Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
31db9f7c AB |
2 | /* |
3 | * Copyright (C) 2012 Alexander Block. All rights reserved. | |
31db9f7c AB |
4 | */ |
5 | ||
6 | #include <linux/bsearch.h> | |
7 | #include <linux/fs.h> | |
8 | #include <linux/file.h> | |
9 | #include <linux/sort.h> | |
10 | #include <linux/mount.h> | |
11 | #include <linux/xattr.h> | |
12 | #include <linux/posix_acl_xattr.h> | |
5b8418b8 | 13 | #include <linux/radix-tree.h> |
a1857ebe | 14 | #include <linux/vmalloc.h> |
ed84885d | 15 | #include <linux/string.h> |
2351f431 | 16 | #include <linux/compat.h> |
9678c543 | 17 | #include <linux/crc32c.h> |
38622010 | 18 | #include <linux/fsverity.h> |
31db9f7c AB |
19 | |
20 | #include "send.h" | |
8234d3f6 | 21 | #include "ctree.h" |
31db9f7c AB |
22 | #include "backref.h" |
23 | #include "locking.h" | |
24 | #include "disk-io.h" | |
25 | #include "btrfs_inode.h" | |
26 | #include "transaction.h" | |
ebb8765b | 27 | #include "compression.h" |
89efda52 | 28 | #include "xattr.h" |
d96b3424 | 29 | #include "print-tree.h" |
07e81dc9 | 30 | #include "accessors.h" |
f2b39277 | 31 | #include "dir-item.h" |
7c8ede16 | 32 | #include "file-item.h" |
7572dec8 | 33 | #include "ioctl.h" |
5c11adcc | 34 | #include "verity.h" |
90b90d4a | 35 | #include "lru_cache.h" |
31db9f7c | 36 | |
fd0ddbe2 FM |
37 | /* |
38 | * Maximum number of references an extent can have in order for us to attempt to | |
39 | * issue clone operations instead of write operations. This currently exists to | |
40 | * avoid hitting limitations of the backreference walking code (taking a lot of | |
41 | * time and using too much memory for extents with large number of references). | |
42 | */ | |
e2a04165 | 43 | #define SEND_MAX_EXTENT_REFS 1024 |
fd0ddbe2 | 44 | |
31db9f7c AB |
45 | /* |
46 | * A fs_path is a helper to dynamically build path names with unknown size. | |
47 | * It reallocates the internal buffer on demand. | |
48 | * It allows fast adding of path elements on the right side (normal path) and | |
49 | * fast adding to the left side (reversed path). A reversed path can also be | |
50 | * unreversed if needed. | |
51 | */ | |
52 | struct fs_path { | |
53 | union { | |
54 | struct { | |
55 | char *start; | |
56 | char *end; | |
31db9f7c AB |
57 | |
58 | char *buf; | |
1f5a7ff9 DS |
59 | unsigned short buf_len:15; |
60 | unsigned short reversed:1; | |
31db9f7c AB |
61 | char inline_buf[]; |
62 | }; | |
ace01050 DS |
63 | /* |
64 | * Average path length does not exceed 200 bytes, we'll have | |
65 | * better packing in the slab and higher chance to satisfy | |
66 | * a allocation later during send. | |
67 | */ | |
68 | char pad[256]; | |
31db9f7c AB |
69 | }; |
70 | }; | |
71 | #define FS_PATH_INLINE_SIZE \ | |
72 | (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) | |
73 | ||
74 | ||
75 | /* reused for each extent */ | |
76 | struct clone_root { | |
77 | struct btrfs_root *root; | |
78 | u64 ino; | |
79 | u64 offset; | |
c7499a64 | 80 | u64 num_bytes; |
88ffb665 | 81 | bool found_ref; |
31db9f7c AB |
82 | }; |
83 | ||
c48545de | 84 | #define SEND_MAX_NAME_CACHE_SIZE 256 |
31db9f7c | 85 | |
66d04209 | 86 | /* |
ace79df8 FM |
87 | * Limit the root_ids array of struct backref_cache_entry to 17 elements. |
88 | * This makes the size of a cache entry to be exactly 192 bytes on x86_64, which | |
89 | * can be satisfied from the kmalloc-192 slab, without wasting any space. | |
66d04209 | 90 | * The most common case is to have a single root for cloning, which corresponds |
ace79df8 | 91 | * to the send root. Having the user specify more than 16 clone roots is not |
66d04209 | 92 | * common, and in such rare cases we simply don't use caching if the number of |
ace79df8 | 93 | * cloning roots that lead down to a leaf is more than 17. |
66d04209 | 94 | */ |
ace79df8 | 95 | #define SEND_MAX_BACKREF_CACHE_ROOTS 17 |
66d04209 FM |
96 | |
97 | /* | |
98 | * Max number of entries in the cache. | |
ace79df8 FM |
99 | * With SEND_MAX_BACKREF_CACHE_ROOTS as 17, the size in bytes, excluding |
100 | * maple tree's internal nodes, is 24K. | |
66d04209 FM |
101 | */ |
102 | #define SEND_MAX_BACKREF_CACHE_SIZE 128 | |
103 | ||
104 | /* | |
105 | * A backref cache entry maps a leaf to a list of IDs of roots from which the | |
106 | * leaf is accessible and we can use for clone operations. | |
107 | * With SEND_MAX_BACKREF_CACHE_ROOTS as 12, each cache entry is 128 bytes (on | |
108 | * x86_64). | |
109 | */ | |
110 | struct backref_cache_entry { | |
90b90d4a | 111 | struct btrfs_lru_cache_entry entry; |
66d04209 FM |
112 | u64 root_ids[SEND_MAX_BACKREF_CACHE_ROOTS]; |
113 | /* Number of valid elements in the root_ids array. */ | |
114 | int num_roots; | |
115 | }; | |
116 | ||
90b90d4a FM |
117 | /* See the comment at lru_cache.h about struct btrfs_lru_cache_entry. */ |
118 | static_assert(offsetof(struct backref_cache_entry, entry) == 0); | |
119 | ||
e8a7f49d FM |
120 | /* |
121 | * Max number of entries in the cache that stores directories that were already | |
122 | * created. The cache uses raw struct btrfs_lru_cache_entry entries, so it uses | |
0da0c560 | 123 | * at most 4096 bytes - sizeof(struct btrfs_lru_cache_entry) is 48 bytes, but |
e8a7f49d FM |
124 | * the kmalloc-64 slab is used, so we get 4096 bytes (64 bytes * 64). |
125 | */ | |
126 | #define SEND_MAX_DIR_CREATED_CACHE_SIZE 64 | |
127 | ||
3e49363b FM |
128 | /* |
129 | * Max number of entries in the cache that stores directories that were already | |
130 | * created. The cache uses raw struct btrfs_lru_cache_entry entries, so it uses | |
131 | * at most 4096 bytes - sizeof(struct btrfs_lru_cache_entry) is 48 bytes, but | |
132 | * the kmalloc-64 slab is used, so we get 4096 bytes (64 bytes * 64). | |
133 | */ | |
134 | #define SEND_MAX_DIR_UTIMES_CACHE_SIZE 64 | |
135 | ||
31db9f7c AB |
136 | struct send_ctx { |
137 | struct file *send_filp; | |
138 | loff_t send_off; | |
139 | char *send_buf; | |
140 | u32 send_size; | |
141 | u32 send_max_size; | |
356bbbb6 OS |
142 | /* |
143 | * Whether BTRFS_SEND_A_DATA attribute was already added to current | |
144 | * command (since protocol v2, data must be the last attribute). | |
145 | */ | |
146 | bool put_data; | |
a4b333f2 | 147 | struct page **send_buf_pages; |
cb95e7bf | 148 | u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ |
e77fbf99 DS |
149 | /* Protocol version compatibility requested */ |
150 | u32 proto; | |
31db9f7c | 151 | |
31db9f7c AB |
152 | struct btrfs_root *send_root; |
153 | struct btrfs_root *parent_root; | |
154 | struct clone_root *clone_roots; | |
155 | int clone_roots_cnt; | |
156 | ||
157 | /* current state of the compare_tree call */ | |
158 | struct btrfs_path *left_path; | |
159 | struct btrfs_path *right_path; | |
160 | struct btrfs_key *cmp_key; | |
161 | ||
d96b3424 FM |
162 | /* |
163 | * Keep track of the generation of the last transaction that was used | |
164 | * for relocating a block group. This is periodically checked in order | |
165 | * to detect if a relocation happened since the last check, so that we | |
166 | * don't operate on stale extent buffers for nodes (level >= 1) or on | |
167 | * stale disk_bytenr values of file extent items. | |
168 | */ | |
169 | u64 last_reloc_trans; | |
170 | ||
31db9f7c AB |
171 | /* |
172 | * infos of the currently processed inode. In case of deleted inodes, | |
173 | * these are the values from the deleted inode. | |
174 | */ | |
175 | u64 cur_ino; | |
176 | u64 cur_inode_gen; | |
31db9f7c AB |
177 | u64 cur_inode_size; |
178 | u64 cur_inode_mode; | |
644d1940 | 179 | u64 cur_inode_rdev; |
16e7549f | 180 | u64 cur_inode_last_extent; |
ffa7c429 | 181 | u64 cur_inode_next_write_offset; |
9555e1f1 DS |
182 | bool cur_inode_new; |
183 | bool cur_inode_new_gen; | |
184 | bool cur_inode_deleted; | |
46b2f459 | 185 | bool ignore_cur_inode; |
38622010 BB |
186 | bool cur_inode_needs_verity; |
187 | void *verity_descriptor; | |
31db9f7c AB |
188 | |
189 | u64 send_progress; | |
190 | ||
191 | struct list_head new_refs; | |
192 | struct list_head deleted_refs; | |
193 | ||
c48545de | 194 | struct btrfs_lru_cache name_cache; |
31db9f7c | 195 | |
521b6803 FM |
196 | /* |
197 | * The inode we are currently processing. It's not NULL only when we | |
198 | * need to issue write commands for data extents from this inode. | |
199 | */ | |
200 | struct inode *cur_inode; | |
2131bcd3 | 201 | struct file_ra_state ra; |
152555b3 FM |
202 | u64 page_cache_clear_start; |
203 | bool clean_page_cache; | |
2131bcd3 | 204 | |
9f03740a FDBM |
205 | /* |
206 | * We process inodes by their increasing order, so if before an | |
207 | * incremental send we reverse the parent/child relationship of | |
208 | * directories such that a directory with a lower inode number was | |
209 | * the parent of a directory with a higher inode number, and the one | |
210 | * becoming the new parent got renamed too, we can't rename/move the | |
211 | * directory with lower inode number when we finish processing it - we | |
212 | * must process the directory with higher inode number first, then | |
213 | * rename/move it and then rename/move the directory with lower inode | |
214 | * number. Example follows. | |
215 | * | |
216 | * Tree state when the first send was performed: | |
217 | * | |
218 | * . | |
219 | * |-- a (ino 257) | |
220 | * |-- b (ino 258) | |
221 | * | | |
222 | * | | |
223 | * |-- c (ino 259) | |
224 | * | |-- d (ino 260) | |
225 | * | | |
226 | * |-- c2 (ino 261) | |
227 | * | |
228 | * Tree state when the second (incremental) send is performed: | |
229 | * | |
230 | * . | |
231 | * |-- a (ino 257) | |
232 | * |-- b (ino 258) | |
233 | * |-- c2 (ino 261) | |
234 | * |-- d2 (ino 260) | |
235 | * |-- cc (ino 259) | |
236 | * | |
237 | * The sequence of steps that lead to the second state was: | |
238 | * | |
239 | * mv /a/b/c/d /a/b/c2/d2 | |
240 | * mv /a/b/c /a/b/c2/d2/cc | |
241 | * | |
242 | * "c" has lower inode number, but we can't move it (2nd mv operation) | |
243 | * before we move "d", which has higher inode number. | |
244 | * | |
245 | * So we just memorize which move/rename operations must be performed | |
246 | * later when their respective parent is processed and moved/renamed. | |
247 | */ | |
248 | ||
249 | /* Indexed by parent directory inode number. */ | |
250 | struct rb_root pending_dir_moves; | |
251 | ||
252 | /* | |
253 | * Reverse index, indexed by the inode number of a directory that | |
254 | * is waiting for the move/rename of its immediate parent before its | |
255 | * own move/rename can be performed. | |
256 | */ | |
257 | struct rb_root waiting_dir_moves; | |
9dc44214 FM |
258 | |
259 | /* | |
260 | * A directory that is going to be rm'ed might have a child directory | |
261 | * which is in the pending directory moves index above. In this case, | |
262 | * the directory can only be removed after the move/rename of its child | |
263 | * is performed. Example: | |
264 | * | |
265 | * Parent snapshot: | |
266 | * | |
267 | * . (ino 256) | |
268 | * |-- a/ (ino 257) | |
269 | * |-- b/ (ino 258) | |
270 | * |-- c/ (ino 259) | |
271 | * | |-- x/ (ino 260) | |
272 | * | | |
273 | * |-- y/ (ino 261) | |
274 | * | |
275 | * Send snapshot: | |
276 | * | |
277 | * . (ino 256) | |
278 | * |-- a/ (ino 257) | |
279 | * |-- b/ (ino 258) | |
280 | * |-- YY/ (ino 261) | |
281 | * |-- x/ (ino 260) | |
282 | * | |
283 | * Sequence of steps that lead to the send snapshot: | |
284 | * rm -f /a/b/c/foo.txt | |
285 | * mv /a/b/y /a/b/YY | |
286 | * mv /a/b/c/x /a/b/YY | |
287 | * rmdir /a/b/c | |
288 | * | |
289 | * When the child is processed, its move/rename is delayed until its | |
290 | * parent is processed (as explained above), but all other operations | |
291 | * like update utimes, chown, chgrp, etc, are performed and the paths | |
292 | * that it uses for those operations must use the orphanized name of | |
293 | * its parent (the directory we're going to rm later), so we need to | |
294 | * memorize that name. | |
295 | * | |
296 | * Indexed by the inode number of the directory to be deleted. | |
297 | */ | |
298 | struct rb_root orphan_dirs; | |
3aa5bd36 BC |
299 | |
300 | struct rb_root rbtree_new_refs; | |
301 | struct rb_root rbtree_deleted_refs; | |
66d04209 | 302 | |
90b90d4a FM |
303 | struct btrfs_lru_cache backref_cache; |
304 | u64 backref_cache_last_reloc_trans; | |
e8a7f49d FM |
305 | |
306 | struct btrfs_lru_cache dir_created_cache; | |
3e49363b | 307 | struct btrfs_lru_cache dir_utimes_cache; |
9f03740a FDBM |
308 | }; |
309 | ||
310 | struct pending_dir_move { | |
311 | struct rb_node node; | |
312 | struct list_head list; | |
313 | u64 parent_ino; | |
314 | u64 ino; | |
315 | u64 gen; | |
316 | struct list_head update_refs; | |
317 | }; | |
318 | ||
319 | struct waiting_dir_move { | |
320 | struct rb_node node; | |
321 | u64 ino; | |
9dc44214 FM |
322 | /* |
323 | * There might be some directory that could not be removed because it | |
324 | * was waiting for this directory inode to be moved first. Therefore | |
325 | * after this directory is moved, we can try to rmdir the ino rmdir_ino. | |
326 | */ | |
327 | u64 rmdir_ino; | |
0b3f407e | 328 | u64 rmdir_gen; |
8b191a68 | 329 | bool orphanized; |
9dc44214 FM |
330 | }; |
331 | ||
332 | struct orphan_dir_info { | |
333 | struct rb_node node; | |
334 | u64 ino; | |
335 | u64 gen; | |
0f96f517 | 336 | u64 last_dir_index_offset; |
474e4761 | 337 | u64 dir_high_seq_ino; |
31db9f7c AB |
338 | }; |
339 | ||
340 | struct name_cache_entry { | |
7e0926fe | 341 | /* |
c48545de FM |
342 | * The key in the entry is an inode number, and the generation matches |
343 | * the inode's generation. | |
7e0926fe | 344 | */ |
c48545de | 345 | struct btrfs_lru_cache_entry entry; |
31db9f7c AB |
346 | u64 parent_ino; |
347 | u64 parent_gen; | |
348 | int ret; | |
349 | int need_later_update; | |
350 | int name_len; | |
351 | char name[]; | |
352 | }; | |
353 | ||
c48545de FM |
354 | /* See the comment at lru_cache.h about struct btrfs_lru_cache_entry. */ |
355 | static_assert(offsetof(struct name_cache_entry, entry) == 0); | |
356 | ||
18d0f5c6 DS |
357 | #define ADVANCE 1 |
358 | #define ADVANCE_ONLY_NEXT -1 | |
359 | ||
360 | enum btrfs_compare_tree_result { | |
361 | BTRFS_COMPARE_TREE_NEW, | |
362 | BTRFS_COMPARE_TREE_DELETED, | |
363 | BTRFS_COMPARE_TREE_CHANGED, | |
364 | BTRFS_COMPARE_TREE_SAME, | |
365 | }; | |
18d0f5c6 | 366 | |
e67c718b | 367 | __cold |
95155585 FM |
368 | static void inconsistent_snapshot_error(struct send_ctx *sctx, |
369 | enum btrfs_compare_tree_result result, | |
370 | const char *what) | |
371 | { | |
372 | const char *result_string; | |
373 | ||
374 | switch (result) { | |
375 | case BTRFS_COMPARE_TREE_NEW: | |
376 | result_string = "new"; | |
377 | break; | |
378 | case BTRFS_COMPARE_TREE_DELETED: | |
379 | result_string = "deleted"; | |
380 | break; | |
381 | case BTRFS_COMPARE_TREE_CHANGED: | |
382 | result_string = "updated"; | |
383 | break; | |
384 | case BTRFS_COMPARE_TREE_SAME: | |
385 | ASSERT(0); | |
386 | result_string = "unchanged"; | |
387 | break; | |
388 | default: | |
389 | ASSERT(0); | |
390 | result_string = "unexpected"; | |
391 | } | |
392 | ||
393 | btrfs_err(sctx->send_root->fs_info, | |
394 | "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu", | |
395 | result_string, what, sctx->cmp_key->objectid, | |
396 | sctx->send_root->root_key.objectid, | |
397 | (sctx->parent_root ? | |
398 | sctx->parent_root->root_key.objectid : 0)); | |
399 | } | |
400 | ||
e77fbf99 DS |
401 | __maybe_unused |
402 | static bool proto_cmd_ok(const struct send_ctx *sctx, int cmd) | |
403 | { | |
404 | switch (sctx->proto) { | |
54cab6af OS |
405 | case 1: return cmd <= BTRFS_SEND_C_MAX_V1; |
406 | case 2: return cmd <= BTRFS_SEND_C_MAX_V2; | |
c86eab81 | 407 | case 3: return cmd <= BTRFS_SEND_C_MAX_V3; |
e77fbf99 DS |
408 | default: return false; |
409 | } | |
410 | } | |
411 | ||
9f03740a FDBM |
412 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); |
413 | ||
9dc44214 FM |
414 | static struct waiting_dir_move * |
415 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino); | |
416 | ||
0b3f407e | 417 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen); |
9dc44214 | 418 | |
16e7549f JB |
419 | static int need_send_hole(struct send_ctx *sctx) |
420 | { | |
421 | return (sctx->parent_root && !sctx->cur_inode_new && | |
422 | !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted && | |
423 | S_ISREG(sctx->cur_inode_mode)); | |
424 | } | |
425 | ||
31db9f7c AB |
426 | static void fs_path_reset(struct fs_path *p) |
427 | { | |
428 | if (p->reversed) { | |
429 | p->start = p->buf + p->buf_len - 1; | |
430 | p->end = p->start; | |
431 | *p->start = 0; | |
432 | } else { | |
433 | p->start = p->buf; | |
434 | p->end = p->start; | |
435 | *p->start = 0; | |
436 | } | |
437 | } | |
438 | ||
924794c9 | 439 | static struct fs_path *fs_path_alloc(void) |
31db9f7c AB |
440 | { |
441 | struct fs_path *p; | |
442 | ||
e780b0d1 | 443 | p = kmalloc(sizeof(*p), GFP_KERNEL); |
31db9f7c AB |
444 | if (!p) |
445 | return NULL; | |
446 | p->reversed = 0; | |
31db9f7c AB |
447 | p->buf = p->inline_buf; |
448 | p->buf_len = FS_PATH_INLINE_SIZE; | |
449 | fs_path_reset(p); | |
450 | return p; | |
451 | } | |
452 | ||
924794c9 | 453 | static struct fs_path *fs_path_alloc_reversed(void) |
31db9f7c AB |
454 | { |
455 | struct fs_path *p; | |
456 | ||
924794c9 | 457 | p = fs_path_alloc(); |
31db9f7c AB |
458 | if (!p) |
459 | return NULL; | |
460 | p->reversed = 1; | |
461 | fs_path_reset(p); | |
462 | return p; | |
463 | } | |
464 | ||
924794c9 | 465 | static void fs_path_free(struct fs_path *p) |
31db9f7c AB |
466 | { |
467 | if (!p) | |
468 | return; | |
ace01050 DS |
469 | if (p->buf != p->inline_buf) |
470 | kfree(p->buf); | |
31db9f7c AB |
471 | kfree(p); |
472 | } | |
473 | ||
474 | static int fs_path_len(struct fs_path *p) | |
475 | { | |
476 | return p->end - p->start; | |
477 | } | |
478 | ||
479 | static int fs_path_ensure_buf(struct fs_path *p, int len) | |
480 | { | |
481 | char *tmp_buf; | |
482 | int path_len; | |
483 | int old_buf_len; | |
484 | ||
485 | len++; | |
486 | ||
487 | if (p->buf_len >= len) | |
488 | return 0; | |
489 | ||
cfd4a535 CM |
490 | if (len > PATH_MAX) { |
491 | WARN_ON(1); | |
492 | return -ENOMEM; | |
493 | } | |
494 | ||
1b2782c8 DS |
495 | path_len = p->end - p->start; |
496 | old_buf_len = p->buf_len; | |
497 | ||
905889bc KC |
498 | /* |
499 | * Allocate to the next largest kmalloc bucket size, to let | |
500 | * the fast path happen most of the time. | |
501 | */ | |
502 | len = kmalloc_size_roundup(len); | |
ace01050 DS |
503 | /* |
504 | * First time the inline_buf does not suffice | |
505 | */ | |
01a9a8a9 | 506 | if (p->buf == p->inline_buf) { |
e780b0d1 | 507 | tmp_buf = kmalloc(len, GFP_KERNEL); |
01a9a8a9 FM |
508 | if (tmp_buf) |
509 | memcpy(tmp_buf, p->buf, old_buf_len); | |
510 | } else { | |
e780b0d1 | 511 | tmp_buf = krealloc(p->buf, len, GFP_KERNEL); |
01a9a8a9 | 512 | } |
9c9ca00b DS |
513 | if (!tmp_buf) |
514 | return -ENOMEM; | |
515 | p->buf = tmp_buf; | |
905889bc | 516 | p->buf_len = len; |
ace01050 | 517 | |
31db9f7c AB |
518 | if (p->reversed) { |
519 | tmp_buf = p->buf + old_buf_len - path_len - 1; | |
520 | p->end = p->buf + p->buf_len - 1; | |
521 | p->start = p->end - path_len; | |
522 | memmove(p->start, tmp_buf, path_len + 1); | |
523 | } else { | |
524 | p->start = p->buf; | |
525 | p->end = p->start + path_len; | |
526 | } | |
527 | return 0; | |
528 | } | |
529 | ||
b23ab57d DS |
530 | static int fs_path_prepare_for_add(struct fs_path *p, int name_len, |
531 | char **prepared) | |
31db9f7c AB |
532 | { |
533 | int ret; | |
534 | int new_len; | |
535 | ||
536 | new_len = p->end - p->start + name_len; | |
537 | if (p->start != p->end) | |
538 | new_len++; | |
539 | ret = fs_path_ensure_buf(p, new_len); | |
540 | if (ret < 0) | |
541 | goto out; | |
542 | ||
543 | if (p->reversed) { | |
544 | if (p->start != p->end) | |
545 | *--p->start = '/'; | |
546 | p->start -= name_len; | |
b23ab57d | 547 | *prepared = p->start; |
31db9f7c AB |
548 | } else { |
549 | if (p->start != p->end) | |
550 | *p->end++ = '/'; | |
b23ab57d | 551 | *prepared = p->end; |
31db9f7c AB |
552 | p->end += name_len; |
553 | *p->end = 0; | |
554 | } | |
555 | ||
556 | out: | |
557 | return ret; | |
558 | } | |
559 | ||
560 | static int fs_path_add(struct fs_path *p, const char *name, int name_len) | |
561 | { | |
562 | int ret; | |
b23ab57d | 563 | char *prepared; |
31db9f7c | 564 | |
b23ab57d | 565 | ret = fs_path_prepare_for_add(p, name_len, &prepared); |
31db9f7c AB |
566 | if (ret < 0) |
567 | goto out; | |
b23ab57d | 568 | memcpy(prepared, name, name_len); |
31db9f7c AB |
569 | |
570 | out: | |
571 | return ret; | |
572 | } | |
573 | ||
574 | static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) | |
575 | { | |
576 | int ret; | |
b23ab57d | 577 | char *prepared; |
31db9f7c | 578 | |
b23ab57d | 579 | ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); |
31db9f7c AB |
580 | if (ret < 0) |
581 | goto out; | |
b23ab57d | 582 | memcpy(prepared, p2->start, p2->end - p2->start); |
31db9f7c AB |
583 | |
584 | out: | |
585 | return ret; | |
586 | } | |
587 | ||
588 | static int fs_path_add_from_extent_buffer(struct fs_path *p, | |
589 | struct extent_buffer *eb, | |
590 | unsigned long off, int len) | |
591 | { | |
592 | int ret; | |
b23ab57d | 593 | char *prepared; |
31db9f7c | 594 | |
b23ab57d | 595 | ret = fs_path_prepare_for_add(p, len, &prepared); |
31db9f7c AB |
596 | if (ret < 0) |
597 | goto out; | |
598 | ||
b23ab57d | 599 | read_extent_buffer(eb, prepared, off, len); |
31db9f7c AB |
600 | |
601 | out: | |
602 | return ret; | |
603 | } | |
604 | ||
31db9f7c AB |
605 | static int fs_path_copy(struct fs_path *p, struct fs_path *from) |
606 | { | |
31db9f7c AB |
607 | p->reversed = from->reversed; |
608 | fs_path_reset(p); | |
609 | ||
0292ecf1 | 610 | return fs_path_add_path(p, from); |
31db9f7c AB |
611 | } |
612 | ||
31db9f7c AB |
613 | static void fs_path_unreverse(struct fs_path *p) |
614 | { | |
615 | char *tmp; | |
616 | int len; | |
617 | ||
618 | if (!p->reversed) | |
619 | return; | |
620 | ||
621 | tmp = p->start; | |
622 | len = p->end - p->start; | |
623 | p->start = p->buf; | |
624 | p->end = p->start + len; | |
625 | memmove(p->start, tmp, len + 1); | |
626 | p->reversed = 0; | |
627 | } | |
628 | ||
629 | static struct btrfs_path *alloc_path_for_send(void) | |
630 | { | |
631 | struct btrfs_path *path; | |
632 | ||
633 | path = btrfs_alloc_path(); | |
634 | if (!path) | |
635 | return NULL; | |
636 | path->search_commit_root = 1; | |
637 | path->skip_locking = 1; | |
3f8a18cc | 638 | path->need_commit_sem = 1; |
31db9f7c AB |
639 | return path; |
640 | } | |
641 | ||
48a3b636 | 642 | static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) |
31db9f7c AB |
643 | { |
644 | int ret; | |
31db9f7c AB |
645 | u32 pos = 0; |
646 | ||
31db9f7c | 647 | while (pos < len) { |
8e93157b | 648 | ret = kernel_write(filp, buf + pos, len - pos, off); |
31db9f7c | 649 | if (ret < 0) |
8e93157b | 650 | return ret; |
cec3dad9 | 651 | if (ret == 0) |
8e93157b | 652 | return -EIO; |
31db9f7c AB |
653 | pos += ret; |
654 | } | |
655 | ||
8e93157b | 656 | return 0; |
31db9f7c AB |
657 | } |
658 | ||
659 | static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) | |
660 | { | |
661 | struct btrfs_tlv_header *hdr; | |
662 | int total_len = sizeof(*hdr) + len; | |
663 | int left = sctx->send_max_size - sctx->send_size; | |
664 | ||
356bbbb6 OS |
665 | if (WARN_ON_ONCE(sctx->put_data)) |
666 | return -EINVAL; | |
667 | ||
31db9f7c AB |
668 | if (unlikely(left < total_len)) |
669 | return -EOVERFLOW; | |
670 | ||
671 | hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); | |
e2f896b3 DS |
672 | put_unaligned_le16(attr, &hdr->tlv_type); |
673 | put_unaligned_le16(len, &hdr->tlv_len); | |
31db9f7c AB |
674 | memcpy(hdr + 1, data, len); |
675 | sctx->send_size += total_len; | |
676 | ||
677 | return 0; | |
678 | } | |
679 | ||
95bc79d5 DS |
680 | #define TLV_PUT_DEFINE_INT(bits) \ |
681 | static int tlv_put_u##bits(struct send_ctx *sctx, \ | |
682 | u##bits attr, u##bits value) \ | |
683 | { \ | |
684 | __le##bits __tmp = cpu_to_le##bits(value); \ | |
685 | return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ | |
686 | } | |
31db9f7c | 687 | |
38622010 | 688 | TLV_PUT_DEFINE_INT(8) |
3ea4dc5b | 689 | TLV_PUT_DEFINE_INT(32) |
95bc79d5 | 690 | TLV_PUT_DEFINE_INT(64) |
31db9f7c AB |
691 | |
692 | static int tlv_put_string(struct send_ctx *sctx, u16 attr, | |
693 | const char *str, int len) | |
694 | { | |
695 | if (len == -1) | |
696 | len = strlen(str); | |
697 | return tlv_put(sctx, attr, str, len); | |
698 | } | |
699 | ||
700 | static int tlv_put_uuid(struct send_ctx *sctx, u16 attr, | |
701 | const u8 *uuid) | |
702 | { | |
703 | return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE); | |
704 | } | |
705 | ||
31db9f7c AB |
706 | static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, |
707 | struct extent_buffer *eb, | |
708 | struct btrfs_timespec *ts) | |
709 | { | |
710 | struct btrfs_timespec bts; | |
711 | read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts)); | |
712 | return tlv_put(sctx, attr, &bts, sizeof(bts)); | |
713 | } | |
714 | ||
715 | ||
895a72be | 716 | #define TLV_PUT(sctx, attrtype, data, attrlen) \ |
31db9f7c | 717 | do { \ |
895a72be | 718 | ret = tlv_put(sctx, attrtype, data, attrlen); \ |
31db9f7c AB |
719 | if (ret < 0) \ |
720 | goto tlv_put_failure; \ | |
721 | } while (0) | |
722 | ||
723 | #define TLV_PUT_INT(sctx, attrtype, bits, value) \ | |
724 | do { \ | |
725 | ret = tlv_put_u##bits(sctx, attrtype, value); \ | |
726 | if (ret < 0) \ | |
727 | goto tlv_put_failure; \ | |
728 | } while (0) | |
729 | ||
730 | #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data) | |
731 | #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data) | |
732 | #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data) | |
733 | #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data) | |
734 | #define TLV_PUT_STRING(sctx, attrtype, str, len) \ | |
735 | do { \ | |
736 | ret = tlv_put_string(sctx, attrtype, str, len); \ | |
737 | if (ret < 0) \ | |
738 | goto tlv_put_failure; \ | |
739 | } while (0) | |
740 | #define TLV_PUT_PATH(sctx, attrtype, p) \ | |
741 | do { \ | |
742 | ret = tlv_put_string(sctx, attrtype, p->start, \ | |
743 | p->end - p->start); \ | |
744 | if (ret < 0) \ | |
745 | goto tlv_put_failure; \ | |
746 | } while(0) | |
747 | #define TLV_PUT_UUID(sctx, attrtype, uuid) \ | |
748 | do { \ | |
749 | ret = tlv_put_uuid(sctx, attrtype, uuid); \ | |
750 | if (ret < 0) \ | |
751 | goto tlv_put_failure; \ | |
752 | } while (0) | |
31db9f7c AB |
753 | #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \ |
754 | do { \ | |
755 | ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \ | |
756 | if (ret < 0) \ | |
757 | goto tlv_put_failure; \ | |
758 | } while (0) | |
759 | ||
760 | static int send_header(struct send_ctx *sctx) | |
761 | { | |
762 | struct btrfs_stream_header hdr; | |
763 | ||
764 | strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); | |
d6815592 | 765 | hdr.version = cpu_to_le32(sctx->proto); |
1bcea355 AJ |
766 | return write_buf(sctx->send_filp, &hdr, sizeof(hdr), |
767 | &sctx->send_off); | |
31db9f7c AB |
768 | } |
769 | ||
770 | /* | |
771 | * For each command/item we want to send to userspace, we call this function. | |
772 | */ | |
773 | static int begin_cmd(struct send_ctx *sctx, int cmd) | |
774 | { | |
775 | struct btrfs_cmd_header *hdr; | |
776 | ||
fae7f21c | 777 | if (WARN_ON(!sctx->send_buf)) |
31db9f7c | 778 | return -EINVAL; |
31db9f7c AB |
779 | |
780 | BUG_ON(sctx->send_size); | |
781 | ||
782 | sctx->send_size += sizeof(*hdr); | |
783 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
e2f896b3 | 784 | put_unaligned_le16(cmd, &hdr->cmd); |
31db9f7c AB |
785 | |
786 | return 0; | |
787 | } | |
788 | ||
789 | static int send_cmd(struct send_ctx *sctx) | |
790 | { | |
791 | int ret; | |
792 | struct btrfs_cmd_header *hdr; | |
793 | u32 crc; | |
794 | ||
795 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
e2f896b3 DS |
796 | put_unaligned_le32(sctx->send_size - sizeof(*hdr), &hdr->len); |
797 | put_unaligned_le32(0, &hdr->crc); | |
31db9f7c | 798 | |
65019df8 | 799 | crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); |
e2f896b3 | 800 | put_unaligned_le32(crc, &hdr->crc); |
31db9f7c | 801 | |
1bcea355 AJ |
802 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, |
803 | &sctx->send_off); | |
31db9f7c | 804 | |
31db9f7c | 805 | sctx->send_size = 0; |
356bbbb6 | 806 | sctx->put_data = false; |
31db9f7c AB |
807 | |
808 | return ret; | |
809 | } | |
810 | ||
811 | /* | |
812 | * Sends a move instruction to user space | |
813 | */ | |
814 | static int send_rename(struct send_ctx *sctx, | |
815 | struct fs_path *from, struct fs_path *to) | |
816 | { | |
04ab956e | 817 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
818 | int ret; |
819 | ||
04ab956e | 820 | btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start); |
31db9f7c AB |
821 | |
822 | ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); | |
823 | if (ret < 0) | |
824 | goto out; | |
825 | ||
826 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); | |
827 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); | |
828 | ||
829 | ret = send_cmd(sctx); | |
830 | ||
831 | tlv_put_failure: | |
832 | out: | |
833 | return ret; | |
834 | } | |
835 | ||
836 | /* | |
837 | * Sends a link instruction to user space | |
838 | */ | |
839 | static int send_link(struct send_ctx *sctx, | |
840 | struct fs_path *path, struct fs_path *lnk) | |
841 | { | |
04ab956e | 842 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
843 | int ret; |
844 | ||
04ab956e | 845 | btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start); |
31db9f7c AB |
846 | |
847 | ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); | |
848 | if (ret < 0) | |
849 | goto out; | |
850 | ||
851 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
852 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); | |
853 | ||
854 | ret = send_cmd(sctx); | |
855 | ||
856 | tlv_put_failure: | |
857 | out: | |
858 | return ret; | |
859 | } | |
860 | ||
861 | /* | |
862 | * Sends an unlink instruction to user space | |
863 | */ | |
864 | static int send_unlink(struct send_ctx *sctx, struct fs_path *path) | |
865 | { | |
04ab956e | 866 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
867 | int ret; |
868 | ||
04ab956e | 869 | btrfs_debug(fs_info, "send_unlink %s", path->start); |
31db9f7c AB |
870 | |
871 | ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); | |
872 | if (ret < 0) | |
873 | goto out; | |
874 | ||
875 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
876 | ||
877 | ret = send_cmd(sctx); | |
878 | ||
879 | tlv_put_failure: | |
880 | out: | |
881 | return ret; | |
882 | } | |
883 | ||
884 | /* | |
885 | * Sends a rmdir instruction to user space | |
886 | */ | |
887 | static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) | |
888 | { | |
04ab956e | 889 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
890 | int ret; |
891 | ||
04ab956e | 892 | btrfs_debug(fs_info, "send_rmdir %s", path->start); |
31db9f7c AB |
893 | |
894 | ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); | |
895 | if (ret < 0) | |
896 | goto out; | |
897 | ||
898 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
899 | ||
900 | ret = send_cmd(sctx); | |
901 | ||
902 | tlv_put_failure: | |
903 | out: | |
904 | return ret; | |
905 | } | |
906 | ||
7e93f6dc BC |
907 | struct btrfs_inode_info { |
908 | u64 size; | |
909 | u64 gen; | |
910 | u64 mode; | |
911 | u64 uid; | |
912 | u64 gid; | |
913 | u64 rdev; | |
914 | u64 fileattr; | |
9ed0a72e | 915 | u64 nlink; |
7e93f6dc BC |
916 | }; |
917 | ||
31db9f7c AB |
918 | /* |
919 | * Helper function to retrieve some fields from an inode item. | |
920 | */ | |
7e93f6dc BC |
921 | static int get_inode_info(struct btrfs_root *root, u64 ino, |
922 | struct btrfs_inode_info *info) | |
31db9f7c AB |
923 | { |
924 | int ret; | |
7e93f6dc | 925 | struct btrfs_path *path; |
31db9f7c AB |
926 | struct btrfs_inode_item *ii; |
927 | struct btrfs_key key; | |
31db9f7c | 928 | |
7e93f6dc BC |
929 | path = alloc_path_for_send(); |
930 | if (!path) | |
931 | return -ENOMEM; | |
932 | ||
31db9f7c AB |
933 | key.objectid = ino; |
934 | key.type = BTRFS_INODE_ITEM_KEY; | |
935 | key.offset = 0; | |
936 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
31db9f7c | 937 | if (ret) { |
3f8a18cc JB |
938 | if (ret > 0) |
939 | ret = -ENOENT; | |
7e93f6dc | 940 | goto out; |
31db9f7c AB |
941 | } |
942 | ||
7e93f6dc BC |
943 | if (!info) |
944 | goto out; | |
945 | ||
31db9f7c AB |
946 | ii = btrfs_item_ptr(path->nodes[0], path->slots[0], |
947 | struct btrfs_inode_item); | |
7e93f6dc BC |
948 | info->size = btrfs_inode_size(path->nodes[0], ii); |
949 | info->gen = btrfs_inode_generation(path->nodes[0], ii); | |
950 | info->mode = btrfs_inode_mode(path->nodes[0], ii); | |
951 | info->uid = btrfs_inode_uid(path->nodes[0], ii); | |
952 | info->gid = btrfs_inode_gid(path->nodes[0], ii); | |
953 | info->rdev = btrfs_inode_rdev(path->nodes[0], ii); | |
9ed0a72e | 954 | info->nlink = btrfs_inode_nlink(path->nodes[0], ii); |
48247359 DS |
955 | /* |
956 | * Transfer the unchanged u64 value of btrfs_inode_item::flags, that's | |
957 | * otherwise logically split to 32/32 parts. | |
958 | */ | |
7e93f6dc | 959 | info->fileattr = btrfs_inode_flags(path->nodes[0], ii); |
31db9f7c | 960 | |
7e93f6dc BC |
961 | out: |
962 | btrfs_free_path(path); | |
3f8a18cc JB |
963 | return ret; |
964 | } | |
965 | ||
7e93f6dc | 966 | static int get_inode_gen(struct btrfs_root *root, u64 ino, u64 *gen) |
3f8a18cc | 967 | { |
3f8a18cc | 968 | int ret; |
ab199013 | 969 | struct btrfs_inode_info info = { 0 }; |
3f8a18cc | 970 | |
ab199013 | 971 | ASSERT(gen); |
7e93f6dc BC |
972 | |
973 | ret = get_inode_info(root, ino, &info); | |
ab199013 | 974 | *gen = info.gen; |
31db9f7c AB |
975 | return ret; |
976 | } | |
977 | ||
978 | typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, | |
979 | struct fs_path *p, | |
980 | void *ctx); | |
981 | ||
982 | /* | |
96b5bd77 JS |
983 | * Helper function to iterate the entries in ONE btrfs_inode_ref or |
984 | * btrfs_inode_extref. | |
31db9f7c AB |
985 | * The iterate callback may return a non zero value to stop iteration. This can |
986 | * be a negative value for error codes or 1 to simply stop it. | |
987 | * | |
96b5bd77 | 988 | * path must point to the INODE_REF or INODE_EXTREF when called. |
31db9f7c | 989 | */ |
924794c9 | 990 | static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
991 | struct btrfs_key *found_key, int resolve, |
992 | iterate_inode_ref_t iterate, void *ctx) | |
993 | { | |
96b5bd77 | 994 | struct extent_buffer *eb = path->nodes[0]; |
31db9f7c | 995 | struct btrfs_inode_ref *iref; |
96b5bd77 | 996 | struct btrfs_inode_extref *extref; |
31db9f7c AB |
997 | struct btrfs_path *tmp_path; |
998 | struct fs_path *p; | |
96b5bd77 | 999 | u32 cur = 0; |
31db9f7c | 1000 | u32 total; |
96b5bd77 | 1001 | int slot = path->slots[0]; |
31db9f7c AB |
1002 | u32 name_len; |
1003 | char *start; | |
1004 | int ret = 0; | |
96b5bd77 | 1005 | int num = 0; |
31db9f7c | 1006 | int index; |
96b5bd77 JS |
1007 | u64 dir; |
1008 | unsigned long name_off; | |
1009 | unsigned long elem_size; | |
1010 | unsigned long ptr; | |
31db9f7c | 1011 | |
924794c9 | 1012 | p = fs_path_alloc_reversed(); |
31db9f7c AB |
1013 | if (!p) |
1014 | return -ENOMEM; | |
1015 | ||
1016 | tmp_path = alloc_path_for_send(); | |
1017 | if (!tmp_path) { | |
924794c9 | 1018 | fs_path_free(p); |
31db9f7c AB |
1019 | return -ENOMEM; |
1020 | } | |
1021 | ||
31db9f7c | 1022 | |
96b5bd77 JS |
1023 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
1024 | ptr = (unsigned long)btrfs_item_ptr(eb, slot, | |
1025 | struct btrfs_inode_ref); | |
3212fa14 | 1026 | total = btrfs_item_size(eb, slot); |
96b5bd77 JS |
1027 | elem_size = sizeof(*iref); |
1028 | } else { | |
1029 | ptr = btrfs_item_ptr_offset(eb, slot); | |
3212fa14 | 1030 | total = btrfs_item_size(eb, slot); |
96b5bd77 JS |
1031 | elem_size = sizeof(*extref); |
1032 | } | |
1033 | ||
31db9f7c AB |
1034 | while (cur < total) { |
1035 | fs_path_reset(p); | |
1036 | ||
96b5bd77 JS |
1037 | if (found_key->type == BTRFS_INODE_REF_KEY) { |
1038 | iref = (struct btrfs_inode_ref *)(ptr + cur); | |
1039 | name_len = btrfs_inode_ref_name_len(eb, iref); | |
1040 | name_off = (unsigned long)(iref + 1); | |
1041 | index = btrfs_inode_ref_index(eb, iref); | |
1042 | dir = found_key->offset; | |
1043 | } else { | |
1044 | extref = (struct btrfs_inode_extref *)(ptr + cur); | |
1045 | name_len = btrfs_inode_extref_name_len(eb, extref); | |
1046 | name_off = (unsigned long)&extref->name; | |
1047 | index = btrfs_inode_extref_index(eb, extref); | |
1048 | dir = btrfs_inode_extref_parent(eb, extref); | |
1049 | } | |
1050 | ||
31db9f7c | 1051 | if (resolve) { |
96b5bd77 JS |
1052 | start = btrfs_ref_to_path(root, tmp_path, name_len, |
1053 | name_off, eb, dir, | |
1054 | p->buf, p->buf_len); | |
31db9f7c AB |
1055 | if (IS_ERR(start)) { |
1056 | ret = PTR_ERR(start); | |
1057 | goto out; | |
1058 | } | |
1059 | if (start < p->buf) { | |
1060 | /* overflow , try again with larger buffer */ | |
1061 | ret = fs_path_ensure_buf(p, | |
1062 | p->buf_len + p->buf - start); | |
1063 | if (ret < 0) | |
1064 | goto out; | |
96b5bd77 JS |
1065 | start = btrfs_ref_to_path(root, tmp_path, |
1066 | name_len, name_off, | |
1067 | eb, dir, | |
1068 | p->buf, p->buf_len); | |
31db9f7c AB |
1069 | if (IS_ERR(start)) { |
1070 | ret = PTR_ERR(start); | |
1071 | goto out; | |
1072 | } | |
1073 | BUG_ON(start < p->buf); | |
1074 | } | |
1075 | p->start = start; | |
1076 | } else { | |
96b5bd77 JS |
1077 | ret = fs_path_add_from_extent_buffer(p, eb, name_off, |
1078 | name_len); | |
31db9f7c AB |
1079 | if (ret < 0) |
1080 | goto out; | |
1081 | } | |
1082 | ||
96b5bd77 JS |
1083 | cur += elem_size + name_len; |
1084 | ret = iterate(num, dir, index, p, ctx); | |
31db9f7c AB |
1085 | if (ret) |
1086 | goto out; | |
31db9f7c AB |
1087 | num++; |
1088 | } | |
1089 | ||
1090 | out: | |
1091 | btrfs_free_path(tmp_path); | |
924794c9 | 1092 | fs_path_free(p); |
31db9f7c AB |
1093 | return ret; |
1094 | } | |
1095 | ||
1096 | typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, | |
1097 | const char *name, int name_len, | |
1098 | const char *data, int data_len, | |
b1dea4e7 | 1099 | void *ctx); |
31db9f7c AB |
1100 | |
1101 | /* | |
1102 | * Helper function to iterate the entries in ONE btrfs_dir_item. | |
1103 | * The iterate callback may return a non zero value to stop iteration. This can | |
1104 | * be a negative value for error codes or 1 to simply stop it. | |
1105 | * | |
1106 | * path must point to the dir item when called. | |
1107 | */ | |
924794c9 | 1108 | static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, |
31db9f7c AB |
