Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (C) 2012 Alexander Block. All rights reserved.
4 : */
5 :
6 : #include <linux/bsearch.h>
7 : #include <linux/fs.h>
8 : #include <linux/file.h>
9 : #include <linux/sort.h>
10 : #include <linux/mount.h>
11 : #include <linux/xattr.h>
12 : #include <linux/posix_acl_xattr.h>
13 : #include <linux/radix-tree.h>
14 : #include <linux/vmalloc.h>
15 : #include <linux/string.h>
16 : #include <linux/compat.h>
17 : #include <linux/crc32c.h>
18 : #include <linux/fsverity.h>
19 :
20 : #include "send.h"
21 : #include "ctree.h"
22 : #include "backref.h"
23 : #include "locking.h"
24 : #include "disk-io.h"
25 : #include "btrfs_inode.h"
26 : #include "transaction.h"
27 : #include "compression.h"
28 : #include "xattr.h"
29 : #include "print-tree.h"
30 : #include "accessors.h"
31 : #include "dir-item.h"
32 : #include "file-item.h"
33 : #include "ioctl.h"
34 : #include "verity.h"
35 : #include "lru_cache.h"
36 :
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 : */
43 : #define SEND_MAX_EXTENT_REFS 1024
44 :
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;
57 :
58 : char *buf;
59 : unsigned short buf_len:15;
60 : unsigned short reversed:1;
61 : char inline_buf[];
62 : };
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];
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;
80 : u64 num_bytes;
81 : bool found_ref;
82 : };
83 :
84 : #define SEND_MAX_NAME_CACHE_SIZE 256
85 :
86 : /*
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.
90 : * The most common case is to have a single root for cloning, which corresponds
91 : * to the send root. Having the user specify more than 16 clone roots is not
92 : * common, and in such rare cases we simply don't use caching if the number of
93 : * cloning roots that lead down to a leaf is more than 17.
94 : */
95 : #define SEND_MAX_BACKREF_CACHE_ROOTS 17
96 :
97 : /*
98 : * Max number of entries in the cache.
99 : * With SEND_MAX_BACKREF_CACHE_ROOTS as 17, the size in bytes, excluding
100 : * maple tree's internal nodes, is 24K.
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 {
111 : struct btrfs_lru_cache_entry entry;
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 :
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 :
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
123 : * at most 4096 bytes - sizeof(struct btrfs_lru_cache_entry) is 48 bytes, but
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 :
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 :
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;
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;
147 : struct page **send_buf_pages;
148 : u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
149 : /* Protocol version compatibility requested */
150 : u32 proto;
151 :
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 :
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 :
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;
177 : u64 cur_inode_size;
178 : u64 cur_inode_mode;
179 : u64 cur_inode_rdev;
180 : u64 cur_inode_last_extent;
181 : u64 cur_inode_next_write_offset;
182 : bool cur_inode_new;
183 : bool cur_inode_new_gen;
184 : bool cur_inode_deleted;
185 : bool ignore_cur_inode;
186 : bool cur_inode_needs_verity;
187 : void *verity_descriptor;
188 :
189 : u64 send_progress;
190 :
191 : struct list_head new_refs;
192 : struct list_head deleted_refs;
193 :
194 : struct btrfs_lru_cache name_cache;
195 :
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;
201 : struct file_ra_state ra;
202 : u64 page_cache_clear_start;
203 : bool clean_page_cache;
204 :
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;
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;
299 :
300 : struct rb_root rbtree_new_refs;
301 : struct rb_root rbtree_deleted_refs;
302 :
303 : struct btrfs_lru_cache backref_cache;
304 : u64 backref_cache_last_reloc_trans;
305 :
306 : struct btrfs_lru_cache dir_created_cache;
307 : struct btrfs_lru_cache dir_utimes_cache;
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;
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;
328 : u64 rmdir_gen;
329 : bool orphanized;
330 : };
331 :
332 : struct orphan_dir_info {
333 : struct rb_node node;
334 : u64 ino;
335 : u64 gen;
336 : u64 last_dir_index_offset;
337 : u64 dir_high_seq_ino;
338 : };
339 :
340 : struct name_cache_entry {
341 : /*
342 : * The key in the entry is an inode number, and the generation matches
343 : * the inode's generation.
344 : */
345 : struct btrfs_lru_cache_entry entry;
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 :
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 :
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 : };
366 :
367 : __cold
368 0 : static void inconsistent_snapshot_error(struct send_ctx *sctx,
369 : enum btrfs_compare_tree_result result,
370 : const char *what)
371 : {
372 0 : const char *result_string;
373 :
374 0 : 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 0 : 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 0 : }
400 :
401 : __maybe_unused
402 : static bool proto_cmd_ok(const struct send_ctx *sctx, int cmd)
403 : {
404 0 : switch (sctx->proto) {
405 : case 1: return cmd <= BTRFS_SEND_C_MAX_V1;
406 : case 2: return cmd <= BTRFS_SEND_C_MAX_V2;
407 0 : case 3: return cmd <= BTRFS_SEND_C_MAX_V3;
408 : default: return false;
409 : }
410 : }
411 :
412 : static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
413 :
414 : static struct waiting_dir_move *
415 : get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
416 :
417 : static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen);
418 :
419 0 : static int need_send_hole(struct send_ctx *sctx)
420 : {
421 0 : return (sctx->parent_root && !sctx->cur_inode_new &&
422 0 : !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
423 0 : S_ISREG(sctx->cur_inode_mode));
424 : }
425 :
426 0 : static void fs_path_reset(struct fs_path *p)
427 : {
428 0 : if (p->reversed) {
429 0 : p->start = p->buf + p->buf_len - 1;
430 0 : p->end = p->start;
431 0 : *p->start = 0;
432 : } else {
433 0 : p->start = p->buf;
434 0 : p->end = p->start;
435 0 : *p->start = 0;
436 : }
437 0 : }
438 :
439 0 : static struct fs_path *fs_path_alloc(void)
440 : {
441 0 : struct fs_path *p;
442 :
443 0 : p = kmalloc(sizeof(*p), GFP_KERNEL);
444 0 : if (!p)
445 : return NULL;
446 0 : p->reversed = 0;
447 0 : p->buf = p->inline_buf;
448 0 : p->buf_len = FS_PATH_INLINE_SIZE;
449 0 : fs_path_reset(p);
450 0 : return p;
451 : }
452 :
453 0 : static struct fs_path *fs_path_alloc_reversed(void)
454 : {
455 0 : struct fs_path *p;
456 :
457 0 : p = fs_path_alloc();
458 0 : if (!p)
459 : return NULL;
460 0 : p->reversed = 1;
461 0 : fs_path_reset(p);
462 0 : return p;
463 : }
464 :
465 0 : static void fs_path_free(struct fs_path *p)
466 : {
467 0 : if (!p)
468 : return;
469 0 : if (p->buf != p->inline_buf)
470 0 : kfree(p->buf);
471 0 : kfree(p);
472 : }
473 :
474 : static int fs_path_len(struct fs_path *p)
475 : {
476 0 : return p->end - p->start;
477 : }
478 :
479 0 : static int fs_path_ensure_buf(struct fs_path *p, int len)
480 : {
481 0 : char *tmp_buf;
482 0 : int path_len;
483 0 : int old_buf_len;
484 :
485 0 : len++;
486 :
487 0 : if (p->buf_len >= len)
488 : return 0;
489 :
490 0 : if (len > PATH_MAX) {
491 0 : WARN_ON(1);
492 0 : return -ENOMEM;
493 : }
494 :
495 0 : path_len = p->end - p->start;
496 0 : old_buf_len = p->buf_len;
497 :
498 : /*
499 : * Allocate to the next largest kmalloc bucket size, to let
500 : * the fast path happen most of the time.
501 : */
502 0 : len = kmalloc_size_roundup(len);
503 : /*
504 : * First time the inline_buf does not suffice
505 : */
506 0 : if (p->buf == p->inline_buf) {
507 0 : tmp_buf = kmalloc(len, GFP_KERNEL);
508 0 : if (tmp_buf)
509 0 : memcpy(tmp_buf, p->buf, old_buf_len);
510 : } else {
511 0 : tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
512 : }
513 0 : if (!tmp_buf)
514 : return -ENOMEM;
515 0 : p->buf = tmp_buf;
516 0 : p->buf_len = len;
517 :
518 0 : if (p->reversed) {
519 0 : tmp_buf = p->buf + old_buf_len - path_len - 1;
520 0 : p->end = p->buf + p->buf_len - 1;
521 0 : p->start = p->end - path_len;
522 0 : memmove(p->start, tmp_buf, path_len + 1);
523 : } else {
524 0 : p->start = p->buf;
525 0 : p->end = p->start + path_len;
526 : }
527 : return 0;
528 : }
529 :
530 0 : static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
531 : char **prepared)
532 : {
533 0 : int ret;
534 0 : int new_len;
535 :
536 0 : new_len = p->end - p->start + name_len;
537 0 : if (p->start != p->end)
538 0 : new_len++;
539 0 : ret = fs_path_ensure_buf(p, new_len);
540 0 : if (ret < 0)
541 0 : goto out;
542 :
543 0 : if (p->reversed) {
544 0 : if (p->start != p->end)
545 0 : *--p->start = '/';
546 0 : p->start -= name_len;
547 0 : *prepared = p->start;
548 : } else {
549 0 : if (p->start != p->end)
550 0 : *p->end++ = '/';
551 0 : *prepared = p->end;
552 0 : p->end += name_len;
553 0 : *p->end = 0;
554 : }
555 :
556 0 : out:
557 0 : return ret;
558 : }
559 :
560 0 : static int fs_path_add(struct fs_path *p, const char *name, int name_len)
561 : {
562 0 : int ret;
563 0 : char *prepared;
564 :
565 0 : ret = fs_path_prepare_for_add(p, name_len, &prepared);
566 0 : if (ret < 0)
567 0 : goto out;
568 0 : memcpy(prepared, name, name_len);
569 :
570 0 : out:
571 0 : return ret;
572 : }
573 :
574 0 : static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
575 : {
576 0 : int ret;
577 0 : char *prepared;
578 :
579 0 : ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
580 0 : if (ret < 0)
581 0 : goto out;
582 0 : memcpy(prepared, p2->start, p2->end - p2->start);
583 :
584 0 : out:
585 0 : return ret;
586 : }
587 :
588 0 : 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 0 : int ret;
593 0 : char *prepared;
594 :
595 0 : ret = fs_path_prepare_for_add(p, len, &prepared);
596 0 : if (ret < 0)
597 0 : goto out;
598 :
599 0 : read_extent_buffer(eb, prepared, off, len);
600 :
601 0 : out:
602 0 : return ret;
603 : }
604 :
605 0 : static int fs_path_copy(struct fs_path *p, struct fs_path *from)
606 : {
607 0 : p->reversed = from->reversed;
608 0 : fs_path_reset(p);
609 :
610 0 : return fs_path_add_path(p, from);
611 : }
612 :
613 0 : static void fs_path_unreverse(struct fs_path *p)
614 : {
615 0 : char *tmp;
616 0 : int len;
617 :
618 0 : if (!p->reversed)
619 : return;
620 :
621 0 : tmp = p->start;
622 0 : len = p->end - p->start;
623 0 : p->start = p->buf;
624 0 : p->end = p->start + len;
625 0 : memmove(p->start, tmp, len + 1);
626 0 : p->reversed = 0;
627 : }
628 :
629 0 : static struct btrfs_path *alloc_path_for_send(void)
630 : {
631 0 : struct btrfs_path *path;
632 :
633 0 : path = btrfs_alloc_path();
634 0 : if (!path)
635 : return NULL;
636 0 : path->search_commit_root = 1;
637 0 : path->skip_locking = 1;
638 0 : path->need_commit_sem = 1;
639 0 : return path;
640 : }
641 :
642 0 : static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
643 : {
644 0 : int ret;
645 0 : u32 pos = 0;
646 :
647 0 : while (pos < len) {
648 0 : ret = kernel_write(filp, buf + pos, len - pos, off);
649 0 : if (ret < 0)
650 0 : return ret;
651 0 : if (ret == 0)
652 : return -EIO;
653 0 : pos += ret;
654 : }
655 :
656 : return 0;
657 : }
658 :
659 0 : static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
660 : {
661 0 : struct btrfs_tlv_header *hdr;
662 0 : int total_len = sizeof(*hdr) + len;
663 0 : int left = sctx->send_max_size - sctx->send_size;
664 :
665 0 : if (WARN_ON_ONCE(sctx->put_data))
666 : return -EINVAL;
667 :
668 0 : if (unlikely(left < total_len))
669 : return -EOVERFLOW;
670 :
671 0 : hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
672 0 : put_unaligned_le16(attr, &hdr->tlv_type);
673 0 : put_unaligned_le16(len, &hdr->tlv_len);
674 0 : memcpy(hdr + 1, data, len);
675 0 : sctx->send_size += total_len;
676 :
677 0 : return 0;
678 : }
679 :
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 : }
687 :
688 : TLV_PUT_DEFINE_INT(8)
689 0 : TLV_PUT_DEFINE_INT(32)
690 0 : TLV_PUT_DEFINE_INT(64)
691 :
692 0 : static int tlv_put_string(struct send_ctx *sctx, u16 attr,
693 : const char *str, int len)
694 : {
695 0 : if (len == -1)
696 0 : len = strlen(str);
697 0 : 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 0 : return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
704 : }
705 :
706 0 : static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
707 : struct extent_buffer *eb,
708 : struct btrfs_timespec *ts)
709 : {
710 0 : struct btrfs_timespec bts;
711 0 : read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
712 0 : return tlv_put(sctx, attr, &bts, sizeof(bts));
713 : }
714 :
715 :
716 : #define TLV_PUT(sctx, attrtype, data, attrlen) \
717 : do { \
718 : ret = tlv_put(sctx, attrtype, data, attrlen); \
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)
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 0 : static int send_header(struct send_ctx *sctx)
761 : {
762 0 : struct btrfs_stream_header hdr;
763 :
764 0 : strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
765 0 : hdr.version = cpu_to_le32(sctx->proto);
766 0 : return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
767 : &sctx->send_off);
768 : }
769 :
770 : /*
771 : * For each command/item we want to send to userspace, we call this function.
772 : */
773 0 : static int begin_cmd(struct send_ctx *sctx, int cmd)
774 : {
775 0 : struct btrfs_cmd_header *hdr;
776 :
777 0 : if (WARN_ON(!sctx->send_buf))
778 : return -EINVAL;
779 :
780 0 : BUG_ON(sctx->send_size);
781 :
782 0 : sctx->send_size += sizeof(*hdr);
783 0 : hdr = (struct btrfs_cmd_header *)sctx->send_buf;
784 0 : put_unaligned_le16(cmd, &hdr->cmd);
785 :
786 0 : return 0;
787 : }
788 :
789 0 : static int send_cmd(struct send_ctx *sctx)
790 : {
791 0 : int ret;
792 0 : struct btrfs_cmd_header *hdr;
793 0 : u32 crc;
794 :
795 0 : hdr = (struct btrfs_cmd_header *)sctx->send_buf;
796 0 : put_unaligned_le32(sctx->send_size - sizeof(*hdr), &hdr->len);
797 0 : put_unaligned_le32(0, &hdr->crc);
798 :
799 0 : crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
800 0 : put_unaligned_le32(crc, &hdr->crc);
801 :
802 0 : ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
803 : &sctx->send_off);
804 :
805 0 : sctx->send_size = 0;
806 0 : sctx->put_data = false;
807 :
808 0 : return ret;
809 : }
810 :
811 : /*
812 : * Sends a move instruction to user space
813 : */
814 0 : static int send_rename(struct send_ctx *sctx,
815 : struct fs_path *from, struct fs_path *to)
816 : {
817 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
818 0 : int ret;
819 :
820 0 : btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
821 :
822 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
823 0 : if (ret < 0)
824 0 : goto out;
825 :
826 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
827 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
828 :
829 0 : ret = send_cmd(sctx);
830 :
831 0 : tlv_put_failure:
832 0 : out:
833 0 : return ret;
834 : }
835 :
836 : /*
837 : * Sends a link instruction to user space
838 : */
839 0 : static int send_link(struct send_ctx *sctx,
840 : struct fs_path *path, struct fs_path *lnk)
841 : {
842 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
843 0 : int ret;
844 :
845 0 : btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
846 :
847 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
848 0 : if (ret < 0)
849 0 : goto out;
850 :
851 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
852 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
853 :
854 0 : ret = send_cmd(sctx);
855 :
856 0 : tlv_put_failure:
857 0 : out:
858 0 : return ret;
859 : }
860 :
861 : /*
862 : * Sends an unlink instruction to user space
863 : */
864 0 : static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
865 : {
866 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
867 0 : int ret;
868 :
869 0 : btrfs_debug(fs_info, "send_unlink %s", path->start);
870 :
871 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
872 0 : if (ret < 0)
873 0 : goto out;
874 :
875 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
876 :
877 0 : ret = send_cmd(sctx);
878 :
879 0 : tlv_put_failure:
880 0 : out:
881 0 : return ret;
882 : }
883 :
884 : /*
885 : * Sends a rmdir instruction to user space
886 : */
887 0 : static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
888 : {
889 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
890 0 : int ret;
891 :
892 0 : btrfs_debug(fs_info, "send_rmdir %s", path->start);
893 :
894 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
895 0 : if (ret < 0)
896 0 : goto out;
897 :
898 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
899 :
900 0 : ret = send_cmd(sctx);
901 :
902 0 : tlv_put_failure:
903 0 : out:
904 0 : return ret;
905 : }
906 :
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;
915 : u64 nlink;
916 : };
917 :
918 : /*
919 : * Helper function to retrieve some fields from an inode item.
920 : */
921 0 : static int get_inode_info(struct btrfs_root *root, u64 ino,
922 : struct btrfs_inode_info *info)
923 : {
924 0 : int ret;
925 0 : struct btrfs_path *path;
926 0 : struct btrfs_inode_item *ii;
927 0 : struct btrfs_key key;
928 :
929 0 : path = alloc_path_for_send();
930 0 : if (!path)
931 : return -ENOMEM;
932 :
933 0 : key.objectid = ino;
934 0 : key.type = BTRFS_INODE_ITEM_KEY;
935 0 : key.offset = 0;
936 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
937 0 : if (ret) {
938 0 : if (ret > 0)
939 0 : ret = -ENOENT;
940 0 : goto out;
941 : }
942 :
943 0 : if (!info)
944 0 : goto out;
945 :
946 0 : ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
947 : struct btrfs_inode_item);
948 0 : info->size = btrfs_inode_size(path->nodes[0], ii);
949 0 : info->gen = btrfs_inode_generation(path->nodes[0], ii);
950 0 : info->mode = btrfs_inode_mode(path->nodes[0], ii);
951 0 : info->uid = btrfs_inode_uid(path->nodes[0], ii);
952 0 : info->gid = btrfs_inode_gid(path->nodes[0], ii);
953 0 : info->rdev = btrfs_inode_rdev(path->nodes[0], ii);
954 0 : info->nlink = btrfs_inode_nlink(path->nodes[0], ii);
955 : /*
956 : * Transfer the unchanged u64 value of btrfs_inode_item::flags, that's
957 : * otherwise logically split to 32/32 parts.
958 : */
959 0 : info->fileattr = btrfs_inode_flags(path->nodes[0], ii);
960 :
961 0 : out:
962 0 : btrfs_free_path(path);
963 0 : return ret;
964 : }
965 :
966 0 : static int get_inode_gen(struct btrfs_root *root, u64 ino, u64 *gen)
967 : {
968 0 : int ret;
969 0 : struct btrfs_inode_info info = { 0 };
970 :
971 0 : ASSERT(gen);
972 :
973 0 : ret = get_inode_info(root, ino, &info);
974 0 : *gen = info.gen;
975 0 : 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 : /*
983 : * Helper function to iterate the entries in ONE btrfs_inode_ref or
984 : * btrfs_inode_extref.
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 : *
988 : * path must point to the INODE_REF or INODE_EXTREF when called.
989 : */
990 0 : static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
991 : struct btrfs_key *found_key, int resolve,
992 : iterate_inode_ref_t iterate, void *ctx)
993 : {
994 0 : struct extent_buffer *eb = path->nodes[0];
995 0 : struct btrfs_inode_ref *iref;
996 0 : struct btrfs_inode_extref *extref;
997 0 : struct btrfs_path *tmp_path;
998 0 : struct fs_path *p;
999 0 : u32 cur = 0;
1000 0 : u32 total;
1001 0 : int slot = path->slots[0];
1002 0 : u32 name_len;
1003 0 : char *start;
1004 0 : int ret = 0;
1005 0 : int num = 0;
1006 0 : int index;
1007 0 : u64 dir;
1008 0 : unsigned long name_off;
1009 0 : unsigned long elem_size;
1010 0 : unsigned long ptr;
1011 :
1012 0 : p = fs_path_alloc_reversed();
1013 0 : if (!p)
1014 : return -ENOMEM;
1015 :
1016 0 : tmp_path = alloc_path_for_send();
1017 0 : if (!tmp_path) {
1018 0 : fs_path_free(p);
1019 0 : return -ENOMEM;
1020 : }
1021 :
1022 :
1023 0 : if (found_key->type == BTRFS_INODE_REF_KEY) {
1024 0 : ptr = (unsigned long)btrfs_item_ptr(eb, slot,
1025 : struct btrfs_inode_ref);
1026 0 : total = btrfs_item_size(eb, slot);
1027 0 : elem_size = sizeof(*iref);
1028 : } else {
1029 0 : ptr = btrfs_item_ptr_offset(eb, slot);
1030 0 : total = btrfs_item_size(eb, slot);
1031 0 : elem_size = sizeof(*extref);
1032 : }
1033 :
1034 0 : while (cur < total) {
1035 0 : fs_path_reset(p);
1036 :
1037 0 : if (found_key->type == BTRFS_INODE_REF_KEY) {
1038 0 : iref = (struct btrfs_inode_ref *)(ptr + cur);
1039 0 : name_len = btrfs_inode_ref_name_len(eb, iref);
1040 0 : name_off = (unsigned long)(iref + 1);
1041 0 : index = btrfs_inode_ref_index(eb, iref);
1042 0 : dir = found_key->offset;
1043 : } else {
1044 0 : extref = (struct btrfs_inode_extref *)(ptr + cur);
1045 0 : name_len = btrfs_inode_extref_name_len(eb, extref);
1046 0 : name_off = (unsigned long)&extref->name;
1047 0 : index = btrfs_inode_extref_index(eb, extref);
1048 0 : dir = btrfs_inode_extref_parent(eb, extref);
1049 : }
1050 :
1051 0 : if (resolve) {
1052 0 : start = btrfs_ref_to_path(root, tmp_path, name_len,
1053 : name_off, eb, dir,
1054 0 : p->buf, p->buf_len);
1055 0 : if (IS_ERR(start)) {
1056 0 : ret = PTR_ERR(start);
1057 0 : goto out;
1058 : }
1059 0 : if (start < p->buf) {
1060 : /* overflow , try again with larger buffer */
1061 0 : ret = fs_path_ensure_buf(p,
1062 0 : p->buf_len + p->buf - start);
1063 0 : if (ret < 0)
1064 0 : goto out;
1065 0 : start = btrfs_ref_to_path(root, tmp_path,
1066 : name_len, name_off,
1067 : eb, dir,
1068 0 : p->buf, p->buf_len);
1069 0 : if (IS_ERR(start)) {
1070 0 : ret = PTR_ERR(start);
1071 0 : goto out;
1072 : }
1073 0 : BUG_ON(start < p->buf);
1074 : }
1075 0 : p->start = start;
1076 : } else {
1077 0 : ret = fs_path_add_from_extent_buffer(p, eb, name_off,
1078 : name_len);
1079 0 : if (ret < 0)
1080 0 : goto out;
1081 : }
1082 :
1083 0 : cur += elem_size + name_len;
1084 0 : ret = iterate(num, dir, index, p, ctx);
1085 0 : if (ret)
1086 0 : goto out;
1087 0 : num++;
1088 : }
1089 :
1090 0 : out:
1091 0 : btrfs_free_path(tmp_path);
1092 0 : fs_path_free(p);
1093 0 : 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,
1099 : void *ctx);
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 : */
1108 0 : static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
1109 : iterate_dir_item_t iterate, void *ctx)
1110 : {
1111 0 : int ret = 0;
1112 0 : struct extent_buffer *eb;
1113 0 : struct btrfs_dir_item *di;
1114 0 : struct btrfs_key di_key;
1115 0 : char *buf = NULL;
1116 0 : int buf_len;
1117 0 : u32 name_len;
1118 0 : u32 data_len;
1119 0 : u32 cur;
1120 0 : u32 len;
1121 0 : u32 total;
1122 0 : int slot;
1123 0 : int num;
1124 :
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 0 : buf_len = PATH_MAX;
1132 0 : buf = kmalloc(buf_len, GFP_KERNEL);
1133 0 : if (!buf) {
1134 0 : ret = -ENOMEM;
1135 0 : goto out;
1136 : }
1137 :
1138 0 : eb = path->nodes[0];
1139 0 : slot = path->slots[0];
1140 0 : di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1141 0 : cur = 0;
1142 0 : len = 0;
1143 0 : total = btrfs_item_size(eb, slot);
1144 :
1145 0 : num = 0;
1146 0 : while (cur < total) {
1147 0 : name_len = btrfs_dir_name_len(eb, di);
1148 0 : data_len = btrfs_dir_data_len(eb, di);
1149 0 : btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1150 :
1151 0 : if (btrfs_dir_ftype(eb, di) == BTRFS_FT_XATTR) {
1152 0 : if (name_len > XATTR_NAME_MAX) {
1153 0 : ret = -ENAMETOOLONG;
1154 0 : goto out;
1155 : }
1156 0 : if (name_len + data_len >
1157 0 : BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
1158 0 : ret = -E2BIG;
1159 0 : goto out;
1160 : }
1161 : } else {
1162 : /*
1163 : * Path too long
1164 : */
1165 0 : if (name_len + data_len > PATH_MAX) {
1166 0 : ret = -ENAMETOOLONG;
1167 0 : goto out;
1168 : }
1169 : }
1170 :
1171 0 : if (name_len + data_len > buf_len) {
1172 0 : buf_len = name_len + data_len;
1173 0 : if (is_vmalloc_addr(buf)) {
1174 0 : vfree(buf);
1175 0 : buf = NULL;
1176 : } else {
1177 0 : char *tmp = krealloc(buf, buf_len,
1178 : GFP_KERNEL | __GFP_NOWARN);
1179 :
1180 0 : if (!tmp)
1181 0 : kfree(buf);
1182 0 : buf = tmp;
1183 : }
1184 0 : if (!buf) {
1185 0 : buf = kvmalloc(buf_len, GFP_KERNEL);
1186 0 : if (!buf) {
1187 0 : ret = -ENOMEM;
1188 0 : goto out;
1189 : }
1190 : }
1191 : }
1192 :
1193 0 : read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1194 : name_len + data_len);
1195 :
1196 0 : len = sizeof(*di) + name_len + data_len;
1197 0 : di = (struct btrfs_dir_item *)((char *)di + len);
1198 0 : cur += len;
1199 :
1200 0 : ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1201 : data_len, ctx);
1202 0 : if (ret < 0)
1203 0 : goto out;
1204 0 : if (ret) {
1205 0 : ret = 0;
1206 0 : goto out;
1207 : }
1208 :
1209 0 : num++;
1210 : }
1211 :
1212 0 : out:
1213 0 : kvfree(buf);
1214 0 : return ret;
1215 : }
1216 :
1217 0 : static int __copy_first_ref(int num, u64 dir, int index,
1218 : struct fs_path *p, void *ctx)
1219 : {
1220 0 : int ret;
1221 0 : struct fs_path *pt = ctx;
1222 :
1223 0 : ret = fs_path_copy(pt, p);
1224 0 : if (ret < 0)
1225 0 : 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 : */
1235 0 : static int get_inode_path(struct btrfs_root *root,
1236 : u64 ino, struct fs_path *path)
1237 : {
1238 0 : int ret;
1239 0 : struct btrfs_key key, found_key;
1240 0 : struct btrfs_path *p;
1241 :
1242 0 : p = alloc_path_for_send();
1243 0 : if (!p)
1244 : return -ENOMEM;
1245 :
1246 0 : fs_path_reset(path);
1247 :
1248 0 : key.objectid = ino;
1249 0 : key.type = BTRFS_INODE_REF_KEY;
1250 0 : key.offset = 0;
1251 :
1252 0 : ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1253 0 : if (ret < 0)
1254 0 : goto out;
1255 0 : if (ret) {
1256 0 : ret = 1;
1257 0 : goto out;
1258 : }
1259 0 : btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1260 0 : if (found_key.objectid != ino ||
1261 0 : (found_key.type != BTRFS_INODE_REF_KEY &&
1262 : found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1263 0 : ret = -ENOENT;
1264 0 : goto out;
1265 : }
1266 :
1267 0 : ret = iterate_inode_ref(root, p, &found_key, 1,
1268 : __copy_first_ref, path);
1269 0 : if (ret < 0)
1270 : goto out;
1271 : ret = 0;
1272 :
1273 0 : out:
1274 0 : btrfs_free_path(p);
1275 0 : 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;
1293 :
1294 : /* The bytenr the file extent item we are processing refers to. */
1295 : u64 bytenr;
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;
1300 : };
1301 :
1302 0 : static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1303 : {
1304 0 : u64 root = (u64)(uintptr_t)key;
1305 0 : const struct clone_root *cr = elt;
1306 :
1307 0 : if (root < cr->root->root_key.objectid)
1308 : return -1;
1309 0 : if (root > cr->root->root_key.objectid)
1310 0 : return 1;
1311 : return 0;
1312 : }
1313 :
1314 0 : static int __clone_root_cmp_sort(const void *e1, const void *e2)
1315 : {
1316 0 : const struct clone_root *cr1 = e1;
1317 0 : const struct clone_root *cr2 = e2;
1318 :
1319 0 : if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
1320 : return -1;
1321 0 : if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
1322 0 : return 1;
1323 : return 0;
1324 : }
1325 :
1326 : /*
1327 : * Called for every backref that is found for the current extent.
1328 : * Results are collected in sctx->clone_roots->ino/offset.
1329 : */
1330 0 : static int iterate_backrefs(u64 ino, u64 offset, u64 num_bytes, u64 root_id,
1331 : void *ctx_)
1332 : {
1333 0 : struct backref_ctx *bctx = ctx_;
1334 0 : struct clone_root *clone_root;
1335 :
1336 : /* First check if the root is in the list of accepted clone sources */
1337 0 : clone_root = bsearch((void *)(uintptr_t)root_id, bctx->sctx->clone_roots,
1338 0 : bctx->sctx->clone_roots_cnt,
1339 : sizeof(struct clone_root),
1340 : __clone_root_cmp_bsearch);
1341 0 : if (!clone_root)
1342 : return 0;
1343 :
1344 : /* This is our own reference, bail out as we can't clone from it. */
1345 0 : if (clone_root->root == bctx->sctx->send_root &&
1346 0 : ino == bctx->cur_objectid &&
1347 0 : offset == bctx->cur_offset)
1348 : return 0;
1349 :
1350 : /*
1351 : * Make sure we don't consider clones from send_root that are
1352 : * behind the current inode/offset.
1353 : */
1354 0 : if (clone_root->root == bctx->sctx->send_root) {
1355 : /*
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.
1359 : */
1360 0 : 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 0 : if (ino == bctx->cur_objectid &&
1369 0 : offset + bctx->extent_len >
1370 0 : bctx->sctx->cur_inode_next_write_offset)
1371 : return 0;
1372 : }
1373 :
1374 0 : bctx->found++;
1375 0 : clone_root->found_ref = true;
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 : */
1383 0 : if (num_bytes > clone_root->num_bytes) {
1384 0 : clone_root->ino = ino;
1385 0 : clone_root->offset = offset;
1386 0 : 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 0 : if (num_bytes >= bctx->extent_len)
1393 0 : return BTRFS_ITERATE_EXTENT_INODES_STOP;
1394 : }
1395 :
1396 : return 0;
1397 : }
1398 :
1399 0 : static bool lookup_backref_cache(u64 leaf_bytenr, void *ctx,
1400 : const u64 **root_ids_ret, int *root_count_ret)
1401 : {
1402 0 : struct backref_ctx *bctx = ctx;
1403 0 : struct send_ctx *sctx = bctx->sctx;
1404 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1405 0 : const u64 key = leaf_bytenr >> fs_info->sectorsize_bits;
1406 0 : struct btrfs_lru_cache_entry *raw_entry;
1407 0 : struct backref_cache_entry *entry;
1408 :
1409 0 : if (btrfs_lru_cache_size(&sctx->backref_cache) == 0)
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 : */
1423 0 : if (fs_info->last_reloc_trans > sctx->backref_cache_last_reloc_trans) {
1424 0 : btrfs_lru_cache_clear(&sctx->backref_cache);
1425 0 : return false;
1426 : }
1427 :
1428 0 : raw_entry = btrfs_lru_cache_lookup(&sctx->backref_cache, key, 0);
1429 0 : if (!raw_entry)
1430 : return false;
1431 :
1432 0 : entry = container_of(raw_entry, struct backref_cache_entry, entry);
1433 0 : *root_ids_ret = entry->root_ids;
1434 0 : *root_count_ret = entry->num_roots;
1435 :
1436 0 : return true;
1437 : }
1438 :
1439 0 : static void store_backref_cache(u64 leaf_bytenr, const struct ulist *root_ids,
1440 : void *ctx)
1441 : {
1442 0 : struct backref_ctx *bctx = ctx;
1443 0 : struct send_ctx *sctx = bctx->sctx;
1444 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1445 0 : struct backref_cache_entry *new_entry;
1446 0 : struct ulist_iterator uiter;
1447 0 : struct ulist_node *node;
1448 0 : 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 0 : new_entry = kmalloc(sizeof(struct backref_cache_entry), GFP_NOFS);
1456 : /* No worries, cache is optional. */
1457 0 : if (!new_entry)
1458 0 : return;
1459 :
1460 0 : new_entry->entry.key = leaf_bytenr >> fs_info->sectorsize_bits;
1461 0 : new_entry->entry.gen = 0;
1462 0 : new_entry->num_roots = 0;
1463 0 : ULIST_ITER_INIT(&uiter);
1464 0 : while ((node = ulist_next(root_ids, &uiter)) != NULL) {
1465 0 : const u64 root_id = node->val;
1466 0 : struct clone_root *root;
1467 :
1468 0 : root = bsearch((void *)(uintptr_t)root_id, sctx->clone_roots,
1469 0 : sctx->clone_roots_cnt, sizeof(struct clone_root),
1470 : __clone_root_cmp_bsearch);
1471 0 : if (!root)
1472 0 : continue;
1473 :
1474 : /* Too many roots, just exit, no worries as caching is optional. */
1475 0 : if (new_entry->num_roots >= SEND_MAX_BACKREF_CACHE_ROOTS) {
1476 0 : kfree(new_entry);
1477 0 : return;
1478 : }
1479 :
1480 0 : new_entry->root_ids[new_entry->num_roots] = root_id;
1481 0 : 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.
1489 : *
1490 : * Also use GFP_NOFS because we're called while holding a transaction
1491 : * handle or while holding fs_info->commit_root_sem.
1492 : */
1493 0 : ret = btrfs_lru_cache_store(&sctx->backref_cache, &new_entry->entry,
1494 : GFP_NOFS);
1495 0 : ASSERT(ret == 0 || ret == -ENOMEM);
1496 0 : if (ret) {
1497 : /* Caching is optional, no worries. */
1498 0 : kfree(new_entry);
1499 0 : return;
1500 : }
1501 :
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 : */
1507 0 : if (btrfs_lru_cache_size(&sctx->backref_cache) == 1)
1508 0 : sctx->backref_cache_last_reloc_trans = fs_info->last_reloc_trans;
1509 : }
1510 :
1511 0 : static int check_extent_item(u64 bytenr, const struct btrfs_extent_item *ei,
1512 : const struct extent_buffer *leaf, void *ctx)
1513 : {
1514 0 : const u64 refs = btrfs_extent_refs(leaf, ei);
1515 0 : const struct backref_ctx *bctx = ctx;
1516 0 : const struct send_ctx *sctx = bctx->sctx;
1517 :
1518 0 : if (bytenr == bctx->bytenr) {
1519 0 : const u64 flags = btrfs_extent_flags(leaf, ei);
1520 :
1521 0 : 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 0 : 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 0 : if (refs > SEND_MAX_EXTENT_REFS)
1543 0 : return -ENOENT;
1544 :
1545 : return 0;
1546 : }
1547 :
1548 0 : static bool skip_self_data_ref(u64 root, u64 ino, u64 offset, void *ctx)
1549 : {
1550 0 : const struct backref_ctx *bctx = ctx;
1551 :
1552 0 : if (ino == bctx->cur_objectid &&
1553 0 : root == bctx->backref_owner &&
1554 0 : offset == bctx->backref_offset)
1555 0 : return true;
1556 :
1557 : return false;
1558 : }
1559 :
1560 : /*
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 : *
1567 : * path must point to the extent item when called.
