 | ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); | |
4711 | if (ret < 0) | |
4712 | goto out; | |
4713 | ret = 0; | |
4714 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
4715 | if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
4716 | goto out; | |
4717 | ||
4718 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4719 | struct btrfs_file_extent_item); | |
4720 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
4721 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
4722 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
4723 | path->slots[0], fi); | |
16e7549f JB |
4724 | extent_end = ALIGN(key.offset + size, |
4725 | sctx->send_root->sectorsize); | |
4726 | } else { | |
4727 | extent_end = key.offset + | |
4728 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
4729 | } | |
4730 | sctx->cur_inode_last_extent = extent_end; | |
4731 | out: | |
4732 | btrfs_free_path(path); | |
4733 | return ret; | |
4734 | } | |
4735 | ||
4736 | static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, | |
4737 | struct btrfs_key *key) | |
4738 | { | |
4739 | struct btrfs_file_extent_item *fi; | |
4740 | u64 extent_end; | |
4741 | u8 type; | |
4742 | int ret = 0; | |
4743 | ||
4744 | if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) | |
4745 | return 0; | |
4746 | ||
4747 | if (sctx->cur_inode_last_extent == (u64)-1) { | |
4748 | ret = get_last_extent(sctx, key->offset - 1); | |
4749 | if (ret) | |
4750 | return ret; | |
4751 | } | |
4752 | ||
4753 | fi = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4754 | struct btrfs_file_extent_item); | |
4755 | type = btrfs_file_extent_type(path->nodes[0], fi); | |
4756 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
514ac8ad CM |
4757 | u64 size = btrfs_file_extent_inline_len(path->nodes[0], |
4758 | path->slots[0], fi); | |
16e7549f JB |
4759 | extent_end = ALIGN(key->offset + size, |
4760 | sctx->send_root->sectorsize); | |
4761 | } else { | |
4762 | extent_end = key->offset + | |
4763 | btrfs_file_extent_num_bytes(path->nodes[0], fi); | |
4764 | } | |
bf54f412 FDBM |
4765 | |
4766 | if (path->slots[0] == 0 && | |
4767 | sctx->cur_inode_last_extent < key->offset) { | |
4768 | /* | |
4769 | * We might have skipped entire leafs that contained only | |
4770 | * file extent items for our current inode. These leafs have | |
4771 | * a generation number smaller (older) than the one in the | |
4772 | * current leaf and the leaf our last extent came from, and | |
4773 | * are located between these 2 leafs. | |
4774 | */ | |
4775 | ret = get_last_extent(sctx, key->offset - 1); | |
4776 | if (ret) | |
4777 | return ret; | |
4778 | } | |
4779 | ||
16e7549f JB |
4780 | if (sctx->cur_inode_last_extent < key->offset) |
4781 | ret = send_hole(sctx, key->offset); | |
4782 | sctx->cur_inode_last_extent = extent_end; | |
4783 | return ret; | |
4784 | } | |
4785 | ||
31db9f7c AB |
4786 | static int process_extent(struct send_ctx *sctx, |
4787 | struct btrfs_path *path, | |
4788 | struct btrfs_key *key) | |
4789 | { | |
31db9f7c | 4790 | struct clone_root *found_clone = NULL; |
57cfd462 | 4791 | int ret = 0; |
31db9f7c AB |
4792 | |
4793 | if (S_ISLNK(sctx->cur_inode_mode)) | |
4794 | return 0; | |
4795 | ||
4796 | if (sctx->parent_root && !sctx->cur_inode_new) { | |
4797 | ret = is_extent_unchanged(sctx, path, key); | |
4798 | if (ret < 0) | |
4799 | goto out; | |
4800 | if (ret) { | |
4801 | ret = 0; | |
16e7549f | 4802 | goto out_hole; |
31db9f7c | 4803 | } |
57cfd462 JB |
4804 | } else { |
4805 | struct btrfs_file_extent_item *ei; | |
4806 | u8 type; | |
4807 | ||
4808 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
4809 | struct btrfs_file_extent_item); | |
4810 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
4811 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
4812 | type == BTRFS_FILE_EXTENT_REG) { | |
4813 | /* | |
4814 | * The send spec does not have a prealloc command yet, | |
4815 | * so just leave a hole for prealloc'ed extents until | |
4816 | * we have enough commands queued up to justify rev'ing | |
4817 | * the send spec. | |
4818 | */ | |
4819 | if (type == BTRFS_FILE_EXTENT_PREALLOC) { | |
4820 | ret = 0; | |
4821 | goto out; | |
4822 | } | |
4823 | ||
4824 | /* Have a hole, just skip it. */ | |
4825 | if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { | |
4826 | ret = 0; | |
4827 | goto out; | |
4828 | } | |
4829 | } | |
31db9f7c AB |
4830 | } |
4831 | ||
4832 | ret = find_extent_clone(sctx, path, key->objectid, key->offset, | |
4833 | sctx->cur_inode_size, &found_clone); | |
4834 | if (ret != -ENOENT && ret < 0) | |
4835 | goto out; | |
4836 | ||
4837 | ret = send_write_or_clone(sctx, path, key, found_clone); | |
16e7549f JB |
4838 | if (ret) |
4839 | goto out; | |
4840 | out_hole: | |
4841 | ret = maybe_send_hole(sctx, path, key); | |
