projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
[linux-2.6-block.git] / fs / btrfs / send.c
CommitLineData
c1d7c514 1// SPDX-License-Identifier: GPL-2.0
31db9f7c
AB		 2/*
3 * Copyright (C) 2012 Alexander Block. All rights reserved.
31db9f7c
AB		 4 */
5
6#include <linux/bsearch.h>
7#include <linux/fs.h>
8#include <linux/file.h>
9#include <linux/sort.h>
10#include <linux/mount.h>
11#include <linux/xattr.h>
12#include <linux/posix_acl_xattr.h>
13#include <linux/radix-tree.h>
a1857ebe 14#include <linux/vmalloc.h>
ed84885d 15#include <linux/string.h>
2351f431 16#include <linux/compat.h>
9678c543 17#include <linux/crc32c.h>
31db9f7c
AB		 18
19#include "send.h"
20#include "backref.h"
21#include "locking.h"
22#include "disk-io.h"
23#include "btrfs_inode.h"
24#include "transaction.h"
ebb8765b 25#include "compression.h"
31db9f7c 26
fd0ddbe2
FM		 27/*
28 * Maximum number of references an extent can have in order for us to attempt to
29 * issue clone operations instead of write operations. This currently exists to
30 * avoid hitting limitations of the backreference walking code (taking a lot of
31 * time and using too much memory for extents with large number of references).
32 */
33#define SEND_MAX_EXTENT_REFS 64
34
31db9f7c
AB		 35/*
36 * A fs_path is a helper to dynamically build path names with unknown size.
37 * It reallocates the internal buffer on demand.
38 * It allows fast adding of path elements on the right side (normal path) and
39 * fast adding to the left side (reversed path). A reversed path can also be
40 * unreversed if needed.
41 */
42struct fs_path {
43 union {
44 struct {
45 char *start;
46 char *end;
31db9f7c
AB		 47
48 char *buf;
1f5a7ff9
DS		 49 unsigned short buf_len:15;
50 unsigned short reversed:1;
31db9f7c
AB		 51 char inline_buf[];
52 };
ace01050
DS		 53 /*
54 * Average path length does not exceed 200 bytes, we'll have
55 * better packing in the slab and higher chance to satisfy
56 * a allocation later during send.
57 */
58 char pad[256];
31db9f7c
AB		 59 };
60};
61#define FS_PATH_INLINE_SIZE \
62 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
63
64
65/* reused for each extent */
66struct clone_root {
67 struct btrfs_root *root;
68 u64 ino;
69 u64 offset;
70
71 u64 found_refs;
72};
73
74#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
75#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
76
77struct send_ctx {
78 struct file *send_filp;
79 loff_t send_off;
80 char *send_buf;
81 u32 send_size;
82 u32 send_max_size;
83 u64 total_send_size;
84 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
cb95e7bf 85 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
31db9f7c 86
31db9f7c
AB		 87 struct btrfs_root *send_root;
88 struct btrfs_root *parent_root;
89 struct clone_root *clone_roots;
90 int clone_roots_cnt;
91
92 /* current state of the compare_tree call */
93 struct btrfs_path *left_path;
94 struct btrfs_path *right_path;
95 struct btrfs_key *cmp_key;
96
97 /*
98 * infos of the currently processed inode. In case of deleted inodes,
99 * these are the values from the deleted inode.
100 */
101 u64 cur_ino;
102 u64 cur_inode_gen;
103 int cur_inode_new;
104 int cur_inode_new_gen;
105 int cur_inode_deleted;
31db9f7c
AB		 106 u64 cur_inode_size;
107 u64 cur_inode_mode;
644d1940 108 u64 cur_inode_rdev;
16e7549f 109 u64 cur_inode_last_extent;
ffa7c429 110 u64 cur_inode_next_write_offset;
46b2f459 111 bool ignore_cur_inode;
31db9f7c
AB		 112
113 u64 send_progress;
114
115 struct list_head new_refs;
116 struct list_head deleted_refs;
117
118 struct radix_tree_root name_cache;
119 struct list_head name_cache_list;
120 int name_cache_size;
121
2131bcd3
LB		 122 struct file_ra_state ra;
123
31db9f7c 124 char *read_buf;
9f03740a
FDBM		 125
126 /*
127 * We process inodes by their increasing order, so if before an
128 * incremental send we reverse the parent/child relationship of
129 * directories such that a directory with a lower inode number was
130 * the parent of a directory with a higher inode number, and the one
131 * becoming the new parent got renamed too, we can't rename/move the
132 * directory with lower inode number when we finish processing it - we
133 * must process the directory with higher inode number first, then
134 * rename/move it and then rename/move the directory with lower inode
135 * number. Example follows.