1568 : */
1569 0 : 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 : {
1575 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1576 0 : int ret;
1577 0 : int extent_type;
1578 0 : u64 logical;
1579 0 : u64 disk_byte;
1580 0 : u64 num_bytes;
1581 0 : struct btrfs_file_extent_item *fi;
1582 0 : struct extent_buffer *eb = path->nodes[0];
1583 0 : struct backref_ctx backref_ctx = { 0 };
1584 0 : struct btrfs_backref_walk_ctx backref_walk_ctx = { 0 };
1585 0 : struct clone_root *cur_clone_root;
1586 0 : int compressed;
1587 0 : u32 i;
1588 :
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 0 : if (data_offset >= ino_size)
1596 : return 0;
1597 :
1598 0 : fi = btrfs_item_ptr(eb, path->slots[0], struct btrfs_file_extent_item);
1599 0 : extent_type = btrfs_file_extent_type(eb, fi);
1600 0 : if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1601 : return -ENOENT;
1602 :
1603 0 : disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1604 0 : if (disk_byte == 0)
1605 : return -ENOENT;
1606 :
1607 0 : compressed = btrfs_file_extent_compression(eb, fi);
1608 0 : num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1609 0 : logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1610 :
1611 : /*
1612 : * Setup the clone roots.
1613 : */
1614 0 : for (i = 0; i < sctx->clone_roots_cnt; i++) {
1615 0 : cur_clone_root = sctx->clone_roots + i;
1616 0 : cur_clone_root->ino = (u64)-1;
1617 0 : cur_clone_root->offset = 0;
1618 0 : cur_clone_root->num_bytes = 0;
1619 0 : cur_clone_root->found_ref = false;
1620 : }
1621 :
1622 0 : backref_ctx.sctx = sctx;
1623 0 : backref_ctx.cur_objectid = ino;
1624 0 : backref_ctx.cur_offset = data_offset;
1625 0 : backref_ctx.bytenr = disk_byte;
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 0 : backref_ctx.backref_owner = btrfs_header_owner(eb);
1631 0 : backref_ctx.backref_offset = data_offset - btrfs_file_extent_offset(eb, fi);
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
1636 : * iterate_backrefs() work.
1637 : */
1638 0 : if (data_offset + num_bytes >= ino_size)
1639 0 : backref_ctx.extent_len = ino_size - data_offset;
1640 : else
1641 0 : backref_ctx.extent_len = num_bytes;
1642 :
1643 : /*
1644 : * Now collect all backrefs.
1645 : */
1646 0 : backref_walk_ctx.bytenr = disk_byte;
1647 0 : if (compressed == BTRFS_COMPRESS_NONE)
1648 0 : backref_walk_ctx.extent_item_pos = btrfs_file_extent_offset(eb, fi);
1649 0 : backref_walk_ctx.fs_info = fs_info;
1650 0 : backref_walk_ctx.cache_lookup = lookup_backref_cache;
1651 0 : backref_walk_ctx.cache_store = store_backref_cache;
1652 0 : backref_walk_ctx.indirect_ref_iterator = iterate_backrefs;
1653 0 : backref_walk_ctx.check_extent_item = check_extent_item;
1654 0 : backref_walk_ctx.user_ctx = &backref_ctx;
1655 :
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 0 : if (sctx->clone_roots_cnt == 1)
1665 0 : backref_walk_ctx.skip_data_ref = skip_self_data_ref;
1666 :
1667 0 : ret = iterate_extent_inodes(&backref_walk_ctx, true, iterate_backrefs,
1668 : &backref_ctx);
1669 0 : if (ret < 0)
1670 : return ret;
1671 :
1672 0 : down_read(&fs_info->commit_root_sem);
1673 0 : 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 0 : up_read(&fs_info->commit_root_sem);
1687 0 : return -ENOENT;
1688 : }
1689 0 : up_read(&fs_info->commit_root_sem);
1690 :
1691 0 : 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);
1694 :
1695 0 : if (!backref_ctx.found) {
1696 : btrfs_debug(fs_info, "no clones found");
1697 : return -ENOENT;
1698 : }
1699 :
1700 : cur_clone_root = NULL;
1701 0 : for (i = 0; i < sctx->clone_roots_cnt; i++) {
1702 0 : struct clone_root *clone_root = &sctx->clone_roots[i];
1703 :
1704 0 : if (!clone_root->found_ref)
1705 0 : 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 0 : if (!cur_clone_root ||
1713 0 : clone_root->num_bytes > cur_clone_root->num_bytes) {
1714 0 : 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 0 : if (clone_root->num_bytes >= backref_ctx.extent_len)
1721 : break;
1722 : }
1723 : }
1724 :
1725 0 : if (cur_clone_root) {
1726 0 : *found = cur_clone_root;
1727 0 : ret = 0;
1728 : } else {
1729 : ret = -ENOENT;
1730 : }
1731 :
1732 : return ret;
1733 : }
1734 :
1735 0 : static int read_symlink(struct btrfs_root *root,
1736 : u64 ino,
1737 : struct fs_path *dest)
1738 : {
1739 0 : int ret;
1740 0 : struct btrfs_path *path;
1741 0 : struct btrfs_key key;
1742 0 : struct btrfs_file_extent_item *ei;
1743 0 : u8 type;
1744 0 : u8 compression;
1745 0 : unsigned long off;
1746 0 : int len;
1747 :
1748 0 : path = alloc_path_for_send();
1749 0 : if (!path)
1750 : return -ENOMEM;
1751 :
1752 0 : key.objectid = ino;
1753 0 : key.type = BTRFS_EXTENT_DATA_KEY;
1754 0 : key.offset = 0;
1755 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1756 0 : if (ret < 0)
1757 0 : goto out;
1758 0 : 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 0 : btrfs_err(root->fs_info,
1768 : "Found empty symlink inode %llu at root %llu",
1769 : ino, root->root_key.objectid);
1770 0 : ret = -EIO;
1771 0 : goto out;
1772 : }
1773 :
1774 0 : ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1775 : struct btrfs_file_extent_item);
1776 0 : type = btrfs_file_extent_type(path->nodes[0], ei);
1777 0 : if (unlikely(type != BTRFS_FILE_EXTENT_INLINE)) {
1778 0 : ret = -EUCLEAN;
1779 0 : btrfs_crit(root->fs_info,
1780 : "send: found symlink extent that is not inline, ino %llu root %llu extent type %d",
1781 : ino, btrfs_root_id(root), type);
1782 0 : goto out;
1783 : }
1784 0 : compression = btrfs_file_extent_compression(path->nodes[0], ei);
1785 0 : if (unlikely(compression != BTRFS_COMPRESS_NONE)) {
1786 0 : ret = -EUCLEAN;
1787 0 : btrfs_crit(root->fs_info,
1788 : "send: found symlink extent with compression, ino %llu root %llu compression type %d",
1789 : ino, btrfs_root_id(root), compression);
1790 0 : goto out;
1791 : }
1792 :
1793 0 : off = btrfs_file_extent_inline_start(ei);
1794 0 : len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
1795 :
1796 0 : ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1797 :
1798 0 : out:
1799 0 : btrfs_free_path(path);
1800 0 : return ret;
1801 : }
1802 :
1803 : /*
1804 : * Helper function to generate a file name that is unique in the root of
1805 : * send_root and parent_root. This is used to generate names for orphan inodes.
1806 : */
1807 0 : static int gen_unique_name(struct send_ctx *sctx,
1808 : u64 ino, u64 gen,
1809 : struct fs_path *dest)
1810 : {
1811 0 : int ret = 0;
1812 0 : struct btrfs_path *path;
1813 0 : struct btrfs_dir_item *di;
1814 0 : char tmp[64];
1815 0 : int len;
1816 0 : u64 idx = 0;
1817 :
1818 0 : path = alloc_path_for_send();
1819 0 : if (!path)
1820 : return -ENOMEM;
1821 :
1822 0 : while (1) {
1823 0 : struct fscrypt_str tmp_name;
1824 :
1825 0 : len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1826 : ino, gen, idx);
1827 0 : ASSERT(len < sizeof(tmp));
1828 0 : tmp_name.name = tmp;
1829 0 : tmp_name.len = strlen(tmp);
1830 :
1831 0 : di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1832 : path, BTRFS_FIRST_FREE_OBJECTID,
1833 : &tmp_name, 0);
1834 0 : btrfs_release_path(path);
1835 0 : if (IS_ERR(di)) {
1836 0 : ret = PTR_ERR(di);
1837 0 : goto out;
1838 : }
1839 0 : if (di) {
1840 : /* not unique, try again */
1841 0 : idx++;
1842 0 : continue;
1843 : }
1844 :
1845 0 : if (!sctx->parent_root) {
1846 : /* unique */
1847 : ret = 0;
1848 0 : break;
1849 : }
1850 :
1851 0 : di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1852 : path, BTRFS_FIRST_FREE_OBJECTID,
1853 : &tmp_name, 0);
1854 0 : btrfs_release_path(path);
1855 0 : if (IS_ERR(di)) {
1856 0 : ret = PTR_ERR(di);
1857 0 : goto out;
1858 : }
1859 0 : if (di) {
1860 : /* not unique, try again */
1861 0 : idx++;
1862 0 : continue;
1863 : }
1864 : /* unique */
1865 : break;
1866 : }
1867 :
1868 0 : ret = fs_path_add(dest, tmp, strlen(tmp));
1869 :
1870 0 : out:
1871 0 : btrfs_free_path(path);
1872 0 : return ret;
1873 : }
1874 :
1875 : enum inode_state {
1876 : inode_state_no_change,
1877 : inode_state_will_create,
1878 : inode_state_did_create,
1879 : inode_state_will_delete,
1880 : inode_state_did_delete,
1881 : };
1882 :
1883 0 : static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen,
1884 : u64 *send_gen, u64 *parent_gen)
1885 : {
1886 0 : int ret;
1887 0 : int left_ret;
1888 0 : int right_ret;
1889 0 : u64 left_gen;
1890 0 : u64 right_gen = 0;
1891 0 : struct btrfs_inode_info info;
1892 :
1893 0 : ret = get_inode_info(sctx->send_root, ino, &info);
1894 0 : if (ret < 0 && ret != -ENOENT)
1895 0 : goto out;
1896 0 : left_ret = (info.nlink == 0) ? -ENOENT : ret;
1897 0 : left_gen = info.gen;
1898 0 : if (send_gen)
1899 0 : *send_gen = ((left_ret == -ENOENT) ? 0 : info.gen);
1900 :
1901 0 : if (!sctx->parent_root) {
1902 : right_ret = -ENOENT;
1903 : } else {
1904 0 : ret = get_inode_info(sctx->parent_root, ino, &info);
1905 0 : if (ret < 0 && ret != -ENOENT)
1906 0 : goto out;
1907 0 : right_ret = (info.nlink == 0) ? -ENOENT : ret;
1908 0 : right_gen = info.gen;
1909 0 : if (parent_gen)
1910 0 : *parent_gen = ((right_ret == -ENOENT) ? 0 : info.gen);
1911 : }
1912 :
1913 0 : if (!left_ret && !right_ret) {
1914 0 : if (left_gen == gen && right_gen == gen) {
1915 : ret = inode_state_no_change;
1916 0 : } else if (left_gen == gen) {
1917 0 : if (ino < sctx->send_progress)
1918 : ret = inode_state_did_create;
1919 : else
1920 0 : ret = inode_state_will_create;
1921 0 : } else if (right_gen == gen) {
1922 0 : if (ino < sctx->send_progress)
1923 : ret = inode_state_did_delete;
1924 : else
1925 0 : ret = inode_state_will_delete;
1926 : } else {
1927 : ret = -ENOENT;
1928 : }
1929 0 : } else if (!left_ret) {
1930 0 : if (left_gen == gen) {
1931 0 : if (ino < sctx->send_progress)
1932 : ret = inode_state_did_create;
1933 : else
1934 0 : ret = inode_state_will_create;
1935 : } else {
1936 : ret = -ENOENT;
1937 : }
1938 0 : } else if (!right_ret) {
1939 0 : if (right_gen == gen) {
1940 0 : if (ino < sctx->send_progress)
1941 : ret = inode_state_did_delete;
1942 : else
1943 0 : ret = inode_state_will_delete;
1944 : } else {
1945 : ret = -ENOENT;
1946 : }
1947 : } else {
1948 : ret = -ENOENT;
1949 : }
1950 :
1951 0 : out:
1952 0 : return ret;
1953 : }
1954 :
1955 0 : static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen,
1956 : u64 *send_gen, u64 *parent_gen)
1957 : {
1958 0 : int ret;
1959 :
1960 0 : if (ino == BTRFS_FIRST_FREE_OBJECTID)
1961 : return 1;
1962 :
1963 0 : ret = get_cur_inode_state(sctx, ino, gen, send_gen, parent_gen);
1964 0 : if (ret < 0)
1965 0 : goto out;
1966 :
1967 0 : if (ret == inode_state_no_change ||
1968 0 : ret == inode_state_did_create ||
1969 : ret == inode_state_will_delete)
1970 : ret = 1;
1971 : else
1972 : ret = 0;
1973 :
1974 : out:
1975 : return ret;
1976 : }
1977 :
1978 : /*
1979 : * Helper function to lookup a dir item in a dir.
1980 : */
1981 0 : static int lookup_dir_item_inode(struct btrfs_root *root,
1982 : u64 dir, const char *name, int name_len,
1983 : u64 *found_inode)
1984 : {
1985 0 : int ret = 0;
1986 0 : struct btrfs_dir_item *di;
1987 0 : struct btrfs_key key;
1988 0 : struct btrfs_path *path;
1989 0 : struct fscrypt_str name_str = FSTR_INIT((char *)name, name_len);
1990 :
1991 0 : path = alloc_path_for_send();
1992 0 : if (!path)
1993 : return -ENOMEM;
1994 :
1995 0 : di = btrfs_lookup_dir_item(NULL, root, path, dir, &name_str, 0);
1996 0 : if (IS_ERR_OR_NULL(di)) {
1997 0 : ret = di ? PTR_ERR(di) : -ENOENT;
1998 0 : goto out;
1999 : }
2000 0 : btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
2001 0 : if (key.type == BTRFS_ROOT_ITEM_KEY) {
2002 0 : ret = -ENOENT;
2003 0 : goto out;
2004 : }
2005 0 : *found_inode = key.objectid;
2006 :
2007 0 : out:
2008 0 : btrfs_free_path(path);
2009 0 : return ret;
2010 : }
2011 :
2012 : /*
2013 : * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
2014 : * generation of the parent dir and the name of the dir entry.
2015 : */
2016 0 : static int get_first_ref(struct btrfs_root *root, u64 ino,
2017 : u64 *dir, u64 *dir_gen, struct fs_path *name)
2018 : {
2019 0 : int ret;
2020 0 : struct btrfs_key key;
2021 0 : struct btrfs_key found_key;
2022 0 : struct btrfs_path *path;
2023 0 : int len;
2024 0 : u64 parent_dir;
2025 :
2026 0 : path = alloc_path_for_send();
2027 0 : if (!path)
2028 : return -ENOMEM;
2029 :
2030 0 : key.objectid = ino;
2031 0 : key.type = BTRFS_INODE_REF_KEY;
2032 0 : key.offset = 0;
2033 :
2034 0 : ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2035 0 : if (ret < 0)
2036 0 : goto out;
2037 0 : if (!ret)
2038 0 : btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2039 : path->slots[0]);
2040 0 : if (ret || found_key.objectid != ino ||
2041 0 : (found_key.type != BTRFS_INODE_REF_KEY &&
2042 : found_key.type != BTRFS_INODE_EXTREF_KEY)) {
2043 0 : ret = -ENOENT;
2044 0 : goto out;
2045 : }
2046 :
2047 0 : if (found_key.type == BTRFS_INODE_REF_KEY) {
2048 0 : struct btrfs_inode_ref *iref;
2049 0 : iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2050 : struct btrfs_inode_ref);
2051 0 : len = btrfs_inode_ref_name_len(path->nodes[0], iref);
2052 0 : ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
2053 0 : (unsigned long)(iref + 1),
2054 : len);
2055 0 : parent_dir = found_key.offset;
2056 : } else {
2057 0 : struct btrfs_inode_extref *extref;
2058 0 : extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2059 : struct btrfs_inode_extref);
2060 0 : len = btrfs_inode_extref_name_len(path->nodes[0], extref);
2061 0 : ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
2062 0 : (unsigned long)&extref->name, len);
2063 0 : parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
2064 : }
2065 0 : if (ret < 0)
2066 0 : goto out;
2067 0 : btrfs_release_path(path);
2068 :
2069 0 : if (dir_gen) {
2070 0 : ret = get_inode_gen(root, parent_dir, dir_gen);
2071 0 : if (ret < 0)
2072 0 : goto out;
2073 : }
2074 :
2075 0 : *dir = parent_dir;
2076 :
2077 0 : out:
2078 0 : btrfs_free_path(path);
2079 0 : return ret;
2080 : }
2081 :
2082 0 : static int is_first_ref(struct btrfs_root *root,
2083 : u64 ino, u64 dir,
2084 : const char *name, int name_len)
2085 : {
2086 0 : int ret;
2087 0 : struct fs_path *tmp_name;
2088 0 : u64 tmp_dir;
2089 :
2090 0 : tmp_name = fs_path_alloc();
2091 0 : if (!tmp_name)
2092 : return -ENOMEM;
2093 :
2094 0 : ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
2095 0 : if (ret < 0)
2096 0 : goto out;
2097 :
2098 0 : if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
2099 0 : ret = 0;
2100 0 : goto out;
2101 : }
2102 :
2103 0 : ret = !memcmp(tmp_name->start, name, name_len);
2104 :
2105 0 : out:
2106 0 : fs_path_free(tmp_name);
2107 0 : return ret;
2108 : }
2109 :
2110 : /*
2111 : * Used by process_recorded_refs to determine if a new ref would overwrite an
2112 : * already existing ref. In case it detects an overwrite, it returns the
2113 : * inode/gen in who_ino/who_gen.
2114 : * When an overwrite is detected, process_recorded_refs does proper orphanizing
2115 : * to make sure later references to the overwritten inode are possible.
2116 : * Orphanizing is however only required for the first ref of an inode.
2117 : * process_recorded_refs does an additional is_first_ref check to see if
2118 : * orphanizing is really required.
2119 : */
2120 0 : static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2121 : const char *name, int name_len,
2122 : u64 *who_ino, u64 *who_gen, u64 *who_mode)
2123 : {
2124 0 : int ret;
2125 0 : u64 parent_root_dir_gen;
2126 0 : u64 other_inode = 0;
2127 0 : struct btrfs_inode_info info;
2128 :
2129 0 : if (!sctx->parent_root)
2130 : return 0;
2131 :
2132 0 : ret = is_inode_existent(sctx, dir, dir_gen, NULL, &parent_root_dir_gen);
2133 0 : if (ret <= 0)
2134 : return 0;
2135 :
2136 : /*
2137 : * If we have a parent root we need to verify that the parent dir was
2138 : * not deleted and then re-created, if it was then we have no overwrite
2139 : * and we can just unlink this entry.
2140 : *
2141 : * @parent_root_dir_gen was set to 0 if the inode does not exist in the
2142 : * parent root.
2143 : */
2144 0 : if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID &&
2145 0 : parent_root_dir_gen != dir_gen)
2146 : return 0;
2147 :
2148 0 : ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
2149 : &other_inode);
2150 0 : if (ret == -ENOENT)
2151 : return 0;
2152 0 : else if (ret < 0)
2153 : return ret;
2154 :
2155 : /*
2156 : * Check if the overwritten ref was already processed. If yes, the ref
2157 : * was already unlinked/moved, so we can safely assume that we will not
2158 : * overwrite anything at this point in time.
2159 : */
2160 0 : if (other_inode > sctx->send_progress ||
2161 0 : is_waiting_for_move(sctx, other_inode)) {
2162 0 : ret = get_inode_info(sctx->parent_root, other_inode, &info);
2163 0 : if (ret < 0)
2164 : return ret;
2165 :
2166 0 : *who_ino = other_inode;
2167 0 : *who_gen = info.gen;
2168 0 : *who_mode = info.mode;
2169 0 : return 1;
2170 : }
2171 :
2172 : return 0;
2173 : }
2174 :
2175 : /*
2176 : * Checks if the ref was overwritten by an already processed inode. This is
2177 : * used by __get_cur_name_and_parent to find out if the ref was orphanized and
2178 : * thus the orphan name needs be used.
2179 : * process_recorded_refs also uses it to avoid unlinking of refs that were
2180 : * overwritten.
2181 : */
2182 0 : static int did_overwrite_ref(struct send_ctx *sctx,
2183 : u64 dir, u64 dir_gen,
2184 : u64 ino, u64 ino_gen,
2185 : const char *name, int name_len)
2186 : {
2187 0 : int ret;
2188 0 : u64 ow_inode;
2189 0 : u64 ow_gen = 0;
2190 0 : u64 send_root_dir_gen;
2191 :
2192 0 : if (!sctx->parent_root)
2193 : return 0;
2194 :
2195 0 : ret = is_inode_existent(sctx, dir, dir_gen, &send_root_dir_gen, NULL);
2196 0 : if (ret <= 0)
2197 : return ret;
2198 :
2199 : /*
2200 : * @send_root_dir_gen was set to 0 if the inode does not exist in the
2201 : * send root.
2202 : */
2203 0 : if (dir != BTRFS_FIRST_FREE_OBJECTID && send_root_dir_gen != dir_gen)
2204 : return 0;
2205 :
2206 : /* check if the ref was overwritten by another ref */
2207 0 : ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
2208 : &ow_inode);
2209 0 : if (ret == -ENOENT) {
2210 : /* was never and will never be overwritten */
2211 : return 0;
2212 0 : } else if (ret < 0) {
2213 : return ret;
2214 : }
2215 :
2216 0 : if (ow_inode == ino) {
2217 0 : ret = get_inode_gen(sctx->send_root, ow_inode, &ow_gen);
2218 0 : if (ret < 0)
2219 : return ret;
2220 :
2221 : /* It's the same inode, so no overwrite happened. */
2222 0 : if (ow_gen == ino_gen)
2223 : return 0;
2224 : }
2225 :
2226 : /*
2227 : * We know that it is or will be overwritten. Check this now.
2228 : * The current inode being processed might have been the one that caused
2229 : * inode 'ino' to be orphanized, therefore check if ow_inode matches
2230 : * the current inode being processed.
2231 : */
2232 0 : if (ow_inode < sctx->send_progress)
2233 : return 1;
2234 :
2235 0 : if (ino != sctx->cur_ino && ow_inode == sctx->cur_ino) {
2236 0 : if (ow_gen == 0) {
2237 0 : ret = get_inode_gen(sctx->send_root, ow_inode, &ow_gen);
2238 0 : if (ret < 0)
2239 : return ret;
2240 : }
2241 0 : if (ow_gen == sctx->cur_inode_gen)
2242 0 : return 1;
2243 : }
2244 :
2245 : return 0;
2246 : }
2247 :
2248 : /*
2249 : * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
2250 : * that got overwritten. This is used by process_recorded_refs to determine
2251 : * if it has to use the path as returned by get_cur_path or the orphan name.
2252 : */
2253 0 : static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
2254 : {
2255 0 : int ret = 0;
2256 0 : struct fs_path *name = NULL;
2257 0 : u64 dir;
2258 0 : u64 dir_gen;
2259 :
2260 0 : if (!sctx->parent_root)
2261 0 : goto out;
2262 :
2263 0 : name = fs_path_alloc();
2264 0 : if (!name)
2265 : return -ENOMEM;
2266 :
2267 0 : ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
2268 0 : if (ret < 0)
2269 0 : goto out;
2270 :
2271 0 : ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
2272 0 : name->start, fs_path_len(name));
2273 :
2274 0 : out:
2275 0 : fs_path_free(name);
2276 0 : return ret;
2277 : }
2278 :
2279 : static inline struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2280 : u64 ino, u64 gen)
2281 : {
2282 0 : struct btrfs_lru_cache_entry *entry;
2283 :
2284 0 : entry = btrfs_lru_cache_lookup(&sctx->name_cache, ino, gen);
2285 0 : if (!entry)
2286 : return NULL;
2287 :
2288 : return container_of(entry, struct name_cache_entry, entry);
2289 : }
2290 :
2291 : /*
2292 : * Used by get_cur_path for each ref up to the root.
2293 : * Returns 0 if it succeeded.
2294 : * Returns 1 if the inode is not existent or got overwritten. In that case, the
2295 : * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2296 : * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2297 : * Returns <0 in case of error.
2298 : */
2299 0 : static int __get_cur_name_and_parent(struct send_ctx *sctx,
2300 : u64 ino, u64 gen,
2301 : u64 *parent_ino,
2302 : u64 *parent_gen,
2303 : struct fs_path *dest)
2304 : {
2305 0 : int ret;
2306 0 : int nce_ret;
2307 0 : struct name_cache_entry *nce;
2308 :
2309 : /*
2310 : * First check if we already did a call to this function with the same
2311 : * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2312 : * return the cached result.
2313 : */
2314 0 : nce = name_cache_search(sctx, ino, gen);
2315 0 : if (nce) {
2316 0 : if (ino < sctx->send_progress && nce->need_later_update) {
2317 0 : btrfs_lru_cache_remove(&sctx->name_cache, &nce->entry);
2318 0 : nce = NULL;
2319 : } else {
2320 0 : *parent_ino = nce->parent_ino;
2321 0 : *parent_gen = nce->parent_gen;
2322 0 : ret = fs_path_add(dest, nce->name, nce->name_len);
2323 0 : if (ret < 0)
2324 0 : goto out;
2325 0 : ret = nce->ret;
2326 0 : goto out;
2327 : }
2328 : }
2329 :
2330 : /*
2331 : * If the inode is not existent yet, add the orphan name and return 1.
2332 : * This should only happen for the parent dir that we determine in
2333 : * record_new_ref_if_needed().
2334 : */
2335 0 : ret = is_inode_existent(sctx, ino, gen, NULL, NULL);
2336 0 : if (ret < 0)
2337 0 : goto out;
2338 :
2339 0 : if (!ret) {
2340 0 : ret = gen_unique_name(sctx, ino, gen, dest);
2341 0 : if (ret < 0)
2342 0 : goto out;
2343 0 : ret = 1;
2344 0 : goto out_cache;
2345 : }
2346 :
2347 : /*
2348 : * Depending on whether the inode was already processed or not, use
2349 : * send_root or parent_root for ref lookup.
2350 : */
2351 0 : if (ino < sctx->send_progress)
2352 0 : ret = get_first_ref(sctx->send_root, ino,
2353 : parent_ino, parent_gen, dest);
2354 : else
2355 0 : ret = get_first_ref(sctx->parent_root, ino,
2356 : parent_ino, parent_gen, dest);
2357 0 : if (ret < 0)
2358 0 : goto out;
2359 :
2360 : /*
2361 : * Check if the ref was overwritten by an inode's ref that was processed
2362 : * earlier. If yes, treat as orphan and return 1.
2363 : */
2364 0 : ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2365 0 : dest->start, dest->end - dest->start);
2366 0 : if (ret < 0)
2367 0 : goto out;
2368 0 : if (ret) {
2369 0 : fs_path_reset(dest);
2370 0 : ret = gen_unique_name(sctx, ino, gen, dest);
2371 0 : if (ret < 0)
2372 0 : goto out;
2373 : ret = 1;
2374 : }
2375 :
2376 0 : out_cache:
2377 : /*
2378 : * Store the result of the lookup in the name cache.
2379 : */
2380 0 : nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2381 0 : if (!nce) {
2382 0 : ret = -ENOMEM;
2383 0 : goto out;
2384 : }
2385 :
2386 0 : nce->entry.key = ino;
2387 0 : nce->entry.gen = gen;
2388 0 : nce->parent_ino = *parent_ino;
2389 0 : nce->parent_gen = *parent_gen;
2390 0 : nce->name_len = fs_path_len(dest);
2391 0 : nce->ret = ret;
2392 0 : strcpy(nce->name, dest->start);
2393 :
2394 0 : if (ino < sctx->send_progress)
2395 0 : nce->need_later_update = 0;
2396 : else
2397 0 : nce->need_later_update = 1;
2398 :
2399 0 : nce_ret = btrfs_lru_cache_store(&sctx->name_cache, &nce->entry, GFP_KERNEL);
2400 0 : if (nce_ret < 0) {
2401 0 : kfree(nce);
2402 0 : ret = nce_ret;
2403 : }
2404 :
2405 0 : out:
2406 0 : return ret;
2407 : }
2408 :
2409 : /*
2410 : * Magic happens here. This function returns the first ref to an inode as it
2411 : * would look like while receiving the stream at this point in time.
2412 : * We walk the path up to the root. For every inode in between, we check if it
2413 : * was already processed/sent. If yes, we continue with the parent as found
2414 : * in send_root. If not, we continue with the parent as found in parent_root.
2415 : * If we encounter an inode that was deleted at this point in time, we use the
2416 : * inodes "orphan" name instead of the real name and stop. Same with new inodes
2417 : * that were not created yet and overwritten inodes/refs.
2418 : *
2419 : * When do we have orphan inodes:
2420 : * 1. When an inode is freshly created and thus no valid refs are available yet
2421 : * 2. When a directory lost all it's refs (deleted) but still has dir items
2422 : * inside which were not processed yet (pending for move/delete). If anyone
2423 : * tried to get the path to the dir items, it would get a path inside that
2424 : * orphan directory.
2425 : * 3. When an inode is moved around or gets new links, it may overwrite the ref
2426 : * of an unprocessed inode. If in that case the first ref would be
2427 : * overwritten, the overwritten inode gets "orphanized". Later when we
2428 : * process this overwritten inode, it is restored at a new place by moving
2429 : * the orphan inode.
2430 : *
2431 : * sctx->send_progress tells this function at which point in time receiving
2432 : * would be.
2433 : */
2434 0 : static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2435 : struct fs_path *dest)
2436 : {
2437 0 : int ret = 0;
2438 0 : struct fs_path *name = NULL;
2439 0 : u64 parent_inode = 0;
2440 0 : u64 parent_gen = 0;
2441 0 : int stop = 0;
2442 :
2443 0 : name = fs_path_alloc();
2444 0 : if (!name) {
2445 0 : ret = -ENOMEM;
2446 0 : goto out;
2447 : }
2448 :
2449 0 : dest->reversed = 1;
2450 0 : fs_path_reset(dest);
2451 :
2452 0 : while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2453 0 : struct waiting_dir_move *wdm;
2454 :
2455 0 : fs_path_reset(name);
2456 :
2457 0 : if (is_waiting_for_rm(sctx, ino, gen)) {
2458 0 : ret = gen_unique_name(sctx, ino, gen, name);
2459 0 : if (ret < 0)
2460 0 : goto out;
2461 0 : ret = fs_path_add_path(dest, name);
2462 0 : break;
2463 : }
2464 :
2465 0 : wdm = get_waiting_dir_move(sctx, ino);
2466 0 : if (wdm && wdm->orphanized) {
2467 0 : ret = gen_unique_name(sctx, ino, gen, name);
2468 0 : stop = 1;
2469 0 : } else if (wdm) {
2470 0 : ret = get_first_ref(sctx->parent_root, ino,
2471 : &parent_inode, &parent_gen, name);
2472 : } else {
2473 0 : ret = __get_cur_name_and_parent(sctx, ino, gen,
2474 : &parent_inode,
2475 : &parent_gen, name);
2476 0 : if (ret)
2477 0 : stop = 1;
2478 : }
2479 :
2480 0 : if (ret < 0)
2481 0 : goto out;
2482 :
2483 0 : ret = fs_path_add_path(dest, name);
2484 0 : if (ret < 0)
2485 0 : goto out;
2486 :
2487 0 : ino = parent_inode;
2488 0 : gen = parent_gen;
2489 : }
2490 :
2491 0 : out:
2492 0 : fs_path_free(name);
2493 0 : if (!ret)
2494 0 : fs_path_unreverse(dest);
2495 0 : return ret;
2496 : }
2497 :
2498 : /*
2499 : * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2500 : */
2501 0 : static int send_subvol_begin(struct send_ctx *sctx)
2502 : {
2503 0 : int ret;
2504 0 : struct btrfs_root *send_root = sctx->send_root;
2505 0 : struct btrfs_root *parent_root = sctx->parent_root;
2506 0 : struct btrfs_path *path;
2507 0 : struct btrfs_key key;
2508 0 : struct btrfs_root_ref *ref;
2509 0 : struct extent_buffer *leaf;
2510 0 : char *name = NULL;
2511 0 : int namelen;
2512 :
2513 0 : path = btrfs_alloc_path();
2514 0 : if (!path)
2515 : return -ENOMEM;
2516 :
2517 0 : name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2518 0 : if (!name) {
2519 0 : btrfs_free_path(path);
2520 0 : return -ENOMEM;
2521 : }
2522 :
2523 0 : key.objectid = send_root->root_key.objectid;
2524 0 : key.type = BTRFS_ROOT_BACKREF_KEY;
2525 0 : key.offset = 0;
2526 :
2527 0 : ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2528 : &key, path, 1, 0);
2529 0 : if (ret < 0)
2530 0 : goto out;
2531 0 : if (ret) {
2532 0 : ret = -ENOENT;
2533 0 : goto out;
2534 : }
2535 :
2536 0 : leaf = path->nodes[0];
2537 0 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2538 0 : if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2539 0 : key.objectid != send_root->root_key.objectid) {
2540 0 : ret = -ENOENT;
2541 0 : goto out;
2542 : }
2543 0 : ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2544 0 : namelen = btrfs_root_ref_name_len(leaf, ref);
2545 0 : read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2546 0 : btrfs_release_path(path);
2547 :
2548 0 : if (parent_root) {
2549 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2550 0 : if (ret < 0)
2551 0 : goto out;
2552 : } else {
2553 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2554 0 : if (ret < 0)
2555 0 : goto out;
2556 : }
2557 :
2558 0 : TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2559 :
2560 0 : if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2561 0 : TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2562 : sctx->send_root->root_item.received_uuid);
2563 : else
2564 0 : TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2565 : sctx->send_root->root_item.uuid);
2566 :
2567 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2568 : btrfs_root_ctransid(&sctx->send_root->root_item));
2569 0 : if (parent_root) {
2570 0 : if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2571 0 : TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2572 : parent_root->root_item.received_uuid);
2573 : else
2574 0 : TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2575 : parent_root->root_item.uuid);
2576 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2577 : btrfs_root_ctransid(&sctx->parent_root->root_item));
2578 : }
2579 :
2580 0 : ret = send_cmd(sctx);
2581 :
2582 0 : tlv_put_failure:
2583 0 : out:
2584 0 : btrfs_free_path(path);
2585 0 : kfree(name);
2586 0 : return ret;
2587 : }
2588 :
2589 0 : static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2590 : {
2591 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2592 0 : int ret = 0;
2593 0 : struct fs_path *p;
2594 :
2595 0 : btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2596 :
2597 0 : p = fs_path_alloc();
2598 0 : if (!p)
2599 : return -ENOMEM;
2600 :
2601 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2602 0 : if (ret < 0)
2603 0 : goto out;
2604 :
2605 0 : ret = get_cur_path(sctx, ino, gen, p);
2606 0 : if (ret < 0)
2607 0 : goto out;
2608 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2609 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2610 :
2611 0 : ret = send_cmd(sctx);
2612 :
2613 0 : tlv_put_failure:
2614 0 : out:
2615 0 : fs_path_free(p);
2616 0 : return ret;
2617 : }
2618 :
2619 0 : static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2620 : {
2621 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2622 0 : int ret = 0;
2623 0 : struct fs_path *p;
2624 :
2625 0 : btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2626 :
2627 0 : p = fs_path_alloc();
2628 0 : if (!p)
2629 : return -ENOMEM;
2630 :
2631 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2632 0 : if (ret < 0)
2633 0 : goto out;
2634 :
2635 0 : ret = get_cur_path(sctx, ino, gen, p);
2636 0 : if (ret < 0)
2637 0 : goto out;
2638 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2639 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2640 :
2641 0 : ret = send_cmd(sctx);
2642 :
2643 0 : tlv_put_failure:
2644 0 : out:
2645 0 : fs_path_free(p);
2646 0 : return ret;
2647 : }
2648 :
2649 0 : static int send_fileattr(struct send_ctx *sctx, u64 ino, u64 gen, u64 fileattr)
2650 : {
2651 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2652 0 : int ret = 0;
2653 0 : struct fs_path *p;
2654 :
2655 0 : if (sctx->proto < 2)
2656 : return 0;
2657 :
2658 0 : btrfs_debug(fs_info, "send_fileattr %llu fileattr=%llu", ino, fileattr);
2659 :
2660 0 : p = fs_path_alloc();
2661 0 : if (!p)
2662 : return -ENOMEM;
2663 :
2664 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_FILEATTR);
2665 0 : if (ret < 0)
2666 0 : goto out;
2667 :
2668 0 : ret = get_cur_path(sctx, ino, gen, p);
2669 0 : if (ret < 0)
2670 0 : goto out;
2671 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2672 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_FILEATTR, fileattr);
2673 :
2674 0 : ret = send_cmd(sctx);
2675 :
2676 0 : tlv_put_failure:
2677 0 : out:
2678 0 : fs_path_free(p);
2679 0 : return ret;
2680 : }
2681 :
2682 0 : static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2683 : {
2684 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2685 0 : int ret = 0;
2686 0 : struct fs_path *p;
2687 :
2688 0 : btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2689 : ino, uid, gid);
2690 :
2691 0 : p = fs_path_alloc();
2692 0 : if (!p)
2693 : return -ENOMEM;
2694 :
2695 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2696 0 : if (ret < 0)
2697 0 : goto out;
2698 :
2699 0 : ret = get_cur_path(sctx, ino, gen, p);
2700 0 : if (ret < 0)
2701 0 : goto out;
2702 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2703 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2704 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2705 :
2706 0 : ret = send_cmd(sctx);
2707 :
2708 0 : tlv_put_failure:
2709 0 : out:
2710 0 : fs_path_free(p);
2711 0 : return ret;
2712 : }
2713 :
2714 0 : static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2715 : {
2716 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2717 0 : int ret = 0;
2718 0 : struct fs_path *p = NULL;
2719 0 : struct btrfs_inode_item *ii;
2720 0 : struct btrfs_path *path = NULL;
2721 0 : struct extent_buffer *eb;
2722 0 : struct btrfs_key key;
2723 0 : int slot;
2724 :
2725 0 : btrfs_debug(fs_info, "send_utimes %llu", ino);
2726 :
2727 0 : p = fs_path_alloc();
2728 0 : if (!p)
2729 : return -ENOMEM;
2730 :
2731 0 : path = alloc_path_for_send();
2732 0 : if (!path) {
2733 0 : ret = -ENOMEM;
2734 0 : goto out;
2735 : }
2736 :
2737 0 : key.objectid = ino;
2738 0 : key.type = BTRFS_INODE_ITEM_KEY;
2739 0 : key.offset = 0;
2740 0 : ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2741 0 : if (ret > 0)
2742 : ret = -ENOENT;
2743 0 : if (ret < 0)
2744 0 : goto out;
2745 :
2746 0 : eb = path->nodes[0];
2747 0 : slot = path->slots[0];
2748 0 : ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2749 :
2750 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2751 0 : if (ret < 0)
2752 0 : goto out;
2753 :
2754 0 : ret = get_cur_path(sctx, ino, gen, p);
2755 0 : if (ret < 0)
2756 0 : goto out;
2757 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2758 0 : TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2759 0 : TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2760 0 : TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2761 0 : if (sctx->proto >= 2)
2762 0 : TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_OTIME, eb, &ii->otime);
2763 :
2764 0 : ret = send_cmd(sctx);
2765 :
2766 0 : tlv_put_failure:
2767 0 : out:
2768 0 : fs_path_free(p);
2769 0 : btrfs_free_path(path);
2770 0 : return ret;
2771 : }
2772 :
2773 : /*
2774 : * If the cache is full, we can't remove entries from it and do a call to
2775 : * send_utimes() for each respective inode, because we might be finishing
2776 : * processing an inode that is a directory and it just got renamed, and existing
2777 : * entries in the cache may refer to inodes that have the directory in their
2778 : * full path - in which case we would generate outdated paths (pre-rename)
2779 : * for the inodes that the cache entries point to. Instead of prunning the
2780 : * cache when inserting, do it after we finish processing each inode at
2781 : * finish_inode_if_needed().