31db9f7c AB |
4842 | out: |
4843 | return ret; | |
4844 | } | |
4845 | ||
4846 | static int process_all_extents(struct send_ctx *sctx) | |
4847 | { | |
4848 | int ret; | |
4849 | struct btrfs_root *root; | |
4850 | struct btrfs_path *path; | |
4851 | struct btrfs_key key; | |
4852 | struct btrfs_key found_key; | |
4853 | struct extent_buffer *eb; | |
4854 | int slot; | |
4855 | ||
4856 | root = sctx->send_root; | |
4857 | path = alloc_path_for_send(); | |
4858 | if (!path) | |
4859 | return -ENOMEM; | |
4860 | ||
4861 | key.objectid = sctx->cmp_key->objectid; | |
4862 | key.type = BTRFS_EXTENT_DATA_KEY; | |
4863 | key.offset = 0; | |
7fdd29d0 FDBM |
4864 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4865 | if (ret < 0) | |
4866 | goto out; | |
31db9f7c | 4867 | |
7fdd29d0 | 4868 | while (1) { |
31db9f7c AB |
4869 | eb = path->nodes[0]; |
4870 | slot = path->slots[0]; | |
7fdd29d0 FDBM |
4871 | |
4872 | if (slot >= btrfs_header_nritems(eb)) { | |
4873 | ret = btrfs_next_leaf(root, path); | |
4874 | if (ret < 0) { | |
4875 | goto out; | |
4876 | } else if (ret > 0) { | |
4877 | ret = 0; | |
4878 | break; | |
4879 | } | |
4880 | continue; | |
4881 | } | |
4882 | ||
31db9f7c AB |
4883 | btrfs_item_key_to_cpu(eb, &found_key, slot); |
4884 | ||
4885 | if (found_key.objectid != key.objectid || | |
4886 | found_key.type != key.type) { | |
4887 | ret = 0; | |
4888 | goto out; | |
4889 | } | |
4890 | ||
4891 | ret = process_extent(sctx, path, &found_key); | |
4892 | if (ret < 0) | |
4893 | goto out; | |
4894 | ||
7fdd29d0 | 4895 | path->slots[0]++; |
31db9f7c AB |
4896 | } |
4897 | ||
4898 | out: | |
4899 | btrfs_free_path(path); | |
4900 | return ret; | |
4901 | } | |
4902 | ||
9f03740a FDBM |
4903 | static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, |
4904 | int *pending_move, | |
4905 | int *refs_processed) | |
31db9f7c AB |
4906 | { |
4907 | int ret = 0; | |
4908 | ||
4909 | if (sctx->cur_ino == 0) | |
4910 | goto out; | |
4911 | if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && | |
96b5bd77 | 4912 | sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) |
31db9f7c AB |
4913 | goto out; |
4914 | if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) | |
4915 | goto out; | |
4916 | ||
9f03740a | 4917 | ret = process_recorded_refs(sctx, pending_move); |
e479d9bb AB |
4918 | if (ret < 0) |
4919 | goto out; | |
4920 | ||
9f03740a | 4921 | *refs_processed = 1; |
31db9f7c AB |
4922 | out: |
4923 | return ret; | |
4924 | } | |
4925 | ||
4926 | static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) | |
4927 | { | |
4928 | int ret = 0; | |
4929 | u64 left_mode; | |
4930 | u64 left_uid; | |
4931 | u64 left_gid; | |
4932 | u64 right_mode; | |
4933 | u64 right_uid; | |
4934 | u64 right_gid; | |
4935 | int need_chmod = 0; | |
4936 | int need_chown = 0; | |
9f03740a FDBM |
4937 | int pending_move = 0; |
4938 | int refs_processed = 0; | |
31db9f7c | 4939 | |
9f03740a FDBM |
4940 | ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, |
4941 | &refs_processed); | |
31db9f7c AB |
4942 | if (ret < 0) |
4943 | goto out; | |
4944 | ||
9f03740a FDBM |
4945 | /* |
4946 | * We have processed the refs and thus need to advance send_progress. | |
4947 | * Now, calls to get_cur_xxx will take the updated refs of the current | |
4948 | * inode into account. | |
4949 | * | |
4950 | * On the other hand, if our current inode is a directory and couldn't | |
4951 | * be moved/renamed because its parent was renamed/moved too and it has | |
4952 | * a higher inode number, we can only move/rename our current inode | |
4953 | * after we moved/renamed its parent. Therefore in this case operate on | |
4954 | * the old path (pre move/rename) of our current inode, and the | |
4955 | * move/rename will be performed later. | |
4956 | */ | |
4957 | if (refs_processed && !pending_move) | |
4958 | sctx->send_progress = sctx->cur_ino + 1; | |
4959 | ||
31db9f7c AB |
4960 | if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) |
4961 | goto out; | |
4962 | if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) | |
4963 | goto out; | |
4964 | ||
4965 | ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, | |
85a7b33b | 4966 | &left_mode, &left_uid, &left_gid, NULL); |
31db9f7c AB |
4967 | if (ret < 0) |
4968 | goto out; | |
4969 | ||
e2d044fe AL |
4970 | if (!sctx->parent_root || sctx->cur_inode_new) { |
4971 | need_chown = 1; | |
4972 | if (!S_ISLNK(sctx->cur_inode_mode)) | |
31db9f7c | 4973 | need_chmod = 1; |
e2d044fe AL |
4974 | } else { |
4975 | ret = get_inode_info(sctx->parent_root, sctx->cur_ino, | |
4976 | NULL, NULL, &right_mode, &right_uid, | |
4977 | &right_gid, NULL); | |
4978 | if (ret < 0) | |
4979 | goto out; | |
31db9f7c | 4980 | |
e2d044fe AL |
4981 | if (left_uid != right_uid || left_gid != right_gid) |