136 *
137 * Tree state when the first send was performed:
138 *
139 * .
140 * |-- a (ino 257)
141 * |-- b (ino 258)
142 * |
143 * |
144 * |-- c (ino 259)
145 * | |-- d (ino 260)
146 * |
147 * |-- c2 (ino 261)
148 *
149 * Tree state when the second (incremental) send is performed:
150 *
151 * .
152 * |-- a (ino 257)
153 * |-- b (ino 258)
154 * |-- c2 (ino 261)
155 * |-- d2 (ino 260)
156 * |-- cc (ino 259)
157 *
158 * The sequence of steps that lead to the second state was:
159 *
160 * mv /a/b/c/d /a/b/c2/d2
161 * mv /a/b/c /a/b/c2/d2/cc
162 *
163 * "c" has lower inode number, but we can't move it (2nd mv operation)
164 * before we move "d", which has higher inode number.
165 *
166 * So we just memorize which move/rename operations must be performed
167 * later when their respective parent is processed and moved/renamed.
168 */
169
170 /* Indexed by parent directory inode number. */
171 struct rb_root pending_dir_moves;
172
173 /*
174 * Reverse index, indexed by the inode number of a directory that
175 * is waiting for the move/rename of its immediate parent before its
176 * own move/rename can be performed.
177 */
178 struct rb_root waiting_dir_moves;
9dc44214
FM		 179
180 /*
181 * A directory that is going to be rm'ed might have a child directory
182 * which is in the pending directory moves index above. In this case,
183 * the directory can only be removed after the move/rename of its child
184 * is performed. Example:
185 *
186 * Parent snapshot:
187 *
188 * . (ino 256)
189 * |-- a/ (ino 257)
190 * |-- b/ (ino 258)
191 * |-- c/ (ino 259)
192 * | |-- x/ (ino 260)
193 * |
194 * |-- y/ (ino 261)
195 *
196 * Send snapshot:
197 *
198 * . (ino 256)
199 * |-- a/ (ino 257)
200 * |-- b/ (ino 258)
201 * |-- YY/ (ino 261)
202 * |-- x/ (ino 260)
203 *
204 * Sequence of steps that lead to the send snapshot:
205 * rm -f /a/b/c/foo.txt
206 * mv /a/b/y /a/b/YY
207 * mv /a/b/c/x /a/b/YY
208 * rmdir /a/b/c
209 *
210 * When the child is processed, its move/rename is delayed until its
211 * parent is processed (as explained above), but all other operations
212 * like update utimes, chown, chgrp, etc, are performed and the paths
213 * that it uses for those operations must use the orphanized name of
214 * its parent (the directory we're going to rm later), so we need to
215 * memorize that name.
216 *
217 * Indexed by the inode number of the directory to be deleted.
218 */
219 struct rb_root orphan_dirs;
9f03740a
FDBM		 220};
221
222struct pending_dir_move {
223 struct rb_node node;
224 struct list_head list;
225 u64 parent_ino;
226 u64 ino;
227 u64 gen;
228 struct list_head update_refs;
229};
230
231struct waiting_dir_move {
232 struct rb_node node;
233 u64 ino;
9dc44214
FM		 234 /*
235 * There might be some directory that could not be removed because it
236 * was waiting for this directory inode to be moved first. Therefore
237 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
238 */
239 u64 rmdir_ino;
8b191a68 240 bool orphanized;
9dc44214
FM		 241};
242
243struct orphan_dir_info {
244 struct rb_node node;
245 u64 ino;
246 u64 gen;
0f96f517 247 u64 last_dir_index_offset;
31db9f7c
AB		 248};
249
250struct name_cache_entry {
251 struct list_head list;
7e0926fe
AB		 252 /*
253 * radix_tree has only 32bit entries but we need to handle 64bit inums.
254 * We use the lower 32bit of the 64bit inum to store it in the tree. If
255 * more then one inum would fall into the same entry, we use radix_list
256 * to store the additional entries. radix_list is also used to store
257 * entries where two entries have the same inum but different
258 * generations.