2782 : */
2783 0 : static int cache_dir_utimes(struct send_ctx *sctx, u64 dir, u64 gen)
2784 : {
2785 0 : struct btrfs_lru_cache_entry *entry;
2786 0 : int ret;
2787 :
2788 0 : entry = btrfs_lru_cache_lookup(&sctx->dir_utimes_cache, dir, gen);
2789 0 : if (entry != NULL)
2790 : return 0;
2791 :
2792 : /* Caching is optional, don't fail if we can't allocate memory. */
2793 0 : entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2794 0 : if (!entry)
2795 0 : return send_utimes(sctx, dir, gen);
2796 :
2797 0 : entry->key = dir;
2798 0 : entry->gen = gen;
2799 :
2800 0 : ret = btrfs_lru_cache_store(&sctx->dir_utimes_cache, entry, GFP_KERNEL);
2801 0 : ASSERT(ret != -EEXIST);
2802 0 : if (ret) {
2803 0 : kfree(entry);
2804 0 : return send_utimes(sctx, dir, gen);
2805 : }
2806 :
2807 : return 0;
2808 : }
2809 :
2810 0 : static int trim_dir_utimes_cache(struct send_ctx *sctx)
2811 : {
2812 0 : while (btrfs_lru_cache_size(&sctx->dir_utimes_cache) >
2813 : SEND_MAX_DIR_UTIMES_CACHE_SIZE) {
2814 0 : struct btrfs_lru_cache_entry *lru;
2815 0 : int ret;
2816 :
2817 0 : lru = btrfs_lru_cache_lru_entry(&sctx->dir_utimes_cache);
2818 0 : ASSERT(lru != NULL);
2819 :
2820 0 : ret = send_utimes(sctx, lru->key, lru->gen);
2821 0 : if (ret)
2822 0 : return ret;
2823 :
2824 0 : btrfs_lru_cache_remove(&sctx->dir_utimes_cache, lru);
2825 : }
2826 :
2827 : return 0;
2828 : }
2829 :
2830 : /*
2831 : * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2832 : * a valid path yet because we did not process the refs yet. So, the inode
2833 : * is created as orphan.
2834 : */
2835 0 : static int send_create_inode(struct send_ctx *sctx, u64 ino)
2836 : {
2837 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2838 0 : int ret = 0;
2839 0 : struct fs_path *p;
2840 0 : int cmd;
2841 0 : struct btrfs_inode_info info;
2842 0 : u64 gen;
2843 0 : u64 mode;
2844 0 : u64 rdev;
2845 :
2846 0 : btrfs_debug(fs_info, "send_create_inode %llu", ino);
2847 :
2848 0 : p = fs_path_alloc();
2849 0 : if (!p)
2850 : return -ENOMEM;
2851 :
2852 0 : if (ino != sctx->cur_ino) {
2853 0 : ret = get_inode_info(sctx->send_root, ino, &info);
2854 0 : if (ret < 0)
2855 0 : goto out;
2856 0 : gen = info.gen;
2857 0 : mode = info.mode;
2858 0 : rdev = info.rdev;
2859 : } else {
2860 0 : gen = sctx->cur_inode_gen;
2861 0 : mode = sctx->cur_inode_mode;
2862 0 : rdev = sctx->cur_inode_rdev;
2863 : }
2864 :
2865 0 : if (S_ISREG(mode)) {
2866 : cmd = BTRFS_SEND_C_MKFILE;
2867 0 : } else if (S_ISDIR(mode)) {
2868 : cmd = BTRFS_SEND_C_MKDIR;
2869 0 : } else if (S_ISLNK(mode)) {
2870 : cmd = BTRFS_SEND_C_SYMLINK;
2871 0 : } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2872 : cmd = BTRFS_SEND_C_MKNOD;
2873 0 : } else if (S_ISFIFO(mode)) {
2874 : cmd = BTRFS_SEND_C_MKFIFO;
2875 0 : } else if (S_ISSOCK(mode)) {
2876 : cmd = BTRFS_SEND_C_MKSOCK;
2877 : } else {
2878 0 : btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2879 : (int)(mode & S_IFMT));
2880 0 : ret = -EOPNOTSUPP;
2881 0 : goto out;
2882 : }
2883 :
2884 0 : ret = begin_cmd(sctx, cmd);
2885 0 : if (ret < 0)
2886 0 : goto out;
2887 :
2888 0 : ret = gen_unique_name(sctx, ino, gen, p);
2889 0 : if (ret < 0)
2890 0 : goto out;
2891 :
2892 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2893 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2894 :
2895 0 : if (S_ISLNK(mode)) {
2896 0 : fs_path_reset(p);
2897 0 : ret = read_symlink(sctx->send_root, ino, p);
2898 0 : if (ret < 0)
2899 0 : goto out;
2900 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2901 0 : } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2902 0 : S_ISFIFO(mode) || S_ISSOCK(mode)) {
2903 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2904 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2905 : }
2906 :
2907 0 : ret = send_cmd(sctx);
2908 0 : if (ret < 0)
2909 0 : goto out;
2910 :
2911 :
2912 0 : tlv_put_failure:
2913 0 : out:
2914 0 : fs_path_free(p);
2915 0 : return ret;
2916 : }
2917 :
2918 0 : static void cache_dir_created(struct send_ctx *sctx, u64 dir)
2919 : {
2920 0 : struct btrfs_lru_cache_entry *entry;
2921 0 : int ret;
2922 :
2923 : /* Caching is optional, ignore any failures. */
2924 0 : entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2925 0 : if (!entry)
2926 : return;
2927 :
2928 0 : entry->key = dir;
2929 0 : entry->gen = 0;
2930 0 : ret = btrfs_lru_cache_store(&sctx->dir_created_cache, entry, GFP_KERNEL);
2931 0 : if (ret < 0)
2932 0 : kfree(entry);
2933 : }
2934 :
2935 : /*
2936 : * We need some special handling for inodes that get processed before the parent
2937 : * directory got created. See process_recorded_refs for details.
2938 : * This function does the check if we already created the dir out of order.
2939 : */
2940 0 : static int did_create_dir(struct send_ctx *sctx, u64 dir)
2941 : {
2942 0 : int ret = 0;
2943 0 : int iter_ret = 0;
2944 0 : struct btrfs_path *path = NULL;
2945 0 : struct btrfs_key key;
2946 0 : struct btrfs_key found_key;
2947 0 : struct btrfs_key di_key;
2948 0 : struct btrfs_dir_item *di;
2949 :
2950 0 : if (btrfs_lru_cache_lookup(&sctx->dir_created_cache, dir, 0))
2951 : return 1;
2952 :
2953 0 : path = alloc_path_for_send();
2954 0 : if (!path)
2955 : return -ENOMEM;
2956 :
2957 0 : key.objectid = dir;
2958 0 : key.type = BTRFS_DIR_INDEX_KEY;
2959 0 : key.offset = 0;
2960 :
2961 0 : btrfs_for_each_slot(sctx->send_root, &key, &found_key, path, iter_ret) {
2962 0 : struct extent_buffer *eb = path->nodes[0];
2963 :
2964 0 : if (found_key.objectid != key.objectid ||
2965 0 : found_key.type != key.type) {
2966 : ret = 0;
2967 : break;
2968 : }
2969 :
2970 0 : di = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dir_item);
2971 0 : btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2972 :
2973 0 : if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2974 0 : di_key.objectid < sctx->send_progress) {
2975 0 : ret = 1;
2976 0 : cache_dir_created(sctx, dir);
2977 0 : break;
2978 : }
2979 : }
2980 : /* Catch error found during iteration */
2981 0 : if (iter_ret < 0)
2982 0 : ret = iter_ret;
2983 :
2984 0 : btrfs_free_path(path);
2985 0 : return ret;
2986 : }
2987 :
2988 : /*
2989 : * Only creates the inode if it is:
2990 : * 1. Not a directory
2991 : * 2. Or a directory which was not created already due to out of order
2992 : * directories. See did_create_dir and process_recorded_refs for details.
2993 : */
2994 0 : static int send_create_inode_if_needed(struct send_ctx *sctx)
2995 : {
2996 0 : int ret;
2997 :
2998 0 : if (S_ISDIR(sctx->cur_inode_mode)) {
2999 0 : ret = did_create_dir(sctx, sctx->cur_ino);
3000 0 : if (ret < 0)
3001 : return ret;
3002 0 : else if (ret > 0)
3003 : return 0;
3004 : }
3005 :
3006 0 : ret = send_create_inode(sctx, sctx->cur_ino);
3007 :
3008 0 : if (ret == 0 && S_ISDIR(sctx->cur_inode_mode))
3009 0 : cache_dir_created(sctx, sctx->cur_ino);
3010 :
3011 : return ret;
3012 : }
3013 :
3014 : struct recorded_ref {
3015 : struct list_head list;
3016 : char *name;
3017 : struct fs_path *full_path;
3018 : u64 dir;
3019 : u64 dir_gen;
3020 : int name_len;
3021 : struct rb_node node;
3022 : struct rb_root *root;
3023 : };
3024 :
3025 0 : static struct recorded_ref *recorded_ref_alloc(void)
3026 : {
3027 0 : struct recorded_ref *ref;
3028 :
3029 0 : ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3030 0 : if (!ref)
3031 : return NULL;
3032 0 : RB_CLEAR_NODE(&ref->node);
3033 0 : INIT_LIST_HEAD(&ref->list);
3034 0 : return ref;
3035 : }
3036 :
3037 0 : static void recorded_ref_free(struct recorded_ref *ref)
3038 : {
3039 0 : if (!ref)
3040 : return;
3041 0 : if (!RB_EMPTY_NODE(&ref->node))
3042 0 : rb_erase(&ref->node, ref->root);
3043 0 : list_del(&ref->list);
3044 0 : fs_path_free(ref->full_path);
3045 0 : kfree(ref);
3046 : }
3047 :
3048 0 : static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
3049 : {
3050 0 : ref->full_path = path;
3051 0 : ref->name = (char *)kbasename(ref->full_path->start);
3052 0 : ref->name_len = ref->full_path->end - ref->name;
3053 0 : }
3054 :
3055 0 : static int dup_ref(struct recorded_ref *ref, struct list_head *list)
3056 : {
3057 0 : struct recorded_ref *new;
3058 :
3059 0 : new = recorded_ref_alloc();
3060 0 : if (!new)
3061 : return -ENOMEM;
3062 :
3063 0 : new->dir = ref->dir;
3064 0 : new->dir_gen = ref->dir_gen;
3065 0 : list_add_tail(&new->list, list);
3066 0 : return 0;
3067 : }
3068 :
3069 : static void __free_recorded_refs(struct list_head *head)
3070 : {
3071 0 : struct recorded_ref *cur;
3072 :
3073 0 : while (!list_empty(head)) {
3074 0 : cur = list_entry(head->next, struct recorded_ref, list);
3075 0 : recorded_ref_free(cur);
3076 : }
3077 : }
3078 :
3079 0 : static void free_recorded_refs(struct send_ctx *sctx)
3080 : {
3081 0 : __free_recorded_refs(&sctx->new_refs);
3082 0 : __free_recorded_refs(&sctx->deleted_refs);
3083 0 : }
3084 :
3085 : /*
3086 : * Renames/moves a file/dir to its orphan name. Used when the first
3087 : * ref of an unprocessed inode gets overwritten and for all non empty
3088 : * directories.
3089 : */
3090 0 : static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
3091 : struct fs_path *path)
3092 : {
3093 0 : int ret;
3094 0 : struct fs_path *orphan;
3095 :
3096 0 : orphan = fs_path_alloc();
3097 0 : if (!orphan)
3098 : return -ENOMEM;
3099 :
3100 0 : ret = gen_unique_name(sctx, ino, gen, orphan);
3101 0 : if (ret < 0)
3102 0 : goto out;
3103 :
3104 0 : ret = send_rename(sctx, path, orphan);
3105 :
3106 0 : out:
3107 0 : fs_path_free(orphan);
3108 0 : return ret;
3109 : }
3110 :
3111 0 : static struct orphan_dir_info *add_orphan_dir_info(struct send_ctx *sctx,
3112 : u64 dir_ino, u64 dir_gen)
3113 : {
3114 0 : struct rb_node **p = &sctx->orphan_dirs.rb_node;
3115 0 : struct rb_node *parent = NULL;
3116 0 : struct orphan_dir_info *entry, *odi;
3117 :
3118 0 : while (*p) {
3119 0 : parent = *p;
3120 0 : entry = rb_entry(parent, struct orphan_dir_info, node);
3121 0 : if (dir_ino < entry->ino)
3122 0 : p = &(*p)->rb_left;
3123 0 : else if (dir_ino > entry->ino)
3124 0 : p = &(*p)->rb_right;
3125 0 : else if (dir_gen < entry->gen)
3126 0 : p = &(*p)->rb_left;
3127 0 : else if (dir_gen > entry->gen)
3128 0 : p = &(*p)->rb_right;
3129 : else
3130 0 : return entry;
3131 : }
3132 :
3133 0 : odi = kmalloc(sizeof(*odi), GFP_KERNEL);
3134 0 : if (!odi)
3135 : return ERR_PTR(-ENOMEM);
3136 0 : odi->ino = dir_ino;
3137 0 : odi->gen = dir_gen;
3138 0 : odi->last_dir_index_offset = 0;
3139 0 : odi->dir_high_seq_ino = 0;
3140 :
3141 0 : rb_link_node(&odi->node, parent, p);
3142 0 : rb_insert_color(&odi->node, &sctx->orphan_dirs);
3143 0 : return odi;
3144 : }
3145 :
3146 0 : static struct orphan_dir_info *get_orphan_dir_info(struct send_ctx *sctx,
3147 : u64 dir_ino, u64 gen)
3148 : {
3149 0 : struct rb_node *n = sctx->orphan_dirs.rb_node;
3150 0 : struct orphan_dir_info *entry;
3151 :
3152 0 : while (n) {
3153 0 : entry = rb_entry(n, struct orphan_dir_info, node);
3154 0 : if (dir_ino < entry->ino)
3155 0 : n = n->rb_left;
3156 0 : else if (dir_ino > entry->ino)
3157 0 : n = n->rb_right;
3158 0 : else if (gen < entry->gen)
3159 0 : n = n->rb_left;
3160 0 : else if (gen > entry->gen)
3161 0 : n = n->rb_right;
3162 : else
3163 0 : return entry;
3164 : }
3165 : return NULL;
3166 : }
3167 :
3168 : static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen)
3169 : {
3170 0 : struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino, gen);
3171 :
3172 0 : return odi != NULL;
3173 : }
3174 :
3175 0 : static void free_orphan_dir_info(struct send_ctx *sctx,
3176 : struct orphan_dir_info *odi)
3177 : {
3178 0 : if (!odi)
3179 : return;
3180 0 : rb_erase(&odi->node, &sctx->orphan_dirs);
3181 0 : kfree(odi);
3182 : }
3183 :
3184 : /*
3185 : * Returns 1 if a directory can be removed at this point in time.
3186 : * We check this by iterating all dir items and checking if the inode behind
3187 : * the dir item was already processed.
3188 : */
3189 0 : static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen)
3190 : {
3191 0 : int ret = 0;
3192 0 : int iter_ret = 0;
3193 0 : struct btrfs_root *root = sctx->parent_root;
3194 0 : struct btrfs_path *path;
3195 0 : struct btrfs_key key;
3196 0 : struct btrfs_key found_key;
3197 0 : struct btrfs_key loc;
3198 0 : struct btrfs_dir_item *di;
3199 0 : struct orphan_dir_info *odi = NULL;
3200 0 : u64 dir_high_seq_ino = 0;
3201 0 : u64 last_dir_index_offset = 0;
3202 :
3203 : /*
3204 : * Don't try to rmdir the top/root subvolume dir.
3205 : */
3206 0 : if (dir == BTRFS_FIRST_FREE_OBJECTID)
3207 : return 0;
3208 :
3209 0 : odi = get_orphan_dir_info(sctx, dir, dir_gen);
3210 0 : if (odi && sctx->cur_ino < odi->dir_high_seq_ino)
3211 : return 0;
3212 :
3213 0 : path = alloc_path_for_send();
3214 0 : if (!path)
3215 : return -ENOMEM;
3216 :
3217 0 : if (!odi) {
3218 : /*
3219 : * Find the inode number associated with the last dir index
3220 : * entry. This is very likely the inode with the highest number
3221 : * of all inodes that have an entry in the directory. We can
3222 : * then use it to avoid future calls to can_rmdir(), when
3223 : * processing inodes with a lower number, from having to search
3224 : * the parent root b+tree for dir index keys.
3225 : */
3226 0 : key.objectid = dir;
3227 0 : key.type = BTRFS_DIR_INDEX_KEY;
3228 0 : key.offset = (u64)-1;
3229 :
3230 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3231 0 : if (ret < 0) {
3232 0 : goto out;
3233 0 : } else if (ret > 0) {
3234 : /* Can't happen, the root is never empty. */
3235 0 : ASSERT(path->slots[0] > 0);
3236 0 : if (WARN_ON(path->slots[0] == 0)) {
3237 0 : ret = -EUCLEAN;
3238 0 : goto out;
3239 : }
3240 0 : path->slots[0]--;
3241 : }
3242 :
3243 0 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3244 0 : if (key.objectid != dir || key.type != BTRFS_DIR_INDEX_KEY) {
3245 : /* No index keys, dir can be removed. */
3246 0 : ret = 1;
3247 0 : goto out;
3248 : }
3249 :
3250 0 : di = btrfs_item_ptr(path->nodes[0], path->slots[0],
3251 : struct btrfs_dir_item);
3252 0 : btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
3253 0 : dir_high_seq_ino = loc.objectid;
3254 0 : if (sctx->cur_ino < dir_high_seq_ino) {
3255 0 : ret = 0;
3256 0 : goto out;
3257 : }
3258 :
3259 0 : btrfs_release_path(path);
3260 : }
3261 :
3262 0 : key.objectid = dir;
3263 0 : key.type = BTRFS_DIR_INDEX_KEY;
3264 0 : key.offset = (odi ? odi->last_dir_index_offset : 0);
3265 :
3266 0 : btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
3267 0 : struct waiting_dir_move *dm;
3268 :
3269 0 : if (found_key.objectid != key.objectid ||
3270 0 : found_key.type != key.type)
3271 : break;
3272 :
3273 0 : di = btrfs_item_ptr(path->nodes[0], path->slots[0],
3274 : struct btrfs_dir_item);
3275 0 : btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
3276 :
3277 0 : dir_high_seq_ino = max(dir_high_seq_ino, loc.objectid);
3278 0 : last_dir_index_offset = found_key.offset;
3279 :
3280 0 : dm = get_waiting_dir_move(sctx, loc.objectid);
3281 0 : if (dm) {
3282 0 : dm->rmdir_ino = dir;
3283 0 : dm->rmdir_gen = dir_gen;
3284 0 : ret = 0;
3285 0 : goto out;
3286 : }
3287 :
3288 0 : if (loc.objectid > sctx->cur_ino) {
3289 0 : ret = 0;
3290 0 : goto out;
3291 : }
3292 : }
3293 0 : if (iter_ret < 0) {
3294 0 : ret = iter_ret;
3295 0 : goto out;
3296 : }
3297 0 : free_orphan_dir_info(sctx, odi);
3298 :
3299 0 : ret = 1;
3300 :
3301 0 : out:
3302 0 : btrfs_free_path(path);
3303 :
3304 0 : if (ret)
3305 : return ret;
3306 :
3307 0 : if (!odi) {
3308 0 : odi = add_orphan_dir_info(sctx, dir, dir_gen);
3309 0 : if (IS_ERR(odi))
3310 0 : return PTR_ERR(odi);
3311 :
3312 0 : odi->gen = dir_gen;
3313 : }
3314 :
3315 0 : odi->last_dir_index_offset = last_dir_index_offset;
3316 0 : odi->dir_high_seq_ino = max(odi->dir_high_seq_ino, dir_high_seq_ino);
3317 :
3318 0 : return 0;
3319 : }
3320 :
3321 0 : static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3322 : {
3323 0 : struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
3324 :
3325 0 : return entry != NULL;
3326 : }
3327 :
3328 0 : static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3329 : {
3330 0 : struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3331 0 : struct rb_node *parent = NULL;
3332 0 : struct waiting_dir_move *entry, *dm;
3333 :
3334 0 : dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3335 0 : if (!dm)
3336 : return -ENOMEM;
3337 0 : dm->ino = ino;
3338 0 : dm->rmdir_ino = 0;
3339 0 : dm->rmdir_gen = 0;
3340 0 : dm->orphanized = orphanized;
3341 :
3342 0 : while (*p) {
3343 0 : parent = *p;
3344 0 : entry = rb_entry(parent, struct waiting_dir_move, node);
3345 0 : if (ino < entry->ino) {
3346 0 : p = &(*p)->rb_left;
3347 0 : } else if (ino > entry->ino) {
3348 0 : p = &(*p)->rb_right;
3349 : } else {
3350 0 : kfree(dm);
3351 0 : return -EEXIST;
3352 : }
3353 : }
3354 :
3355 0 : rb_link_node(&dm->node, parent, p);
3356 0 : rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3357 0 : return 0;
3358 : }
3359 :
3360 : static struct waiting_dir_move *
3361 : get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3362 : {
3363 0 : struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3364 0 : struct waiting_dir_move *entry;
3365 :
3366 0 : while (n) {
3367 0 : entry = rb_entry(n, struct waiting_dir_move, node);
3368 0 : if (ino < entry->ino)
3369 0 : n = n->rb_left;
3370 0 : else if (ino > entry->ino)
3371 0 : n = n->rb_right;
3372 : else
3373 : return entry;
3374 : }
3375 : return NULL;
3376 : }
3377 :
3378 0 : static void free_waiting_dir_move(struct send_ctx *sctx,
3379 : struct waiting_dir_move *dm)
3380 : {
3381 0 : if (!dm)
3382 : return;
3383 0 : rb_erase(&dm->node, &sctx->waiting_dir_moves);
3384 0 : kfree(dm);
3385 : }
3386 :
3387 0 : static int add_pending_dir_move(struct send_ctx *sctx,
3388 : u64 ino,
3389 : u64 ino_gen,
3390 : u64 parent_ino,
3391 : struct list_head *new_refs,
3392 : struct list_head *deleted_refs,
3393 : const bool is_orphan)
3394 : {
3395 0 : struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3396 0 : struct rb_node *parent = NULL;
3397 0 : struct pending_dir_move *entry = NULL, *pm;
3398 0 : struct recorded_ref *cur;
3399 0 : int exists = 0;
3400 0 : int ret;
3401 :
3402 0 : pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3403 0 : if (!pm)
3404 : return -ENOMEM;
3405 0 : pm->parent_ino = parent_ino;
3406 0 : pm->ino = ino;
3407 0 : pm->gen = ino_gen;
3408 0 : INIT_LIST_HEAD(&pm->list);
3409 0 : INIT_LIST_HEAD(&pm->update_refs);
3410 0 : RB_CLEAR_NODE(&pm->node);
3411 :
3412 0 : while (*p) {
3413 0 : parent = *p;
3414 0 : entry = rb_entry(parent, struct pending_dir_move, node);
3415 0 : if (parent_ino < entry->parent_ino) {
3416 0 : p = &(*p)->rb_left;
3417 0 : } else if (parent_ino > entry->parent_ino) {
3418 0 : p = &(*p)->rb_right;
3419 : } else {
3420 : exists = 1;
3421 : break;
3422 : }
3423 : }
3424 :
3425 0 : list_for_each_entry(cur, deleted_refs, list) {
3426 0 : ret = dup_ref(cur, &pm->update_refs);
3427 0 : if (ret < 0)
3428 0 : goto out;
3429 : }
3430 0 : list_for_each_entry(cur, new_refs, list) {
3431 0 : ret = dup_ref(cur, &pm->update_refs);
3432 0 : if (ret < 0)
3433 0 : goto out;
3434 : }
3435 :
3436 0 : ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3437 0 : if (ret)
3438 0 : goto out;
3439 :
3440 0 : if (exists) {
3441 0 : list_add_tail(&pm->list, &entry->list);
3442 : } else {
3443 0 : rb_link_node(&pm->node, parent, p);
3444 0 : rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3445 : }
3446 : ret = 0;
3447 0 : out:
3448 0 : if (ret) {
3449 : __free_recorded_refs(&pm->update_refs);
3450 0 : kfree(pm);
3451 : }
3452 : return ret;
3453 : }
3454 :
3455 : static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3456 : u64 parent_ino)
3457 : {
3458 0 : struct rb_node *n = sctx->pending_dir_moves.rb_node;
3459 0 : struct pending_dir_move *entry;
3460 :
3461 0 : while (n) {
3462 0 : entry = rb_entry(n, struct pending_dir_move, node);
3463 0 : if (parent_ino < entry->parent_ino)
3464 0 : n = n->rb_left;
3465 0 : else if (parent_ino > entry->parent_ino)
3466 0 : n = n->rb_right;
3467 : else
3468 : return entry;
3469 : }
3470 : return NULL;
3471 : }
3472 :
3473 0 : static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3474 : u64 ino, u64 gen, u64 *ancestor_ino)
3475 : {
3476 0 : int ret = 0;
3477 0 : u64 parent_inode = 0;
3478 0 : u64 parent_gen = 0;
3479 0 : u64 start_ino = ino;
3480 :
3481 0 : *ancestor_ino = 0;
3482 0 : while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3483 0 : fs_path_reset(name);
3484 :
3485 0 : if (is_waiting_for_rm(sctx, ino, gen))
3486 : break;
3487 0 : if (is_waiting_for_move(sctx, ino)) {
3488 0 : if (*ancestor_ino == 0)
3489 0 : *ancestor_ino = ino;
3490 0 : ret = get_first_ref(sctx->parent_root, ino,
3491 : &parent_inode, &parent_gen, name);
3492 : } else {
3493 0 : ret = __get_cur_name_and_parent(sctx, ino, gen,
3494 : &parent_inode,
3495 : &parent_gen, name);
3496 0 : if (ret > 0) {
3497 : ret = 0;
3498 : break;
3499 : }
3500 : }
3501 0 : if (ret < 0)
3502 : break;
3503 0 : if (parent_inode == start_ino) {
3504 0 : ret = 1;
3505 0 : if (*ancestor_ino == 0)
3506 0 : *ancestor_ino = ino;
3507 : break;
3508 : }
3509 0 : ino = parent_inode;
3510 0 : gen = parent_gen;
3511 : }
3512 0 : return ret;
3513 : }
3514 :
3515 0 : static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3516 : {
3517 0 : struct fs_path *from_path = NULL;
3518 0 : struct fs_path *to_path = NULL;
3519 0 : struct fs_path *name = NULL;
3520 0 : u64 orig_progress = sctx->send_progress;
3521 0 : struct recorded_ref *cur;
3522 0 : u64 parent_ino, parent_gen;
3523 0 : struct waiting_dir_move *dm = NULL;
3524 0 : u64 rmdir_ino = 0;
3525 0 : u64 rmdir_gen;
3526 0 : u64 ancestor;
3527 0 : bool is_orphan;
3528 0 : int ret;
3529 :
3530 0 : name = fs_path_alloc();
3531 0 : from_path = fs_path_alloc();
3532 0 : if (!name || !from_path) {
3533 0 : ret = -ENOMEM;
3534 0 : goto out;
3535 : }
3536 :
3537 0 : dm = get_waiting_dir_move(sctx, pm->ino);
3538 0 : ASSERT(dm);
3539 0 : rmdir_ino = dm->rmdir_ino;
3540 0 : rmdir_gen = dm->rmdir_gen;
3541 0 : is_orphan = dm->orphanized;
3542 0 : free_waiting_dir_move(sctx, dm);
3543 :
3544 0 : if (is_orphan) {
3545 0 : ret = gen_unique_name(sctx, pm->ino,
3546 : pm->gen, from_path);
3547 : } else {
3548 0 : ret = get_first_ref(sctx->parent_root, pm->ino,
3549 : &parent_ino, &parent_gen, name);
3550 0 : if (ret < 0)
3551 0 : goto out;
3552 0 : ret = get_cur_path(sctx, parent_ino, parent_gen,
3553 : from_path);
3554 0 : if (ret < 0)
3555 0 : goto out;
3556 0 : ret = fs_path_add_path(from_path, name);
3557 : }
3558 0 : if (ret < 0)
3559 0 : goto out;
3560 :
3561 0 : sctx->send_progress = sctx->cur_ino + 1;
3562 0 : ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3563 0 : if (ret < 0)
3564 0 : goto out;
3565 0 : if (ret) {
3566 0 : LIST_HEAD(deleted_refs);
3567 0 : ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3568 0 : ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3569 : &pm->update_refs, &deleted_refs,
3570 : is_orphan);
3571 0 : if (ret < 0)
3572 0 : goto out;
3573 0 : if (rmdir_ino) {
3574 0 : dm = get_waiting_dir_move(sctx, pm->ino);
3575 0 : ASSERT(dm);
3576 0 : dm->rmdir_ino = rmdir_ino;
3577 0 : dm->rmdir_gen = rmdir_gen;
3578 : }
3579 0 : goto out;
3580 : }
3581 0 : fs_path_reset(name);
3582 0 : to_path = name;
3583 0 : name = NULL;
3584 0 : ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3585 0 : if (ret < 0)
3586 0 : goto out;
3587 :
3588 0 : ret = send_rename(sctx, from_path, to_path);
3589 0 : if (ret < 0)
3590 0 : goto out;
3591 :
3592 0 : if (rmdir_ino) {
3593 0 : struct orphan_dir_info *odi;
3594 0 : u64 gen;
3595 :
3596 0 : odi = get_orphan_dir_info(sctx, rmdir_ino, rmdir_gen);
3597 0 : if (!odi) {
3598 : /* already deleted */
3599 0 : goto finish;
3600 : }
3601 0 : gen = odi->gen;
3602 :
3603 0 : ret = can_rmdir(sctx, rmdir_ino, gen);
3604 0 : if (ret < 0)
3605 0 : goto out;
3606 0 : if (!ret)
3607 0 : goto finish;
3608 :
3609 0 : name = fs_path_alloc();
3610 0 : if (!name) {
3611 0 : ret = -ENOMEM;
3612 0 : goto out;
3613 : }
3614 0 : ret = get_cur_path(sctx, rmdir_ino, gen, name);
3615 0 : if (ret < 0)
3616 0 : goto out;
3617 0 : ret = send_rmdir(sctx, name);
3618 0 : if (ret < 0)
3619 0 : goto out;
3620 : }
3621 :
3622 0 : finish:
3623 0 : ret = cache_dir_utimes(sctx, pm->ino, pm->gen);
3624 0 : if (ret < 0)
3625 0 : goto out;
3626 :
3627 : /*
3628 : * After rename/move, need to update the utimes of both new parent(s)
3629 : * and old parent(s).