1109 | iterate_dir_item_t iterate, void *ctx) |
1110 | { | |
1111 | int ret = 0; | |
1112 | struct extent_buffer *eb; | |
31db9f7c | 1113 | struct btrfs_dir_item *di; |
31db9f7c AB |
1114 | struct btrfs_key di_key; |
1115 | char *buf = NULL; | |
7e3ae33e | 1116 | int buf_len; |
31db9f7c AB |
1117 | u32 name_len; |
1118 | u32 data_len; | |
1119 | u32 cur; | |
1120 | u32 len; | |
1121 | u32 total; | |
1122 | int slot; | |
1123 | int num; | |
31db9f7c | 1124 | |
4395e0c4 FM |
1125 | /* |
1126 | * Start with a small buffer (1 page). If later we end up needing more | |
1127 | * space, which can happen for xattrs on a fs with a leaf size greater | |
1128 | * then the page size, attempt to increase the buffer. Typically xattr | |
1129 | * values are small. | |
1130 | */ | |
1131 | buf_len = PATH_MAX; | |
e780b0d1 | 1132 | buf = kmalloc(buf_len, GFP_KERNEL); |
31db9f7c AB |
1133 | if (!buf) { |
1134 | ret = -ENOMEM; | |
1135 | goto out; | |
1136 | } | |
1137 | ||
31db9f7c AB |
1138 | eb = path->nodes[0]; |
1139 | slot = path->slots[0]; | |
31db9f7c AB |
1140 | di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); |
1141 | cur = 0; | |
1142 | len = 0; | |
3212fa14 | 1143 | total = btrfs_item_size(eb, slot); |
31db9f7c AB |
1144 | |
1145 | num = 0; | |
1146 | while (cur < total) { | |
1147 | name_len = btrfs_dir_name_len(eb, di); | |
1148 | data_len = btrfs_dir_data_len(eb, di); | |
31db9f7c AB |
1149 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); |
1150 | ||
94a48aef | 1151 | if (btrfs_dir_ftype(eb, di) == BTRFS_FT_XATTR) { |
7e3ae33e FM |
1152 | if (name_len > XATTR_NAME_MAX) { |
1153 | ret = -ENAMETOOLONG; | |
1154 | goto out; | |
1155 | } | |
da17066c JM |
1156 | if (name_len + data_len > |
1157 | BTRFS_MAX_XATTR_SIZE(root->fs_info)) { | |
7e3ae33e FM |
1158 | ret = -E2BIG; |
1159 | goto out; | |
1160 | } | |
1161 | } else { | |
1162 | /* | |
1163 | * Path too long | |
1164 | */ | |
4395e0c4 | 1165 | if (name_len + data_len > PATH_MAX) { |
7e3ae33e FM |
1166 | ret = -ENAMETOOLONG; |
1167 | goto out; | |
1168 | } | |
31db9f7c AB |
1169 | } |
1170 | ||
4395e0c4 FM |
1171 | if (name_len + data_len > buf_len) { |
1172 | buf_len = name_len + data_len; | |
1173 | if (is_vmalloc_addr(buf)) { | |
1174 | vfree(buf); | |
1175 | buf = NULL; | |
1176 | } else { | |
1177 | char *tmp = krealloc(buf, buf_len, | |
e780b0d1 | 1178 | GFP_KERNEL | __GFP_NOWARN); |
4395e0c4 FM |
1179 | |
1180 | if (!tmp) | |
1181 | kfree(buf); | |
1182 | buf = tmp; | |
1183 | } | |
1184 | if (!buf) { | |
f11f7441 | 1185 | buf = kvmalloc(buf_len, GFP_KERNEL); |
4395e0c4 FM |
1186 | if (!buf) { |
1187 | ret = -ENOMEM; | |
1188 | goto out; | |
1189 | } | |
1190 | } | |
1191 | } | |
1192 | ||
31db9f7c AB |
1193 | read_extent_buffer(eb, buf, (unsigned long)(di + 1), |
1194 | name_len + data_len); | |
1195 | ||
1196 | len = sizeof(*di) + name_len + data_len; | |
1197 | di = (struct btrfs_dir_item *)((char *)di + len); | |
1198 | cur += len; | |
1199 | ||
1200 | ret = iterate(num, &di_key, buf, name_len, buf + name_len, | |
b1dea4e7 | 1201 | data_len, ctx); |
31db9f7c AB |
1202 | if (ret < 0) |
1203 | goto out; | |
1204 | if (ret) { | |
1205 | ret = 0; | |
1206 | goto out; | |
1207 | } | |
1208 | ||
1209 | num++; | |
1210 | } | |
1211 | ||
1212 | out: | |
4395e0c4 | 1213 | kvfree(buf); |
31db9f7c AB |
1214 | return ret; |
1215 | } | |
1216 | ||
1217 | static int __copy_first_ref(int num, u64 dir, int index, | |
1218 | struct fs_path *p, void *ctx) | |
1219 | { | |
1220 | int ret; | |
1221 | struct fs_path *pt = ctx; | |
1222 | ||
1223 | ret = fs_path_copy(pt, p); | |
1224 | if (ret < 0) | |
1225 | return ret; | |
1226 | ||
1227 | /* we want the first only */ | |
1228 | return 1; | |
1229 | } | |
1230 | ||
1231 | /* | |
1232 | * Retrieve the first path of an inode. If an inode has more then one | |
1233 | * ref/hardlink, this is ignored. | |
1234 | */ | |
924794c9 | 1235 | static int get_inode_path(struct btrfs_root *root, |
31db9f7c AB |
1236 | u64 ino, struct fs_path *path) |
1237 | { | |
1238 | int ret; | |
1239 | struct btrfs_key key, found_key; | |
1240 | struct btrfs_path *p; | |
1241 | ||
1242 | p = alloc_path_for_send(); | |
1243 | if (!p) | |
1244 | return -ENOMEM; | |
1245 | ||
1246 | fs_path_reset(path); | |
1247 | ||
1248 | key.objectid = ino; | |
1249 | key.type = BTRFS_INODE_REF_KEY; | |
1250 | key.offset = 0; | |
1251 | ||
1252 | ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); | |
1253 | if (ret < 0) | |
1254 | goto out; | |
1255 | if (ret) { | |
1256 | ret = 1; | |
1257 | goto out; | |
1258 | } | |
1259 | btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); | |
1260 | if (found_key.objectid != ino || | |
96b5bd77 JS |
1261 | (found_key.type != BTRFS_INODE_REF_KEY && |
1262 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
1263 | ret = -ENOENT; |
1264 | goto out; | |
1265 | } | |
1266 | ||
924794c9 TI |
1267 | ret = iterate_inode_ref(root, p, &found_key, 1, |
1268 | __copy_first_ref, path); | |
31db9f7c AB |
1269 | if (ret < 0) |
1270 | goto out; | |
1271 | ret = 0; | |
1272 | ||
1273 | out: | |
1274 | btrfs_free_path(p); | |
1275 | return ret; | |
1276 | } | |
1277 | ||
1278 | struct backref_ctx { | |
1279 | struct send_ctx *sctx; | |
1280 | ||
1281 | /* number of total found references */ | |
1282 | u64 found; | |
1283 | ||
1284 | /* | |
1285 | * used for clones found in send_root. clones found behind cur_objectid | |
1286 | * and cur_offset are not considered as allowed clones. | |
1287 | */ | |
1288 | u64 cur_objectid; | |
1289 | u64 cur_offset; | |
1290 | ||
1291 | /* may be truncated in case it's the last extent in a file */ | |
1292 | u64 extent_len; | |
f73853c7 FM |
1293 | |
1294 | /* The bytenr the file extent item we are processing refers to. */ | |
1295 | u64 bytenr; | |
adf02418 FM |
1296 | /* The owner (root id) of the data backref for the current extent. */ |
1297 | u64 backref_owner; | |
1298 | /* The offset of the data backref for the current extent. */ | |
1299 | u64 backref_offset; | |
31db9f7c AB |
1300 | }; |
1301 | ||
1302 | static int __clone_root_cmp_bsearch(const void *key, const void *elt) | |
1303 | { | |
995e01b7 | 1304 | u64 root = (u64)(uintptr_t)key; |
214cc184 | 1305 | const struct clone_root *cr = elt; |
31db9f7c | 1306 | |
4fd786e6 | 1307 | if (root < cr->root->root_key.objectid) |
31db9f7c | 1308 | return -1; |
4fd786e6 | 1309 | if (root > cr->root->root_key.objectid) |
31db9f7c AB |
1310 | return 1; |
1311 | return 0; | |
1312 | } | |
1313 | ||
1314 | static int __clone_root_cmp_sort(const void *e1, const void *e2) | |
1315 | { | |
214cc184 DS |
1316 | const struct clone_root *cr1 = e1; |
1317 | const struct clone_root *cr2 = e2; | |
31db9f7c | 1318 | |
4fd786e6 | 1319 | if (cr1->root->root_key.objectid < cr2->root->root_key.objectid) |
31db9f7c | 1320 | return -1; |
4fd786e6 | 1321 | if (cr1->root->root_key.objectid > cr2->root->root_key.objectid) |
31db9f7c AB |
1322 | return 1; |
1323 | return 0; | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Called for every backref that is found for the current extent. | |
88ffb665 | 1328 | * Results are collected in sctx->clone_roots->ino/offset. |
31db9f7c | 1329 | */ |
88ffb665 FM |
1330 | static int iterate_backrefs(u64 ino, u64 offset, u64 num_bytes, u64 root_id, |
1331 | void *ctx_) | |
31db9f7c AB |
1332 | { |
1333 | struct backref_ctx *bctx = ctx_; | |
88ffb665 | 1334 | struct clone_root *clone_root; |
31db9f7c AB |
1335 | |
1336 | /* First check if the root is in the list of accepted clone sources */ | |
88ffb665 FM |
1337 | clone_root = bsearch((void *)(uintptr_t)root_id, bctx->sctx->clone_roots, |
1338 | bctx->sctx->clone_roots_cnt, | |
1339 | sizeof(struct clone_root), | |
1340 | __clone_root_cmp_bsearch); | |
1341 | if (!clone_root) | |
31db9f7c AB |
1342 | return 0; |
1343 | ||
88ffb665 FM |
1344 | /* This is our own reference, bail out as we can't clone from it. */ |
1345 | if (clone_root->root == bctx->sctx->send_root && | |
31db9f7c | 1346 | ino == bctx->cur_objectid && |
88ffb665 FM |
1347 | offset == bctx->cur_offset) |
1348 | return 0; | |
31db9f7c | 1349 | |
31db9f7c AB |
1350 | /* |
1351 | * Make sure we don't consider clones from send_root that are | |
1352 | * behind the current inode/offset. | |
1353 | */ | |
88ffb665 | 1354 | if (clone_root->root == bctx->sctx->send_root) { |
31db9f7c | 1355 | /* |
11f2069c FM |
1356 | * If the source inode was not yet processed we can't issue a |
1357 | * clone operation, as the source extent does not exist yet at | |
1358 | * the destination of the stream. | |
31db9f7c | 1359 | */ |
11f2069c FM |
1360 | if (ino > bctx->cur_objectid) |
1361 | return 0; | |
1362 | /* | |
1363 | * We clone from the inode currently being sent as long as the | |
1364 | * source extent is already processed, otherwise we could try | |
1365 | * to clone from an extent that does not exist yet at the | |
1366 | * destination of the stream. | |
1367 | */ | |
1368 | if (ino == bctx->cur_objectid && | |
9722b101 FM |
1369 | offset + bctx->extent_len > |
1370 | bctx->sctx->cur_inode_next_write_offset) | |
31db9f7c | 1371 | return 0; |
31db9f7c AB |
1372 | } |
1373 | ||
1374 | bctx->found++; | |
88ffb665 | 1375 | clone_root->found_ref = true; |
c7499a64 FM |
1376 | |
1377 | /* | |
1378 | * If the given backref refers to a file extent item with a larger | |
1379 | * number of bytes than what we found before, use the new one so that | |
1380 | * we clone more optimally and end up doing less writes and getting | |
1381 | * less exclusive, non-shared extents at the destination. | |
1382 | */ | |
88ffb665 FM |
1383 | if (num_bytes > clone_root->num_bytes) { |
1384 | clone_root->ino = ino; | |
1385 | clone_root->offset = offset; | |
1386 | clone_root->num_bytes = num_bytes; | |
1387 | ||
1388 | /* | |
1389 | * Found a perfect candidate, so there's no need to continue | |
1390 | * backref walking. | |
1391 | */ | |
1392 | if (num_bytes >= bctx->extent_len) | |
1393 | return BTRFS_ITERATE_EXTENT_INODES_STOP; | |
31db9f7c AB |
1394 | } |
1395 | ||
1396 | return 0; | |
1397 | } | |
1398 | ||
66d04209 FM |
1399 | static bool lookup_backref_cache(u64 leaf_bytenr, void *ctx, |
1400 | const u64 **root_ids_ret, int *root_count_ret) | |
1401 | { | |
88ffb665 FM |
1402 | struct backref_ctx *bctx = ctx; |
1403 | struct send_ctx *sctx = bctx->sctx; | |
66d04209 FM |
1404 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
1405 | const u64 key = leaf_bytenr >> fs_info->sectorsize_bits; | |
90b90d4a | 1406 | struct btrfs_lru_cache_entry *raw_entry; |
66d04209 FM |
1407 | struct backref_cache_entry *entry; |
1408 | ||
90b90d4a | 1409 | if (btrfs_lru_cache_size(&sctx->backref_cache) == 0) |
66d04209 FM |
1410 | return false; |
1411 | ||
1412 | /* | |
1413 | * If relocation happened since we first filled the cache, then we must | |
1414 | * empty the cache and can not use it, because even though we operate on | |
1415 | * read-only roots, their leaves and nodes may have been reallocated and | |
1416 | * now be used for different nodes/leaves of the same tree or some other | |
1417 | * tree. | |
1418 | * | |
1419 | * We are called from iterate_extent_inodes() while either holding a | |
1420 | * transaction handle or holding fs_info->commit_root_sem, so no need | |
1421 | * to take any lock here. | |
1422 | */ | |
90b90d4a FM |
1423 | if (fs_info->last_reloc_trans > sctx->backref_cache_last_reloc_trans) { |
1424 | btrfs_lru_cache_clear(&sctx->backref_cache); | |
66d04209 FM |
1425 | return false; |
1426 | } | |
1427 | ||
0da0c560 | 1428 | raw_entry = btrfs_lru_cache_lookup(&sctx->backref_cache, key, 0); |
90b90d4a | 1429 | if (!raw_entry) |
66d04209 FM |
1430 | return false; |
1431 | ||
90b90d4a | 1432 | entry = container_of(raw_entry, struct backref_cache_entry, entry); |
66d04209 FM |
1433 | *root_ids_ret = entry->root_ids; |
1434 | *root_count_ret = entry->num_roots; | |
66d04209 FM |
1435 | |
1436 | return true; | |
1437 | } | |
1438 | ||
1439 | static void store_backref_cache(u64 leaf_bytenr, const struct ulist *root_ids, | |
1440 | void *ctx) | |
1441 | { | |
88ffb665 FM |
1442 | struct backref_ctx *bctx = ctx; |
1443 | struct send_ctx *sctx = bctx->sctx; | |
66d04209 FM |
1444 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
1445 | struct backref_cache_entry *new_entry; | |
1446 | struct ulist_iterator uiter; | |
1447 | struct ulist_node *node; | |
1448 | int ret; | |
1449 | ||
1450 | /* | |
1451 | * We're called while holding a transaction handle or while holding | |
1452 | * fs_info->commit_root_sem (at iterate_extent_inodes()), so must do a | |
1453 | * NOFS allocation. | |
1454 | */ | |
1455 | new_entry = kmalloc(sizeof(struct backref_cache_entry), GFP_NOFS); | |
1456 | /* No worries, cache is optional. */ | |
1457 | if (!new_entry) | |
1458 | return; | |
1459 | ||
90b90d4a | 1460 | new_entry->entry.key = leaf_bytenr >> fs_info->sectorsize_bits; |
0da0c560 | 1461 | new_entry->entry.gen = 0; |
66d04209 FM |
1462 | new_entry->num_roots = 0; |
1463 | ULIST_ITER_INIT(&uiter); | |
1464 | while ((node = ulist_next(root_ids, &uiter)) != NULL) { | |
1465 | const u64 root_id = node->val; | |
1466 | struct clone_root *root; | |
1467 | ||
1468 | root = bsearch((void *)(uintptr_t)root_id, sctx->clone_roots, | |
1469 | sctx->clone_roots_cnt, sizeof(struct clone_root), | |
1470 | __clone_root_cmp_bsearch); | |
1471 | if (!root) | |
1472 | continue; | |
1473 | ||
1474 | /* Too many roots, just exit, no worries as caching is optional. */ | |
1475 | if (new_entry->num_roots >= SEND_MAX_BACKREF_CACHE_ROOTS) { | |
1476 | kfree(new_entry); | |
1477 | return; | |
1478 | } | |
1479 | ||
1480 | new_entry->root_ids[new_entry->num_roots] = root_id; | |
1481 | new_entry->num_roots++; | |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * We may have not added any roots to the new cache entry, which means | |
1486 | * none of the roots is part of the list of roots from which we are | |
1487 | * allowed to clone. Cache the new entry as it's still useful to avoid | |
1488 | * backref walking to determine which roots have a path to the leaf. | |
90b90d4a FM |
1489 | * |
1490 | * Also use GFP_NOFS because we're called while holding a transaction | |
1491 | * handle or while holding fs_info->commit_root_sem. | |
66d04209 | 1492 | */ |
90b90d4a FM |
1493 | ret = btrfs_lru_cache_store(&sctx->backref_cache, &new_entry->entry, |
1494 | GFP_NOFS); | |
66d04209 FM |
1495 | ASSERT(ret == 0 || ret == -ENOMEM); |
1496 | if (ret) { | |
1497 | /* Caching is optional, no worries. */ | |
1498 | kfree(new_entry); | |
1499 | return; | |
1500 | } | |
1501 | ||
66d04209 FM |
1502 | /* |
1503 | * We are called from iterate_extent_inodes() while either holding a | |
1504 | * transaction handle or holding fs_info->commit_root_sem, so no need | |
1505 | * to take any lock here. | |
1506 | */ | |
90b90d4a FM |
1507 | if (btrfs_lru_cache_size(&sctx->backref_cache) == 1) |
1508 | sctx->backref_cache_last_reloc_trans = fs_info->last_reloc_trans; | |
66d04209 FM |
1509 | } |
1510 | ||
f73853c7 FM |
1511 | static int check_extent_item(u64 bytenr, const struct btrfs_extent_item *ei, |
1512 | const struct extent_buffer *leaf, void *ctx) | |
1513 | { | |
1514 | const u64 refs = btrfs_extent_refs(leaf, ei); | |
1515 | const struct backref_ctx *bctx = ctx; | |
1516 | const struct send_ctx *sctx = bctx->sctx; | |
1517 | ||
1518 | if (bytenr == bctx->bytenr) { | |
1519 | const u64 flags = btrfs_extent_flags(leaf, ei); | |
1520 | ||
1521 | if (WARN_ON(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) | |
1522 | return -EUCLEAN; | |
1523 | ||
1524 | /* | |
1525 | * If we have only one reference and only the send root as a | |
1526 | * clone source - meaning no clone roots were given in the | |
1527 | * struct btrfs_ioctl_send_args passed to the send ioctl - then | |
1528 | * it's our reference and there's no point in doing backref | |
1529 | * walking which is expensive, so exit early. | |
1530 | */ | |
1531 | if (refs == 1 && sctx->clone_roots_cnt == 1) | |
1532 | return -ENOENT; | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | * Backreference walking (iterate_extent_inodes() below) is currently | |
1537 | * too expensive when an extent has a large number of references, both | |
1538 | * in time spent and used memory. So for now just fallback to write | |
1539 | * operations instead of clone operations when an extent has more than | |
1540 | * a certain amount of references. | |
1541 | */ | |
1542 | if (refs > SEND_MAX_EXTENT_REFS) | |
1543 | return -ENOENT; | |
1544 | ||
1545 | return 0; | |
1546 | } | |
1547 | ||
adf02418 FM |
1548 | static bool skip_self_data_ref(u64 root, u64 ino, u64 offset, void *ctx) |
1549 | { | |
1550 | const struct backref_ctx *bctx = ctx; | |
1551 | ||
1552 | if (ino == bctx->cur_objectid && | |
1553 | root == bctx->backref_owner && | |
1554 | offset == bctx->backref_offset) | |
1555 | return true; | |
1556 | ||
1557 | return false; | |
1558 | } | |
1559 | ||
31db9f7c | 1560 | /* |
766702ef AB |
1561 | * Given an inode, offset and extent item, it finds a good clone for a clone |
1562 | * instruction. Returns -ENOENT when none could be found. The function makes | |
1563 | * sure that the returned clone is usable at the point where sending is at the | |
1564 | * moment. This means, that no clones are accepted which lie behind the current | |
1565 | * inode+offset. | |
1566 | * | |
31db9f7c AB |
1567 | * path must point to the extent item when called. |
1568 | */ | |
1569 | static int find_extent_clone(struct send_ctx *sctx, | |
1570 | struct btrfs_path *path, | |
1571 | u64 ino, u64 data_offset, | |
1572 | u64 ino_size, | |
1573 | struct clone_root **found) | |
1574 | { | |
04ab956e | 1575 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
1576 | int ret; |
1577 | int extent_type; | |
1578 | u64 logical; | |
74dd17fb | 1579 | u64 disk_byte; |
31db9f7c | 1580 | u64 num_bytes; |
31db9f7c AB |
1581 | struct btrfs_file_extent_item *fi; |
1582 | struct extent_buffer *eb = path->nodes[0]; | |
a2c8d27e FM |
1583 | struct backref_ctx backref_ctx = { 0 }; |
1584 | struct btrfs_backref_walk_ctx backref_walk_ctx = { 0 }; | |
31db9f7c | 1585 | struct clone_root *cur_clone_root; |
74dd17fb | 1586 | int compressed; |
31db9f7c AB |
1587 | u32 i; |
1588 | ||
d3f41317 FM |
1589 | /* |
1590 | * With fallocate we can get prealloc extents beyond the inode's i_size, | |
1591 | * so we don't do anything here because clone operations can not clone | |
1592 | * to a range beyond i_size without increasing the i_size of the | |
1593 | * destination inode. | |
1594 | */ | |
1595 | if (data_offset >= ino_size) | |
61ce908a | 1596 | return 0; |
31db9f7c | 1597 | |
61ce908a | 1598 | fi = btrfs_item_ptr(eb, path->slots[0], struct btrfs_file_extent_item); |
31db9f7c | 1599 | extent_type = btrfs_file_extent_type(eb, fi); |
61ce908a FM |
1600 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) |
1601 | return -ENOENT; | |
31db9f7c | 1602 | |
74dd17fb | 1603 | disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); |
61ce908a FM |
1604 | if (disk_byte == 0) |
1605 | return -ENOENT; | |
1606 | ||
1607 | compressed = btrfs_file_extent_compression(eb, fi); | |
1608 | num_bytes = btrfs_file_extent_num_bytes(eb, fi); | |
74dd17fb | 1609 | logical = disk_byte + btrfs_file_extent_offset(eb, fi); |
31db9f7c | 1610 | |
31db9f7c AB |
1611 | /* |
1612 | * Setup the clone roots. | |
1613 | */ | |
1614 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
1615 | cur_clone_root = sctx->clone_roots + i; | |
1616 | cur_clone_root->ino = (u64)-1; | |
1617 | cur_clone_root->offset = 0; | |
c7499a64 | 1618 | cur_clone_root->num_bytes = 0; |
88ffb665 | 1619 | cur_clone_root->found_ref = false; |
31db9f7c AB |
1620 | } |
1621 | ||
dce28150 | 1622 | backref_ctx.sctx = sctx; |
dce28150 GR |
1623 | backref_ctx.cur_objectid = ino; |
1624 | backref_ctx.cur_offset = data_offset; | |
f73853c7 | 1625 | backref_ctx.bytenr = disk_byte; |
adf02418 FM |
1626 | /* |
1627 | * Use the header owner and not the send root's id, because in case of a | |
1628 | * snapshot we can have shared subtrees. | |
1629 | */ | |
1630 | backref_ctx.backref_owner = btrfs_header_owner(eb); | |
1631 | backref_ctx.backref_offset = data_offset - btrfs_file_extent_offset(eb, fi); | |
31db9f7c AB |
1632 | |
1633 | /* | |
1634 | * The last extent of a file may be too large due to page alignment. | |
1635 | * We need to adjust extent_len in this case so that the checks in | |
88ffb665 | 1636 | * iterate_backrefs() work. |
31db9f7c AB |
1637 | */ |
1638 | if (data_offset + num_bytes >= ino_size) | |
dce28150 | 1639 | backref_ctx.extent_len = ino_size - data_offset; |
344174a1 FM |
1640 | else |
1641 | backref_ctx.extent_len = num_bytes; | |
31db9f7c AB |
1642 | |
1643 | /* | |
1644 | * Now collect all backrefs. | |
1645 | */ | |
f73853c7 | 1646 | backref_walk_ctx.bytenr = disk_byte; |
74dd17fb | 1647 | if (compressed == BTRFS_COMPRESS_NONE) |
f73853c7 | 1648 | backref_walk_ctx.extent_item_pos = btrfs_file_extent_offset(eb, fi); |
a2c8d27e | 1649 | backref_walk_ctx.fs_info = fs_info; |
66d04209 FM |
1650 | backref_walk_ctx.cache_lookup = lookup_backref_cache; |
1651 | backref_walk_ctx.cache_store = store_backref_cache; | |
88ffb665 | 1652 | backref_walk_ctx.indirect_ref_iterator = iterate_backrefs; |
f73853c7 | 1653 | backref_walk_ctx.check_extent_item = check_extent_item; |
88ffb665 | 1654 | backref_walk_ctx.user_ctx = &backref_ctx; |
74dd17fb | 1655 | |
adf02418 FM |
1656 | /* |
1657 | * If have a single clone root, then it's the send root and we can tell | |
1658 | * the backref walking code to skip our own backref and not resolve it, | |
1659 | * since we can not use it for cloning - the source and destination | |
1660 | * ranges can't overlap and in case the leaf is shared through a subtree | |
1661 | * due to snapshots, we can't use those other roots since they are not | |
1662 | * in the list of clone roots. | |
1663 | */ | |
1664 | if (sctx->clone_roots_cnt == 1) | |
1665 | backref_walk_ctx.skip_data_ref = skip_self_data_ref; | |
1666 | ||
88ffb665 | 1667 | ret = iterate_extent_inodes(&backref_walk_ctx, true, iterate_backrefs, |
a2c8d27e | 1668 | &backref_ctx); |
31db9f7c | 1669 | if (ret < 0) |
f73853c7 | 1670 | return ret; |
31db9f7c | 1671 | |
d96b3424 FM |
1672 | down_read(&fs_info->commit_root_sem); |
1673 | if (fs_info->last_reloc_trans > sctx->last_reloc_trans) { | |
1674 | /* | |
1675 | * A transaction commit for a transaction in which block group | |
1676 | * relocation was done just happened. | |
1677 | * The disk_bytenr of the file extent item we processed is | |
1678 | * possibly stale, referring to the extent's location before | |
1679 | * relocation. So act as if we haven't found any clone sources | |
1680 | * and fallback to write commands, which will read the correct | |
1681 | * data from the new extent location. Otherwise we will fail | |
1682 | * below because we haven't found our own back reference or we | |
1683 | * could be getting incorrect sources in case the old extent | |
1684 | * was already reallocated after the relocation. | |
1685 | */ | |
1686 | up_read(&fs_info->commit_root_sem); | |
f73853c7 | 1687 | return -ENOENT; |
d96b3424 FM |
1688 | } |
1689 | up_read(&fs_info->commit_root_sem); | |
1690 | ||
04ab956e JM |
1691 | btrfs_debug(fs_info, |
1692 | "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu", | |
1693 | data_offset, ino, num_bytes, logical); | |
31db9f7c | 1694 | |
88ffb665 | 1695 | if (!backref_ctx.found) { |
04ab956e | 1696 | btrfs_debug(fs_info, "no clones found"); |
f73853c7 | 1697 | return -ENOENT; |
88ffb665 | 1698 | } |
31db9f7c AB |
1699 | |
1700 | cur_clone_root = NULL; | |
1701 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
c7499a64 | 1702 | struct clone_root *clone_root = &sctx->clone_roots[i]; |
31db9f7c | 1703 | |
88ffb665 | 1704 | if (!clone_root->found_ref) |
c7499a64 FM |
1705 | continue; |
1706 | ||
1707 | /* | |
1708 | * Choose the root from which we can clone more bytes, to | |
1709 | * minimize write operations and therefore have more extent | |
1710 | * sharing at the destination (the same as in the source). | |
1711 | */ | |
1712 | if (!cur_clone_root || | |
1713 | clone_root->num_bytes > cur_clone_root->num_bytes) { | |
1714 | cur_clone_root = clone_root; | |
1715 | ||
1716 | /* | |
1717 | * We found an optimal clone candidate (any inode from | |
1718 | * any root is fine), so we're done. | |
1719 | */ | |
1720 | if (clone_root->num_bytes >= backref_ctx.extent_len) | |
1721 | break; | |
1722 | } | |
31db9f7c AB |
1723 | } |
1724 | ||
1725 | if (cur_clone_root) { | |
1726 | *found = cur_clone_root; | |
1727 | ret = 0; | |
1728 | } else { | |
1729 | ret = -ENOENT; | |
1730 | } | |
1731 | ||
31db9f7c AB |
1732 | return ret; |
1733 | } | |
1734 | ||
924794c9 | 1735 | static int read_symlink(struct btrfs_root *root, |
31db9f7c AB |
1736 | u64 ino, |
1737 | struct fs_path *dest) | |
1738 | { | |
1739 | int ret; | |
1740 | struct btrfs_path *path; | |
1741 | struct btrfs_key key; | |
1742 | struct btrfs_file_extent_item *ei; | |
1743 | u8 type; | |
1744 | u8 compression; | |
1745 | unsigned long off; | |
1746 | int len; | |
1747 | ||
1748 | path = alloc_path_for_send(); | |
1749 | if (!path) | |
1750 | return -ENOMEM; | |
1751 | ||
1752 | key.objectid = ino; | |
1753 | key.type = BTRFS_EXTENT_DATA_KEY; | |
1754 | key.offset = 0; | |
1755 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1756 | if (ret < 0) | |
1757 | goto out; | |
a879719b FM |
1758 | if (ret) { |
1759 | /* | |
1760 | * An empty symlink inode. Can happen in rare error paths when | |
1761 | * creating a symlink (transaction committed before the inode | |
1762 | * eviction handler removed the symlink inode items and a crash | |
1763 | * happened in between or the subvol was snapshoted in between). | |
1764 | * Print an informative message to dmesg/syslog so that the user | |
1765 | * can delete the symlink. | |
1766 | */ | |
1767 | btrfs_err(root->fs_info, | |
1768 | "Found empty symlink inode %llu at root %llu", | |
1769 | ino, root->root_key.objectid); | |
1770 | ret = -EIO; | |
1771 | goto out; | |
1772 | } | |
31db9f7c AB |
1773 | |
1774 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1775 | struct btrfs_file_extent_item); | |
1776 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
1777 | compression = btrfs_file_extent_compression(path->nodes[0], ei); | |
1778 | BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); | |
1779 | BUG_ON(compression); | |
1780 | ||
1781 | off = btrfs_file_extent_inline_start(ei); | |
e41ca589 | 1782 | len = btrfs_file_extent_ram_bytes(path->nodes[0], ei); |
31db9f7c AB |
1783 | |
1784 | ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); | |
31db9f7c AB |
1785 | |
1786 | out: | |
1787 | btrfs_free_path(path); | |
1788 | return ret; | |
1789 | } | |
1790 | ||
1791 | /* | |
1792 | * Helper function to generate a file name that is unique in the root of | |
1793 | * send_root and parent_root. This is used to generate names for orphan inodes. | |
1794 | */ | |
1795 | static int gen_unique_name(struct send_ctx *sctx, | |
1796 | u64 ino, u64 gen, | |
1797 | struct fs_path *dest) | |
1798 | { | |
1799 | int ret = 0; | |
1800 | struct btrfs_path *path; | |
1801 | struct btrfs_dir_item *di; | |
1802 | char tmp[64]; | |
1803 | int len; | |
1804 | u64 idx = 0; | |
1805 | ||
1806 | path = alloc_path_for_send(); | |
1807 | if (!path) | |
1808 | return -ENOMEM; | |
1809 | ||
1810 | while (1) { | |
6db75318 | 1811 | struct fscrypt_str tmp_name; |
e43eec81 | 1812 | |
f74b86d8 | 1813 | len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", |
31db9f7c | 1814 | ino, gen, idx); |
64792f25 | 1815 | ASSERT(len < sizeof(tmp)); |
e43eec81 STD |
1816 | tmp_name.name = tmp; |
1817 | tmp_name.len = strlen(tmp); | |
31db9f7c AB |
1818 | |
1819 | di = btrfs_lookup_dir_item(NULL, sctx->send_root, | |
1820 | path, BTRFS_FIRST_FREE_OBJECTID, | |
e43eec81 | 1821 | &tmp_name, 0); |
31db9f7c AB |
1822 | btrfs_release_path(path); |
1823 | if (IS_ERR(di)) { | |
1824 | ret = PTR_ERR(di); | |
1825 | goto out; | |
1826 | } | |
1827 | if (di) { | |
1828 | /* not unique, try again */ | |
1829 | idx++; | |
1830 | continue; | |
1831 | } | |
1832 | ||
1833 | if (!sctx->parent_root) { | |
1834 | /* unique */ | |
1835 | ret = 0; | |
1836 | break; | |
1837 | } | |
1838 | ||
1839 | di = btrfs_lookup_dir_item(NULL, sctx->parent_root, | |
1840 | path, BTRFS_FIRST_FREE_OBJECTID, | |
e43eec81 | 1841 | &tmp_name, 0); |
31db9f7c AB |
1842 | btrfs_release_path(path); |
1843 | if (IS_ERR(di)) { | |
1844 | ret = PTR_ERR(di); | |
1845 | goto out; | |
1846 | } | |
1847 | if (di) { | |
1848 | /* not unique, try again */ | |
1849 | idx++; | |
1850 | continue; | |
1851 | } | |
1852 | /* unique */ | |
1853 | break; | |
1854 | } | |
1855 | ||
1856 | ret = fs_path_add(dest, tmp, strlen(tmp)); | |
1857 | ||
1858 | out: | |
1859 | btrfs_free_path(path); | |
1860 | return ret; | |
1861 | } | |
1862 | ||
1863 | enum inode_state { | |
1864 | inode_state_no_change, | |
1865 | inode_state_will_create, | |
1866 | inode_state_did_create, | |
1867 | inode_state_will_delete, | |
1868 | inode_state_did_delete, | |
1869 | }; | |
1870 | ||
498581f3 FM |
1871 | static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen, |
1872 | u64 *send_gen, u64 *parent_gen) | |
31db9f7c AB |
1873 | { |
1874 | int ret; | |
1875 | int left_ret; | |
1876 | int right_ret; | |
1877 | u64 left_gen; | |
1878 | u64 right_gen; | |
9ed0a72e | 1879 | struct btrfs_inode_info info; |
31db9f7c | 1880 | |
9ed0a72e | 1881 | ret = get_inode_info(sctx->send_root, ino, &info); |
31db9f7c AB |
1882 | if (ret < 0 && ret != -ENOENT) |
1883 | goto out; | |
9ed0a72e BC |
1884 | left_ret = (info.nlink == 0) ? -ENOENT : ret; |
1885 | left_gen = info.gen; | |
498581f3 FM |
1886 | if (send_gen) |
1887 | *send_gen = ((left_ret == -ENOENT) ? 0 : info.gen); | |
31db9f7c AB |
1888 | |
1889 | if (!sctx->parent_root) { | |
1890 | right_ret = -ENOENT; | |
1891 | } else { | |
9ed0a72e | 1892 | ret = get_inode_info(sctx->parent_root, ino, &info); |
31db9f7c AB |
1893 | if (ret < 0 && ret != -ENOENT) |
1894 | goto out; | |
9ed0a72e BC |
1895 | right_ret = (info.nlink == 0) ? -ENOENT : ret; |
1896 | right_gen = info.gen; | |
498581f3 FM |
1897 | if (parent_gen) |
1898 | *parent_gen = ((right_ret == -ENOENT) ? 0 : info.gen); | |
31db9f7c AB |
1899 | } |
1900 | ||
1901 | if (!left_ret && !right_ret) { | |
e938c8ad | 1902 | if (left_gen == gen && right_gen == gen) { |
31db9f7c | 1903 | ret = inode_state_no_change; |
e938c8ad | 1904 | } else if (left_gen == gen) { |
31db9f7c AB |
1905 | if (ino < sctx->send_progress) |
1906 | ret = inode_state_did_create; | |
1907 | else | |
1908 | ret = inode_state_will_create; | |
1909 | } else if (right_gen == gen) { | |
1910 | if (ino < sctx->send_progress) | |
1911 | ret = inode_state_did_delete; | |
1912 | else | |
1913 | ret = inode_state_will_delete; | |
1914 | } else { | |
1915 | ret = -ENOENT; | |
1916 | } | |
1917 | } else if (!left_ret) { | |
1918 | if (left_gen == gen) { | |
1919 | if (ino < sctx->send_progress) | |
1920 | ret = inode_state_did_create; | |
1921 | else | |
1922 | ret = inode_state_will_create; | |
1923 | } else { | |
1924 | ret = -ENOENT; | |
1925 | } | |
1926 | } else if (!right_ret) { | |
1927 | if (right_gen == gen) { | |
1928 | if (ino < sctx->send_progress) | |
1929 | ret = inode_state_did_delete; | |
1930 | else | |
1931 | ret = inode_state_will_delete; | |
1932 | } else { | |
1933 | ret = -ENOENT; | |
1934 | } | |
1935 | } else { | |
1936 | ret = -ENOENT; | |
1937 | } | |
1938 | ||
1939 | out: | |
1940 | return ret; | |
1941 | } | |
1942 | ||
498581f3 FM |
1943 | static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen, |
1944 | u64 *send_gen, u64 *parent_gen) | |
31db9f7c AB |
1945 | { |
1946 | int ret; | |
1947 | ||
4dd9920d RK |
1948 | if (ino == BTRFS_FIRST_FREE_OBJECTID) |
1949 | return 1; | |
1950 | ||
498581f3 | 1951 | ret = get_cur_inode_state(sctx, ino, gen, send_gen, parent_gen); |
31db9f7c AB |
1952 | if (ret < 0) |
1953 | goto out; | |
1954 | ||
1955 | if (ret == inode_state_no_change || | |
1956 | ret == inode_state_did_create || | |
1957 | ret == inode_state_will_delete) | |
1958 | ret = 1; | |
1959 | else | |
1960 | ret = 0; | |
1961 | ||
1962 | out: | |
1963 | return ret; | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Helper function to lookup a dir item in a dir. | |
1968 | */ | |
1969 | static int lookup_dir_item_inode(struct btrfs_root *root, | |
1970 | u64 dir, const char *name, int name_len, | |
eab67c06 | 1971 | u64 *found_inode) |
31db9f7c AB |
1972 | { |
1973 | int ret = 0; | |
1974 | struct btrfs_dir_item *di; | |
1975 | struct btrfs_key key; | |
1976 | struct btrfs_path *path; | |
6db75318 | 1977 | struct fscrypt_str name_str = FSTR_INIT((char *)name, name_len); |
31db9f7c AB |
1978 | |
1979 | path = alloc_path_for_send(); | |
1980 | if (!path) | |
1981 | return -ENOMEM; | |
1982 | ||
e43eec81 | 1983 | di = btrfs_lookup_dir_item(NULL, root, path, dir, &name_str, 0); |
3cf5068f LB |
1984 | if (IS_ERR_OR_NULL(di)) { |
1985 | ret = di ? PTR_ERR(di) : -ENOENT; | |
31db9f7c AB |
1986 | goto out; |
1987 | } | |
1988 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); | |
1af56070 FM |
1989 | if (key.type == BTRFS_ROOT_ITEM_KEY) { |
1990 | ret = -ENOENT; | |
1991 | goto out; | |
1992 | } | |
31db9f7c | 1993 | *found_inode = key.objectid; |
31db9f7c AB |
1994 | |
1995 | out: | |
1996 | btrfs_free_path(path); | |
1997 | return ret; | |
1998 | } | |
1999 | ||
766702ef AB |
2000 | /* |
2001 | * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir, | |
2002 | * generation of the parent dir and the name of the dir entry. | |
2003 | */ | |
924794c9 | 2004 | static int get_first_ref(struct btrfs_root *root, u64 ino, |
31db9f7c AB |
2005 | u64 *dir, u64 *dir_gen, struct fs_path *name) |
2006 | { | |
2007 | int ret; | |
2008 | struct btrfs_key key; | |
2009 | struct btrfs_key found_key; | |
2010 | struct btrfs_path *path; | |
31db9f7c | 2011 | int len; |
96b5bd77 | 2012 | u64 parent_dir; |
31db9f7c AB |
2013 | |
2014 | path = alloc_path_for_send(); | |
2015 | if (!path) | |
2016 | return -ENOMEM; | |
2017 | ||
2018 | key.objectid = ino; | |
2019 | key.type = BTRFS_INODE_REF_KEY; | |
2020 | key.offset = 0; | |
2021 | ||
2022 | ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); | |
2023 | if (ret < 0) | |
2024 | goto out; | |
2025 | if (!ret) | |
2026 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
2027 | path->slots[0]); | |
96b5bd77 JS |
2028 | if (ret || found_key.objectid != ino || |
2029 | (found_key.type != BTRFS_INODE_REF_KEY && | |
2030 | found_key.type != BTRFS_INODE_EXTREF_KEY)) { | |
31db9f7c AB |
2031 | ret = -ENOENT; |
2032 | goto out; | |
2033 | } | |
2034 | ||
51a60253 | 2035 | if (found_key.type == BTRFS_INODE_REF_KEY) { |
96b5bd77 JS |
2036 | struct btrfs_inode_ref *iref; |
2037 | iref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2038 | struct btrfs_inode_ref); | |
2039 | len = btrfs_inode_ref_name_len(path->nodes[0], iref); | |
2040 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
2041 | (unsigned long)(iref + 1), | |
2042 | len); | |
2043 | parent_dir = found_key.offset; | |
2044 | } else { | |
2045 | struct btrfs_inode_extref *extref; | |
2046 | extref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2047 | struct btrfs_inode_extref); | |
2048 | len = btrfs_inode_extref_name_len(path->nodes[0], extref); | |
2049 | ret = fs_path_add_from_extent_buffer(name, path->nodes[0], | |
2050 | (unsigned long)&extref->name, len); | |
2051 | parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); | |
2052 | } | |
31db9f7c AB |
2053 | if (ret < 0) |
2054 | goto out; | |
2055 | btrfs_release_path(path); | |
2056 | ||
b46ab97b | 2057 | if (dir_gen) { |
7e93f6dc | 2058 | ret = get_inode_gen(root, parent_dir, dir_gen); |
b46ab97b FM |
2059 | if (ret < 0) |
2060 | goto out; | |
2061 | } | |
31db9f7c | 2062 | |
96b5bd77 | 2063 | *dir = parent_dir; |
31db9f7c AB |
2064 | |
2065 | out: | |
2066 | btrfs_free_path(path); | |
2067 | return ret; | |
2068 | } | |
2069 | ||
924794c9 | 2070 | static int is_first_ref(struct btrfs_root *root, |
31db9f7c AB |
2071 | u64 ino, u64 dir, |
2072 | const char *name, int name_len) | |
2073 | { | |
2074 | int ret; | |
2075 | struct fs_path *tmp_name; | |
2076 | u64 tmp_dir; | |
31db9f7c | 2077 | |
924794c9 | 2078 | tmp_name = fs_path_alloc(); |
31db9f7c AB |
2079 | if (!tmp_name) |
2080 | return -ENOMEM; | |
2081 | ||
b46ab97b | 2082 | ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name); |
31db9f7c AB |
2083 | if (ret < 0) |
2084 | goto out; | |
2085 | ||
b9291aff | 2086 | if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) { |
31db9f7c AB |
2087 | ret = 0; |
2088 | goto out; | |
2089 | } | |
2090 | ||
e938c8ad | 2091 | ret = !memcmp(tmp_name->start, name, name_len); |
31db9f7c AB |
2092 | |
2093 | out: | |
924794c9 | 2094 | fs_path_free(tmp_name); |
31db9f7c AB |
2095 | return ret; |
2096 | } | |
2097 | ||
766702ef AB |
2098 | /* |
2099 | * Used by process_recorded_refs to determine if a new ref would overwrite an | |
2100 | * already existing ref. In case it detects an overwrite, it returns the | |
2101 | * inode/gen in who_ino/who_gen. | |
2102 | * When an overwrite is detected, process_recorded_refs does proper orphanizing | |
2103 | * to make sure later references to the overwritten inode are possible. | |
2104 | * Orphanizing is however only required for the first ref of an inode. | |
2105 | * process_recorded_refs does an additional is_first_ref check to see if | |
2106 | * orphanizing is really required. | |
2107 | */ | |
31db9f7c AB |
2108 | static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, |
2109 | const char *name, int name_len, | |
f5962781 | 2110 | u64 *who_ino, u64 *who_gen, u64 *who_mode) |
31db9f7c | 2111 | { |
b3047a42 | 2112 | int ret; |
498581f3 | 2113 | u64 parent_root_dir_gen; |
31db9f7c | 2114 | u64 other_inode = 0; |
7e93f6dc | 2115 | struct btrfs_inode_info info; |
31db9f7c AB |
2116 | |
2117 | if (!sctx->parent_root) | |
b3047a42 | 2118 | return 0; |
31db9f7c | 2119 | |
498581f3 | 2120 | ret = is_inode_existent(sctx, dir, dir_gen, NULL, &parent_root_dir_gen); |
31db9f7c | 2121 | if (ret <= 0) |
b3047a42 | 2122 | return 0; |
31db9f7c | 2123 | |
ebdad913 JB |
2124 | /* |
2125 | * If we have a parent root we need to verify that the parent dir was | |
01327610 | 2126 | * not deleted and then re-created, if it was then we have no overwrite |
ebdad913 | 2127 | * and we can just unlink this entry. |
498581f3 FM |
2128 | * |
2129 | * @parent_root_dir_gen was set to 0 if the inode does not exist in the | |
2130 | * parent root. | |
ebdad913 | 2131 | */ |
498581f3 FM |
2132 | if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID && |
2133 | parent_root_dir_gen != dir_gen) | |
2134 | return 0; | |
ebdad913 | 2135 | |
31db9f7c | 2136 | ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, |
eab67c06 | 2137 | &other_inode); |
b3047a42 FM |
2138 | if (ret == -ENOENT) |
2139 | return 0; | |
2140 | else if (ret < 0) | |
2141 | return ret; | |
31db9f7c | 2142 | |
766702ef AB |
2143 | /* |