4982 | need_chown = 1; | |
4983 | if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) | |
4984 | need_chmod = 1; | |
31db9f7c AB |
4985 | } |
4986 | ||
4987 | if (S_ISREG(sctx->cur_inode_mode)) { | |
16e7549f | 4988 | if (need_send_hole(sctx)) { |
766b5e5a FM |
4989 | if (sctx->cur_inode_last_extent == (u64)-1 || |
4990 | sctx->cur_inode_last_extent < | |
4991 | sctx->cur_inode_size) { | |
16e7549f JB |
4992 | ret = get_last_extent(sctx, (u64)-1); |
4993 | if (ret) | |
4994 | goto out; | |
4995 | } | |
4996 | if (sctx->cur_inode_last_extent < | |
4997 | sctx->cur_inode_size) { | |
4998 | ret = send_hole(sctx, sctx->cur_inode_size); | |
4999 | if (ret) | |
5000 | goto out; | |
5001 | } | |
5002 | } | |
31db9f7c AB |
5003 | ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen, |
5004 | sctx->cur_inode_size); | |
5005 | if (ret < 0) | |
5006 | goto out; | |
5007 | } | |
5008 | ||
5009 | if (need_chown) { | |
5010 | ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5011 | left_uid, left_gid); | |
5012 | if (ret < 0) | |
5013 | goto out; | |
5014 | } | |
5015 | if (need_chmod) { | |
5016 | ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
5017 | left_mode); | |
5018 | if (ret < 0) | |
5019 | goto out; | |
5020 | } | |
5021 | ||
5022 | /* | |
9f03740a FDBM |
5023 | * If other directory inodes depended on our current directory |
5024 | * inode's move/rename, now do their move/rename operations. | |
31db9f7c | 5025 | */ |
9f03740a FDBM |
5026 | if (!is_waiting_for_move(sctx, sctx->cur_ino)) { |
5027 | ret = apply_children_dir_moves(sctx); | |
5028 | if (ret) | |
5029 | goto out; | |
fcbd2154 FM |
5030 | /* |
5031 | * Need to send that every time, no matter if it actually | |
5032 | * changed between the two trees as we have done changes to | |
5033 | * the inode before. If our inode is a directory and it's | |
5034 | * waiting to be moved/renamed, we will send its utimes when | |
5035 | * it's moved/renamed, therefore we don't need to do it here. | |
5036 | */ | |
5037 | sctx->send_progress = sctx->cur_ino + 1; | |
5038 | ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
5039 | if (ret < 0) | |
5040 | goto out; | |
9f03740a FDBM |
5041 | } |
5042 | ||
31db9f7c AB |
5043 | out: |
5044 | return ret; | |
5045 | } | |
5046 | ||
5047 | static int changed_inode(struct send_ctx *sctx, | |
5048 | enum btrfs_compare_tree_result result) | |
5049 | { | |
5050 | int ret = 0; | |
5051 | struct btrfs_key *key = sctx->cmp_key; | |
5052 | struct btrfs_inode_item *left_ii = NULL; | |
5053 | struct btrfs_inode_item *right_ii = NULL; | |
5054 | u64 left_gen = 0; | |
5055 | u64 right_gen = 0; | |
5056 | ||
31db9f7c AB |
5057 | sctx->cur_ino = key->objectid; |
5058 | sctx->cur_inode_new_gen = 0; | |
16e7549f | 5059 | sctx->cur_inode_last_extent = (u64)-1; |
e479d9bb AB |
5060 | |
5061 | /* | |
5062 | * Set send_progress to current inode. This will tell all get_cur_xxx | |
5063 | * functions that the current inode's refs are not updated yet. Later, | |
5064 | * when process_recorded_refs is finished, it is set to cur_ino + 1. | |
5065 | */ | |
31db9f7c AB |
5066 | sctx->send_progress = sctx->cur_ino; |
5067 | ||
5068 | if (result == BTRFS_COMPARE_TREE_NEW || | |
5069 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
5070 | left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], | |
5071 | sctx->left_path->slots[0], | |
5072 | struct btrfs_inode_item); | |
5073 | left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], | |
5074 | left_ii); | |
5075 | } else { | |
5076 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
5077 | sctx->right_path->slots[0], | |
5078 | struct btrfs_inode_item); | |
5079 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
5080 | right_ii); | |
5081 | } | |
5082 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
5083 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
5084 | sctx->right_path->slots[0], | |
5085 | struct btrfs_inode_item); | |
5086 | ||
5087 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
5088 | right_ii); | |
6d85ed05 AB |
5089 | |
5090 | /* | |
5091 | * The cur_ino = root dir case is special here. We can't treat | |
5092 | * the inode as deleted+reused because it would generate a | |
5093 | * stream that tries to delete/mkdir the root dir. | |
5094 | */ | |
5095 | if (left_gen != right_gen && | |
5096 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) | |
31db9f7c AB |
5097 | sctx->cur_inode_new_gen = 1; |
5098 | } | |
5099 | ||
5100 | if (result == BTRFS_COMPARE_TREE_NEW) { | |
5101 | sctx->cur_inode_gen = left_gen; | |
5102 | sctx->cur_inode_new = 1; | |
5103 | sctx->cur_inode_deleted = 0; | |
5104 | sctx->cur_inode_size = btrfs_inode_size( | |
5105 | sctx->left_path->nodes[0], left_ii); | |