3630 : */
3631 0 : list_for_each_entry(cur, &pm->update_refs, list) {
3632 : /*
3633 : * The parent inode might have been deleted in the send snapshot
3634 : */
3635 0 : ret = get_inode_info(sctx->send_root, cur->dir, NULL);
3636 0 : if (ret == -ENOENT) {
3637 0 : ret = 0;
3638 0 : continue;
3639 : }
3640 0 : if (ret < 0)
3641 0 : goto out;
3642 :
3643 0 : ret = cache_dir_utimes(sctx, cur->dir, cur->dir_gen);
3644 0 : if (ret < 0)
3645 0 : goto out;
3646 : }
3647 :
3648 0 : out:
3649 0 : fs_path_free(name);
3650 0 : fs_path_free(from_path);
3651 0 : fs_path_free(to_path);
3652 0 : sctx->send_progress = orig_progress;
3653 :
3654 0 : return ret;
3655 : }
3656 :
3657 0 : static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3658 : {
3659 0 : if (!list_empty(&m->list))
3660 0 : list_del(&m->list);
3661 0 : if (!RB_EMPTY_NODE(&m->node))
3662 0 : rb_erase(&m->node, &sctx->pending_dir_moves);
3663 0 : __free_recorded_refs(&m->update_refs);
3664 0 : kfree(m);
3665 0 : }
3666 :
3667 0 : static void tail_append_pending_moves(struct send_ctx *sctx,
3668 : struct pending_dir_move *moves,
3669 : struct list_head *stack)
3670 : {
3671 0 : if (list_empty(&moves->list)) {
3672 0 : list_add_tail(&moves->list, stack);
3673 : } else {
3674 0 : LIST_HEAD(list);
3675 0 : list_splice_init(&moves->list, &list);
3676 0 : list_add_tail(&moves->list, stack);
3677 0 : list_splice_tail(&list, stack);
3678 : }
3679 0 : if (!RB_EMPTY_NODE(&moves->node)) {
3680 0 : rb_erase(&moves->node, &sctx->pending_dir_moves);
3681 0 : RB_CLEAR_NODE(&moves->node);
3682 : }
3683 0 : }
3684 :
3685 0 : static int apply_children_dir_moves(struct send_ctx *sctx)
3686 : {
3687 0 : struct pending_dir_move *pm;
3688 0 : struct list_head stack;
3689 0 : u64 parent_ino = sctx->cur_ino;
3690 0 : int ret = 0;
3691 :
3692 0 : pm = get_pending_dir_moves(sctx, parent_ino);
3693 0 : if (!pm)
3694 : return 0;
3695 :
3696 0 : INIT_LIST_HEAD(&stack);
3697 0 : tail_append_pending_moves(sctx, pm, &stack);
3698 :
3699 0 : while (!list_empty(&stack)) {
3700 0 : pm = list_first_entry(&stack, struct pending_dir_move, list);
3701 0 : parent_ino = pm->ino;
3702 0 : ret = apply_dir_move(sctx, pm);
3703 0 : free_pending_move(sctx, pm);
3704 0 : if (ret)
3705 0 : goto out;
3706 0 : pm = get_pending_dir_moves(sctx, parent_ino);
3707 0 : if (pm)
3708 0 : tail_append_pending_moves(sctx, pm, &stack);
3709 : }
3710 : return 0;
3711 :
3712 : out:
3713 0 : while (!list_empty(&stack)) {
3714 0 : pm = list_first_entry(&stack, struct pending_dir_move, list);
3715 0 : free_pending_move(sctx, pm);
3716 : }
3717 : return ret;
3718 : }
3719 :
3720 : /*
3721 : * We might need to delay a directory rename even when no ancestor directory
3722 : * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3723 : * renamed. This happens when we rename a directory to the old name (the name
3724 : * in the parent root) of some other unrelated directory that got its rename
3725 : * delayed due to some ancestor with higher number that got renamed.
3726 : *
3727 : * Example:
3728 : *
3729 : * Parent snapshot:
3730 : * . (ino 256)
3731 : * |---- a/ (ino 257)
3732 : * | |---- file (ino 260)
3733 : * |
3734 : * |---- b/ (ino 258)
3735 : * |---- c/ (ino 259)
3736 : *
3737 : * Send snapshot:
3738 : * . (ino 256)
3739 : * |---- a/ (ino 258)
3740 : * |---- x/ (ino 259)
3741 : * |---- y/ (ino 257)
3742 : * |----- file (ino 260)
3743 : *
3744 : * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3745 : * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3746 : * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3747 : * must issue is:
3748 : *
3749 : * 1 - rename 259 from 'c' to 'x'
3750 : * 2 - rename 257 from 'a' to 'x/y'
3751 : * 3 - rename 258 from 'b' to 'a'
3752 : *
3753 : * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3754 : * be done right away and < 0 on error.
3755 : */
3756 0 : static int wait_for_dest_dir_move(struct send_ctx *sctx,
3757 : struct recorded_ref *parent_ref,
3758 : const bool is_orphan)
3759 : {
3760 0 : struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3761 0 : struct btrfs_path *path;
3762 0 : struct btrfs_key key;
3763 0 : struct btrfs_key di_key;
3764 0 : struct btrfs_dir_item *di;
3765 0 : u64 left_gen;
3766 0 : u64 right_gen;
3767 0 : int ret = 0;
3768 0 : struct waiting_dir_move *wdm;
3769 :
3770 0 : if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3771 : return 0;
3772 :
3773 0 : path = alloc_path_for_send();
3774 0 : if (!path)
3775 : return -ENOMEM;
3776 :
3777 0 : key.objectid = parent_ref->dir;
3778 0 : key.type = BTRFS_DIR_ITEM_KEY;
3779 0 : key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3780 :
3781 0 : ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3782 0 : if (ret < 0) {
3783 0 : goto out;
3784 0 : } else if (ret > 0) {
3785 0 : ret = 0;
3786 0 : goto out;
3787 : }
3788 :
3789 0 : di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3790 : parent_ref->name_len);
3791 0 : if (!di) {
3792 0 : ret = 0;
3793 0 : goto out;
3794 : }
3795 : /*
3796 : * di_key.objectid has the number of the inode that has a dentry in the
3797 : * parent directory with the same name that sctx->cur_ino is being
3798 : * renamed to. We need to check if that inode is in the send root as
3799 : * well and if it is currently marked as an inode with a pending rename,
3800 : * if it is, we need to delay the rename of sctx->cur_ino as well, so
3801 : * that it happens after that other inode is renamed.
3802 : */
3803 0 : btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3804 0 : if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3805 0 : ret = 0;
3806 0 : goto out;
3807 : }
3808 :
3809 0 : ret = get_inode_gen(sctx->parent_root, di_key.objectid, &left_gen);
3810 0 : if (ret < 0)
3811 0 : goto out;
3812 0 : ret = get_inode_gen(sctx->send_root, di_key.objectid, &right_gen);
3813 0 : if (ret < 0) {
3814 0 : if (ret == -ENOENT)
3815 0 : ret = 0;
3816 0 : goto out;
3817 : }
3818 :
3819 : /* Different inode, no need to delay the rename of sctx->cur_ino */
3820 0 : if (right_gen != left_gen) {
3821 0 : ret = 0;
3822 0 : goto out;
3823 : }
3824 :
3825 0 : wdm = get_waiting_dir_move(sctx, di_key.objectid);
3826 0 : if (wdm && !wdm->orphanized) {
3827 0 : ret = add_pending_dir_move(sctx,
3828 : sctx->cur_ino,
3829 : sctx->cur_inode_gen,
3830 : di_key.objectid,
3831 : &sctx->new_refs,
3832 : &sctx->deleted_refs,
3833 : is_orphan);
3834 0 : if (!ret)
3835 0 : ret = 1;
3836 : }
3837 0 : out:
3838 0 : btrfs_free_path(path);
3839 0 : return ret;
3840 : }
3841 :
3842 : /*
3843 : * Check if inode ino2, or any of its ancestors, is inode ino1.
3844 : * Return 1 if true, 0 if false and < 0 on error.
3845 : */
3846 0 : static int check_ino_in_path(struct btrfs_root *root,
3847 : const u64 ino1,
3848 : const u64 ino1_gen,
3849 : const u64 ino2,
3850 : const u64 ino2_gen,
3851 : struct fs_path *fs_path)
3852 : {
3853 0 : u64 ino = ino2;
3854 :
3855 0 : if (ino1 == ino2)
3856 0 : return ino1_gen == ino2_gen;
3857 :
3858 0 : while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3859 0 : u64 parent;
3860 0 : u64 parent_gen;
3861 0 : int ret;
3862 :
3863 0 : fs_path_reset(fs_path);
3864 0 : ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3865 0 : if (ret < 0)
3866 0 : return ret;
3867 0 : if (parent == ino1)
3868 0 : return parent_gen == ino1_gen;
3869 0 : ino = parent;
3870 : }
3871 : return 0;
3872 : }
3873 :
3874 : /*
3875 : * Check if inode ino1 is an ancestor of inode ino2 in the given root for any
3876 : * possible path (in case ino2 is not a directory and has multiple hard links).
3877 : * Return 1 if true, 0 if false and < 0 on error.
3878 : */
3879 0 : static int is_ancestor(struct btrfs_root *root,
3880 : const u64 ino1,
3881 : const u64 ino1_gen,
3882 : const u64 ino2,
3883 : struct fs_path *fs_path)
3884 : {
3885 0 : bool free_fs_path = false;
3886 0 : int ret = 0;
3887 0 : int iter_ret = 0;
3888 0 : struct btrfs_path *path = NULL;
3889 0 : struct btrfs_key key;
3890 :
3891 0 : if (!fs_path) {
3892 0 : fs_path = fs_path_alloc();
3893 0 : if (!fs_path)
3894 : return -ENOMEM;
3895 : free_fs_path = true;
3896 : }
3897 :
3898 0 : path = alloc_path_for_send();
3899 0 : if (!path) {
3900 0 : ret = -ENOMEM;
3901 0 : goto out;
3902 : }
3903 :
3904 0 : key.objectid = ino2;
3905 0 : key.type = BTRFS_INODE_REF_KEY;
3906 0 : key.offset = 0;
3907 :
3908 0 : btrfs_for_each_slot(root, &key, &key, path, iter_ret) {
3909 0 : struct extent_buffer *leaf = path->nodes[0];
3910 0 : int slot = path->slots[0];
3911 0 : u32 cur_offset = 0;
3912 0 : u32 item_size;
3913 :
3914 0 : if (key.objectid != ino2)
3915 : break;
3916 0 : if (key.type != BTRFS_INODE_REF_KEY &&
3917 : key.type != BTRFS_INODE_EXTREF_KEY)
3918 : break;
3919 :
3920 0 : item_size = btrfs_item_size(leaf, slot);
3921 0 : while (cur_offset < item_size) {
3922 0 : u64 parent;
3923 0 : u64 parent_gen;
3924 :
3925 0 : if (key.type == BTRFS_INODE_EXTREF_KEY) {
3926 0 : unsigned long ptr;
3927 0 : struct btrfs_inode_extref *extref;
3928 :
3929 0 : ptr = btrfs_item_ptr_offset(leaf, slot);
3930 0 : extref = (struct btrfs_inode_extref *)
3931 0 : (ptr + cur_offset);
3932 0 : parent = btrfs_inode_extref_parent(leaf,
3933 : extref);
3934 0 : cur_offset += sizeof(*extref);
3935 0 : cur_offset += btrfs_inode_extref_name_len(leaf,
3936 : extref);
3937 : } else {
3938 0 : parent = key.offset;
3939 0 : cur_offset = item_size;
3940 : }
3941 :
3942 0 : ret = get_inode_gen(root, parent, &parent_gen);
3943 0 : if (ret < 0)
3944 0 : goto out;
3945 0 : ret = check_ino_in_path(root, ino1, ino1_gen,
3946 : parent, parent_gen, fs_path);
3947 0 : if (ret)
3948 0 : goto out;
3949 : }
3950 : }
3951 0 : ret = 0;
3952 0 : if (iter_ret < 0)
3953 : ret = iter_ret;
3954 :
3955 0 : out:
3956 0 : btrfs_free_path(path);
3957 0 : if (free_fs_path)
3958 0 : fs_path_free(fs_path);
3959 : return ret;
3960 : }
3961 :
3962 0 : static int wait_for_parent_move(struct send_ctx *sctx,
3963 : struct recorded_ref *parent_ref,
3964 : const bool is_orphan)
3965 : {
3966 0 : int ret = 0;
3967 0 : u64 ino = parent_ref->dir;
3968 0 : u64 ino_gen = parent_ref->dir_gen;
3969 0 : u64 parent_ino_before, parent_ino_after;
3970 0 : struct fs_path *path_before = NULL;
3971 0 : struct fs_path *path_after = NULL;
3972 0 : int len1, len2;
3973 :
3974 0 : path_after = fs_path_alloc();
3975 0 : path_before = fs_path_alloc();
3976 0 : if (!path_after || !path_before) {
3977 0 : ret = -ENOMEM;
3978 0 : goto out;
3979 : }
3980 :
3981 : /*
3982 : * Our current directory inode may not yet be renamed/moved because some
3983 : * ancestor (immediate or not) has to be renamed/moved first. So find if
3984 : * such ancestor exists and make sure our own rename/move happens after
3985 : * that ancestor is processed to avoid path build infinite loops (done
3986 : * at get_cur_path()).
3987 : */
3988 0 : while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3989 0 : u64 parent_ino_after_gen;
3990 :
3991 0 : if (is_waiting_for_move(sctx, ino)) {
3992 : /*
3993 : * If the current inode is an ancestor of ino in the
3994 : * parent root, we need to delay the rename of the
3995 : * current inode, otherwise don't delayed the rename
3996 : * because we can end up with a circular dependency
3997 : * of renames, resulting in some directories never
3998 : * getting the respective rename operations issued in
3999 : * the send stream or getting into infinite path build
4000 : * loops.
4001 : */
4002 0 : ret = is_ancestor(sctx->parent_root,
4003 : sctx->cur_ino, sctx->cur_inode_gen,
4004 : ino, path_before);
4005 0 : if (ret)
4006 : break;
4007 : }
4008 :
4009 0 : fs_path_reset(path_before);
4010 0 : fs_path_reset(path_after);
4011 :
4012 0 : ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
4013 : &parent_ino_after_gen, path_after);
4014 0 : if (ret < 0)
4015 0 : goto out;
4016 0 : ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
4017 : NULL, path_before);
4018 0 : if (ret < 0 && ret != -ENOENT) {
4019 0 : goto out;
4020 0 : } else if (ret == -ENOENT) {
4021 : ret = 0;
4022 : break;
4023 : }
4024 :
4025 0 : len1 = fs_path_len(path_before);
4026 0 : len2 = fs_path_len(path_after);
4027 0 : if (ino > sctx->cur_ino &&
4028 0 : (parent_ino_before != parent_ino_after || len1 != len2 ||
4029 0 : memcmp(path_before->start, path_after->start, len1))) {
4030 0 : u64 parent_ino_gen;
4031 :
4032 0 : ret = get_inode_gen(sctx->parent_root, ino, &parent_ino_gen);
4033 0 : if (ret < 0)
4034 0 : goto out;
4035 0 : if (ino_gen == parent_ino_gen) {
4036 0 : ret = 1;
4037 0 : break;
4038 : }
4039 : }
4040 0 : ino = parent_ino_after;
4041 0 : ino_gen = parent_ino_after_gen;
4042 : }
4043 :
4044 0 : out:
4045 0 : fs_path_free(path_before);
4046 0 : fs_path_free(path_after);
4047 :
4048 0 : if (ret == 1) {
4049 0 : ret = add_pending_dir_move(sctx,
4050 : sctx->cur_ino,
4051 : sctx->cur_inode_gen,
4052 : ino,
4053 : &sctx->new_refs,
4054 : &sctx->deleted_refs,
4055 : is_orphan);
4056 0 : if (!ret)
4057 0 : ret = 1;
4058 : }
4059 :
4060 0 : return ret;
4061 : }
4062 :
4063 0 : static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
4064 : {
4065 0 : int ret;
4066 0 : struct fs_path *new_path;
4067 :
4068 : /*
4069 : * Our reference's name member points to its full_path member string, so
4070 : * we use here a new path.
4071 : */
4072 0 : new_path = fs_path_alloc();
4073 0 : if (!new_path)
4074 : return -ENOMEM;
4075 :
4076 0 : ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
4077 0 : if (ret < 0) {
4078 0 : fs_path_free(new_path);
4079 0 : return ret;
4080 : }
4081 0 : ret = fs_path_add(new_path, ref->name, ref->name_len);
4082 0 : if (ret < 0) {
4083 0 : fs_path_free(new_path);
4084 0 : return ret;
4085 : }
4086 :
4087 0 : fs_path_free(ref->full_path);
4088 0 : set_ref_path(ref, new_path);
4089 :
4090 0 : return 0;
4091 : }
4092 :
4093 : /*
4094 : * When processing the new references for an inode we may orphanize an existing
4095 : * directory inode because its old name conflicts with one of the new references
4096 : * of the current inode. Later, when processing another new reference of our
4097 : * inode, we might need to orphanize another inode, but the path we have in the
4098 : * reference reflects the pre-orphanization name of the directory we previously
4099 : * orphanized. For example:
4100 : *
4101 : * parent snapshot looks like:
4102 : *
4103 : * . (ino 256)
4104 : * |----- f1 (ino 257)
4105 : * |----- f2 (ino 258)
4106 : * |----- d1/ (ino 259)
4107 : * |----- d2/ (ino 260)
4108 : *
4109 : * send snapshot looks like:
4110 : *
4111 : * . (ino 256)
4112 : * |----- d1 (ino 258)
4113 : * |----- f2/ (ino 259)
4114 : * |----- f2_link/ (ino 260)
4115 : * | |----- f1 (ino 257)
4116 : * |
4117 : * |----- d2 (ino 258)
4118 : *
4119 : * When processing inode 257 we compute the name for inode 259 as "d1", and we
4120 : * cache it in the name cache. Later when we start processing inode 258, when
4121 : * collecting all its new references we set a full path of "d1/d2" for its new
4122 : * reference with name "d2". When we start processing the new references we
4123 : * start by processing the new reference with name "d1", and this results in
4124 : * orphanizing inode 259, since its old reference causes a conflict. Then we
4125 : * move on the next new reference, with name "d2", and we find out we must
4126 : * orphanize inode 260, as its old reference conflicts with ours - but for the
4127 : * orphanization we use a source path corresponding to the path we stored in the
4128 : * new reference, which is "d1/d2" and not "o259-6-0/d2" - this makes the
4129 : * receiver fail since the path component "d1/" no longer exists, it was renamed
4130 : * to "o259-6-0/" when processing the previous new reference. So in this case we
4131 : * must recompute the path in the new reference and use it for the new
4132 : * orphanization operation.
4133 : */
4134 0 : static int refresh_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
4135 : {
4136 0 : char *name;
4137 0 : int ret;
4138 :
4139 0 : name = kmemdup(ref->name, ref->name_len, GFP_KERNEL);
4140 0 : if (!name)
4141 : return -ENOMEM;
4142 :
4143 0 : fs_path_reset(ref->full_path);
4144 0 : ret = get_cur_path(sctx, ref->dir, ref->dir_gen, ref->full_path);
4145 0 : if (ret < 0)
4146 0 : goto out;
4147 :
4148 0 : ret = fs_path_add(ref->full_path, name, ref->name_len);
4149 0 : if (ret < 0)
4150 0 : goto out;
4151 :
4152 : /* Update the reference's base name pointer. */
4153 0 : set_ref_path(ref, ref->full_path);
4154 0 : out:
4155 0 : kfree(name);
4156 0 : return ret;
4157 : }
4158 :
4159 : /*
4160 : * This does all the move/link/unlink/rmdir magic.
4161 : */
4162 0 : static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
4163 : {
4164 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4165 0 : int ret = 0;
4166 0 : struct recorded_ref *cur;
4167 0 : struct recorded_ref *cur2;
4168 0 : struct list_head check_dirs;
4169 0 : struct fs_path *valid_path = NULL;
4170 0 : u64 ow_inode = 0;
4171 0 : u64 ow_gen;
4172 0 : u64 ow_mode;
4173 0 : int did_overwrite = 0;
4174 0 : int is_orphan = 0;
4175 0 : u64 last_dir_ino_rm = 0;
4176 0 : bool can_rename = true;
4177 0 : bool orphanized_dir = false;
4178 0 : bool orphanized_ancestor = false;
4179 :
4180 0 : btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
4181 :
4182 : /*
4183 : * This should never happen as the root dir always has the same ref
4184 : * which is always '..'
4185 : */
4186 0 : BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
4187 0 : INIT_LIST_HEAD(&check_dirs);
4188 :
4189 0 : valid_path = fs_path_alloc();
4190 0 : if (!valid_path) {
4191 0 : ret = -ENOMEM;
4192 0 : goto out;
4193 : }
4194 :
4195 : /*
4196 : * First, check if the first ref of the current inode was overwritten
4197 : * before. If yes, we know that the current inode was already orphanized
4198 : * and thus use the orphan name. If not, we can use get_cur_path to
4199 : * get the path of the first ref as it would like while receiving at
4200 : * this point in time.
4201 : * New inodes are always orphan at the beginning, so force to use the
4202 : * orphan name in this case.
4203 : * The first ref is stored in valid_path and will be updated if it
4204 : * gets moved around.
4205 : */
4206 0 : if (!sctx->cur_inode_new) {
4207 0 : ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
4208 : sctx->cur_inode_gen);
4209 0 : if (ret < 0)
4210 0 : goto out;
4211 0 : if (ret)
4212 0 : did_overwrite = 1;
4213 : }
4214 0 : if (sctx->cur_inode_new || did_overwrite) {
4215 0 : ret = gen_unique_name(sctx, sctx->cur_ino,
4216 : sctx->cur_inode_gen, valid_path);
4217 0 : if (ret < 0)
4218 0 : goto out;
4219 : is_orphan = 1;
4220 : } else {
4221 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4222 : valid_path);
4223 0 : if (ret < 0)
4224 0 : goto out;
4225 : }
4226 :
4227 : /*
4228 : * Before doing any rename and link operations, do a first pass on the
4229 : * new references to orphanize any unprocessed inodes that may have a
4230 : * reference that conflicts with one of the new references of the current
4231 : * inode. This needs to happen first because a new reference may conflict
4232 : * with the old reference of a parent directory, so we must make sure
4233 : * that the path used for link and rename commands don't use an
4234 : * orphanized name when an ancestor was not yet orphanized.
4235 : *
4236 : * Example:
4237 : *
4238 : * Parent snapshot:
4239 : *
4240 : * . (ino 256)
4241 : * |----- testdir/ (ino 259)
4242 : * | |----- a (ino 257)
4243 : * |
4244 : * |----- b (ino 258)
4245 : *
4246 : * Send snapshot:
4247 : *
4248 : * . (ino 256)
4249 : * |----- testdir_2/ (ino 259)
4250 : * | |----- a (ino 260)
4251 : * |
4252 : * |----- testdir (ino 257)
4253 : * |----- b (ino 257)
4254 : * |----- b2 (ino 258)
4255 : *
4256 : * Processing the new reference for inode 257 with name "b" may happen
4257 : * before processing the new reference with name "testdir". If so, we
4258 : * must make sure that by the time we send a link command to create the
4259 : * hard link "b", inode 259 was already orphanized, since the generated
4260 : * path in "valid_path" already contains the orphanized name for 259.
4261 : * We are processing inode 257, so only later when processing 259 we do
4262 : * the rename operation to change its temporary (orphanized) name to
4263 : * "testdir_2".
4264 : */
4265 0 : list_for_each_entry(cur, &sctx->new_refs, list) {
4266 0 : ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL);
4267 0 : if (ret < 0)
4268 0 : goto out;
4269 0 : if (ret == inode_state_will_create)
4270 0 : continue;
4271 :
4272 : /*
4273 : * Check if this new ref would overwrite the first ref of another
4274 : * unprocessed inode. If yes, orphanize the overwritten inode.
4275 : * If we find an overwritten ref that is not the first ref,
4276 : * simply unlink it.
4277 : */
4278 0 : ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4279 0 : cur->name, cur->name_len,
4280 : &ow_inode, &ow_gen, &ow_mode);
4281 0 : if (ret < 0)
4282 0 : goto out;
4283 0 : if (ret) {
4284 0 : ret = is_first_ref(sctx->parent_root,
4285 0 : ow_inode, cur->dir, cur->name,
4286 : cur->name_len);
4287 0 : if (ret < 0)
4288 0 : goto out;
4289 0 : if (ret) {
4290 0 : struct name_cache_entry *nce;
4291 0 : struct waiting_dir_move *wdm;
4292 :
4293 0 : if (orphanized_dir) {
4294 0 : ret = refresh_ref_path(sctx, cur);
4295 0 : if (ret < 0)
4296 0 : goto out;
4297 : }
4298 :
4299 0 : ret = orphanize_inode(sctx, ow_inode, ow_gen,
4300 : cur->full_path);
4301 0 : if (ret < 0)
4302 0 : goto out;
4303 0 : if (S_ISDIR(ow_mode))
4304 0 : orphanized_dir = true;
4305 :
4306 : /*
4307 : * If ow_inode has its rename operation delayed
4308 : * make sure that its orphanized name is used in
4309 : * the source path when performing its rename
4310 : * operation.
4311 : */
4312 0 : wdm = get_waiting_dir_move(sctx, ow_inode);
4313 0 : if (wdm)
4314 0 : wdm->orphanized = true;
4315 :
4316 : /*
4317 : * Make sure we clear our orphanized inode's
4318 : * name from the name cache. This is because the
4319 : * inode ow_inode might be an ancestor of some
4320 : * other inode that will be orphanized as well
4321 : * later and has an inode number greater than
4322 : * sctx->send_progress. We need to prevent
4323 : * future name lookups from using the old name
4324 : * and get instead the orphan name.
4325 : */
4326 0 : nce = name_cache_search(sctx, ow_inode, ow_gen);
4327 0 : if (nce)
4328 0 : btrfs_lru_cache_remove(&sctx->name_cache,
4329 : &nce->entry);
4330 :
4331 : /*
4332 : * ow_inode might currently be an ancestor of
4333 : * cur_ino, therefore compute valid_path (the
4334 : * current path of cur_ino) again because it
4335 : * might contain the pre-orphanization name of
4336 : * ow_inode, which is no longer valid.
4337 : */
4338 0 : ret = is_ancestor(sctx->parent_root,
4339 : ow_inode, ow_gen,
4340 : sctx->cur_ino, NULL);
4341 0 : if (ret > 0) {
4342 0 : orphanized_ancestor = true;
4343 0 : fs_path_reset(valid_path);
4344 0 : ret = get_cur_path(sctx, sctx->cur_ino,
4345 : sctx->cur_inode_gen,
4346 : valid_path);
4347 : }
4348 0 : if (ret < 0)
4349 0 : goto out;
4350 : } else {
4351 : /*
4352 : * If we previously orphanized a directory that
4353 : * collided with a new reference that we already
4354 : * processed, recompute the current path because
4355 : * that directory may be part of the path.
4356 : */
4357 0 : if (orphanized_dir) {
4358 0 : ret = refresh_ref_path(sctx, cur);
4359 0 : if (ret < 0)
4360 0 : goto out;
4361 : }
4362 0 : ret = send_unlink(sctx, cur->full_path);
4363 0 : if (ret < 0)
4364 0 : goto out;
4365 : }
4366 : }
4367 :
4368 : }
4369 :
4370 0 : list_for_each_entry(cur, &sctx->new_refs, list) {
4371 : /*
4372 : * We may have refs where the parent directory does not exist
4373 : * yet. This happens if the parent directories inum is higher
4374 : * than the current inum. To handle this case, we create the
4375 : * parent directory out of order. But we need to check if this
4376 : * did already happen before due to other refs in the same dir.
4377 : */
4378 0 : ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL);
4379 0 : if (ret < 0)
4380 0 : goto out;
4381 0 : if (ret == inode_state_will_create) {
4382 0 : ret = 0;
4383 : /*
4384 : * First check if any of the current inodes refs did
4385 : * already create the dir.
4386 : */
4387 0 : list_for_each_entry(cur2, &sctx->new_refs, list) {
4388 0 : if (cur == cur2)
4389 : break;
4390 0 : if (cur2->dir == cur->dir) {
4391 : ret = 1;
4392 : break;
4393 : }
4394 : }
4395 :
4396 : /*
4397 : * If that did not happen, check if a previous inode
4398 : * did already create the dir.
4399 : */
4400 0 : if (!ret)
4401 0 : ret = did_create_dir(sctx, cur->dir);
4402 0 : if (ret < 0)
4403 0 : goto out;
4404 0 : if (!ret) {
4405 0 : ret = send_create_inode(sctx, cur->dir);
4406 0 : if (ret < 0)
4407 0 : goto out;
4408 0 : cache_dir_created(sctx, cur->dir);
4409 : }
4410 : }
4411 :
4412 0 : if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
4413 0 : ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
4414 0 : if (ret < 0)
4415 0 : goto out;
4416 0 : if (ret == 1) {
4417 0 : can_rename = false;
4418 0 : *pending_move = 1;
4419 : }
4420 : }
4421 :
4422 0 : if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
4423 : can_rename) {
4424 0 : ret = wait_for_parent_move(sctx, cur, is_orphan);
4425 0 : if (ret < 0)
4426 0 : goto out;
4427 0 : if (ret == 1) {
4428 0 : can_rename = false;
4429 0 : *pending_move = 1;
4430 : }
4431 : }
4432 :
4433 : /*
4434 : * link/move the ref to the new place. If we have an orphan
4435 : * inode, move it and update valid_path. If not, link or move
4436 : * it depending on the inode mode.
4437 : */
4438 0 : if (is_orphan && can_rename) {
4439 0 : ret = send_rename(sctx, valid_path, cur->full_path);
4440 0 : if (ret < 0)
4441 0 : goto out;
4442 0 : is_orphan = 0;
4443 0 : ret = fs_path_copy(valid_path, cur->full_path);
4444 0 : if (ret < 0)
4445 0 : goto out;
4446 0 : } else if (can_rename) {
4447 0 : if (S_ISDIR(sctx->cur_inode_mode)) {
4448 : /*
4449 : * Dirs can't be linked, so move it. For moved
4450 : * dirs, we always have one new and one deleted
4451 : * ref. The deleted ref is ignored later.
4452 : */
4453 0 : ret = send_rename(sctx, valid_path,
4454 : cur->full_path);
4455 0 : if (!ret)
4456 0 : ret = fs_path_copy(valid_path,
4457 : cur->full_path);
4458 0 : if (ret < 0)
4459 0 : goto out;
4460 : } else {
4461 : /*
4462 : * We might have previously orphanized an inode
4463 : * which is an ancestor of our current inode,
4464 : * so our reference's full path, which was
4465 : * computed before any such orphanizations, must
4466 : * be updated.
4467 : */
4468 0 : if (orphanized_dir) {
4469 0 : ret = update_ref_path(sctx, cur);
4470 0 : if (ret < 0)
4471 0 : goto out;
4472 : }
4473 0 : ret = send_link(sctx, cur->full_path,
4474 : valid_path);
4475 0 : if (ret < 0)
4476 0 : goto out;
4477 : }
4478 : }
4479 0 : ret = dup_ref(cur, &check_dirs);
4480 0 : if (ret < 0)
4481 0 : goto out;
4482 : }
4483 :
4484 0 : if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4485 : /*
4486 : * Check if we can already rmdir the directory. If not,
4487 : * orphanize it. For every dir item inside that gets deleted
4488 : * later, we do this check again and rmdir it then if possible.
4489 : * See the use of check_dirs for more details.
4490 : */
4491 0 : ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4492 0 : if (ret < 0)
4493 0 : goto out;
4494 0 : if (ret) {
4495 0 : ret = send_rmdir(sctx, valid_path);
4496 0 : if (ret < 0)
4497 0 : goto out;
4498 0 : } else if (!is_orphan) {
4499 0 : ret = orphanize_inode(sctx, sctx->cur_ino,
4500 : sctx->cur_inode_gen, valid_path);
4501 0 : if (ret < 0)
4502 0 : goto out;
4503 : is_orphan = 1;
4504 : }
4505 :
4506 0 : list_for_each_entry(cur, &sctx->deleted_refs, list) {
4507 0 : ret = dup_ref(cur, &check_dirs);
4508 0 : if (ret < 0)
4509 0 : goto out;
4510 : }
4511 0 : } else if (S_ISDIR(sctx->cur_inode_mode) &&
4512 0 : !list_empty(&sctx->deleted_refs)) {
4513 : /*
4514 : * We have a moved dir. Add the old parent to check_dirs
4515 : */
4516 0 : cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4517 : list);
4518 0 : ret = dup_ref(cur, &check_dirs);
4519 0 : if (ret < 0)
4520 0 : goto out;
4521 0 : } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4522 : /*
4523 : * We have a non dir inode. Go through all deleted refs and
4524 : * unlink them if they were not already overwritten by other
4525 : * inodes.
4526 : */
4527 0 : list_for_each_entry(cur, &sctx->deleted_refs, list) {
4528 0 : ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4529 : sctx->cur_ino, sctx->cur_inode_gen,
4530 0 : cur->name, cur->name_len);
4531 0 : if (ret < 0)
4532 0 : goto out;
4533 0 : if (!ret) {
4534 : /*
4535 : * If we orphanized any ancestor before, we need
4536 : * to recompute the full path for deleted names,
4537 : * since any such path was computed before we
4538 : * processed any references and orphanized any
4539 : * ancestor inode.
4540 : */
4541 0 : if (orphanized_ancestor) {
4542 0 : ret = update_ref_path(sctx, cur);
4543 0 : if (ret < 0)
4544 0 : goto out;
4545 : }
4546 0 : ret = send_unlink(sctx, cur->full_path);
4547 0 : if (ret < 0)
4548 0 : goto out;
4549 : }
4550 0 : ret = dup_ref(cur, &check_dirs);
4551 0 : if (ret < 0)
4552 0 : goto out;
4553 : }
4554 : /*
4555 : * If the inode is still orphan, unlink the orphan. This may
4556 : * happen when a previous inode did overwrite the first ref
4557 : * of this inode and no new refs were added for the current
4558 : * inode. Unlinking does not mean that the inode is deleted in
4559 : * all cases. There may still be links to this inode in other
4560 : * places.
4561 : */
4562 0 : if (is_orphan) {
4563 0 : ret = send_unlink(sctx, valid_path);
4564 0 : if (ret < 0)
4565 0 : goto out;
4566 : }
4567 : }
4568 :
4569 : /*
4570 : * We did collect all parent dirs where cur_inode was once located. We
4571 : * now go through all these dirs and check if they are pending for
4572 : * deletion and if it's finally possible to perform the rmdir now.
4573 : * We also update the inode stats of the parent dirs here.
4574 : */
4575 0 : list_for_each_entry(cur, &check_dirs, list) {
4576 : /*
4577 : * In case we had refs into dirs that were not processed yet,
4578 : * we don't need to do the utime and rmdir logic for these dirs.
4579 : * The dir will be processed later.