2144 | * Check if the overwritten ref was already processed. If yes, the ref | |
2145 | * was already unlinked/moved, so we can safely assume that we will not | |
2146 | * overwrite anything at this point in time. | |
2147 | */ | |
801bec36 RK |
2148 | if (other_inode > sctx->send_progress || |
2149 | is_waiting_for_move(sctx, other_inode)) { | |
7e93f6dc | 2150 | ret = get_inode_info(sctx->parent_root, other_inode, &info); |
31db9f7c | 2151 | if (ret < 0) |
b3047a42 | 2152 | return ret; |
31db9f7c | 2153 | |
31db9f7c | 2154 | *who_ino = other_inode; |
7e93f6dc BC |
2155 | *who_gen = info.gen; |
2156 | *who_mode = info.mode; | |
b3047a42 | 2157 | return 1; |
31db9f7c AB |
2158 | } |
2159 | ||
b3047a42 | 2160 | return 0; |
31db9f7c AB |
2161 | } |
2162 | ||
766702ef AB |
2163 | /* |
2164 | * Checks if the ref was overwritten by an already processed inode. This is | |
2165 | * used by __get_cur_name_and_parent to find out if the ref was orphanized and | |
2166 | * thus the orphan name needs be used. | |
2167 | * process_recorded_refs also uses it to avoid unlinking of refs that were | |
2168 | * overwritten. | |
2169 | */ | |
31db9f7c AB |
2170 | static int did_overwrite_ref(struct send_ctx *sctx, |
2171 | u64 dir, u64 dir_gen, | |
2172 | u64 ino, u64 ino_gen, | |
2173 | const char *name, int name_len) | |
2174 | { | |
e739ba30 | 2175 | int ret; |
31db9f7c | 2176 | u64 ow_inode; |
cb689481 | 2177 | u64 ow_gen = 0; |
498581f3 | 2178 | u64 send_root_dir_gen; |
31db9f7c AB |
2179 | |
2180 | if (!sctx->parent_root) | |
e739ba30 | 2181 | return 0; |
31db9f7c | 2182 | |
498581f3 | 2183 | ret = is_inode_existent(sctx, dir, dir_gen, &send_root_dir_gen, NULL); |
31db9f7c | 2184 | if (ret <= 0) |
e739ba30 | 2185 | return ret; |
31db9f7c | 2186 | |
498581f3 FM |
2187 | /* |
2188 | * @send_root_dir_gen was set to 0 if the inode does not exist in the | |
2189 | * send root. | |
2190 | */ | |
2191 | if (dir != BTRFS_FIRST_FREE_OBJECTID && send_root_dir_gen != dir_gen) | |
2192 | return 0; | |
01914101 | 2193 | |
31db9f7c AB |
2194 | /* check if the ref was overwritten by another ref */ |
2195 | ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, | |
eab67c06 | 2196 | &ow_inode); |
e739ba30 | 2197 | if (ret == -ENOENT) { |
31db9f7c | 2198 | /* was never and will never be overwritten */ |
e739ba30 FM |
2199 | return 0; |
2200 | } else if (ret < 0) { | |
2201 | return ret; | |
31db9f7c AB |
2202 | } |
2203 | ||
cb689481 FM |
2204 | if (ow_inode == ino) { |
2205 | ret = get_inode_gen(sctx->send_root, ow_inode, &ow_gen); | |
2206 | if (ret < 0) | |
2207 | return ret; | |
31db9f7c | 2208 | |
cb689481 FM |
2209 | /* It's the same inode, so no overwrite happened. */ |
2210 | if (ow_gen == ino_gen) | |
2211 | return 0; | |
2212 | } | |
31db9f7c | 2213 | |
8b191a68 FM |
2214 | /* |
2215 | * We know that it is or will be overwritten. Check this now. | |
2216 | * The current inode being processed might have been the one that caused | |
b786f16a FM |
2217 | * inode 'ino' to be orphanized, therefore check if ow_inode matches |
2218 | * the current inode being processed. | |
8b191a68 | 2219 | */ |
cb689481 | 2220 | if (ow_inode < sctx->send_progress) |
e739ba30 | 2221 | return 1; |
31db9f7c | 2222 | |
cb689481 FM |
2223 | if (ino != sctx->cur_ino && ow_inode == sctx->cur_ino) { |
2224 | if (ow_gen == 0) { | |
2225 | ret = get_inode_gen(sctx->send_root, ow_inode, &ow_gen); | |
2226 | if (ret < 0) | |
2227 | return ret; | |
2228 | } | |
2229 | if (ow_gen == sctx->cur_inode_gen) | |
2230 | return 1; | |
2231 | } | |
2232 | ||
e739ba30 | 2233 | return 0; |
31db9f7c AB |
2234 | } |
2235 | ||
766702ef AB |
2236 | /* |
2237 | * Same as did_overwrite_ref, but also checks if it is the first ref of an inode | |
2238 | * that got overwritten. This is used by process_recorded_refs to determine | |
2239 | * if it has to use the path as returned by get_cur_path or the orphan name. | |
2240 | */ | |
31db9f7c AB |
2241 | static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen) |
2242 | { | |
2243 | int ret = 0; | |
2244 | struct fs_path *name = NULL; | |
2245 | u64 dir; | |
2246 | u64 dir_gen; | |
2247 | ||
2248 | if (!sctx->parent_root) | |
2249 | goto out; | |
2250 | ||
924794c9 | 2251 | name = fs_path_alloc(); |
31db9f7c AB |
2252 | if (!name) |
2253 | return -ENOMEM; | |
2254 | ||
924794c9 | 2255 | ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name); |
31db9f7c AB |
2256 | if (ret < 0) |
2257 | goto out; | |
2258 | ||
2259 | ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen, | |
2260 | name->start, fs_path_len(name)); | |
31db9f7c AB |
2261 | |
2262 | out: | |
924794c9 | 2263 | fs_path_free(name); |
31db9f7c AB |
2264 | return ret; |
2265 | } | |
2266 | ||
c48545de FM |
2267 | static inline struct name_cache_entry *name_cache_search(struct send_ctx *sctx, |
2268 | u64 ino, u64 gen) | |
31db9f7c | 2269 | { |
c48545de | 2270 | struct btrfs_lru_cache_entry *entry; |
31db9f7c | 2271 | |
c48545de FM |
2272 | entry = btrfs_lru_cache_lookup(&sctx->name_cache, ino, gen); |
2273 | if (!entry) | |
31db9f7c AB |
2274 | return NULL; |
2275 | ||
c48545de | 2276 | return container_of(entry, struct name_cache_entry, entry); |
31db9f7c AB |
2277 | } |
2278 | ||
766702ef AB |
2279 | /* |
2280 | * Used by get_cur_path for each ref up to the root. | |
2281 | * Returns 0 if it succeeded. | |
2282 | * Returns 1 if the inode is not existent or got overwritten. In that case, the | |
2283 | * name is an orphan name. This instructs get_cur_path to stop iterating. If 1 | |
2284 | * is returned, parent_ino/parent_gen are not guaranteed to be valid. | |
2285 | * Returns <0 in case of error. | |
2286 | */ | |
31db9f7c AB |
2287 | static int __get_cur_name_and_parent(struct send_ctx *sctx, |
2288 | u64 ino, u64 gen, | |
2289 | u64 *parent_ino, | |
2290 | u64 *parent_gen, | |
2291 | struct fs_path *dest) | |
2292 | { | |
2293 | int ret; | |
2294 | int nce_ret; | |
c48545de | 2295 | struct name_cache_entry *nce; |
31db9f7c | 2296 | |
766702ef AB |
2297 | /* |
2298 | * First check if we already did a call to this function with the same | |
2299 | * ino/gen. If yes, check if the cache entry is still up-to-date. If yes | |
2300 | * return the cached result. | |
2301 | */ | |
31db9f7c AB |
2302 | nce = name_cache_search(sctx, ino, gen); |
2303 | if (nce) { | |
2304 | if (ino < sctx->send_progress && nce->need_later_update) { | |
c48545de | 2305 | btrfs_lru_cache_remove(&sctx->name_cache, &nce->entry); |
31db9f7c AB |
2306 | nce = NULL; |
2307 | } else { | |
31db9f7c AB |
2308 | *parent_ino = nce->parent_ino; |
2309 | *parent_gen = nce->parent_gen; | |
2310 | ret = fs_path_add(dest, nce->name, nce->name_len); | |
2311 | if (ret < 0) | |
2312 | goto out; | |
2313 | ret = nce->ret; | |
2314 | goto out; | |
2315 | } | |
2316 | } | |
2317 | ||
766702ef AB |
2318 | /* |
2319 | * If the inode is not existent yet, add the orphan name and return 1. | |
2320 | * This should only happen for the parent dir that we determine in | |
0d8869fb | 2321 | * record_new_ref_if_needed(). |
766702ef | 2322 | */ |
498581f3 | 2323 | ret = is_inode_existent(sctx, ino, gen, NULL, NULL); |
31db9f7c AB |
2324 | if (ret < 0) |
2325 | goto out; | |
2326 | ||
2327 | if (!ret) { | |
2328 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2329 | if (ret < 0) | |
2330 | goto out; | |
2331 | ret = 1; | |
2332 | goto out_cache; | |
2333 | } | |
2334 | ||
766702ef AB |
2335 | /* |
2336 | * Depending on whether the inode was already processed or not, use | |
2337 | * send_root or parent_root for ref lookup. | |
2338 | */ | |
bf0d1f44 | 2339 | if (ino < sctx->send_progress) |
924794c9 TI |
2340 | ret = get_first_ref(sctx->send_root, ino, |
2341 | parent_ino, parent_gen, dest); | |
31db9f7c | 2342 | else |
924794c9 TI |
2343 | ret = get_first_ref(sctx->parent_root, ino, |
2344 | parent_ino, parent_gen, dest); | |
31db9f7c AB |
2345 | if (ret < 0) |
2346 | goto out; | |
2347 | ||
766702ef AB |
2348 | /* |
2349 | * Check if the ref was overwritten by an inode's ref that was processed | |
2350 | * earlier. If yes, treat as orphan and return 1. | |
2351 | */ | |
31db9f7c AB |
2352 | ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, |
2353 | dest->start, dest->end - dest->start); | |
2354 | if (ret < 0) | |
2355 | goto out; | |
2356 | if (ret) { | |
2357 | fs_path_reset(dest); | |
2358 | ret = gen_unique_name(sctx, ino, gen, dest); | |
2359 | if (ret < 0) | |
2360 | goto out; | |
2361 | ret = 1; | |
2362 | } | |
2363 | ||
2364 | out_cache: | |
766702ef AB |
2365 | /* |
2366 | * Store the result of the lookup in the name cache. | |
2367 | */ | |
e780b0d1 | 2368 | nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); |
31db9f7c AB |
2369 | if (!nce) { |
2370 | ret = -ENOMEM; | |
2371 | goto out; | |
2372 | } | |
2373 | ||
c48545de FM |
2374 | nce->entry.key = ino; |
2375 | nce->entry.gen = gen; | |
31db9f7c AB |
2376 | nce->parent_ino = *parent_ino; |
2377 | nce->parent_gen = *parent_gen; | |
2378 | nce->name_len = fs_path_len(dest); | |
2379 | nce->ret = ret; | |
2380 | strcpy(nce->name, dest->start); | |
31db9f7c AB |
2381 | |
2382 | if (ino < sctx->send_progress) | |
2383 | nce->need_later_update = 0; | |
2384 | else | |
2385 | nce->need_later_update = 1; | |
2386 | ||
c48545de FM |
2387 | nce_ret = btrfs_lru_cache_store(&sctx->name_cache, &nce->entry, GFP_KERNEL); |
2388 | if (nce_ret < 0) { | |
2389 | kfree(nce); | |
31db9f7c | 2390 | ret = nce_ret; |
c48545de | 2391 | } |
31db9f7c AB |
2392 | |
2393 | out: | |
31db9f7c AB |
2394 | return ret; |
2395 | } | |
2396 | ||
2397 | /* | |
2398 | * Magic happens here. This function returns the first ref to an inode as it | |
2399 | * would look like while receiving the stream at this point in time. | |
2400 | * We walk the path up to the root. For every inode in between, we check if it | |
2401 | * was already processed/sent. If yes, we continue with the parent as found | |
2402 | * in send_root. If not, we continue with the parent as found in parent_root. | |
2403 | * If we encounter an inode that was deleted at this point in time, we use the | |
2404 | * inodes "orphan" name instead of the real name and stop. Same with new inodes | |
2405 | * that were not created yet and overwritten inodes/refs. | |
2406 | * | |
52042d8e | 2407 | * When do we have orphan inodes: |
31db9f7c AB |
2408 | * 1. When an inode is freshly created and thus no valid refs are available yet |
2409 | * 2. When a directory lost all it's refs (deleted) but still has dir items | |
2410 | * inside which were not processed yet (pending for move/delete). If anyone | |
2411 | * tried to get the path to the dir items, it would get a path inside that | |
2412 | * orphan directory. | |
2413 | * 3. When an inode is moved around or gets new links, it may overwrite the ref | |
2414 | * of an unprocessed inode. If in that case the first ref would be | |
2415 | * overwritten, the overwritten inode gets "orphanized". Later when we | |
2416 | * process this overwritten inode, it is restored at a new place by moving | |
2417 | * the orphan inode. | |
2418 | * | |
2419 | * sctx->send_progress tells this function at which point in time receiving | |
2420 | * would be. | |
2421 | */ | |
2422 | static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, | |
2423 | struct fs_path *dest) | |
2424 | { | |
2425 | int ret = 0; | |
2426 | struct fs_path *name = NULL; | |
2427 | u64 parent_inode = 0; | |
2428 | u64 parent_gen = 0; | |
2429 | int stop = 0; | |
2430 | ||
924794c9 | 2431 | name = fs_path_alloc(); |
31db9f7c AB |
2432 | if (!name) { |
2433 | ret = -ENOMEM; | |
2434 | goto out; | |
2435 | } | |
2436 | ||
2437 | dest->reversed = 1; | |
2438 | fs_path_reset(dest); | |
2439 | ||
2440 | while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) { | |
8b191a68 FM |
2441 | struct waiting_dir_move *wdm; |
2442 | ||
31db9f7c AB |
2443 | fs_path_reset(name); |
2444 | ||
0b3f407e | 2445 | if (is_waiting_for_rm(sctx, ino, gen)) { |
9dc44214 FM |
2446 | ret = gen_unique_name(sctx, ino, gen, name); |
2447 | if (ret < 0) | |
2448 | goto out; | |
2449 | ret = fs_path_add_path(dest, name); | |
2450 | break; | |
2451 | } | |
2452 | ||
8b191a68 FM |
2453 | wdm = get_waiting_dir_move(sctx, ino); |
2454 | if (wdm && wdm->orphanized) { | |
2455 | ret = gen_unique_name(sctx, ino, gen, name); | |
2456 | stop = 1; | |
2457 | } else if (wdm) { | |
bf0d1f44 FM |
2458 | ret = get_first_ref(sctx->parent_root, ino, |
2459 | &parent_inode, &parent_gen, name); | |
2460 | } else { | |
2461 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
2462 | &parent_inode, | |
2463 | &parent_gen, name); | |
2464 | if (ret) | |
2465 | stop = 1; | |
2466 | } | |
2467 | ||
31db9f7c AB |
2468 | if (ret < 0) |
2469 | goto out; | |
9f03740a | 2470 | |
31db9f7c AB |
2471 | ret = fs_path_add_path(dest, name); |
2472 | if (ret < 0) | |
2473 | goto out; | |
2474 | ||
2475 | ino = parent_inode; | |
2476 | gen = parent_gen; | |
2477 | } | |
2478 | ||
2479 | out: | |
924794c9 | 2480 | fs_path_free(name); |
31db9f7c AB |
2481 | if (!ret) |
2482 | fs_path_unreverse(dest); | |
2483 | return ret; | |
2484 | } | |
2485 | ||
31db9f7c AB |
2486 | /* |
2487 | * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace | |
2488 | */ | |
2489 | static int send_subvol_begin(struct send_ctx *sctx) | |
2490 | { | |
2491 | int ret; | |
2492 | struct btrfs_root *send_root = sctx->send_root; | |
2493 | struct btrfs_root *parent_root = sctx->parent_root; | |
2494 | struct btrfs_path *path; | |
2495 | struct btrfs_key key; | |
2496 | struct btrfs_root_ref *ref; | |
2497 | struct extent_buffer *leaf; | |
2498 | char *name = NULL; | |
2499 | int namelen; | |
2500 | ||
ffcfaf81 | 2501 | path = btrfs_alloc_path(); |
31db9f7c AB |
2502 | if (!path) |
2503 | return -ENOMEM; | |
2504 | ||
e780b0d1 | 2505 | name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); |
31db9f7c AB |
2506 | if (!name) { |
2507 | btrfs_free_path(path); | |
2508 | return -ENOMEM; | |
2509 | } | |
2510 | ||
4fd786e6 | 2511 | key.objectid = send_root->root_key.objectid; |
31db9f7c AB |
2512 | key.type = BTRFS_ROOT_BACKREF_KEY; |
2513 | key.offset = 0; | |
2514 | ||
2515 | ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, | |
2516 | &key, path, 1, 0); | |
2517 | if (ret < 0) | |
2518 | goto out; | |
2519 | if (ret) { | |
2520 | ret = -ENOENT; | |
2521 | goto out; | |
2522 | } | |
2523 | ||
2524 | leaf = path->nodes[0]; | |
2525 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2526 | if (key.type != BTRFS_ROOT_BACKREF_KEY || | |
4fd786e6 | 2527 | key.objectid != send_root->root_key.objectid) { |
31db9f7c AB |
2528 | ret = -ENOENT; |
2529 | goto out; | |
2530 | } | |
2531 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
2532 | namelen = btrfs_root_ref_name_len(leaf, ref); | |
2533 | read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen); | |
2534 | btrfs_release_path(path); | |
2535 | ||
31db9f7c AB |
2536 | if (parent_root) { |
2537 | ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); | |
2538 | if (ret < 0) | |
2539 | goto out; | |
2540 | } else { | |
2541 | ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); | |
2542 | if (ret < 0) | |
2543 | goto out; | |
2544 | } | |
2545 | ||
2546 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); | |
b96b1db0 RR |
2547 | |
2548 | if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid)) | |
2549 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2550 | sctx->send_root->root_item.received_uuid); | |
2551 | else | |
2552 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID, | |
2553 | sctx->send_root->root_item.uuid); | |
2554 | ||
31db9f7c | 2555 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, |
09e3a288 | 2556 | btrfs_root_ctransid(&sctx->send_root->root_item)); |
31db9f7c | 2557 | if (parent_root) { |
37b8d27d JB |
2558 | if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) |
2559 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2560 | parent_root->root_item.received_uuid); | |
2561 | else | |
2562 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
2563 | parent_root->root_item.uuid); | |
31db9f7c | 2564 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
09e3a288 | 2565 | btrfs_root_ctransid(&sctx->parent_root->root_item)); |
31db9f7c AB |
2566 | } |
2567 | ||
2568 | ret = send_cmd(sctx); | |
2569 | ||
2570 | tlv_put_failure: | |
2571 | out: | |
2572 | btrfs_free_path(path); | |
2573 | kfree(name); | |
2574 | return ret; | |
2575 | } | |
2576 | ||
2577 | static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) | |
2578 | { | |
04ab956e | 2579 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2580 | int ret = 0; |
2581 | struct fs_path *p; | |
2582 | ||
04ab956e | 2583 | btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size); |
31db9f7c | 2584 | |
924794c9 | 2585 | p = fs_path_alloc(); |
31db9f7c AB |
2586 | if (!p) |
2587 | return -ENOMEM; | |
2588 | ||
2589 | ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); | |
2590 | if (ret < 0) | |
2591 | goto out; | |
2592 | ||
2593 | ret = get_cur_path(sctx, ino, gen, p); | |
2594 | if (ret < 0) | |
2595 | goto out; | |
2596 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2597 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); | |
2598 | ||
2599 | ret = send_cmd(sctx); | |
2600 | ||
2601 | tlv_put_failure: | |
2602 | out: | |
924794c9 | 2603 | fs_path_free(p); |
31db9f7c AB |
2604 | return ret; |
2605 | } | |
2606 | ||
2607 | static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) | |
2608 | { | |
04ab956e | 2609 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2610 | int ret = 0; |
2611 | struct fs_path *p; | |
2612 | ||
04ab956e | 2613 | btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode); |
31db9f7c | 2614 | |
924794c9 | 2615 | p = fs_path_alloc(); |
31db9f7c AB |
2616 | if (!p) |
2617 | return -ENOMEM; | |
2618 | ||
2619 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); | |
2620 | if (ret < 0) | |
2621 | goto out; | |
2622 | ||
2623 | ret = get_cur_path(sctx, ino, gen, p); | |
2624 | if (ret < 0) | |
2625 | goto out; | |
2626 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2627 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); | |
2628 | ||
2629 | ret = send_cmd(sctx); | |
2630 | ||
2631 | tlv_put_failure: | |
2632 | out: | |
924794c9 | 2633 | fs_path_free(p); |
31db9f7c AB |
2634 | return ret; |
2635 | } | |
2636 | ||
48247359 DS |
2637 | static int send_fileattr(struct send_ctx *sctx, u64 ino, u64 gen, u64 fileattr) |
2638 | { | |
2639 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; | |
2640 | int ret = 0; | |
2641 | struct fs_path *p; | |
2642 | ||
2643 | if (sctx->proto < 2) | |
2644 | return 0; | |
2645 | ||
2646 | btrfs_debug(fs_info, "send_fileattr %llu fileattr=%llu", ino, fileattr); | |
2647 | ||
2648 | p = fs_path_alloc(); | |
2649 | if (!p) | |
2650 | return -ENOMEM; | |
2651 | ||
2652 | ret = begin_cmd(sctx, BTRFS_SEND_C_FILEATTR); | |
2653 | if (ret < 0) | |
2654 | goto out; | |
2655 | ||
2656 | ret = get_cur_path(sctx, ino, gen, p); | |
2657 | if (ret < 0) | |
2658 | goto out; | |
2659 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2660 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILEATTR, fileattr); | |
2661 | ||
2662 | ret = send_cmd(sctx); | |
2663 | ||
2664 | tlv_put_failure: | |
2665 | out: | |
2666 | fs_path_free(p); | |
2667 | return ret; | |
2668 | } | |
2669 | ||
31db9f7c AB |
2670 | static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) |
2671 | { | |
04ab956e | 2672 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2673 | int ret = 0; |
2674 | struct fs_path *p; | |
2675 | ||
04ab956e JM |
2676 | btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu", |
2677 | ino, uid, gid); | |
31db9f7c | 2678 | |
924794c9 | 2679 | p = fs_path_alloc(); |
31db9f7c AB |
2680 | if (!p) |
2681 | return -ENOMEM; | |
2682 | ||
2683 | ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); | |
2684 | if (ret < 0) | |
2685 | goto out; | |
2686 | ||
2687 | ret = get_cur_path(sctx, ino, gen, p); | |
2688 | if (ret < 0) | |
2689 | goto out; | |
2690 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
2691 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); | |
2692 | TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); | |
2693 | ||
2694 | ret = send_cmd(sctx); | |
2695 | ||
2696 | tlv_put_failure: | |
2697 | out: | |
924794c9 | 2698 | fs_path_free(p); |
31db9f7c AB |
2699 | return ret; |
2700 | } | |
2701 | ||
2702 | static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) | |
2703 | { | |
04ab956e | 2704 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
2705 | int ret = 0; |
2706 | struct fs_path *p = NULL; | |
2707 | struct btrfs_inode_item *ii; | |
2708 | struct btrfs_path *path = NULL; | |
2709 | struct extent_buffer *eb; | |
2710 | struct btrfs_key key; | |
2711 | int slot; | |
2712 | ||
04ab956e | 2713 | btrfs_debug(fs_info, "send_utimes %llu", ino); |
31db9f7c | 2714 | |
924794c9 | 2715 | p = fs_path_alloc(); |
31db9f7c AB |
2716 | if (!p) |
2717 | return -ENOMEM; | |
2718 | ||
2719 | path = alloc_path_for_send(); | |
2720 | if (!path) { | |
2721 | ret = -ENOMEM; | |
2722 | goto out; | |
2723 | } | |
2724 | ||
2725 | key.objectid = ino; | |
2726 | key.type = BTRFS_INODE_ITEM_KEY; | |
2727 | key.offset = 0; | |
2728 | ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); | |
15b253ea FM |
2729 | if (ret > 0) |
2730 | ret = -ENOENT; | |
31db9f7c AB |
2731 | if (ret < 0) |
2732 | goto out; | |
2733 | ||
2734 | eb = path->nodes[0]; | |
2735 | slot = path->slots[0]; | |
2736 | ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item); | |
2737 | ||
2738 | ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES); | |
2739 | if (ret < 0) | |
2740 | goto out; | |
2741 | ||
2742 | ret = get_cur_path(sctx, ino, gen, p); | |
2743 | if (ret < 0) | |
2744 | goto out; | |
2745 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
a937b979 DS |
2746 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); |
2747 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); | |
2748 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); | |
22a5b2ab DS |
2749 | if (sctx->proto >= 2) |
2750 | TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_OTIME, eb, &ii->otime); | |
31db9f7c AB |
2751 | |
2752 | ret = send_cmd(sctx); | |
2753 | ||
2754 | tlv_put_failure: | |
2755 | out: | |
924794c9 | 2756 | fs_path_free(p); |
31db9f7c AB |
2757 | btrfs_free_path(path); |
2758 | return ret; | |
2759 | } | |
2760 | ||
3e49363b FM |
2761 | /* |
2762 | * If the cache is full, we can't remove entries from it and do a call to | |
2763 | * send_utimes() for each respective inode, because we might be finishing | |
2764 | * processing an inode that is a directory and it just got renamed, and existing | |
2765 | * entries in the cache may refer to inodes that have the directory in their | |
2766 | * full path - in which case we would generate outdated paths (pre-rename) | |
2767 | * for the inodes that the cache entries point to. Instead of prunning the | |
2768 | * cache when inserting, do it after we finish processing each inode at | |
2769 | * finish_inode_if_needed(). | |
2770 | */ | |
2771 | static int cache_dir_utimes(struct send_ctx *sctx, u64 dir, u64 gen) | |
2772 | { | |
2773 | struct btrfs_lru_cache_entry *entry; | |
2774 | int ret; | |
2775 | ||
2776 | entry = btrfs_lru_cache_lookup(&sctx->dir_utimes_cache, dir, gen); | |
2777 | if (entry != NULL) | |
2778 | return 0; | |
2779 | ||
2780 | /* Caching is optional, don't fail if we can't allocate memory. */ | |
2781 | entry = kmalloc(sizeof(*entry), GFP_KERNEL); | |
2782 | if (!entry) | |
2783 | return send_utimes(sctx, dir, gen); | |
2784 | ||
2785 | entry->key = dir; | |
2786 | entry->gen = gen; | |
2787 | ||
2788 | ret = btrfs_lru_cache_store(&sctx->dir_utimes_cache, entry, GFP_KERNEL); | |
2789 | ASSERT(ret != -EEXIST); | |
2790 | if (ret) { | |
2791 | kfree(entry); | |
2792 | return send_utimes(sctx, dir, gen); | |
2793 | } | |
2794 | ||
2795 | return 0; | |
2796 | } | |
2797 | ||
2798 | static int trim_dir_utimes_cache(struct send_ctx *sctx) | |
2799 | { | |
2800 | while (btrfs_lru_cache_size(&sctx->dir_utimes_cache) > | |
2801 | SEND_MAX_DIR_UTIMES_CACHE_SIZE) { | |
2802 | struct btrfs_lru_cache_entry *lru; | |
2803 | int ret; | |
2804 | ||
2805 | lru = btrfs_lru_cache_lru_entry(&sctx->dir_utimes_cache); | |
2806 | ASSERT(lru != NULL); | |
2807 | ||
2808 | ret = send_utimes(sctx, lru->key, lru->gen); | |
2809 | if (ret) | |
2810 | return ret; | |
2811 | ||
2812 | btrfs_lru_cache_remove(&sctx->dir_utimes_cache, lru); | |
2813 | } | |
2814 | ||
2815 | return 0; | |
2816 | } | |
2817 | ||
31db9f7c AB |
2818 | /* |
2819 | * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have | |
2820 | * a valid path yet because we did not process the refs yet. So, the inode | |
2821 | * is created as orphan. | |
2822 | */ | |
1f4692da | 2823 | static int send_create_inode(struct send_ctx *sctx, u64 ino) |
31db9f7c | 2824 | { |
04ab956e | 2825 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c | 2826 | int ret = 0; |
31db9f7c | 2827 | struct fs_path *p; |
31db9f7c | 2828 | int cmd; |
7e93f6dc | 2829 | struct btrfs_inode_info info; |
1f4692da | 2830 | u64 gen; |
31db9f7c | 2831 | u64 mode; |
1f4692da | 2832 | u64 rdev; |
31db9f7c | 2833 | |
04ab956e | 2834 | btrfs_debug(fs_info, "send_create_inode %llu", ino); |
31db9f7c | 2835 | |
924794c9 | 2836 | p = fs_path_alloc(); |
31db9f7c AB |
2837 | if (!p) |
2838 | return -ENOMEM; | |
2839 | ||
644d1940 | 2840 | if (ino != sctx->cur_ino) { |
7e93f6dc | 2841 | ret = get_inode_info(sctx->send_root, ino, &info); |
644d1940 LB |
2842 | if (ret < 0) |
2843 | goto out; | |
7e93f6dc BC |
2844 | gen = info.gen; |
2845 | mode = info.mode; | |
2846 | rdev = info.rdev; | |
644d1940 LB |
2847 | } else { |
2848 | gen = sctx->cur_inode_gen; | |
2849 | mode = sctx->cur_inode_mode; | |
2850 | rdev = sctx->cur_inode_rdev; | |
2851 | } | |
31db9f7c | 2852 | |
e938c8ad | 2853 | if (S_ISREG(mode)) { |
31db9f7c | 2854 | cmd = BTRFS_SEND_C_MKFILE; |
e938c8ad | 2855 | } else if (S_ISDIR(mode)) { |
31db9f7c | 2856 | cmd = BTRFS_SEND_C_MKDIR; |
e938c8ad | 2857 | } else if (S_ISLNK(mode)) { |
31db9f7c | 2858 | cmd = BTRFS_SEND_C_SYMLINK; |
e938c8ad | 2859 | } else if (S_ISCHR(mode) || S_ISBLK(mode)) { |
31db9f7c | 2860 | cmd = BTRFS_SEND_C_MKNOD; |
e938c8ad | 2861 | } else if (S_ISFIFO(mode)) { |
31db9f7c | 2862 | cmd = BTRFS_SEND_C_MKFIFO; |
e938c8ad | 2863 | } else if (S_ISSOCK(mode)) { |
31db9f7c | 2864 | cmd = BTRFS_SEND_C_MKSOCK; |
e938c8ad | 2865 | } else { |
f14d104d | 2866 | btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o", |
31db9f7c | 2867 | (int)(mode & S_IFMT)); |
ca6842bf | 2868 | ret = -EOPNOTSUPP; |
31db9f7c AB |
2869 | goto out; |
2870 | } | |
2871 | ||
2872 | ret = begin_cmd(sctx, cmd); | |
2873 | if (ret < 0) | |
2874 | goto out; | |
2875 | ||
1f4692da | 2876 | ret = gen_unique_name(sctx, ino, gen, p); |
31db9f7c AB |
2877 | if (ret < 0) |
2878 | goto out; | |
2879 | ||
2880 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
1f4692da | 2881 | TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino); |
31db9f7c AB |
2882 | |
2883 | if (S_ISLNK(mode)) { | |
2884 | fs_path_reset(p); | |
924794c9 | 2885 | ret = read_symlink(sctx->send_root, ino, p); |
31db9f7c AB |
2886 | if (ret < 0) |
2887 | goto out; | |
2888 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p); | |
2889 | } else if (S_ISCHR(mode) || S_ISBLK(mode) || | |
2890 | S_ISFIFO(mode) || S_ISSOCK(mode)) { | |
d79e5043 AJ |
2891 | TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev)); |
2892 | TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode); | |
31db9f7c AB |
2893 | } |
2894 | ||
2895 | ret = send_cmd(sctx); | |
2896 | if (ret < 0) | |
2897 | goto out; | |
2898 | ||
2899 | ||
2900 | tlv_put_failure: | |
2901 | out: | |
924794c9 | 2902 | fs_path_free(p); |
31db9f7c AB |
2903 | return ret; |
2904 | } | |
2905 | ||
e8a7f49d FM |
2906 | static void cache_dir_created(struct send_ctx *sctx, u64 dir) |
2907 | { | |
2908 | struct btrfs_lru_cache_entry *entry; | |
2909 | int ret; | |
2910 | ||
2911 | /* Caching is optional, ignore any failures. */ | |
2912 | entry = kmalloc(sizeof(*entry), GFP_KERNEL); | |
2913 | if (!entry) | |
2914 | return; | |
2915 | ||
2916 | entry->key = dir; | |
0da0c560 | 2917 | entry->gen = 0; |
e8a7f49d FM |
2918 | ret = btrfs_lru_cache_store(&sctx->dir_created_cache, entry, GFP_KERNEL); |
2919 | if (ret < 0) | |
2920 | kfree(entry); | |
2921 | } | |
2922 | ||
1f4692da AB |
2923 | /* |
2924 | * We need some special handling for inodes that get processed before the parent | |
2925 | * directory got created. See process_recorded_refs for details. | |
2926 | * This function does the check if we already created the dir out of order. | |
2927 | */ | |
2928 | static int did_create_dir(struct send_ctx *sctx, u64 dir) | |
2929 | { | |
2930 | int ret = 0; | |
6dcee260 | 2931 | int iter_ret = 0; |
1f4692da AB |
2932 | struct btrfs_path *path = NULL; |
2933 | struct btrfs_key key; | |
2934 | struct btrfs_key found_key; | |
2935 | struct btrfs_key di_key; | |
1f4692da | 2936 | struct btrfs_dir_item *di; |
1f4692da | 2937 | |
0da0c560 | 2938 | if (btrfs_lru_cache_lookup(&sctx->dir_created_cache, dir, 0)) |
e8a7f49d FM |
2939 | return 1; |
2940 | ||
1f4692da | 2941 | path = alloc_path_for_send(); |
6dcee260 GN |
2942 | if (!path) |
2943 | return -ENOMEM; | |
1f4692da AB |
2944 | |
2945 | key.objectid = dir; | |
2946 | key.type = BTRFS_DIR_INDEX_KEY; | |
2947 | key.offset = 0; | |
dff6d0ad | 2948 | |
6dcee260 GN |
2949 | btrfs_for_each_slot(sctx->send_root, &key, &found_key, path, iter_ret) { |
2950 | struct extent_buffer *eb = path->nodes[0]; | |
dff6d0ad | 2951 | |
dff6d0ad | 2952 | if (found_key.objectid != key.objectid || |
1f4692da AB |
2953 | found_key.type != key.type) { |
2954 | ret = 0; | |
6dcee260 | 2955 | break; |
1f4692da AB |
2956 | } |
2957 | ||
6dcee260 | 2958 | di = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dir_item); |
1f4692da AB |
2959 | btrfs_dir_item_key_to_cpu(eb, di, &di_key); |
2960 | ||
a0525414 JB |
2961 | if (di_key.type != BTRFS_ROOT_ITEM_KEY && |
2962 | di_key.objectid < sctx->send_progress) { | |
1f4692da | 2963 | ret = 1; |
e8a7f49d | 2964 | cache_dir_created(sctx, dir); |
6dcee260 | 2965 | break; |
1f4692da | 2966 | } |
1f4692da | 2967 | } |
6dcee260 GN |
2968 | /* Catch error found during iteration */ |
2969 | if (iter_ret < 0) | |
2970 | ret = iter_ret; | |
1f4692da | 2971 | |
1f4692da AB |
2972 | btrfs_free_path(path); |
2973 | return ret; | |
2974 | } | |
2975 | ||
2976 | /* | |
2977 | * Only creates the inode if it is: | |
2978 | * 1. Not a directory | |
2979 | * 2. Or a directory which was not created already due to out of order | |
2980 | * directories. See did_create_dir and process_recorded_refs for details. | |
2981 | */ | |
2982 | static int send_create_inode_if_needed(struct send_ctx *sctx) | |
2983 | { | |
2984 | int ret; | |
2985 | ||
2986 | if (S_ISDIR(sctx->cur_inode_mode)) { | |
2987 | ret = did_create_dir(sctx, sctx->cur_ino); | |
2988 | if (ret < 0) | |
0e3dd5bc MPS |
2989 | return ret; |
2990 | else if (ret > 0) | |
2991 | return 0; | |
1f4692da AB |
2992 | } |
2993 | ||
e8a7f49d FM |
2994 | ret = send_create_inode(sctx, sctx->cur_ino); |
2995 | ||
2996 | if (ret == 0 && S_ISDIR(sctx->cur_inode_mode)) | |
2997 | cache_dir_created(sctx, sctx->cur_ino); | |
2998 | ||
2999 | return ret; | |
1f4692da AB |
3000 | } |
3001 | ||
31db9f7c AB |
3002 | struct recorded_ref { |
3003 | struct list_head list; | |
31db9f7c AB |
3004 | char *name; |
3005 | struct fs_path *full_path; | |
3006 | u64 dir; | |
3007 | u64 dir_gen; | |
31db9f7c | 3008 | int name_len; |
3aa5bd36 BC |
3009 | struct rb_node node; |
3010 | struct rb_root *root; | |
31db9f7c AB |
3011 | }; |
3012 | ||
71ecfc13 BC |
3013 | static struct recorded_ref *recorded_ref_alloc(void) |
3014 | { | |
3015 | struct recorded_ref *ref; | |
3016 | ||
3017 | ref = kzalloc(sizeof(*ref), GFP_KERNEL); | |
3018 | if (!ref) | |
3019 | return NULL; | |
3aa5bd36 | 3020 | RB_CLEAR_NODE(&ref->node); |
71ecfc13 BC |
3021 | INIT_LIST_HEAD(&ref->list); |
3022 | return ref; | |
3023 | } | |
3024 | ||
3025 | static void recorded_ref_free(struct recorded_ref *ref) | |
3026 | { | |
3027 | if (!ref) | |
3028 | return; | |
3aa5bd36 BC |
3029 | if (!RB_EMPTY_NODE(&ref->node)) |
3030 | rb_erase(&ref->node, ref->root); | |
71ecfc13 BC |
3031 | list_del(&ref->list); |
3032 | fs_path_free(ref->full_path); | |
3033 | kfree(ref); | |
3034 | } | |
3035 | ||
fdb13889 FM |
3036 | static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) |
3037 | { | |
3038 | ref->full_path = path; | |
3039 | ref->name = (char *)kbasename(ref->full_path->start); | |
3040 | ref->name_len = ref->full_path->end - ref->name; | |
3041 | } | |
3042 | ||
ba5e8f2e JB |
3043 | static int dup_ref(struct recorded_ref *ref, struct list_head *list) |
3044 | { | |
3045 | struct recorded_ref *new; | |
3046 | ||
71ecfc13 | 3047 | new = recorded_ref_alloc(); |
ba5e8f2e JB |
3048 | if (!new) |
3049 | return -ENOMEM; | |
3050 | ||
3051 | new->dir = ref->dir; | |
3052 | new->dir_gen = ref->dir_gen; | |
ba5e8f2e JB |
3053 | list_add_tail(&new->list, list); |
3054 | return 0; | |
3055 | } | |
3056 | ||
924794c9 | 3057 | static void __free_recorded_refs(struct list_head *head) |
31db9f7c AB |
3058 | { |
3059 | struct recorded_ref *cur; | |
31db9f7c | 3060 | |
e938c8ad AB |
3061 | while (!list_empty(head)) { |
3062 | cur = list_entry(head->next, struct recorded_ref, list); | |
71ecfc13 | 3063 | recorded_ref_free(cur); |
31db9f7c | 3064 | } |
31db9f7c AB |
3065 | } |
3066 | ||
3067 | static void free_recorded_refs(struct send_ctx *sctx) | |
3068 | { | |
924794c9 TI |
3069 | __free_recorded_refs(&sctx->new_refs); |
3070 | __free_recorded_refs(&sctx->deleted_refs); | |
31db9f7c AB |
3071 | } |
3072 | ||
3073 | /* | |
766702ef | 3074 | * Renames/moves a file/dir to its orphan name. Used when the first |
31db9f7c AB |
3075 | * ref of an unprocessed inode gets overwritten and for all non empty |
3076 | * directories. | |
3077 | */ | |
3078 | static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, | |
3079 | struct fs_path *path) | |
3080 | { | |
3081 | int ret; | |
3082 | struct fs_path *orphan; | |
3083 | ||
924794c9 | 3084 | orphan = fs_path_alloc(); |
31db9f7c AB |
3085 | if (!orphan) |
3086 | return -ENOMEM; | |
3087 | ||
3088 | ret = gen_unique_name(sctx, ino, gen, orphan); | |
3089 | if (ret < 0) | |
3090 | goto out; | |
3091 | ||
3092 | ret = send_rename(sctx, path, orphan); | |
3093 | ||
3094 | out: | |
924794c9 | 3095 | fs_path_free(orphan); |
31db9f7c AB |
3096 | return ret; |
3097 | } | |
3098 | ||
0b3f407e FM |
3099 | static struct orphan_dir_info *add_orphan_dir_info(struct send_ctx *sctx, |
3100 | u64 dir_ino, u64 dir_gen) | |
9dc44214 FM |
3101 | { |
3102 | struct rb_node **p = &sctx->orphan_dirs.rb_node; | |
3103 | struct rb_node *parent = NULL; | |
3104 | struct orphan_dir_info *entry, *odi; | |
3105 | ||
9dc44214 FM |
3106 | while (*p) { |
3107 | parent = *p; | |
3108 | entry = rb_entry(parent, struct orphan_dir_info, node); | |
0b3f407e | 3109 | if (dir_ino < entry->ino) |
9dc44214 | 3110 | p = &(*p)->rb_left; |
0b3f407e | 3111 | else if (dir_ino > entry->ino) |
9dc44214 | 3112 | p = &(*p)->rb_right; |
0b3f407e FM |
3113 | else if (dir_gen < entry->gen) |
3114 | p = &(*p)->rb_left; | |
3115 | else if (dir_gen > entry->gen) | |
3116 | p = &(*p)->rb_right; | |
3117 | else | |
9dc44214 | 3118 | return entry; |
9dc44214 FM |
3119 | } |
3120 | ||
35c8eda1 RK |
3121 | odi = kmalloc(sizeof(*odi), GFP_KERNEL); |
3122 | if (!odi) | |
3123 | return ERR_PTR(-ENOMEM); | |
3124 | odi->ino = dir_ino; | |
0b3f407e | 3125 | odi->gen = dir_gen; |
0f96f517 | 3126 | odi->last_dir_index_offset = 0; |
474e4761 | 3127 | odi->dir_high_seq_ino = 0; |
35c8eda1 | 3128 | |
9dc44214 FM |
3129 | rb_link_node(&odi->node, parent, p); |
3130 | rb_insert_color(&odi->node, &sctx->orphan_dirs); | |
3131 | return odi; | |
3132 | } | |
3133 | ||
0b3f407e FM |
3134 | static struct orphan_dir_info *get_orphan_dir_info(struct send_ctx *sctx, |
3135 | u64 dir_ino, u64 gen) | |
9dc44214 FM |
3136 | { |
3137 | struct rb_node *n = sctx->orphan_dirs.rb_node; | |
3138 | struct orphan_dir_info *entry; | |
3139 | ||
3140 | while (n) { | |
3141 | entry = rb_entry(n, struct orphan_dir_info, node); | |
3142 | if (dir_ino < entry->ino) | |
3143 | n = n->rb_left; | |
3144 | else if (dir_ino > entry->ino) | |
3145 | n = n->rb_right; | |
0b3f407e FM |
3146 | else if (gen < entry->gen) |
3147 | n = n->rb_left; | |
3148 | else if (gen > entry->gen) | |
3149 | n = n->rb_right; | |
9dc44214 FM |
3150 | else |
3151 | return entry; | |
3152 | } | |
3153 | return NULL; | |
3154 | } | |
3155 | ||
0b3f407e | 3156 | static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen) |
9dc44214 | 3157 | { |
0b3f407e | 3158 | struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino, gen); |
9dc44214 FM |
3159 | |
3160 | return odi != NULL; | |
3161 | } | |
3162 | ||
3163 | static void free_orphan_dir_info(struct send_ctx *sctx, | |
3164 | struct orphan_dir_info *odi) | |
3165 | { | |
3166 | if (!odi) | |
3167 | return; | |
3168 | rb_erase(&odi->node, &sctx->orphan_dirs); | |
3169 | kfree(odi); | |
3170 | } | |
3171 | ||
31db9f7c AB |
3172 | /* |
3173 | * Returns 1 if a directory can be removed at this point in time. | |
3174 | * We check this by iterating all dir items and checking if the inode behind | |
3175 | * the dir item was already processed. | |
3176 | */ | |
24970ccb | 3177 | static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen) |
31db9f7c AB |
3178 | { |
3179 | int ret = 0; | |
18f80f1f | 3180 | int iter_ret = 0; |
31db9f7c AB |
3181 | struct btrfs_root *root = sctx->parent_root; |
3182 | struct btrfs_path *path; | |
3183 | struct btrfs_key key; | |
3184 | struct btrfs_key found_key; | |
3185 | struct btrfs_key loc; | |
3186 | struct btrfs_dir_item *di; | |
0f96f517 | 3187 | struct orphan_dir_info *odi = NULL; |
474e4761 FM |
3188 | u64 dir_high_seq_ino = 0; |
3189 | u64 last_dir_index_offset = 0; | |
31db9f7c | 3190 | |
6d85ed05 AB |
3191 | /* |
3192 | * Don't try to rmdir the top/root subvolume dir. | |
3193 | */ | |
3194 | if (dir == BTRFS_FIRST_FREE_OBJECTID) | |
3195 | return 0; | |
3196 | ||
474e4761 FM |
3197 | odi = get_orphan_dir_info(sctx, dir, dir_gen); |
3198 | if (odi && sctx->cur_ino < odi->dir_high_seq_ino) | |
3199 | return 0; | |
3200 | ||
31db9f7c AB |
3201 | path = alloc_path_for_send(); |
3202 | if (!path) | |
3203 | return -ENOMEM; | |
3204 | ||
474e4761 FM |
3205 | if (!odi) { |
3206 | /* | |
3207 | * Find the inode number associated with the last dir index | |
3208 | * entry. This is very likely the inode with the highest number | |
3209 | * of all inodes that have an entry in the directory. We can | |
3210 | * then use it to avoid future calls to can_rmdir(), when | |
3211 | * processing inodes with a lower number, from having to search | |
3212 | * the parent root b+tree for dir index keys. | |
3213 | */ | |
3214 | key.objectid = dir; | |
3215 | key.type = BTRFS_DIR_INDEX_KEY; | |