259 */
260 struct list_head radix_list;
31db9f7c
AB		 261 u64 ino;
262 u64 gen;
263 u64 parent_ino;
264 u64 parent_gen;
265 int ret;
266 int need_later_update;
267 int name_len;
268 char name[];
269};
270
18d0f5c6
DS		 271#define ADVANCE 1
272#define ADVANCE_ONLY_NEXT -1
273
274enum btrfs_compare_tree_result {
275 BTRFS_COMPARE_TREE_NEW,
276 BTRFS_COMPARE_TREE_DELETED,
277 BTRFS_COMPARE_TREE_CHANGED,
278 BTRFS_COMPARE_TREE_SAME,
279};
280typedef int (*btrfs_changed_cb_t)(struct btrfs_path *left_path,
281 struct btrfs_path *right_path,
282 struct btrfs_key *key,
283 enum btrfs_compare_tree_result result,
284 void *ctx);
285
e67c718b 286__cold
95155585
FM		 287static void inconsistent_snapshot_error(struct send_ctx *sctx,
288 enum btrfs_compare_tree_result result,
289 const char *what)
290{
291 const char *result_string;
292
293 switch (result) {
294 case BTRFS_COMPARE_TREE_NEW:
295 result_string = "new";
296 break;
297 case BTRFS_COMPARE_TREE_DELETED:
298 result_string = "deleted";
299 break;
300 case BTRFS_COMPARE_TREE_CHANGED:
301 result_string = "updated";
302 break;
303 case BTRFS_COMPARE_TREE_SAME:
304 ASSERT(0);
305 result_string = "unchanged";
306 break;
307 default:
308 ASSERT(0);
309 result_string = "unexpected";
310 }
311
312 btrfs_err(sctx->send_root->fs_info,
313 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
314 result_string, what, sctx->cmp_key->objectid,
315 sctx->send_root->root_key.objectid,
316 (sctx->parent_root ?
317 sctx->parent_root->root_key.objectid : 0));
318}
319
9f03740a
FDBM		 320static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
321
9dc44214
FM		 322static struct waiting_dir_move *
323get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
324
325static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
326
16e7549f
JB		 327static int need_send_hole(struct send_ctx *sctx)
328{
329 return (sctx->parent_root && !sctx->cur_inode_new &&
330 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
331 S_ISREG(sctx->cur_inode_mode));
332}
333
31db9f7c
AB		 334static void fs_path_reset(struct fs_path *p)
335{
336 if (p->reversed) {
337 p->start = p->buf + p->buf_len - 1;
338 p->end = p->start;
339 *p->start = 0;
340 } else {
341 p->start = p->buf;
342 p->end = p->start;
343 *p->start = 0;
344 }
345}
346
924794c9 347static struct fs_path *fs_path_alloc(void)
31db9f7c
AB		 348{
349 struct fs_path *p;
350
e780b0d1 351 p = kmalloc(sizeof(*p), GFP_KERNEL);
31db9f7c
AB		 352 if (!p)
353 return NULL;
354 p->reversed = 0;
31db9f7c
AB		 355 p->buf = p->inline_buf;
356 p->buf_len = FS_PATH_INLINE_SIZE;
357 fs_path_reset(p);
358 return p;
359}
360
924794c9 361static struct fs_path *fs_path_alloc_reversed(void)
31db9f7c
AB		 362{
363 struct fs_path *p;
364
924794c9 365 p = fs_path_alloc();
31db9f7c
AB		 366 if (!p)
367 return NULL;
368 p->reversed = 1;
369 fs_path_reset(p);
370 return p;
371}
372
924794c9 373static void fs_path_free(struct fs_path *p)
31db9f7c
AB		 374{
375 if (!p)
376 return;
ace01050
DS		 377 if (p->buf != p->inline_buf)
378 kfree(p->buf);
31db9f7c
AB		 379 kfree(p);
380}
381
382static int fs_path_len(struct fs_path *p)
383{
384 return p->end - p->start;
385}
386
387static int fs_path_ensure_buf(struct fs_path *p, int len)
388{
389 char *tmp_buf;
390 int path_len;
391 int old_buf_len;
392
393 len++;
394
395 if (p->buf_len >= len)
396 return 0;
397
cfd4a535
CM		 398 if (len > PATH_MAX) {
399 WARN_ON(1);
400 return -ENOMEM;
401 }
402
1b2782c8
DS		 403 path_len = p->end - p->start;
404 old_buf_len = p->buf_len;
405
ace01050
DS		 406 /*
407 * First time the inline_buf does not suffice
408 */