4580 : */
4581 0 : if (cur->dir > sctx->cur_ino)
4582 0 : continue;
4583 :
4584 0 : ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL);
4585 0 : if (ret < 0)
4586 0 : goto out;
4587 :
4588 0 : if (ret == inode_state_did_create ||
4589 0 : ret == inode_state_no_change) {
4590 0 : ret = cache_dir_utimes(sctx, cur->dir, cur->dir_gen);
4591 0 : if (ret < 0)
4592 0 : goto out;
4593 0 : } else if (ret == inode_state_did_delete &&
4594 0 : cur->dir != last_dir_ino_rm) {
4595 0 : ret = can_rmdir(sctx, cur->dir, cur->dir_gen);
4596 0 : if (ret < 0)
4597 0 : goto out;
4598 0 : if (ret) {
4599 0 : ret = get_cur_path(sctx, cur->dir,
4600 : cur->dir_gen, valid_path);
4601 0 : if (ret < 0)
4602 0 : goto out;
4603 0 : ret = send_rmdir(sctx, valid_path);
4604 0 : if (ret < 0)
4605 0 : goto out;
4606 0 : last_dir_ino_rm = cur->dir;
4607 : }
4608 : }
4609 : }
4610 :
4611 : ret = 0;
4612 :
4613 0 : out:
4614 0 : __free_recorded_refs(&check_dirs);
4615 0 : free_recorded_refs(sctx);
4616 0 : fs_path_free(valid_path);
4617 0 : return ret;
4618 : }
4619 :
4620 0 : static int rbtree_ref_comp(const void *k, const struct rb_node *node)
4621 : {
4622 0 : const struct recorded_ref *data = k;
4623 0 : const struct recorded_ref *ref = rb_entry(node, struct recorded_ref, node);
4624 0 : int result;
4625 :
4626 0 : if (data->dir > ref->dir)
4627 : return 1;
4628 0 : if (data->dir < ref->dir)
4629 : return -1;
4630 0 : if (data->dir_gen > ref->dir_gen)
4631 : return 1;
4632 0 : if (data->dir_gen < ref->dir_gen)
4633 : return -1;
4634 0 : if (data->name_len > ref->name_len)
4635 : return 1;
4636 0 : if (data->name_len < ref->name_len)
4637 : return -1;
4638 0 : result = strcmp(data->name, ref->name);
4639 0 : if (result > 0)
4640 : return 1;
4641 0 : if (result < 0)
4642 0 : return -1;
4643 : return 0;
4644 : }
4645 :
4646 : static bool rbtree_ref_less(struct rb_node *node, const struct rb_node *parent)
4647 : {
4648 0 : const struct recorded_ref *entry = rb_entry(node, struct recorded_ref, node);
4649 :
4650 0 : return rbtree_ref_comp(entry, parent) < 0;
4651 : }
4652 :
4653 0 : static int record_ref_in_tree(struct rb_root *root, struct list_head *refs,
4654 : struct fs_path *name, u64 dir, u64 dir_gen,
4655 : struct send_ctx *sctx)
4656 : {
4657 0 : int ret = 0;
4658 0 : struct fs_path *path = NULL;
4659 0 : struct recorded_ref *ref = NULL;
4660 :
4661 0 : path = fs_path_alloc();
4662 0 : if (!path) {
4663 0 : ret = -ENOMEM;
4664 0 : goto out;
4665 : }
4666 :
4667 0 : ref = recorded_ref_alloc();
4668 0 : if (!ref) {
4669 0 : ret = -ENOMEM;
4670 0 : goto out;
4671 : }
4672 :
4673 0 : ret = get_cur_path(sctx, dir, dir_gen, path);
4674 0 : if (ret < 0)
4675 0 : goto out;
4676 0 : ret = fs_path_add_path(path, name);
4677 0 : if (ret < 0)
4678 0 : goto out;
4679 :
4680 0 : ref->dir = dir;
4681 0 : ref->dir_gen = dir_gen;
4682 0 : set_ref_path(ref, path);
4683 0 : list_add_tail(&ref->list, refs);
4684 0 : rb_add(&ref->node, root, rbtree_ref_less);
4685 0 : ref->root = root;
4686 0 : out:
4687 0 : if (ret) {
4688 0 : if (path && (!ref || !ref->full_path))
4689 0 : fs_path_free(path);
4690 0 : recorded_ref_free(ref);
4691 : }
4692 0 : return ret;
4693 : }
4694 :
4695 0 : static int record_new_ref_if_needed(int num, u64 dir, int index,
4696 : struct fs_path *name, void *ctx)
4697 : {
4698 0 : int ret = 0;
4699 0 : struct send_ctx *sctx = ctx;
4700 0 : struct rb_node *node = NULL;
4701 0 : struct recorded_ref data;
4702 0 : struct recorded_ref *ref;
4703 0 : u64 dir_gen;
4704 :
4705 0 : ret = get_inode_gen(sctx->send_root, dir, &dir_gen);
4706 0 : if (ret < 0)
4707 0 : goto out;
4708 :
4709 0 : data.dir = dir;
4710 0 : data.dir_gen = dir_gen;
4711 0 : set_ref_path(&data, name);
4712 0 : node = rb_find(&data, &sctx->rbtree_deleted_refs, rbtree_ref_comp);
4713 0 : if (node) {
4714 0 : ref = rb_entry(node, struct recorded_ref, node);
4715 0 : recorded_ref_free(ref);
4716 : } else {
4717 0 : ret = record_ref_in_tree(&sctx->rbtree_new_refs,
4718 : &sctx->new_refs, name, dir, dir_gen,
4719 : sctx);
4720 : }
4721 0 : out:
4722 0 : return ret;
4723 : }
4724 :
4725 0 : static int record_deleted_ref_if_needed(int num, u64 dir, int index,
4726 : struct fs_path *name, void *ctx)
4727 : {
4728 0 : int ret = 0;
4729 0 : struct send_ctx *sctx = ctx;
4730 0 : struct rb_node *node = NULL;
4731 0 : struct recorded_ref data;
4732 0 : struct recorded_ref *ref;
4733 0 : u64 dir_gen;
4734 :
4735 0 : ret = get_inode_gen(sctx->parent_root, dir, &dir_gen);
4736 0 : if (ret < 0)
4737 0 : goto out;
4738 :
4739 0 : data.dir = dir;
4740 0 : data.dir_gen = dir_gen;
4741 0 : set_ref_path(&data, name);
4742 0 : node = rb_find(&data, &sctx->rbtree_new_refs, rbtree_ref_comp);
4743 0 : if (node) {
4744 0 : ref = rb_entry(node, struct recorded_ref, node);
4745 0 : recorded_ref_free(ref);
4746 : } else {
4747 0 : ret = record_ref_in_tree(&sctx->rbtree_deleted_refs,
4748 : &sctx->deleted_refs, name, dir,
4749 : dir_gen, sctx);
4750 : }
4751 0 : out:
4752 0 : return ret;
4753 : }
4754 :
4755 0 : static int record_new_ref(struct send_ctx *sctx)
4756 : {
4757 0 : int ret;
4758 :
4759 0 : ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4760 : sctx->cmp_key, 0, record_new_ref_if_needed, sctx);
4761 0 : if (ret < 0)
4762 : goto out;
4763 : ret = 0;
4764 :
4765 : out:
4766 0 : return ret;
4767 : }
4768 :
4769 0 : static int record_deleted_ref(struct send_ctx *sctx)
4770 : {
4771 0 : int ret;
4772 :
4773 0 : ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4774 : sctx->cmp_key, 0, record_deleted_ref_if_needed,
4775 : sctx);
4776 0 : if (ret < 0)
4777 : goto out;
4778 : ret = 0;
4779 :
4780 : out:
4781 0 : return ret;
4782 : }
4783 :
4784 0 : static int record_changed_ref(struct send_ctx *sctx)
4785 : {
4786 0 : int ret = 0;
4787 :
4788 0 : ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4789 : sctx->cmp_key, 0, record_new_ref_if_needed, sctx);
4790 0 : if (ret < 0)
4791 0 : goto out;
4792 0 : ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4793 : sctx->cmp_key, 0, record_deleted_ref_if_needed, sctx);
4794 0 : if (ret < 0)
4795 : goto out;
4796 : ret = 0;
4797 :
4798 0 : out:
4799 0 : return ret;
4800 : }
4801 :
4802 : /*
4803 : * Record and process all refs at once. Needed when an inode changes the
4804 : * generation number, which means that it was deleted and recreated.
4805 : */
4806 0 : static int process_all_refs(struct send_ctx *sctx,
4807 : enum btrfs_compare_tree_result cmd)
4808 : {
4809 0 : int ret = 0;
4810 0 : int iter_ret = 0;
4811 0 : struct btrfs_root *root;
4812 0 : struct btrfs_path *path;
4813 0 : struct btrfs_key key;
4814 0 : struct btrfs_key found_key;
4815 0 : iterate_inode_ref_t cb;
4816 0 : int pending_move = 0;
4817 :
4818 0 : path = alloc_path_for_send();
4819 0 : if (!path)
4820 : return -ENOMEM;
4821 :
4822 0 : if (cmd == BTRFS_COMPARE_TREE_NEW) {
4823 0 : root = sctx->send_root;
4824 0 : cb = record_new_ref_if_needed;
4825 0 : } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4826 0 : root = sctx->parent_root;
4827 0 : cb = record_deleted_ref_if_needed;
4828 : } else {
4829 0 : btrfs_err(sctx->send_root->fs_info,
4830 : "Wrong command %d in process_all_refs", cmd);
4831 0 : ret = -EINVAL;
4832 0 : goto out;
4833 : }
4834 :
4835 0 : key.objectid = sctx->cmp_key->objectid;
4836 0 : key.type = BTRFS_INODE_REF_KEY;
4837 0 : key.offset = 0;
4838 0 : btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
4839 0 : if (found_key.objectid != key.objectid ||
4840 0 : (found_key.type != BTRFS_INODE_REF_KEY &&
4841 : found_key.type != BTRFS_INODE_EXTREF_KEY))
4842 : break;
4843 :
4844 0 : ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4845 0 : if (ret < 0)
4846 0 : goto out;
4847 : }
4848 : /* Catch error found during iteration */
4849 0 : if (iter_ret < 0) {
4850 0 : ret = iter_ret;
4851 0 : goto out;
4852 : }
4853 0 : btrfs_release_path(path);
4854 :
4855 : /*
4856 : * We don't actually care about pending_move as we are simply
4857 : * re-creating this inode and will be rename'ing it into place once we
4858 : * rename the parent directory.
4859 : */
4860 0 : ret = process_recorded_refs(sctx, &pending_move);
4861 0 : out:
4862 0 : btrfs_free_path(path);
4863 0 : return ret;
4864 : }
4865 :
4866 0 : static int send_set_xattr(struct send_ctx *sctx,
4867 : struct fs_path *path,
4868 : const char *name, int name_len,
4869 : const char *data, int data_len)
4870 : {
4871 0 : int ret = 0;
4872 :
4873 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4874 0 : if (ret < 0)
4875 0 : goto out;
4876 :
4877 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4878 0 : TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4879 0 : TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4880 :
4881 0 : ret = send_cmd(sctx);
4882 :
4883 0 : tlv_put_failure:
4884 0 : out:
4885 0 : return ret;
4886 : }
4887 :
4888 0 : static int send_remove_xattr(struct send_ctx *sctx,
4889 : struct fs_path *path,
4890 : const char *name, int name_len)
4891 : {
4892 0 : int ret = 0;
4893 :
4894 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4895 0 : if (ret < 0)
4896 0 : goto out;
4897 :
4898 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4899 0 : TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4900 :
4901 0 : ret = send_cmd(sctx);
4902 :
4903 0 : tlv_put_failure:
4904 0 : out:
4905 0 : return ret;
4906 : }
4907 :
4908 0 : static int __process_new_xattr(int num, struct btrfs_key *di_key,
4909 : const char *name, int name_len, const char *data,
4910 : int data_len, void *ctx)
4911 : {
4912 0 : int ret;
4913 0 : struct send_ctx *sctx = ctx;
4914 0 : struct fs_path *p;
4915 0 : struct posix_acl_xattr_header dummy_acl;
4916 :
4917 : /* Capabilities are emitted by finish_inode_if_needed */
4918 0 : if (!strncmp(name, XATTR_NAME_CAPS, name_len))
4919 : return 0;
4920 :
4921 0 : p = fs_path_alloc();
4922 0 : if (!p)
4923 : return -ENOMEM;
4924 :
4925 : /*
4926 : * This hack is needed because empty acls are stored as zero byte
4927 : * data in xattrs. Problem with that is, that receiving these zero byte
4928 : * acls will fail later. To fix this, we send a dummy acl list that
4929 : * only contains the version number and no entries.
4930 : */
4931 0 : if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4932 0 : !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4933 0 : if (data_len == 0) {
4934 0 : dummy_acl.a_version =
4935 : cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4936 0 : data = (char *)&dummy_acl;
4937 0 : data_len = sizeof(dummy_acl);
4938 : }
4939 : }
4940 :
4941 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4942 0 : if (ret < 0)
4943 0 : goto out;
4944 :
4945 0 : ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4946 :
4947 0 : out:
4948 0 : fs_path_free(p);
4949 0 : return ret;
4950 : }
4951 :
4952 0 : static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4953 : const char *name, int name_len,
4954 : const char *data, int data_len, void *ctx)
4955 : {
4956 0 : int ret;
4957 0 : struct send_ctx *sctx = ctx;
4958 0 : struct fs_path *p;
4959 :
4960 0 : p = fs_path_alloc();
4961 0 : if (!p)
4962 : return -ENOMEM;
4963 :
4964 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4965 0 : if (ret < 0)
4966 0 : goto out;
4967 :
4968 0 : ret = send_remove_xattr(sctx, p, name, name_len);
4969 :
4970 0 : out:
4971 0 : fs_path_free(p);
4972 0 : return ret;
4973 : }
4974 :
4975 0 : static int process_new_xattr(struct send_ctx *sctx)
4976 : {
4977 0 : int ret = 0;
4978 :
4979 0 : ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4980 : __process_new_xattr, sctx);
4981 :
4982 0 : return ret;
4983 : }
4984 :
4985 0 : static int process_deleted_xattr(struct send_ctx *sctx)
4986 : {
4987 0 : return iterate_dir_item(sctx->parent_root, sctx->right_path,
4988 : __process_deleted_xattr, sctx);
4989 : }
4990 :
4991 : struct find_xattr_ctx {
4992 : const char *name;
4993 : int name_len;
4994 : int found_idx;
4995 : char *found_data;
4996 : int found_data_len;
4997 : };
4998 :
4999 0 : static int __find_xattr(int num, struct btrfs_key *di_key, const char *name,
5000 : int name_len, const char *data, int data_len, void *vctx)
5001 : {
5002 0 : struct find_xattr_ctx *ctx = vctx;
5003 :
5004 0 : if (name_len == ctx->name_len &&
5005 0 : strncmp(name, ctx->name, name_len) == 0) {
5006 0 : ctx->found_idx = num;
5007 0 : ctx->found_data_len = data_len;
5008 0 : ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
5009 0 : if (!ctx->found_data)
5010 : return -ENOMEM;
5011 0 : return 1;
5012 : }
5013 : return 0;
5014 : }
5015 :
5016 0 : static int find_xattr(struct btrfs_root *root,
5017 : struct btrfs_path *path,
5018 : struct btrfs_key *key,
5019 : const char *name, int name_len,
5020 : char **data, int *data_len)
5021 : {
5022 0 : int ret;
5023 0 : struct find_xattr_ctx ctx;
5024 :
5025 0 : ctx.name = name;
5026 0 : ctx.name_len = name_len;
5027 0 : ctx.found_idx = -1;
5028 0 : ctx.found_data = NULL;
5029 0 : ctx.found_data_len = 0;
5030 :
5031 0 : ret = iterate_dir_item(root, path, __find_xattr, &ctx);
5032 0 : if (ret < 0)
5033 : return ret;
5034 :
5035 0 : if (ctx.found_idx == -1)
5036 : return -ENOENT;
5037 0 : if (data) {
5038 0 : *data = ctx.found_data;
5039 0 : *data_len = ctx.found_data_len;
5040 : } else {
5041 0 : kfree(ctx.found_data);
5042 : }
5043 0 : return ctx.found_idx;
5044 : }
5045 :
5046 :
5047 0 : static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
5048 : const char *name, int name_len,
5049 : const char *data, int data_len,
5050 : void *ctx)
5051 : {
5052 0 : int ret;
5053 0 : struct send_ctx *sctx = ctx;
5054 0 : char *found_data = NULL;
5055 0 : int found_data_len = 0;
5056 :
5057 0 : ret = find_xattr(sctx->parent_root, sctx->right_path,
5058 : sctx->cmp_key, name, name_len, &found_data,
5059 : &found_data_len);
5060 0 : if (ret == -ENOENT) {
5061 0 : ret = __process_new_xattr(num, di_key, name, name_len, data,
5062 : data_len, ctx);
5063 0 : } else if (ret >= 0) {
5064 0 : if (data_len != found_data_len ||
5065 0 : memcmp(data, found_data, data_len)) {
5066 0 : ret = __process_new_xattr(num, di_key, name, name_len,
5067 : data, data_len, ctx);
5068 : } else {
5069 : ret = 0;
5070 : }
5071 : }
5072 :
5073 0 : kfree(found_data);
5074 0 : return ret;
5075 : }
5076 :
5077 0 : static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
5078 : const char *name, int name_len,
5079 : const char *data, int data_len,
5080 : void *ctx)
5081 : {
5082 0 : int ret;
5083 0 : struct send_ctx *sctx = ctx;
5084 :
5085 0 : ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
5086 : name, name_len, NULL, NULL);
5087 0 : if (ret == -ENOENT)
5088 0 : ret = __process_deleted_xattr(num, di_key, name, name_len, data,
5089 : data_len, ctx);
5090 0 : else if (ret >= 0)
5091 : ret = 0;
5092 :
5093 0 : return ret;
5094 : }
5095 :
5096 0 : static int process_changed_xattr(struct send_ctx *sctx)
5097 : {
5098 0 : int ret = 0;
5099 :
5100 0 : ret = iterate_dir_item(sctx->send_root, sctx->left_path,
5101 : __process_changed_new_xattr, sctx);
5102 0 : if (ret < 0)
5103 0 : goto out;
5104 0 : ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
5105 : __process_changed_deleted_xattr, sctx);
5106 :
5107 0 : out:
5108 0 : return ret;
5109 : }
5110 :
5111 0 : static int process_all_new_xattrs(struct send_ctx *sctx)
5112 : {
5113 0 : int ret = 0;
5114 0 : int iter_ret = 0;
5115 0 : struct btrfs_root *root;
5116 0 : struct btrfs_path *path;
5117 0 : struct btrfs_key key;
5118 0 : struct btrfs_key found_key;
5119 :
5120 0 : path = alloc_path_for_send();
5121 0 : if (!path)
5122 : return -ENOMEM;
5123 :
5124 0 : root = sctx->send_root;
5125 :
5126 0 : key.objectid = sctx->cmp_key->objectid;
5127 0 : key.type = BTRFS_XATTR_ITEM_KEY;
5128 0 : key.offset = 0;
5129 0 : btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
5130 0 : if (found_key.objectid != key.objectid ||
5131 0 : found_key.type != key.type) {
5132 : ret = 0;
5133 : break;
5134 : }
5135 :
5136 0 : ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
5137 0 : if (ret < 0)
5138 : break;
5139 : }
5140 : /* Catch error found during iteration */
5141 0 : if (iter_ret < 0)
5142 0 : ret = iter_ret;
5143 :
5144 0 : btrfs_free_path(path);
5145 0 : return ret;
5146 : }
5147 :
5148 : static int send_verity(struct send_ctx *sctx, struct fs_path *path,
5149 : struct fsverity_descriptor *desc)
5150 : {
5151 : int ret;
5152 :
5153 : ret = begin_cmd(sctx, BTRFS_SEND_C_ENABLE_VERITY);
5154 : if (ret < 0)
5155 : goto out;
5156 :
5157 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
5158 : TLV_PUT_U8(sctx, BTRFS_SEND_A_VERITY_ALGORITHM,
5159 : le8_to_cpu(desc->hash_algorithm));
5160 : TLV_PUT_U32(sctx, BTRFS_SEND_A_VERITY_BLOCK_SIZE,
5161 : 1U << le8_to_cpu(desc->log_blocksize));
5162 : TLV_PUT(sctx, BTRFS_SEND_A_VERITY_SALT_DATA, desc->salt,
5163 : le8_to_cpu(desc->salt_size));
5164 : TLV_PUT(sctx, BTRFS_SEND_A_VERITY_SIG_DATA, desc->signature,
5165 : le32_to_cpu(desc->sig_size));
5166 :
5167 : ret = send_cmd(sctx);
5168 :
5169 : tlv_put_failure:
5170 : out:
5171 : return ret;
5172 : }
5173 :
5174 0 : static int process_verity(struct send_ctx *sctx)
5175 : {
5176 0 : int ret = 0;
5177 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
5178 0 : struct inode *inode;
5179 0 : struct fs_path *p;
5180 :
5181 0 : inode = btrfs_iget(fs_info->sb, sctx->cur_ino, sctx->send_root);
5182 0 : if (IS_ERR(inode))
5183 0 : return PTR_ERR(inode);
5184 :
5185 0 : ret = btrfs_get_verity_descriptor(inode, NULL, 0);
5186 0 : if (ret < 0)
5187 0 : goto iput;
5188 :
5189 : if (ret > FS_VERITY_MAX_DESCRIPTOR_SIZE) {
5190 : ret = -EMSGSIZE;
5191 : goto iput;
5192 : }
5193 : if (!sctx->verity_descriptor) {
5194 : sctx->verity_descriptor = kvmalloc(FS_VERITY_MAX_DESCRIPTOR_SIZE,
5195 : GFP_KERNEL);
5196 : if (!sctx->verity_descriptor) {
5197 : ret = -ENOMEM;
5198 : goto iput;
5199 : }
5200 : }
5201 :
5202 : ret = btrfs_get_verity_descriptor(inode, sctx->verity_descriptor, ret);
5203 : if (ret < 0)
5204 : goto iput;
5205 :
5206 : p = fs_path_alloc();
5207 : if (!p) {
5208 : ret = -ENOMEM;
5209 : goto iput;
5210 : }
5211 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5212 : if (ret < 0)
5213 : goto free_path;
5214 :
5215 : ret = send_verity(sctx, p, sctx->verity_descriptor);
5216 : if (ret < 0)
5217 : goto free_path;
5218 :
5219 : free_path:
5220 : fs_path_free(p);
5221 : iput:
5222 0 : iput(inode);
5223 0 : return ret;
5224 : }
5225 :
5226 : static inline u64 max_send_read_size(const struct send_ctx *sctx)
5227 : {
5228 0 : return sctx->send_max_size - SZ_16K;
5229 : }
5230 :
5231 0 : static int put_data_header(struct send_ctx *sctx, u32 len)
5232 : {
5233 0 : if (WARN_ON_ONCE(sctx->put_data))
5234 : return -EINVAL;
5235 0 : sctx->put_data = true;
5236 0 : if (sctx->proto >= 2) {
5237 : /*
5238 : * Since v2, the data attribute header doesn't include a length,
5239 : * it is implicitly to the end of the command.
5240 : */
5241 0 : if (sctx->send_max_size - sctx->send_size < sizeof(__le16) + len)
5242 : return -EOVERFLOW;
5243 0 : put_unaligned_le16(BTRFS_SEND_A_DATA, sctx->send_buf + sctx->send_size);
5244 0 : sctx->send_size += sizeof(__le16);
5245 : } else {
5246 0 : struct btrfs_tlv_header *hdr;
5247 :
5248 0 : if (sctx->send_max_size - sctx->send_size < sizeof(*hdr) + len)
5249 : return -EOVERFLOW;
5250 0 : hdr = (struct btrfs_tlv_header *)(sctx->send_buf + sctx->send_size);
5251 0 : put_unaligned_le16(BTRFS_SEND_A_DATA, &hdr->tlv_type);
5252 0 : put_unaligned_le16(len, &hdr->tlv_len);
5253 0 : sctx->send_size += sizeof(*hdr);
5254 : }
5255 : return 0;
5256 : }
5257 :
5258 0 : static int put_file_data(struct send_ctx *sctx, u64 offset, u32 len)
5259 : {
5260 0 : struct btrfs_root *root = sctx->send_root;
5261 0 : struct btrfs_fs_info *fs_info = root->fs_info;
5262 0 : struct page *page;
5263 0 : pgoff_t index = offset >> PAGE_SHIFT;
5264 0 : pgoff_t last_index;
5265 0 : unsigned pg_offset = offset_in_page(offset);
5266 0 : int ret;
5267 :
5268 0 : ret = put_data_header(sctx, len);
5269 0 : if (ret)
5270 : return ret;
5271 :
5272 0 : last_index = (offset + len - 1) >> PAGE_SHIFT;
5273 :
5274 0 : while (index <= last_index) {
5275 0 : unsigned cur_len = min_t(unsigned, len,
5276 : PAGE_SIZE - pg_offset);
5277 :
5278 0 : page = find_lock_page(sctx->cur_inode->i_mapping, index);
5279 0 : if (!page) {
5280 0 : page_cache_sync_readahead(sctx->cur_inode->i_mapping,
5281 : &sctx->ra, NULL, index,
5282 0 : last_index + 1 - index);
5283 :
5284 0 : page = find_or_create_page(sctx->cur_inode->i_mapping,
5285 : index, GFP_KERNEL);
5286 0 : if (!page) {
5287 : ret = -ENOMEM;
5288 : break;
5289 : }
5290 : }
5291 :
5292 0 : if (PageReadahead(page))
5293 0 : page_cache_async_readahead(sctx->cur_inode->i_mapping,
5294 : &sctx->ra, NULL, page_folio(page),
5295 0 : index, last_index + 1 - index);
5296 :
5297 0 : if (!PageUptodate(page)) {
5298 0 : btrfs_read_folio(NULL, page_folio(page));
5299 0 : lock_page(page);
5300 0 : if (!PageUptodate(page)) {
5301 0 : unlock_page(page);
5302 0 : btrfs_err(fs_info,
5303 : "send: IO error at offset %llu for inode %llu root %llu",
5304 : page_offset(page), sctx->cur_ino,
5305 : sctx->send_root->root_key.objectid);
5306 0 : put_page(page);
5307 0 : ret = -EIO;
5308 0 : break;
5309 : }
5310 : }
5311 :
5312 0 : memcpy_from_page(sctx->send_buf + sctx->send_size, page,
5313 : pg_offset, cur_len);
5314 0 : unlock_page(page);
5315 0 : put_page(page);
5316 0 : index++;
5317 0 : pg_offset = 0;
5318 0 : len -= cur_len;
5319 0 : sctx->send_size += cur_len;
5320 : }
5321 :
5322 : return ret;
5323 : }
5324 :
5325 : /*
5326 : * Read some bytes from the current inode/file and send a write command to
5327 : * user space.
5328 : */
5329 0 : static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
5330 : {
5331 0 : struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
5332 0 : int ret = 0;
5333 0 : struct fs_path *p;
5334 :
5335 0 : p = fs_path_alloc();
5336 0 : if (!p)
5337 : return -ENOMEM;
5338 :
5339 0 : btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
5340 :
5341 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5342 0 : if (ret < 0)
5343 0 : goto out;
5344 :
5345 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5346 0 : if (ret < 0)
5347 0 : goto out;
5348 :
5349 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5350 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5351 0 : ret = put_file_data(sctx, offset, len);
5352 0 : if (ret < 0)
5353 0 : goto out;
5354 :
5355 0 : ret = send_cmd(sctx);
5356 :
5357 0 : tlv_put_failure:
5358 0 : out:
5359 0 : fs_path_free(p);
5360 0 : return ret;
5361 : }
5362 :
5363 : /*
5364 : * Send a clone command to user space.
5365 : */
5366 0 : static int send_clone(struct send_ctx *sctx,
5367 : u64 offset, u32 len,
5368 : struct clone_root *clone_root)
5369 : {
5370 0 : int ret = 0;
5371 0 : struct fs_path *p;
5372 0 : u64 gen;
5373 :
5374 0 : btrfs_debug(sctx->send_root->fs_info,
5375 : "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
5376 : offset, len, clone_root->root->root_key.objectid,
5377 : clone_root->ino, clone_root->offset);
5378 :
5379 0 : p = fs_path_alloc();
5380 0 : if (!p)
5381 : return -ENOMEM;
5382 :
5383 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
5384 0 : if (ret < 0)
5385 0 : goto out;
5386 :
5387 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5388 0 : if (ret < 0)
5389 0 : goto out;
5390 :
5391 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5392 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
5393 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5394 :
5395 0 : if (clone_root->root == sctx->send_root) {
5396 0 : ret = get_inode_gen(sctx->send_root, clone_root->ino, &gen);
5397 0 : if (ret < 0)
5398 0 : goto out;
5399 0 : ret = get_cur_path(sctx, clone_root->ino, gen, p);
5400 : } else {
5401 0 : ret = get_inode_path(clone_root->root, clone_root->ino, p);
5402 : }
5403 0 : if (ret < 0)
5404 0 : goto out;
5405 :
5406 : /*
5407 : * If the parent we're using has a received_uuid set then use that as
5408 : * our clone source as that is what we will look for when doing a
5409 : * receive.
5410 : *
5411 : * This covers the case that we create a snapshot off of a received
5412 : * subvolume and then use that as the parent and try to receive on a
5413 : * different host.
5414 : */
5415 0 : if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
5416 0 : TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
5417 : clone_root->root->root_item.received_uuid);
5418 : else
5419 0 : TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
5420 : clone_root->root->root_item.uuid);
5421 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
5422 : btrfs_root_ctransid(&clone_root->root->root_item));
5423 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
5424 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
5425 : clone_root->offset);
5426 :
5427 0 : ret = send_cmd(sctx);
5428 :
5429 0 : tlv_put_failure:
5430 0 : out:
5431 0 : fs_path_free(p);
5432 0 : return ret;
5433 : }
5434 :
5435 : /*
5436 : * Send an update extent command to user space.
5437 : */
5438 0 : static int send_update_extent(struct send_ctx *sctx,
5439 : u64 offset, u32 len)
5440 : {
5441 0 : int ret = 0;
5442 0 : struct fs_path *p;
5443 :
5444 0 : p = fs_path_alloc();
5445 0 : if (!p)
5446 : return -ENOMEM;
5447 :
5448 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
5449 0 : if (ret < 0)
5450 0 : goto out;
5451 :
5452 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5453 0 : if (ret < 0)
5454 0 : goto out;
5455 :
5456 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5457 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5458 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
5459 :
5460 0 : ret = send_cmd(sctx);
5461 :
5462 0 : tlv_put_failure:
5463 0 : out:
5464 0 : fs_path_free(p);
5465 0 : return ret;
5466 : }
5467 :
5468 0 : static int send_hole(struct send_ctx *sctx, u64 end)
5469 : {
5470 0 : struct fs_path *p = NULL;
5471 0 : u64 read_size = max_send_read_size(sctx);
5472 0 : u64 offset = sctx->cur_inode_last_extent;
5473 0 : int ret = 0;
5474 :
5475 : /*
5476 : * A hole that starts at EOF or beyond it. Since we do not yet support
5477 : * fallocate (for extent preallocation and hole punching), sending a
5478 : * write of zeroes starting at EOF or beyond would later require issuing
5479 : * a truncate operation which would undo the write and achieve nothing.
5480 : */
5481 0 : if (offset >= sctx->cur_inode_size)
5482 : return 0;
5483 :
5484 : /*
5485 : * Don't go beyond the inode's i_size due to prealloc extents that start
5486 : * after the i_size.
5487 : */
5488 0 : end = min_t(u64, end, sctx->cur_inode_size);
5489 :
5490 0 : if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5491 0 : return send_update_extent(sctx, offset, end - offset);
5492 :
5493 0 : p = fs_path_alloc();
5494 0 : if (!p)
5495 : return -ENOMEM;
5496 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5497 0 : if (ret < 0)
5498 0 : goto tlv_put_failure;
5499 0 : while (offset < end) {
5500 0 : u64 len = min(end - offset, read_size);
5501 :
5502 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5503 0 : if (ret < 0)
5504 : break;
5505 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5506 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5507 0 : ret = put_data_header(sctx, len);
5508 0 : if (ret < 0)
5509 : break;
5510 0 : memset(sctx->send_buf + sctx->send_size, 0, len);
5511 0 : sctx->send_size += len;
5512 0 : ret = send_cmd(sctx);
5513 0 : if (ret < 0)
5514 : break;
5515 0 : offset += len;
5516 : }
5517 0 : sctx->cur_inode_next_write_offset = offset;
5518 0 : tlv_put_failure:
5519 0 : fs_path_free(p);
5520 0 : return ret;
5521 : }
5522 :
5523 0 : static int send_encoded_inline_extent(struct send_ctx *sctx,
5524 : struct btrfs_path *path, u64 offset,
5525 : u64 len)
5526 : {
5527 0 : struct btrfs_root *root = sctx->send_root;
5528 0 : struct btrfs_fs_info *fs_info = root->fs_info;
5529 0 : struct inode *inode;
5530 0 : struct fs_path *fspath;
5531 0 : struct extent_buffer *leaf = path->nodes[0];
5532 0 : struct btrfs_key key;
5533 0 : struct btrfs_file_extent_item *ei;
5534 0 : u64 ram_bytes;
5535 0 : size_t inline_size;
5536 0 : int ret;
5537 :
5538 0 : inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root);
5539 0 : if (IS_ERR(inode))
5540 0 : return PTR_ERR(inode);
5541 :
5542 0 : fspath = fs_path_alloc();
5543 0 : if (!fspath) {
5544 0 : ret = -ENOMEM;
5545 0 : goto out;
5546 : }
5547 :
5548 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_ENCODED_WRITE);
5549 0 : if (ret < 0)
5550 0 : goto out;
5551 :
5552 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath);
5553 0 : if (ret < 0)
5554 0 : goto out;
5555 :
5556 0 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5557 0 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
5558 0 : ram_bytes = btrfs_file_extent_ram_bytes(leaf, ei);
5559 0 : inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
5560 :
5561 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, fspath);
5562 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5563 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_FILE_LEN,
5564 : min(key.offset + ram_bytes - offset, len));
5565 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_LEN, ram_bytes);
5566 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_OFFSET, offset - key.offset);
5567 0 : ret = btrfs_encoded_io_compression_from_extent(fs_info,
5568 : btrfs_file_extent_compression(leaf, ei));
5569 0 : if (ret < 0)
5570 0 : goto out;
5571 0 : TLV_PUT_U32(sctx, BTRFS_SEND_A_COMPRESSION, ret);
5572 :
5573 0 : ret = put_data_header(sctx, inline_size);
5574 0 : if (ret < 0)
5575 0 : goto out;
5576 0 : read_extent_buffer(leaf, sctx->send_buf + sctx->send_size,
5577 : btrfs_file_extent_inline_start(ei), inline_size);
5578 0 : sctx->send_size += inline_size;
5579 :
5580 0 : ret = send_cmd(sctx);
5581 :
5582 0 : tlv_put_failure:
5583 0 : out:
5584 0 : fs_path_free(fspath);
5585 0 : iput(inode);
5586 0 : return ret;
5587 : }
5588 :
5589 0 : static int send_encoded_extent(struct send_ctx *sctx, struct btrfs_path *path,
5590 : u64 offset, u64 len)
5591 : {
5592 0 : struct btrfs_root *root = sctx->send_root;
5593 0 : struct btrfs_fs_info *fs_info = root->fs_info;
5594 0 : struct inode *inode;
5595 0 : struct fs_path *fspath;
5596 0 : struct extent_buffer *leaf = path->nodes[0];
5597 0 : struct btrfs_key key;
5598 0 : struct btrfs_file_extent_item *ei;
5599 0 : u64 disk_bytenr, disk_num_bytes;
5600 0 : u32 data_offset;
5601 0 : struct btrfs_cmd_header *hdr;
5602 0 : u32 crc;
5603 0 : int ret;
5604 :
5605 0 : inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root);
5606 0 : if (IS_ERR(inode))
5607 0 : return PTR_ERR(inode);
5608 :
5609 0 : fspath = fs_path_alloc();
5610 0 : if (!fspath) {
5611 0 : ret = -ENOMEM;
5612 0 : goto out;
5613 : }
5614 :
5615 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_ENCODED_WRITE);
5616 0 : if (ret < 0)
5617 0 : goto out;
5618 :
5619 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath);
5620 0 : if (ret < 0)
5621 0 : goto out;
5622 :
5623 0 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5624 0 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
5625 0 : disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
5626 0 : disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, ei);
5627 :
5628 0 : TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, fspath);
5629 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5630 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_FILE_LEN,
5631 : min(key.offset + btrfs_file_extent_num_bytes(leaf, ei) - offset,
5632 : len));
5633 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_LEN,
5634 : btrfs_file_extent_ram_bytes(leaf, ei));
5635 0 : TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_OFFSET,
5636 : offset - key.offset + btrfs_file_extent_offset(leaf, ei));
5637 0 : ret = btrfs_encoded_io_compression_from_extent(fs_info,
5638 : btrfs_file_extent_compression(leaf, ei));
5639 0 : if (ret < 0)
5640 0 : goto out;
5641 0 : TLV_PUT_U32(sctx, BTRFS_SEND_A_COMPRESSION, ret);
5642 0 : TLV_PUT_U32(sctx, BTRFS_SEND_A_ENCRYPTION, 0);
5643 :
5644 0 : ret = put_data_header(sctx, disk_num_bytes);
5645 0 : if (ret < 0)
5646 0 : goto out;
5647 :
5648 : /*
5649 : * We want to do I/O directly into the send buffer, so get the next page
5650 : * boundary in the send buffer. This means that there may be a gap
5651 : * between the beginning of the command and the file data.
5652 : */
5653 0 : data_offset = PAGE_ALIGN(sctx->send_size);
5654 0 : if (data_offset > sctx->send_max_size ||
5655 0 : sctx->send_max_size - data_offset < disk_num_bytes) {
5656 0 : ret = -EOVERFLOW;
5657 0 : goto out;
5658 : }
5659 :
5660 : /*
5661 : * Note that send_buf is a mapping of send_buf_pages, so this is really
5662 : * reading into send_buf.