3216 | key.offset = (u64)-1; | |
3217 | ||
3218 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3219 | if (ret < 0) { | |
3220 | goto out; | |
3221 | } else if (ret > 0) { | |
3222 | /* Can't happen, the root is never empty. */ | |
3223 | ASSERT(path->slots[0] > 0); | |
3224 | if (WARN_ON(path->slots[0] == 0)) { | |
3225 | ret = -EUCLEAN; | |
3226 | goto out; | |
3227 | } | |
3228 | path->slots[0]--; | |
3229 | } | |
3230 | ||
3231 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
3232 | if (key.objectid != dir || key.type != BTRFS_DIR_INDEX_KEY) { | |
3233 | /* No index keys, dir can be removed. */ | |
3234 | ret = 1; | |
3235 | goto out; | |
3236 | } | |
3237 | ||
3238 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3239 | struct btrfs_dir_item); | |
3240 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
3241 | dir_high_seq_ino = loc.objectid; | |
3242 | if (sctx->cur_ino < dir_high_seq_ino) { | |
3243 | ret = 0; | |
3244 | goto out; | |
3245 | } | |
3246 | ||
3247 | btrfs_release_path(path); | |
3248 | } | |
3249 | ||
31db9f7c AB |
3250 | key.objectid = dir; |
3251 | key.type = BTRFS_DIR_INDEX_KEY; | |
474e4761 | 3252 | key.offset = (odi ? odi->last_dir_index_offset : 0); |
0f96f517 | 3253 | |
18f80f1f | 3254 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
9dc44214 FM |
3255 | struct waiting_dir_move *dm; |
3256 | ||
dff6d0ad FDBM |
3257 | if (found_key.objectid != key.objectid || |
3258 | found_key.type != key.type) | |
31db9f7c | 3259 | break; |
31db9f7c AB |
3260 | |
3261 | di = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3262 | struct btrfs_dir_item); | |
3263 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); | |
3264 | ||
474e4761 FM |
3265 | dir_high_seq_ino = max(dir_high_seq_ino, loc.objectid); |
3266 | last_dir_index_offset = found_key.offset; | |
3267 | ||
9dc44214 FM |
3268 | dm = get_waiting_dir_move(sctx, loc.objectid); |
3269 | if (dm) { | |
9dc44214 | 3270 | dm->rmdir_ino = dir; |
0b3f407e | 3271 | dm->rmdir_gen = dir_gen; |
9dc44214 FM |
3272 | ret = 0; |
3273 | goto out; | |
3274 | } | |
3275 | ||
24970ccb | 3276 | if (loc.objectid > sctx->cur_ino) { |
31db9f7c AB |
3277 | ret = 0; |
3278 | goto out; | |
3279 | } | |
18f80f1f GN |
3280 | } |
3281 | if (iter_ret < 0) { | |
3282 | ret = iter_ret; | |
3283 | goto out; | |
31db9f7c | 3284 | } |
0f96f517 | 3285 | free_orphan_dir_info(sctx, odi); |
31db9f7c AB |
3286 | |
3287 | ret = 1; | |
3288 | ||
3289 | out: | |
3290 | btrfs_free_path(path); | |
d921b9cf FM |
3291 | |
3292 | if (ret) | |
3293 | return ret; | |
3294 | ||
78cf1a95 FM |
3295 | if (!odi) { |
3296 | odi = add_orphan_dir_info(sctx, dir, dir_gen); | |
3297 | if (IS_ERR(odi)) | |
3298 | return PTR_ERR(odi); | |
3299 | ||
3300 | odi->gen = dir_gen; | |
3301 | } | |
d921b9cf | 3302 | |
474e4761 FM |
3303 | odi->last_dir_index_offset = last_dir_index_offset; |
3304 | odi->dir_high_seq_ino = max(odi->dir_high_seq_ino, dir_high_seq_ino); | |
d921b9cf FM |
3305 | |
3306 | return 0; | |
31db9f7c AB |
3307 | } |
3308 | ||
9f03740a FDBM |
3309 | static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) |
3310 | { | |
9dc44214 | 3311 | struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino); |
9f03740a | 3312 | |
9dc44214 | 3313 | return entry != NULL; |
9f03740a FDBM |
3314 | } |
3315 | ||
8b191a68 | 3316 | static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) |
9f03740a FDBM |
3317 | { |
3318 | struct rb_node **p = &sctx->waiting_dir_moves.rb_node; | |
3319 | struct rb_node *parent = NULL; | |
3320 | struct waiting_dir_move *entry, *dm; | |
3321 | ||
e780b0d1 | 3322 | dm = kmalloc(sizeof(*dm), GFP_KERNEL); |
9f03740a FDBM |
3323 | if (!dm) |
3324 | return -ENOMEM; | |
3325 | dm->ino = ino; | |
9dc44214 | 3326 | dm->rmdir_ino = 0; |
0b3f407e | 3327 | dm->rmdir_gen = 0; |
8b191a68 | 3328 | dm->orphanized = orphanized; |
9f03740a FDBM |
3329 | |
3330 | while (*p) { | |
3331 | parent = *p; | |
3332 | entry = rb_entry(parent, struct waiting_dir_move, node); | |
3333 | if (ino < entry->ino) { | |
3334 | p = &(*p)->rb_left; | |
3335 | } else if (ino > entry->ino) { | |
3336 | p = &(*p)->rb_right; | |
3337 | } else { | |
3338 | kfree(dm); | |
3339 | return -EEXIST; | |
3340 | } | |
3341 | } | |
3342 | ||
3343 | rb_link_node(&dm->node, parent, p); | |
3344 | rb_insert_color(&dm->node, &sctx->waiting_dir_moves); | |
3345 | return 0; | |
3346 | } | |
3347 | ||
9dc44214 FM |
3348 | static struct waiting_dir_move * |
3349 | get_waiting_dir_move(struct send_ctx *sctx, u64 ino) | |
9f03740a FDBM |
3350 | { |
3351 | struct rb_node *n = sctx->waiting_dir_moves.rb_node; | |
3352 | struct waiting_dir_move *entry; | |
3353 | ||
3354 | while (n) { | |
3355 | entry = rb_entry(n, struct waiting_dir_move, node); | |
9dc44214 | 3356 | if (ino < entry->ino) |
9f03740a | 3357 | n = n->rb_left; |
9dc44214 | 3358 | else if (ino > entry->ino) |
9f03740a | 3359 | n = n->rb_right; |
9dc44214 FM |
3360 | else |
3361 | return entry; | |
9f03740a | 3362 | } |
9dc44214 FM |
3363 | return NULL; |
3364 | } | |
3365 | ||
3366 | static void free_waiting_dir_move(struct send_ctx *sctx, | |
3367 | struct waiting_dir_move *dm) | |
3368 | { | |
3369 | if (!dm) | |
3370 | return; | |
3371 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
3372 | kfree(dm); | |
9f03740a FDBM |
3373 | } |
3374 | ||
bfa7e1f8 FM |
3375 | static int add_pending_dir_move(struct send_ctx *sctx, |
3376 | u64 ino, | |
3377 | u64 ino_gen, | |
f959492f FM |
3378 | u64 parent_ino, |
3379 | struct list_head *new_refs, | |
84471e24 FM |
3380 | struct list_head *deleted_refs, |
3381 | const bool is_orphan) | |
9f03740a FDBM |
3382 | { |
3383 | struct rb_node **p = &sctx->pending_dir_moves.rb_node; | |
3384 | struct rb_node *parent = NULL; | |
73b802f4 | 3385 | struct pending_dir_move *entry = NULL, *pm; |
9f03740a FDBM |
3386 | struct recorded_ref *cur; |
3387 | int exists = 0; | |
3388 | int ret; | |
3389 | ||
e780b0d1 | 3390 | pm = kmalloc(sizeof(*pm), GFP_KERNEL); |
9f03740a FDBM |
3391 | if (!pm) |
3392 | return -ENOMEM; | |
3393 | pm->parent_ino = parent_ino; | |
bfa7e1f8 FM |
3394 | pm->ino = ino; |
3395 | pm->gen = ino_gen; | |
9f03740a FDBM |
3396 | INIT_LIST_HEAD(&pm->list); |
3397 | INIT_LIST_HEAD(&pm->update_refs); | |
3398 | RB_CLEAR_NODE(&pm->node); | |
3399 | ||
3400 | while (*p) { | |
3401 | parent = *p; | |
3402 | entry = rb_entry(parent, struct pending_dir_move, node); | |
3403 | if (parent_ino < entry->parent_ino) { | |
3404 | p = &(*p)->rb_left; | |
3405 | } else if (parent_ino > entry->parent_ino) { | |
3406 | p = &(*p)->rb_right; | |
3407 | } else { | |
3408 | exists = 1; | |
3409 | break; | |
3410 | } | |
3411 | } | |
3412 | ||
f959492f | 3413 | list_for_each_entry(cur, deleted_refs, list) { |
9f03740a FDBM |
3414 | ret = dup_ref(cur, &pm->update_refs); |
3415 | if (ret < 0) | |
3416 | goto out; | |
3417 | } | |
f959492f | 3418 | list_for_each_entry(cur, new_refs, list) { |
9f03740a FDBM |
3419 | ret = dup_ref(cur, &pm->update_refs); |
3420 | if (ret < 0) | |
3421 | goto out; | |
3422 | } | |
3423 | ||
8b191a68 | 3424 | ret = add_waiting_dir_move(sctx, pm->ino, is_orphan); |
9f03740a FDBM |
3425 | if (ret) |
3426 | goto out; | |
3427 | ||
3428 | if (exists) { | |
3429 | list_add_tail(&pm->list, &entry->list); | |
3430 | } else { | |
3431 | rb_link_node(&pm->node, parent, p); | |
3432 | rb_insert_color(&pm->node, &sctx->pending_dir_moves); | |
3433 | } | |
3434 | ret = 0; | |
3435 | out: | |
3436 | if (ret) { | |
3437 | __free_recorded_refs(&pm->update_refs); | |
3438 | kfree(pm); | |
3439 | } | |
3440 | return ret; | |
3441 | } | |
3442 | ||
3443 | static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx, | |
3444 | u64 parent_ino) | |
3445 | { | |
3446 | struct rb_node *n = sctx->pending_dir_moves.rb_node; | |
3447 | struct pending_dir_move *entry; | |
3448 | ||
3449 | while (n) { | |
3450 | entry = rb_entry(n, struct pending_dir_move, node); | |
3451 | if (parent_ino < entry->parent_ino) | |
3452 | n = n->rb_left; | |
3453 | else if (parent_ino > entry->parent_ino) | |
3454 | n = n->rb_right; | |
3455 | else | |
3456 | return entry; | |
3457 | } | |
3458 | return NULL; | |
3459 | } | |
3460 | ||
801bec36 RK |
3461 | static int path_loop(struct send_ctx *sctx, struct fs_path *name, |
3462 | u64 ino, u64 gen, u64 *ancestor_ino) | |
3463 | { | |
3464 | int ret = 0; | |
3465 | u64 parent_inode = 0; | |
3466 | u64 parent_gen = 0; | |
3467 | u64 start_ino = ino; | |
3468 | ||
3469 | *ancestor_ino = 0; | |
3470 | while (ino != BTRFS_FIRST_FREE_OBJECTID) { | |
3471 | fs_path_reset(name); | |
3472 | ||
0b3f407e | 3473 | if (is_waiting_for_rm(sctx, ino, gen)) |
801bec36 RK |
3474 | break; |
3475 | if (is_waiting_for_move(sctx, ino)) { | |
3476 | if (*ancestor_ino == 0) | |
3477 | *ancestor_ino = ino; | |
3478 | ret = get_first_ref(sctx->parent_root, ino, | |
3479 | &parent_inode, &parent_gen, name); | |
3480 | } else { | |
3481 | ret = __get_cur_name_and_parent(sctx, ino, gen, | |
3482 | &parent_inode, | |
3483 | &parent_gen, name); | |
3484 | if (ret > 0) { | |
3485 | ret = 0; | |
3486 | break; | |
3487 | } | |
3488 | } | |
3489 | if (ret < 0) | |
3490 | break; | |
3491 | if (parent_inode == start_ino) { | |
3492 | ret = 1; | |
3493 | if (*ancestor_ino == 0) | |
3494 | *ancestor_ino = ino; | |
3495 | break; | |
3496 | } | |
3497 | ino = parent_inode; | |
3498 | gen = parent_gen; | |
3499 | } | |
3500 | return ret; | |
3501 | } | |
3502 | ||
9f03740a FDBM |
3503 | static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) |
3504 | { | |
3505 | struct fs_path *from_path = NULL; | |
3506 | struct fs_path *to_path = NULL; | |
2b863a13 | 3507 | struct fs_path *name = NULL; |
9f03740a FDBM |
3508 | u64 orig_progress = sctx->send_progress; |
3509 | struct recorded_ref *cur; | |
2b863a13 | 3510 | u64 parent_ino, parent_gen; |
9dc44214 FM |
3511 | struct waiting_dir_move *dm = NULL; |
3512 | u64 rmdir_ino = 0; | |
0b3f407e | 3513 | u64 rmdir_gen; |
801bec36 RK |
3514 | u64 ancestor; |
3515 | bool is_orphan; | |
9f03740a FDBM |
3516 | int ret; |
3517 | ||
2b863a13 | 3518 | name = fs_path_alloc(); |
9f03740a | 3519 | from_path = fs_path_alloc(); |
2b863a13 FM |
3520 | if (!name || !from_path) { |
3521 | ret = -ENOMEM; | |
3522 | goto out; | |
3523 | } | |
9f03740a | 3524 | |
9dc44214 FM |
3525 | dm = get_waiting_dir_move(sctx, pm->ino); |
3526 | ASSERT(dm); | |
3527 | rmdir_ino = dm->rmdir_ino; | |
0b3f407e | 3528 | rmdir_gen = dm->rmdir_gen; |
801bec36 | 3529 | is_orphan = dm->orphanized; |
9dc44214 | 3530 | free_waiting_dir_move(sctx, dm); |
2b863a13 | 3531 | |
801bec36 | 3532 | if (is_orphan) { |
84471e24 FM |
3533 | ret = gen_unique_name(sctx, pm->ino, |
3534 | pm->gen, from_path); | |
3535 | } else { | |
3536 | ret = get_first_ref(sctx->parent_root, pm->ino, | |
3537 | &parent_ino, &parent_gen, name); | |
3538 | if (ret < 0) | |
3539 | goto out; | |
3540 | ret = get_cur_path(sctx, parent_ino, parent_gen, | |
3541 | from_path); | |
3542 | if (ret < 0) | |
3543 | goto out; | |
3544 | ret = fs_path_add_path(from_path, name); | |
3545 | } | |
c992ec94 FM |
3546 | if (ret < 0) |
3547 | goto out; | |
2b863a13 | 3548 | |
f959492f | 3549 | sctx->send_progress = sctx->cur_ino + 1; |
801bec36 | 3550 | ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor); |
7969e77a FM |
3551 | if (ret < 0) |
3552 | goto out; | |
801bec36 RK |
3553 | if (ret) { |
3554 | LIST_HEAD(deleted_refs); | |
3555 | ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID); | |
3556 | ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor, | |
3557 | &pm->update_refs, &deleted_refs, | |
3558 | is_orphan); | |
3559 | if (ret < 0) | |
3560 | goto out; | |
3561 | if (rmdir_ino) { | |
3562 | dm = get_waiting_dir_move(sctx, pm->ino); | |
3563 | ASSERT(dm); | |
3564 | dm->rmdir_ino = rmdir_ino; | |
0b3f407e | 3565 | dm->rmdir_gen = rmdir_gen; |
801bec36 RK |
3566 | } |
3567 | goto out; | |
3568 | } | |
c992ec94 FM |
3569 | fs_path_reset(name); |
3570 | to_path = name; | |
2b863a13 | 3571 | name = NULL; |
9f03740a FDBM |
3572 | ret = get_cur_path(sctx, pm->ino, pm->gen, to_path); |
3573 | if (ret < 0) | |
3574 | goto out; | |
3575 | ||
3576 | ret = send_rename(sctx, from_path, to_path); | |
3577 | if (ret < 0) | |
3578 | goto out; | |
3579 | ||
9dc44214 FM |
3580 | if (rmdir_ino) { |
3581 | struct orphan_dir_info *odi; | |
0f96f517 | 3582 | u64 gen; |
9dc44214 | 3583 | |
0b3f407e | 3584 | odi = get_orphan_dir_info(sctx, rmdir_ino, rmdir_gen); |
9dc44214 FM |
3585 | if (!odi) { |
3586 | /* already deleted */ | |
3587 | goto finish; | |
3588 | } | |
0f96f517 RK |
3589 | gen = odi->gen; |
3590 | ||
24970ccb | 3591 | ret = can_rmdir(sctx, rmdir_ino, gen); |
9dc44214 FM |
3592 | if (ret < 0) |
3593 | goto out; | |
3594 | if (!ret) | |
3595 | goto finish; | |
3596 | ||
3597 | name = fs_path_alloc(); | |
3598 | if (!name) { | |
3599 | ret = -ENOMEM; | |
3600 | goto out; | |
3601 | } | |
0f96f517 | 3602 | ret = get_cur_path(sctx, rmdir_ino, gen, name); |
9dc44214 FM |
3603 | if (ret < 0) |
3604 | goto out; | |
3605 | ret = send_rmdir(sctx, name); | |
3606 | if (ret < 0) | |
3607 | goto out; | |
9dc44214 FM |
3608 | } |
3609 | ||
3610 | finish: | |
3e49363b | 3611 | ret = cache_dir_utimes(sctx, pm->ino, pm->gen); |
9f03740a FDBM |
3612 | if (ret < 0) |
3613 | goto out; | |
3614 | ||
3615 | /* | |
3616 | * After rename/move, need to update the utimes of both new parent(s) | |
3617 | * and old parent(s). | |
3618 | */ | |
3619 | list_for_each_entry(cur, &pm->update_refs, list) { | |
764433a1 RK |
3620 | /* |
3621 | * The parent inode might have been deleted in the send snapshot | |
3622 | */ | |
7e93f6dc | 3623 | ret = get_inode_info(sctx->send_root, cur->dir, NULL); |
764433a1 RK |
3624 | if (ret == -ENOENT) { |
3625 | ret = 0; | |
9dc44214 | 3626 | continue; |
764433a1 RK |
3627 | } |
3628 | if (ret < 0) | |
3629 | goto out; | |
3630 | ||
3e49363b | 3631 | ret = cache_dir_utimes(sctx, cur->dir, cur->dir_gen); |
9f03740a FDBM |
3632 | if (ret < 0) |
3633 | goto out; | |
3634 | } | |
3635 | ||
3636 | out: | |
2b863a13 | 3637 | fs_path_free(name); |
9f03740a FDBM |
3638 | fs_path_free(from_path); |
3639 | fs_path_free(to_path); | |
3640 | sctx->send_progress = orig_progress; | |
3641 | ||
3642 | return ret; | |
3643 | } | |
3644 | ||
3645 | static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m) | |
3646 | { | |
3647 | if (!list_empty(&m->list)) | |
3648 | list_del(&m->list); | |
3649 | if (!RB_EMPTY_NODE(&m->node)) | |
3650 | rb_erase(&m->node, &sctx->pending_dir_moves); | |
3651 | __free_recorded_refs(&m->update_refs); | |
3652 | kfree(m); | |
3653 | } | |
3654 | ||
a4390aee RK |
3655 | static void tail_append_pending_moves(struct send_ctx *sctx, |
3656 | struct pending_dir_move *moves, | |
9f03740a FDBM |
3657 | struct list_head *stack) |
3658 | { | |
3659 | if (list_empty(&moves->list)) { | |
3660 | list_add_tail(&moves->list, stack); | |
3661 | } else { | |
3662 | LIST_HEAD(list); | |
3663 | list_splice_init(&moves->list, &list); | |
3664 | list_add_tail(&moves->list, stack); | |
3665 | list_splice_tail(&list, stack); | |
3666 | } | |
a4390aee RK |
3667 | if (!RB_EMPTY_NODE(&moves->node)) { |
3668 | rb_erase(&moves->node, &sctx->pending_dir_moves); | |
3669 | RB_CLEAR_NODE(&moves->node); | |
3670 | } | |
9f03740a FDBM |
3671 | } |
3672 | ||
3673 | static int apply_children_dir_moves(struct send_ctx *sctx) | |
3674 | { | |
3675 | struct pending_dir_move *pm; | |
3676 | struct list_head stack; | |
3677 | u64 parent_ino = sctx->cur_ino; | |
3678 | int ret = 0; | |
3679 | ||
3680 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3681 | if (!pm) | |
3682 | return 0; | |
3683 | ||
3684 | INIT_LIST_HEAD(&stack); | |
a4390aee | 3685 | tail_append_pending_moves(sctx, pm, &stack); |
9f03740a FDBM |
3686 | |
3687 | while (!list_empty(&stack)) { | |
3688 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3689 | parent_ino = pm->ino; | |
3690 | ret = apply_dir_move(sctx, pm); | |
3691 | free_pending_move(sctx, pm); | |
3692 | if (ret) | |
3693 | goto out; | |
3694 | pm = get_pending_dir_moves(sctx, parent_ino); | |
3695 | if (pm) | |
a4390aee | 3696 | tail_append_pending_moves(sctx, pm, &stack); |
9f03740a FDBM |
3697 | } |
3698 | return 0; | |
3699 | ||
3700 | out: | |
3701 | while (!list_empty(&stack)) { | |
3702 | pm = list_first_entry(&stack, struct pending_dir_move, list); | |
3703 | free_pending_move(sctx, pm); | |
3704 | } | |
3705 | return ret; | |
3706 | } | |
3707 | ||
84471e24 FM |
3708 | /* |
3709 | * We might need to delay a directory rename even when no ancestor directory | |
3710 | * (in the send root) with a higher inode number than ours (sctx->cur_ino) was | |
3711 | * renamed. This happens when we rename a directory to the old name (the name | |
3712 | * in the parent root) of some other unrelated directory that got its rename | |
3713 | * delayed due to some ancestor with higher number that got renamed. | |
3714 | * | |
3715 | * Example: | |
3716 | * | |
3717 | * Parent snapshot: | |
3718 | * . (ino 256) | |
3719 | * |---- a/ (ino 257) | |
3720 | * | |---- file (ino 260) | |
3721 | * | | |
3722 | * |---- b/ (ino 258) | |
3723 | * |---- c/ (ino 259) | |
3724 | * | |
3725 | * Send snapshot: | |
3726 | * . (ino 256) | |
3727 | * |---- a/ (ino 258) | |
3728 | * |---- x/ (ino 259) | |
3729 | * |---- y/ (ino 257) | |
3730 | * |----- file (ino 260) | |
3731 | * | |
3732 | * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257 | |
3733 | * from 'a' to 'x/y' happening first, which in turn depends on the rename of | |
3734 | * inode 259 from 'c' to 'x'. So the order of rename commands the send stream | |
3735 | * must issue is: | |
3736 | * | |
3737 | * 1 - rename 259 from 'c' to 'x' | |
3738 | * 2 - rename 257 from 'a' to 'x/y' | |
3739 | * 3 - rename 258 from 'b' to 'a' | |
3740 | * | |
3741 | * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can | |
3742 | * be done right away and < 0 on error. | |
3743 | */ | |
3744 | static int wait_for_dest_dir_move(struct send_ctx *sctx, | |
3745 | struct recorded_ref *parent_ref, | |
3746 | const bool is_orphan) | |
3747 | { | |
2ff7e61e | 3748 | struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info; |
84471e24 FM |
3749 | struct btrfs_path *path; |
3750 | struct btrfs_key key; | |
3751 | struct btrfs_key di_key; | |
3752 | struct btrfs_dir_item *di; | |
3753 | u64 left_gen; | |
3754 | u64 right_gen; | |
3755 | int ret = 0; | |
801bec36 | 3756 | struct waiting_dir_move *wdm; |
84471e24 FM |
3757 | |
3758 | if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) | |
3759 | return 0; | |
3760 | ||
3761 | path = alloc_path_for_send(); | |
3762 | if (!path) | |
3763 | return -ENOMEM; | |
3764 | ||
3765 | key.objectid = parent_ref->dir; | |
3766 | key.type = BTRFS_DIR_ITEM_KEY; | |
3767 | key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); | |
3768 | ||
3769 | ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); | |
3770 | if (ret < 0) { | |
3771 | goto out; | |
3772 | } else if (ret > 0) { | |
3773 | ret = 0; | |
3774 | goto out; | |
3775 | } | |
3776 | ||
2ff7e61e JM |
3777 | di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name, |
3778 | parent_ref->name_len); | |
84471e24 FM |
3779 | if (!di) { |
3780 | ret = 0; | |
3781 | goto out; | |
3782 | } | |
3783 | /* | |
3784 | * di_key.objectid has the number of the inode that has a dentry in the | |
3785 | * parent directory with the same name that sctx->cur_ino is being | |
3786 | * renamed to. We need to check if that inode is in the send root as | |
3787 | * well and if it is currently marked as an inode with a pending rename, | |
3788 | * if it is, we need to delay the rename of sctx->cur_ino as well, so | |
3789 | * that it happens after that other inode is renamed. | |
3790 | */ | |
3791 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); | |
3792 | if (di_key.type != BTRFS_INODE_ITEM_KEY) { | |
3793 | ret = 0; | |
3794 | goto out; | |
3795 | } | |
3796 | ||
7e93f6dc | 3797 | ret = get_inode_gen(sctx->parent_root, di_key.objectid, &left_gen); |
84471e24 FM |
3798 | if (ret < 0) |
3799 | goto out; | |
7e93f6dc | 3800 | ret = get_inode_gen(sctx->send_root, di_key.objectid, &right_gen); |
84471e24 FM |
3801 | if (ret < 0) { |
3802 | if (ret == -ENOENT) | |
3803 | ret = 0; | |
3804 | goto out; | |
3805 | } | |
3806 | ||
3807 | /* Different inode, no need to delay the rename of sctx->cur_ino */ | |
3808 | if (right_gen != left_gen) { | |
3809 | ret = 0; | |
3810 | goto out; | |
3811 | } | |
3812 | ||
801bec36 RK |
3813 | wdm = get_waiting_dir_move(sctx, di_key.objectid); |
3814 | if (wdm && !wdm->orphanized) { | |
84471e24 FM |
3815 | ret = add_pending_dir_move(sctx, |
3816 | sctx->cur_ino, | |
3817 | sctx->cur_inode_gen, | |
3818 | di_key.objectid, | |
3819 | &sctx->new_refs, | |
3820 | &sctx->deleted_refs, | |
3821 | is_orphan); | |
3822 | if (!ret) | |
3823 | ret = 1; | |
3824 | } | |
3825 | out: | |
3826 | btrfs_free_path(path); | |
3827 | return ret; | |
3828 | } | |
3829 | ||
80aa6027 | 3830 | /* |
ea37d599 FM |
3831 | * Check if inode ino2, or any of its ancestors, is inode ino1. |
3832 | * Return 1 if true, 0 if false and < 0 on error. | |
3833 | */ | |
3834 | static int check_ino_in_path(struct btrfs_root *root, | |
3835 | const u64 ino1, | |
3836 | const u64 ino1_gen, | |
3837 | const u64 ino2, | |
3838 | const u64 ino2_gen, | |
3839 | struct fs_path *fs_path) | |
3840 | { | |
3841 | u64 ino = ino2; | |
3842 | ||
3843 | if (ino1 == ino2) | |
3844 | return ino1_gen == ino2_gen; | |
3845 | ||
3846 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { | |
3847 | u64 parent; | |
3848 | u64 parent_gen; | |
3849 | int ret; | |
3850 | ||
3851 | fs_path_reset(fs_path); | |
3852 | ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path); | |
3853 | if (ret < 0) | |
3854 | return ret; | |
3855 | if (parent == ino1) | |
3856 | return parent_gen == ino1_gen; | |
3857 | ino = parent; | |
3858 | } | |
3859 | return 0; | |
3860 | } | |
3861 | ||
3862 | /* | |
35a68080 | 3863 | * Check if inode ino1 is an ancestor of inode ino2 in the given root for any |
ea37d599 | 3864 | * possible path (in case ino2 is not a directory and has multiple hard links). |
80aa6027 FM |
3865 | * Return 1 if true, 0 if false and < 0 on error. |
3866 | */ | |
3867 | static int is_ancestor(struct btrfs_root *root, | |
3868 | const u64 ino1, | |
3869 | const u64 ino1_gen, | |
3870 | const u64 ino2, | |
3871 | struct fs_path *fs_path) | |
3872 | { | |
ea37d599 | 3873 | bool free_fs_path = false; |
72c3668f | 3874 | int ret = 0; |
35a68080 | 3875 | int iter_ret = 0; |
ea37d599 FM |
3876 | struct btrfs_path *path = NULL; |
3877 | struct btrfs_key key; | |
72c3668f FM |
3878 | |
3879 | if (!fs_path) { | |
3880 | fs_path = fs_path_alloc(); | |
3881 | if (!fs_path) | |
3882 | return -ENOMEM; | |
ea37d599 | 3883 | free_fs_path = true; |
72c3668f | 3884 | } |
80aa6027 | 3885 | |
ea37d599 FM |
3886 | path = alloc_path_for_send(); |
3887 | if (!path) { | |
3888 | ret = -ENOMEM; | |
3889 | goto out; | |
3890 | } | |
80aa6027 | 3891 | |
ea37d599 FM |
3892 | key.objectid = ino2; |
3893 | key.type = BTRFS_INODE_REF_KEY; | |
3894 | key.offset = 0; | |
3895 | ||
35a68080 | 3896 | btrfs_for_each_slot(root, &key, &key, path, iter_ret) { |
ea37d599 FM |
3897 | struct extent_buffer *leaf = path->nodes[0]; |
3898 | int slot = path->slots[0]; | |
3899 | u32 cur_offset = 0; | |
3900 | u32 item_size; | |
3901 | ||
ea37d599 FM |
3902 | if (key.objectid != ino2) |
3903 | break; | |
3904 | if (key.type != BTRFS_INODE_REF_KEY && | |
3905 | key.type != BTRFS_INODE_EXTREF_KEY) | |
3906 | break; | |
3907 | ||
3212fa14 | 3908 | item_size = btrfs_item_size(leaf, slot); |
ea37d599 FM |
3909 | while (cur_offset < item_size) { |
3910 | u64 parent; | |
3911 | u64 parent_gen; | |
3912 | ||
3913 | if (key.type == BTRFS_INODE_EXTREF_KEY) { | |
3914 | unsigned long ptr; | |
3915 | struct btrfs_inode_extref *extref; | |
3916 | ||
3917 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3918 | extref = (struct btrfs_inode_extref *) | |
3919 | (ptr + cur_offset); | |
3920 | parent = btrfs_inode_extref_parent(leaf, | |
3921 | extref); | |
3922 | cur_offset += sizeof(*extref); | |
3923 | cur_offset += btrfs_inode_extref_name_len(leaf, | |
3924 | extref); | |
3925 | } else { | |
3926 | parent = key.offset; | |
3927 | cur_offset = item_size; | |
3928 | } | |
3929 | ||
7e93f6dc | 3930 | ret = get_inode_gen(root, parent, &parent_gen); |
ea37d599 FM |
3931 | if (ret < 0) |
3932 | goto out; | |
3933 | ret = check_ino_in_path(root, ino1, ino1_gen, | |
3934 | parent, parent_gen, fs_path); | |
3935 | if (ret) | |
3936 | goto out; | |
80aa6027 | 3937 | } |
80aa6027 | 3938 | } |
ea37d599 | 3939 | ret = 0; |
35a68080 GN |
3940 | if (iter_ret < 0) |
3941 | ret = iter_ret; | |
3942 | ||
3943 | out: | |
ea37d599 FM |
3944 | btrfs_free_path(path); |
3945 | if (free_fs_path) | |
72c3668f FM |
3946 | fs_path_free(fs_path); |
3947 | return ret; | |
80aa6027 FM |
3948 | } |
3949 | ||
9f03740a | 3950 | static int wait_for_parent_move(struct send_ctx *sctx, |
8b191a68 FM |
3951 | struct recorded_ref *parent_ref, |
3952 | const bool is_orphan) | |
9f03740a | 3953 | { |
f959492f | 3954 | int ret = 0; |
9f03740a | 3955 | u64 ino = parent_ref->dir; |
fe9c798d | 3956 | u64 ino_gen = parent_ref->dir_gen; |
9f03740a | 3957 | u64 parent_ino_before, parent_ino_after; |
9f03740a FDBM |
3958 | struct fs_path *path_before = NULL; |
3959 | struct fs_path *path_after = NULL; | |
3960 | int len1, len2; | |
9f03740a FDBM |
3961 | |
3962 | path_after = fs_path_alloc(); | |
f959492f FM |
3963 | path_before = fs_path_alloc(); |
3964 | if (!path_after || !path_before) { | |
9f03740a FDBM |
3965 | ret = -ENOMEM; |
3966 | goto out; | |
3967 | } | |
3968 | ||
bfa7e1f8 | 3969 | /* |
f959492f FM |
3970 | * Our current directory inode may not yet be renamed/moved because some |
3971 | * ancestor (immediate or not) has to be renamed/moved first. So find if | |
3972 | * such ancestor exists and make sure our own rename/move happens after | |
80aa6027 FM |
3973 | * that ancestor is processed to avoid path build infinite loops (done |
3974 | * at get_cur_path()). | |
bfa7e1f8 | 3975 | */ |
f959492f | 3976 | while (ino > BTRFS_FIRST_FREE_OBJECTID) { |
fe9c798d FM |
3977 | u64 parent_ino_after_gen; |
3978 | ||
f959492f | 3979 | if (is_waiting_for_move(sctx, ino)) { |
80aa6027 FM |
3980 | /* |
3981 | * If the current inode is an ancestor of ino in the | |
3982 | * parent root, we need to delay the rename of the | |
3983 | * current inode, otherwise don't delayed the rename | |
3984 | * because we can end up with a circular dependency | |
3985 | * of renames, resulting in some directories never | |
3986 | * getting the respective rename operations issued in | |
3987 | * the send stream or getting into infinite path build | |
3988 | * loops. | |
3989 | */ | |
3990 | ret = is_ancestor(sctx->parent_root, | |
3991 | sctx->cur_ino, sctx->cur_inode_gen, | |
3992 | ino, path_before); | |
4122ea64 FM |
3993 | if (ret) |
3994 | break; | |
f959492f | 3995 | } |
bfa7e1f8 FM |
3996 | |
3997 | fs_path_reset(path_before); | |
3998 | fs_path_reset(path_after); | |
3999 | ||
4000 | ret = get_first_ref(sctx->send_root, ino, &parent_ino_after, | |
fe9c798d | 4001 | &parent_ino_after_gen, path_after); |
bfa7e1f8 FM |
4002 | if (ret < 0) |
4003 | goto out; | |
4004 | ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before, | |
4005 | NULL, path_before); | |
f959492f | 4006 | if (ret < 0 && ret != -ENOENT) { |
bfa7e1f8 | 4007 | goto out; |
f959492f | 4008 | } else if (ret == -ENOENT) { |
bf8e8ca6 | 4009 | ret = 0; |
f959492f | 4010 | break; |
bfa7e1f8 FM |
4011 | } |
4012 | ||
4013 | len1 = fs_path_len(path_before); | |
4014 | len2 = fs_path_len(path_after); | |
f959492f FM |
4015 | if (ino > sctx->cur_ino && |
4016 | (parent_ino_before != parent_ino_after || len1 != len2 || | |
4017 | memcmp(path_before->start, path_after->start, len1))) { | |
fe9c798d FM |
4018 | u64 parent_ino_gen; |
4019 | ||
7e93f6dc | 4020 | ret = get_inode_gen(sctx->parent_root, ino, &parent_ino_gen); |
fe9c798d FM |
4021 | if (ret < 0) |
4022 | goto out; | |
4023 | if (ino_gen == parent_ino_gen) { | |
4024 | ret = 1; | |
4025 | break; | |
4026 | } | |
bfa7e1f8 | 4027 | } |
bfa7e1f8 | 4028 | ino = parent_ino_after; |
fe9c798d | 4029 | ino_gen = parent_ino_after_gen; |
bfa7e1f8 FM |
4030 | } |
4031 | ||
9f03740a FDBM |
4032 | out: |
4033 | fs_path_free(path_before); | |
4034 | fs_path_free(path_after); | |
4035 | ||
f959492f FM |
4036 | if (ret == 1) { |
4037 | ret = add_pending_dir_move(sctx, | |
4038 | sctx->cur_ino, | |
4039 | sctx->cur_inode_gen, | |
4040 | ino, | |
4041 | &sctx->new_refs, | |
84471e24 | 4042 | &sctx->deleted_refs, |
8b191a68 | 4043 | is_orphan); |
f959492f FM |
4044 | if (!ret) |
4045 | ret = 1; | |
4046 | } | |
4047 | ||
9f03740a FDBM |
4048 | return ret; |
4049 | } | |
4050 | ||
f5962781 FM |
4051 | static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) |
4052 | { | |
4053 | int ret; | |
4054 | struct fs_path *new_path; | |
4055 | ||
4056 | /* | |
4057 | * Our reference's name member points to its full_path member string, so | |
4058 | * we use here a new path. | |
4059 | */ | |
4060 | new_path = fs_path_alloc(); | |
4061 | if (!new_path) | |
4062 | return -ENOMEM; | |
4063 | ||
4064 | ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path); | |
4065 | if (ret < 0) { | |
4066 | fs_path_free(new_path); | |
4067 | return ret; | |
4068 | } | |
4069 | ret = fs_path_add(new_path, ref->name, ref->name_len); | |
4070 | if (ret < 0) { | |
4071 | fs_path_free(new_path); | |
4072 | return ret; | |
4073 | } | |
4074 | ||
4075 | fs_path_free(ref->full_path); | |
4076 | set_ref_path(ref, new_path); | |
4077 | ||
4078 | return 0; | |
4079 | } | |
4080 | ||
9c2b4e03 FM |
4081 | /* |
4082 | * When processing the new references for an inode we may orphanize an existing | |
4083 | * directory inode because its old name conflicts with one of the new references | |
4084 | * of the current inode. Later, when processing another new reference of our | |
4085 | * inode, we might need to orphanize another inode, but the path we have in the | |
4086 | * reference reflects the pre-orphanization name of the directory we previously | |
4087 | * orphanized. For example: | |
4088 | * | |
4089 | * parent snapshot looks like: | |
4090 | * | |
4091 | * . (ino 256) | |
4092 | * |----- f1 (ino 257) | |
4093 | * |----- f2 (ino 258) | |
4094 | * |----- d1/ (ino 259) | |
4095 | * |----- d2/ (ino 260) | |
4096 | * | |
4097 | * send snapshot looks like: | |
4098 | * | |
4099 | * . (ino 256) | |
4100 | * |----- d1 (ino 258) | |
4101 | * |----- f2/ (ino 259) | |
4102 | * |----- f2_link/ (ino 260) | |
4103 | * | |----- f1 (ino 257) | |
4104 | * | | |
4105 | * |----- d2 (ino 258) | |
4106 | * | |
4107 | * When processing inode 257 we compute the name for inode 259 as "d1", and we | |
4108 | * cache it in the name cache. Later when we start processing inode 258, when | |
4109 | * collecting all its new references we set a full path of "d1/d2" for its new | |
4110 | * reference with name "d2". When we start processing the new references we | |
4111 | * start by processing the new reference with name "d1", and this results in | |
4112 | * orphanizing inode 259, since its old reference causes a conflict. Then we | |
4113 | * move on the next new reference, with name "d2", and we find out we must | |
4114 | * orphanize inode 260, as its old reference conflicts with ours - but for the | |
4115 | * orphanization we use a source path corresponding to the path we stored in the | |
4116 | * new reference, which is "d1/d2" and not "o259-6-0/d2" - this makes the | |
4117 | * receiver fail since the path component "d1/" no longer exists, it was renamed | |
4118 | * to "o259-6-0/" when processing the previous new reference. So in this case we | |
4119 | * must recompute the path in the new reference and use it for the new | |
4120 | * orphanization operation. | |
4121 | */ | |
4122 | static int refresh_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) | |
4123 | { | |
4124 | char *name; | |
4125 | int ret; | |
4126 | ||
4127 | name = kmemdup(ref->name, ref->name_len, GFP_KERNEL); | |
4128 | if (!name) | |
4129 | return -ENOMEM; | |
4130 | ||
4131 | fs_path_reset(ref->full_path); | |
4132 | ret = get_cur_path(sctx, ref->dir, ref->dir_gen, ref->full_path); | |
4133 | if (ret < 0) | |
4134 | goto out; | |
4135 | ||
4136 | ret = fs_path_add(ref->full_path, name, ref->name_len); | |
4137 | if (ret < 0) | |
4138 | goto out; | |
4139 | ||
4140 | /* Update the reference's base name pointer. */ | |
4141 | set_ref_path(ref, ref->full_path); | |
4142 | out: | |
4143 | kfree(name); | |
4144 | return ret; | |
4145 | } | |
4146 | ||
31db9f7c AB |
4147 | /* |
4148 | * This does all the move/link/unlink/rmdir magic. | |
4149 | */ | |
9f03740a | 4150 | static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) |
31db9f7c | 4151 | { |
04ab956e | 4152 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
4153 | int ret = 0; |
4154 | struct recorded_ref *cur; | |
1f4692da | 4155 | struct recorded_ref *cur2; |
ba5e8f2e | 4156 | struct list_head check_dirs; |
31db9f7c | 4157 | struct fs_path *valid_path = NULL; |
b24baf69 | 4158 | u64 ow_inode = 0; |
31db9f7c | 4159 | u64 ow_gen; |
f5962781 | 4160 | u64 ow_mode; |
31db9f7c AB |
4161 | int did_overwrite = 0; |
4162 | int is_orphan = 0; | |
29d6d30f | 4163 | u64 last_dir_ino_rm = 0; |
84471e24 | 4164 | bool can_rename = true; |
f5962781 | 4165 | bool orphanized_dir = false; |
fdb13889 | 4166 | bool orphanized_ancestor = false; |
31db9f7c | 4167 | |
04ab956e | 4168 | btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino); |
31db9f7c | 4169 | |
6d85ed05 AB |
4170 | /* |
4171 | * This should never happen as the root dir always has the same ref | |
4172 | * which is always '..' | |
4173 | */ | |
4174 | BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); | |
ba5e8f2e | 4175 | INIT_LIST_HEAD(&check_dirs); |
6d85ed05 | 4176 | |
924794c9 | 4177 | valid_path = fs_path_alloc(); |
31db9f7c AB |
4178 | if (!valid_path) { |
4179 | ret = -ENOMEM; | |
4180 | goto out; | |
4181 | } | |
4182 | ||
31db9f7c AB |
4183 | /* |
4184 | * First, check if the first ref of the current inode was overwritten | |
4185 | * before. If yes, we know that the current inode was already orphanized | |
4186 | * and thus use the orphan name. If not, we can use get_cur_path to | |
4187 | * get the path of the first ref as it would like while receiving at | |
4188 | * this point in time. | |
4189 | * New inodes are always orphan at the beginning, so force to use the | |
4190 | * orphan name in this case. | |
4191 | * The first ref is stored in valid_path and will be updated if it | |
4192 | * gets moved around. | |
4193 | */ | |
4194 | if (!sctx->cur_inode_new) { | |
4195 | ret = did_overwrite_first_ref(sctx, sctx->cur_ino, | |
4196 | sctx->cur_inode_gen); | |
4197 | if (ret < 0) | |
4198 | goto out; | |
4199 | if (ret) | |
4200 | did_overwrite = 1; | |
4201 | } | |
4202 | if (sctx->cur_inode_new || did_overwrite) { | |
4203 | ret = gen_unique_name(sctx, sctx->cur_ino, | |
4204 | sctx->cur_inode_gen, valid_path); | |
4205 | if (ret < 0) | |
4206 | goto out; | |
4207 | is_orphan = 1; | |
4208 | } else { | |
4209 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
4210 | valid_path); | |
4211 | if (ret < 0) | |
4212 | goto out; | |
4213 | } | |
4214 | ||
98272bb7 FM |
4215 | /* |
4216 | * Before doing any rename and link operations, do a first pass on the | |
4217 | * new references to orphanize any unprocessed inodes that may have a | |
4218 | * reference that conflicts with one of the new references of the current | |
4219 | * inode. This needs to happen first because a new reference may conflict | |
4220 | * with the old reference of a parent directory, so we must make sure | |
4221 | * that the path used for link and rename commands don't use an | |
4222 | * orphanized name when an ancestor was not yet orphanized. | |
4223 | * | |
4224 | * Example: | |
4225 | * | |
4226 | * Parent snapshot: | |
4227 | * | |
4228 | * . (ino 256) | |
4229 | * |----- testdir/ (ino 259) | |
4230 | * | |----- a (ino 257) | |
4231 | * | | |
4232 | * |----- b (ino 258) | |
4233 | * | |
4234 | * Send snapshot: | |
4235 | * | |
4236 | * . (ino 256) | |
4237 | * |----- testdir_2/ (ino 259) | |
4238 | * | |----- a (ino 260) | |
4239 | * | | |
4240 | * |----- testdir (ino 257) | |
4241 | * |----- b (ino 257) | |
4242 | * |----- b2 (ino 258) | |
4243 | * | |
4244 | * Processing the new reference for inode 257 with name "b" may happen | |
4245 | * before processing the new reference with name "testdir". If so, we | |
4246 | * must make sure that by the time we send a link command to create the | |
4247 | * hard link "b", inode 259 was already orphanized, since the generated | |
4248 | * path in "valid_path" already contains the orphanized name for 259. | |
4249 | * We are processing inode 257, so only later when processing 259 we do | |
4250 | * the rename operation to change its temporary (orphanized) name to | |
4251 | * "testdir_2". | |
4252 | */ | |
31db9f7c | 4253 | list_for_each_entry(cur, &sctx->new_refs, list) { |
498581f3 | 4254 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL); |
1f4692da AB |
4255 | if (ret < 0) |
4256 | goto out; | |
98272bb7 FM |
4257 | if (ret == inode_state_will_create) |
4258 | continue; | |
1f4692da | 4259 | |
31db9f7c | 4260 | /* |
98272bb7 FM |
4261 | * Check if this new ref would overwrite the first ref of another |
4262 | * unprocessed inode. If yes, orphanize the overwritten inode. | |
4263 | * If we find an overwritten ref that is not the first ref, | |
4264 | * simply unlink it. | |
31db9f7c AB |
4265 | */ |
4266 | ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
4267 | cur->name, cur->name_len, | |
f5962781 | 4268 | &ow_inode, &ow_gen, &ow_mode); |
31db9f7c AB |
4269 | if (ret < 0) |
4270 | goto out; | |
4271 | if (ret) { | |
924794c9 TI |
4272 | ret = is_first_ref(sctx->parent_root, |
4273 | ow_inode, cur->dir, cur->name, | |
4274 | cur->name_len); | |
31db9f7c AB |
4275 | if (ret < 0) |
4276 | goto out; | |
4277 | if (ret) { | |
8996a48c | 4278 | struct name_cache_entry *nce; |
801bec36 | 4279 | struct waiting_dir_move *wdm; |
8996a48c | 4280 | |
9c2b4e03 FM |
4281 | if (orphanized_dir) { |
4282 | ret = refresh_ref_path(sctx, cur); | |
4283 | if (ret < 0) | |
4284 | goto out; | |
4285 | } | |
4286 | ||
31db9f7c AB |
4287 | ret = orphanize_inode(sctx, ow_inode, ow_gen, |
4288 | cur->full_path); | |
4289 | if (ret < 0) | |
4290 | goto out; | |
f5962781 FM |
4291 | if (S_ISDIR(ow_mode)) |
4292 | orphanized_dir = true; | |
801bec36 RK |
4293 | |
4294 | /* | |
4295 | * If ow_inode has its rename operation delayed | |
4296 | * make sure that its orphanized name is used in | |
4297 | * the source path when performing its rename | |
4298 | * operation. | |
4299 | */ | |
8c139e1d FM |
4300 | wdm = get_waiting_dir_move(sctx, ow_inode); |
4301 | if (wdm) | |
801bec36 | 4302 | wdm->orphanized = true; |
801bec36 | 4303 | |
8996a48c FM |
4304 | /* |