5106 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5107 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
5108 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
5109 | sctx->left_path->nodes[0], left_ii); | |
31db9f7c | 5110 | if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) |
1f4692da | 5111 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
5112 | } else if (result == BTRFS_COMPARE_TREE_DELETED) { |
5113 | sctx->cur_inode_gen = right_gen; | |
5114 | sctx->cur_inode_new = 0; | |
5115 | sctx->cur_inode_deleted = 1; | |
5116 | sctx->cur_inode_size = btrfs_inode_size( | |
5117 | sctx->right_path->nodes[0], right_ii); | |
5118 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5119 | sctx->right_path->nodes[0], right_ii); | |
5120 | } else if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
766702ef AB |
5121 | /* |
5122 | * We need to do some special handling in case the inode was | |
5123 | * reported as changed with a changed generation number. This | |
5124 | * means that the original inode was deleted and new inode | |
5125 | * reused the same inum. So we have to treat the old inode as | |
5126 | * deleted and the new one as new. | |
5127 | */ | |
31db9f7c | 5128 | if (sctx->cur_inode_new_gen) { |
766702ef AB |
5129 | /* |
5130 | * First, process the inode as if it was deleted. | |
5131 | */ | |
31db9f7c AB |
5132 | sctx->cur_inode_gen = right_gen; |
5133 | sctx->cur_inode_new = 0; | |
5134 | sctx->cur_inode_deleted = 1; | |
5135 | sctx->cur_inode_size = btrfs_inode_size( | |
5136 | sctx->right_path->nodes[0], right_ii); | |
5137 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5138 | sctx->right_path->nodes[0], right_ii); | |
5139 | ret = process_all_refs(sctx, | |
5140 | BTRFS_COMPARE_TREE_DELETED); | |
5141 | if (ret < 0) | |
5142 | goto out; | |
5143 | ||
766702ef AB |
5144 | /* |
5145 | * Now process the inode as if it was new. | |
5146 | */ | |
31db9f7c AB |
5147 | sctx->cur_inode_gen = left_gen; |
5148 | sctx->cur_inode_new = 1; | |
5149 | sctx->cur_inode_deleted = 0; | |
5150 | sctx->cur_inode_size = btrfs_inode_size( | |
5151 | sctx->left_path->nodes[0], left_ii); | |
5152 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5153 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
5154 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
5155 | sctx->left_path->nodes[0], left_ii); | |
1f4692da | 5156 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
5157 | if (ret < 0) |
5158 | goto out; | |
5159 | ||
5160 | ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); | |
5161 | if (ret < 0) | |
5162 | goto out; | |
e479d9bb AB |
5163 | /* |
5164 | * Advance send_progress now as we did not get into | |
5165 | * process_recorded_refs_if_needed in the new_gen case. | |
5166 | */ | |
5167 | sctx->send_progress = sctx->cur_ino + 1; | |
766702ef AB |
5168 | |
5169 | /* | |
5170 | * Now process all extents and xattrs of the inode as if | |
5171 | * they were all new. | |
5172 | */ | |
31db9f7c AB |
5173 | ret = process_all_extents(sctx); |
5174 | if (ret < 0) | |
5175 | goto out; | |
5176 | ret = process_all_new_xattrs(sctx); | |
5177 | if (ret < 0) | |
5178 | goto out; | |
5179 | } else { | |
5180 | sctx->cur_inode_gen = left_gen; | |
5181 | sctx->cur_inode_new = 0; | |
5182 | sctx->cur_inode_new_gen = 0; | |
5183 | sctx->cur_inode_deleted = 0; | |
5184 | sctx->cur_inode_size = btrfs_inode_size( | |
5185 | sctx->left_path->nodes[0], left_ii); | |
5186 | sctx->cur_inode_mode = btrfs_inode_mode( | |
5187 | sctx->left_path->nodes[0], left_ii); | |
5188 | } | |
5189 | } | |
5190 | ||
5191 | out: | |
5192 | return ret; | |
5193 | } | |
5194 | ||
766702ef AB |
5195 | /* |
5196 | * We have to process new refs before deleted refs, but compare_trees gives us | |
5197 | * the new and deleted refs mixed. To fix this, we record the new/deleted refs | |
5198 | * first and later process them in process_recorded_refs. | |
5199 | * For the cur_inode_new_gen case, we skip recording completely because | |
5200 | * changed_inode did already initiate processing of refs. The reason for this is | |
5201 | * that in this case, compare_tree actually compares the refs of 2 different | |
5202 | * inodes. To fix this, process_all_refs is used in changed_inode to handle all | |
5203 | * refs of the right tree as deleted and all refs of the left tree as new. | |
5204 | */ | |
31db9f7c AB |
5205 | static int changed_ref(struct send_ctx *sctx, |
5206 | enum btrfs_compare_tree_result result) | |
5207 | { | |
5208 | int ret = 0; | |
5209 | ||
5210 | BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid); | |
5211 | ||
5212 | if (!sctx->cur_inode_new_gen && | |
5213 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { | |
5214 | if (result == BTRFS_COMPARE_TREE_NEW) | |
5215 | ret = record_new_ref(sctx); | |
5216 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