5663 : */
5664 0 : ret = btrfs_encoded_read_regular_fill_pages(BTRFS_I(inode), offset,
5665 : disk_bytenr, disk_num_bytes,
5666 0 : sctx->send_buf_pages +
5667 0 : (data_offset >> PAGE_SHIFT));
5668 0 : if (ret)
5669 0 : goto out;
5670 :
5671 0 : hdr = (struct btrfs_cmd_header *)sctx->send_buf;
5672 0 : hdr->len = cpu_to_le32(sctx->send_size + disk_num_bytes - sizeof(*hdr));
5673 0 : hdr->crc = 0;
5674 0 : crc = btrfs_crc32c(0, sctx->send_buf, sctx->send_size);
5675 0 : crc = btrfs_crc32c(crc, sctx->send_buf + data_offset, disk_num_bytes);
5676 0 : hdr->crc = cpu_to_le32(crc);
5677 :
5678 0 : ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
5679 : &sctx->send_off);
5680 0 : if (!ret) {
5681 0 : ret = write_buf(sctx->send_filp, sctx->send_buf + data_offset,
5682 : disk_num_bytes, &sctx->send_off);
5683 : }
5684 0 : sctx->send_size = 0;
5685 0 : sctx->put_data = false;
5686 :
5687 0 : tlv_put_failure:
5688 0 : out:
5689 0 : fs_path_free(fspath);
5690 0 : iput(inode);
5691 0 : return ret;
5692 : }
5693 :
5694 0 : static int send_extent_data(struct send_ctx *sctx, struct btrfs_path *path,
5695 : const u64 offset, const u64 len)
5696 : {
5697 0 : const u64 end = offset + len;
5698 0 : struct extent_buffer *leaf = path->nodes[0];
5699 0 : struct btrfs_file_extent_item *ei;
5700 0 : u64 read_size = max_send_read_size(sctx);
5701 0 : u64 sent = 0;
5702 :
5703 0 : if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5704 0 : return send_update_extent(sctx, offset, len);
5705 :
5706 0 : ei = btrfs_item_ptr(leaf, path->slots[0],
5707 : struct btrfs_file_extent_item);
5708 0 : if ((sctx->flags & BTRFS_SEND_FLAG_COMPRESSED) &&
5709 : btrfs_file_extent_compression(leaf, ei) != BTRFS_COMPRESS_NONE) {
5710 0 : bool is_inline = (btrfs_file_extent_type(leaf, ei) ==
5711 : BTRFS_FILE_EXTENT_INLINE);
5712 :
5713 : /*
5714 : * Send the compressed extent unless the compressed data is
5715 : * larger than the decompressed data. This can happen if we're
5716 : * not sending the entire extent, either because it has been
5717 : * partially overwritten/truncated or because this is a part of
5718 : * the extent that we couldn't clone in clone_range().
5719 : */
5720 0 : if (is_inline &&
5721 0 : btrfs_file_extent_inline_item_len(leaf,
5722 : path->slots[0]) <= len) {
5723 0 : return send_encoded_inline_extent(sctx, path, offset,
5724 : len);
5725 0 : } else if (!is_inline &&
5726 : btrfs_file_extent_disk_num_bytes(leaf, ei) <= len) {
5727 0 : return send_encoded_extent(sctx, path, offset, len);
5728 : }
5729 : }
5730 :
5731 0 : if (sctx->cur_inode == NULL) {
5732 0 : struct btrfs_root *root = sctx->send_root;
5733 :
5734 0 : sctx->cur_inode = btrfs_iget(root->fs_info->sb, sctx->cur_ino, root);
5735 0 : if (IS_ERR(sctx->cur_inode)) {
5736 0 : int err = PTR_ERR(sctx->cur_inode);
5737 :
5738 0 : sctx->cur_inode = NULL;
5739 0 : return err;
5740 : }
5741 0 : memset(&sctx->ra, 0, sizeof(struct file_ra_state));
5742 0 : file_ra_state_init(&sctx->ra, sctx->cur_inode->i_mapping);
5743 :
5744 : /*
5745 : * It's very likely there are no pages from this inode in the page
5746 : * cache, so after reading extents and sending their data, we clean
5747 : * the page cache to avoid trashing the page cache (adding pressure
5748 : * to the page cache and forcing eviction of other data more useful
5749 : * for applications).
5750 : *
5751 : * We decide if we should clean the page cache simply by checking
5752 : * if the inode's mapping nrpages is 0 when we first open it, and
5753 : * not by using something like filemap_range_has_page() before
5754 : * reading an extent because when we ask the readahead code to
5755 : * read a given file range, it may (and almost always does) read
5756 : * pages from beyond that range (see the documentation for
5757 : * page_cache_sync_readahead()), so it would not be reliable,
5758 : * because after reading the first extent future calls to
5759 : * filemap_range_has_page() would return true because the readahead
5760 : * on the previous extent resulted in reading pages of the current
5761 : * extent as well.
5762 : */
5763 0 : sctx->clean_page_cache = (sctx->cur_inode->i_mapping->nrpages == 0);
5764 0 : sctx->page_cache_clear_start = round_down(offset, PAGE_SIZE);
5765 : }
5766 :
5767 0 : while (sent < len) {
5768 0 : u64 size = min(len - sent, read_size);
5769 0 : int ret;
5770 :
5771 0 : ret = send_write(sctx, offset + sent, size);
5772 0 : if (ret < 0)
5773 0 : return ret;
5774 0 : sent += size;
5775 : }
5776 :
5777 0 : if (sctx->clean_page_cache && PAGE_ALIGNED(end)) {
5778 : /*
5779 : * Always operate only on ranges that are a multiple of the page
5780 : * size. This is not only to prevent zeroing parts of a page in
5781 : * the case of subpage sector size, but also to guarantee we evict
5782 : * pages, as passing a range that is smaller than page size does
5783 : * not evict the respective page (only zeroes part of its content).
5784 : *
5785 : * Always start from the end offset of the last range cleared.
5786 : * This is because the readahead code may (and very often does)
5787 : * reads pages beyond the range we request for readahead. So if
5788 : * we have an extent layout like this:
5789 : *
5790 : * [ extent A ] [ extent B ] [ extent C ]
5791 : *
5792 : * When we ask page_cache_sync_readahead() to read extent A, it
5793 : * may also trigger reads for pages of extent B. If we are doing
5794 : * an incremental send and extent B has not changed between the
5795 : * parent and send snapshots, some or all of its pages may end
5796 : * up being read and placed in the page cache. So when truncating
5797 : * the page cache we always start from the end offset of the
5798 : * previously processed extent up to the end of the current
5799 : * extent.
5800 : */
5801 0 : truncate_inode_pages_range(&sctx->cur_inode->i_data,
5802 0 : sctx->page_cache_clear_start,
5803 0 : end - 1);
5804 0 : sctx->page_cache_clear_start = end;
5805 : }
5806 :
5807 : return 0;
5808 : }
5809 :
5810 : /*
5811 : * Search for a capability xattr related to sctx->cur_ino. If the capability is
5812 : * found, call send_set_xattr function to emit it.
5813 : *
5814 : * Return 0 if there isn't a capability, or when the capability was emitted
5815 : * successfully, or < 0 if an error occurred.
5816 : */
5817 0 : static int send_capabilities(struct send_ctx *sctx)
5818 : {
5819 0 : struct fs_path *fspath = NULL;
5820 0 : struct btrfs_path *path;
5821 0 : struct btrfs_dir_item *di;
5822 0 : struct extent_buffer *leaf;
5823 0 : unsigned long data_ptr;
5824 0 : char *buf = NULL;
5825 0 : int buf_len;
5826 0 : int ret = 0;
5827 :
5828 0 : path = alloc_path_for_send();
5829 0 : if (!path)
5830 : return -ENOMEM;
5831 :
5832 0 : di = btrfs_lookup_xattr(NULL, sctx->send_root, path, sctx->cur_ino,
5833 : XATTR_NAME_CAPS, strlen(XATTR_NAME_CAPS), 0);
5834 0 : if (!di) {
5835 : /* There is no xattr for this inode */
5836 0 : goto out;
5837 0 : } else if (IS_ERR(di)) {
5838 0 : ret = PTR_ERR(di);
5839 0 : goto out;
5840 : }
5841 :
5842 0 : leaf = path->nodes[0];
5843 0 : buf_len = btrfs_dir_data_len(leaf, di);
5844 :
5845 0 : fspath = fs_path_alloc();
5846 0 : buf = kmalloc(buf_len, GFP_KERNEL);
5847 0 : if (!fspath || !buf) {
5848 0 : ret = -ENOMEM;
5849 0 : goto out;
5850 : }
5851 :
5852 0 : ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath);
5853 0 : if (ret < 0)
5854 0 : goto out;
5855 :
5856 0 : data_ptr = (unsigned long)(di + 1) + btrfs_dir_name_len(leaf, di);
5857 0 : read_extent_buffer(leaf, buf, data_ptr, buf_len);
5858 :
5859 0 : ret = send_set_xattr(sctx, fspath, XATTR_NAME_CAPS,
5860 : strlen(XATTR_NAME_CAPS), buf, buf_len);
5861 0 : out:
5862 0 : kfree(buf);
5863 0 : fs_path_free(fspath);
5864 0 : btrfs_free_path(path);
5865 0 : return ret;
5866 : }
5867 :
5868 0 : static int clone_range(struct send_ctx *sctx, struct btrfs_path *dst_path,
5869 : struct clone_root *clone_root, const u64 disk_byte,
5870 : u64 data_offset, u64 offset, u64 len)
5871 : {
5872 0 : struct btrfs_path *path;
5873 0 : struct btrfs_key key;
5874 0 : int ret;
5875 0 : struct btrfs_inode_info info;
5876 0 : u64 clone_src_i_size = 0;
5877 :
5878 : /*
5879 : * Prevent cloning from a zero offset with a length matching the sector
5880 : * size because in some scenarios this will make the receiver fail.
5881 : *
5882 : * For example, if in the source filesystem the extent at offset 0
5883 : * has a length of sectorsize and it was written using direct IO, then
5884 : * it can never be an inline extent (even if compression is enabled).
5885 : * Then this extent can be cloned in the original filesystem to a non
5886 : * zero file offset, but it may not be possible to clone in the
5887 : * destination filesystem because it can be inlined due to compression
5888 : * on the destination filesystem (as the receiver's write operations are
5889 : * always done using buffered IO). The same happens when the original
5890 : * filesystem does not have compression enabled but the destination
5891 : * filesystem has.
5892 : */
5893 0 : if (clone_root->offset == 0 &&
5894 0 : len == sctx->send_root->fs_info->sectorsize)
5895 0 : return send_extent_data(sctx, dst_path, offset, len);
5896 :
5897 0 : path = alloc_path_for_send();
5898 0 : if (!path)
5899 : return -ENOMEM;
5900 :
5901 : /*
5902 : * There are inodes that have extents that lie behind its i_size. Don't
5903 : * accept clones from these extents.
5904 : */
5905 0 : ret = get_inode_info(clone_root->root, clone_root->ino, &info);
5906 0 : btrfs_release_path(path);
5907 0 : if (ret < 0)
5908 0 : goto out;
5909 0 : clone_src_i_size = info.size;
5910 :
5911 : /*
5912 : * We can't send a clone operation for the entire range if we find
5913 : * extent items in the respective range in the source file that
5914 : * refer to different extents or if we find holes.
5915 : * So check for that and do a mix of clone and regular write/copy
5916 : * operations if needed.
5917 : *
5918 : * Example:
5919 : *
5920 : * mkfs.btrfs -f /dev/sda
5921 : * mount /dev/sda /mnt
5922 : * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5923 : * cp --reflink=always /mnt/foo /mnt/bar
5924 : * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5925 : * btrfs subvolume snapshot -r /mnt /mnt/snap
5926 : *
5927 : * If when we send the snapshot and we are processing file bar (which
5928 : * has a higher inode number than foo) we blindly send a clone operation
5929 : * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5930 : * a file bar that matches the content of file foo - iow, doesn't match
5931 : * the content from bar in the original filesystem.
5932 : */
5933 0 : key.objectid = clone_root->ino;
5934 0 : key.type = BTRFS_EXTENT_DATA_KEY;
5935 0 : key.offset = clone_root->offset;
5936 0 : ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5937 0 : if (ret < 0)
5938 0 : goto out;
5939 0 : if (ret > 0 && path->slots[0] > 0) {
5940 0 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5941 0 : if (key.objectid == clone_root->ino &&
5942 0 : key.type == BTRFS_EXTENT_DATA_KEY)
5943 0 : path->slots[0]--;
5944 : }
5945 :
5946 0 : while (true) {
5947 0 : struct extent_buffer *leaf = path->nodes[0];
5948 0 : int slot = path->slots[0];
5949 0 : struct btrfs_file_extent_item *ei;
5950 0 : u8 type;
5951 0 : u64 ext_len;
5952 0 : u64 clone_len;
5953 0 : u64 clone_data_offset;
5954 0 : bool crossed_src_i_size = false;
5955 :
5956 0 : if (slot >= btrfs_header_nritems(leaf)) {
5957 0 : ret = btrfs_next_leaf(clone_root->root, path);
5958 0 : if (ret < 0)
5959 0 : goto out;
5960 0 : else if (ret > 0)
5961 : break;
5962 0 : continue;
5963 : }
5964 :
5965 0 : btrfs_item_key_to_cpu(leaf, &key, slot);
5966 :
5967 : /*
5968 : * We might have an implicit trailing hole (NO_HOLES feature
5969 : * enabled). We deal with it after leaving this loop.
5970 : */
5971 0 : if (key.objectid != clone_root->ino ||
5972 0 : key.type != BTRFS_EXTENT_DATA_KEY)
5973 : break;
5974 :
5975 0 : ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5976 0 : type = btrfs_file_extent_type(leaf, ei);
5977 0 : if (type == BTRFS_FILE_EXTENT_INLINE) {
5978 0 : ext_len = btrfs_file_extent_ram_bytes(leaf, ei);
5979 0 : ext_len = PAGE_ALIGN(ext_len);
5980 : } else {
5981 0 : ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5982 : }
5983 :
5984 0 : if (key.offset + ext_len <= clone_root->offset)
5985 0 : goto next;
5986 :
5987 0 : if (key.offset > clone_root->offset) {
5988 : /* Implicit hole, NO_HOLES feature enabled. */
5989 0 : u64 hole_len = key.offset - clone_root->offset;
5990 :
5991 0 : if (hole_len > len)
5992 : hole_len = len;
5993 0 : ret = send_extent_data(sctx, dst_path, offset,
5994 : hole_len);
5995 0 : if (ret < 0)
5996 0 : goto out;
5997 :
5998 0 : len -= hole_len;
5999 0 : if (len == 0)
6000 : break;
6001 0 : offset += hole_len;
6002 0 : clone_root->offset += hole_len;
6003 0 : data_offset += hole_len;
6004 : }
6005 :
6006 0 : if (key.offset >= clone_root->offset + len)
6007 : break;
6008 :
6009 0 : if (key.offset >= clone_src_i_size)
6010 : break;
6011 :
6012 0 : if (key.offset + ext_len > clone_src_i_size) {
6013 0 : ext_len = clone_src_i_size - key.offset;
6014 0 : crossed_src_i_size = true;
6015 : }
6016 :
6017 0 : clone_data_offset = btrfs_file_extent_offset(leaf, ei);
6018 0 : if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
6019 0 : clone_root->offset = key.offset;
6020 0 : if (clone_data_offset < data_offset &&
6021 0 : clone_data_offset + ext_len > data_offset) {
6022 0 : u64 extent_offset;
6023 :
6024 0 : extent_offset = data_offset - clone_data_offset;
6025 0 : ext_len -= extent_offset;
6026 0 : clone_data_offset += extent_offset;
6027 0 : clone_root->offset += extent_offset;
6028 : }
6029 : }
6030 :
6031 0 : clone_len = min_t(u64, ext_len, len);
6032 :
6033 0 : if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
6034 : clone_data_offset == data_offset) {
6035 0 : const u64 src_end = clone_root->offset + clone_len;
6036 0 : const u64 sectorsize = SZ_64K;
6037 :
6038 : /*
6039 : * We can't clone the last block, when its size is not
6040 : * sector size aligned, into the middle of a file. If we
6041 : * do so, the receiver will get a failure (-EINVAL) when
6042 : * trying to clone or will silently corrupt the data in
6043 : * the destination file if it's on a kernel without the
6044 : * fix introduced by commit ac765f83f1397646
6045 : * ("Btrfs: fix data corruption due to cloning of eof
6046 : * block).
6047 : *
6048 : * So issue a clone of the aligned down range plus a
6049 : * regular write for the eof block, if we hit that case.
6050 : *
6051 : * Also, we use the maximum possible sector size, 64K,
6052 : * because we don't know what's the sector size of the
6053 : * filesystem that receives the stream, so we have to
6054 : * assume the largest possible sector size.
6055 : */
6056 0 : if (src_end == clone_src_i_size &&
6057 0 : !IS_ALIGNED(src_end, sectorsize) &&
6058 0 : offset + clone_len < sctx->cur_inode_size) {
6059 0 : u64 slen;
6060 :
6061 0 : slen = ALIGN_DOWN(src_end - clone_root->offset,
6062 : sectorsize);
6063 0 : if (slen > 0) {
6064 0 : ret = send_clone(sctx, offset, slen,
6065 : clone_root);
6066 0 : if (ret < 0)
6067 0 : goto out;
6068 : }
6069 0 : ret = send_extent_data(sctx, dst_path,
6070 : offset + slen,
6071 : clone_len - slen);
6072 : } else {
6073 0 : ret = send_clone(sctx, offset, clone_len,
6074 : clone_root);
6075 : }
6076 0 : } else if (crossed_src_i_size && clone_len < len) {
6077 : /*
6078 : * If we are at i_size of the clone source inode and we
6079 : * can not clone from it, terminate the loop. This is
6080 : * to avoid sending two write operations, one with a
6081 : * length matching clone_len and the final one after
6082 : * this loop with a length of len - clone_len.
6083 : *
6084 : * When using encoded writes (BTRFS_SEND_FLAG_COMPRESSED
6085 : * was passed to the send ioctl), this helps avoid
6086 : * sending an encoded write for an offset that is not
6087 : * sector size aligned, in case the i_size of the source
6088 : * inode is not sector size aligned. That will make the
6089 : * receiver fallback to decompression of the data and
6090 : * writing it using regular buffered IO, therefore while
6091 : * not incorrect, it's not optimal due decompression and
6092 : * possible re-compression at the receiver.
6093 : */
6094 : break;
6095 : } else {
6096 0 : ret = send_extent_data(sctx, dst_path, offset,
6097 : clone_len);
6098 : }
6099 :
6100 0 : if (ret < 0)
6101 0 : goto out;
6102 :
6103 0 : len -= clone_len;
6104 0 : if (len == 0)
6105 : break;
6106 0 : offset += clone_len;
6107 0 : clone_root->offset += clone_len;
6108 :
6109 : /*
6110 : * If we are cloning from the file we are currently processing,
6111 : * and using the send root as the clone root, we must stop once
6112 : * the current clone offset reaches the current eof of the file
6113 : * at the receiver, otherwise we would issue an invalid clone
6114 : * operation (source range going beyond eof) and cause the
6115 : * receiver to fail. So if we reach the current eof, bail out
6116 : * and fallback to a regular write.
6117 : */
6118 0 : if (clone_root->root == sctx->send_root &&
6119 0 : clone_root->ino == sctx->cur_ino &&
6120 0 : clone_root->offset >= sctx->cur_inode_next_write_offset)
6121 : break;
6122 :
6123 0 : data_offset += clone_len;
6124 0 : next:
6125 0 : path->slots[0]++;
6126 : }
6127 :
6128 0 : if (len > 0)
6129 0 : ret = send_extent_data(sctx, dst_path, offset, len);
6130 : else
6131 : ret = 0;
6132 0 : out:
6133 0 : btrfs_free_path(path);
6134 0 : return ret;
6135 : }
6136 :
6137 0 : static int send_write_or_clone(struct send_ctx *sctx,
6138 : struct btrfs_path *path,
6139 : struct btrfs_key *key,
6140 : struct clone_root *clone_root)
6141 : {
6142 0 : int ret = 0;
6143 0 : u64 offset = key->offset;
6144 0 : u64 end;
6145 0 : u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
6146 :
6147 0 : end = min_t(u64, btrfs_file_extent_end(path), sctx->cur_inode_size);
6148 0 : if (offset >= end)
6149 : return 0;
6150 :
6151 0 : if (clone_root && IS_ALIGNED(end, bs)) {
6152 0 : struct btrfs_file_extent_item *ei;
6153 0 : u64 disk_byte;
6154 0 : u64 data_offset;
6155 :
6156 0 : ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
6157 : struct btrfs_file_extent_item);
6158 0 : disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
6159 0 : data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
6160 0 : ret = clone_range(sctx, path, clone_root, disk_byte,
6161 : data_offset, offset, end - offset);
6162 : } else {
6163 0 : ret = send_extent_data(sctx, path, offset, end - offset);
6164 : }
6165 0 : sctx->cur_inode_next_write_offset = end;
6166 0 : return ret;
6167 : }
6168 :
6169 0 : static int is_extent_unchanged(struct send_ctx *sctx,
6170 : struct btrfs_path *left_path,
6171 : struct btrfs_key *ekey)
6172 : {
6173 0 : int ret = 0;
6174 0 : struct btrfs_key key;
6175 0 : struct btrfs_path *path = NULL;
6176 0 : struct extent_buffer *eb;
6177 0 : int slot;
6178 0 : struct btrfs_key found_key;
6179 0 : struct btrfs_file_extent_item *ei;
6180 0 : u64 left_disknr;
6181 0 : u64 right_disknr;
6182 0 : u64 left_offset;
6183 0 : u64 right_offset;
6184 0 : u64 left_offset_fixed;
6185 0 : u64 left_len;
6186 0 : u64 right_len;
6187 0 : u64 left_gen;
6188 0 : u64 right_gen;
6189 0 : u8 left_type;
6190 0 : u8 right_type;
6191 :
6192 0 : path = alloc_path_for_send();
6193 0 : if (!path)
6194 : return -ENOMEM;
6195 :
6196 0 : eb = left_path->nodes[0];
6197 0 : slot = left_path->slots[0];
6198 0 : ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
6199 0 : left_type = btrfs_file_extent_type(eb, ei);
6200 :
6201 0 : if (left_type != BTRFS_FILE_EXTENT_REG) {
6202 0 : ret = 0;
6203 0 : goto out;
6204 : }
6205 0 : left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
6206 0 : left_len = btrfs_file_extent_num_bytes(eb, ei);
6207 0 : left_offset = btrfs_file_extent_offset(eb, ei);
6208 0 : left_gen = btrfs_file_extent_generation(eb, ei);
6209 :
6210 : /*
6211 : * Following comments will refer to these graphics. L is the left
6212 : * extents which we are checking at the moment. 1-8 are the right
6213 : * extents that we iterate.
6214 : *
6215 : * |-----L-----|
6216 : * |-1-|-2a-|-3-|-4-|-5-|-6-|
6217 : *
6218 : * |-----L-----|
6219 : * |--1--|-2b-|...(same as above)
6220 : *
6221 : * Alternative situation. Happens on files where extents got split.
6222 : * |-----L-----|
6223 : * |-----------7-----------|-6-|
6224 : *
6225 : * Alternative situation. Happens on files which got larger.
6226 : * |-----L-----|
6227 : * |-8-|
6228 : * Nothing follows after 8.
6229 : */
6230 :
6231 0 : key.objectid = ekey->objectid;
6232 0 : key.type = BTRFS_EXTENT_DATA_KEY;
6233 0 : key.offset = ekey->offset;
6234 0 : ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
6235 0 : if (ret < 0)
6236 0 : goto out;
6237 0 : if (ret) {
6238 0 : ret = 0;
6239 0 : goto out;
6240 : }
6241 :
6242 : /*
6243 : * Handle special case where the right side has no extents at all.
6244 : */
6245 0 : eb = path->nodes[0];
6246 0 : slot = path->slots[0];
6247 0 : btrfs_item_key_to_cpu(eb, &found_key, slot);
6248 0 : if (found_key.objectid != key.objectid ||
6249 0 : found_key.type != key.type) {
6250 : /* If we're a hole then just pretend nothing changed */
6251 0 : ret = (left_disknr) ? 0 : 1;
6252 0 : goto out;
6253 : }
6254 :
6255 : /*
6256 : * We're now on 2a, 2b or 7.
6257 : */
6258 0 : key = found_key;
6259 0 : while (key.offset < ekey->offset + left_len) {
6260 0 : ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
6261 0 : right_type = btrfs_file_extent_type(eb, ei);
6262 0 : if (right_type != BTRFS_FILE_EXTENT_REG &&
6263 : right_type != BTRFS_FILE_EXTENT_INLINE) {
6264 0 : ret = 0;
6265 0 : goto out;
6266 : }
6267 :
6268 0 : if (right_type == BTRFS_FILE_EXTENT_INLINE) {
6269 0 : right_len = btrfs_file_extent_ram_bytes(eb, ei);
6270 0 : right_len = PAGE_ALIGN(right_len);
6271 : } else {
6272 0 : right_len = btrfs_file_extent_num_bytes(eb, ei);
6273 : }
6274 :
6275 : /*
6276 : * Are we at extent 8? If yes, we know the extent is changed.
6277 : * This may only happen on the first iteration.
6278 : */
6279 0 : if (found_key.offset + right_len <= ekey->offset) {
6280 : /* If we're a hole just pretend nothing changed */
6281 0 : ret = (left_disknr) ? 0 : 1;
6282 0 : goto out;
6283 : }
6284 :
6285 : /*
6286 : * We just wanted to see if when we have an inline extent, what
6287 : * follows it is a regular extent (wanted to check the above
6288 : * condition for inline extents too). This should normally not
6289 : * happen but it's possible for example when we have an inline
6290 : * compressed extent representing data with a size matching
6291 : * the page size (currently the same as sector size).
6292 : */
6293 0 : if (right_type == BTRFS_FILE_EXTENT_INLINE) {
6294 0 : ret = 0;
6295 0 : goto out;
6296 : }
6297 :
6298 0 : right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
6299 0 : right_offset = btrfs_file_extent_offset(eb, ei);
6300 0 : right_gen = btrfs_file_extent_generation(eb, ei);
6301 :
6302 0 : left_offset_fixed = left_offset;
6303 0 : if (key.offset < ekey->offset) {
6304 : /* Fix the right offset for 2a and 7. */
6305 0 : right_offset += ekey->offset - key.offset;
6306 : } else {
6307 : /* Fix the left offset for all behind 2a and 2b */
6308 0 : left_offset_fixed += key.offset - ekey->offset;
6309 : }
6310 :
6311 : /*
6312 : * Check if we have the same extent.
6313 : */
6314 0 : if (left_disknr != right_disknr ||
6315 0 : left_offset_fixed != right_offset ||
6316 : left_gen != right_gen) {
6317 0 : ret = 0;
6318 0 : goto out;
6319 : }
6320 :
6321 : /*
6322 : * Go to the next extent.
6323 : */
6324 0 : ret = btrfs_next_item(sctx->parent_root, path);
6325 0 : if (ret < 0)
6326 0 : goto out;
6327 0 : if (!ret) {
6328 0 : eb = path->nodes[0];
6329 0 : slot = path->slots[0];
6330 0 : btrfs_item_key_to_cpu(eb, &found_key, slot);
6331 : }
6332 0 : if (ret || found_key.objectid != key.objectid ||
6333 0 : found_key.type != key.type) {
6334 0 : key.offset += right_len;
6335 0 : break;
6336 : }
6337 0 : if (found_key.offset != key.offset + right_len) {
6338 0 : ret = 0;
6339 0 : goto out;
6340 : }
6341 0 : key = found_key;
6342 : }
6343 :
6344 : /*
6345 : * We're now behind the left extent (treat as unchanged) or at the end
6346 : * of the right side (treat as changed).
6347 : */
6348 0 : if (key.offset >= ekey->offset + left_len)
6349 : ret = 1;
6350 : else
6351 0 : ret = 0;
6352 :
6353 :
6354 0 : out:
6355 0 : btrfs_free_path(path);
6356 0 : return ret;
6357 : }
6358 :
6359 0 : static int get_last_extent(struct send_ctx *sctx, u64 offset)
6360 : {
6361 0 : struct btrfs_path *path;
6362 0 : struct btrfs_root *root = sctx->send_root;
6363 0 : struct btrfs_key key;
6364 0 : int ret;
6365 :
6366 0 : path = alloc_path_for_send();
6367 0 : if (!path)
6368 : return -ENOMEM;
6369 :
6370 0 : sctx->cur_inode_last_extent = 0;
6371 :
6372 0 : key.objectid = sctx->cur_ino;
6373 0 : key.type = BTRFS_EXTENT_DATA_KEY;
6374 0 : key.offset = offset;
6375 0 : ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
6376 0 : if (ret < 0)
6377 0 : goto out;
6378 0 : ret = 0;
6379 0 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
6380 0 : if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
6381 0 : goto out;
6382 :
6383 0 : sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
6384 0 : out:
6385 0 : btrfs_free_path(path);
6386 0 : return ret;
6387 : }
6388 :
6389 0 : static int range_is_hole_in_parent(struct send_ctx *sctx,
6390 : const u64 start,
6391 : const u64 end)
6392 : {
6393 0 : struct btrfs_path *path;
6394 0 : struct btrfs_key key;
6395 0 : struct btrfs_root *root = sctx->parent_root;
6396 0 : u64 search_start = start;
6397 0 : int ret;
6398 :
6399 0 : path = alloc_path_for_send();
6400 0 : if (!path)
6401 : return -ENOMEM;
6402 :
6403 0 : key.objectid = sctx->cur_ino;
6404 0 : key.type = BTRFS_EXTENT_DATA_KEY;
6405 0 : key.offset = search_start;
6406 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6407 0 : if (ret < 0)
6408 0 : goto out;
6409 0 : if (ret > 0 && path->slots[0] > 0)
6410 0 : path->slots[0]--;
6411 :
6412 0 : while (search_start < end) {
6413 0 : struct extent_buffer *leaf = path->nodes[0];
6414 0 : int slot = path->slots[0];
6415 0 : struct btrfs_file_extent_item *fi;
6416 0 : u64 extent_end;
6417 :
6418 0 : if (slot >= btrfs_header_nritems(leaf)) {
6419 0 : ret = btrfs_next_leaf(root, path);
6420 0 : if (ret < 0)
6421 0 : goto out;
6422 0 : else if (ret > 0)
6423 : break;
6424 0 : continue;
6425 : }
6426 :
6427 0 : btrfs_item_key_to_cpu(leaf, &key, slot);
6428 0 : if (key.objectid < sctx->cur_ino ||
6429 0 : key.type < BTRFS_EXTENT_DATA_KEY)
6430 0 : goto next;
6431 0 : if (key.objectid > sctx->cur_ino ||
6432 0 : key.type > BTRFS_EXTENT_DATA_KEY ||
6433 0 : key.offset >= end)
6434 : break;
6435 :
6436 0 : fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6437 0 : extent_end = btrfs_file_extent_end(path);
6438 0 : if (extent_end <= start)
6439 0 : goto next;
6440 0 : if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
6441 0 : search_start = extent_end;
6442 0 : goto next;
6443 : }
6444 0 : ret = 0;
6445 0 : goto out;
6446 0 : next:
6447 0 : path->slots[0]++;
6448 : }
6449 : ret = 1;
6450 0 : out:
6451 0 : btrfs_free_path(path);
6452 0 : return ret;
6453 : }
6454 :
6455 0 : static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
6456 : struct btrfs_key *key)
6457 : {
6458 0 : int ret = 0;
6459 :
6460 0 : if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
6461 : return 0;
6462 :
6463 0 : if (sctx->cur_inode_last_extent == (u64)-1) {
6464 0 : ret = get_last_extent(sctx, key->offset - 1);
6465 0 : if (ret)
6466 : return ret;
6467 : }
6468 :
6469 0 : if (path->slots[0] == 0 &&
6470 0 : sctx->cur_inode_last_extent < key->offset) {
6471 : /*
6472 : * We might have skipped entire leafs that contained only
6473 : * file extent items for our current inode. These leafs have
6474 : * a generation number smaller (older) than the one in the
6475 : * current leaf and the leaf our last extent came from, and
6476 : * are located between these 2 leafs.
6477 : */
6478 0 : ret = get_last_extent(sctx, key->offset - 1);
6479 0 : if (ret)
6480 : return ret;
6481 : }
6482 :
6483 0 : if (sctx->cur_inode_last_extent < key->offset) {
6484 0 : ret = range_is_hole_in_parent(sctx,
6485 : sctx->cur_inode_last_extent,
6486 : key->offset);
6487 0 : if (ret < 0)
6488 : return ret;
6489 0 : else if (ret == 0)
6490 0 : ret = send_hole(sctx, key->offset);
6491 : else
6492 : ret = 0;
6493 : }
6494 0 : sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
6495 0 : return ret;
6496 : }
6497 :
6498 0 : static int process_extent(struct send_ctx *sctx,
6499 : struct btrfs_path *path,
6500 : struct btrfs_key *key)
6501 : {
6502 0 : struct clone_root *found_clone = NULL;
6503 0 : int ret = 0;
6504 :
6505 0 : if (S_ISLNK(sctx->cur_inode_mode))
6506 : return 0;
6507 :
6508 0 : if (sctx->parent_root && !sctx->cur_inode_new) {
6509 0 : ret = is_extent_unchanged(sctx, path, key);
6510 0 : if (ret < 0)
6511 0 : goto out;
6512 0 : if (ret) {
6513 0 : ret = 0;
6514 0 : goto out_hole;
6515 : }
6516 : } else {
6517 0 : struct btrfs_file_extent_item *ei;
6518 0 : u8 type;
6519 :
6520 0 : ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
6521 : struct btrfs_file_extent_item);
6522 0 : type = btrfs_file_extent_type(path->nodes[0], ei);
6523 0 : if (type == BTRFS_FILE_EXTENT_PREALLOC ||
6524 : type == BTRFS_FILE_EXTENT_REG) {
6525 : /*
6526 : * The send spec does not have a prealloc command yet,
6527 : * so just leave a hole for prealloc'ed extents until
6528 : * we have enough commands queued up to justify rev'ing
6529 : * the send spec.
6530 : */
6531 0 : if (type == BTRFS_FILE_EXTENT_PREALLOC) {
6532 0 : ret = 0;
6533 0 : goto out;
6534 : }
6535 :
6536 : /* Have a hole, just skip it. */
6537 0 : if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
6538 0 : ret = 0;
6539 0 : goto out;
6540 : }
6541 : }
6542 : }
6543 :
6544 0 : ret = find_extent_clone(sctx, path, key->objectid, key->offset,
6545 : sctx->cur_inode_size, &found_clone);
6546 0 : if (ret != -ENOENT && ret < 0)
6547 0 : goto out;
6548 :
6549 0 : ret = send_write_or_clone(sctx, path, key, found_clone);
6550 0 : if (ret)
6551 0 : goto out;
6552 0 : out_hole:
6553 0 : ret = maybe_send_hole(sctx, path, key);
6554 : out:
6555 : return ret;
6556 : }
6557 :
6558 0 : static int process_all_extents(struct send_ctx *sctx)
6559 : {
6560 0 : int ret = 0;
6561 0 : int iter_ret = 0;
6562 0 : struct btrfs_root *root;
6563 0 : struct btrfs_path *path;
6564 0 : struct btrfs_key key;
6565 0 : struct btrfs_key found_key;
6566 :
6567 0 : root = sctx->send_root;
6568 0 : path = alloc_path_for_send();
6569 0 : if (!path)
6570 : return -ENOMEM;
6571 :
6572 0 : key.objectid = sctx->cmp_key->objectid;
6573 0 : key.type = BTRFS_EXTENT_DATA_KEY;
6574 0 : key.offset = 0;
6575 0 : btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
6576 0 : if (found_key.objectid != key.objectid ||
6577 0 : found_key.type != key.type) {
6578 : ret = 0;
6579 : break;
6580 : }
6581 :
6582 0 : ret = process_extent(sctx, path, &found_key);
6583 0 : if (ret < 0)
6584 : break;
6585 : }
6586 : /* Catch error found during iteration */
6587 0 : if (iter_ret < 0)
6588 0 : ret = iter_ret;
6589 :
6590 0 : btrfs_free_path(path);
6591 0 : return ret;
6592 : }
6593 :
6594 0 : static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
6595 : int *pending_move,
6596 : int *refs_processed)
6597 : {
6598 0 : int ret = 0;
6599 :
6600 0 : if (sctx->cur_ino == 0)
6601 0 : goto out;
6602 0 : if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
6603 0 : sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
6604 0 : goto out;
6605 0 : if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
6606 0 : goto out;
6607 :
6608 0 : ret = process_recorded_refs(sctx, pending_move);
6609 0 : if (ret < 0)
6610 0 : goto out;
6611 :
6612 0 : *refs_processed = 1;
6613 0 : out:
6614 0 : return ret;
6615 : }
6616 :
6617 0 : static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
6618 : {
6619 0 : int ret = 0;
6620 0 : struct btrfs_inode_info info;
6621 0 : u64 left_mode;
6622 0 : u64 left_uid;
6623 0 : u64 left_gid;
6624 0 : u64 left_fileattr;
6625 0 : u64 right_mode;
6626 0 : u64 right_uid;
6627 0 : u64 right_gid;
6628 0 : u64 right_fileattr;
6629 0 : int need_chmod = 0;
6630 0 : int need_chown = 0;
6631 0 : bool need_fileattr = false;
6632 0 : int need_truncate = 1;
6633 0 : int pending_move = 0;
6634 0 : int refs_processed = 0;
6635 :
6636 0 : if (sctx->ignore_cur_inode)
6637 : return 0;
6638 :
6639 0 : ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
6640 : &refs_processed);
6641 0 : if (ret < 0)
6642 0 : goto out;
6643 :
6644 : /*
6645 : * We have processed the refs and thus need to advance send_progress.