01a9a8a9 409 if (p->buf == p->inline_buf) {
e780b0d1 410 tmp_buf = kmalloc(len, GFP_KERNEL);
01a9a8a9
FM		 411 if (tmp_buf)
412 memcpy(tmp_buf, p->buf, old_buf_len);
413 } else {
e780b0d1 414 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
01a9a8a9 415 }
9c9ca00b
DS		 416 if (!tmp_buf)
417 return -ENOMEM;
418 p->buf = tmp_buf;
419 /*
420 * The real size of the buffer is bigger, this will let the fast path
421 * happen most of the time
422 */
423 p->buf_len = ksize(p->buf);
ace01050 424
31db9f7c
AB		 425 if (p->reversed) {
426 tmp_buf = p->buf + old_buf_len - path_len - 1;
427 p->end = p->buf + p->buf_len - 1;
428 p->start = p->end - path_len;
429 memmove(p->start, tmp_buf, path_len + 1);
430 } else {
431 p->start = p->buf;
432 p->end = p->start + path_len;
433 }
434 return 0;
435}
436
b23ab57d
DS		 437static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
438 char **prepared)
31db9f7c
AB		 439{
440 int ret;
441 int new_len;
442
443 new_len = p->end - p->start + name_len;
444 if (p->start != p->end)
445 new_len++;
446 ret = fs_path_ensure_buf(p, new_len);
447 if (ret < 0)
448 goto out;
449
450 if (p->reversed) {
451 if (p->start != p->end)
452 *--p->start = '/';
453 p->start -= name_len;
b23ab57d 454 *prepared = p->start;
31db9f7c
AB		 455 } else {
456 if (p->start != p->end)
457 *p->end++ = '/';
b23ab57d 458 *prepared = p->end;
31db9f7c
AB		 459 p->end += name_len;
460 *p->end = 0;
461 }
462
463out:
464 return ret;
465}
466
467static int fs_path_add(struct fs_path *p, const char *name, int name_len)
468{
469 int ret;
b23ab57d 470 char *prepared;
31db9f7c 471
b23ab57d 472 ret = fs_path_prepare_for_add(p, name_len, &prepared);
31db9f7c
AB		 473 if (ret < 0)
474 goto out;
b23ab57d 475 memcpy(prepared, name, name_len);
31db9f7c
AB		 476
477out:
478 return ret;
479}
480
481static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
482{
483 int ret;
b23ab57d 484 char *prepared;
31db9f7c 485
b23ab57d 486 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
31db9f7c
AB		 487 if (ret < 0)
488 goto out;
b23ab57d 489 memcpy(prepared, p2->start, p2->end - p2->start);
31db9f7c
AB		 490
491out:
492 return ret;
493}
494
495static int fs_path_add_from_extent_buffer(struct fs_path *p,
496 struct extent_buffer *eb,
497 unsigned long off, int len)
498{
499 int ret;
b23ab57d 500 char *prepared;
31db9f7c 501
b23ab57d 502 ret = fs_path_prepare_for_add(p, len, &prepared);
31db9f7c
AB		 503 if (ret < 0)
504 goto out;
505
b23ab57d 506 read_extent_buffer(eb, prepared, off, len);
31db9f7c
AB		 507
508out:
509 return ret;
510}
511
31db9f7c
AB		 512static int fs_path_copy(struct fs_path *p, struct fs_path *from)
513{
514 int ret;
515
516 p->reversed = from->reversed;
517 fs_path_reset(p);
518
519 ret = fs_path_add_path(p, from);
520
521 return ret;
522}
523
524
525static void fs_path_unreverse(struct fs_path *p)
526{
527 char *tmp;
528 int len;
529
530 if (!p->reversed)
531 return;
532
533 tmp = p->start;
534 len = p->end - p->start;
535 p->start = p->buf;
536 p->end = p->start + len;
537 memmove(p->start, tmp, len + 1);
538 p->reversed = 0;
539}
540
541static struct btrfs_path *alloc_path_for_send(void)
542{
543 struct btrfs_path *path;
544
545 path = btrfs_alloc_path();
546 if (!path)
547 return NULL;
548 path->search_commit_root = 1;
549 path->skip_locking = 1;
3f8a18cc 550 path->need_commit_sem = 1;
31db9f7c
AB		 551 return path;
552}
553
48a3b636 554static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
31db9f7c
AB		 555{
556 int ret;
31db9f7c
AB		 557 u32 pos = 0;
558
31db9f7c 559 while (pos < len) {
8e93157b 560 ret = kernel_write(filp, buf + pos, len - pos, off);