4305 | * Make sure we clear our orphanized inode's | |
4306 | * name from the name cache. This is because the | |
4307 | * inode ow_inode might be an ancestor of some | |
4308 | * other inode that will be orphanized as well | |
4309 | * later and has an inode number greater than | |
4310 | * sctx->send_progress. We need to prevent | |
4311 | * future name lookups from using the old name | |
4312 | * and get instead the orphan name. | |
4313 | */ | |
4314 | nce = name_cache_search(sctx, ow_inode, ow_gen); | |
c48545de FM |
4315 | if (nce) |
4316 | btrfs_lru_cache_remove(&sctx->name_cache, | |
4317 | &nce->entry); | |
801bec36 RK |
4318 | |
4319 | /* | |
4320 | * ow_inode might currently be an ancestor of | |
4321 | * cur_ino, therefore compute valid_path (the | |
4322 | * current path of cur_ino) again because it | |
4323 | * might contain the pre-orphanization name of | |
4324 | * ow_inode, which is no longer valid. | |
4325 | */ | |
72c3668f FM |
4326 | ret = is_ancestor(sctx->parent_root, |
4327 | ow_inode, ow_gen, | |
4328 | sctx->cur_ino, NULL); | |
4329 | if (ret > 0) { | |
fdb13889 | 4330 | orphanized_ancestor = true; |
72c3668f FM |
4331 | fs_path_reset(valid_path); |
4332 | ret = get_cur_path(sctx, sctx->cur_ino, | |
4333 | sctx->cur_inode_gen, | |
4334 | valid_path); | |
4335 | } | |
801bec36 RK |
4336 | if (ret < 0) |
4337 | goto out; | |
31db9f7c | 4338 | } else { |
d8ac76cd FM |
4339 | /* |
4340 | * If we previously orphanized a directory that | |
4341 | * collided with a new reference that we already | |
4342 | * processed, recompute the current path because | |
4343 | * that directory may be part of the path. | |
4344 | */ | |
4345 | if (orphanized_dir) { | |
4346 | ret = refresh_ref_path(sctx, cur); | |
4347 | if (ret < 0) | |
4348 | goto out; | |
4349 | } | |
31db9f7c AB |
4350 | ret = send_unlink(sctx, cur->full_path); |
4351 | if (ret < 0) | |
4352 | goto out; | |
4353 | } | |
4354 | } | |
4355 | ||
98272bb7 FM |
4356 | } |
4357 | ||
4358 | list_for_each_entry(cur, &sctx->new_refs, list) { | |
4359 | /* | |
4360 | * We may have refs where the parent directory does not exist | |
4361 | * yet. This happens if the parent directories inum is higher | |
4362 | * than the current inum. To handle this case, we create the | |
4363 | * parent directory out of order. But we need to check if this | |
4364 | * did already happen before due to other refs in the same dir. | |
4365 | */ | |
498581f3 | 4366 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL); |
98272bb7 FM |
4367 | if (ret < 0) |
4368 | goto out; | |
4369 | if (ret == inode_state_will_create) { | |
4370 | ret = 0; | |
4371 | /* | |
4372 | * First check if any of the current inodes refs did | |
4373 | * already create the dir. | |
4374 | */ | |
4375 | list_for_each_entry(cur2, &sctx->new_refs, list) { | |
4376 | if (cur == cur2) | |
4377 | break; | |
4378 | if (cur2->dir == cur->dir) { | |
4379 | ret = 1; | |
4380 | break; | |
4381 | } | |
4382 | } | |
4383 | ||
4384 | /* | |
4385 | * If that did not happen, check if a previous inode | |
4386 | * did already create the dir. | |
4387 | */ | |
4388 | if (!ret) | |
4389 | ret = did_create_dir(sctx, cur->dir); | |
4390 | if (ret < 0) | |
4391 | goto out; | |
4392 | if (!ret) { | |
4393 | ret = send_create_inode(sctx, cur->dir); | |
4394 | if (ret < 0) | |
4395 | goto out; | |
e8a7f49d | 4396 | cache_dir_created(sctx, cur->dir); |
98272bb7 FM |
4397 | } |
4398 | } | |
4399 | ||
84471e24 FM |
4400 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { |
4401 | ret = wait_for_dest_dir_move(sctx, cur, is_orphan); | |
4402 | if (ret < 0) | |
4403 | goto out; | |
4404 | if (ret == 1) { | |
4405 | can_rename = false; | |
4406 | *pending_move = 1; | |
4407 | } | |
4408 | } | |
4409 | ||
8b191a68 FM |
4410 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root && |
4411 | can_rename) { | |
4412 | ret = wait_for_parent_move(sctx, cur, is_orphan); | |
4413 | if (ret < 0) | |
4414 | goto out; | |
4415 | if (ret == 1) { | |
4416 | can_rename = false; | |
4417 | *pending_move = 1; | |
4418 | } | |
4419 | } | |
4420 | ||
31db9f7c AB |
4421 | /* |
4422 | * link/move the ref to the new place. If we have an orphan | |
4423 | * inode, move it and update valid_path. If not, link or move | |
4424 | * it depending on the inode mode. | |
4425 | */ | |
84471e24 | 4426 | if (is_orphan && can_rename) { |
31db9f7c AB |
4427 | ret = send_rename(sctx, valid_path, cur->full_path); |
4428 | if (ret < 0) | |
4429 | goto out; | |
4430 | is_orphan = 0; | |
4431 | ret = fs_path_copy(valid_path, cur->full_path); | |
4432 | if (ret < 0) | |
4433 | goto out; | |
84471e24 | 4434 | } else if (can_rename) { |
31db9f7c AB |
4435 | if (S_ISDIR(sctx->cur_inode_mode)) { |
4436 | /* | |
4437 | * Dirs can't be linked, so move it. For moved | |
4438 | * dirs, we always have one new and one deleted | |
4439 | * ref. The deleted ref is ignored later. | |
4440 | */ | |
8b191a68 FM |
4441 | ret = send_rename(sctx, valid_path, |
4442 | cur->full_path); | |
4443 | if (!ret) | |
4444 | ret = fs_path_copy(valid_path, | |
4445 | cur->full_path); | |
31db9f7c AB |
4446 | if (ret < 0) |
4447 | goto out; | |
4448 | } else { | |
f5962781 FM |
4449 | /* |
4450 | * We might have previously orphanized an inode | |
4451 | * which is an ancestor of our current inode, | |
4452 | * so our reference's full path, which was | |
4453 | * computed before any such orphanizations, must | |
4454 | * be updated. | |
4455 | */ | |
4456 | if (orphanized_dir) { | |
4457 | ret = update_ref_path(sctx, cur); | |
4458 | if (ret < 0) | |
4459 | goto out; | |
4460 | } | |
31db9f7c AB |
4461 | ret = send_link(sctx, cur->full_path, |
4462 | valid_path); | |
4463 | if (ret < 0) | |
4464 | goto out; | |
4465 | } | |
4466 | } | |
ba5e8f2e | 4467 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4468 | if (ret < 0) |
4469 | goto out; | |
4470 | } | |
4471 | ||
4472 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) { | |
4473 | /* | |
4474 | * Check if we can already rmdir the directory. If not, | |
4475 | * orphanize it. For every dir item inside that gets deleted | |
4476 | * later, we do this check again and rmdir it then if possible. | |
4477 | * See the use of check_dirs for more details. | |
4478 | */ | |
24970ccb | 4479 | ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen); |
31db9f7c AB |
4480 | if (ret < 0) |
4481 | goto out; | |
4482 | if (ret) { | |
4483 | ret = send_rmdir(sctx, valid_path); | |
4484 | if (ret < 0) | |
4485 | goto out; | |
4486 | } else if (!is_orphan) { | |
4487 | ret = orphanize_inode(sctx, sctx->cur_ino, | |
4488 | sctx->cur_inode_gen, valid_path); | |
4489 | if (ret < 0) | |
4490 | goto out; | |
4491 | is_orphan = 1; | |
4492 | } | |
4493 | ||
4494 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
ba5e8f2e | 4495 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4496 | if (ret < 0) |
4497 | goto out; | |
4498 | } | |
ccf1626b AB |
4499 | } else if (S_ISDIR(sctx->cur_inode_mode) && |
4500 | !list_empty(&sctx->deleted_refs)) { | |
4501 | /* | |
4502 | * We have a moved dir. Add the old parent to check_dirs | |
4503 | */ | |
4504 | cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, | |
4505 | list); | |
ba5e8f2e | 4506 | ret = dup_ref(cur, &check_dirs); |
ccf1626b AB |
4507 | if (ret < 0) |
4508 | goto out; | |
31db9f7c AB |
4509 | } else if (!S_ISDIR(sctx->cur_inode_mode)) { |
4510 | /* | |
4511 | * We have a non dir inode. Go through all deleted refs and | |
4512 | * unlink them if they were not already overwritten by other | |
4513 | * inodes. | |
4514 | */ | |
4515 | list_for_each_entry(cur, &sctx->deleted_refs, list) { | |
4516 | ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen, | |
4517 | sctx->cur_ino, sctx->cur_inode_gen, | |
4518 | cur->name, cur->name_len); | |
4519 | if (ret < 0) | |
4520 | goto out; | |
4521 | if (!ret) { | |
fdb13889 FM |
4522 | /* |
4523 | * If we orphanized any ancestor before, we need | |
4524 | * to recompute the full path for deleted names, | |
4525 | * since any such path was computed before we | |
4526 | * processed any references and orphanized any | |
4527 | * ancestor inode. | |
4528 | */ | |
4529 | if (orphanized_ancestor) { | |
f5962781 FM |
4530 | ret = update_ref_path(sctx, cur); |
4531 | if (ret < 0) | |
fdb13889 | 4532 | goto out; |
fdb13889 | 4533 | } |
1f4692da AB |
4534 | ret = send_unlink(sctx, cur->full_path); |
4535 | if (ret < 0) | |
4536 | goto out; | |
31db9f7c | 4537 | } |
ba5e8f2e | 4538 | ret = dup_ref(cur, &check_dirs); |
31db9f7c AB |
4539 | if (ret < 0) |
4540 | goto out; | |
4541 | } | |
31db9f7c AB |
4542 | /* |
4543 | * If the inode is still orphan, unlink the orphan. This may | |
4544 | * happen when a previous inode did overwrite the first ref | |
4545 | * of this inode and no new refs were added for the current | |
766702ef AB |
4546 | * inode. Unlinking does not mean that the inode is deleted in |
4547 | * all cases. There may still be links to this inode in other | |
4548 | * places. | |
31db9f7c | 4549 | */ |
1f4692da | 4550 | if (is_orphan) { |
31db9f7c AB |
4551 | ret = send_unlink(sctx, valid_path); |
4552 | if (ret < 0) | |
4553 | goto out; | |
4554 | } | |
4555 | } | |
4556 | ||
4557 | /* | |
4558 | * We did collect all parent dirs where cur_inode was once located. We | |
4559 | * now go through all these dirs and check if they are pending for | |
4560 | * deletion and if it's finally possible to perform the rmdir now. | |
4561 | * We also update the inode stats of the parent dirs here. | |
4562 | */ | |
ba5e8f2e | 4563 | list_for_each_entry(cur, &check_dirs, list) { |
766702ef AB |
4564 | /* |
4565 | * In case we had refs into dirs that were not processed yet, | |
4566 | * we don't need to do the utime and rmdir logic for these dirs. | |
4567 | * The dir will be processed later. | |
4568 | */ | |
ba5e8f2e | 4569 | if (cur->dir > sctx->cur_ino) |
31db9f7c AB |
4570 | continue; |
4571 | ||
498581f3 | 4572 | ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL); |
31db9f7c AB |
4573 | if (ret < 0) |
4574 | goto out; | |
4575 | ||
4576 | if (ret == inode_state_did_create || | |
4577 | ret == inode_state_no_change) { | |
3e49363b | 4578 | ret = cache_dir_utimes(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4579 | if (ret < 0) |
4580 | goto out; | |
29d6d30f FM |
4581 | } else if (ret == inode_state_did_delete && |
4582 | cur->dir != last_dir_ino_rm) { | |
24970ccb | 4583 | ret = can_rmdir(sctx, cur->dir, cur->dir_gen); |
31db9f7c AB |
4584 | if (ret < 0) |
4585 | goto out; | |
4586 | if (ret) { | |
ba5e8f2e JB |
4587 | ret = get_cur_path(sctx, cur->dir, |
4588 | cur->dir_gen, valid_path); | |
31db9f7c AB |
4589 | if (ret < 0) |
4590 | goto out; | |
4591 | ret = send_rmdir(sctx, valid_path); | |
4592 | if (ret < 0) | |
4593 | goto out; | |
29d6d30f | 4594 | last_dir_ino_rm = cur->dir; |
31db9f7c AB |
4595 | } |
4596 | } | |
4597 | } | |
4598 | ||
31db9f7c AB |
4599 | ret = 0; |
4600 | ||
4601 | out: | |
ba5e8f2e | 4602 | __free_recorded_refs(&check_dirs); |
31db9f7c | 4603 | free_recorded_refs(sctx); |
924794c9 | 4604 | fs_path_free(valid_path); |
31db9f7c AB |
4605 | return ret; |
4606 | } | |
4607 | ||
3aa5bd36 BC |
4608 | static int rbtree_ref_comp(const void *k, const struct rb_node *node) |
4609 | { | |
4610 | const struct recorded_ref *data = k; | |
4611 | const struct recorded_ref *ref = rb_entry(node, struct recorded_ref, node); | |
4612 | int result; | |
4613 | ||
4614 | if (data->dir > ref->dir) | |
4615 | return 1; | |
4616 | if (data->dir < ref->dir) | |
4617 | return -1; | |
4618 | if (data->dir_gen > ref->dir_gen) | |
4619 | return 1; | |
4620 | if (data->dir_gen < ref->dir_gen) | |
4621 | return -1; | |
4622 | if (data->name_len > ref->name_len) | |
4623 | return 1; | |
4624 | if (data->name_len < ref->name_len) | |
4625 | return -1; | |
4626 | result = strcmp(data->name, ref->name); | |
4627 | if (result > 0) | |
4628 | return 1; | |
4629 | if (result < 0) | |
4630 | return -1; | |
4631 | return 0; | |
4632 | } | |
4633 | ||
4634 | static bool rbtree_ref_less(struct rb_node *node, const struct rb_node *parent) | |
4635 | { | |
4636 | const struct recorded_ref *entry = rb_entry(node, struct recorded_ref, node); | |
4637 | ||
4638 | return rbtree_ref_comp(entry, parent) < 0; | |
4639 | } | |
4640 | ||
4641 | static int record_ref_in_tree(struct rb_root *root, struct list_head *refs, | |
4642 | struct fs_path *name, u64 dir, u64 dir_gen, | |
4643 | struct send_ctx *sctx) | |
4644 | { | |
4645 | int ret = 0; | |
4646 | struct fs_path *path = NULL; | |
4647 | struct recorded_ref *ref = NULL; | |
4648 | ||
4649 | path = fs_path_alloc(); | |
4650 | if (!path) { | |
4651 | ret = -ENOMEM; | |
4652 | goto out; | |
4653 | } | |
4654 | ||
4655 | ref = recorded_ref_alloc(); | |
4656 | if (!ref) { | |
4657 | ret = -ENOMEM; | |
4658 | goto out; | |
4659 | } | |
4660 | ||
4661 | ret = get_cur_path(sctx, dir, dir_gen, path); | |
4662 | if (ret < 0) | |
4663 | goto out; | |
4664 | ret = fs_path_add_path(path, name); | |
4665 | if (ret < 0) | |
4666 | goto out; | |
4667 | ||
4668 | ref->dir = dir; | |
4669 | ref->dir_gen = dir_gen; | |
4670 | set_ref_path(ref, path); | |
4671 | list_add_tail(&ref->list, refs); | |
4672 | rb_add(&ref->node, root, rbtree_ref_less); | |
4673 | ref->root = root; | |
4674 | out: | |
4675 | if (ret) { | |
4676 | if (path && (!ref || !ref->full_path)) | |
4677 | fs_path_free(path); | |
4678 | recorded_ref_free(ref); | |
4679 | } | |
4680 | return ret; | |
4681 | } | |
4682 | ||
4683 | static int record_new_ref_if_needed(int num, u64 dir, int index, | |
4684 | struct fs_path *name, void *ctx) | |
4685 | { | |
4686 | int ret = 0; | |
4687 | struct send_ctx *sctx = ctx; | |
4688 | struct rb_node *node = NULL; | |
4689 | struct recorded_ref data; | |
4690 | struct recorded_ref *ref; | |
4691 | u64 dir_gen; | |
4692 | ||
7e93f6dc | 4693 | ret = get_inode_gen(sctx->send_root, dir, &dir_gen); |
3aa5bd36 BC |
4694 | if (ret < 0) |
4695 | goto out; | |
4696 | ||
4697 | data.dir = dir; | |
4698 | data.dir_gen = dir_gen; | |
4699 | set_ref_path(&data, name); | |
4700 | node = rb_find(&data, &sctx->rbtree_deleted_refs, rbtree_ref_comp); | |
4701 | if (node) { | |
4702 | ref = rb_entry(node, struct recorded_ref, node); | |
4703 | recorded_ref_free(ref); | |
4704 | } else { | |
4705 | ret = record_ref_in_tree(&sctx->rbtree_new_refs, | |
4706 | &sctx->new_refs, name, dir, dir_gen, | |
4707 | sctx); | |
4708 | } | |
4709 | out: | |
4710 | return ret; | |
4711 | } | |
4712 | ||
4713 | static int record_deleted_ref_if_needed(int num, u64 dir, int index, | |
4714 | struct fs_path *name, void *ctx) | |
4715 | { | |
4716 | int ret = 0; | |
4717 | struct send_ctx *sctx = ctx; | |
4718 | struct rb_node *node = NULL; | |
4719 | struct recorded_ref data; | |
4720 | struct recorded_ref *ref; | |
4721 | u64 dir_gen; | |
4722 | ||
7e93f6dc | 4723 | ret = get_inode_gen(sctx->parent_root, dir, &dir_gen); |
3aa5bd36 BC |
4724 | if (ret < 0) |
4725 | goto out; | |
4726 | ||
4727 | data.dir = dir; | |
4728 | data.dir_gen = dir_gen; | |
4729 | set_ref_path(&data, name); | |
4730 | node = rb_find(&data, &sctx->rbtree_new_refs, rbtree_ref_comp); | |
4731 | if (node) { | |
4732 | ref = rb_entry(node, struct recorded_ref, node); | |
4733 | recorded_ref_free(ref); | |
4734 | } else { | |
4735 | ret = record_ref_in_tree(&sctx->rbtree_deleted_refs, | |
4736 | &sctx->deleted_refs, name, dir, | |
4737 | dir_gen, sctx); | |
4738 | } | |
4739 | out: | |
4740 | return ret; | |
4741 | } | |
4742 | ||
31db9f7c AB |
4743 | static int record_new_ref(struct send_ctx *sctx) |
4744 | { | |
4745 | int ret; | |
4746 | ||
924794c9 | 4747 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
3aa5bd36 | 4748 | sctx->cmp_key, 0, record_new_ref_if_needed, sctx); |
31db9f7c AB |
4749 | if (ret < 0) |
4750 | goto out; | |
4751 | ret = 0; | |
4752 | ||
4753 | out: | |
4754 | return ret; | |
4755 | } | |
4756 | ||
4757 | static int record_deleted_ref(struct send_ctx *sctx) | |
4758 | { | |
4759 | int ret; | |
4760 | ||
924794c9 | 4761 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
3aa5bd36 BC |
4762 | sctx->cmp_key, 0, record_deleted_ref_if_needed, |
4763 | sctx); | |
31db9f7c AB |
4764 | if (ret < 0) |
4765 | goto out; | |
4766 | ret = 0; | |
4767 | ||
4768 | out: | |
4769 | return ret; | |
4770 | } | |
4771 | ||
31db9f7c AB |
4772 | static int record_changed_ref(struct send_ctx *sctx) |
4773 | { | |
4774 | int ret = 0; | |
4775 | ||
924794c9 | 4776 | ret = iterate_inode_ref(sctx->send_root, sctx->left_path, |
0d8869fb | 4777 | sctx->cmp_key, 0, record_new_ref_if_needed, sctx); |
31db9f7c AB |
4778 | if (ret < 0) |
4779 | goto out; | |
924794c9 | 4780 | ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, |
0d8869fb | 4781 | sctx->cmp_key, 0, record_deleted_ref_if_needed, sctx); |
31db9f7c AB |
4782 | if (ret < 0) |
4783 | goto out; | |
4784 | ret = 0; | |
4785 | ||
4786 | out: | |
4787 | return ret; | |
4788 | } | |
4789 | ||
4790 | /* | |
4791 | * Record and process all refs at once. Needed when an inode changes the | |
4792 | * generation number, which means that it was deleted and recreated. | |
4793 | */ | |
4794 | static int process_all_refs(struct send_ctx *sctx, | |
4795 | enum btrfs_compare_tree_result cmd) | |
4796 | { | |
649b9635 GN |
4797 | int ret = 0; |
4798 | int iter_ret = 0; | |
31db9f7c AB |
4799 | struct btrfs_root *root; |
4800 | struct btrfs_path *path; | |
4801 | struct btrfs_key key; | |
4802 | struct btrfs_key found_key; | |
31db9f7c | 4803 | iterate_inode_ref_t cb; |
9f03740a | 4804 | int pending_move = 0; |
31db9f7c AB |
4805 | |
4806 | path = alloc_path_for_send(); | |
4807 | if (!path) | |
4808 | return -ENOMEM; | |
4809 | ||
4810 | if (cmd == BTRFS_COMPARE_TREE_NEW) { | |
4811 | root = sctx->send_root; | |
0d8869fb | 4812 | cb = record_new_ref_if_needed; |
31db9f7c AB |
4813 | } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { |
4814 | root = sctx->parent_root; | |
0d8869fb | 4815 | cb = record_deleted_ref_if_needed; |
31db9f7c | 4816 | } else { |
4d1a63b2 DS |
4817 | btrfs_err(sctx->send_root->fs_info, |
4818 | "Wrong command %d in process_all_refs", cmd); | |
4819 | ret = -EINVAL; | |
4820 | goto out; | |
31db9f7c AB |
4821 | } |
4822 | ||
4823 | key.objectid = sctx->cmp_key->objectid; | |
4824 | key.type = BTRFS_INODE_REF_KEY; | |
4825 | key.offset = 0; | |
649b9635 | 4826 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
31db9f7c | 4827 | if (found_key.objectid != key.objectid || |
96b5bd77 JS |
4828 | (found_key.type != BTRFS_INODE_REF_KEY && |
4829 | found_key.type != BTRFS_INODE_EXTREF_KEY)) | |
31db9f7c | 4830 | break; |
31db9f7c | 4831 | |
924794c9 | 4832 | ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); |
31db9f7c AB |
4833 | if (ret < 0) |
4834 | goto out; | |
649b9635 GN |
4835 | } |
4836 | /* Catch error found during iteration */ | |
4837 | if (iter_ret < 0) { | |
4838 | ret = iter_ret; | |
4839 | goto out; | |
31db9f7c | 4840 | } |
e938c8ad | 4841 | btrfs_release_path(path); |
31db9f7c | 4842 | |
3dc09ec8 JB |
4843 | /* |
4844 | * We don't actually care about pending_move as we are simply | |
4845 | * re-creating this inode and will be rename'ing it into place once we | |
4846 | * rename the parent directory. | |
4847 | */ | |
9f03740a | 4848 | ret = process_recorded_refs(sctx, &pending_move); |
31db9f7c AB |
4849 | out: |
4850 | btrfs_free_path(path); | |
4851 | return ret; | |
4852 | } | |
4853 | ||
4854 | static int send_set_xattr(struct send_ctx *sctx, | |
4855 | struct fs_path *path, | |
4856 | const char *name, int name_len, | |
4857 | const char *data, int data_len) | |
4858 | { | |
4859 | int ret = 0; | |
4860 | ||
4861 | ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); | |
4862 | if (ret < 0) | |
4863 | goto out; | |
4864 | ||
4865 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4866 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4867 | TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len); | |
4868 | ||
4869 | ret = send_cmd(sctx); | |
4870 | ||
4871 | tlv_put_failure: | |
4872 | out: | |
4873 | return ret; | |
4874 | } | |
4875 | ||
4876 | static int send_remove_xattr(struct send_ctx *sctx, | |
4877 | struct fs_path *path, | |
4878 | const char *name, int name_len) | |
4879 | { | |
4880 | int ret = 0; | |
4881 | ||
4882 | ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); | |
4883 | if (ret < 0) | |
4884 | goto out; | |
4885 | ||
4886 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
4887 | TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); | |
4888 | ||
4889 | ret = send_cmd(sctx); | |
4890 | ||
4891 | tlv_put_failure: | |
4892 | out: | |
4893 | return ret; | |
4894 | } | |
4895 | ||
4896 | static int __process_new_xattr(int num, struct btrfs_key *di_key, | |
b1dea4e7 OS |
4897 | const char *name, int name_len, const char *data, |
4898 | int data_len, void *ctx) | |
31db9f7c AB |
4899 | { |
4900 | int ret; | |
4901 | struct send_ctx *sctx = ctx; | |
4902 | struct fs_path *p; | |
2211d5ba | 4903 | struct posix_acl_xattr_header dummy_acl; |
31db9f7c | 4904 | |
89efda52 MPS |
4905 | /* Capabilities are emitted by finish_inode_if_needed */ |
4906 | if (!strncmp(name, XATTR_NAME_CAPS, name_len)) | |
4907 | return 0; | |
4908 | ||
924794c9 | 4909 | p = fs_path_alloc(); |
31db9f7c AB |
4910 | if (!p) |
4911 | return -ENOMEM; | |
4912 | ||
4913 | /* | |
01327610 | 4914 | * This hack is needed because empty acls are stored as zero byte |
31db9f7c | 4915 | * data in xattrs. Problem with that is, that receiving these zero byte |
01327610 | 4916 | * acls will fail later. To fix this, we send a dummy acl list that |
31db9f7c AB |
4917 | * only contains the version number and no entries. |
4918 | */ | |
4919 | if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) || | |
4920 | !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) { | |
4921 | if (data_len == 0) { | |
4922 | dummy_acl.a_version = | |
4923 | cpu_to_le32(POSIX_ACL_XATTR_VERSION); | |
4924 | data = (char *)&dummy_acl; | |
4925 | data_len = sizeof(dummy_acl); | |
4926 | } | |
4927 | } | |
4928 | ||
4929 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4930 | if (ret < 0) | |
4931 | goto out; | |
4932 | ||
4933 | ret = send_set_xattr(sctx, p, name, name_len, data, data_len); | |
4934 | ||
4935 | out: | |
924794c9 | 4936 | fs_path_free(p); |
31db9f7c AB |
4937 | return ret; |
4938 | } | |
4939 | ||
4940 | static int __process_deleted_xattr(int num, struct btrfs_key *di_key, | |
4941 | const char *name, int name_len, | |
b1dea4e7 | 4942 | const char *data, int data_len, void *ctx) |
31db9f7c AB |
4943 | { |
4944 | int ret; | |
4945 | struct send_ctx *sctx = ctx; | |
4946 | struct fs_path *p; | |
4947 | ||
924794c9 | 4948 | p = fs_path_alloc(); |
31db9f7c AB |
4949 | if (!p) |
4950 | return -ENOMEM; | |
4951 | ||
4952 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
4953 | if (ret < 0) | |
4954 | goto out; | |
4955 | ||
4956 | ret = send_remove_xattr(sctx, p, name, name_len); | |
4957 | ||
4958 | out: | |
924794c9 | 4959 | fs_path_free(p); |
31db9f7c AB |
4960 | return ret; |
4961 | } | |
4962 | ||
4963 | static int process_new_xattr(struct send_ctx *sctx) | |
4964 | { | |
4965 | int ret = 0; | |
4966 | ||
924794c9 | 4967 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 4968 | __process_new_xattr, sctx); |
31db9f7c AB |
4969 | |
4970 | return ret; | |
4971 | } | |
4972 | ||
4973 | static int process_deleted_xattr(struct send_ctx *sctx) | |
4974 | { | |
e2c89907 | 4975 | return iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 4976 | __process_deleted_xattr, sctx); |
31db9f7c AB |
4977 | } |
4978 | ||
4979 | struct find_xattr_ctx { | |
4980 | const char *name; | |
4981 | int name_len; | |
4982 | int found_idx; | |
4983 | char *found_data; | |
4984 | int found_data_len; | |
4985 | }; | |
4986 | ||
b1dea4e7 OS |
4987 | static int __find_xattr(int num, struct btrfs_key *di_key, const char *name, |
4988 | int name_len, const char *data, int data_len, void *vctx) | |
31db9f7c AB |
4989 | { |
4990 | struct find_xattr_ctx *ctx = vctx; | |
4991 | ||
4992 | if (name_len == ctx->name_len && | |
4993 | strncmp(name, ctx->name, name_len) == 0) { | |
4994 | ctx->found_idx = num; | |
4995 | ctx->found_data_len = data_len; | |
e780b0d1 | 4996 | ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); |
31db9f7c AB |
4997 | if (!ctx->found_data) |
4998 | return -ENOMEM; | |
31db9f7c AB |
4999 | return 1; |
5000 | } | |
5001 | return 0; | |
5002 | } | |
5003 | ||
924794c9 | 5004 | static int find_xattr(struct btrfs_root *root, |
31db9f7c AB |
5005 | struct btrfs_path *path, |
5006 | struct btrfs_key *key, | |
5007 | const char *name, int name_len, | |
5008 | char **data, int *data_len) | |
5009 | { | |
5010 | int ret; | |
5011 | struct find_xattr_ctx ctx; | |
5012 | ||
5013 | ctx.name = name; | |
5014 | ctx.name_len = name_len; | |
5015 | ctx.found_idx = -1; | |
5016 | ctx.found_data = NULL; | |
5017 | ctx.found_data_len = 0; | |
5018 | ||
a0357511 | 5019 | ret = iterate_dir_item(root, path, __find_xattr, &ctx); |
31db9f7c AB |
5020 | if (ret < 0) |
5021 | return ret; | |
5022 | ||
5023 | if (ctx.found_idx == -1) | |
5024 | return -ENOENT; | |
5025 | if (data) { | |
5026 | *data = ctx.found_data; | |
5027 | *data_len = ctx.found_data_len; | |
5028 | } else { | |
5029 | kfree(ctx.found_data); | |
5030 | } | |
5031 | return ctx.found_idx; | |
5032 | } | |
5033 | ||
5034 | ||
5035 | static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, | |
5036 | const char *name, int name_len, | |
5037 | const char *data, int data_len, | |
b1dea4e7 | 5038 | void *ctx) |
31db9f7c AB |
5039 | { |
5040 | int ret; | |
5041 | struct send_ctx *sctx = ctx; | |
5042 | char *found_data = NULL; | |
5043 | int found_data_len = 0; | |
31db9f7c | 5044 | |
924794c9 TI |
5045 | ret = find_xattr(sctx->parent_root, sctx->right_path, |
5046 | sctx->cmp_key, name, name_len, &found_data, | |
5047 | &found_data_len); | |
31db9f7c AB |
5048 | if (ret == -ENOENT) { |
5049 | ret = __process_new_xattr(num, di_key, name, name_len, data, | |
b1dea4e7 | 5050 | data_len, ctx); |
31db9f7c AB |
5051 | } else if (ret >= 0) { |
5052 | if (data_len != found_data_len || | |
5053 | memcmp(data, found_data, data_len)) { | |
5054 | ret = __process_new_xattr(num, di_key, name, name_len, | |
b1dea4e7 | 5055 | data, data_len, ctx); |
31db9f7c AB |
5056 | } else { |
5057 | ret = 0; | |
5058 | } | |
5059 | } | |
5060 | ||
5061 | kfree(found_data); | |
31db9f7c AB |
5062 | return ret; |
5063 | } | |
5064 | ||
5065 | static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, | |
5066 | const char *name, int name_len, | |
5067 | const char *data, int data_len, | |
b1dea4e7 | 5068 | void *ctx) |
31db9f7c AB |
5069 | { |
5070 | int ret; | |
5071 | struct send_ctx *sctx = ctx; | |
5072 | ||
924794c9 TI |
5073 | ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key, |
5074 | name, name_len, NULL, NULL); | |
31db9f7c AB |
5075 | if (ret == -ENOENT) |
5076 | ret = __process_deleted_xattr(num, di_key, name, name_len, data, | |
b1dea4e7 | 5077 | data_len, ctx); |
31db9f7c AB |
5078 | else if (ret >= 0) |
5079 | ret = 0; | |
5080 | ||
5081 | return ret; | |
5082 | } | |
5083 | ||
5084 | static int process_changed_xattr(struct send_ctx *sctx) | |
5085 | { | |
5086 | int ret = 0; | |
5087 | ||
924794c9 | 5088 | ret = iterate_dir_item(sctx->send_root, sctx->left_path, |
a0357511 | 5089 | __process_changed_new_xattr, sctx); |
31db9f7c AB |
5090 | if (ret < 0) |
5091 | goto out; | |
924794c9 | 5092 | ret = iterate_dir_item(sctx->parent_root, sctx->right_path, |
a0357511 | 5093 | __process_changed_deleted_xattr, sctx); |
31db9f7c AB |
5094 | |
5095 | out: | |
5096 | return ret; | |
5097 | } | |
5098 | ||
5099 | static int process_all_new_xattrs(struct send_ctx *sctx) | |
5100 | { | |
69e43177 GN |
5101 | int ret = 0; |
5102 | int iter_ret = 0; | |
31db9f7c AB |
5103 | struct btrfs_root *root; |
5104 | struct btrfs_path *path; | |
5105 | struct btrfs_key key; | |
5106 | struct btrfs_key found_key; | |
31db9f7c AB |
5107 | |
5108 | path = alloc_path_for_send(); | |
5109 | if (!path) | |
5110 | return -ENOMEM; | |
5111 | ||
5112 | root = sctx->send_root; | |
5113 | ||
5114 | key.objectid = sctx->cmp_key->objectid; | |
5115 | key.type = BTRFS_XATTR_ITEM_KEY; | |
5116 | key.offset = 0; | |
69e43177 | 5117 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
31db9f7c AB |
5118 | if (found_key.objectid != key.objectid || |
5119 | found_key.type != key.type) { | |
5120 | ret = 0; | |
69e43177 | 5121 | break; |
31db9f7c AB |
5122 | } |
5123 | ||
a0357511 | 5124 | ret = iterate_dir_item(root, path, __process_new_xattr, sctx); |
31db9f7c | 5125 | if (ret < 0) |
69e43177 | 5126 | break; |
31db9f7c | 5127 | } |
69e43177 GN |
5128 | /* Catch error found during iteration */ |
5129 | if (iter_ret < 0) | |
5130 | ret = iter_ret; | |
31db9f7c | 5131 | |
31db9f7c AB |
5132 | btrfs_free_path(path); |
5133 | return ret; | |
5134 | } | |
5135 | ||
38622010 BB |
5136 | static int send_verity(struct send_ctx *sctx, struct fs_path *path, |
5137 | struct fsverity_descriptor *desc) | |
5138 | { | |
5139 | int ret; | |
5140 | ||
5141 | ret = begin_cmd(sctx, BTRFS_SEND_C_ENABLE_VERITY); | |
5142 | if (ret < 0) | |
5143 | goto out; | |
5144 | ||
5145 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); | |
5146 | TLV_PUT_U8(sctx, BTRFS_SEND_A_VERITY_ALGORITHM, | |
5147 | le8_to_cpu(desc->hash_algorithm)); | |
5148 | TLV_PUT_U32(sctx, BTRFS_SEND_A_VERITY_BLOCK_SIZE, | |
5149 | 1U << le8_to_cpu(desc->log_blocksize)); | |
5150 | TLV_PUT(sctx, BTRFS_SEND_A_VERITY_SALT_DATA, desc->salt, | |
5151 | le8_to_cpu(desc->salt_size)); | |
5152 | TLV_PUT(sctx, BTRFS_SEND_A_VERITY_SIG_DATA, desc->signature, | |
5153 | le32_to_cpu(desc->sig_size)); | |
5154 | ||
5155 | ret = send_cmd(sctx); | |
5156 | ||
5157 | tlv_put_failure: | |
5158 | out: | |
5159 | return ret; | |
5160 | } | |
5161 | ||
5162 | static int process_verity(struct send_ctx *sctx) | |
5163 | { | |
5164 | int ret = 0; | |
5165 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; | |
5166 | struct inode *inode; | |
5167 | struct fs_path *p; | |
5168 | ||
5169 | inode = btrfs_iget(fs_info->sb, sctx->cur_ino, sctx->send_root); | |
5170 | if (IS_ERR(inode)) | |
5171 | return PTR_ERR(inode); | |
5172 | ||
5173 | ret = btrfs_get_verity_descriptor(inode, NULL, 0); | |
5174 | if (ret < 0) | |
5175 | goto iput; | |
5176 | ||
5177 | if (ret > FS_VERITY_MAX_DESCRIPTOR_SIZE) { | |
5178 | ret = -EMSGSIZE; | |
5179 | goto iput; | |
5180 | } | |
5181 | if (!sctx->verity_descriptor) { | |
5182 | sctx->verity_descriptor = kvmalloc(FS_VERITY_MAX_DESCRIPTOR_SIZE, | |
5183 | GFP_KERNEL); | |
5184 | if (!sctx->verity_descriptor) { | |
5185 | ret = -ENOMEM; | |
5186 | goto iput; | |
5187 | } | |
5188 | } | |
5189 | ||
5190 | ret = btrfs_get_verity_descriptor(inode, sctx->verity_descriptor, ret); | |
5191 | if (ret < 0) | |
5192 | goto iput; | |
5193 | ||
5194 | p = fs_path_alloc(); | |
5195 | if (!p) { | |
5196 | ret = -ENOMEM; | |
5197 | goto iput; | |
5198 | } | |
5199 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
5200 | if (ret < 0) | |
5201 | goto free_path; | |
5202 | ||
5203 | ret = send_verity(sctx, p, sctx->verity_descriptor); | |
5204 | if (ret < 0) | |
5205 | goto free_path; | |
5206 | ||
5207 | free_path: | |
5208 | fs_path_free(p); | |
5209 | iput: | |
5210 | iput(inode); | |
5211 | return ret; | |
5212 | } | |
5213 | ||
8c7d9fe0 OS |
5214 | static inline u64 max_send_read_size(const struct send_ctx *sctx) |
5215 | { | |
5216 | return sctx->send_max_size - SZ_16K; | |
5217 | } | |
5218 | ||
5219 | static int put_data_header(struct send_ctx *sctx, u32 len) | |
5220 | { | |
356bbbb6 OS |
5221 | if (WARN_ON_ONCE(sctx->put_data)) |
5222 | return -EINVAL; | |
5223 | sctx->put_data = true; | |
5224 | if (sctx->proto >= 2) { | |
5225 | /* | |
5226 | * Since v2, the data attribute header doesn't include a length, | |
5227 | * it is implicitly to the end of the command. | |
5228 | */ | |
5229 | if (sctx->send_max_size - sctx->send_size < sizeof(__le16) + len) | |
5230 | return -EOVERFLOW; | |
5231 | put_unaligned_le16(BTRFS_SEND_A_DATA, sctx->send_buf + sctx->send_size); | |
5232 | sctx->send_size += sizeof(__le16); | |
5233 | } else { | |
5234 | struct btrfs_tlv_header *hdr; | |
8c7d9fe0 | 5235 | |
356bbbb6 OS |
5236 | if (sctx->send_max_size - sctx->send_size < sizeof(*hdr) + len) |
5237 | return -EOVERFLOW; | |
5238 | hdr = (struct btrfs_tlv_header *)(sctx->send_buf + sctx->send_size); | |
5239 | put_unaligned_le16(BTRFS_SEND_A_DATA, &hdr->tlv_type); | |
5240 | put_unaligned_le16(len, &hdr->tlv_len); | |
5241 | sctx->send_size += sizeof(*hdr); | |
5242 | } | |
8c7d9fe0 OS |
5243 | return 0; |
5244 | } | |
5245 | ||
5246 | static int put_file_data(struct send_ctx *sctx, u64 offset, u32 len) | |
ed259095 JB |
5247 | { |
5248 | struct btrfs_root *root = sctx->send_root; | |
5249 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ed259095 | 5250 | struct page *page; |
09cbfeaf | 5251 | pgoff_t index = offset >> PAGE_SHIFT; |
ed259095 | 5252 | pgoff_t last_index; |
7073017a | 5253 | unsigned pg_offset = offset_in_page(offset); |
8c7d9fe0 OS |
5254 | int ret; |
5255 | ||
5256 | ret = put_data_header(sctx, len); | |
5257 | if (ret) | |
5258 | return ret; | |
ed259095 | 5259 | |
09cbfeaf | 5260 | last_index = (offset + len - 1) >> PAGE_SHIFT; |
2131bcd3 | 5261 | |
ed259095 JB |
5262 | while (index <= last_index) { |
5263 | unsigned cur_len = min_t(unsigned, len, | |
09cbfeaf | 5264 | PAGE_SIZE - pg_offset); |
eef16ba2 | 5265 | |
521b6803 | 5266 | page = find_lock_page(sctx->cur_inode->i_mapping, index); |
ed259095 | 5267 | if (!page) { |
521b6803 FM |
5268 | page_cache_sync_readahead(sctx->cur_inode->i_mapping, |
5269 | &sctx->ra, NULL, index, | |
5270 | last_index + 1 - index); | |
eef16ba2 | 5271 | |
521b6803 FM |
5272 | page = find_or_create_page(sctx->cur_inode->i_mapping, |
5273 | index, GFP_KERNEL); | |
eef16ba2 KH |
5274 | if (!page) { |
5275 | ret = -ENOMEM; | |
5276 | break; | |
5277 | } | |
5278 | } | |
5279 | ||
521b6803 FM |
5280 | if (PageReadahead(page)) |
5281 | page_cache_async_readahead(sctx->cur_inode->i_mapping, | |
fdaf9a58 LT |
5282 | &sctx->ra, NULL, page_folio(page), |
5283 | index, last_index + 1 - index); | |
ed259095 JB |
5284 | |
5285 | if (!PageUptodate(page)) { | |
fb12489b | 5286 | btrfs_read_folio(NULL, page_folio(page)); |
ed259095 JB |
5287 | lock_page(page); |
5288 | if (!PageUptodate(page)) { | |
5289 | unlock_page(page); | |
2e7be9db DM |
5290 | btrfs_err(fs_info, |
5291 | "send: IO error at offset %llu for inode %llu root %llu", | |
5292 | page_offset(page), sctx->cur_ino, | |
5293 | sctx->send_root->root_key.objectid); | |
09cbfeaf | 5294 | put_page(page); |
ed259095 JB |
5295 | ret = -EIO; |
5296 | break; | |
5297 | } | |
5298 | } | |
5299 | ||
3590ec58 IW |
5300 | memcpy_from_page(sctx->send_buf + sctx->send_size, page, |
5301 | pg_offset, cur_len); | |
ed259095 | 5302 | unlock_page(page); |
09cbfeaf | 5303 | put_page(page); |
ed259095 JB |
5304 | index++; |
5305 | pg_offset = 0; | |
5306 | len -= cur_len; | |
8c7d9fe0 | 5307 | sctx->send_size += cur_len; |
ed259095 | 5308 | } |
521b6803 | 5309 | |
ed259095 JB |
5310 | return ret; |
5311 | } | |
5312 | ||
31db9f7c AB |
5313 | /* |
5314 | * Read some bytes from the current inode/file and send a write command to | |
5315 | * user space. | |
5316 | */ | |
5317 | static int send_write(struct send_ctx *sctx, u64 offset, u32 len) | |
5318 | { | |
04ab956e | 5319 | struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; |
31db9f7c AB |
5320 | int ret = 0; |
5321 | struct fs_path *p; | |
31db9f7c | 5322 | |
924794c9 | 5323 | p = fs_path_alloc(); |
31db9f7c AB |
5324 | if (!p) |
5325 | return -ENOMEM; | |
5326 | ||
04ab956e | 5327 | btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len); |
31db9f7c | 5328 | |
31db9f7c AB |
5329 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); |
5330 | if (ret < 0) | |
5331 | goto out; | |
5332 | ||
5333 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
5334 | if (ret < 0) | |
5335 | goto out; | |
5336 | ||
5337 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5338 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
8c7d9fe0 OS |
5339 | ret = put_file_data(sctx, offset, len); |
5340 | if (ret < 0) | |
5341 | goto out; | |
31db9f7c AB |
5342 | |
5343 | ret = send_cmd(sctx); | |
5344 | ||
5345 | tlv_put_failure: | |
5346 | out: | |
924794c9 | 5347 | fs_path_free(p); |
a9b2e0de | 5348 | return ret; |
31db9f7c AB |
5349 | } |
5350 | ||
5351 | /* | |
5352 | * Send a clone command to user space. | |
5353 | */ | |
5354 | static int send_clone(struct send_ctx *sctx, | |
5355 | u64 offset, u32 len, | |
5356 | struct clone_root *clone_root) | |
5357 | { | |
5358 | int ret = 0; | |
31db9f7c AB |
5359 | struct fs_path *p; |
5360 | u64 gen; | |
5361 | ||
04ab956e JM |
5362 | btrfs_debug(sctx->send_root->fs_info, |
5363 | "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu", | |
4fd786e6 MT |
5364 | offset, len, clone_root->root->root_key.objectid, |
5365 | clone_root->ino, clone_root->offset); | |
31db9f7c | 5366 | |
924794c9 | 5367 | p = fs_path_alloc(); |
31db9f7c AB |
5368 | if (!p) |
5369 | return -ENOMEM; | |
5370 | ||
5371 | ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE); | |
5372 | if (ret < 0) | |
5373 | goto out; | |
5374 | ||
5375 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
5376 | if (ret < 0) | |
5377 | goto out; | |
5378 | ||
5379 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5380 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); | |
5381 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5382 | ||
e938c8ad | 5383 | if (clone_root->root == sctx->send_root) { |
7e93f6dc | 5384 | ret = get_inode_gen(sctx->send_root, clone_root->ino, &gen); |
31db9f7c AB |
5385 | if (ret < 0) |
5386 | goto out; | |
5387 | ret = get_cur_path(sctx, clone_root->ino, gen, p); | |
5388 | } else { | |
924794c9 | 5389 | ret = get_inode_path(clone_root->root, clone_root->ino, p); |
31db9f7c AB |
5390 | } |
5391 | if (ret < 0) | |
5392 | goto out; | |
5393 | ||
37b8d27d JB |
5394 | /* |
5395 | * If the parent we're using has a received_uuid set then use that as | |
5396 | * our clone source as that is what we will look for when doing a | |
5397 | * receive. | |
5398 | * | |
5399 | * This covers the case that we create a snapshot off of a received | |
5400 | * subvolume and then use that as the parent and try to receive on a | |
5401 | * different host. | |
5402 | */ | |
5403 | if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid)) | |
5404 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
5405 | clone_root->root->root_item.received_uuid); | |
5406 | else | |
5407 | TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, | |
5408 | clone_root->root->root_item.uuid); | |
31db9f7c | 5409 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, |
09e3a288 | 5410 | btrfs_root_ctransid(&clone_root->root->root_item)); |
31db9f7c AB |
5411 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); |
5412 | TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, | |