6646 : * Now, calls to get_cur_xxx will take the updated refs of the current
6647 : * inode into account.
6648 : *
6649 : * On the other hand, if our current inode is a directory and couldn't
6650 : * be moved/renamed because its parent was renamed/moved too and it has
6651 : * a higher inode number, we can only move/rename our current inode
6652 : * after we moved/renamed its parent. Therefore in this case operate on
6653 : * the old path (pre move/rename) of our current inode, and the
6654 : * move/rename will be performed later.
6655 : */
6656 0 : if (refs_processed && !pending_move)
6657 0 : sctx->send_progress = sctx->cur_ino + 1;
6658 :
6659 0 : if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
6660 0 : goto out;
6661 0 : if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
6662 0 : goto out;
6663 0 : ret = get_inode_info(sctx->send_root, sctx->cur_ino, &info);
6664 0 : if (ret < 0)
6665 0 : goto out;
6666 0 : left_mode = info.mode;
6667 0 : left_uid = info.uid;
6668 0 : left_gid = info.gid;
6669 0 : left_fileattr = info.fileattr;
6670 :
6671 0 : if (!sctx->parent_root || sctx->cur_inode_new) {
6672 0 : need_chown = 1;
6673 0 : if (!S_ISLNK(sctx->cur_inode_mode))
6674 0 : need_chmod = 1;
6675 0 : if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
6676 0 : need_truncate = 0;
6677 : } else {
6678 0 : u64 old_size;
6679 :
6680 0 : ret = get_inode_info(sctx->parent_root, sctx->cur_ino, &info);
6681 0 : if (ret < 0)
6682 0 : goto out;
6683 0 : old_size = info.size;
6684 0 : right_mode = info.mode;
6685 0 : right_uid = info.uid;
6686 0 : right_gid = info.gid;
6687 0 : right_fileattr = info.fileattr;
6688 :
6689 0 : if (left_uid != right_uid || left_gid != right_gid)
6690 0 : need_chown = 1;
6691 0 : if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
6692 0 : need_chmod = 1;
6693 0 : if (!S_ISLNK(sctx->cur_inode_mode) && left_fileattr != right_fileattr)
6694 0 : need_fileattr = true;
6695 0 : if ((old_size == sctx->cur_inode_size) ||
6696 0 : (sctx->cur_inode_size > old_size &&
6697 0 : sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
6698 0 : need_truncate = 0;
6699 : }
6700 :
6701 0 : if (S_ISREG(sctx->cur_inode_mode)) {
6702 0 : if (need_send_hole(sctx)) {
6703 0 : if (sctx->cur_inode_last_extent == (u64)-1 ||
6704 : sctx->cur_inode_last_extent <
6705 0 : sctx->cur_inode_size) {
6706 0 : ret = get_last_extent(sctx, (u64)-1);
6707 0 : if (ret)
6708 0 : goto out;
6709 : }
6710 0 : if (sctx->cur_inode_last_extent <
6711 0 : sctx->cur_inode_size) {
6712 0 : ret = send_hole(sctx, sctx->cur_inode_size);
6713 0 : if (ret)
6714 0 : goto out;
6715 : }
6716 : }
6717 0 : if (need_truncate) {
6718 0 : ret = send_truncate(sctx, sctx->cur_ino,
6719 : sctx->cur_inode_gen,
6720 : sctx->cur_inode_size);
6721 0 : if (ret < 0)
6722 0 : goto out;
6723 : }
6724 : }
6725 :
6726 0 : if (need_chown) {
6727 0 : ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6728 : left_uid, left_gid);
6729 0 : if (ret < 0)
6730 0 : goto out;
6731 : }
6732 0 : if (need_chmod) {
6733 0 : ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6734 : left_mode);
6735 0 : if (ret < 0)
6736 0 : goto out;
6737 : }
6738 0 : if (need_fileattr) {
6739 0 : ret = send_fileattr(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6740 : left_fileattr);
6741 0 : if (ret < 0)
6742 0 : goto out;
6743 : }
6744 :
6745 0 : if (proto_cmd_ok(sctx, BTRFS_SEND_C_ENABLE_VERITY)
6746 0 : && sctx->cur_inode_needs_verity) {
6747 0 : ret = process_verity(sctx);
6748 0 : if (ret < 0)
6749 0 : goto out;
6750 : }
6751 :
6752 0 : ret = send_capabilities(sctx);
6753 0 : if (ret < 0)
6754 0 : goto out;
6755 :
6756 : /*
6757 : * If other directory inodes depended on our current directory
6758 : * inode's move/rename, now do their move/rename operations.
6759 : */
6760 0 : if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
6761 0 : ret = apply_children_dir_moves(sctx);
6762 0 : if (ret)
6763 0 : goto out;
6764 : /*
6765 : * Need to send that every time, no matter if it actually
6766 : * changed between the two trees as we have done changes to
6767 : * the inode before. If our inode is a directory and it's
6768 : * waiting to be moved/renamed, we will send its utimes when
6769 : * it's moved/renamed, therefore we don't need to do it here.
6770 : */
6771 0 : sctx->send_progress = sctx->cur_ino + 1;
6772 :
6773 : /*
6774 : * If the current inode is a non-empty directory, delay issuing
6775 : * the utimes command for it, as it's very likely we have inodes
6776 : * with an higher number inside it. We want to issue the utimes
6777 : * command only after adding all dentries to it.
6778 : */
6779 0 : if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_size > 0)
6780 0 : ret = cache_dir_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
6781 : else
6782 0 : ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
6783 :
6784 0 : if (ret < 0)
6785 0 : goto out;
6786 : }
6787 :
6788 0 : out:
6789 0 : if (!ret)
6790 0 : ret = trim_dir_utimes_cache(sctx);
6791 :
6792 : return ret;
6793 : }
6794 :
6795 0 : static void close_current_inode(struct send_ctx *sctx)
6796 : {
6797 0 : u64 i_size;
6798 :
6799 0 : if (sctx->cur_inode == NULL)
6800 : return;
6801 :
6802 0 : i_size = i_size_read(sctx->cur_inode);
6803 :
6804 : /*
6805 : * If we are doing an incremental send, we may have extents between the
6806 : * last processed extent and the i_size that have not been processed
6807 : * because they haven't changed but we may have read some of their pages
6808 : * through readahead, see the comments at send_extent_data().
6809 : */
6810 0 : if (sctx->clean_page_cache && sctx->page_cache_clear_start < i_size)
6811 0 : truncate_inode_pages_range(&sctx->cur_inode->i_data,
6812 : sctx->page_cache_clear_start,
6813 0 : round_up(i_size, PAGE_SIZE) - 1);
6814 :
6815 0 : iput(sctx->cur_inode);
6816 0 : sctx->cur_inode = NULL;
6817 : }
6818 :
6819 0 : static int changed_inode(struct send_ctx *sctx,
6820 : enum btrfs_compare_tree_result result)
6821 : {
6822 0 : int ret = 0;
6823 0 : struct btrfs_key *key = sctx->cmp_key;
6824 0 : struct btrfs_inode_item *left_ii = NULL;
6825 0 : struct btrfs_inode_item *right_ii = NULL;
6826 0 : u64 left_gen = 0;
6827 0 : u64 right_gen = 0;
6828 :
6829 0 : close_current_inode(sctx);
6830 :
6831 0 : sctx->cur_ino = key->objectid;
6832 0 : sctx->cur_inode_new_gen = false;
6833 0 : sctx->cur_inode_last_extent = (u64)-1;
6834 0 : sctx->cur_inode_next_write_offset = 0;
6835 0 : sctx->ignore_cur_inode = false;
6836 :
6837 : /*
6838 : * Set send_progress to current inode. This will tell all get_cur_xxx
6839 : * functions that the current inode's refs are not updated yet. Later,
6840 : * when process_recorded_refs is finished, it is set to cur_ino + 1.
6841 : */
6842 0 : sctx->send_progress = sctx->cur_ino;
6843 :
6844 0 : if (result == BTRFS_COMPARE_TREE_NEW ||
6845 0 : result == BTRFS_COMPARE_TREE_CHANGED) {
6846 0 : left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6847 : sctx->left_path->slots[0],
6848 : struct btrfs_inode_item);
6849 0 : left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6850 : left_ii);
6851 : } else {
6852 0 : right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6853 : sctx->right_path->slots[0],
6854 : struct btrfs_inode_item);
6855 0 : right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6856 : right_ii);
6857 : }
6858 0 : if (result == BTRFS_COMPARE_TREE_CHANGED) {
6859 0 : right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6860 : sctx->right_path->slots[0],
6861 : struct btrfs_inode_item);
6862 :
6863 0 : right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6864 : right_ii);
6865 :
6866 : /*
6867 : * The cur_ino = root dir case is special here. We can't treat
6868 : * the inode as deleted+reused because it would generate a
6869 : * stream that tries to delete/mkdir the root dir.
6870 : */
6871 0 : if (left_gen != right_gen &&
6872 0 : sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6873 0 : sctx->cur_inode_new_gen = true;
6874 : }
6875 :
6876 : /*
6877 : * Normally we do not find inodes with a link count of zero (orphans)
6878 : * because the most common case is to create a snapshot and use it
6879 : * for a send operation. However other less common use cases involve
6880 : * using a subvolume and send it after turning it to RO mode just
6881 : * after deleting all hard links of a file while holding an open
6882 : * file descriptor against it or turning a RO snapshot into RW mode,
6883 : * keep an open file descriptor against a file, delete it and then
6884 : * turn the snapshot back to RO mode before using it for a send
6885 : * operation. The former is what the receiver operation does.
6886 : * Therefore, if we want to send these snapshots soon after they're
6887 : * received, we need to handle orphan inodes as well. Moreover, orphans
6888 : * can appear not only in the send snapshot but also in the parent
6889 : * snapshot. Here are several cases:
6890 : *
6891 : * Case 1: BTRFS_COMPARE_TREE_NEW
6892 : * | send snapshot | action
6893 : * --------------------------------
6894 : * nlink | 0 | ignore
6895 : *
6896 : * Case 2: BTRFS_COMPARE_TREE_DELETED
6897 : * | parent snapshot | action
6898 : * ----------------------------------
6899 : * nlink | 0 | as usual
6900 : * Note: No unlinks will be sent because there're no paths for it.
6901 : *
6902 : * Case 3: BTRFS_COMPARE_TREE_CHANGED
6903 : * | | parent snapshot | send snapshot | action
6904 : * -----------------------------------------------------------------------
6905 : * subcase 1 | nlink | 0 | 0 | ignore
6906 : * subcase 2 | nlink | >0 | 0 | new_gen(deletion)
6907 : * subcase 3 | nlink | 0 | >0 | new_gen(creation)
6908 : *
6909 : */
6910 0 : if (result == BTRFS_COMPARE_TREE_NEW) {
6911 0 : if (btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii) == 0) {
6912 0 : sctx->ignore_cur_inode = true;
6913 0 : goto out;
6914 : }
6915 0 : sctx->cur_inode_gen = left_gen;
6916 0 : sctx->cur_inode_new = true;
6917 0 : sctx->cur_inode_deleted = false;
6918 0 : sctx->cur_inode_size = btrfs_inode_size(
6919 0 : sctx->left_path->nodes[0], left_ii);
6920 0 : sctx->cur_inode_mode = btrfs_inode_mode(
6921 0 : sctx->left_path->nodes[0], left_ii);
6922 0 : sctx->cur_inode_rdev = btrfs_inode_rdev(
6923 0 : sctx->left_path->nodes[0], left_ii);
6924 0 : if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6925 0 : ret = send_create_inode_if_needed(sctx);
6926 0 : } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6927 0 : sctx->cur_inode_gen = right_gen;
6928 0 : sctx->cur_inode_new = false;
6929 0 : sctx->cur_inode_deleted = true;
6930 0 : sctx->cur_inode_size = btrfs_inode_size(
6931 0 : sctx->right_path->nodes[0], right_ii);
6932 0 : sctx->cur_inode_mode = btrfs_inode_mode(
6933 0 : sctx->right_path->nodes[0], right_ii);
6934 0 : } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
6935 0 : u32 new_nlinks, old_nlinks;
6936 :
6937 0 : new_nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6938 0 : old_nlinks = btrfs_inode_nlink(sctx->right_path->nodes[0], right_ii);
6939 0 : if (new_nlinks == 0 && old_nlinks == 0) {
6940 0 : sctx->ignore_cur_inode = true;
6941 0 : goto out;
6942 0 : } else if (new_nlinks == 0 || old_nlinks == 0) {
6943 0 : sctx->cur_inode_new_gen = 1;
6944 : }
6945 : /*
6946 : * We need to do some special handling in case the inode was
6947 : * reported as changed with a changed generation number. This
6948 : * means that the original inode was deleted and new inode
6949 : * reused the same inum. So we have to treat the old inode as
6950 : * deleted and the new one as new.
6951 : */
6952 0 : if (sctx->cur_inode_new_gen) {
6953 : /*
6954 : * First, process the inode as if it was deleted.
6955 : */
6956 0 : if (old_nlinks > 0) {
6957 0 : sctx->cur_inode_gen = right_gen;
6958 0 : sctx->cur_inode_new = false;
6959 0 : sctx->cur_inode_deleted = true;
6960 0 : sctx->cur_inode_size = btrfs_inode_size(
6961 0 : sctx->right_path->nodes[0], right_ii);
6962 0 : sctx->cur_inode_mode = btrfs_inode_mode(
6963 0 : sctx->right_path->nodes[0], right_ii);
6964 0 : ret = process_all_refs(sctx,
6965 : BTRFS_COMPARE_TREE_DELETED);
6966 0 : if (ret < 0)
6967 0 : goto out;
6968 : }
6969 :
6970 : /*
6971 : * Now process the inode as if it was new.
6972 : */
6973 0 : if (new_nlinks > 0) {
6974 0 : sctx->cur_inode_gen = left_gen;
6975 0 : sctx->cur_inode_new = true;
6976 0 : sctx->cur_inode_deleted = false;
6977 0 : sctx->cur_inode_size = btrfs_inode_size(
6978 0 : sctx->left_path->nodes[0],
6979 : left_ii);
6980 0 : sctx->cur_inode_mode = btrfs_inode_mode(
6981 0 : sctx->left_path->nodes[0],
6982 : left_ii);
6983 0 : sctx->cur_inode_rdev = btrfs_inode_rdev(
6984 0 : sctx->left_path->nodes[0],
6985 : left_ii);
6986 0 : ret = send_create_inode_if_needed(sctx);
6987 0 : if (ret < 0)
6988 0 : goto out;
6989 :
6990 0 : ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6991 0 : if (ret < 0)
6992 0 : goto out;
6993 : /*
6994 : * Advance send_progress now as we did not get
6995 : * into process_recorded_refs_if_needed in the
6996 : * new_gen case.
6997 : */
6998 0 : sctx->send_progress = sctx->cur_ino + 1;
6999 :
7000 : /*
7001 : * Now process all extents and xattrs of the
7002 : * inode as if they were all new.
7003 : */
7004 0 : ret = process_all_extents(sctx);
7005 0 : if (ret < 0)
7006 0 : goto out;
7007 0 : ret = process_all_new_xattrs(sctx);
7008 0 : if (ret < 0)
7009 0 : goto out;
7010 : }
7011 : } else {
7012 0 : sctx->cur_inode_gen = left_gen;
7013 0 : sctx->cur_inode_new = false;
7014 0 : sctx->cur_inode_new_gen = false;
7015 0 : sctx->cur_inode_deleted = false;
7016 0 : sctx->cur_inode_size = btrfs_inode_size(
7017 0 : sctx->left_path->nodes[0], left_ii);
7018 0 : sctx->cur_inode_mode = btrfs_inode_mode(
7019 0 : sctx->left_path->nodes[0], left_ii);
7020 : }
7021 : }
7022 :
7023 0 : out:
7024 0 : return ret;
7025 : }
7026 :
7027 : /*
7028 : * We have to process new refs before deleted refs, but compare_trees gives us
7029 : * the new and deleted refs mixed. To fix this, we record the new/deleted refs
7030 : * first and later process them in process_recorded_refs.
7031 : * For the cur_inode_new_gen case, we skip recording completely because
7032 : * changed_inode did already initiate processing of refs. The reason for this is
7033 : * that in this case, compare_tree actually compares the refs of 2 different
7034 : * inodes. To fix this, process_all_refs is used in changed_inode to handle all
7035 : * refs of the right tree as deleted and all refs of the left tree as new.
7036 : */
7037 0 : static int changed_ref(struct send_ctx *sctx,
7038 : enum btrfs_compare_tree_result result)
7039 : {
7040 0 : int ret = 0;
7041 :
7042 0 : if (sctx->cur_ino != sctx->cmp_key->objectid) {
7043 0 : inconsistent_snapshot_error(sctx, result, "reference");
7044 0 : return -EIO;
7045 : }
7046 :
7047 0 : if (!sctx->cur_inode_new_gen &&
7048 : sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
7049 0 : if (result == BTRFS_COMPARE_TREE_NEW)
7050 0 : ret = record_new_ref(sctx);
7051 0 : else if (result == BTRFS_COMPARE_TREE_DELETED)
7052 0 : ret = record_deleted_ref(sctx);
7053 0 : else if (result == BTRFS_COMPARE_TREE_CHANGED)
7054 0 : ret = record_changed_ref(sctx);
7055 : }
7056 :
7057 : return ret;
7058 : }
7059 :
7060 : /*
7061 : * Process new/deleted/changed xattrs. We skip processing in the
7062 : * cur_inode_new_gen case because changed_inode did already initiate processing
7063 : * of xattrs. The reason is the same as in changed_ref
7064 : */
7065 0 : static int changed_xattr(struct send_ctx *sctx,
7066 : enum btrfs_compare_tree_result result)
7067 : {
7068 0 : int ret = 0;
7069 :
7070 0 : if (sctx->cur_ino != sctx->cmp_key->objectid) {
7071 0 : inconsistent_snapshot_error(sctx, result, "xattr");
7072 0 : return -EIO;
7073 : }
7074 :
7075 0 : if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
7076 0 : if (result == BTRFS_COMPARE_TREE_NEW)
7077 0 : ret = process_new_xattr(sctx);
7078 0 : else if (result == BTRFS_COMPARE_TREE_DELETED)
7079 0 : ret = process_deleted_xattr(sctx);
7080 0 : else if (result == BTRFS_COMPARE_TREE_CHANGED)
7081 0 : ret = process_changed_xattr(sctx);
7082 : }
7083 :
7084 : return ret;
7085 : }
7086 :
7087 : /*
7088 : * Process new/deleted/changed extents. We skip processing in the
7089 : * cur_inode_new_gen case because changed_inode did already initiate processing
7090 : * of extents. The reason is the same as in changed_ref
7091 : */
7092 0 : static int changed_extent(struct send_ctx *sctx,
7093 : enum btrfs_compare_tree_result result)
7094 : {
7095 0 : int ret = 0;
7096 :
7097 : /*
7098 : * We have found an extent item that changed without the inode item
7099 : * having changed. This can happen either after relocation (where the
7100 : * disk_bytenr of an extent item is replaced at
7101 : * relocation.c:replace_file_extents()) or after deduplication into a
7102 : * file in both the parent and send snapshots (where an extent item can
7103 : * get modified or replaced with a new one). Note that deduplication
7104 : * updates the inode item, but it only changes the iversion (sequence
7105 : * field in the inode item) of the inode, so if a file is deduplicated
7106 : * the same amount of times in both the parent and send snapshots, its
7107 : * iversion becomes the same in both snapshots, whence the inode item is
7108 : * the same on both snapshots.
7109 : */
7110 0 : if (sctx->cur_ino != sctx->cmp_key->objectid)
7111 : return 0;
7112 :
7113 0 : if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
7114 0 : if (result != BTRFS_COMPARE_TREE_DELETED)
7115 0 : ret = process_extent(sctx, sctx->left_path,
7116 : sctx->cmp_key);
7117 : }
7118 :
7119 : return ret;
7120 : }
7121 :
7122 : static int changed_verity(struct send_ctx *sctx, enum btrfs_compare_tree_result result)
7123 : {
7124 0 : int ret = 0;
7125 :
7126 0 : if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
7127 0 : if (result == BTRFS_COMPARE_TREE_NEW)
7128 0 : sctx->cur_inode_needs_verity = true;
7129 : }
7130 : return ret;
7131 : }
7132 :
7133 0 : static int dir_changed(struct send_ctx *sctx, u64 dir)
7134 : {
7135 0 : u64 orig_gen, new_gen;
7136 0 : int ret;
7137 :
7138 0 : ret = get_inode_gen(sctx->send_root, dir, &new_gen);
7139 0 : if (ret)
7140 : return ret;
7141 :
7142 0 : ret = get_inode_gen(sctx->parent_root, dir, &orig_gen);
7143 0 : if (ret)
7144 : return ret;
7145 :
7146 0 : return (orig_gen != new_gen) ? 1 : 0;
7147 : }
7148 :
7149 0 : static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
7150 : struct btrfs_key *key)
7151 : {
7152 0 : struct btrfs_inode_extref *extref;
7153 0 : struct extent_buffer *leaf;
7154 0 : u64 dirid = 0, last_dirid = 0;
7155 0 : unsigned long ptr;
7156 0 : u32 item_size;
7157 0 : u32 cur_offset = 0;
7158 0 : int ref_name_len;
7159 0 : int ret = 0;
7160 :
7161 : /* Easy case, just check this one dirid */
7162 0 : if (key->type == BTRFS_INODE_REF_KEY) {
7163 0 : dirid = key->offset;
7164 :
7165 0 : ret = dir_changed(sctx, dirid);
7166 0 : goto out;
7167 : }
7168 :
7169 0 : leaf = path->nodes[0];
7170 0 : item_size = btrfs_item_size(leaf, path->slots[0]);
7171 0 : ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
7172 0 : while (cur_offset < item_size) {
7173 0 : extref = (struct btrfs_inode_extref *)(ptr +
7174 : cur_offset);
7175 0 : dirid = btrfs_inode_extref_parent(leaf, extref);
7176 0 : ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
7177 0 : cur_offset += ref_name_len + sizeof(*extref);
7178 0 : if (dirid == last_dirid)
7179 0 : continue;
7180 0 : ret = dir_changed(sctx, dirid);
7181 0 : if (ret)
7182 : break;
7183 : last_dirid = dirid;
7184 : }
7185 0 : out:
7186 0 : return ret;
7187 : }
7188 :
7189 : /*
7190 : * Updates compare related fields in sctx and simply forwards to the actual
7191 : * changed_xxx functions.
7192 : */
7193 0 : static int changed_cb(struct btrfs_path *left_path,
7194 : struct btrfs_path *right_path,
7195 : struct btrfs_key *key,
7196 : enum btrfs_compare_tree_result result,
7197 : struct send_ctx *sctx)
7198 : {
7199 0 : int ret = 0;
7200 :
7201 : /*
7202 : * We can not hold the commit root semaphore here. This is because in
7203 : * the case of sending and receiving to the same filesystem, using a
7204 : * pipe, could result in a deadlock:
7205 : *
7206 : * 1) The task running send blocks on the pipe because it's full;
7207 : *
7208 : * 2) The task running receive, which is the only consumer of the pipe,
7209 : * is waiting for a transaction commit (for example due to a space
7210 : * reservation when doing a write or triggering a transaction commit
7211 : * when creating a subvolume);
7212 : *
7213 : * 3) The transaction is waiting to write lock the commit root semaphore,
7214 : * but can not acquire it since it's being held at 1).
7215 : *
7216 : * Down this call chain we write to the pipe through kernel_write().
7217 : * The same type of problem can also happen when sending to a file that
7218 : * is stored in the same filesystem - when reserving space for a write
7219 : * into the file, we can trigger a transaction commit.
7220 : *
7221 : * Our caller has supplied us with clones of leaves from the send and
7222 : * parent roots, so we're safe here from a concurrent relocation and
7223 : * further reallocation of metadata extents while we are here. Below we
7224 : * also assert that the leaves are clones.
7225 : */
7226 0 : lockdep_assert_not_held(&sctx->send_root->fs_info->commit_root_sem);
7227 :
7228 : /*
7229 : * We always have a send root, so left_path is never NULL. We will not
7230 : * have a leaf when we have reached the end of the send root but have
7231 : * not yet reached the end of the parent root.
7232 : */
7233 0 : if (left_path->nodes[0])
7234 0 : ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED,
7235 : &left_path->nodes[0]->bflags));
7236 : /*
7237 : * When doing a full send we don't have a parent root, so right_path is
7238 : * NULL. When doing an incremental send, we may have reached the end of
7239 : * the parent root already, so we don't have a leaf at right_path.
7240 : */
7241 0 : if (right_path && right_path->nodes[0])
7242 0 : ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED,
7243 : &right_path->nodes[0]->bflags));
7244 :
7245 0 : if (result == BTRFS_COMPARE_TREE_SAME) {
7246 0 : if (key->type == BTRFS_INODE_REF_KEY ||
7247 : key->type == BTRFS_INODE_EXTREF_KEY) {
7248 0 : ret = compare_refs(sctx, left_path, key);
7249 0 : if (!ret)
7250 : return 0;
7251 0 : if (ret < 0)
7252 : return ret;
7253 0 : } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
7254 0 : return maybe_send_hole(sctx, left_path, key);
7255 : } else {
7256 : return 0;
7257 : }
7258 : result = BTRFS_COMPARE_TREE_CHANGED;
7259 : ret = 0;
7260 : }
7261 :
7262 0 : sctx->left_path = left_path;
7263 0 : sctx->right_path = right_path;
7264 0 : sctx->cmp_key = key;
7265 :
7266 0 : ret = finish_inode_if_needed(sctx, 0);
7267 0 : if (ret < 0)
7268 0 : goto out;
7269 :
7270 : /* Ignore non-FS objects */
7271 0 : if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
7272 : key->objectid == BTRFS_FREE_SPACE_OBJECTID)
7273 0 : goto out;
7274 :
7275 0 : if (key->type == BTRFS_INODE_ITEM_KEY) {
7276 0 : ret = changed_inode(sctx, result);
7277 0 : } else if (!sctx->ignore_cur_inode) {
7278 0 : if (key->type == BTRFS_INODE_REF_KEY ||
7279 : key->type == BTRFS_INODE_EXTREF_KEY)
7280 0 : ret = changed_ref(sctx, result);
7281 0 : else if (key->type == BTRFS_XATTR_ITEM_KEY)
7282 0 : ret = changed_xattr(sctx, result);
7283 0 : else if (key->type == BTRFS_EXTENT_DATA_KEY)
7284 0 : ret = changed_extent(sctx, result);
7285 0 : else if (key->type == BTRFS_VERITY_DESC_ITEM_KEY &&
7286 0 : key->offset == 0)
7287 0 : ret = changed_verity(sctx, result);
7288 : }
7289 :
7290 0 : out:
7291 : return ret;
7292 : }
7293 :
7294 0 : static int search_key_again(const struct send_ctx *sctx,
7295 : struct btrfs_root *root,
7296 : struct btrfs_path *path,
7297 : const struct btrfs_key *key)
7298 : {
7299 0 : int ret;
7300 :
7301 0 : if (!path->need_commit_sem)
7302 0 : lockdep_assert_held_read(&root->fs_info->commit_root_sem);
7303 :
7304 : /*
7305 : * Roots used for send operations are readonly and no one can add,
7306 : * update or remove keys from them, so we should be able to find our
7307 : * key again. The only exception is deduplication, which can operate on
7308 : * readonly roots and add, update or remove keys to/from them - but at
7309 : * the moment we don't allow it to run in parallel with send.
7310 : */
7311 0 : ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7312 0 : ASSERT(ret <= 0);
7313 0 : if (ret > 0) {
7314 0 : btrfs_print_tree(path->nodes[path->lowest_level], false);
7315 0 : btrfs_err(root->fs_info,
7316 : "send: key (%llu %u %llu) not found in %s root %llu, lowest_level %d, slot %d",
7317 : key->objectid, key->type, key->offset,
7318 : (root == sctx->parent_root ? "parent" : "send"),
7319 : root->root_key.objectid, path->lowest_level,
7320 : path->slots[path->lowest_level]);
7321 0 : return -EUCLEAN;
7322 : }
7323 :
7324 : return ret;
7325 : }
7326 :
7327 0 : static int full_send_tree(struct send_ctx *sctx)
7328 : {
7329 0 : int ret;
7330 0 : struct btrfs_root *send_root = sctx->send_root;
7331 0 : struct btrfs_key key;
7332 0 : struct btrfs_fs_info *fs_info = send_root->fs_info;
7333 0 : struct btrfs_path *path;
7334 :
7335 0 : path = alloc_path_for_send();
7336 0 : if (!path)
7337 : return -ENOMEM;
7338 0 : path->reada = READA_FORWARD_ALWAYS;
7339 :
7340 0 : key.objectid = BTRFS_FIRST_FREE_OBJECTID;
7341 0 : key.type = BTRFS_INODE_ITEM_KEY;
7342 0 : key.offset = 0;
7343 :
7344 0 : down_read(&fs_info->commit_root_sem);
7345 0 : sctx->last_reloc_trans = fs_info->last_reloc_trans;
7346 0 : up_read(&fs_info->commit_root_sem);
7347 :
7348 0 : ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
7349 0 : if (ret < 0)
7350 0 : goto out;
7351 0 : if (ret)
7352 0 : goto out_finish;
7353 :
7354 0 : while (1) {
7355 0 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
7356 :
7357 0 : ret = changed_cb(path, NULL, &key,
7358 : BTRFS_COMPARE_TREE_NEW, sctx);
7359 0 : if (ret < 0)
7360 0 : goto out;
7361 :
7362 0 : down_read(&fs_info->commit_root_sem);
7363 0 : if (fs_info->last_reloc_trans > sctx->last_reloc_trans) {
7364 0 : sctx->last_reloc_trans = fs_info->last_reloc_trans;
7365 0 : up_read(&fs_info->commit_root_sem);
7366 : /*
7367 : * A transaction used for relocating a block group was
7368 : * committed or is about to finish its commit. Release
7369 : * our path (leaf) and restart the search, so that we
7370 : * avoid operating on any file extent items that are
7371 : * stale, with a disk_bytenr that reflects a pre
7372 : * relocation value. This way we avoid as much as
7373 : * possible to fallback to regular writes when checking
7374 : * if we can clone file ranges.
7375 : */
7376 0 : btrfs_release_path(path);
7377 0 : ret = search_key_again(sctx, send_root, path, &key);
7378 0 : if (ret < 0)
7379 0 : goto out;
7380 : } else {
7381 0 : up_read(&fs_info->commit_root_sem);
7382 : }
7383 :
7384 0 : ret = btrfs_next_item(send_root, path);
7385 0 : if (ret < 0)
7386 0 : goto out;
7387 0 : if (ret) {
7388 : ret = 0;
7389 : break;
7390 : }
7391 : }
7392 :
7393 0 : out_finish:
7394 0 : ret = finish_inode_if_needed(sctx, 1);
7395 :
7396 0 : out:
7397 0 : btrfs_free_path(path);
7398 0 : return ret;
7399 : }
7400 :
7401 0 : static int replace_node_with_clone(struct btrfs_path *path, int level)
7402 : {
7403 0 : struct extent_buffer *clone;
7404 :
7405 0 : clone = btrfs_clone_extent_buffer(path->nodes[level]);
7406 0 : if (!clone)
7407 : return -ENOMEM;
7408 :
7409 0 : free_extent_buffer(path->nodes[level]);
7410 0 : path->nodes[level] = clone;
7411 :
7412 0 : return 0;
7413 : }
7414 :
7415 0 : static int tree_move_down(struct btrfs_path *path, int *level, u64 reada_min_gen)
7416 : {
7417 0 : struct extent_buffer *eb;
7418 0 : struct extent_buffer *parent = path->nodes[*level];
7419 0 : int slot = path->slots[*level];
7420 0 : const int nritems = btrfs_header_nritems(parent);
7421 0 : u64 reada_max;
7422 0 : u64 reada_done = 0;
7423 :
7424 0 : lockdep_assert_held_read(&parent->fs_info->commit_root_sem);
7425 :
7426 0 : BUG_ON(*level == 0);
7427 0 : eb = btrfs_read_node_slot(parent, slot);
7428 0 : if (IS_ERR(eb))
7429 0 : return PTR_ERR(eb);
7430 :
7431 : /*
7432 : * Trigger readahead for the next leaves we will process, so that it is
7433 : * very likely that when we need them they are already in memory and we
7434 : * will not block on disk IO. For nodes we only do readahead for one,
7435 : * since the time window between processing nodes is typically larger.
7436 : */
7437 0 : reada_max = (*level == 1 ? SZ_128K : eb->fs_info->nodesize);
7438 :
7439 0 : for (slot++; slot < nritems && reada_done < reada_max; slot++) {
7440 0 : if (btrfs_node_ptr_generation(parent, slot) > reada_min_gen) {
7441 0 : btrfs_readahead_node_child(parent, slot);
7442 0 : reada_done += eb->fs_info->nodesize;
7443 : }
7444 : }
7445 :
7446 0 : path->nodes[*level - 1] = eb;
7447 0 : path->slots[*level - 1] = 0;
7448 0 : (*level)--;
7449 :
7450 0 : if (*level == 0)
7451 0 : return replace_node_with_clone(path, 0);
7452 :
7453 : return 0;
7454 : }
7455 :
7456 0 : static int tree_move_next_or_upnext(struct btrfs_path *path,
7457 : int *level, int root_level)
7458 : {
7459 0 : int ret = 0;
7460 0 : int nritems;
7461 0 : nritems = btrfs_header_nritems(path->nodes[*level]);
7462 :
7463 0 : path->slots[*level]++;
7464 :
7465 0 : while (path->slots[*level] >= nritems) {
7466 0 : if (*level == root_level) {
7467 0 : path->slots[*level] = nritems - 1;
7468 0 : return -1;
7469 : }
7470 :
7471 : /* move upnext */
7472 0 : path->slots[*level] = 0;
7473 0 : free_extent_buffer(path->nodes[*level]);
7474 0 : path->nodes[*level] = NULL;
7475 0 : (*level)++;
7476 0 : path->slots[*level]++;
7477 :
7478 0 : nritems = btrfs_header_nritems(path->nodes[*level]);
7479 0 : ret = 1;
7480 : }
7481 : return ret;
7482 : }
7483 :
7484 : /*
7485 : * Returns 1 if it had to move up and next. 0 is returned if it moved only next
7486 : * or down.
7487 : */
7488 0 : static int tree_advance(struct btrfs_path *path,
7489 : int *level, int root_level,
7490 : int allow_down,
7491 : struct btrfs_key *key,
7492 : u64 reada_min_gen)
7493 : {
7494 0 : int ret;
7495 :
7496 0 : if (*level == 0 || !allow_down) {
7497 0 : ret = tree_move_next_or_upnext(path, level, root_level);
7498 : } else {
7499 0 : ret = tree_move_down(path, level, reada_min_gen);
7500 : }
7501 :
7502 : /*
7503 : * Even if we have reached the end of a tree, ret is -1, update the key
7504 : * anyway, so that in case we need to restart due to a block group
7505 : * relocation, we can assert that the last key of the root node still
7506 : * exists in the tree.