31db9f7c
AB		 561 /* TODO handle that correctly */
562 /*if (ret == -ERESTARTSYS) {
563 continue;
564 }*/
565 if (ret < 0)
8e93157b 566 return ret;
31db9f7c 567 if (ret == 0) {
8e93157b 568 return -EIO;
31db9f7c
AB		 569 }
570 pos += ret;
571 }
572
8e93157b 573 return 0;
31db9f7c
AB		 574}
575
576static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
577{
578 struct btrfs_tlv_header *hdr;
579 int total_len = sizeof(*hdr) + len;
580 int left = sctx->send_max_size - sctx->send_size;
581
582 if (unlikely(left < total_len))
583 return -EOVERFLOW;
584
585 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
586 hdr->tlv_type = cpu_to_le16(attr);
587 hdr->tlv_len = cpu_to_le16(len);
588 memcpy(hdr + 1, data, len);
589 sctx->send_size += total_len;
590
591 return 0;
592}
593
95bc79d5
DS		 594#define TLV_PUT_DEFINE_INT(bits) \
595 static int tlv_put_u##bits(struct send_ctx *sctx, \
596 u##bits attr, u##bits value) \
597 { \
598 __le##bits __tmp = cpu_to_le##bits(value); \
599 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
600 }
31db9f7c 601
95bc79d5 602TLV_PUT_DEFINE_INT(64)
31db9f7c
AB		 603
604static int tlv_put_string(struct send_ctx *sctx, u16 attr,
605 const char *str, int len)
606{
607 if (len == -1)
608 len = strlen(str);
609 return tlv_put(sctx, attr, str, len);
610}
611
612static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
613 const u8 *uuid)
614{
615 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
616}
617
31db9f7c
AB		 618static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
619 struct extent_buffer *eb,
620 struct btrfs_timespec *ts)
621{
622 struct btrfs_timespec bts;
623 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
624 return tlv_put(sctx, attr, &bts, sizeof(bts));
625}
626
627
895a72be 628#define TLV_PUT(sctx, attrtype, data, attrlen) \
31db9f7c 629 do { \
895a72be 630 ret = tlv_put(sctx, attrtype, data, attrlen); \
31db9f7c
AB		 631 if (ret < 0) \
632 goto tlv_put_failure; \
633 } while (0)
634
635#define TLV_PUT_INT(sctx, attrtype, bits, value) \
636 do { \
637 ret = tlv_put_u##bits(sctx, attrtype, value); \
638 if (ret < 0) \
639 goto tlv_put_failure; \
640 } while (0)
641
642#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
643#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
644#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
645#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
646#define TLV_PUT_STRING(sctx, attrtype, str, len) \
647 do { \
648 ret = tlv_put_string(sctx, attrtype, str, len); \
649 if (ret < 0) \
650 goto tlv_put_failure; \
651 } while (0)
652#define TLV_PUT_PATH(sctx, attrtype, p) \
653 do { \
654 ret = tlv_put_string(sctx, attrtype, p->start, \
655 p->end - p->start); \
656 if (ret < 0) \
657 goto tlv_put_failure; \
658 } while(0)
659#define TLV_PUT_UUID(sctx, attrtype, uuid) \
660 do { \
661 ret = tlv_put_uuid(sctx, attrtype, uuid); \
662 if (ret < 0) \
663 goto tlv_put_failure; \
664 } while (0)
31db9f7c
AB		 665#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
666 do { \
667 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
668 if (ret < 0) \
669 goto tlv_put_failure; \
670 } while (0)
671
672static int send_header(struct send_ctx *sctx)
673{
674 struct btrfs_stream_header hdr;
675
676 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
677 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
678
1bcea355
AJ		 679 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
680 &sctx->send_off);
31db9f7c
AB		 681}
682
683/*
684 * For each command/item we want to send to userspace, we call this function.