5217 | ret = record_deleted_ref(sctx); | |
5218 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
5219 | ret = record_changed_ref(sctx); | |
5220 | } | |
5221 | ||
5222 | return ret; | |
5223 | } | |
5224 | ||
766702ef AB |
5225 | /* |
5226 | * Process new/deleted/changed xattrs. We skip processing in the | |
5227 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
5228 | * of xattrs. The reason is the same as in changed_ref | |
5229 | */ | |
31db9f7c AB |
5230 | static int changed_xattr(struct send_ctx *sctx, |
5231 | enum btrfs_compare_tree_result result) | |
5232 | { | |
5233 | int ret = 0; | |
5234 | ||
5235 | BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid); | |
5236 | ||
5237 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
5238 | if (result == BTRFS_COMPARE_TREE_NEW) | |
5239 | ret = process_new_xattr(sctx); | |
5240 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
5241 | ret = process_deleted_xattr(sctx); | |
5242 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
5243 | ret = process_changed_xattr(sctx); | |
5244 | } | |
5245 | ||
5246 | return ret; | |
5247 | } | |
5248 | ||
766702ef AB |
5249 | /* |
5250 | * Process new/deleted/changed extents. We skip processing in the | |
5251 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
5252 | * of extents. The reason is the same as in changed_ref | |
5253 | */ | |
31db9f7c AB |
5254 | static int changed_extent(struct send_ctx *sctx, |
5255 | enum btrfs_compare_tree_result result) | |
5256 | { | |
5257 | int ret = 0; | |
5258 | ||
5259 | BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid); | |
5260 | ||
5261 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
5262 | if (result != BTRFS_COMPARE_TREE_DELETED) | |
5263 | ret = process_extent(sctx, sctx->left_path, | |
5264 | sctx->cmp_key); | |
5265 | } | |
5266 | ||
5267 | return ret; | |
5268 | } | |
5269 | ||
ba5e8f2e JB |
5270 | static int dir_changed(struct send_ctx *sctx, u64 dir) |
5271 | { | |
5272 | u64 orig_gen, new_gen; | |
5273 | int ret; | |
5274 | ||
5275 | ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, | |
5276 | NULL, NULL); | |
5277 | if (ret) | |
5278 | return ret; | |
5279 | ||
5280 | ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, | |
5281 | NULL, NULL, NULL); | |
5282 | if (ret) | |
5283 | return ret; | |
5284 | ||
5285 | return (orig_gen != new_gen) ? 1 : 0; | |
5286 | } | |
5287 | ||
5288 | static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, | |
5289 | struct btrfs_key *key) | |
5290 | { | |
5291 | struct btrfs_inode_extref *extref; | |
5292 | struct extent_buffer *leaf; | |
5293 | u64 dirid = 0, last_dirid = 0; | |
5294 | unsigned long ptr; | |
5295 | u32 item_size; | |
5296 | u32 cur_offset = 0; | |
5297 | int ref_name_len; | |
5298 | int ret = 0; | |
5299 | ||
5300 | /* Easy case, just check this one dirid */ | |
5301 | if (key->type == BTRFS_INODE_REF_KEY) { | |
5302 | dirid = key->offset; | |
5303 | ||
5304 | ret = dir_changed(sctx, dirid); | |
5305 | goto out; | |
5306 | } | |
5307 | ||
5308 | leaf = path->nodes[0]; | |
5309 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5310 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
5311 | while (cur_offset < item_size) { | |
5312 | extref = (struct btrfs_inode_extref *)(ptr + | |
5313 | cur_offset); | |
5314 | dirid = btrfs_inode_extref_parent(leaf, extref); | |
5315 | ref_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
5316 | cur_offset += ref_name_len + sizeof(*extref); | |
5317 | if (dirid == last_dirid) | |
5318 | continue; | |
5319 | ret = dir_changed(sctx, dirid); | |
5320 | if (ret) | |
5321 | break; | |
5322 | last_dirid = dirid; | |
5323 | } | |
5324 | out: | |
5325 | return ret; | |
5326 | } | |
5327 | ||
766702ef AB |
5328 | /* |
5329 | * Updates compare related fields in sctx and simply forwards to the actual | |
5330 | * changed_xxx functions. | |
5331 | */ | |
31db9f7c AB |
5332 | static int changed_cb(struct btrfs_root *left_root, |
5333 | struct btrfs_root *right_root, | |
5334 | struct btrfs_path *left_path, | |
5335 | struct btrfs_path *right_path, | |
5336 | struct btrfs_key *key, | |
5337 | enum btrfs_compare_tree_result result, | |
5338 | void *ctx) | |
5339 | { | |
5340 | int ret = 0; | |
5341 | struct send_ctx *sctx = ctx; | |
5342 | ||
ba5e8f2e | 5343 | if (result == BTRFS_COMPARE_TREE_SAME) { |
16e7549f JB |
5344 | if (key->type == BTRFS_INODE_REF_KEY || |
5345 | key->type == BTRFS_INODE_EXTREF_KEY) { | |
5346 | ret = compare_refs(sctx, left_path, key); | |
5347 | if (!ret) | |
5348 | return 0; | |
5349 | if (ret < 0) | |
5350 | return ret; | |
5351 | } else if (key->type == BTRFS_EXTENT_DATA_KEY) { | |
5352 | return maybe_send_hole(sctx, left_path, key); | |
5353 | } else { | |
ba5e8f2e | 5354 | return 0; |
16e7549f | 5355 | } |