7507 : */
7508 0 : if (*level == 0)
7509 0 : btrfs_item_key_to_cpu(path->nodes[*level], key,
7510 : path->slots[*level]);
7511 : else
7512 0 : btrfs_node_key_to_cpu(path->nodes[*level], key,
7513 : path->slots[*level]);
7514 :
7515 0 : return ret;
7516 : }
7517 :
7518 0 : static int tree_compare_item(struct btrfs_path *left_path,
7519 : struct btrfs_path *right_path,
7520 : char *tmp_buf)
7521 : {
7522 0 : int cmp;
7523 0 : int len1, len2;
7524 0 : unsigned long off1, off2;
7525 :
7526 0 : len1 = btrfs_item_size(left_path->nodes[0], left_path->slots[0]);
7527 0 : len2 = btrfs_item_size(right_path->nodes[0], right_path->slots[0]);
7528 0 : if (len1 != len2)
7529 : return 1;
7530 :
7531 0 : off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
7532 0 : off2 = btrfs_item_ptr_offset(right_path->nodes[0],
7533 : right_path->slots[0]);
7534 :
7535 0 : read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
7536 :
7537 0 : cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
7538 0 : if (cmp)
7539 0 : return 1;
7540 : return 0;
7541 : }
7542 :
7543 : /*
7544 : * A transaction used for relocating a block group was committed or is about to
7545 : * finish its commit. Release our paths and restart the search, so that we are
7546 : * not using stale extent buffers:
7547 : *
7548 : * 1) For levels > 0, we are only holding references of extent buffers, without
7549 : * any locks on them, which does not prevent them from having been relocated
7550 : * and reallocated after the last time we released the commit root semaphore.
7551 : * The exception are the root nodes, for which we always have a clone, see
7552 : * the comment at btrfs_compare_trees();
7553 : *
7554 : * 2) For leaves, level 0, we are holding copies (clones) of extent buffers, so
7555 : * we are safe from the concurrent relocation and reallocation. However they
7556 : * can have file extent items with a pre relocation disk_bytenr value, so we
7557 : * restart the start from the current commit roots and clone the new leaves so
7558 : * that we get the post relocation disk_bytenr values. Not doing so, could
7559 : * make us clone the wrong data in case there are new extents using the old
7560 : * disk_bytenr that happen to be shared.
7561 : */
7562 0 : static int restart_after_relocation(struct btrfs_path *left_path,
7563 : struct btrfs_path *right_path,
7564 : const struct btrfs_key *left_key,
7565 : const struct btrfs_key *right_key,
7566 : int left_level,
7567 : int right_level,
7568 : const struct send_ctx *sctx)
7569 : {
7570 0 : int root_level;
7571 0 : int ret;
7572 :
7573 0 : lockdep_assert_held_read(&sctx->send_root->fs_info->commit_root_sem);
7574 :
7575 0 : btrfs_release_path(left_path);
7576 0 : btrfs_release_path(right_path);
7577 :
7578 : /*
7579 : * Since keys can not be added or removed to/from our roots because they
7580 : * are readonly and we do not allow deduplication to run in parallel
7581 : * (which can add, remove or change keys), the layout of the trees should
7582 : * not change.
7583 : */
7584 0 : left_path->lowest_level = left_level;
7585 0 : ret = search_key_again(sctx, sctx->send_root, left_path, left_key);
7586 0 : if (ret < 0)
7587 : return ret;
7588 :
7589 0 : right_path->lowest_level = right_level;
7590 0 : ret = search_key_again(sctx, sctx->parent_root, right_path, right_key);
7591 0 : if (ret < 0)
7592 : return ret;
7593 :
7594 : /*
7595 : * If the lowest level nodes are leaves, clone them so that they can be
7596 : * safely used by changed_cb() while not under the protection of the
7597 : * commit root semaphore, even if relocation and reallocation happens in
7598 : * parallel.
7599 : */
7600 0 : if (left_level == 0) {
7601 0 : ret = replace_node_with_clone(left_path, 0);
7602 0 : if (ret < 0)
7603 : return ret;
7604 : }
7605 :
7606 0 : if (right_level == 0) {
7607 0 : ret = replace_node_with_clone(right_path, 0);
7608 0 : if (ret < 0)
7609 : return ret;
7610 : }
7611 :
7612 : /*
7613 : * Now clone the root nodes (unless they happen to be the leaves we have
7614 : * already cloned). This is to protect against concurrent snapshotting of
7615 : * the send and parent roots (see the comment at btrfs_compare_trees()).
7616 : */
7617 0 : root_level = btrfs_header_level(sctx->send_root->commit_root);
7618 0 : if (root_level > 0) {
7619 0 : ret = replace_node_with_clone(left_path, root_level);
7620 0 : if (ret < 0)
7621 : return ret;
7622 : }
7623 :
7624 0 : root_level = btrfs_header_level(sctx->parent_root->commit_root);
7625 0 : if (root_level > 0) {
7626 0 : ret = replace_node_with_clone(right_path, root_level);
7627 0 : if (ret < 0)
7628 : return ret;
7629 : }
7630 :
7631 : return 0;
7632 : }
7633 :
7634 : /*
7635 : * This function compares two trees and calls the provided callback for
7636 : * every changed/new/deleted item it finds.
7637 : * If shared tree blocks are encountered, whole subtrees are skipped, making
7638 : * the compare pretty fast on snapshotted subvolumes.
7639 : *
7640 : * This currently works on commit roots only. As commit roots are read only,
7641 : * we don't do any locking. The commit roots are protected with transactions.
7642 : * Transactions are ended and rejoined when a commit is tried in between.
7643 : *
7644 : * This function checks for modifications done to the trees while comparing.
7645 : * If it detects a change, it aborts immediately.
7646 : */
7647 0 : static int btrfs_compare_trees(struct btrfs_root *left_root,
7648 : struct btrfs_root *right_root, struct send_ctx *sctx)
7649 : {
7650 0 : struct btrfs_fs_info *fs_info = left_root->fs_info;
7651 0 : int ret;
7652 0 : int cmp;
7653 0 : struct btrfs_path *left_path = NULL;
7654 0 : struct btrfs_path *right_path = NULL;
7655 0 : struct btrfs_key left_key;
7656 0 : struct btrfs_key right_key;
7657 0 : char *tmp_buf = NULL;
7658 0 : int left_root_level;
7659 0 : int right_root_level;
7660 0 : int left_level;
7661 0 : int right_level;
7662 0 : int left_end_reached = 0;
7663 0 : int right_end_reached = 0;
7664 0 : int advance_left = 0;
7665 0 : int advance_right = 0;
7666 0 : u64 left_blockptr;
7667 0 : u64 right_blockptr;
7668 0 : u64 left_gen;
7669 0 : u64 right_gen;
7670 0 : u64 reada_min_gen;
7671 :
7672 0 : left_path = btrfs_alloc_path();
7673 0 : if (!left_path) {
7674 0 : ret = -ENOMEM;
7675 0 : goto out;
7676 : }
7677 0 : right_path = btrfs_alloc_path();
7678 0 : if (!right_path) {
7679 0 : ret = -ENOMEM;
7680 0 : goto out;
7681 : }
7682 :
7683 0 : tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
7684 0 : if (!tmp_buf) {
7685 0 : ret = -ENOMEM;
7686 0 : goto out;
7687 : }
7688 :
7689 0 : left_path->search_commit_root = 1;
7690 0 : left_path->skip_locking = 1;
7691 0 : right_path->search_commit_root = 1;
7692 0 : right_path->skip_locking = 1;
7693 :
7694 : /*
7695 : * Strategy: Go to the first items of both trees. Then do
7696 : *
7697 : * If both trees are at level 0
7698 : * Compare keys of current items
7699 : * If left < right treat left item as new, advance left tree
7700 : * and repeat
7701 : * If left > right treat right item as deleted, advance right tree
7702 : * and repeat
7703 : * If left == right do deep compare of items, treat as changed if
7704 : * needed, advance both trees and repeat
7705 : * If both trees are at the same level but not at level 0
7706 : * Compare keys of current nodes/leafs
7707 : * If left < right advance left tree and repeat
7708 : * If left > right advance right tree and repeat
7709 : * If left == right compare blockptrs of the next nodes/leafs
7710 : * If they match advance both trees but stay at the same level
7711 : * and repeat
7712 : * If they don't match advance both trees while allowing to go
7713 : * deeper and repeat
7714 : * If tree levels are different
7715 : * Advance the tree that needs it and repeat
7716 : *
7717 : * Advancing a tree means:
7718 : * If we are at level 0, try to go to the next slot. If that's not
7719 : * possible, go one level up and repeat. Stop when we found a level
7720 : * where we could go to the next slot. We may at this point be on a
7721 : * node or a leaf.
7722 : *
7723 : * If we are not at level 0 and not on shared tree blocks, go one
7724 : * level deeper.
7725 : *
7726 : * If we are not at level 0 and on shared tree blocks, go one slot to
7727 : * the right if possible or go up and right.
7728 : */
7729 :
7730 0 : down_read(&fs_info->commit_root_sem);
7731 0 : left_level = btrfs_header_level(left_root->commit_root);
7732 0 : left_root_level = left_level;
7733 : /*
7734 : * We clone the root node of the send and parent roots to prevent races
7735 : * with snapshot creation of these roots. Snapshot creation COWs the
7736 : * root node of a tree, so after the transaction is committed the old
7737 : * extent can be reallocated while this send operation is still ongoing.
7738 : * So we clone them, under the commit root semaphore, to be race free.
7739 : */
7740 0 : left_path->nodes[left_level] =
7741 0 : btrfs_clone_extent_buffer(left_root->commit_root);
7742 0 : if (!left_path->nodes[left_level]) {
7743 0 : ret = -ENOMEM;
7744 0 : goto out_unlock;
7745 : }
7746 :
7747 0 : right_level = btrfs_header_level(right_root->commit_root);
7748 0 : right_root_level = right_level;
7749 0 : right_path->nodes[right_level] =
7750 0 : btrfs_clone_extent_buffer(right_root->commit_root);
7751 0 : if (!right_path->nodes[right_level]) {
7752 0 : ret = -ENOMEM;
7753 0 : goto out_unlock;
7754 : }
7755 : /*
7756 : * Our right root is the parent root, while the left root is the "send"
7757 : * root. We know that all new nodes/leaves in the left root must have
7758 : * a generation greater than the right root's generation, so we trigger
7759 : * readahead for those nodes and leaves of the left root, as we know we
7760 : * will need to read them at some point.
7761 : */
7762 0 : reada_min_gen = btrfs_header_generation(right_root->commit_root);
7763 :
7764 0 : if (left_level == 0)
7765 0 : btrfs_item_key_to_cpu(left_path->nodes[left_level],
7766 : &left_key, left_path->slots[left_level]);
7767 : else
7768 0 : btrfs_node_key_to_cpu(left_path->nodes[left_level],
7769 : &left_key, left_path->slots[left_level]);
7770 0 : if (right_level == 0)
7771 0 : btrfs_item_key_to_cpu(right_path->nodes[right_level],
7772 : &right_key, right_path->slots[right_level]);
7773 : else
7774 0 : btrfs_node_key_to_cpu(right_path->nodes[right_level],
7775 : &right_key, right_path->slots[right_level]);
7776 :
7777 0 : sctx->last_reloc_trans = fs_info->last_reloc_trans;
7778 :
7779 0 : while (1) {
7780 0 : if (need_resched() ||
7781 : rwsem_is_contended(&fs_info->commit_root_sem)) {
7782 0 : up_read(&fs_info->commit_root_sem);
7783 0 : cond_resched();
7784 0 : down_read(&fs_info->commit_root_sem);
7785 : }
7786 :
7787 0 : if (fs_info->last_reloc_trans > sctx->last_reloc_trans) {
7788 0 : ret = restart_after_relocation(left_path, right_path,
7789 : &left_key, &right_key,
7790 : left_level, right_level,
7791 : sctx);
7792 0 : if (ret < 0)
7793 0 : goto out_unlock;
7794 0 : sctx->last_reloc_trans = fs_info->last_reloc_trans;
7795 : }
7796 :
7797 0 : if (advance_left && !left_end_reached) {
7798 0 : ret = tree_advance(left_path, &left_level,
7799 : left_root_level,
7800 : advance_left != ADVANCE_ONLY_NEXT,
7801 : &left_key, reada_min_gen);
7802 0 : if (ret == -1)
7803 : left_end_reached = ADVANCE;
7804 0 : else if (ret < 0)
7805 0 : goto out_unlock;
7806 : advance_left = 0;
7807 : }
7808 0 : if (advance_right && !right_end_reached) {
7809 0 : ret = tree_advance(right_path, &right_level,
7810 : right_root_level,
7811 : advance_right != ADVANCE_ONLY_NEXT,
7812 : &right_key, reada_min_gen);
7813 0 : if (ret == -1)
7814 : right_end_reached = ADVANCE;
7815 0 : else if (ret < 0)
7816 0 : goto out_unlock;
7817 : advance_right = 0;
7818 : }
7819 :
7820 0 : if (left_end_reached && right_end_reached) {
7821 0 : ret = 0;
7822 0 : goto out_unlock;
7823 0 : } else if (left_end_reached) {
7824 0 : if (right_level == 0) {
7825 0 : up_read(&fs_info->commit_root_sem);
7826 0 : ret = changed_cb(left_path, right_path,
7827 : &right_key,
7828 : BTRFS_COMPARE_TREE_DELETED,
7829 : sctx);
7830 0 : if (ret < 0)
7831 0 : goto out;
7832 0 : down_read(&fs_info->commit_root_sem);
7833 : }
7834 0 : advance_right = ADVANCE;
7835 0 : continue;
7836 0 : } else if (right_end_reached) {
7837 0 : if (left_level == 0) {
7838 0 : up_read(&fs_info->commit_root_sem);
7839 0 : ret = changed_cb(left_path, right_path,
7840 : &left_key,
7841 : BTRFS_COMPARE_TREE_NEW,
7842 : sctx);
7843 0 : if (ret < 0)
7844 0 : goto out;
7845 0 : down_read(&fs_info->commit_root_sem);
7846 : }
7847 0 : advance_left = ADVANCE;
7848 0 : continue;
7849 : }
7850 :
7851 0 : if (left_level == 0 && right_level == 0) {
7852 0 : up_read(&fs_info->commit_root_sem);
7853 0 : cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
7854 0 : if (cmp < 0) {
7855 0 : ret = changed_cb(left_path, right_path,
7856 : &left_key,
7857 : BTRFS_COMPARE_TREE_NEW,
7858 : sctx);
7859 0 : advance_left = ADVANCE;
7860 0 : } else if (cmp > 0) {
7861 0 : ret = changed_cb(left_path, right_path,
7862 : &right_key,
7863 : BTRFS_COMPARE_TREE_DELETED,
7864 : sctx);
7865 0 : advance_right = ADVANCE;
7866 : } else {
7867 0 : enum btrfs_compare_tree_result result;
7868 :
7869 0 : WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
7870 0 : ret = tree_compare_item(left_path, right_path,
7871 : tmp_buf);
7872 0 : if (ret)
7873 : result = BTRFS_COMPARE_TREE_CHANGED;
7874 : else
7875 0 : result = BTRFS_COMPARE_TREE_SAME;
7876 0 : ret = changed_cb(left_path, right_path,
7877 : &left_key, result, sctx);
7878 0 : advance_left = ADVANCE;
7879 0 : advance_right = ADVANCE;
7880 : }
7881 :
7882 0 : if (ret < 0)
7883 0 : goto out;
7884 0 : down_read(&fs_info->commit_root_sem);
7885 0 : } else if (left_level == right_level) {
7886 0 : cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
7887 0 : if (cmp < 0) {
7888 : advance_left = ADVANCE;
7889 0 : } else if (cmp > 0) {
7890 : advance_right = ADVANCE;
7891 : } else {
7892 0 : left_blockptr = btrfs_node_blockptr(
7893 0 : left_path->nodes[left_level],
7894 : left_path->slots[left_level]);
7895 0 : right_blockptr = btrfs_node_blockptr(
7896 0 : right_path->nodes[right_level],
7897 : right_path->slots[right_level]);
7898 0 : left_gen = btrfs_node_ptr_generation(
7899 0 : left_path->nodes[left_level],
7900 : left_path->slots[left_level]);
7901 0 : right_gen = btrfs_node_ptr_generation(
7902 0 : right_path->nodes[right_level],
7903 : right_path->slots[right_level]);
7904 0 : if (left_blockptr == right_blockptr &&
7905 0 : left_gen == right_gen) {
7906 : /*
7907 : * As we're on a shared block, don't
7908 : * allow to go deeper.
7909 : */
7910 : advance_left = ADVANCE_ONLY_NEXT;
7911 : advance_right = ADVANCE_ONLY_NEXT;
7912 : } else {
7913 0 : advance_left = ADVANCE;
7914 0 : advance_right = ADVANCE;
7915 : }
7916 : }
7917 0 : } else if (left_level < right_level) {
7918 : advance_right = ADVANCE;
7919 : } else {
7920 0 : advance_left = ADVANCE;
7921 : }
7922 : }
7923 :
7924 0 : out_unlock:
7925 0 : up_read(&fs_info->commit_root_sem);
7926 0 : out:
7927 0 : btrfs_free_path(left_path);
7928 0 : btrfs_free_path(right_path);
7929 0 : kvfree(tmp_buf);
7930 0 : return ret;
7931 : }
7932 :
7933 0 : static int send_subvol(struct send_ctx *sctx)
7934 : {
7935 0 : int ret;
7936 :
7937 0 : if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
7938 0 : ret = send_header(sctx);
7939 0 : if (ret < 0)
7940 0 : goto out;
7941 : }
7942 :
7943 0 : ret = send_subvol_begin(sctx);
7944 0 : if (ret < 0)
7945 0 : goto out;
7946 :
7947 0 : if (sctx->parent_root) {
7948 0 : ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, sctx);
7949 0 : if (ret < 0)
7950 0 : goto out;
7951 0 : ret = finish_inode_if_needed(sctx, 1);
7952 0 : if (ret < 0)
7953 0 : goto out;
7954 : } else {
7955 0 : ret = full_send_tree(sctx);
7956 0 : if (ret < 0)
7957 0 : goto out;
7958 : }
7959 :
7960 0 : out:
7961 0 : free_recorded_refs(sctx);
7962 0 : return ret;
7963 : }
7964 :
7965 : /*
7966 : * If orphan cleanup did remove any orphans from a root, it means the tree
7967 : * was modified and therefore the commit root is not the same as the current
7968 : * root anymore. This is a problem, because send uses the commit root and
7969 : * therefore can see inode items that don't exist in the current root anymore,
7970 : * and for example make calls to btrfs_iget, which will do tree lookups based
7971 : * on the current root and not on the commit root. Those lookups will fail,
7972 : * returning a -ESTALE error, and making send fail with that error. So make
7973 : * sure a send does not see any orphans we have just removed, and that it will
7974 : * see the same inodes regardless of whether a transaction commit happened
7975 : * before it started (meaning that the commit root will be the same as the
7976 : * current root) or not.
7977 : */
7978 0 : static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
7979 : {
7980 0 : int i;
7981 0 : struct btrfs_trans_handle *trans = NULL;
7982 :
7983 0 : again:
7984 0 : if (sctx->parent_root &&
7985 0 : sctx->parent_root->node != sctx->parent_root->commit_root)
7986 0 : goto commit_trans;
7987 :
7988 0 : for (i = 0; i < sctx->clone_roots_cnt; i++)
7989 0 : if (sctx->clone_roots[i].root->node !=
7990 0 : sctx->clone_roots[i].root->commit_root)
7991 0 : goto commit_trans;
7992 :
7993 0 : if (trans)
7994 0 : return btrfs_end_transaction(trans);
7995 :
7996 : return 0;
7997 :
7998 0 : commit_trans:
7999 : /* Use any root, all fs roots will get their commit roots updated. */
8000 0 : if (!trans) {
8001 0 : trans = btrfs_join_transaction(sctx->send_root);
8002 0 : if (IS_ERR(trans))
8003 0 : return PTR_ERR(trans);
8004 0 : goto again;
8005 : }
8006 :
8007 0 : return btrfs_commit_transaction(trans);
8008 : }
8009 :
8010 : /*
8011 : * Make sure any existing dellaloc is flushed for any root used by a send
8012 : * operation so that we do not miss any data and we do not race with writeback
8013 : * finishing and changing a tree while send is using the tree. This could
8014 : * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and
8015 : * a send operation then uses the subvolume.
8016 : * After flushing delalloc ensure_commit_roots_uptodate() must be called.
8017 : */
8018 0 : static int flush_delalloc_roots(struct send_ctx *sctx)
8019 : {
8020 0 : struct btrfs_root *root = sctx->parent_root;
8021 0 : int ret;
8022 0 : int i;
8023 :
8024 0 : if (root) {
8025 0 : ret = btrfs_start_delalloc_snapshot(root, false);
8026 0 : if (ret)
8027 : return ret;
8028 0 : btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
8029 : }
8030 :
8031 0 : for (i = 0; i < sctx->clone_roots_cnt; i++) {
8032 0 : root = sctx->clone_roots[i].root;
8033 0 : ret = btrfs_start_delalloc_snapshot(root, false);
8034 0 : if (ret)
8035 0 : return ret;
8036 0 : btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
8037 : }
8038 :
8039 : return 0;
8040 : }
8041 :
8042 0 : static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
8043 : {
8044 0 : spin_lock(&root->root_item_lock);
8045 0 : root->send_in_progress--;
8046 : /*
8047 : * Not much left to do, we don't know why it's unbalanced and
8048 : * can't blindly reset it to 0.
8049 : */
8050 0 : if (root->send_in_progress < 0)
8051 0 : btrfs_err(root->fs_info,
8052 : "send_in_progress unbalanced %d root %llu",
8053 : root->send_in_progress, root->root_key.objectid);
8054 0 : spin_unlock(&root->root_item_lock);
8055 0 : }
8056 :
8057 0 : static void dedupe_in_progress_warn(const struct btrfs_root *root)
8058 : {
8059 0 : btrfs_warn_rl(root->fs_info,
8060 : "cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
8061 : root->root_key.objectid, root->dedupe_in_progress);
8062 0 : }
8063 :
8064 0 : long btrfs_ioctl_send(struct inode *inode, struct btrfs_ioctl_send_args *arg)
8065 : {
8066 0 : int ret = 0;
8067 0 : struct btrfs_root *send_root = BTRFS_I(inode)->root;
8068 0 : struct btrfs_fs_info *fs_info = send_root->fs_info;
8069 0 : struct btrfs_root *clone_root;
8070 0 : struct send_ctx *sctx = NULL;
8071 0 : u32 i;
8072 0 : u64 *clone_sources_tmp = NULL;
8073 0 : int clone_sources_to_rollback = 0;
8074 0 : size_t alloc_size;
8075 0 : int sort_clone_roots = 0;
8076 0 : struct btrfs_lru_cache_entry *entry;
8077 0 : struct btrfs_lru_cache_entry *tmp;
8078 :
8079 0 : if (!capable(CAP_SYS_ADMIN))
8080 : return -EPERM;
8081 :
8082 : /*
8083 : * The subvolume must remain read-only during send, protect against
8084 : * making it RW. This also protects against deletion.
8085 : */
8086 0 : spin_lock(&send_root->root_item_lock);
8087 0 : if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
8088 0 : dedupe_in_progress_warn(send_root);
8089 0 : spin_unlock(&send_root->root_item_lock);
8090 0 : return -EAGAIN;
8091 : }
8092 0 : send_root->send_in_progress++;
8093 0 : spin_unlock(&send_root->root_item_lock);
8094 :
8095 : /*
8096 : * Userspace tools do the checks and warn the user if it's
8097 : * not RO.
8098 : */
8099 0 : if (!btrfs_root_readonly(send_root)) {
8100 0 : ret = -EPERM;
8101 0 : goto out;
8102 : }
8103 :
8104 : /*
8105 : * Check that we don't overflow at later allocations, we request
8106 : * clone_sources_count + 1 items, and compare to unsigned long inside
8107 : * access_ok. Also set an upper limit for allocation size so this can't
8108 : * easily exhaust memory. Max number of clone sources is about 200K.
8109 : */
8110 0 : if (arg->clone_sources_count > SZ_8M / sizeof(struct clone_root)) {
8111 0 : ret = -EINVAL;
8112 0 : goto out;
8113 : }
8114 :
8115 0 : if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
8116 0 : ret = -EINVAL;
8117 0 : goto out;
8118 : }
8119 :
8120 0 : sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
8121 0 : if (!sctx) {
8122 0 : ret = -ENOMEM;
8123 0 : goto out;
8124 : }
8125 :
8126 0 : INIT_LIST_HEAD(&sctx->new_refs);
8127 0 : INIT_LIST_HEAD(&sctx->deleted_refs);
8128 :
8129 0 : btrfs_lru_cache_init(&sctx->name_cache, SEND_MAX_NAME_CACHE_SIZE);
8130 0 : btrfs_lru_cache_init(&sctx->backref_cache, SEND_MAX_BACKREF_CACHE_SIZE);
8131 0 : btrfs_lru_cache_init(&sctx->dir_created_cache,
8132 : SEND_MAX_DIR_CREATED_CACHE_SIZE);
8133 : /*
8134 : * This cache is periodically trimmed to a fixed size elsewhere, see
8135 : * cache_dir_utimes() and trim_dir_utimes_cache().
8136 : */
8137 0 : btrfs_lru_cache_init(&sctx->dir_utimes_cache, 0);
8138 :
8139 0 : sctx->pending_dir_moves = RB_ROOT;
8140 0 : sctx->waiting_dir_moves = RB_ROOT;
8141 0 : sctx->orphan_dirs = RB_ROOT;
8142 0 : sctx->rbtree_new_refs = RB_ROOT;
8143 0 : sctx->rbtree_deleted_refs = RB_ROOT;
8144 :
8145 0 : sctx->flags = arg->flags;
8146 :
8147 0 : if (arg->flags & BTRFS_SEND_FLAG_VERSION) {
8148 0 : if (arg->version > BTRFS_SEND_STREAM_VERSION) {
8149 0 : ret = -EPROTO;
8150 0 : goto out;
8151 : }
8152 : /* Zero means "use the highest version" */
8153 0 : sctx->proto = arg->version ?: BTRFS_SEND_STREAM_VERSION;
8154 : } else {
8155 0 : sctx->proto = 1;
8156 : }
8157 0 : if ((arg->flags & BTRFS_SEND_FLAG_COMPRESSED) && sctx->proto < 2) {
8158 0 : ret = -EINVAL;
8159 0 : goto out;
8160 : }
8161 :
8162 0 : sctx->send_filp = fget(arg->send_fd);
8163 0 : if (!sctx->send_filp) {
8164 0 : ret = -EBADF;
8165 0 : goto out;
8166 : }
8167 :
8168 0 : sctx->send_root = send_root;
8169 : /*
8170 : * Unlikely but possible, if the subvolume is marked for deletion but
8171 : * is slow to remove the directory entry, send can still be started
8172 : */
8173 0 : if (btrfs_root_dead(sctx->send_root)) {
8174 0 : ret = -EPERM;
8175 0 : goto out;
8176 : }
8177 :
8178 0 : sctx->clone_roots_cnt = arg->clone_sources_count;
8179 :
8180 0 : if (sctx->proto >= 2) {
8181 0 : u32 send_buf_num_pages;
8182 :
8183 0 : sctx->send_max_size = BTRFS_SEND_BUF_SIZE_V2;
8184 0 : sctx->send_buf = vmalloc(sctx->send_max_size);
8185 0 : if (!sctx->send_buf) {
8186 0 : ret = -ENOMEM;
8187 0 : goto out;
8188 : }
8189 0 : send_buf_num_pages = sctx->send_max_size >> PAGE_SHIFT;
8190 0 : sctx->send_buf_pages = kcalloc(send_buf_num_pages,
8191 : sizeof(*sctx->send_buf_pages),
8192 : GFP_KERNEL);
8193 0 : if (!sctx->send_buf_pages) {
8194 0 : ret = -ENOMEM;
8195 0 : goto out;
8196 : }
8197 0 : for (i = 0; i < send_buf_num_pages; i++) {
8198 0 : sctx->send_buf_pages[i] =
8199 0 : vmalloc_to_page(sctx->send_buf + (i << PAGE_SHIFT));
8200 : }
8201 : } else {
8202 0 : sctx->send_max_size = BTRFS_SEND_BUF_SIZE_V1;
8203 0 : sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
8204 : }
8205 0 : if (!sctx->send_buf) {
8206 0 : ret = -ENOMEM;
8207 0 : goto out;
8208 : }
8209 :
8210 0 : sctx->clone_roots = kvcalloc(sizeof(*sctx->clone_roots),
8211 0 : arg->clone_sources_count + 1,
8212 : GFP_KERNEL);
8213 0 : if (!sctx->clone_roots) {
8214 0 : ret = -ENOMEM;
8215 0 : goto out;
8216 : }
8217 :
8218 0 : alloc_size = array_size(sizeof(*arg->clone_sources),
8219 : arg->clone_sources_count);
8220 :
8221 0 : if (arg->clone_sources_count) {
8222 0 : clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
8223 0 : if (!clone_sources_tmp) {
8224 0 : ret = -ENOMEM;
8225 0 : goto out;
8226 : }
8227 :
8228 0 : ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
8229 : alloc_size);
8230 0 : if (ret) {
8231 0 : ret = -EFAULT;
8232 0 : goto out;
8233 : }
8234 :
8235 0 : for (i = 0; i < arg->clone_sources_count; i++) {
8236 0 : clone_root = btrfs_get_fs_root(fs_info,
8237 0 : clone_sources_tmp[i], true);
8238 0 : if (IS_ERR(clone_root)) {
8239 0 : ret = PTR_ERR(clone_root);
8240 0 : goto out;
8241 : }
8242 0 : spin_lock(&clone_root->root_item_lock);
8243 0 : if (!btrfs_root_readonly(clone_root) ||
8244 : btrfs_root_dead(clone_root)) {
8245 0 : spin_unlock(&clone_root->root_item_lock);
8246 0 : btrfs_put_root(clone_root);
8247 0 : ret = -EPERM;
8248 0 : goto out;
8249 : }
8250 0 : if (clone_root->dedupe_in_progress) {
8251 0 : dedupe_in_progress_warn(clone_root);
8252 0 : spin_unlock(&clone_root->root_item_lock);
8253 0 : btrfs_put_root(clone_root);
8254 0 : ret = -EAGAIN;
8255 0 : goto out;
8256 : }
8257 0 : clone_root->send_in_progress++;
8258 0 : spin_unlock(&clone_root->root_item_lock);
8259 :
8260 0 : sctx->clone_roots[i].root = clone_root;
8261 0 : clone_sources_to_rollback = i + 1;
8262 : }
8263 0 : kvfree(clone_sources_tmp);
8264 0 : clone_sources_tmp = NULL;
8265 : }
8266 :
8267 0 : if (arg->parent_root) {
8268 0 : sctx->parent_root = btrfs_get_fs_root(fs_info, arg->parent_root,
8269 : true);
8270 0 : if (IS_ERR(sctx->parent_root)) {
8271 0 : ret = PTR_ERR(sctx->parent_root);
8272 0 : goto out;
8273 : }
8274 :
8275 0 : spin_lock(&sctx->parent_root->root_item_lock);
8276 0 : sctx->parent_root->send_in_progress++;
8277 0 : if (!btrfs_root_readonly(sctx->parent_root) ||
8278 : btrfs_root_dead(sctx->parent_root)) {
8279 0 : spin_unlock(&sctx->parent_root->root_item_lock);
8280 0 : ret = -EPERM;
8281 0 : goto out;
8282 : }
8283 0 : if (sctx->parent_root->dedupe_in_progress) {
8284 0 : dedupe_in_progress_warn(sctx->parent_root);
8285 0 : spin_unlock(&sctx->parent_root->root_item_lock);
8286 0 : ret = -EAGAIN;
8287 0 : goto out;
8288 : }
8289 0 : spin_unlock(&sctx->parent_root->root_item_lock);
8290 : }
8291 :
8292 : /*
8293 : * Clones from send_root are allowed, but only if the clone source
8294 : * is behind the current send position. This is checked while searching
8295 : * for possible clone sources.
8296 : */
8297 0 : sctx->clone_roots[sctx->clone_roots_cnt++].root =
8298 0 : btrfs_grab_root(sctx->send_root);
8299 :
8300 : /* We do a bsearch later */
8301 0 : sort(sctx->clone_roots, sctx->clone_roots_cnt,
8302 : sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
8303 : NULL);
8304 0 : sort_clone_roots = 1;
8305 :
8306 0 : ret = flush_delalloc_roots(sctx);
8307 0 : if (ret)
8308 0 : goto out;
8309 :
8310 0 : ret = ensure_commit_roots_uptodate(sctx);
8311 0 : if (ret)
8312 0 : goto out;
8313 :
8314 0 : ret = send_subvol(sctx);
8315 0 : if (ret < 0)
8316 0 : goto out;
8317 :
8318 0 : btrfs_lru_cache_for_each_entry_safe(&sctx->dir_utimes_cache, entry, tmp) {
8319 0 : ret = send_utimes(sctx, entry->key, entry->gen);
8320 0 : if (ret < 0)
8321 0 : goto out;
8322 0 : btrfs_lru_cache_remove(&sctx->dir_utimes_cache, entry);
8323 : }
8324 :
8325 0 : if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
8326 0 : ret = begin_cmd(sctx, BTRFS_SEND_C_END);
8327 0 : if (ret < 0)
8328 0 : goto out;
8329 0 : ret = send_cmd(sctx);
8330 0 : if (ret < 0)
8331 0 : goto out;
8332 : }
8333 :
8334 0 : out:
8335 0 : WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
8336 0 : while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
8337 0 : struct rb_node *n;
8338 0 : struct pending_dir_move *pm;
8339 :
8340 0 : n = rb_first(&sctx->pending_dir_moves);
8341 0 : pm = rb_entry(n, struct pending_dir_move, node);
8342 0 : while (!list_empty(&pm->list)) {
8343 0 : struct pending_dir_move *pm2;
8344 :
8345 0 : pm2 = list_first_entry(&pm->list,
8346 : struct pending_dir_move, list);
8347 0 : free_pending_move(sctx, pm2);
8348 : }
8349 0 : free_pending_move(sctx, pm);
8350 : }
8351 :
8352 0 : WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
8353 0 : while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
8354 0 : struct rb_node *n;
8355 0 : struct waiting_dir_move *dm;
8356 :
8357 0 : n = rb_first(&sctx->waiting_dir_moves);
8358 0 : dm = rb_entry(n, struct waiting_dir_move, node);
8359 0 : rb_erase(&dm->node, &sctx->waiting_dir_moves);
8360 0 : kfree(dm);
8361 : }
8362 :
8363 0 : WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
8364 0 : while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
8365 0 : struct rb_node *n;
8366 0 : struct orphan_dir_info *odi;
8367 :
8368 0 : n = rb_first(&sctx->orphan_dirs);
8369 0 : odi = rb_entry(n, struct orphan_dir_info, node);
8370 0 : free_orphan_dir_info(sctx, odi);
8371 : }
8372 :
8373 0 : if (sort_clone_roots) {
8374 0 : for (i = 0; i < sctx->clone_roots_cnt; i++) {
8375 0 : btrfs_root_dec_send_in_progress(
8376 0 : sctx->clone_roots[i].root);
8377 0 : btrfs_put_root(sctx->clone_roots[i].root);
8378 : }
8379 : } else {
8380 0 : for (i = 0; sctx && i < clone_sources_to_rollback; i++) {
8381 0 : btrfs_root_dec_send_in_progress(
8382 0 : sctx->clone_roots[i].root);
8383 0 : btrfs_put_root(sctx->clone_roots[i].root);
8384 : }
8385 :
8386 0 : btrfs_root_dec_send_in_progress(send_root);
8387 : }
8388 0 : if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) {
8389 0 : btrfs_root_dec_send_in_progress(sctx->parent_root);
8390 0 : btrfs_put_root(sctx->parent_root);
8391 : }
8392 :
8393 0 : kvfree(clone_sources_tmp);
8394 :
8395 0 : if (sctx) {
8396 0 : if (sctx->send_filp)
8397 0 : fput(sctx->send_filp);
8398 :
8399 0 : kvfree(sctx->clone_roots);
8400 0 : kfree(sctx->send_buf_pages);
8401 0 : kvfree(sctx->send_buf);
8402 0 : kvfree(sctx->verity_descriptor);
8403 :
8404 0 : close_current_inode(sctx);
8405 :
8406 0 : btrfs_lru_cache_clear(&sctx->name_cache);
8407 0 : btrfs_lru_cache_clear(&sctx->backref_cache);
8408 0 : btrfs_lru_cache_clear(&sctx->dir_created_cache);
8409 0 : btrfs_lru_cache_clear(&sctx->dir_utimes_cache);
8410 :
8411 0 : kfree(sctx);
8412 : }
8413 :
8414 0 : return ret;
8415 : }
|