685 */
686static int begin_cmd(struct send_ctx *sctx, int cmd)
687{
688 struct btrfs_cmd_header *hdr;
689
fae7f21c 690 if (WARN_ON(!sctx->send_buf))
31db9f7c 691 return -EINVAL;
31db9f7c
AB		 692
693 BUG_ON(sctx->send_size);
694
695 sctx->send_size += sizeof(*hdr);
696 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
697 hdr->cmd = cpu_to_le16(cmd);
698
699 return 0;
700}
701
702static int send_cmd(struct send_ctx *sctx)
703{
704 int ret;
705 struct btrfs_cmd_header *hdr;
706 u32 crc;
707
708 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
709 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
710 hdr->crc = 0;
711
65019df8 712 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
31db9f7c
AB		 713 hdr->crc = cpu_to_le32(crc);
714
1bcea355
AJ		 715 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
716 &sctx->send_off);
31db9f7c
AB		 717
718 sctx->total_send_size += sctx->send_size;
719 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
720 sctx->send_size = 0;
721
722 return ret;
723}
724
725/*
726 * Sends a move instruction to user space
727 */
728static int send_rename(struct send_ctx *sctx,
729 struct fs_path *from, struct fs_path *to)
730{
04ab956e 731 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 732 int ret;
733
04ab956e 734 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
31db9f7c
AB		 735
736 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
737 if (ret < 0)
738 goto out;
739
740 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
741 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
742
743 ret = send_cmd(sctx);
744
745tlv_put_failure:
746out:
747 return ret;
748}
749
750/*
751 * Sends a link instruction to user space
752 */
753static int send_link(struct send_ctx *sctx,
754 struct fs_path *path, struct fs_path *lnk)
755{
04ab956e 756 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 757 int ret;
758
04ab956e 759 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
31db9f7c
AB		 760
761 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
762 if (ret < 0)
763 goto out;
764
765 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
766 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
767
768 ret = send_cmd(sctx);
769
770tlv_put_failure:
771out:
772 return ret;
773}
774
775/*
776 * Sends an unlink instruction to user space
777 */
778static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
779{
04ab956e 780 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 781 int ret;
782
04ab956e 783 btrfs_debug(fs_info, "send_unlink %s", path->start);
31db9f7c
AB		 784
785 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
786 if (ret < 0)
787 goto out;
788
789 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
790
791 ret = send_cmd(sctx);
792
793tlv_put_failure:
794out:
795 return ret;
796}
797
798/*
799 * Sends a rmdir instruction to user space
800 */
801static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
802{
04ab956e 803 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 804 int ret;
805
04ab956e 806 btrfs_debug(fs_info, "send_rmdir %s", path->start);
31db9f7c
AB		 807
808 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
809 if (ret < 0)
810 goto out;
811
812 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
813
814 ret = send_cmd(sctx);
815
816tlv_put_failure:
817out:
818 return ret;
819}
820
821/*
822 * Helper function to retrieve some fields from an inode item.
823 */
3f8a18cc
JB		 824static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
825 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
826 u64 *gid, u64 *rdev)
31db9f7c
AB		 827{
828 int ret;
829 struct btrfs_inode_item *ii;
830 struct btrfs_key key;
31db9f7c
AB		 831
832 key.objectid = ino;
833 key.type = BTRFS_INODE_ITEM_KEY;
834 key.offset = 0;
835 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
31db9f7c 836 if (ret) {
3f8a18cc
JB		 837 if (ret > 0)
838 ret = -ENOENT;
839 return ret;
31db9f7c
AB		 840 }
841
842 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
843 struct btrfs_inode_item);
844 if (size)
845 *size = btrfs_inode_size(path->nodes[0], ii);
846 if (gen)
847 *gen = btrfs_inode_generation(path->nodes[0], ii);
848 if (mode)
849 *mode = btrfs_inode_mode(path->nodes[0], ii);
850 if (uid)
851 *uid = btrfs_inode_uid(path->nodes[0], ii);
852 if (gid)
853 *gid = btrfs_inode_gid(path->nodes[0], ii);
85a7b33b
AB		 854 if (rdev)
855 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
31db9f7c 856
3f8a18cc
JB		 857 return ret;
858}
859
860static int get_inode_info(struct btrfs_root *root,
861 u64 ino, u64 *size, u64 *gen,
862 u64 *mode, u64 *uid, u64 *gid,
863 u64 *rdev)
864{
865 struct btrfs_path *path;
866 int ret;
867
868 path = alloc_path_for_send();
869 if (!path)
870 return -ENOMEM;
871 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
872 rdev);
31db9f7c
AB		 873 btrfs_free_path(path);
874 return ret;
875}
876
877typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
878 struct fs_path *p,
879 void *ctx);
880
881/*
96b5bd77
JS		 882 * Helper function to iterate the entries in ONE btrfs_inode_ref or
883 * btrfs_inode_extref.
31db9f7c
AB		 884 * The iterate callback may return a non zero value to stop iteration. This can
885 * be a negative value for error codes or 1 to simply stop it.
886 *
96b5bd77 887 * path must point to the INODE_REF or INODE_EXTREF when called.