ba5e8f2e JB |
5356 | result = BTRFS_COMPARE_TREE_CHANGED; |
5357 | ret = 0; | |
5358 | } | |
5359 | ||
31db9f7c AB |
5360 | sctx->left_path = left_path; |
5361 | sctx->right_path = right_path; | |
5362 | sctx->cmp_key = key; | |
5363 | ||
5364 | ret = finish_inode_if_needed(sctx, 0); | |
5365 | if (ret < 0) | |
5366 | goto out; | |
5367 | ||
2981e225 AB |
5368 | /* Ignore non-FS objects */ |
5369 | if (key->objectid == BTRFS_FREE_INO_OBJECTID || | |
5370 | key->objectid == BTRFS_FREE_SPACE_OBJECTID) | |
5371 | goto out; | |
5372 | ||
31db9f7c AB |
5373 | if (key->type == BTRFS_INODE_ITEM_KEY) |
5374 | ret = changed_inode(sctx, result); | |
96b5bd77 JS |
5375 | else if (key->type == BTRFS_INODE_REF_KEY || |
5376 | key->type == BTRFS_INODE_EXTREF_KEY) | |
31db9f7c AB |
5377 | ret = changed_ref(sctx, result); |
5378 | else if (key->type == BTRFS_XATTR_ITEM_KEY) | |
5379 | ret = changed_xattr(sctx, result); | |
5380 | else if (key->type == BTRFS_EXTENT_DATA_KEY) | |
5381 | ret = changed_extent(sctx, result); | |
5382 | ||
5383 | out: | |
5384 | return ret; | |
5385 | } | |
5386 | ||
5387 | static int full_send_tree(struct send_ctx *sctx) | |
5388 | { | |
5389 | int ret; | |
31db9f7c AB |
5390 | struct btrfs_root *send_root = sctx->send_root; |
5391 | struct btrfs_key key; | |
5392 | struct btrfs_key found_key; | |
5393 | struct btrfs_path *path; | |
5394 | struct extent_buffer *eb; | |
5395 | int slot; | |
31db9f7c AB |
5396 | |
5397 | path = alloc_path_for_send(); | |
5398 | if (!path) | |
5399 | return -ENOMEM; | |
5400 | ||
31db9f7c AB |
5401 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; |
5402 | key.type = BTRFS_INODE_ITEM_KEY; | |
5403 | key.offset = 0; | |
5404 | ||
31db9f7c AB |
5405 | ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); |
5406 | if (ret < 0) | |
5407 | goto out; | |
5408 | if (ret) | |
5409 | goto out_finish; | |
5410 | ||
5411 | while (1) { | |
31db9f7c AB |
5412 | eb = path->nodes[0]; |
5413 | slot = path->slots[0]; | |
5414 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
5415 | ||
5416 | ret = changed_cb(send_root, NULL, path, NULL, | |
5417 | &found_key, BTRFS_COMPARE_TREE_NEW, sctx); | |
5418 | if (ret < 0) | |
5419 | goto out; | |
5420 | ||
5421 | key.objectid = found_key.objectid; | |
5422 | key.type = found_key.type; | |
5423 | key.offset = found_key.offset + 1; | |
5424 | ||
5425 | ret = btrfs_next_item(send_root, path); | |
5426 | if (ret < 0) | |
5427 | goto out; | |
5428 | if (ret) { | |
5429 | ret = 0; | |
5430 | break; | |
5431 | } | |
5432 | } | |
5433 | ||
5434 | out_finish: | |
5435 | ret = finish_inode_if_needed(sctx, 1); | |
5436 | ||
5437 | out: | |
5438 | btrfs_free_path(path); | |
31db9f7c AB |
5439 | return ret; |
5440 | } | |
5441 | ||
5442 | static int send_subvol(struct send_ctx *sctx) | |
5443 | { | |
5444 | int ret; | |
5445 | ||
c2c71324 SB |
5446 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { |
5447 | ret = send_header(sctx); | |
5448 | if (ret < 0) | |
5449 | goto out; | |
5450 | } | |
31db9f7c AB |
5451 | |
5452 | ret = send_subvol_begin(sctx); | |
5453 | if (ret < 0) | |
5454 | goto out; | |
5455 | ||
5456 | if (sctx->parent_root) { | |
5457 | ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, | |
5458 | changed_cb, sctx); | |
5459 | if (ret < 0) | |
5460 | goto out; | |
5461 | ret = finish_inode_if_needed(sctx, 1); | |
5462 | if (ret < 0) | |
5463 | goto out; | |
5464 | } else { | |
5465 | ret = full_send_tree(sctx); | |
5466 | if (ret < 0) | |
5467 | goto out; | |
5468 | } | |
5469 | ||
5470 | out: | |
31db9f7c AB |
5471 | free_recorded_refs(sctx); |
5472 | return ret; | |
5473 | } | |
5474 | ||
66ef7d65 DS |
5475 | static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) |
5476 | { | |
5477 | spin_lock(&root->root_item_lock); | |
5478 | root->send_in_progress--; | |
5479 | /* | |
5480 | * Not much left to do, we don't know why it's unbalanced and | |
5481 | * can't blindly reset it to 0. | |
5482 | */ | |
5483 | if (root->send_in_progress < 0) | |
5484 | btrfs_err(root->fs_info, | |
5485 | "send_in_progres unbalanced %d root %llu\n", | |
5486 | root->send_in_progress, root->root_key.objectid); | |
5487 | spin_unlock(&root->root_item_lock); | |
5488 | } | |
5489 | ||
31db9f7c AB |
5490 | long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_) |
5491 | { | |
5492 | int ret = 0; | |
5493 | struct btrfs_root *send_root; | |
5494 | struct btrfs_root *clone_root; | |
5495 | struct btrfs_fs_info *fs_info; | |
5496 | struct btrfs_ioctl_send_args *arg = NULL; | |
5497 | struct btrfs_key key; | |
31db9f7c AB |
5498 | struct send_ctx *sctx = NULL; |
5499 | u32 i; | |
5500 | u64 *clone_sources_tmp = NULL; | |
2c686537 | 5501 | int clone_sources_to_rollback = 0; |
896c14f9 | 5502 | int sort_clone_roots = 0; |
18f687d5 | 5503 | int index; |