5413 | clone_root->offset); | |
5414 | ||
5415 | ret = send_cmd(sctx); | |
5416 | ||
5417 | tlv_put_failure: | |
5418 | out: | |
924794c9 | 5419 | fs_path_free(p); |
31db9f7c AB |
5420 | return ret; |
5421 | } | |
5422 | ||
cb95e7bf MF |
5423 | /* |
5424 | * Send an update extent command to user space. | |
5425 | */ | |
5426 | static int send_update_extent(struct send_ctx *sctx, | |
5427 | u64 offset, u32 len) | |
5428 | { | |
5429 | int ret = 0; | |
5430 | struct fs_path *p; | |
5431 | ||
924794c9 | 5432 | p = fs_path_alloc(); |
cb95e7bf MF |
5433 | if (!p) |
5434 | return -ENOMEM; | |
5435 | ||
5436 | ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); | |
5437 | if (ret < 0) | |
5438 | goto out; | |
5439 | ||
5440 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); | |
5441 | if (ret < 0) | |
5442 | goto out; | |
5443 | ||
5444 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5445 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5446 | TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); | |
5447 | ||
5448 | ret = send_cmd(sctx); | |
5449 | ||
5450 | tlv_put_failure: | |
5451 | out: | |
924794c9 | 5452 | fs_path_free(p); |
cb95e7bf MF |
5453 | return ret; |
5454 | } | |
5455 | ||
16e7549f JB |
5456 | static int send_hole(struct send_ctx *sctx, u64 end) |
5457 | { | |
5458 | struct fs_path *p = NULL; | |
8c7d9fe0 | 5459 | u64 read_size = max_send_read_size(sctx); |
16e7549f | 5460 | u64 offset = sctx->cur_inode_last_extent; |
16e7549f JB |
5461 | int ret = 0; |
5462 | ||
22d3151c FM |
5463 | /* |
5464 | * A hole that starts at EOF or beyond it. Since we do not yet support | |
5465 | * fallocate (for extent preallocation and hole punching), sending a | |
5466 | * write of zeroes starting at EOF or beyond would later require issuing | |
5467 | * a truncate operation which would undo the write and achieve nothing. | |
5468 | */ | |
5469 | if (offset >= sctx->cur_inode_size) | |
5470 | return 0; | |
5471 | ||
6b1f72e5 FM |
5472 | /* |
5473 | * Don't go beyond the inode's i_size due to prealloc extents that start | |
5474 | * after the i_size. | |
5475 | */ | |
5476 | end = min_t(u64, end, sctx->cur_inode_size); | |
5477 | ||
d4dfc0f4 FM |
5478 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) |
5479 | return send_update_extent(sctx, offset, end - offset); | |
5480 | ||
16e7549f JB |
5481 | p = fs_path_alloc(); |
5482 | if (!p) | |
5483 | return -ENOMEM; | |
c715e155 FM |
5484 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); |
5485 | if (ret < 0) | |
5486 | goto tlv_put_failure; | |
16e7549f | 5487 | while (offset < end) { |
8c7d9fe0 | 5488 | u64 len = min(end - offset, read_size); |
16e7549f JB |
5489 | |
5490 | ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); | |
16e7549f JB |
5491 | if (ret < 0) |
5492 | break; | |
5493 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); | |
5494 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
8c7d9fe0 OS |
5495 | ret = put_data_header(sctx, len); |
5496 | if (ret < 0) | |
5497 | break; | |
5498 | memset(sctx->send_buf + sctx->send_size, 0, len); | |
5499 | sctx->send_size += len; | |
16e7549f JB |
5500 | ret = send_cmd(sctx); |
5501 | if (ret < 0) | |
5502 | break; | |
5503 | offset += len; | |
5504 | } | |
ffa7c429 | 5505 | sctx->cur_inode_next_write_offset = offset; |
16e7549f JB |
5506 | tlv_put_failure: |
5507 | fs_path_free(p); | |
5508 | return ret; | |
5509 | } | |
5510 | ||
3ea4dc5b OS |
5511 | static int send_encoded_inline_extent(struct send_ctx *sctx, |
5512 | struct btrfs_path *path, u64 offset, | |
5513 | u64 len) | |
5514 | { | |
5515 | struct btrfs_root *root = sctx->send_root; | |
5516 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5517 | struct inode *inode; | |
5518 | struct fs_path *fspath; | |
5519 | struct extent_buffer *leaf = path->nodes[0]; | |
5520 | struct btrfs_key key; | |
5521 | struct btrfs_file_extent_item *ei; | |
5522 | u64 ram_bytes; | |
5523 | size_t inline_size; | |
5524 | int ret; | |
5525 | ||
5526 | inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root); | |
5527 | if (IS_ERR(inode)) | |
5528 | return PTR_ERR(inode); | |
5529 | ||
5530 | fspath = fs_path_alloc(); | |
5531 | if (!fspath) { | |
5532 | ret = -ENOMEM; | |
5533 | goto out; | |
5534 | } | |
5535 | ||
5536 | ret = begin_cmd(sctx, BTRFS_SEND_C_ENCODED_WRITE); | |
5537 | if (ret < 0) | |
5538 | goto out; | |
5539 | ||
5540 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath); | |
5541 | if (ret < 0) | |
5542 | goto out; | |
5543 | ||
5544 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
5545 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
5546 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, ei); | |
5547 | inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]); | |
5548 | ||
5549 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, fspath); | |
5550 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5551 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_FILE_LEN, | |
5552 | min(key.offset + ram_bytes - offset, len)); | |
5553 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_LEN, ram_bytes); | |
5554 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_OFFSET, offset - key.offset); | |
5555 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
5556 | btrfs_file_extent_compression(leaf, ei)); | |
5557 | if (ret < 0) | |
5558 | goto out; | |
5559 | TLV_PUT_U32(sctx, BTRFS_SEND_A_COMPRESSION, ret); | |
5560 | ||
5561 | ret = put_data_header(sctx, inline_size); | |
5562 | if (ret < 0) | |
5563 | goto out; | |
5564 | read_extent_buffer(leaf, sctx->send_buf + sctx->send_size, | |
5565 | btrfs_file_extent_inline_start(ei), inline_size); | |
5566 | sctx->send_size += inline_size; | |
5567 | ||
5568 | ret = send_cmd(sctx); | |
5569 | ||
5570 | tlv_put_failure: | |
5571 | out: | |
5572 | fs_path_free(fspath); | |
5573 | iput(inode); | |
5574 | return ret; | |
5575 | } | |
5576 | ||
5577 | static int send_encoded_extent(struct send_ctx *sctx, struct btrfs_path *path, | |
5578 | u64 offset, u64 len) | |
5579 | { | |
5580 | struct btrfs_root *root = sctx->send_root; | |
5581 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5582 | struct inode *inode; | |
5583 | struct fs_path *fspath; | |
5584 | struct extent_buffer *leaf = path->nodes[0]; | |
5585 | struct btrfs_key key; | |
5586 | struct btrfs_file_extent_item *ei; | |
5587 | u64 disk_bytenr, disk_num_bytes; | |
5588 | u32 data_offset; | |
5589 | struct btrfs_cmd_header *hdr; | |
5590 | u32 crc; | |
5591 | int ret; | |
5592 | ||
5593 | inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root); | |
5594 | if (IS_ERR(inode)) | |
5595 | return PTR_ERR(inode); | |
5596 | ||
5597 | fspath = fs_path_alloc(); | |
5598 | if (!fspath) { | |
5599 | ret = -ENOMEM; | |
5600 | goto out; | |
5601 | } | |
5602 | ||
5603 | ret = begin_cmd(sctx, BTRFS_SEND_C_ENCODED_WRITE); | |
5604 | if (ret < 0) | |
5605 | goto out; | |
5606 | ||
5607 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath); | |
5608 | if (ret < 0) | |
5609 | goto out; | |
5610 | ||
5611 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
5612 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); | |
5613 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); | |
5614 | disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, ei); | |
5615 | ||
5616 | TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, fspath); | |
5617 | TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); | |
5618 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_FILE_LEN, | |
5619 | min(key.offset + btrfs_file_extent_num_bytes(leaf, ei) - offset, | |
5620 | len)); | |
5621 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_LEN, | |
5622 | btrfs_file_extent_ram_bytes(leaf, ei)); | |
5623 | TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_OFFSET, | |
5624 | offset - key.offset + btrfs_file_extent_offset(leaf, ei)); | |
5625 | ret = btrfs_encoded_io_compression_from_extent(fs_info, | |
5626 | btrfs_file_extent_compression(leaf, ei)); | |
5627 | if (ret < 0) | |
5628 | goto out; | |
5629 | TLV_PUT_U32(sctx, BTRFS_SEND_A_COMPRESSION, ret); | |
5630 | TLV_PUT_U32(sctx, BTRFS_SEND_A_ENCRYPTION, 0); | |
5631 | ||
5632 | ret = put_data_header(sctx, disk_num_bytes); | |
5633 | if (ret < 0) | |
5634 | goto out; | |
5635 | ||
5636 | /* | |
5637 | * We want to do I/O directly into the send buffer, so get the next page | |
5638 | * boundary in the send buffer. This means that there may be a gap | |
5639 | * between the beginning of the command and the file data. | |
5640 | */ | |
ce394a7f | 5641 | data_offset = PAGE_ALIGN(sctx->send_size); |
3ea4dc5b OS |
5642 | if (data_offset > sctx->send_max_size || |
5643 | sctx->send_max_size - data_offset < disk_num_bytes) { | |
5644 | ret = -EOVERFLOW; | |
5645 | goto out; | |
5646 | } | |
5647 | ||
5648 | /* | |
5649 | * Note that send_buf is a mapping of send_buf_pages, so this is really | |
5650 | * reading into send_buf. | |
5651 | */ | |
5652 | ret = btrfs_encoded_read_regular_fill_pages(BTRFS_I(inode), offset, | |
5653 | disk_bytenr, disk_num_bytes, | |
5654 | sctx->send_buf_pages + | |
5655 | (data_offset >> PAGE_SHIFT)); | |
5656 | if (ret) | |
5657 | goto out; | |
5658 | ||
5659 | hdr = (struct btrfs_cmd_header *)sctx->send_buf; | |
5660 | hdr->len = cpu_to_le32(sctx->send_size + disk_num_bytes - sizeof(*hdr)); | |
5661 | hdr->crc = 0; | |
5662 | crc = btrfs_crc32c(0, sctx->send_buf, sctx->send_size); | |
5663 | crc = btrfs_crc32c(crc, sctx->send_buf + data_offset, disk_num_bytes); | |
5664 | hdr->crc = cpu_to_le32(crc); | |
5665 | ||
5666 | ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, | |
5667 | &sctx->send_off); | |
5668 | if (!ret) { | |
5669 | ret = write_buf(sctx->send_filp, sctx->send_buf + data_offset, | |
5670 | disk_num_bytes, &sctx->send_off); | |
5671 | } | |
5672 | sctx->send_size = 0; | |
5673 | sctx->put_data = false; | |
5674 | ||
5675 | tlv_put_failure: | |
5676 | out: | |
5677 | fs_path_free(fspath); | |
5678 | iput(inode); | |
5679 | return ret; | |
5680 | } | |
5681 | ||
5682 | static int send_extent_data(struct send_ctx *sctx, struct btrfs_path *path, | |
5683 | const u64 offset, const u64 len) | |
d906d49f | 5684 | { |
152555b3 | 5685 | const u64 end = offset + len; |
3ea4dc5b OS |
5686 | struct extent_buffer *leaf = path->nodes[0]; |
5687 | struct btrfs_file_extent_item *ei; | |
8c7d9fe0 | 5688 | u64 read_size = max_send_read_size(sctx); |
d906d49f FM |
5689 | u64 sent = 0; |
5690 | ||
5691 | if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) | |
5692 | return send_update_extent(sctx, offset, len); | |
5693 | ||
3ea4dc5b OS |
5694 | ei = btrfs_item_ptr(leaf, path->slots[0], |
5695 | struct btrfs_file_extent_item); | |
5696 | if ((sctx->flags & BTRFS_SEND_FLAG_COMPRESSED) && | |
5697 | btrfs_file_extent_compression(leaf, ei) != BTRFS_COMPRESS_NONE) { | |
5698 | bool is_inline = (btrfs_file_extent_type(leaf, ei) == | |
5699 | BTRFS_FILE_EXTENT_INLINE); | |
5700 | ||
5701 | /* | |
5702 | * Send the compressed extent unless the compressed data is | |
5703 | * larger than the decompressed data. This can happen if we're | |
5704 | * not sending the entire extent, either because it has been | |
5705 | * partially overwritten/truncated or because this is a part of | |
5706 | * the extent that we couldn't clone in clone_range(). | |
5707 | */ | |
5708 | if (is_inline && | |
5709 | btrfs_file_extent_inline_item_len(leaf, | |
5710 | path->slots[0]) <= len) { | |
5711 | return send_encoded_inline_extent(sctx, path, offset, | |
5712 | len); | |
5713 | } else if (!is_inline && | |
5714 | btrfs_file_extent_disk_num_bytes(leaf, ei) <= len) { | |
5715 | return send_encoded_extent(sctx, path, offset, len); | |
5716 | } | |
5717 | } | |
5718 | ||
521b6803 FM |
5719 | if (sctx->cur_inode == NULL) { |
5720 | struct btrfs_root *root = sctx->send_root; | |
5721 | ||
5722 | sctx->cur_inode = btrfs_iget(root->fs_info->sb, sctx->cur_ino, root); | |
5723 | if (IS_ERR(sctx->cur_inode)) { | |
5724 | int err = PTR_ERR(sctx->cur_inode); | |
5725 | ||
5726 | sctx->cur_inode = NULL; | |
5727 | return err; | |
5728 | } | |
5729 | memset(&sctx->ra, 0, sizeof(struct file_ra_state)); | |
5730 | file_ra_state_init(&sctx->ra, sctx->cur_inode->i_mapping); | |
152555b3 FM |
5731 | |
5732 | /* | |
5733 | * It's very likely there are no pages from this inode in the page | |
5734 | * cache, so after reading extents and sending their data, we clean | |
5735 | * the page cache to avoid trashing the page cache (adding pressure | |
5736 | * to the page cache and forcing eviction of other data more useful | |
5737 | * for applications). | |
5738 | * | |
5739 | * We decide if we should clean the page cache simply by checking | |
5740 | * if the inode's mapping nrpages is 0 when we first open it, and | |
5741 | * not by using something like filemap_range_has_page() before | |
5742 | * reading an extent because when we ask the readahead code to | |
5743 | * read a given file range, it may (and almost always does) read | |
5744 | * pages from beyond that range (see the documentation for | |
5745 | * page_cache_sync_readahead()), so it would not be reliable, | |
5746 | * because after reading the first extent future calls to | |
5747 | * filemap_range_has_page() would return true because the readahead | |
5748 | * on the previous extent resulted in reading pages of the current | |
5749 | * extent as well. | |
5750 | */ | |
5751 | sctx->clean_page_cache = (sctx->cur_inode->i_mapping->nrpages == 0); | |
5752 | sctx->page_cache_clear_start = round_down(offset, PAGE_SIZE); | |
521b6803 FM |
5753 | } |
5754 | ||
d906d49f | 5755 | while (sent < len) { |
8c7d9fe0 | 5756 | u64 size = min(len - sent, read_size); |
d906d49f FM |
5757 | int ret; |
5758 | ||
d906d49f FM |
5759 | ret = send_write(sctx, offset + sent, size); |
5760 | if (ret < 0) | |
5761 | return ret; | |
a9b2e0de | 5762 | sent += size; |
d906d49f | 5763 | } |
152555b3 | 5764 | |
ce394a7f | 5765 | if (sctx->clean_page_cache && PAGE_ALIGNED(end)) { |
152555b3 FM |
5766 | /* |
5767 | * Always operate only on ranges that are a multiple of the page | |
5768 | * size. This is not only to prevent zeroing parts of a page in | |
5769 | * the case of subpage sector size, but also to guarantee we evict | |
5770 | * pages, as passing a range that is smaller than page size does | |
5771 | * not evict the respective page (only zeroes part of its content). | |
5772 | * | |
5773 | * Always start from the end offset of the last range cleared. | |
5774 | * This is because the readahead code may (and very often does) | |
5775 | * reads pages beyond the range we request for readahead. So if | |
5776 | * we have an extent layout like this: | |
5777 | * | |
5778 | * [ extent A ] [ extent B ] [ extent C ] | |
5779 | * | |
5780 | * When we ask page_cache_sync_readahead() to read extent A, it | |
5781 | * may also trigger reads for pages of extent B. If we are doing | |
5782 | * an incremental send and extent B has not changed between the | |
5783 | * parent and send snapshots, some or all of its pages may end | |
5784 | * up being read and placed in the page cache. So when truncating | |
5785 | * the page cache we always start from the end offset of the | |
5786 | * previously processed extent up to the end of the current | |
5787 | * extent. | |
5788 | */ | |
5789 | truncate_inode_pages_range(&sctx->cur_inode->i_data, | |
5790 | sctx->page_cache_clear_start, | |
5791 | end - 1); | |
5792 | sctx->page_cache_clear_start = end; | |
5793 | } | |
5794 | ||
d906d49f FM |
5795 | return 0; |
5796 | } | |
5797 | ||
89efda52 MPS |
5798 | /* |
5799 | * Search for a capability xattr related to sctx->cur_ino. If the capability is | |
5800 | * found, call send_set_xattr function to emit it. | |
5801 | * | |
5802 | * Return 0 if there isn't a capability, or when the capability was emitted | |
5803 | * successfully, or < 0 if an error occurred. | |
5804 | */ | |
5805 | static int send_capabilities(struct send_ctx *sctx) | |
5806 | { | |
5807 | struct fs_path *fspath = NULL; | |
5808 | struct btrfs_path *path; | |
5809 | struct btrfs_dir_item *di; | |
5810 | struct extent_buffer *leaf; | |
5811 | unsigned long data_ptr; | |
5812 | char *buf = NULL; | |
5813 | int buf_len; | |
5814 | int ret = 0; | |
5815 | ||
5816 | path = alloc_path_for_send(); | |
5817 | if (!path) | |
5818 | return -ENOMEM; | |
5819 | ||
5820 | di = btrfs_lookup_xattr(NULL, sctx->send_root, path, sctx->cur_ino, | |
5821 | XATTR_NAME_CAPS, strlen(XATTR_NAME_CAPS), 0); | |
5822 | if (!di) { | |
5823 | /* There is no xattr for this inode */ | |
5824 | goto out; | |
5825 | } else if (IS_ERR(di)) { | |
5826 | ret = PTR_ERR(di); | |
5827 | goto out; | |
5828 | } | |
5829 | ||
5830 | leaf = path->nodes[0]; | |
5831 | buf_len = btrfs_dir_data_len(leaf, di); | |
5832 | ||
5833 | fspath = fs_path_alloc(); | |
5834 | buf = kmalloc(buf_len, GFP_KERNEL); | |
5835 | if (!fspath || !buf) { | |
5836 | ret = -ENOMEM; | |
5837 | goto out; | |
5838 | } | |
5839 | ||
5840 | ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath); | |
5841 | if (ret < 0) | |
5842 | goto out; | |
5843 | ||
5844 | data_ptr = (unsigned long)(di + 1) + btrfs_dir_name_len(leaf, di); | |
5845 | read_extent_buffer(leaf, buf, data_ptr, buf_len); | |
5846 | ||
5847 | ret = send_set_xattr(sctx, fspath, XATTR_NAME_CAPS, | |
5848 | strlen(XATTR_NAME_CAPS), buf, buf_len); | |
5849 | out: | |
5850 | kfree(buf); | |
5851 | fs_path_free(fspath); | |
5852 | btrfs_free_path(path); | |
5853 | return ret; | |
5854 | } | |
5855 | ||
3ea4dc5b OS |
5856 | static int clone_range(struct send_ctx *sctx, struct btrfs_path *dst_path, |
5857 | struct clone_root *clone_root, const u64 disk_byte, | |
5858 | u64 data_offset, u64 offset, u64 len) | |
d906d49f FM |
5859 | { |
5860 | struct btrfs_path *path; | |
5861 | struct btrfs_key key; | |
5862 | int ret; | |
7e93f6dc | 5863 | struct btrfs_inode_info info; |
431d3988 | 5864 | u64 clone_src_i_size = 0; |
d906d49f | 5865 | |
72610b1b FM |
5866 | /* |
5867 | * Prevent cloning from a zero offset with a length matching the sector | |
5868 | * size because in some scenarios this will make the receiver fail. | |
5869 | * | |
5870 | * For example, if in the source filesystem the extent at offset 0 | |
5871 | * has a length of sectorsize and it was written using direct IO, then | |
5872 | * it can never be an inline extent (even if compression is enabled). | |
5873 | * Then this extent can be cloned in the original filesystem to a non | |
5874 | * zero file offset, but it may not be possible to clone in the | |
5875 | * destination filesystem because it can be inlined due to compression | |
5876 | * on the destination filesystem (as the receiver's write operations are | |
5877 | * always done using buffered IO). The same happens when the original | |
5878 | * filesystem does not have compression enabled but the destination | |
5879 | * filesystem has. | |
5880 | */ | |
5881 | if (clone_root->offset == 0 && | |
5882 | len == sctx->send_root->fs_info->sectorsize) | |
3ea4dc5b | 5883 | return send_extent_data(sctx, dst_path, offset, len); |
72610b1b | 5884 | |
d906d49f FM |
5885 | path = alloc_path_for_send(); |
5886 | if (!path) | |
5887 | return -ENOMEM; | |
5888 | ||
040ee612 RK |
5889 | /* |
5890 | * There are inodes that have extents that lie behind its i_size. Don't | |
5891 | * accept clones from these extents. | |
5892 | */ | |
7e93f6dc | 5893 | ret = get_inode_info(clone_root->root, clone_root->ino, &info); |
040ee612 RK |
5894 | btrfs_release_path(path); |
5895 | if (ret < 0) | |
5896 | goto out; | |
7e93f6dc | 5897 | clone_src_i_size = info.size; |
040ee612 | 5898 | |
d906d49f FM |
5899 | /* |
5900 | * We can't send a clone operation for the entire range if we find | |
5901 | * extent items in the respective range in the source file that | |
5902 | * refer to different extents or if we find holes. | |
5903 | * So check for that and do a mix of clone and regular write/copy | |
5904 | * operations if needed. | |
5905 | * | |
5906 | * Example: | |
5907 | * | |
5908 | * mkfs.btrfs -f /dev/sda | |
5909 | * mount /dev/sda /mnt | |
5910 | * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo | |
5911 | * cp --reflink=always /mnt/foo /mnt/bar | |
5912 | * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo | |
5913 | * btrfs subvolume snapshot -r /mnt /mnt/snap | |
5914 | * | |
5915 | * If when we send the snapshot and we are processing file bar (which | |
5916 | * has a higher inode number than foo) we blindly send a clone operation | |
5917 | * for the [0, 100K[ range from foo to bar, the receiver ends up getting | |
5918 | * a file bar that matches the content of file foo - iow, doesn't match | |
5919 | * the content from bar in the original filesystem. | |
5920 | */ | |
5921 | key.objectid = clone_root->ino; | |
5922 | key.type = BTRFS_EXTENT_DATA_KEY; | |
5923 | key.offset = clone_root->offset; | |
5924 | ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); | |
5925 | if (ret < 0) | |
5926 | goto out; | |
5927 | if (ret > 0 && path->slots[0] > 0) { | |
5928 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
5929 | if (key.objectid == clone_root->ino && | |
5930 | key.type == BTRFS_EXTENT_DATA_KEY) | |
5931 | path->slots[0]--; | |
5932 | } | |
5933 | ||
5934 | while (true) { | |
5935 | struct extent_buffer *leaf = path->nodes[0]; | |
5936 | int slot = path->slots[0]; | |
5937 | struct btrfs_file_extent_item *ei; | |
5938 | u8 type; | |
5939 | u64 ext_len; | |
5940 | u64 clone_len; | |
040ee612 | 5941 | u64 clone_data_offset; |
a11452a3 | 5942 | bool crossed_src_i_size = false; |
d906d49f FM |
5943 | |
5944 | if (slot >= btrfs_header_nritems(leaf)) { | |
5945 | ret = btrfs_next_leaf(clone_root->root, path); | |
5946 | if (ret < 0) | |
5947 | goto out; | |
5948 | else if (ret > 0) | |
5949 | break; | |
5950 | continue; | |
5951 | } | |
5952 | ||
5953 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
5954 | ||
5955 | /* | |
5956 | * We might have an implicit trailing hole (NO_HOLES feature | |
5957 | * enabled). We deal with it after leaving this loop. | |
5958 | */ | |
5959 | if (key.objectid != clone_root->ino || | |
5960 | key.type != BTRFS_EXTENT_DATA_KEY) | |
5961 | break; | |
5962 | ||
5963 | ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5964 | type = btrfs_file_extent_type(leaf, ei); | |
5965 | if (type == BTRFS_FILE_EXTENT_INLINE) { | |
e41ca589 | 5966 | ext_len = btrfs_file_extent_ram_bytes(leaf, ei); |
09cbfeaf | 5967 | ext_len = PAGE_ALIGN(ext_len); |
d906d49f FM |
5968 | } else { |
5969 | ext_len = btrfs_file_extent_num_bytes(leaf, ei); | |
5970 | } | |
5971 | ||
5972 | if (key.offset + ext_len <= clone_root->offset) | |
5973 | goto next; | |
5974 | ||
5975 | if (key.offset > clone_root->offset) { | |
5976 | /* Implicit hole, NO_HOLES feature enabled. */ | |
5977 | u64 hole_len = key.offset - clone_root->offset; | |
5978 | ||
5979 | if (hole_len > len) | |
5980 | hole_len = len; | |
3ea4dc5b OS |
5981 | ret = send_extent_data(sctx, dst_path, offset, |
5982 | hole_len); | |
d906d49f FM |
5983 | if (ret < 0) |
5984 | goto out; | |
5985 | ||
5986 | len -= hole_len; | |
5987 | if (len == 0) | |
5988 | break; | |
5989 | offset += hole_len; | |
5990 | clone_root->offset += hole_len; | |
5991 | data_offset += hole_len; | |
5992 | } | |
5993 | ||
5994 | if (key.offset >= clone_root->offset + len) | |
5995 | break; | |
5996 | ||
040ee612 RK |
5997 | if (key.offset >= clone_src_i_size) |
5998 | break; | |
5999 | ||
a11452a3 | 6000 | if (key.offset + ext_len > clone_src_i_size) { |
040ee612 | 6001 | ext_len = clone_src_i_size - key.offset; |
a11452a3 FM |
6002 | crossed_src_i_size = true; |
6003 | } | |
040ee612 RK |
6004 | |
6005 | clone_data_offset = btrfs_file_extent_offset(leaf, ei); | |
6006 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) { | |
6007 | clone_root->offset = key.offset; | |
6008 | if (clone_data_offset < data_offset && | |
6009 | clone_data_offset + ext_len > data_offset) { | |
6010 | u64 extent_offset; | |
6011 | ||
6012 | extent_offset = data_offset - clone_data_offset; | |
6013 | ext_len -= extent_offset; | |
6014 | clone_data_offset += extent_offset; | |
6015 | clone_root->offset += extent_offset; | |
6016 | } | |
6017 | } | |
6018 | ||
d906d49f FM |
6019 | clone_len = min_t(u64, ext_len, len); |
6020 | ||
6021 | if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && | |
3c850b45 FM |
6022 | clone_data_offset == data_offset) { |
6023 | const u64 src_end = clone_root->offset + clone_len; | |
6024 | const u64 sectorsize = SZ_64K; | |
6025 | ||
6026 | /* | |
6027 | * We can't clone the last block, when its size is not | |
6028 | * sector size aligned, into the middle of a file. If we | |
6029 | * do so, the receiver will get a failure (-EINVAL) when | |
6030 | * trying to clone or will silently corrupt the data in | |
6031 | * the destination file if it's on a kernel without the | |
6032 | * fix introduced by commit ac765f83f1397646 | |
6033 | * ("Btrfs: fix data corruption due to cloning of eof | |
6034 | * block). | |
6035 | * | |
6036 | * So issue a clone of the aligned down range plus a | |
6037 | * regular write for the eof block, if we hit that case. | |
6038 | * | |
6039 | * Also, we use the maximum possible sector size, 64K, | |
6040 | * because we don't know what's the sector size of the | |
6041 | * filesystem that receives the stream, so we have to | |
6042 | * assume the largest possible sector size. | |
6043 | */ | |
6044 | if (src_end == clone_src_i_size && | |
6045 | !IS_ALIGNED(src_end, sectorsize) && | |
6046 | offset + clone_len < sctx->cur_inode_size) { | |
6047 | u64 slen; | |
6048 | ||
6049 | slen = ALIGN_DOWN(src_end - clone_root->offset, | |
6050 | sectorsize); | |
6051 | if (slen > 0) { | |
6052 | ret = send_clone(sctx, offset, slen, | |
6053 | clone_root); | |
6054 | if (ret < 0) | |
6055 | goto out; | |
6056 | } | |
3ea4dc5b OS |
6057 | ret = send_extent_data(sctx, dst_path, |
6058 | offset + slen, | |
3c850b45 FM |
6059 | clone_len - slen); |
6060 | } else { | |
6061 | ret = send_clone(sctx, offset, clone_len, | |
6062 | clone_root); | |
6063 | } | |
a11452a3 FM |
6064 | } else if (crossed_src_i_size && clone_len < len) { |
6065 | /* | |
6066 | * If we are at i_size of the clone source inode and we | |
6067 | * can not clone from it, terminate the loop. This is | |
6068 | * to avoid sending two write operations, one with a | |
6069 | * length matching clone_len and the final one after | |
6070 | * this loop with a length of len - clone_len. | |
6071 | * | |
6072 | * When using encoded writes (BTRFS_SEND_FLAG_COMPRESSED | |
6073 | * was passed to the send ioctl), this helps avoid | |
6074 | * sending an encoded write for an offset that is not | |
6075 | * sector size aligned, in case the i_size of the source | |
6076 | * inode is not sector size aligned. That will make the | |
6077 | * receiver fallback to decompression of the data and | |
6078 | * writing it using regular buffered IO, therefore while | |
6079 | * not incorrect, it's not optimal due decompression and | |
6080 | * possible re-compression at the receiver. | |
6081 | */ | |
6082 | break; | |
3c850b45 | 6083 | } else { |
3ea4dc5b OS |
6084 | ret = send_extent_data(sctx, dst_path, offset, |
6085 | clone_len); | |
3c850b45 | 6086 | } |
d906d49f FM |
6087 | |
6088 | if (ret < 0) | |
6089 | goto out; | |
6090 | ||
6091 | len -= clone_len; | |
6092 | if (len == 0) | |
6093 | break; | |
6094 | offset += clone_len; | |
6095 | clone_root->offset += clone_len; | |
518837e6 FM |
6096 | |
6097 | /* | |
6098 | * If we are cloning from the file we are currently processing, | |
6099 | * and using the send root as the clone root, we must stop once | |
6100 | * the current clone offset reaches the current eof of the file | |
6101 | * at the receiver, otherwise we would issue an invalid clone | |
6102 | * operation (source range going beyond eof) and cause the | |
6103 | * receiver to fail. So if we reach the current eof, bail out | |
6104 | * and fallback to a regular write. | |
6105 | */ | |
6106 | if (clone_root->root == sctx->send_root && | |
6107 | clone_root->ino == sctx->cur_ino && | |
6108 | clone_root->offset >= sctx->cur_inode_next_write_offset) | |
6109 | break; | |
6110 | ||
d906d49f FM |
6111 | data_offset += clone_len; |
6112 | next: | |
6113 | path->slots[0]++; | |
6114 | } | |
6115 | ||
6116 | if (len > 0) | |
3ea4dc5b | 6117 | ret = send_extent_data(sctx, dst_path, offset, len); |
d906d49f FM |
6118 | else |
6119 | ret = 0; | |
6120 | out: | |
6121 | btrfs_free_path(path); | |
6122 | return ret; | |
6123 | } | |
6124 | ||
31db9f7c AB |
6125 | static int send_write_or_clone(struct send_ctx *sctx, |
6126 | struct btrfs_path *path, | |
6127 | struct btrfs_key *key, | |
6128 | struct clone_root *clone_root) | |
6129 | { | |
6130 | int ret = 0; | |
31db9f7c | 6131 | u64 offset = key->offset; |
c9a949af | 6132 | u64 end; |
28e5dd8f | 6133 | u64 bs = sctx->send_root->fs_info->sb->s_blocksize; |
31db9f7c | 6134 | |
c9a949af OS |
6135 | end = min_t(u64, btrfs_file_extent_end(path), sctx->cur_inode_size); |
6136 | if (offset >= end) | |
6137 | return 0; | |
31db9f7c | 6138 | |
c9a949af OS |
6139 | if (clone_root && IS_ALIGNED(end, bs)) { |
6140 | struct btrfs_file_extent_item *ei; | |
d906d49f FM |
6141 | u64 disk_byte; |
6142 | u64 data_offset; | |
6143 | ||
c9a949af OS |
6144 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6145 | struct btrfs_file_extent_item); | |
d906d49f FM |
6146 | disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); |
6147 | data_offset = btrfs_file_extent_offset(path->nodes[0], ei); | |
3ea4dc5b OS |
6148 | ret = clone_range(sctx, path, clone_root, disk_byte, |
6149 | data_offset, offset, end - offset); | |
cb95e7bf | 6150 | } else { |
3ea4dc5b | 6151 | ret = send_extent_data(sctx, path, offset, end - offset); |
31db9f7c | 6152 | } |
c9a949af | 6153 | sctx->cur_inode_next_write_offset = end; |
31db9f7c AB |
6154 | return ret; |
6155 | } | |
6156 | ||
6157 | static int is_extent_unchanged(struct send_ctx *sctx, | |
6158 | struct btrfs_path *left_path, | |
6159 | struct btrfs_key *ekey) | |
6160 | { | |
6161 | int ret = 0; | |
6162 | struct btrfs_key key; | |
6163 | struct btrfs_path *path = NULL; | |
6164 | struct extent_buffer *eb; | |
6165 | int slot; | |
6166 | struct btrfs_key found_key; | |
6167 | struct btrfs_file_extent_item *ei; | |
6168 | u64 left_disknr; | |
6169 | u64 right_disknr; | |
6170 | u64 left_offset; | |
6171 | u64 right_offset; | |
6172 | u64 left_offset_fixed; | |
6173 | u64 left_len; | |
6174 | u64 right_len; | |
74dd17fb CM |
6175 | u64 left_gen; |
6176 | u64 right_gen; | |
31db9f7c AB |
6177 | u8 left_type; |
6178 | u8 right_type; | |
6179 | ||
6180 | path = alloc_path_for_send(); | |
6181 | if (!path) | |
6182 | return -ENOMEM; | |
6183 | ||
6184 | eb = left_path->nodes[0]; | |
6185 | slot = left_path->slots[0]; | |
31db9f7c AB |
6186 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
6187 | left_type = btrfs_file_extent_type(eb, ei); | |
31db9f7c AB |
6188 | |
6189 | if (left_type != BTRFS_FILE_EXTENT_REG) { | |
6190 | ret = 0; | |
6191 | goto out; | |
6192 | } | |
74dd17fb CM |
6193 | left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
6194 | left_len = btrfs_file_extent_num_bytes(eb, ei); | |
6195 | left_offset = btrfs_file_extent_offset(eb, ei); | |
6196 | left_gen = btrfs_file_extent_generation(eb, ei); | |
31db9f7c AB |
6197 | |
6198 | /* | |
6199 | * Following comments will refer to these graphics. L is the left | |
6200 | * extents which we are checking at the moment. 1-8 are the right | |
6201 | * extents that we iterate. | |
6202 | * | |
6203 | * |-----L-----| | |
6204 | * |-1-|-2a-|-3-|-4-|-5-|-6-| | |
6205 | * | |
6206 | * |-----L-----| | |
6207 | * |--1--|-2b-|...(same as above) | |
6208 | * | |
6209 | * Alternative situation. Happens on files where extents got split. | |
6210 | * |-----L-----| | |
6211 | * |-----------7-----------|-6-| | |
6212 | * | |
6213 | * Alternative situation. Happens on files which got larger. | |
6214 | * |-----L-----| | |
6215 | * |-8-| | |
6216 | * Nothing follows after 8. | |
6217 | */ | |
6218 | ||
6219 | key.objectid = ekey->objectid; | |
6220 | key.type = BTRFS_EXTENT_DATA_KEY; | |
6221 | key.offset = ekey->offset; | |
6222 | ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); | |
6223 | if (ret < 0) | |
6224 | goto out; | |
6225 | if (ret) { | |
6226 | ret = 0; | |
6227 | goto out; | |
6228 | } | |
6229 | ||
6230 | /* | |
6231 | * Handle special case where the right side has no extents at all. | |
6232 | */ | |
6233 | eb = path->nodes[0]; | |
6234 | slot = path->slots[0]; | |
6235 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6236 | if (found_key.objectid != key.objectid || | |
6237 | found_key.type != key.type) { | |
57cfd462 JB |
6238 | /* If we're a hole then just pretend nothing changed */ |
6239 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
6240 | goto out; |
6241 | } | |
6242 | ||
6243 | /* | |
6244 | * We're now on 2a, 2b or 7. | |
6245 | */ | |
6246 | key = found_key; | |
6247 | while (key.offset < ekey->offset + left_len) { | |
6248 | ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); | |
6249 | right_type = btrfs_file_extent_type(eb, ei); | |
e1cbfd7b FM |
6250 | if (right_type != BTRFS_FILE_EXTENT_REG && |
6251 | right_type != BTRFS_FILE_EXTENT_INLINE) { | |
31db9f7c AB |
6252 | ret = 0; |
6253 | goto out; | |
6254 | } | |
6255 | ||
e1cbfd7b | 6256 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { |
e41ca589 | 6257 | right_len = btrfs_file_extent_ram_bytes(eb, ei); |
e1cbfd7b FM |
6258 | right_len = PAGE_ALIGN(right_len); |
6259 | } else { | |
6260 | right_len = btrfs_file_extent_num_bytes(eb, ei); | |
6261 | } | |
007d31f7 | 6262 | |
31db9f7c AB |
6263 | /* |
6264 | * Are we at extent 8? If yes, we know the extent is changed. | |
6265 | * This may only happen on the first iteration. | |
6266 | */ | |
d8347fa4 | 6267 | if (found_key.offset + right_len <= ekey->offset) { |
57cfd462 JB |
6268 | /* If we're a hole just pretend nothing changed */ |
6269 | ret = (left_disknr) ? 0 : 1; | |
31db9f7c AB |
6270 | goto out; |
6271 | } | |
6272 | ||
e1cbfd7b FM |
6273 | /* |
6274 | * We just wanted to see if when we have an inline extent, what | |
6275 | * follows it is a regular extent (wanted to check the above | |
6276 | * condition for inline extents too). This should normally not | |
6277 | * happen but it's possible for example when we have an inline | |
6278 | * compressed extent representing data with a size matching | |
6279 | * the page size (currently the same as sector size). | |
6280 | */ | |
6281 | if (right_type == BTRFS_FILE_EXTENT_INLINE) { | |
6282 | ret = 0; | |
6283 | goto out; | |
6284 | } | |
6285 | ||
24e52b11 FM |
6286 | right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); |
6287 | right_offset = btrfs_file_extent_offset(eb, ei); | |
6288 | right_gen = btrfs_file_extent_generation(eb, ei); | |
6289 | ||
31db9f7c AB |
6290 | left_offset_fixed = left_offset; |
6291 | if (key.offset < ekey->offset) { | |
6292 | /* Fix the right offset for 2a and 7. */ | |
6293 | right_offset += ekey->offset - key.offset; | |
6294 | } else { | |
6295 | /* Fix the left offset for all behind 2a and 2b */ | |
6296 | left_offset_fixed += key.offset - ekey->offset; | |
6297 | } | |
6298 | ||
6299 | /* | |
6300 | * Check if we have the same extent. | |
6301 | */ | |
3954096d | 6302 | if (left_disknr != right_disknr || |
74dd17fb CM |
6303 | left_offset_fixed != right_offset || |
6304 | left_gen != right_gen) { | |
31db9f7c AB |
6305 | ret = 0; |
6306 | goto out; | |
6307 | } | |
6308 | ||
6309 | /* | |
6310 | * Go to the next extent. | |
6311 | */ | |
6312 | ret = btrfs_next_item(sctx->parent_root, path); | |
6313 | if (ret < 0) | |
6314 | goto out; | |
6315 | if (!ret) { | |
6316 | eb = path->nodes[0]; | |
6317 | slot = path->slots[0]; | |
6318 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6319 | } | |
6320 | if (ret || found_key.objectid != key.objectid || | |
6321 | found_key.type != key.type) { | |
6322 | key.offset += right_len; | |
6323 | break; | |
adaa4b8e JS |
6324 | } |
6325 | if (found_key.offset != key.offset + right_len) { | |
6326 | ret = 0; | |
6327 | goto out; | |
31db9f7c AB |
6328 | } |
6329 | key = found_key; | |
6330 | } | |
6331 | ||
6332 | /* | |
6333 | * We're now behind the left extent (treat as unchanged) or at the end | |
6334 | * of the right side (treat as changed). | |
6335 | */ | |
6336 | if (key.offset >= ekey->offset + left_len) | |
6337 | ret = 1; | |
6338 | else | |
6339 | ret = 0; | |
6340 | ||
6341 | ||
6342 | out: | |
6343 | btrfs_free_path(path); | |
6344 | return ret; | |
6345 | } | |
6346 | ||
16e7549f JB |
6347 | static int get_last_extent(struct send_ctx *sctx, u64 offset) |
6348 | { | |
6349 | struct btrfs_path *path; | |
6350 | struct btrfs_root *root = sctx->send_root; | |
16e7549f | 6351 | struct btrfs_key key; |
16e7549f JB |
6352 | int ret; |
6353 | ||
6354 | path = alloc_path_for_send(); | |
6355 | if (!path) | |
6356 | return -ENOMEM; | |
6357 | ||
6358 | sctx->cur_inode_last_extent = 0; | |
6359 | ||
6360 | key.objectid = sctx->cur_ino; | |
6361 | key.type = BTRFS_EXTENT_DATA_KEY; | |
6362 | key.offset = offset; | |
6363 | ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); | |
6364 | if (ret < 0) | |
6365 | goto out; | |
6366 | ret = 0; | |
6367 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
6368 | if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
6369 | goto out; | |
6370 | ||
a5eeb3d1 | 6371 | sctx->cur_inode_last_extent = btrfs_file_extent_end(path); |
16e7549f JB |
6372 | out: |
6373 | btrfs_free_path(path); | |
6374 | return ret; | |
6375 | } | |
6376 | ||
82bfb2e7 FM |
6377 | static int range_is_hole_in_parent(struct send_ctx *sctx, |
6378 | const u64 start, | |
6379 | const u64 end) | |
6380 | { | |
6381 | struct btrfs_path *path; | |
6382 | struct btrfs_key key; | |
6383 | struct btrfs_root *root = sctx->parent_root; | |
6384 | u64 search_start = start; | |
6385 | int ret; | |
6386 | ||
6387 | path = alloc_path_for_send(); | |
6388 | if (!path) | |
6389 | return -ENOMEM; | |
6390 | ||
6391 | key.objectid = sctx->cur_ino; | |
6392 | key.type = BTRFS_EXTENT_DATA_KEY; | |
6393 | key.offset = search_start; | |
6394 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6395 | if (ret < 0) | |
6396 | goto out; | |
6397 | if (ret > 0 && path->slots[0] > 0) | |
6398 | path->slots[0]--; | |
6399 | ||
6400 | while (search_start < end) { | |
6401 | struct extent_buffer *leaf = path->nodes[0]; | |
6402 | int slot = path->slots[0]; | |
6403 | struct btrfs_file_extent_item *fi; | |
6404 | u64 extent_end; | |
6405 | ||
6406 | if (slot >= btrfs_header_nritems(leaf)) { | |
6407 | ret = btrfs_next_leaf(root, path); | |
6408 | if (ret < 0) | |
6409 | goto out; | |
6410 | else if (ret > 0) | |
6411 | break; | |
6412 | continue; | |
6413 | } | |
6414 | ||
6415 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
6416 | if (key.objectid < sctx->cur_ino || | |
6417 | key.type < BTRFS_EXTENT_DATA_KEY) | |
6418 | goto next; | |