31db9f7c 888 */
924794c9 889static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 890 struct btrfs_key *found_key, int resolve,
891 iterate_inode_ref_t iterate, void *ctx)
892{
96b5bd77 893 struct extent_buffer *eb = path->nodes[0];
31db9f7c
AB		 894 struct btrfs_item *item;
895 struct btrfs_inode_ref *iref;
96b5bd77 896 struct btrfs_inode_extref *extref;
31db9f7c
AB		 897 struct btrfs_path *tmp_path;
898 struct fs_path *p;
96b5bd77 899 u32 cur = 0;
31db9f7c 900 u32 total;
96b5bd77 901 int slot = path->slots[0];
31db9f7c
AB		 902 u32 name_len;
903 char *start;
904 int ret = 0;
96b5bd77 905 int num = 0;
31db9f7c 906 int index;
96b5bd77
JS		 907 u64 dir;
908 unsigned long name_off;
909 unsigned long elem_size;
910 unsigned long ptr;
31db9f7c 911
924794c9 912 p = fs_path_alloc_reversed();
31db9f7c
AB		 913 if (!p)
914 return -ENOMEM;
915
916 tmp_path = alloc_path_for_send();
917 if (!tmp_path) {
924794c9 918 fs_path_free(p);
31db9f7c
AB		 919 return -ENOMEM;
920 }
921
31db9f7c 922
96b5bd77
JS		 923 if (found_key->type == BTRFS_INODE_REF_KEY) {
924 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
925 struct btrfs_inode_ref);
dd3cc16b 926 item = btrfs_item_nr(slot);
96b5bd77
JS		 927 total = btrfs_item_size(eb, item);
928 elem_size = sizeof(*iref);
929 } else {
930 ptr = btrfs_item_ptr_offset(eb, slot);
931 total = btrfs_item_size_nr(eb, slot);
932 elem_size = sizeof(*extref);
933 }
934
31db9f7c
AB		 935 while (cur < total) {
936 fs_path_reset(p);
937
96b5bd77
JS		 938 if (found_key->type == BTRFS_INODE_REF_KEY) {
939 iref = (struct btrfs_inode_ref *)(ptr + cur);
940 name_len = btrfs_inode_ref_name_len(eb, iref);
941 name_off = (unsigned long)(iref + 1);
942 index = btrfs_inode_ref_index(eb, iref);
943 dir = found_key->offset;
944 } else {
945 extref = (struct btrfs_inode_extref *)(ptr + cur);
946 name_len = btrfs_inode_extref_name_len(eb, extref);
947 name_off = (unsigned long)&extref->name;
948 index = btrfs_inode_extref_index(eb, extref);
949 dir = btrfs_inode_extref_parent(eb, extref);
950 }
951
31db9f7c 952 if (resolve) {
96b5bd77
JS		 953 start = btrfs_ref_to_path(root, tmp_path, name_len,
954 name_off, eb, dir,
955 p->buf, p->buf_len);
31db9f7c
AB		 956 if (IS_ERR(start)) {
957 ret = PTR_ERR(start);
958 goto out;
959 }
960 if (start < p->buf) {
961 /* overflow , try again with larger buffer */
962 ret = fs_path_ensure_buf(p,
963 p->buf_len + p->buf - start);
964 if (ret < 0)
965 goto out;
96b5bd77
JS		 966 start = btrfs_ref_to_path(root, tmp_path,
967 name_len, name_off,
968 eb, dir,
969 p->buf, p->buf_len);
31db9f7c
AB		 970 if (IS_ERR(start)) {
971 ret = PTR_ERR(start);
972 goto out;
973 }
974 BUG_ON(start < p->buf);
975 }
976 p->start = start;
977 } else {
96b5bd77
JS		 978 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
979 name_len);
31db9f7c
AB		 980 if (ret < 0)
981 goto out;
982 }
983
96b5bd77
JS		 984 cur += elem_size + name_len;
985 ret = iterate(num, dir, index, p, ctx);
31db9f7c
AB		 986 if (ret)
987 goto out;
31db9f7c
AB		 988 num++;
989 }
990
991out:
992 btrfs_free_path(tmp_path);
924794c9 993 fs_path_free(p);
31db9f7c
AB		 994 return ret;
995}
996
997typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
998 const char *name, int name_len,
999 const char *data, int data_len,
1000 u8 type, void *ctx);
1001
1002/*
1003 * Helper function to iterate the entries in ONE btrfs_dir_item.
1004 * The iterate callback may return a non zero value to stop iteration. This can
1005 * be a negative value for error codes or 1 to simply stop it.
1006 *
1007 * path must point to the dir item when called.