31db9f7c AB |
5504 | |
5505 | if (!capable(CAP_SYS_ADMIN)) | |
5506 | return -EPERM; | |
5507 | ||
496ad9aa | 5508 | send_root = BTRFS_I(file_inode(mnt_file))->root; |
31db9f7c AB |
5509 | fs_info = send_root->fs_info; |
5510 | ||
2c686537 DS |
5511 | /* |
5512 | * The subvolume must remain read-only during send, protect against | |
5513 | * making it RW. | |
5514 | */ | |
5515 | spin_lock(&send_root->root_item_lock); | |
5516 | send_root->send_in_progress++; | |
5517 | spin_unlock(&send_root->root_item_lock); | |
5518 | ||
139f807a JB |
5519 | /* |
5520 | * This is done when we lookup the root, it should already be complete | |
5521 | * by the time we get here. | |
5522 | */ | |
5523 | WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE); | |
5524 | ||
2c686537 DS |
5525 | /* |
5526 | * Userspace tools do the checks and warn the user if it's | |
5527 | * not RO. | |
5528 | */ | |
5529 | if (!btrfs_root_readonly(send_root)) { | |
5530 | ret = -EPERM; | |
5531 | goto out; | |
5532 | } | |
5533 | ||
31db9f7c AB |
5534 | arg = memdup_user(arg_, sizeof(*arg)); |
5535 | if (IS_ERR(arg)) { | |
5536 | ret = PTR_ERR(arg); | |
5537 | arg = NULL; | |
5538 | goto out; | |
5539 | } | |
5540 | ||
5541 | if (!access_ok(VERIFY_READ, arg->clone_sources, | |
700ff4f0 DC |
5542 | sizeof(*arg->clone_sources) * |
5543 | arg->clone_sources_count)) { | |
31db9f7c AB |
5544 | ret = -EFAULT; |
5545 | goto out; | |
5546 | } | |
5547 | ||
c2c71324 | 5548 | if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { |
cb95e7bf MF |
5549 | ret = -EINVAL; |
5550 | goto out; | |
5551 | } | |
5552 | ||
31db9f7c AB |
5553 | sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS); |
5554 | if (!sctx) { | |
5555 | ret = -ENOMEM; | |
5556 | goto out; | |
5557 | } | |
5558 | ||
5559 | INIT_LIST_HEAD(&sctx->new_refs); | |
5560 | INIT_LIST_HEAD(&sctx->deleted_refs); | |
5561 | INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS); | |
5562 | INIT_LIST_HEAD(&sctx->name_cache_list); | |
5563 | ||
cb95e7bf MF |
5564 | sctx->flags = arg->flags; |
5565 | ||
31db9f7c | 5566 | sctx->send_filp = fget(arg->send_fd); |
ecc7ada7 TI |
5567 | if (!sctx->send_filp) { |
5568 | ret = -EBADF; | |
31db9f7c AB |
5569 | goto out; |
5570 | } | |
5571 | ||
31db9f7c AB |
5572 | sctx->send_root = send_root; |
5573 | sctx->clone_roots_cnt = arg->clone_sources_count; | |
5574 | ||
5575 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE; | |
5576 | sctx->send_buf = vmalloc(sctx->send_max_size); | |
5577 | if (!sctx->send_buf) { | |
5578 | ret = -ENOMEM; | |
5579 | goto out; | |
5580 | } | |
5581 | ||
5582 | sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE); | |
5583 | if (!sctx->read_buf) { | |
5584 | ret = -ENOMEM; | |
5585 | goto out; | |
5586 | } | |
5587 | ||
9f03740a FDBM |
5588 | sctx->pending_dir_moves = RB_ROOT; |
5589 | sctx->waiting_dir_moves = RB_ROOT; | |
9dc44214 | 5590 | sctx->orphan_dirs = RB_ROOT; |
9f03740a | 5591 | |
31db9f7c AB |
5592 | sctx->clone_roots = vzalloc(sizeof(struct clone_root) * |
5593 | (arg->clone_sources_count + 1)); | |
5594 | if (!sctx->clone_roots) { | |
5595 | ret = -ENOMEM; | |
5596 | goto out; | |
5597 | } | |
5598 | ||
5599 | if (arg->clone_sources_count) { | |
5600 | clone_sources_tmp = vmalloc(arg->clone_sources_count * | |
5601 | sizeof(*arg->clone_sources)); | |
5602 | if (!clone_sources_tmp) { | |
5603 | ret = -ENOMEM; | |
5604 | goto out; | |
5605 | } | |
5606 | ||
5607 | ret = copy_from_user(clone_sources_tmp, arg->clone_sources, | |
5608 | arg->clone_sources_count * | |
5609 | sizeof(*arg->clone_sources)); | |
5610 | if (ret) { | |
5611 | ret = -EFAULT; | |
5612 | goto out; | |
5613 | } | |
5614 | ||
5615 | for (i = 0; i < arg->clone_sources_count; i++) { | |
5616 | key.objectid = clone_sources_tmp[i]; | |
5617 | key.type = BTRFS_ROOT_ITEM_KEY; | |
5618 | key.offset = (u64)-1; | |
18f687d5 WS |
5619 | |
5620 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
5621 | ||
31db9f7c | 5622 | clone_root = btrfs_read_fs_root_no_name(fs_info, &key); |
31db9f7c | 5623 | if (IS_ERR(clone_root)) { |
18f687d5 | 5624 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
31db9f7c AB |
5625 | ret = PTR_ERR(clone_root); |
5626 | goto out; | |
5627 | } | |
2c686537 DS |
5628 | clone_sources_to_rollback = i + 1; |
5629 | spin_lock(&clone_root->root_item_lock); | |
5630 | clone_root->send_in_progress++; | |
5631 | if (!btrfs_root_readonly(clone_root)) { | |
5632 | spin_unlock(&clone_root->root_item_lock); | |
18f687d5 | 5633 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
5634 | ret = -EPERM; |
5635 | goto out; | |
5636 | } | |
5637 | spin_unlock(&clone_root->root_item_lock); | |
18f687d5 WS |
5638 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
5639 | ||
31db9f7c AB |
5640 | sctx->clone_roots[i].root = clone_root; |
5641 | } | |