6419 | if (key.objectid > sctx->cur_ino || | |
6420 | key.type > BTRFS_EXTENT_DATA_KEY || | |
6421 | key.offset >= end) | |
6422 | break; | |
6423 | ||
6424 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
a5eeb3d1 | 6425 | extent_end = btrfs_file_extent_end(path); |
82bfb2e7 FM |
6426 | if (extent_end <= start) |
6427 | goto next; | |
6428 | if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) { | |
6429 | search_start = extent_end; | |
6430 | goto next; | |
6431 | } | |
6432 | ret = 0; | |
6433 | goto out; | |
6434 | next: | |
6435 | path->slots[0]++; | |
6436 | } | |
6437 | ret = 1; | |
6438 | out: | |
6439 | btrfs_free_path(path); | |
6440 | return ret; | |
6441 | } | |
6442 | ||
16e7549f JB |
6443 | static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, |
6444 | struct btrfs_key *key) | |
6445 | { | |
16e7549f JB |
6446 | int ret = 0; |
6447 | ||
6448 | if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) | |
6449 | return 0; | |
6450 | ||
6451 | if (sctx->cur_inode_last_extent == (u64)-1) { | |
6452 | ret = get_last_extent(sctx, key->offset - 1); | |
6453 | if (ret) | |
6454 | return ret; | |
6455 | } | |
6456 | ||
bf54f412 FDBM |
6457 | if (path->slots[0] == 0 && |
6458 | sctx->cur_inode_last_extent < key->offset) { | |
6459 | /* | |
6460 | * We might have skipped entire leafs that contained only | |
6461 | * file extent items for our current inode. These leafs have | |
6462 | * a generation number smaller (older) than the one in the | |
6463 | * current leaf and the leaf our last extent came from, and | |
6464 | * are located between these 2 leafs. | |
6465 | */ | |
6466 | ret = get_last_extent(sctx, key->offset - 1); | |
6467 | if (ret) | |
6468 | return ret; | |
6469 | } | |
6470 | ||
82bfb2e7 FM |
6471 | if (sctx->cur_inode_last_extent < key->offset) { |
6472 | ret = range_is_hole_in_parent(sctx, | |
6473 | sctx->cur_inode_last_extent, | |
6474 | key->offset); | |
6475 | if (ret < 0) | |
6476 | return ret; | |
6477 | else if (ret == 0) | |
6478 | ret = send_hole(sctx, key->offset); | |
6479 | else | |
6480 | ret = 0; | |
6481 | } | |
a5eeb3d1 | 6482 | sctx->cur_inode_last_extent = btrfs_file_extent_end(path); |
16e7549f JB |
6483 | return ret; |
6484 | } | |
6485 | ||
31db9f7c AB |
6486 | static int process_extent(struct send_ctx *sctx, |
6487 | struct btrfs_path *path, | |
6488 | struct btrfs_key *key) | |
6489 | { | |
31db9f7c | 6490 | struct clone_root *found_clone = NULL; |
57cfd462 | 6491 | int ret = 0; |
31db9f7c AB |
6492 | |
6493 | if (S_ISLNK(sctx->cur_inode_mode)) | |
6494 | return 0; | |
6495 | ||
6496 | if (sctx->parent_root && !sctx->cur_inode_new) { | |
6497 | ret = is_extent_unchanged(sctx, path, key); | |
6498 | if (ret < 0) | |
6499 | goto out; | |
6500 | if (ret) { | |
6501 | ret = 0; | |
16e7549f | 6502 | goto out_hole; |
31db9f7c | 6503 | } |
57cfd462 JB |
6504 | } else { |
6505 | struct btrfs_file_extent_item *ei; | |
6506 | u8 type; | |
6507 | ||
6508 | ei = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
6509 | struct btrfs_file_extent_item); | |
6510 | type = btrfs_file_extent_type(path->nodes[0], ei); | |
6511 | if (type == BTRFS_FILE_EXTENT_PREALLOC || | |
6512 | type == BTRFS_FILE_EXTENT_REG) { | |
6513 | /* | |
6514 | * The send spec does not have a prealloc command yet, | |
6515 | * so just leave a hole for prealloc'ed extents until | |
6516 | * we have enough commands queued up to justify rev'ing | |
6517 | * the send spec. | |
6518 | */ | |
6519 | if (type == BTRFS_FILE_EXTENT_PREALLOC) { | |
6520 | ret = 0; | |
6521 | goto out; | |
6522 | } | |
6523 | ||
6524 | /* Have a hole, just skip it. */ | |
6525 | if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) { | |
6526 | ret = 0; | |
6527 | goto out; | |
6528 | } | |
6529 | } | |
31db9f7c AB |
6530 | } |
6531 | ||
6532 | ret = find_extent_clone(sctx, path, key->objectid, key->offset, | |
6533 | sctx->cur_inode_size, &found_clone); | |
6534 | if (ret != -ENOENT && ret < 0) | |
6535 | goto out; | |
6536 | ||
6537 | ret = send_write_or_clone(sctx, path, key, found_clone); | |
16e7549f JB |
6538 | if (ret) |
6539 | goto out; | |
6540 | out_hole: | |
6541 | ret = maybe_send_hole(sctx, path, key); | |
31db9f7c AB |
6542 | out: |
6543 | return ret; | |
6544 | } | |
6545 | ||
6546 | static int process_all_extents(struct send_ctx *sctx) | |
6547 | { | |
9930e9d4 GN |
6548 | int ret = 0; |
6549 | int iter_ret = 0; | |
31db9f7c AB |
6550 | struct btrfs_root *root; |
6551 | struct btrfs_path *path; | |
6552 | struct btrfs_key key; | |
6553 | struct btrfs_key found_key; | |
31db9f7c AB |
6554 | |
6555 | root = sctx->send_root; | |
6556 | path = alloc_path_for_send(); | |
6557 | if (!path) | |
6558 | return -ENOMEM; | |
6559 | ||
6560 | key.objectid = sctx->cmp_key->objectid; | |
6561 | key.type = BTRFS_EXTENT_DATA_KEY; | |
6562 | key.offset = 0; | |
9930e9d4 | 6563 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
31db9f7c AB |
6564 | if (found_key.objectid != key.objectid || |
6565 | found_key.type != key.type) { | |
6566 | ret = 0; | |
9930e9d4 | 6567 | break; |
31db9f7c AB |
6568 | } |
6569 | ||
6570 | ret = process_extent(sctx, path, &found_key); | |
6571 | if (ret < 0) | |
9930e9d4 | 6572 | break; |
31db9f7c | 6573 | } |
9930e9d4 GN |
6574 | /* Catch error found during iteration */ |
6575 | if (iter_ret < 0) | |
6576 | ret = iter_ret; | |
31db9f7c | 6577 | |
31db9f7c AB |
6578 | btrfs_free_path(path); |
6579 | return ret; | |
6580 | } | |
6581 | ||
9f03740a FDBM |
6582 | static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, |
6583 | int *pending_move, | |
6584 | int *refs_processed) | |
31db9f7c AB |
6585 | { |
6586 | int ret = 0; | |
6587 | ||
6588 | if (sctx->cur_ino == 0) | |
6589 | goto out; | |
6590 | if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid && | |
96b5bd77 | 6591 | sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY) |
31db9f7c AB |
6592 | goto out; |
6593 | if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs)) | |
6594 | goto out; | |
6595 | ||
9f03740a | 6596 | ret = process_recorded_refs(sctx, pending_move); |
e479d9bb AB |
6597 | if (ret < 0) |
6598 | goto out; | |
6599 | ||
9f03740a | 6600 | *refs_processed = 1; |
31db9f7c AB |
6601 | out: |
6602 | return ret; | |
6603 | } | |
6604 | ||
6605 | static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) | |
6606 | { | |
6607 | int ret = 0; | |
7e93f6dc | 6608 | struct btrfs_inode_info info; |
31db9f7c AB |
6609 | u64 left_mode; |
6610 | u64 left_uid; | |
6611 | u64 left_gid; | |
48247359 | 6612 | u64 left_fileattr; |
31db9f7c AB |
6613 | u64 right_mode; |
6614 | u64 right_uid; | |
6615 | u64 right_gid; | |
48247359 | 6616 | u64 right_fileattr; |
31db9f7c AB |
6617 | int need_chmod = 0; |
6618 | int need_chown = 0; | |
48247359 | 6619 | bool need_fileattr = false; |
ffa7c429 | 6620 | int need_truncate = 1; |
9f03740a FDBM |
6621 | int pending_move = 0; |
6622 | int refs_processed = 0; | |
31db9f7c | 6623 | |
46b2f459 FM |
6624 | if (sctx->ignore_cur_inode) |
6625 | return 0; | |
6626 | ||
9f03740a FDBM |
6627 | ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move, |
6628 | &refs_processed); | |
31db9f7c AB |
6629 | if (ret < 0) |
6630 | goto out; | |
6631 | ||
9f03740a FDBM |
6632 | /* |
6633 | * We have processed the refs and thus need to advance send_progress. | |
6634 | * Now, calls to get_cur_xxx will take the updated refs of the current | |
6635 | * inode into account. | |
6636 | * | |
6637 | * On the other hand, if our current inode is a directory and couldn't | |
6638 | * be moved/renamed because its parent was renamed/moved too and it has | |
6639 | * a higher inode number, we can only move/rename our current inode | |
6640 | * after we moved/renamed its parent. Therefore in this case operate on | |
6641 | * the old path (pre move/rename) of our current inode, and the | |
6642 | * move/rename will be performed later. | |
6643 | */ | |
6644 | if (refs_processed && !pending_move) | |
6645 | sctx->send_progress = sctx->cur_ino + 1; | |
6646 | ||
31db9f7c AB |
6647 | if (sctx->cur_ino == 0 || sctx->cur_inode_deleted) |
6648 | goto out; | |
6649 | if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) | |
6650 | goto out; | |
7e93f6dc | 6651 | ret = get_inode_info(sctx->send_root, sctx->cur_ino, &info); |
31db9f7c AB |
6652 | if (ret < 0) |
6653 | goto out; | |
7e93f6dc BC |
6654 | left_mode = info.mode; |
6655 | left_uid = info.uid; | |
6656 | left_gid = info.gid; | |
6657 | left_fileattr = info.fileattr; | |
31db9f7c | 6658 | |
e2d044fe AL |
6659 | if (!sctx->parent_root || sctx->cur_inode_new) { |
6660 | need_chown = 1; | |
6661 | if (!S_ISLNK(sctx->cur_inode_mode)) | |
31db9f7c | 6662 | need_chmod = 1; |
ffa7c429 FM |
6663 | if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size) |
6664 | need_truncate = 0; | |
e2d044fe | 6665 | } else { |
ffa7c429 FM |
6666 | u64 old_size; |
6667 | ||
7e93f6dc | 6668 | ret = get_inode_info(sctx->parent_root, sctx->cur_ino, &info); |
e2d044fe AL |
6669 | if (ret < 0) |
6670 | goto out; | |
7e93f6dc BC |
6671 | old_size = info.size; |
6672 | right_mode = info.mode; | |
6673 | right_uid = info.uid; | |
6674 | right_gid = info.gid; | |
6675 | right_fileattr = info.fileattr; | |
31db9f7c | 6676 | |
e2d044fe AL |
6677 | if (left_uid != right_uid || left_gid != right_gid) |
6678 | need_chown = 1; | |
6679 | if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) | |
6680 | need_chmod = 1; | |
48247359 DS |
6681 | if (!S_ISLNK(sctx->cur_inode_mode) && left_fileattr != right_fileattr) |
6682 | need_fileattr = true; | |
ffa7c429 FM |
6683 | if ((old_size == sctx->cur_inode_size) || |
6684 | (sctx->cur_inode_size > old_size && | |
6685 | sctx->cur_inode_next_write_offset == sctx->cur_inode_size)) | |
6686 | need_truncate = 0; | |
31db9f7c AB |
6687 | } |
6688 | ||
6689 | if (S_ISREG(sctx->cur_inode_mode)) { | |
16e7549f | 6690 | if (need_send_hole(sctx)) { |
766b5e5a FM |
6691 | if (sctx->cur_inode_last_extent == (u64)-1 || |
6692 | sctx->cur_inode_last_extent < | |
6693 | sctx->cur_inode_size) { | |
16e7549f JB |
6694 | ret = get_last_extent(sctx, (u64)-1); |
6695 | if (ret) | |
6696 | goto out; | |
6697 | } | |
6698 | if (sctx->cur_inode_last_extent < | |
6699 | sctx->cur_inode_size) { | |
6700 | ret = send_hole(sctx, sctx->cur_inode_size); | |
6701 | if (ret) | |
6702 | goto out; | |
6703 | } | |
6704 | } | |
ffa7c429 FM |
6705 | if (need_truncate) { |
6706 | ret = send_truncate(sctx, sctx->cur_ino, | |
6707 | sctx->cur_inode_gen, | |
6708 | sctx->cur_inode_size); | |
6709 | if (ret < 0) | |
6710 | goto out; | |
6711 | } | |
31db9f7c AB |
6712 | } |
6713 | ||
6714 | if (need_chown) { | |
6715 | ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
6716 | left_uid, left_gid); | |
6717 | if (ret < 0) | |
6718 | goto out; | |
6719 | } | |
6720 | if (need_chmod) { | |
6721 | ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
6722 | left_mode); | |
6723 | if (ret < 0) | |
6724 | goto out; | |
6725 | } | |
48247359 DS |
6726 | if (need_fileattr) { |
6727 | ret = send_fileattr(sctx, sctx->cur_ino, sctx->cur_inode_gen, | |
6728 | left_fileattr); | |
6729 | if (ret < 0) | |
6730 | goto out; | |
6731 | } | |
c86eab81 DS |
6732 | |
6733 | if (proto_cmd_ok(sctx, BTRFS_SEND_C_ENABLE_VERITY) | |
6734 | && sctx->cur_inode_needs_verity) { | |
38622010 BB |
6735 | ret = process_verity(sctx); |
6736 | if (ret < 0) | |
6737 | goto out; | |
6738 | } | |
31db9f7c | 6739 | |
89efda52 MPS |
6740 | ret = send_capabilities(sctx); |
6741 | if (ret < 0) | |
6742 | goto out; | |
6743 | ||
31db9f7c | 6744 | /* |
9f03740a FDBM |
6745 | * If other directory inodes depended on our current directory |
6746 | * inode's move/rename, now do their move/rename operations. | |
31db9f7c | 6747 | */ |
9f03740a FDBM |
6748 | if (!is_waiting_for_move(sctx, sctx->cur_ino)) { |
6749 | ret = apply_children_dir_moves(sctx); | |
6750 | if (ret) | |
6751 | goto out; | |
fcbd2154 FM |
6752 | /* |
6753 | * Need to send that every time, no matter if it actually | |
6754 | * changed between the two trees as we have done changes to | |
6755 | * the inode before. If our inode is a directory and it's | |
6756 | * waiting to be moved/renamed, we will send its utimes when | |
6757 | * it's moved/renamed, therefore we don't need to do it here. | |
6758 | */ | |
6759 | sctx->send_progress = sctx->cur_ino + 1; | |
3e49363b FM |
6760 | |
6761 | /* | |
6762 | * If the current inode is a non-empty directory, delay issuing | |
6763 | * the utimes command for it, as it's very likely we have inodes | |
6764 | * with an higher number inside it. We want to issue the utimes | |
6765 | * command only after adding all dentries to it. | |
6766 | */ | |
6767 | if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_size > 0) | |
6768 | ret = cache_dir_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
6769 | else | |
6770 | ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); | |
6771 | ||
fcbd2154 FM |
6772 | if (ret < 0) |
6773 | goto out; | |
9f03740a FDBM |
6774 | } |
6775 | ||
31db9f7c | 6776 | out: |
3e49363b FM |
6777 | if (!ret) |
6778 | ret = trim_dir_utimes_cache(sctx); | |
6779 | ||
31db9f7c AB |
6780 | return ret; |
6781 | } | |
6782 | ||
152555b3 FM |
6783 | static void close_current_inode(struct send_ctx *sctx) |
6784 | { | |
6785 | u64 i_size; | |
6786 | ||
6787 | if (sctx->cur_inode == NULL) | |
6788 | return; | |
6789 | ||
6790 | i_size = i_size_read(sctx->cur_inode); | |
6791 | ||
6792 | /* | |
6793 | * If we are doing an incremental send, we may have extents between the | |
6794 | * last processed extent and the i_size that have not been processed | |
6795 | * because they haven't changed but we may have read some of their pages | |
6796 | * through readahead, see the comments at send_extent_data(). | |
6797 | */ | |
6798 | if (sctx->clean_page_cache && sctx->page_cache_clear_start < i_size) | |
6799 | truncate_inode_pages_range(&sctx->cur_inode->i_data, | |
6800 | sctx->page_cache_clear_start, | |
6801 | round_up(i_size, PAGE_SIZE) - 1); | |
6802 | ||
6803 | iput(sctx->cur_inode); | |
6804 | sctx->cur_inode = NULL; | |
6805 | } | |
6806 | ||
31db9f7c AB |
6807 | static int changed_inode(struct send_ctx *sctx, |
6808 | enum btrfs_compare_tree_result result) | |
6809 | { | |
6810 | int ret = 0; | |
6811 | struct btrfs_key *key = sctx->cmp_key; | |
6812 | struct btrfs_inode_item *left_ii = NULL; | |
6813 | struct btrfs_inode_item *right_ii = NULL; | |
6814 | u64 left_gen = 0; | |
6815 | u64 right_gen = 0; | |
6816 | ||
152555b3 | 6817 | close_current_inode(sctx); |
521b6803 | 6818 | |
31db9f7c | 6819 | sctx->cur_ino = key->objectid; |
9555e1f1 | 6820 | sctx->cur_inode_new_gen = false; |
16e7549f | 6821 | sctx->cur_inode_last_extent = (u64)-1; |
ffa7c429 | 6822 | sctx->cur_inode_next_write_offset = 0; |
46b2f459 | 6823 | sctx->ignore_cur_inode = false; |
e479d9bb AB |
6824 | |
6825 | /* | |
6826 | * Set send_progress to current inode. This will tell all get_cur_xxx | |
6827 | * functions that the current inode's refs are not updated yet. Later, | |
6828 | * when process_recorded_refs is finished, it is set to cur_ino + 1. | |
6829 | */ | |
31db9f7c AB |
6830 | sctx->send_progress = sctx->cur_ino; |
6831 | ||
6832 | if (result == BTRFS_COMPARE_TREE_NEW || | |
6833 | result == BTRFS_COMPARE_TREE_CHANGED) { | |
6834 | left_ii = btrfs_item_ptr(sctx->left_path->nodes[0], | |
6835 | sctx->left_path->slots[0], | |
6836 | struct btrfs_inode_item); | |
6837 | left_gen = btrfs_inode_generation(sctx->left_path->nodes[0], | |
6838 | left_ii); | |
6839 | } else { | |
6840 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6841 | sctx->right_path->slots[0], | |
6842 | struct btrfs_inode_item); | |
6843 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6844 | right_ii); | |
6845 | } | |
6846 | if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
6847 | right_ii = btrfs_item_ptr(sctx->right_path->nodes[0], | |
6848 | sctx->right_path->slots[0], | |
6849 | struct btrfs_inode_item); | |
6850 | ||
6851 | right_gen = btrfs_inode_generation(sctx->right_path->nodes[0], | |
6852 | right_ii); | |
6d85ed05 AB |
6853 | |
6854 | /* | |
6855 | * The cur_ino = root dir case is special here. We can't treat | |
6856 | * the inode as deleted+reused because it would generate a | |
6857 | * stream that tries to delete/mkdir the root dir. | |
6858 | */ | |
6859 | if (left_gen != right_gen && | |
6860 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) | |
9555e1f1 | 6861 | sctx->cur_inode_new_gen = true; |
31db9f7c AB |
6862 | } |
6863 | ||
46b2f459 FM |
6864 | /* |
6865 | * Normally we do not find inodes with a link count of zero (orphans) | |
6866 | * because the most common case is to create a snapshot and use it | |
6867 | * for a send operation. However other less common use cases involve | |
6868 | * using a subvolume and send it after turning it to RO mode just | |
6869 | * after deleting all hard links of a file while holding an open | |
6870 | * file descriptor against it or turning a RO snapshot into RW mode, | |
6871 | * keep an open file descriptor against a file, delete it and then | |
6872 | * turn the snapshot back to RO mode before using it for a send | |
9ed0a72e BC |
6873 | * operation. The former is what the receiver operation does. |
6874 | * Therefore, if we want to send these snapshots soon after they're | |
6875 | * received, we need to handle orphan inodes as well. Moreover, orphans | |
6876 | * can appear not only in the send snapshot but also in the parent | |
6877 | * snapshot. Here are several cases: | |
6878 | * | |
6879 | * Case 1: BTRFS_COMPARE_TREE_NEW | |
6880 | * | send snapshot | action | |
6881 | * -------------------------------- | |
6882 | * nlink | 0 | ignore | |
6883 | * | |
6884 | * Case 2: BTRFS_COMPARE_TREE_DELETED | |
6885 | * | parent snapshot | action | |
6886 | * ---------------------------------- | |
6887 | * nlink | 0 | as usual | |
6888 | * Note: No unlinks will be sent because there're no paths for it. | |
6889 | * | |
6890 | * Case 3: BTRFS_COMPARE_TREE_CHANGED | |
6891 | * | | parent snapshot | send snapshot | action | |
6892 | * ----------------------------------------------------------------------- | |
6893 | * subcase 1 | nlink | 0 | 0 | ignore | |
6894 | * subcase 2 | nlink | >0 | 0 | new_gen(deletion) | |
6895 | * subcase 3 | nlink | 0 | >0 | new_gen(creation) | |
6896 | * | |
46b2f459 | 6897 | */ |
9ed0a72e BC |
6898 | if (result == BTRFS_COMPARE_TREE_NEW) { |
6899 | if (btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii) == 0) { | |
46b2f459 | 6900 | sctx->ignore_cur_inode = true; |
46b2f459 FM |
6901 | goto out; |
6902 | } | |
31db9f7c | 6903 | sctx->cur_inode_gen = left_gen; |
9555e1f1 DS |
6904 | sctx->cur_inode_new = true; |
6905 | sctx->cur_inode_deleted = false; | |
31db9f7c AB |
6906 | sctx->cur_inode_size = btrfs_inode_size( |
6907 | sctx->left_path->nodes[0], left_ii); | |
6908 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6909 | sctx->left_path->nodes[0], left_ii); | |
644d1940 LB |
6910 | sctx->cur_inode_rdev = btrfs_inode_rdev( |
6911 | sctx->left_path->nodes[0], left_ii); | |
31db9f7c | 6912 | if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) |
1f4692da | 6913 | ret = send_create_inode_if_needed(sctx); |
31db9f7c AB |
6914 | } else if (result == BTRFS_COMPARE_TREE_DELETED) { |
6915 | sctx->cur_inode_gen = right_gen; | |
9555e1f1 DS |
6916 | sctx->cur_inode_new = false; |
6917 | sctx->cur_inode_deleted = true; | |
31db9f7c AB |
6918 | sctx->cur_inode_size = btrfs_inode_size( |
6919 | sctx->right_path->nodes[0], right_ii); | |
6920 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6921 | sctx->right_path->nodes[0], right_ii); | |
6922 | } else if (result == BTRFS_COMPARE_TREE_CHANGED) { | |
9ed0a72e BC |
6923 | u32 new_nlinks, old_nlinks; |
6924 | ||
6925 | new_nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii); | |
6926 | old_nlinks = btrfs_inode_nlink(sctx->right_path->nodes[0], right_ii); | |
6927 | if (new_nlinks == 0 && old_nlinks == 0) { | |
6928 | sctx->ignore_cur_inode = true; | |
6929 | goto out; | |
6930 | } else if (new_nlinks == 0 || old_nlinks == 0) { | |
6931 | sctx->cur_inode_new_gen = 1; | |
6932 | } | |
766702ef AB |
6933 | /* |
6934 | * We need to do some special handling in case the inode was | |
6935 | * reported as changed with a changed generation number. This | |
6936 | * means that the original inode was deleted and new inode | |
6937 | * reused the same inum. So we have to treat the old inode as | |
6938 | * deleted and the new one as new. | |
6939 | */ | |
31db9f7c | 6940 | if (sctx->cur_inode_new_gen) { |
766702ef AB |
6941 | /* |
6942 | * First, process the inode as if it was deleted. | |
6943 | */ | |
9b8be45f BC |
6944 | if (old_nlinks > 0) { |
6945 | sctx->cur_inode_gen = right_gen; | |
6946 | sctx->cur_inode_new = false; | |
6947 | sctx->cur_inode_deleted = true; | |
6948 | sctx->cur_inode_size = btrfs_inode_size( | |
6949 | sctx->right_path->nodes[0], right_ii); | |
6950 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6951 | sctx->right_path->nodes[0], right_ii); | |
6952 | ret = process_all_refs(sctx, | |
6953 | BTRFS_COMPARE_TREE_DELETED); | |
6954 | if (ret < 0) | |
6955 | goto out; | |
6956 | } | |
31db9f7c | 6957 | |
766702ef AB |
6958 | /* |
6959 | * Now process the inode as if it was new. | |
6960 | */ | |
9ed0a72e BC |
6961 | if (new_nlinks > 0) { |
6962 | sctx->cur_inode_gen = left_gen; | |
6963 | sctx->cur_inode_new = true; | |
6964 | sctx->cur_inode_deleted = false; | |
6965 | sctx->cur_inode_size = btrfs_inode_size( | |
6966 | sctx->left_path->nodes[0], | |
6967 | left_ii); | |
6968 | sctx->cur_inode_mode = btrfs_inode_mode( | |
6969 | sctx->left_path->nodes[0], | |
6970 | left_ii); | |
6971 | sctx->cur_inode_rdev = btrfs_inode_rdev( | |
6972 | sctx->left_path->nodes[0], | |
6973 | left_ii); | |
6974 | ret = send_create_inode_if_needed(sctx); | |
6975 | if (ret < 0) | |
6976 | goto out; | |
31db9f7c | 6977 | |
9ed0a72e BC |
6978 | ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); |
6979 | if (ret < 0) | |
6980 | goto out; | |
6981 | /* | |
6982 | * Advance send_progress now as we did not get | |
6983 | * into process_recorded_refs_if_needed in the | |
6984 | * new_gen case. | |
6985 | */ | |
6986 | sctx->send_progress = sctx->cur_ino + 1; | |
766702ef | 6987 | |
9ed0a72e BC |
6988 | /* |
6989 | * Now process all extents and xattrs of the | |
6990 | * inode as if they were all new. | |
6991 | */ | |
6992 | ret = process_all_extents(sctx); | |
6993 | if (ret < 0) | |
6994 | goto out; | |
6995 | ret = process_all_new_xattrs(sctx); | |
6996 | if (ret < 0) | |
6997 | goto out; | |
6998 | } | |
31db9f7c AB |
6999 | } else { |
7000 | sctx->cur_inode_gen = left_gen; | |
9555e1f1 DS |
7001 | sctx->cur_inode_new = false; |
7002 | sctx->cur_inode_new_gen = false; | |
7003 | sctx->cur_inode_deleted = false; | |
31db9f7c AB |
7004 | sctx->cur_inode_size = btrfs_inode_size( |
7005 | sctx->left_path->nodes[0], left_ii); | |
7006 | sctx->cur_inode_mode = btrfs_inode_mode( | |
7007 | sctx->left_path->nodes[0], left_ii); | |
7008 | } | |
7009 | } | |
7010 | ||
7011 | out: | |
7012 | return ret; | |
7013 | } | |
7014 | ||
766702ef AB |
7015 | /* |
7016 | * We have to process new refs before deleted refs, but compare_trees gives us | |
7017 | * the new and deleted refs mixed. To fix this, we record the new/deleted refs | |
7018 | * first and later process them in process_recorded_refs. | |
7019 | * For the cur_inode_new_gen case, we skip recording completely because | |
7020 | * changed_inode did already initiate processing of refs. The reason for this is | |
7021 | * that in this case, compare_tree actually compares the refs of 2 different | |
7022 | * inodes. To fix this, process_all_refs is used in changed_inode to handle all | |
7023 | * refs of the right tree as deleted and all refs of the left tree as new. | |
7024 | */ | |
31db9f7c AB |
7025 | static int changed_ref(struct send_ctx *sctx, |
7026 | enum btrfs_compare_tree_result result) | |
7027 | { | |
7028 | int ret = 0; | |
7029 | ||
95155585 FM |
7030 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
7031 | inconsistent_snapshot_error(sctx, result, "reference"); | |
7032 | return -EIO; | |
7033 | } | |
31db9f7c AB |
7034 | |
7035 | if (!sctx->cur_inode_new_gen && | |
7036 | sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) { | |
7037 | if (result == BTRFS_COMPARE_TREE_NEW) | |
7038 | ret = record_new_ref(sctx); | |
7039 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
7040 | ret = record_deleted_ref(sctx); | |
7041 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
7042 | ret = record_changed_ref(sctx); | |
7043 | } | |
7044 | ||
7045 | return ret; | |
7046 | } | |
7047 | ||
766702ef AB |
7048 | /* |
7049 | * Process new/deleted/changed xattrs. We skip processing in the | |
7050 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
7051 | * of xattrs. The reason is the same as in changed_ref | |
7052 | */ | |
31db9f7c AB |
7053 | static int changed_xattr(struct send_ctx *sctx, |
7054 | enum btrfs_compare_tree_result result) | |
7055 | { | |
7056 | int ret = 0; | |
7057 | ||
95155585 FM |
7058 | if (sctx->cur_ino != sctx->cmp_key->objectid) { |
7059 | inconsistent_snapshot_error(sctx, result, "xattr"); | |
7060 | return -EIO; | |
7061 | } | |
31db9f7c AB |
7062 | |
7063 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
7064 | if (result == BTRFS_COMPARE_TREE_NEW) | |
7065 | ret = process_new_xattr(sctx); | |
7066 | else if (result == BTRFS_COMPARE_TREE_DELETED) | |
7067 | ret = process_deleted_xattr(sctx); | |
7068 | else if (result == BTRFS_COMPARE_TREE_CHANGED) | |
7069 | ret = process_changed_xattr(sctx); | |
7070 | } | |
7071 | ||
7072 | return ret; | |
7073 | } | |
7074 | ||
766702ef AB |
7075 | /* |
7076 | * Process new/deleted/changed extents. We skip processing in the | |
7077 | * cur_inode_new_gen case because changed_inode did already initiate processing | |
7078 | * of extents. The reason is the same as in changed_ref | |
7079 | */ | |
31db9f7c AB |
7080 | static int changed_extent(struct send_ctx *sctx, |
7081 | enum btrfs_compare_tree_result result) | |
7082 | { | |
7083 | int ret = 0; | |
7084 | ||
b4f9a1a8 FM |
7085 | /* |
7086 | * We have found an extent item that changed without the inode item | |
7087 | * having changed. This can happen either after relocation (where the | |
7088 | * disk_bytenr of an extent item is replaced at | |
7089 | * relocation.c:replace_file_extents()) or after deduplication into a | |
7090 | * file in both the parent and send snapshots (where an extent item can | |
7091 | * get modified or replaced with a new one). Note that deduplication | |
7092 | * updates the inode item, but it only changes the iversion (sequence | |
7093 | * field in the inode item) of the inode, so if a file is deduplicated | |
7094 | * the same amount of times in both the parent and send snapshots, its | |
1a9fd417 | 7095 | * iversion becomes the same in both snapshots, whence the inode item is |
b4f9a1a8 FM |
7096 | * the same on both snapshots. |
7097 | */ | |
7098 | if (sctx->cur_ino != sctx->cmp_key->objectid) | |
7099 | return 0; | |
31db9f7c AB |
7100 | |
7101 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
7102 | if (result != BTRFS_COMPARE_TREE_DELETED) | |
7103 | ret = process_extent(sctx, sctx->left_path, | |
7104 | sctx->cmp_key); | |
7105 | } | |
7106 | ||
7107 | return ret; | |
7108 | } | |
7109 | ||
38622010 BB |
7110 | static int changed_verity(struct send_ctx *sctx, enum btrfs_compare_tree_result result) |
7111 | { | |
7112 | int ret = 0; | |
7113 | ||
7114 | if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { | |
7115 | if (result == BTRFS_COMPARE_TREE_NEW) | |
7116 | sctx->cur_inode_needs_verity = true; | |
7117 | } | |
7118 | return ret; | |
7119 | } | |
7120 | ||
ba5e8f2e JB |
7121 | static int dir_changed(struct send_ctx *sctx, u64 dir) |
7122 | { | |
7123 | u64 orig_gen, new_gen; | |
7124 | int ret; | |
7125 | ||
7e93f6dc | 7126 | ret = get_inode_gen(sctx->send_root, dir, &new_gen); |
ba5e8f2e JB |
7127 | if (ret) |
7128 | return ret; | |
7129 | ||
7e93f6dc | 7130 | ret = get_inode_gen(sctx->parent_root, dir, &orig_gen); |
ba5e8f2e JB |
7131 | if (ret) |
7132 | return ret; | |
7133 | ||
7134 | return (orig_gen != new_gen) ? 1 : 0; | |
7135 | } | |
7136 | ||
7137 | static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, | |
7138 | struct btrfs_key *key) | |
7139 | { | |
7140 | struct btrfs_inode_extref *extref; | |
7141 | struct extent_buffer *leaf; | |
7142 | u64 dirid = 0, last_dirid = 0; | |
7143 | unsigned long ptr; | |
7144 | u32 item_size; | |
7145 | u32 cur_offset = 0; | |
7146 | int ref_name_len; | |
7147 | int ret = 0; | |
7148 | ||
7149 | /* Easy case, just check this one dirid */ | |
7150 | if (key->type == BTRFS_INODE_REF_KEY) { | |
7151 | dirid = key->offset; | |
7152 | ||
7153 | ret = dir_changed(sctx, dirid); | |
7154 | goto out; | |
7155 | } | |
7156 | ||
7157 | leaf = path->nodes[0]; | |
3212fa14 | 7158 | item_size = btrfs_item_size(leaf, path->slots[0]); |
ba5e8f2e JB |
7159 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); |
7160 | while (cur_offset < item_size) { | |
7161 | extref = (struct btrfs_inode_extref *)(ptr + | |
7162 | cur_offset); | |
7163 | dirid = btrfs_inode_extref_parent(leaf, extref); | |
7164 | ref_name_len = btrfs_inode_extref_name_len(leaf, extref); | |
7165 | cur_offset += ref_name_len + sizeof(*extref); | |
7166 | if (dirid == last_dirid) | |
7167 | continue; | |
7168 | ret = dir_changed(sctx, dirid); | |
7169 | if (ret) | |
7170 | break; | |
7171 | last_dirid = dirid; | |
7172 | } | |
7173 | out: | |
7174 | return ret; | |
7175 | } | |
7176 | ||
766702ef AB |
7177 | /* |
7178 | * Updates compare related fields in sctx and simply forwards to the actual | |
7179 | * changed_xxx functions. | |
7180 | */ | |
ee8c494f | 7181 | static int changed_cb(struct btrfs_path *left_path, |
31db9f7c AB |
7182 | struct btrfs_path *right_path, |
7183 | struct btrfs_key *key, | |
7184 | enum btrfs_compare_tree_result result, | |
88980383 | 7185 | struct send_ctx *sctx) |
31db9f7c AB |
7186 | { |
7187 | int ret = 0; | |
31db9f7c | 7188 | |
d96b3424 FM |
7189 | /* |
7190 | * We can not hold the commit root semaphore here. This is because in | |
7191 | * the case of sending and receiving to the same filesystem, using a | |
7192 | * pipe, could result in a deadlock: | |
7193 | * | |
7194 | * 1) The task running send blocks on the pipe because it's full; | |
7195 | * | |
7196 | * 2) The task running receive, which is the only consumer of the pipe, | |
7197 | * is waiting for a transaction commit (for example due to a space | |
7198 | * reservation when doing a write or triggering a transaction commit | |
7199 | * when creating a subvolume); | |
7200 | * | |
7201 | * 3) The transaction is waiting to write lock the commit root semaphore, | |
7202 | * but can not acquire it since it's being held at 1). | |
7203 | * | |
7204 | * Down this call chain we write to the pipe through kernel_write(). | |
7205 | * The same type of problem can also happen when sending to a file that | |
7206 | * is stored in the same filesystem - when reserving space for a write | |
7207 | * into the file, we can trigger a transaction commit. | |
7208 | * | |
7209 | * Our caller has supplied us with clones of leaves from the send and | |
7210 | * parent roots, so we're safe here from a concurrent relocation and | |
7211 | * further reallocation of metadata extents while we are here. Below we | |
7212 | * also assert that the leaves are clones. | |
7213 | */ | |
7214 | lockdep_assert_not_held(&sctx->send_root->fs_info->commit_root_sem); | |
7215 | ||
7216 | /* | |
7217 | * We always have a send root, so left_path is never NULL. We will not | |
7218 | * have a leaf when we have reached the end of the send root but have | |
7219 | * not yet reached the end of the parent root. | |
7220 | */ | |
7221 | if (left_path->nodes[0]) | |
7222 | ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED, | |
7223 | &left_path->nodes[0]->bflags)); | |
7224 | /* | |
7225 | * When doing a full send we don't have a parent root, so right_path is | |
7226 | * NULL. When doing an incremental send, we may have reached the end of | |
7227 | * the parent root already, so we don't have a leaf at right_path. | |
7228 | */ | |
7229 | if (right_path && right_path->nodes[0]) | |
7230 | ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED, | |
7231 | &right_path->nodes[0]->bflags)); | |
7232 | ||
ba5e8f2e | 7233 | if (result == BTRFS_COMPARE_TREE_SAME) { |
16e7549f JB |
7234 | if (key->type == BTRFS_INODE_REF_KEY || |
7235 | key->type == BTRFS_INODE_EXTREF_KEY) { | |
7236 | ret = compare_refs(sctx, left_path, key); | |
7237 | if (!ret) | |
7238 | return 0; | |
7239 | if (ret < 0) | |
7240 | return ret; | |
7241 | } else if (key->type == BTRFS_EXTENT_DATA_KEY) { | |
7242 | return maybe_send_hole(sctx, left_path, key); | |
7243 | } else { | |
ba5e8f2e | 7244 | return 0; |
16e7549f | 7245 | } |
ba5e8f2e JB |
7246 | result = BTRFS_COMPARE_TREE_CHANGED; |
7247 | ret = 0; | |
7248 | } | |
7249 | ||
31db9f7c AB |
7250 | sctx->left_path = left_path; |
7251 | sctx->right_path = right_path; | |
7252 | sctx->cmp_key = key; | |
7253 | ||
7254 | ret = finish_inode_if_needed(sctx, 0); | |
7255 | if (ret < 0) | |
7256 | goto out; | |
7257 | ||
2981e225 AB |
7258 | /* Ignore non-FS objects */ |
7259 | if (key->objectid == BTRFS_FREE_INO_OBJECTID || | |
7260 | key->objectid == BTRFS_FREE_SPACE_OBJECTID) | |
7261 | goto out; | |
7262 | ||
46b2f459 | 7263 | if (key->type == BTRFS_INODE_ITEM_KEY) { |
31db9f7c | 7264 | ret = changed_inode(sctx, result); |
46b2f459 FM |
7265 | } else if (!sctx->ignore_cur_inode) { |
7266 | if (key->type == BTRFS_INODE_REF_KEY || | |
7267 | key->type == BTRFS_INODE_EXTREF_KEY) | |
7268 | ret = changed_ref(sctx, result); | |
7269 | else if (key->type == BTRFS_XATTR_ITEM_KEY) | |
7270 | ret = changed_xattr(sctx, result); | |
7271 | else if (key->type == BTRFS_EXTENT_DATA_KEY) | |
7272 | ret = changed_extent(sctx, result); | |
38622010 BB |
7273 | else if (key->type == BTRFS_VERITY_DESC_ITEM_KEY && |
7274 | key->offset == 0) | |
7275 | ret = changed_verity(sctx, result); | |
46b2f459 | 7276 | } |
31db9f7c AB |
7277 | |
7278 | out: | |
7279 | return ret; | |
7280 | } | |
7281 | ||
d96b3424 FM |
7282 | static int search_key_again(const struct send_ctx *sctx, |
7283 | struct btrfs_root *root, | |
7284 | struct btrfs_path *path, | |
7285 | const struct btrfs_key *key) | |
7286 | { | |
7287 | int ret; | |
7288 | ||
7289 | if (!path->need_commit_sem) | |
7290 | lockdep_assert_held_read(&root->fs_info->commit_root_sem); | |
7291 | ||
7292 | /* | |
7293 | * Roots used for send operations are readonly and no one can add, | |
7294 | * update or remove keys from them, so we should be able to find our | |
7295 | * key again. The only exception is deduplication, which can operate on | |
7296 | * readonly roots and add, update or remove keys to/from them - but at | |
7297 | * the moment we don't allow it to run in parallel with send. | |
7298 | */ | |
7299 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
7300 | ASSERT(ret <= 0); | |
7301 | if (ret > 0) { | |
7302 | btrfs_print_tree(path->nodes[path->lowest_level], false); | |
7303 | btrfs_err(root->fs_info, | |
7304 | "send: key (%llu %u %llu) not found in %s root %llu, lowest_level %d, slot %d", | |
7305 | key->objectid, key->type, key->offset, | |
7306 | (root == sctx->parent_root ? "parent" : "send"), | |
7307 | root->root_key.objectid, path->lowest_level, | |
7308 | path->slots[path->lowest_level]); | |
7309 | return -EUCLEAN; | |
7310 | } | |
7311 | ||
7312 | return ret; | |
7313 | } | |
7314 | ||
31db9f7c AB |
7315 | static int full_send_tree(struct send_ctx *sctx) |
7316 | { | |
7317 | int ret; | |
31db9f7c AB |
7318 | struct btrfs_root *send_root = sctx->send_root; |
7319 | struct btrfs_key key; | |
d96b3424 | 7320 | struct btrfs_fs_info *fs_info = send_root->fs_info; |
31db9f7c | 7321 | struct btrfs_path *path; |
31db9f7c AB |
7322 | |
7323 | path = alloc_path_for_send(); | |
7324 | if (!path) | |
7325 | return -ENOMEM; | |
ace75066 | 7326 | path->reada = READA_FORWARD_ALWAYS; |
31db9f7c | 7327 | |
31db9f7c AB |
7328 | key.objectid = BTRFS_FIRST_FREE_OBJECTID; |
7329 | key.type = BTRFS_INODE_ITEM_KEY; | |
7330 | key.offset = 0; | |
7331 | ||
d96b3424 FM |
7332 | down_read(&fs_info->commit_root_sem); |
7333 | sctx->last_reloc_trans = fs_info->last_reloc_trans; | |
7334 | up_read(&fs_info->commit_root_sem); | |
7335 | ||
31db9f7c AB |
7336 | ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); |
7337 | if (ret < 0) | |
7338 | goto out; | |
7339 | if (ret) | |
7340 | goto out_finish; | |
7341 | ||
7342 | while (1) { | |
d96b3424 | 7343 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); |
31db9f7c | 7344 | |
ca5d2ba1 | 7345 | ret = changed_cb(path, NULL, &key, |
ee8c494f | 7346 | BTRFS_COMPARE_TREE_NEW, sctx); |
31db9f7c AB |
7347 | if (ret < 0) |
7348 | goto out; | |
7349 | ||
d96b3424 FM |
7350 | down_read(&fs_info->commit_root_sem); |
7351 | if (fs_info->last_reloc_trans > sctx->last_reloc_trans) { | |
7352 | sctx->last_reloc_trans = fs_info->last_reloc_trans; | |
7353 | up_read(&fs_info->commit_root_sem); | |
7354 | /* | |
7355 | * A transaction used for relocating a block group was | |
7356 | * committed or is about to finish its commit. Release | |
7357 | * our path (leaf) and restart the search, so that we | |
7358 | * avoid operating on any file extent items that are | |
7359 | * stale, with a disk_bytenr that reflects a pre | |
7360 | * relocation value. This way we avoid as much as | |
7361 | * possible to fallback to regular writes when checking | |
7362 | * if we can clone file ranges. | |
7363 | */ | |
7364 | btrfs_release_path(path); | |
7365 | ret = search_key_again(sctx, send_root, path, &key); | |
7366 | if (ret < 0) | |
7367 | goto out; | |
7368 | } else { | |
7369 | up_read(&fs_info->commit_root_sem); | |
7370 | } | |
7371 | ||
31db9f7c AB |
7372 | ret = btrfs_next_item(send_root, path); |
7373 | if (ret < 0) | |
7374 | goto out; | |
7375 | if (ret) { | |
7376 | ret = 0; | |
7377 | break; | |
7378 | } | |
7379 | } | |
7380 | ||
7381 | out_finish: | |
7382 | ret = finish_inode_if_needed(sctx, 1); | |
7383 | ||
7384 | out: | |
7385 | btrfs_free_path(path); | |
31db9f7c AB |
7386 | return ret; |
7387 | } | |
7388 | ||
d96b3424 FM |
7389 | static int replace_node_with_clone(struct btrfs_path *path, int level) |
7390 | { | |
7391 | struct extent_buffer *clone; | |
7392 | ||
7393 | clone = btrfs_clone_extent_buffer(path->nodes[level]); | |
7394 | if (!clone) | |
7395 | return -ENOMEM; | |
7396 | ||
7397 | free_extent_buffer(path->nodes[level]); | |
7398 | path->nodes[level] = clone; | |
7399 | ||
7400 | return 0; | |
7401 | } | |
7402 | ||
2ce73c63 | 7403 | static int tree_move_down(struct btrfs_path *path, int *level, u64 reada_min_gen) |
18d0f5c6 DS |
7404 | { |
7405 | struct extent_buffer *eb; | |
2ce73c63 FM |
7406 | struct extent_buffer *parent = path->nodes[*level]; |
7407 | int slot = path->slots[*level]; | |