5642 | vfree(clone_sources_tmp); | |
5643 | clone_sources_tmp = NULL; | |
5644 | } | |
5645 | ||
5646 | if (arg->parent_root) { | |
5647 | key.objectid = arg->parent_root; | |
5648 | key.type = BTRFS_ROOT_ITEM_KEY; | |
5649 | key.offset = (u64)-1; | |
18f687d5 WS |
5650 | |
5651 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
5652 | ||
31db9f7c | 5653 | sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key); |
b1b19596 | 5654 | if (IS_ERR(sctx->parent_root)) { |
18f687d5 | 5655 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
b1b19596 | 5656 | ret = PTR_ERR(sctx->parent_root); |
31db9f7c AB |
5657 | goto out; |
5658 | } | |
18f687d5 | 5659 | |
2c686537 DS |
5660 | spin_lock(&sctx->parent_root->root_item_lock); |
5661 | sctx->parent_root->send_in_progress++; | |
5662 | if (!btrfs_root_readonly(sctx->parent_root)) { | |
5663 | spin_unlock(&sctx->parent_root->root_item_lock); | |
18f687d5 | 5664 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
2c686537 DS |
5665 | ret = -EPERM; |
5666 | goto out; | |
5667 | } | |
5668 | spin_unlock(&sctx->parent_root->root_item_lock); | |
18f687d5 WS |
5669 | |
5670 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
31db9f7c AB |
5671 | } |
5672 | ||
5673 | /* | |
5674 | * Clones from send_root are allowed, but only if the clone source | |
5675 | * is behind the current send position. This is checked while searching | |
5676 | * for possible clone sources. | |
5677 | */ | |
5678 | sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root; | |
5679 | ||
5680 | /* We do a bsearch later */ | |
5681 | sort(sctx->clone_roots, sctx->clone_roots_cnt, | |
5682 | sizeof(*sctx->clone_roots), __clone_root_cmp_sort, | |
5683 | NULL); | |
896c14f9 | 5684 | sort_clone_roots = 1; |
31db9f7c | 5685 | |
a26e8c9f | 5686 | current->journal_info = (void *)BTRFS_SEND_TRANS_STUB; |
31db9f7c | 5687 | ret = send_subvol(sctx); |
a26e8c9f | 5688 | current->journal_info = NULL; |
31db9f7c AB |
5689 | if (ret < 0) |
5690 | goto out; | |
5691 | ||
c2c71324 SB |
5692 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { |
5693 | ret = begin_cmd(sctx, BTRFS_SEND_C_END); | |
5694 | if (ret < 0) | |
5695 | goto out; | |
5696 | ret = send_cmd(sctx); | |
5697 | if (ret < 0) | |
5698 | goto out; | |
5699 | } | |
31db9f7c AB |
5700 | |
5701 | out: | |
9f03740a FDBM |
5702 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); |
5703 | while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { | |
5704 | struct rb_node *n; | |
5705 | struct pending_dir_move *pm; | |
5706 | ||
5707 | n = rb_first(&sctx->pending_dir_moves); | |
5708 | pm = rb_entry(n, struct pending_dir_move, node); | |
5709 | while (!list_empty(&pm->list)) { | |
5710 | struct pending_dir_move *pm2; | |
5711 | ||
5712 | pm2 = list_first_entry(&pm->list, | |
5713 | struct pending_dir_move, list); | |
5714 | free_pending_move(sctx, pm2); | |
5715 | } | |
5716 | free_pending_move(sctx, pm); | |
5717 | } | |
5718 | ||
5719 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); | |
5720 | while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { | |
5721 | struct rb_node *n; | |
5722 | struct waiting_dir_move *dm; | |
5723 | ||
5724 | n = rb_first(&sctx->waiting_dir_moves); | |
5725 | dm = rb_entry(n, struct waiting_dir_move, node); | |
5726 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
5727 | kfree(dm); | |
5728 | } | |
5729 | ||
9dc44214 FM |
5730 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); |
5731 | while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { | |
5732 | struct rb_node *n; | |
5733 | struct orphan_dir_info *odi; | |
5734 | ||
5735 | n = rb_first(&sctx->orphan_dirs); | |
5736 | odi = rb_entry(n, struct orphan_dir_info, node); | |
5737 | free_orphan_dir_info(sctx, odi); | |
5738 | } | |
5739 | ||
896c14f9 WS |
5740 | if (sort_clone_roots) { |
5741 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
5742 | btrfs_root_dec_send_in_progress( | |
5743 | sctx->clone_roots[i].root); | |
5744 | } else { | |
5745 | for (i = 0; sctx && i < clone_sources_to_rollback; i++) | |
5746 | btrfs_root_dec_send_in_progress( | |
5747 | sctx->clone_roots[i].root); | |
5748 | ||
5749 | btrfs_root_dec_send_in_progress(send_root); | |
5750 | } | |
66ef7d65 DS |
5751 | if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) |
5752 | btrfs_root_dec_send_in_progress(sctx->parent_root); | |
2c686537 | 5753 | |
31db9f7c AB |
5754 | kfree(arg); |
5755 | vfree(clone_sources_tmp); | |
5756 | ||
5757 | if (sctx) { | |
5758 | if (sctx->send_filp) | |
5759 | fput(sctx->send_filp); | |
5760 | ||
5761 | vfree(sctx->clone_roots); | |
5762 | vfree(sctx->send_buf); | |
5763 | vfree(sctx->read_buf); | |
5764 | ||
5765 | name_cache_free(sctx); | |
5766 | ||
5767 | kfree(sctx); | |
5768 | } | |
5769 | ||
5770 | return ret; | |
5771 | } |