7408 | const int nritems = btrfs_header_nritems(parent); | |
7409 | u64 reada_max; | |
7410 | u64 reada_done = 0; | |
18d0f5c6 | 7411 | |
d96b3424 FM |
7412 | lockdep_assert_held_read(&parent->fs_info->commit_root_sem); |
7413 | ||
18d0f5c6 | 7414 | BUG_ON(*level == 0); |
2ce73c63 | 7415 | eb = btrfs_read_node_slot(parent, slot); |
18d0f5c6 DS |
7416 | if (IS_ERR(eb)) |
7417 | return PTR_ERR(eb); | |
7418 | ||
2ce73c63 FM |
7419 | /* |
7420 | * Trigger readahead for the next leaves we will process, so that it is | |
7421 | * very likely that when we need them they are already in memory and we | |
7422 | * will not block on disk IO. For nodes we only do readahead for one, | |
7423 | * since the time window between processing nodes is typically larger. | |
7424 | */ | |
7425 | reada_max = (*level == 1 ? SZ_128K : eb->fs_info->nodesize); | |
7426 | ||
7427 | for (slot++; slot < nritems && reada_done < reada_max; slot++) { | |
7428 | if (btrfs_node_ptr_generation(parent, slot) > reada_min_gen) { | |
7429 | btrfs_readahead_node_child(parent, slot); | |
7430 | reada_done += eb->fs_info->nodesize; | |
7431 | } | |
7432 | } | |
7433 | ||
18d0f5c6 DS |
7434 | path->nodes[*level - 1] = eb; |
7435 | path->slots[*level - 1] = 0; | |
7436 | (*level)--; | |
d96b3424 FM |
7437 | |
7438 | if (*level == 0) | |
7439 | return replace_node_with_clone(path, 0); | |
7440 | ||
18d0f5c6 DS |
7441 | return 0; |
7442 | } | |
7443 | ||
7444 | static int tree_move_next_or_upnext(struct btrfs_path *path, | |
7445 | int *level, int root_level) | |
7446 | { | |
7447 | int ret = 0; | |
7448 | int nritems; | |
7449 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
7450 | ||
7451 | path->slots[*level]++; | |
7452 | ||
7453 | while (path->slots[*level] >= nritems) { | |
d96b3424 FM |
7454 | if (*level == root_level) { |
7455 | path->slots[*level] = nritems - 1; | |
18d0f5c6 | 7456 | return -1; |
d96b3424 | 7457 | } |
18d0f5c6 DS |
7458 | |
7459 | /* move upnext */ | |
7460 | path->slots[*level] = 0; | |
7461 | free_extent_buffer(path->nodes[*level]); | |
7462 | path->nodes[*level] = NULL; | |
7463 | (*level)++; | |
7464 | path->slots[*level]++; | |
7465 | ||
7466 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
7467 | ret = 1; | |
7468 | } | |
7469 | return ret; | |
7470 | } | |
7471 | ||
7472 | /* | |
7473 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
7474 | * or down. | |
7475 | */ | |
7476 | static int tree_advance(struct btrfs_path *path, | |
7477 | int *level, int root_level, | |
7478 | int allow_down, | |
2ce73c63 FM |
7479 | struct btrfs_key *key, |
7480 | u64 reada_min_gen) | |
18d0f5c6 DS |
7481 | { |
7482 | int ret; | |
7483 | ||
7484 | if (*level == 0 || !allow_down) { | |
7485 | ret = tree_move_next_or_upnext(path, level, root_level); | |
7486 | } else { | |
2ce73c63 | 7487 | ret = tree_move_down(path, level, reada_min_gen); |
18d0f5c6 | 7488 | } |
d96b3424 FM |
7489 | |
7490 | /* | |
7491 | * Even if we have reached the end of a tree, ret is -1, update the key | |
7492 | * anyway, so that in case we need to restart due to a block group | |
7493 | * relocation, we can assert that the last key of the root node still | |
7494 | * exists in the tree. | |
7495 | */ | |
7496 | if (*level == 0) | |
7497 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
7498 | path->slots[*level]); | |
7499 | else | |
7500 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
7501 | path->slots[*level]); | |
7502 | ||
18d0f5c6 DS |
7503 | return ret; |
7504 | } | |
7505 | ||
7506 | static int tree_compare_item(struct btrfs_path *left_path, | |
7507 | struct btrfs_path *right_path, | |
7508 | char *tmp_buf) | |
7509 | { | |
7510 | int cmp; | |
7511 | int len1, len2; | |
7512 | unsigned long off1, off2; | |
7513 | ||
3212fa14 JB |
7514 | len1 = btrfs_item_size(left_path->nodes[0], left_path->slots[0]); |
7515 | len2 = btrfs_item_size(right_path->nodes[0], right_path->slots[0]); | |
18d0f5c6 DS |
7516 | if (len1 != len2) |
7517 | return 1; | |
7518 | ||
7519 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
7520 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
7521 | right_path->slots[0]); | |
7522 | ||
7523 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
7524 | ||
7525 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
7526 | if (cmp) | |
7527 | return 1; | |
7528 | return 0; | |
7529 | } | |
7530 | ||
d96b3424 FM |
7531 | /* |
7532 | * A transaction used for relocating a block group was committed or is about to | |
7533 | * finish its commit. Release our paths and restart the search, so that we are | |
7534 | * not using stale extent buffers: | |
7535 | * | |
7536 | * 1) For levels > 0, we are only holding references of extent buffers, without | |
7537 | * any locks on them, which does not prevent them from having been relocated | |
7538 | * and reallocated after the last time we released the commit root semaphore. | |
7539 | * The exception are the root nodes, for which we always have a clone, see | |
7540 | * the comment at btrfs_compare_trees(); | |
7541 | * | |
7542 | * 2) For leaves, level 0, we are holding copies (clones) of extent buffers, so | |
7543 | * we are safe from the concurrent relocation and reallocation. However they | |
7544 | * can have file extent items with a pre relocation disk_bytenr value, so we | |
7545 | * restart the start from the current commit roots and clone the new leaves so | |
7546 | * that we get the post relocation disk_bytenr values. Not doing so, could | |
7547 | * make us clone the wrong data in case there are new extents using the old | |
7548 | * disk_bytenr that happen to be shared. | |
7549 | */ | |
7550 | static int restart_after_relocation(struct btrfs_path *left_path, | |
7551 | struct btrfs_path *right_path, | |
7552 | const struct btrfs_key *left_key, | |
7553 | const struct btrfs_key *right_key, | |
7554 | int left_level, | |
7555 | int right_level, | |
7556 | const struct send_ctx *sctx) | |
7557 | { | |
7558 | int root_level; | |
7559 | int ret; | |
7560 | ||
7561 | lockdep_assert_held_read(&sctx->send_root->fs_info->commit_root_sem); | |
7562 | ||
7563 | btrfs_release_path(left_path); | |
7564 | btrfs_release_path(right_path); | |
7565 | ||
7566 | /* | |
7567 | * Since keys can not be added or removed to/from our roots because they | |
7568 | * are readonly and we do not allow deduplication to run in parallel | |
7569 | * (which can add, remove or change keys), the layout of the trees should | |
7570 | * not change. | |
7571 | */ | |
7572 | left_path->lowest_level = left_level; | |
7573 | ret = search_key_again(sctx, sctx->send_root, left_path, left_key); | |
7574 | if (ret < 0) | |
7575 | return ret; | |
7576 | ||
7577 | right_path->lowest_level = right_level; | |
7578 | ret = search_key_again(sctx, sctx->parent_root, right_path, right_key); | |
7579 | if (ret < 0) | |
7580 | return ret; | |
7581 | ||
7582 | /* | |
7583 | * If the lowest level nodes are leaves, clone them so that they can be | |
7584 | * safely used by changed_cb() while not under the protection of the | |
7585 | * commit root semaphore, even if relocation and reallocation happens in | |
7586 | * parallel. | |
7587 | */ | |
7588 | if (left_level == 0) { | |
7589 | ret = replace_node_with_clone(left_path, 0); | |
7590 | if (ret < 0) | |
7591 | return ret; | |
7592 | } | |
7593 | ||
7594 | if (right_level == 0) { | |
7595 | ret = replace_node_with_clone(right_path, 0); | |
7596 | if (ret < 0) | |
7597 | return ret; | |
7598 | } | |
7599 | ||
7600 | /* | |
7601 | * Now clone the root nodes (unless they happen to be the leaves we have | |
7602 | * already cloned). This is to protect against concurrent snapshotting of | |
7603 | * the send and parent roots (see the comment at btrfs_compare_trees()). | |
7604 | */ | |
7605 | root_level = btrfs_header_level(sctx->send_root->commit_root); | |
7606 | if (root_level > 0) { | |
7607 | ret = replace_node_with_clone(left_path, root_level); | |
7608 | if (ret < 0) | |
7609 | return ret; | |
7610 | } | |
7611 | ||
7612 | root_level = btrfs_header_level(sctx->parent_root->commit_root); | |
7613 | if (root_level > 0) { | |
7614 | ret = replace_node_with_clone(right_path, root_level); | |
7615 | if (ret < 0) | |
7616 | return ret; | |
7617 | } | |
7618 | ||
7619 | return 0; | |
7620 | } | |
7621 | ||
18d0f5c6 DS |
7622 | /* |
7623 | * This function compares two trees and calls the provided callback for | |
7624 | * every changed/new/deleted item it finds. | |
7625 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
7626 | * the compare pretty fast on snapshotted subvolumes. | |
7627 | * | |
7628 | * This currently works on commit roots only. As commit roots are read only, | |
7629 | * we don't do any locking. The commit roots are protected with transactions. | |
7630 | * Transactions are ended and rejoined when a commit is tried in between. | |
7631 | * | |
7632 | * This function checks for modifications done to the trees while comparing. | |
7633 | * If it detects a change, it aborts immediately. | |
7634 | */ | |
7635 | static int btrfs_compare_trees(struct btrfs_root *left_root, | |
88980383 | 7636 | struct btrfs_root *right_root, struct send_ctx *sctx) |
18d0f5c6 DS |
7637 | { |
7638 | struct btrfs_fs_info *fs_info = left_root->fs_info; | |
7639 | int ret; | |
7640 | int cmp; | |
7641 | struct btrfs_path *left_path = NULL; | |
7642 | struct btrfs_path *right_path = NULL; | |
7643 | struct btrfs_key left_key; | |
7644 | struct btrfs_key right_key; | |
7645 | char *tmp_buf = NULL; | |
7646 | int left_root_level; | |
7647 | int right_root_level; | |
7648 | int left_level; | |
7649 | int right_level; | |
d96b3424 FM |
7650 | int left_end_reached = 0; |
7651 | int right_end_reached = 0; | |
7652 | int advance_left = 0; | |
7653 | int advance_right = 0; | |
18d0f5c6 DS |
7654 | u64 left_blockptr; |
7655 | u64 right_blockptr; | |
7656 | u64 left_gen; | |
7657 | u64 right_gen; | |
2ce73c63 | 7658 | u64 reada_min_gen; |
18d0f5c6 DS |
7659 | |
7660 | left_path = btrfs_alloc_path(); | |
7661 | if (!left_path) { | |
7662 | ret = -ENOMEM; | |
7663 | goto out; | |
7664 | } | |
7665 | right_path = btrfs_alloc_path(); | |
7666 | if (!right_path) { | |
7667 | ret = -ENOMEM; | |
7668 | goto out; | |
7669 | } | |
7670 | ||
7671 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); | |
7672 | if (!tmp_buf) { | |
7673 | ret = -ENOMEM; | |
7674 | goto out; | |
7675 | } | |
7676 | ||
7677 | left_path->search_commit_root = 1; | |
7678 | left_path->skip_locking = 1; | |
7679 | right_path->search_commit_root = 1; | |
7680 | right_path->skip_locking = 1; | |
7681 | ||
7682 | /* | |
7683 | * Strategy: Go to the first items of both trees. Then do | |
7684 | * | |
7685 | * If both trees are at level 0 | |
7686 | * Compare keys of current items | |
7687 | * If left < right treat left item as new, advance left tree | |
7688 | * and repeat | |
7689 | * If left > right treat right item as deleted, advance right tree | |
7690 | * and repeat | |
7691 | * If left == right do deep compare of items, treat as changed if | |
7692 | * needed, advance both trees and repeat | |
7693 | * If both trees are at the same level but not at level 0 | |
7694 | * Compare keys of current nodes/leafs | |
7695 | * If left < right advance left tree and repeat | |
7696 | * If left > right advance right tree and repeat | |
7697 | * If left == right compare blockptrs of the next nodes/leafs | |
7698 | * If they match advance both trees but stay at the same level | |
7699 | * and repeat | |
7700 | * If they don't match advance both trees while allowing to go | |
7701 | * deeper and repeat | |
7702 | * If tree levels are different | |
7703 | * Advance the tree that needs it and repeat | |
7704 | * | |
7705 | * Advancing a tree means: | |
7706 | * If we are at level 0, try to go to the next slot. If that's not | |
7707 | * possible, go one level up and repeat. Stop when we found a level | |
7708 | * where we could go to the next slot. We may at this point be on a | |
7709 | * node or a leaf. | |
7710 | * | |
7711 | * If we are not at level 0 and not on shared tree blocks, go one | |
7712 | * level deeper. | |
7713 | * | |
7714 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
7715 | * the right if possible or go up and right. | |
7716 | */ | |
7717 | ||
7718 | down_read(&fs_info->commit_root_sem); | |
7719 | left_level = btrfs_header_level(left_root->commit_root); | |
7720 | left_root_level = left_level; | |
d96b3424 FM |
7721 | /* |
7722 | * We clone the root node of the send and parent roots to prevent races | |
7723 | * with snapshot creation of these roots. Snapshot creation COWs the | |
7724 | * root node of a tree, so after the transaction is committed the old | |
7725 | * extent can be reallocated while this send operation is still ongoing. | |
7726 | * So we clone them, under the commit root semaphore, to be race free. | |
7727 | */ | |
18d0f5c6 DS |
7728 | left_path->nodes[left_level] = |
7729 | btrfs_clone_extent_buffer(left_root->commit_root); | |
7730 | if (!left_path->nodes[left_level]) { | |
18d0f5c6 | 7731 | ret = -ENOMEM; |
d96b3424 | 7732 | goto out_unlock; |
18d0f5c6 DS |
7733 | } |
7734 | ||
7735 | right_level = btrfs_header_level(right_root->commit_root); | |
7736 | right_root_level = right_level; | |
7737 | right_path->nodes[right_level] = | |
7738 | btrfs_clone_extent_buffer(right_root->commit_root); | |
7739 | if (!right_path->nodes[right_level]) { | |
18d0f5c6 | 7740 | ret = -ENOMEM; |
d96b3424 | 7741 | goto out_unlock; |
18d0f5c6 | 7742 | } |
2ce73c63 FM |
7743 | /* |
7744 | * Our right root is the parent root, while the left root is the "send" | |
7745 | * root. We know that all new nodes/leaves in the left root must have | |
7746 | * a generation greater than the right root's generation, so we trigger | |
7747 | * readahead for those nodes and leaves of the left root, as we know we | |
7748 | * will need to read them at some point. | |
7749 | */ | |
7750 | reada_min_gen = btrfs_header_generation(right_root->commit_root); | |
18d0f5c6 DS |
7751 | |
7752 | if (left_level == 0) | |
7753 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
7754 | &left_key, left_path->slots[left_level]); | |
7755 | else | |
7756 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
7757 | &left_key, left_path->slots[left_level]); | |
7758 | if (right_level == 0) | |
7759 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
7760 | &right_key, right_path->slots[right_level]); | |
7761 | else | |
7762 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
7763 | &right_key, right_path->slots[right_level]); | |
7764 | ||
d96b3424 | 7765 | sctx->last_reloc_trans = fs_info->last_reloc_trans; |
18d0f5c6 DS |
7766 | |
7767 | while (1) { | |
d96b3424 FM |
7768 | if (need_resched() || |
7769 | rwsem_is_contended(&fs_info->commit_root_sem)) { | |
7770 | up_read(&fs_info->commit_root_sem); | |
7771 | cond_resched(); | |
7772 | down_read(&fs_info->commit_root_sem); | |
7773 | } | |
7774 | ||
7775 | if (fs_info->last_reloc_trans > sctx->last_reloc_trans) { | |
7776 | ret = restart_after_relocation(left_path, right_path, | |
7777 | &left_key, &right_key, | |
7778 | left_level, right_level, | |
7779 | sctx); | |
7780 | if (ret < 0) | |
7781 | goto out_unlock; | |
7782 | sctx->last_reloc_trans = fs_info->last_reloc_trans; | |
7783 | } | |
7784 | ||
18d0f5c6 DS |
7785 | if (advance_left && !left_end_reached) { |
7786 | ret = tree_advance(left_path, &left_level, | |
7787 | left_root_level, | |
7788 | advance_left != ADVANCE_ONLY_NEXT, | |
2ce73c63 | 7789 | &left_key, reada_min_gen); |
18d0f5c6 DS |
7790 | if (ret == -1) |
7791 | left_end_reached = ADVANCE; | |
7792 | else if (ret < 0) | |
d96b3424 | 7793 | goto out_unlock; |
18d0f5c6 DS |
7794 | advance_left = 0; |
7795 | } | |
7796 | if (advance_right && !right_end_reached) { | |
7797 | ret = tree_advance(right_path, &right_level, | |
7798 | right_root_level, | |
7799 | advance_right != ADVANCE_ONLY_NEXT, | |
2ce73c63 | 7800 | &right_key, reada_min_gen); |
18d0f5c6 DS |
7801 | if (ret == -1) |
7802 | right_end_reached = ADVANCE; | |
7803 | else if (ret < 0) | |
d96b3424 | 7804 | goto out_unlock; |
18d0f5c6 DS |
7805 | advance_right = 0; |
7806 | } | |
7807 | ||
7808 | if (left_end_reached && right_end_reached) { | |
7809 | ret = 0; | |
d96b3424 | 7810 | goto out_unlock; |
18d0f5c6 DS |
7811 | } else if (left_end_reached) { |
7812 | if (right_level == 0) { | |
d96b3424 | 7813 | up_read(&fs_info->commit_root_sem); |
18d0f5c6 DS |
7814 | ret = changed_cb(left_path, right_path, |
7815 | &right_key, | |
7816 | BTRFS_COMPARE_TREE_DELETED, | |
88980383 | 7817 | sctx); |
18d0f5c6 DS |
7818 | if (ret < 0) |
7819 | goto out; | |
d96b3424 | 7820 | down_read(&fs_info->commit_root_sem); |
18d0f5c6 DS |
7821 | } |
7822 | advance_right = ADVANCE; | |
7823 | continue; | |
7824 | } else if (right_end_reached) { | |
7825 | if (left_level == 0) { | |
d96b3424 | 7826 | up_read(&fs_info->commit_root_sem); |
18d0f5c6 DS |
7827 | ret = changed_cb(left_path, right_path, |
7828 | &left_key, | |
7829 | BTRFS_COMPARE_TREE_NEW, | |
88980383 | 7830 | sctx); |
18d0f5c6 DS |
7831 | if (ret < 0) |
7832 | goto out; | |
d96b3424 | 7833 | down_read(&fs_info->commit_root_sem); |
18d0f5c6 DS |
7834 | } |
7835 | advance_left = ADVANCE; | |
7836 | continue; | |
7837 | } | |
7838 | ||
7839 | if (left_level == 0 && right_level == 0) { | |
d96b3424 | 7840 | up_read(&fs_info->commit_root_sem); |
18d0f5c6 DS |
7841 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); |
7842 | if (cmp < 0) { | |
7843 | ret = changed_cb(left_path, right_path, | |
7844 | &left_key, | |
7845 | BTRFS_COMPARE_TREE_NEW, | |
88980383 | 7846 | sctx); |
18d0f5c6 DS |
7847 | advance_left = ADVANCE; |
7848 | } else if (cmp > 0) { | |
7849 | ret = changed_cb(left_path, right_path, | |
7850 | &right_key, | |
7851 | BTRFS_COMPARE_TREE_DELETED, | |
88980383 | 7852 | sctx); |
18d0f5c6 DS |
7853 | advance_right = ADVANCE; |
7854 | } else { | |
7855 | enum btrfs_compare_tree_result result; | |
7856 | ||
7857 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); | |
7858 | ret = tree_compare_item(left_path, right_path, | |
7859 | tmp_buf); | |
7860 | if (ret) | |
7861 | result = BTRFS_COMPARE_TREE_CHANGED; | |
7862 | else | |
7863 | result = BTRFS_COMPARE_TREE_SAME; | |
7864 | ret = changed_cb(left_path, right_path, | |
88980383 | 7865 | &left_key, result, sctx); |
18d0f5c6 DS |
7866 | advance_left = ADVANCE; |
7867 | advance_right = ADVANCE; | |
7868 | } | |
d96b3424 FM |
7869 | |
7870 | if (ret < 0) | |
7871 | goto out; | |
7872 | down_read(&fs_info->commit_root_sem); | |
18d0f5c6 DS |
7873 | } else if (left_level == right_level) { |
7874 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
7875 | if (cmp < 0) { | |
7876 | advance_left = ADVANCE; | |
7877 | } else if (cmp > 0) { | |
7878 | advance_right = ADVANCE; | |
7879 | } else { | |
7880 | left_blockptr = btrfs_node_blockptr( | |
7881 | left_path->nodes[left_level], | |
7882 | left_path->slots[left_level]); | |
7883 | right_blockptr = btrfs_node_blockptr( | |
7884 | right_path->nodes[right_level], | |
7885 | right_path->slots[right_level]); | |
7886 | left_gen = btrfs_node_ptr_generation( | |
7887 | left_path->nodes[left_level], | |
7888 | left_path->slots[left_level]); | |
7889 | right_gen = btrfs_node_ptr_generation( | |
7890 | right_path->nodes[right_level], | |
7891 | right_path->slots[right_level]); | |
7892 | if (left_blockptr == right_blockptr && | |
7893 | left_gen == right_gen) { | |
7894 | /* | |
7895 | * As we're on a shared block, don't | |
7896 | * allow to go deeper. | |
7897 | */ | |
7898 | advance_left = ADVANCE_ONLY_NEXT; | |
7899 | advance_right = ADVANCE_ONLY_NEXT; | |
7900 | } else { | |
7901 | advance_left = ADVANCE; | |
7902 | advance_right = ADVANCE; | |
7903 | } | |
7904 | } | |
7905 | } else if (left_level < right_level) { | |
7906 | advance_right = ADVANCE; | |
7907 | } else { | |
7908 | advance_left = ADVANCE; | |
7909 | } | |
7910 | } | |
7911 | ||
d96b3424 FM |
7912 | out_unlock: |
7913 | up_read(&fs_info->commit_root_sem); | |
18d0f5c6 DS |
7914 | out: |
7915 | btrfs_free_path(left_path); | |
7916 | btrfs_free_path(right_path); | |
7917 | kvfree(tmp_buf); | |
7918 | return ret; | |
7919 | } | |
7920 | ||
31db9f7c AB |
7921 | static int send_subvol(struct send_ctx *sctx) |
7922 | { | |
7923 | int ret; | |
7924 | ||
c2c71324 SB |
7925 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) { |
7926 | ret = send_header(sctx); | |
7927 | if (ret < 0) | |
7928 | goto out; | |
7929 | } | |
31db9f7c AB |
7930 | |
7931 | ret = send_subvol_begin(sctx); | |
7932 | if (ret < 0) | |
7933 | goto out; | |
7934 | ||
7935 | if (sctx->parent_root) { | |
1b51d6fc | 7936 | ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, sctx); |
31db9f7c AB |
7937 | if (ret < 0) |
7938 | goto out; | |
7939 | ret = finish_inode_if_needed(sctx, 1); | |
7940 | if (ret < 0) | |
7941 | goto out; | |
7942 | } else { | |
7943 | ret = full_send_tree(sctx); | |
7944 | if (ret < 0) | |
7945 | goto out; | |
7946 | } | |
7947 | ||
7948 | out: | |
31db9f7c AB |
7949 | free_recorded_refs(sctx); |
7950 | return ret; | |
7951 | } | |
7952 | ||
e5fa8f86 FM |
7953 | /* |
7954 | * If orphan cleanup did remove any orphans from a root, it means the tree | |
7955 | * was modified and therefore the commit root is not the same as the current | |
7956 | * root anymore. This is a problem, because send uses the commit root and | |
7957 | * therefore can see inode items that don't exist in the current root anymore, | |
7958 | * and for example make calls to btrfs_iget, which will do tree lookups based | |
7959 | * on the current root and not on the commit root. Those lookups will fail, | |
7960 | * returning a -ESTALE error, and making send fail with that error. So make | |
7961 | * sure a send does not see any orphans we have just removed, and that it will | |
7962 | * see the same inodes regardless of whether a transaction commit happened | |
7963 | * before it started (meaning that the commit root will be the same as the | |
7964 | * current root) or not. | |
7965 | */ | |
7966 | static int ensure_commit_roots_uptodate(struct send_ctx *sctx) | |
7967 | { | |
7968 | int i; | |
7969 | struct btrfs_trans_handle *trans = NULL; | |
7970 | ||
7971 | again: | |
7972 | if (sctx->parent_root && | |
7973 | sctx->parent_root->node != sctx->parent_root->commit_root) | |
7974 | goto commit_trans; | |
7975 | ||
7976 | for (i = 0; i < sctx->clone_roots_cnt; i++) | |
7977 | if (sctx->clone_roots[i].root->node != | |
7978 | sctx->clone_roots[i].root->commit_root) | |
7979 | goto commit_trans; | |
7980 | ||
7981 | if (trans) | |
3a45bb20 | 7982 | return btrfs_end_transaction(trans); |
e5fa8f86 FM |
7983 | |
7984 | return 0; | |
7985 | ||
7986 | commit_trans: | |
7987 | /* Use any root, all fs roots will get their commit roots updated. */ | |
7988 | if (!trans) { | |
7989 | trans = btrfs_join_transaction(sctx->send_root); | |
7990 | if (IS_ERR(trans)) | |
7991 | return PTR_ERR(trans); | |
7992 | goto again; | |
7993 | } | |
7994 | ||
3a45bb20 | 7995 | return btrfs_commit_transaction(trans); |
e5fa8f86 FM |
7996 | } |
7997 | ||
9f89d5de FM |
7998 | /* |
7999 | * Make sure any existing dellaloc is flushed for any root used by a send | |
8000 | * operation so that we do not miss any data and we do not race with writeback | |
8001 | * finishing and changing a tree while send is using the tree. This could | |
8002 | * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and | |
8003 | * a send operation then uses the subvolume. | |
8004 | * After flushing delalloc ensure_commit_roots_uptodate() must be called. | |
8005 | */ | |
8006 | static int flush_delalloc_roots(struct send_ctx *sctx) | |
8007 | { | |
8008 | struct btrfs_root *root = sctx->parent_root; | |
8009 | int ret; | |
8010 | int i; | |
8011 | ||
8012 | if (root) { | |
f9baa501 | 8013 | ret = btrfs_start_delalloc_snapshot(root, false); |
9f89d5de FM |
8014 | if (ret) |
8015 | return ret; | |
8016 | btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX); | |
8017 | } | |
8018 | ||
8019 | for (i = 0; i < sctx->clone_roots_cnt; i++) { | |
8020 | root = sctx->clone_roots[i].root; | |
f9baa501 | 8021 | ret = btrfs_start_delalloc_snapshot(root, false); |
9f89d5de FM |
8022 | if (ret) |
8023 | return ret; | |
8024 | btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX); | |
8025 | } | |
8026 | ||
8027 | return 0; | |
8028 | } | |
8029 | ||
66ef7d65 DS |
8030 | static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) |
8031 | { | |
8032 | spin_lock(&root->root_item_lock); | |
8033 | root->send_in_progress--; | |
8034 | /* | |
8035 | * Not much left to do, we don't know why it's unbalanced and | |
8036 | * can't blindly reset it to 0. | |
8037 | */ | |
8038 | if (root->send_in_progress < 0) | |
8039 | btrfs_err(root->fs_info, | |
f5686e3a | 8040 | "send_in_progress unbalanced %d root %llu", |
0b246afa | 8041 | root->send_in_progress, root->root_key.objectid); |
66ef7d65 DS |
8042 | spin_unlock(&root->root_item_lock); |
8043 | } | |
8044 | ||
62d54f3a FM |
8045 | static void dedupe_in_progress_warn(const struct btrfs_root *root) |
8046 | { | |
8047 | btrfs_warn_rl(root->fs_info, | |
8048 | "cannot use root %llu for send while deduplications on it are in progress (%d in progress)", | |
8049 | root->root_key.objectid, root->dedupe_in_progress); | |
8050 | } | |
8051 | ||
9ad12305 | 8052 | long btrfs_ioctl_send(struct inode *inode, struct btrfs_ioctl_send_args *arg) |
31db9f7c AB |
8053 | { |
8054 | int ret = 0; | |
9ad12305 | 8055 | struct btrfs_root *send_root = BTRFS_I(inode)->root; |
0b246afa | 8056 | struct btrfs_fs_info *fs_info = send_root->fs_info; |
31db9f7c | 8057 | struct btrfs_root *clone_root; |
31db9f7c AB |
8058 | struct send_ctx *sctx = NULL; |
8059 | u32 i; | |
8060 | u64 *clone_sources_tmp = NULL; | |
2c686537 | 8061 | int clone_sources_to_rollback = 0; |
bae12df9 | 8062 | size_t alloc_size; |
896c14f9 | 8063 | int sort_clone_roots = 0; |
3e49363b FM |
8064 | struct btrfs_lru_cache_entry *entry; |
8065 | struct btrfs_lru_cache_entry *tmp; | |
31db9f7c AB |
8066 | |
8067 | if (!capable(CAP_SYS_ADMIN)) | |
8068 | return -EPERM; | |
8069 | ||
2c686537 DS |
8070 | /* |
8071 | * The subvolume must remain read-only during send, protect against | |
521e0546 | 8072 | * making it RW. This also protects against deletion. |
2c686537 DS |
8073 | */ |
8074 | spin_lock(&send_root->root_item_lock); | |
62d54f3a FM |
8075 | if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) { |
8076 | dedupe_in_progress_warn(send_root); | |
8077 | spin_unlock(&send_root->root_item_lock); | |
8078 | return -EAGAIN; | |
8079 | } | |
2c686537 DS |
8080 | send_root->send_in_progress++; |
8081 | spin_unlock(&send_root->root_item_lock); | |
8082 | ||
2c686537 DS |
8083 | /* |
8084 | * Userspace tools do the checks and warn the user if it's | |
8085 | * not RO. | |
8086 | */ | |
8087 | if (!btrfs_root_readonly(send_root)) { | |
8088 | ret = -EPERM; | |
8089 | goto out; | |
8090 | } | |
8091 | ||
457ae726 DC |
8092 | /* |
8093 | * Check that we don't overflow at later allocations, we request | |
8094 | * clone_sources_count + 1 items, and compare to unsigned long inside | |
33e17b3f DS |
8095 | * access_ok. Also set an upper limit for allocation size so this can't |
8096 | * easily exhaust memory. Max number of clone sources is about 200K. | |
457ae726 | 8097 | */ |
33e17b3f | 8098 | if (arg->clone_sources_count > SZ_8M / sizeof(struct clone_root)) { |
f5ecec3c DC |
8099 | ret = -EINVAL; |
8100 | goto out; | |
8101 | } | |
8102 | ||
c2c71324 | 8103 | if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { |
cb95e7bf MF |
8104 | ret = -EINVAL; |
8105 | goto out; | |
8106 | } | |
8107 | ||
e780b0d1 | 8108 | sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL); |
31db9f7c AB |
8109 | if (!sctx) { |
8110 | ret = -ENOMEM; | |
8111 | goto out; | |
8112 | } | |
8113 | ||
8114 | INIT_LIST_HEAD(&sctx->new_refs); | |
8115 | INIT_LIST_HEAD(&sctx->deleted_refs); | |
31db9f7c | 8116 | |
c48545de | 8117 | btrfs_lru_cache_init(&sctx->name_cache, SEND_MAX_NAME_CACHE_SIZE); |
90b90d4a | 8118 | btrfs_lru_cache_init(&sctx->backref_cache, SEND_MAX_BACKREF_CACHE_SIZE); |
e8a7f49d FM |
8119 | btrfs_lru_cache_init(&sctx->dir_created_cache, |
8120 | SEND_MAX_DIR_CREATED_CACHE_SIZE); | |
3e49363b FM |
8121 | /* |
8122 | * This cache is periodically trimmed to a fixed size elsewhere, see | |
8123 | * cache_dir_utimes() and trim_dir_utimes_cache(). | |
8124 | */ | |
8125 | btrfs_lru_cache_init(&sctx->dir_utimes_cache, 0); | |
66d04209 | 8126 | |
d307d2f3 FM |
8127 | sctx->pending_dir_moves = RB_ROOT; |
8128 | sctx->waiting_dir_moves = RB_ROOT; | |
8129 | sctx->orphan_dirs = RB_ROOT; | |
8130 | sctx->rbtree_new_refs = RB_ROOT; | |
8131 | sctx->rbtree_deleted_refs = RB_ROOT; | |
8132 | ||
cb95e7bf MF |
8133 | sctx->flags = arg->flags; |
8134 | ||
e77fbf99 DS |
8135 | if (arg->flags & BTRFS_SEND_FLAG_VERSION) { |
8136 | if (arg->version > BTRFS_SEND_STREAM_VERSION) { | |
8137 | ret = -EPROTO; | |
8138 | goto out; | |
8139 | } | |
8140 | /* Zero means "use the highest version" */ | |
8141 | sctx->proto = arg->version ?: BTRFS_SEND_STREAM_VERSION; | |
8142 | } else { | |
8143 | sctx->proto = 1; | |
8144 | } | |
d6815592 OS |
8145 | if ((arg->flags & BTRFS_SEND_FLAG_COMPRESSED) && sctx->proto < 2) { |
8146 | ret = -EINVAL; | |
8147 | goto out; | |
8148 | } | |
e77fbf99 | 8149 | |
31db9f7c | 8150 | sctx->send_filp = fget(arg->send_fd); |
ecc7ada7 TI |
8151 | if (!sctx->send_filp) { |
8152 | ret = -EBADF; | |
31db9f7c AB |
8153 | goto out; |
8154 | } | |
8155 | ||
31db9f7c | 8156 | sctx->send_root = send_root; |
521e0546 DS |
8157 | /* |
8158 | * Unlikely but possible, if the subvolume is marked for deletion but | |
8159 | * is slow to remove the directory entry, send can still be started | |
8160 | */ | |
8161 | if (btrfs_root_dead(sctx->send_root)) { | |
8162 | ret = -EPERM; | |
8163 | goto out; | |
8164 | } | |
8165 | ||
31db9f7c AB |
8166 | sctx->clone_roots_cnt = arg->clone_sources_count; |
8167 | ||
a4b333f2 OS |
8168 | if (sctx->proto >= 2) { |
8169 | u32 send_buf_num_pages; | |
8170 | ||
875c627c | 8171 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE_V2; |
a4b333f2 OS |
8172 | sctx->send_buf = vmalloc(sctx->send_max_size); |
8173 | if (!sctx->send_buf) { | |
8174 | ret = -ENOMEM; | |
8175 | goto out; | |
8176 | } | |
8177 | send_buf_num_pages = sctx->send_max_size >> PAGE_SHIFT; | |
8178 | sctx->send_buf_pages = kcalloc(send_buf_num_pages, | |
8179 | sizeof(*sctx->send_buf_pages), | |
8180 | GFP_KERNEL); | |
8181 | if (!sctx->send_buf_pages) { | |
8182 | ret = -ENOMEM; | |
8183 | goto out; | |
8184 | } | |
8185 | for (i = 0; i < send_buf_num_pages; i++) { | |
8186 | sctx->send_buf_pages[i] = | |
8187 | vmalloc_to_page(sctx->send_buf + (i << PAGE_SHIFT)); | |
8188 | } | |
8189 | } else { | |
356bbbb6 | 8190 | sctx->send_max_size = BTRFS_SEND_BUF_SIZE_V1; |
a4b333f2 OS |
8191 | sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); |
8192 | } | |
31db9f7c | 8193 | if (!sctx->send_buf) { |
752ade68 MH |
8194 | ret = -ENOMEM; |
8195 | goto out; | |
31db9f7c AB |
8196 | } |
8197 | ||
bae12df9 DE |
8198 | sctx->clone_roots = kvcalloc(sizeof(*sctx->clone_roots), |
8199 | arg->clone_sources_count + 1, | |
8200 | GFP_KERNEL); | |
31db9f7c | 8201 | if (!sctx->clone_roots) { |
818e010b DS |
8202 | ret = -ENOMEM; |
8203 | goto out; | |
31db9f7c AB |
8204 | } |
8205 | ||
bae12df9 DE |
8206 | alloc_size = array_size(sizeof(*arg->clone_sources), |
8207 | arg->clone_sources_count); | |
e55d1153 | 8208 | |
31db9f7c | 8209 | if (arg->clone_sources_count) { |
752ade68 | 8210 | clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL); |
31db9f7c | 8211 | if (!clone_sources_tmp) { |
752ade68 MH |
8212 | ret = -ENOMEM; |
8213 | goto out; | |
31db9f7c AB |
8214 | } |
8215 | ||
8216 | ret = copy_from_user(clone_sources_tmp, arg->clone_sources, | |
e55d1153 | 8217 | alloc_size); |
31db9f7c AB |
8218 | if (ret) { |
8219 | ret = -EFAULT; | |
8220 | goto out; | |
8221 | } | |
8222 | ||
8223 | for (i = 0; i < arg->clone_sources_count; i++) { | |
56e9357a DS |
8224 | clone_root = btrfs_get_fs_root(fs_info, |
8225 | clone_sources_tmp[i], true); | |
31db9f7c AB |
8226 | if (IS_ERR(clone_root)) { |
8227 | ret = PTR_ERR(clone_root); | |
8228 | goto out; | |
8229 | } | |
2c686537 | 8230 | spin_lock(&clone_root->root_item_lock); |
5cc2b17e FM |
8231 | if (!btrfs_root_readonly(clone_root) || |
8232 | btrfs_root_dead(clone_root)) { | |
2c686537 | 8233 | spin_unlock(&clone_root->root_item_lock); |
00246528 | 8234 | btrfs_put_root(clone_root); |
2c686537 DS |
8235 | ret = -EPERM; |
8236 | goto out; | |
8237 | } | |
62d54f3a FM |
8238 | if (clone_root->dedupe_in_progress) { |
8239 | dedupe_in_progress_warn(clone_root); | |
8240 | spin_unlock(&clone_root->root_item_lock); | |
00246528 | 8241 | btrfs_put_root(clone_root); |
62d54f3a FM |
8242 | ret = -EAGAIN; |
8243 | goto out; | |
8244 | } | |
2f1f465a | 8245 | clone_root->send_in_progress++; |
2c686537 | 8246 | spin_unlock(&clone_root->root_item_lock); |
18f687d5 | 8247 | |
31db9f7c | 8248 | sctx->clone_roots[i].root = clone_root; |
2f1f465a | 8249 | clone_sources_to_rollback = i + 1; |
31db9f7c | 8250 | } |
2f91306a | 8251 | kvfree(clone_sources_tmp); |
31db9f7c AB |
8252 | clone_sources_tmp = NULL; |
8253 | } | |
8254 | ||
8255 | if (arg->parent_root) { | |
56e9357a DS |
8256 | sctx->parent_root = btrfs_get_fs_root(fs_info, arg->parent_root, |
8257 | true); | |
b1b19596 SB |
8258 | if (IS_ERR(sctx->parent_root)) { |
8259 | ret = PTR_ERR(sctx->parent_root); | |
31db9f7c AB |
8260 | goto out; |
8261 | } | |
18f687d5 | 8262 | |
2c686537 DS |
8263 | spin_lock(&sctx->parent_root->root_item_lock); |
8264 | sctx->parent_root->send_in_progress++; | |
521e0546 DS |
8265 | if (!btrfs_root_readonly(sctx->parent_root) || |
8266 | btrfs_root_dead(sctx->parent_root)) { | |
2c686537 DS |
8267 | spin_unlock(&sctx->parent_root->root_item_lock); |
8268 | ret = -EPERM; | |
8269 | goto out; | |
8270 | } | |
62d54f3a FM |
8271 | if (sctx->parent_root->dedupe_in_progress) { |
8272 | dedupe_in_progress_warn(sctx->parent_root); | |
8273 | spin_unlock(&sctx->parent_root->root_item_lock); | |
62d54f3a FM |
8274 | ret = -EAGAIN; |
8275 | goto out; | |
8276 | } | |
2c686537 | 8277 | spin_unlock(&sctx->parent_root->root_item_lock); |
31db9f7c AB |
8278 | } |
8279 | ||
8280 | /* | |
8281 | * Clones from send_root are allowed, but only if the clone source | |
8282 | * is behind the current send position. This is checked while searching | |
8283 | * for possible clone sources. | |
8284 | */ | |
6f9a3da5 | 8285 | sctx->clone_roots[sctx->clone_roots_cnt++].root = |
00246528 | 8286 | btrfs_grab_root(sctx->send_root); |
31db9f7c AB |
8287 | |
8288 | /* We do a bsearch later */ | |
8289 | sort(sctx->clone_roots, sctx->clone_roots_cnt, | |
8290 | sizeof(*sctx->clone_roots), __clone_root_cmp_sort, | |
8291 | NULL); | |
896c14f9 | 8292 | sort_clone_roots = 1; |
31db9f7c | 8293 | |
9f89d5de FM |
8294 | ret = flush_delalloc_roots(sctx); |
8295 | if (ret) | |
8296 | goto out; | |
8297 | ||
e5fa8f86 FM |
8298 | ret = ensure_commit_roots_uptodate(sctx); |
8299 | if (ret) | |
8300 | goto out; | |
8301 | ||
31db9f7c AB |
8302 | ret = send_subvol(sctx); |
8303 | if (ret < 0) | |
8304 | goto out; | |
8305 | ||
3e49363b FM |
8306 | btrfs_lru_cache_for_each_entry_safe(&sctx->dir_utimes_cache, entry, tmp) { |
8307 | ret = send_utimes(sctx, entry->key, entry->gen); | |
8308 | if (ret < 0) | |
8309 | goto out; | |
8310 | btrfs_lru_cache_remove(&sctx->dir_utimes_cache, entry); | |
8311 | } | |
8312 | ||
c2c71324 SB |
8313 | if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { |
8314 | ret = begin_cmd(sctx, BTRFS_SEND_C_END); | |
8315 | if (ret < 0) | |
8316 | goto out; | |
8317 | ret = send_cmd(sctx); | |
8318 | if (ret < 0) | |
8319 | goto out; | |
8320 | } | |
31db9f7c AB |
8321 | |
8322 | out: | |
9f03740a FDBM |
8323 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)); |
8324 | while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) { | |
8325 | struct rb_node *n; | |
8326 | struct pending_dir_move *pm; | |
8327 | ||
8328 | n = rb_first(&sctx->pending_dir_moves); | |
8329 | pm = rb_entry(n, struct pending_dir_move, node); | |
8330 | while (!list_empty(&pm->list)) { | |
8331 | struct pending_dir_move *pm2; | |
8332 | ||
8333 | pm2 = list_first_entry(&pm->list, | |
8334 | struct pending_dir_move, list); | |
8335 | free_pending_move(sctx, pm2); | |
8336 | } | |
8337 | free_pending_move(sctx, pm); | |
8338 | } | |
8339 | ||
8340 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)); | |
8341 | while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) { | |
8342 | struct rb_node *n; | |
8343 | struct waiting_dir_move *dm; | |
8344 | ||
8345 | n = rb_first(&sctx->waiting_dir_moves); | |
8346 | dm = rb_entry(n, struct waiting_dir_move, node); | |
8347 | rb_erase(&dm->node, &sctx->waiting_dir_moves); | |
8348 | kfree(dm); | |
8349 | } | |
8350 | ||
9dc44214 FM |
8351 | WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs)); |
8352 | while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) { | |
8353 | struct rb_node *n; | |
8354 | struct orphan_dir_info *odi; | |
8355 | ||
8356 | n = rb_first(&sctx->orphan_dirs); | |
8357 | odi = rb_entry(n, struct orphan_dir_info, node); | |
8358 | free_orphan_dir_info(sctx, odi); | |
8359 | } | |
8360 | ||
896c14f9 | 8361 | if (sort_clone_roots) { |
6f9a3da5 | 8362 | for (i = 0; i < sctx->clone_roots_cnt; i++) { |
896c14f9 WS |
8363 | btrfs_root_dec_send_in_progress( |
8364 | sctx->clone_roots[i].root); | |
00246528 | 8365 | btrfs_put_root(sctx->clone_roots[i].root); |
6f9a3da5 | 8366 | } |
896c14f9 | 8367 | } else { |
6f9a3da5 | 8368 | for (i = 0; sctx && i < clone_sources_to_rollback; i++) { |
896c14f9 WS |
8369 | btrfs_root_dec_send_in_progress( |
8370 | sctx->clone_roots[i].root); | |
00246528 | 8371 | btrfs_put_root(sctx->clone_roots[i].root); |
6f9a3da5 | 8372 | } |
896c14f9 WS |
8373 | |
8374 | btrfs_root_dec_send_in_progress(send_root); | |
8375 | } | |
6f9a3da5 | 8376 | if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) { |
66ef7d65 | 8377 | btrfs_root_dec_send_in_progress(sctx->parent_root); |
00246528 | 8378 | btrfs_put_root(sctx->parent_root); |
6f9a3da5 | 8379 | } |
2c686537 | 8380 | |
2f91306a | 8381 | kvfree(clone_sources_tmp); |
31db9f7c AB |
8382 | |
8383 | if (sctx) { | |
8384 | if (sctx->send_filp) | |
8385 | fput(sctx->send_filp); | |
8386 | ||
c03d01f3 | 8387 | kvfree(sctx->clone_roots); |
a4b333f2 | 8388 | kfree(sctx->send_buf_pages); |
6ff48ce0 | 8389 | kvfree(sctx->send_buf); |
38622010 | 8390 | kvfree(sctx->verity_descriptor); |
31db9f7c | 8391 | |
152555b3 | 8392 | close_current_inode(sctx); |
521b6803 | 8393 | |
c48545de | 8394 | btrfs_lru_cache_clear(&sctx->name_cache); |
90b90d4a | 8395 | btrfs_lru_cache_clear(&sctx->backref_cache); |
e8a7f49d | 8396 | btrfs_lru_cache_clear(&sctx->dir_created_cache); |
3e49363b | 8397 | btrfs_lru_cache_clear(&sctx->dir_utimes_cache); |
66d04209 | 8398 | |
31db9f7c AB |
8399 | kfree(sctx); |
8400 | } | |
8401 | ||
8402 | return ret; | |
8403 | } |