1008 */
924794c9 1009static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 1010 iterate_dir_item_t iterate, void *ctx)
1011{
1012 int ret = 0;
1013 struct extent_buffer *eb;
1014 struct btrfs_item *item;
1015 struct btrfs_dir_item *di;
31db9f7c
AB		 1016 struct btrfs_key di_key;
1017 char *buf = NULL;
7e3ae33e 1018 int buf_len;
31db9f7c
AB		 1019 u32 name_len;
1020 u32 data_len;
1021 u32 cur;
1022 u32 len;
1023 u32 total;
1024 int slot;
1025 int num;
1026 u8 type;
1027
4395e0c4
FM		 1028 /*
1029 * Start with a small buffer (1 page). If later we end up needing more
1030 * space, which can happen for xattrs on a fs with a leaf size greater
1031 * then the page size, attempt to increase the buffer. Typically xattr
1032 * values are small.
1033 */
1034 buf_len = PATH_MAX;
e780b0d1 1035 buf = kmalloc(buf_len, GFP_KERNEL);
31db9f7c
AB		 1036 if (!buf) {
1037 ret = -ENOMEM;
1038 goto out;
1039 }
1040
31db9f7c
AB		 1041 eb = path->nodes[0];
1042 slot = path->slots[0];
dd3cc16b 1043 item = btrfs_item_nr(slot);
31db9f7c
AB		 1044 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1045 cur = 0;
1046 len = 0;
1047 total = btrfs_item_size(eb, item);
1048
1049 num = 0;
1050 while (cur < total) {
1051 name_len = btrfs_dir_name_len(eb, di);
1052 data_len = btrfs_dir_data_len(eb, di);
1053 type = btrfs_dir_type(eb, di);
1054 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1055
7e3ae33e
FM		 1056 if (type == BTRFS_FT_XATTR) {
1057 if (name_len > XATTR_NAME_MAX) {
1058 ret = -ENAMETOOLONG;
1059 goto out;
1060 }
da17066c
JM		 1061 if (name_len + data_len >
1062 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
7e3ae33e
FM		 1063 ret = -E2BIG;
1064 goto out;
1065 }
1066 } else {
1067 /*
1068 * Path too long
1069 */
4395e0c4 1070 if (name_len + data_len > PATH_MAX) {
7e3ae33e
FM		 1071 ret = -ENAMETOOLONG;
1072 goto out;
1073 }
31db9f7c
AB		 1074 }
1075
4395e0c4
FM		 1076 if (name_len + data_len > buf_len) {
1077 buf_len = name_len + data_len;
1078 if (is_vmalloc_addr(buf)) {
1079 vfree(buf);
1080 buf = NULL;
1081 } else {
1082 char *tmp = krealloc(buf, buf_len,
e780b0d1 1083 GFP_KERNEL | __GFP_NOWARN);
4395e0c4
FM		 1084
1085 if (!tmp)
1086 kfree(buf);
1087 buf = tmp;
1088 }
1089 if (!buf) {
f11f7441 1090 buf = kvmalloc(buf_len, GFP_KERNEL);
4395e0c4
FM		 1091 if (!buf) {
1092 ret = -ENOMEM;
1093 goto out;
1094 }
1095 }
1096 }
1097
31db9f7c
AB		 1098 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1099 name_len + data_len);
1100
1101 len = sizeof(*di) + name_len + data_len;
1102 di = (struct btrfs_dir_item *)((char *)di + len);
1103 cur += len;
1104
1105 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1106 data_len, type, ctx);
1107 if (ret < 0)
1108 goto out;
1109 if (ret) {
1110 ret = 0;
1111 goto out;
1112 }
1113
1114 num++;
1115 }
1116
1117out:
4395e0c4 1118 kvfree(buf);
31db9f7c
AB		 1119 return ret;
1120}
1121
1122static int __copy_first_ref(int num, u64 dir, int index,
1123 struct fs_path *p, void *ctx)
1124{
1125 int ret;
1126 struct fs_path *pt = ctx;
1127
1128 ret = fs_path_copy(pt, p);
1129 if (ret < 0)
1130 return ret;
1131
1132 /* we want the first only */
1133 return 1;
1134}
1135
1136/*
1137 * Retrieve the first path of an inode. If an inode has more then one
1138 * ref/hardlink, this is ignored.
1139 */
924794c9 1140static int get_inode_path(struct btrfs_root *root,
31db9f7c
AB		 1141 u64 ino, struct fs_path *path)
1142{
1143 int ret;
1144 struct btrfs_key key, found_key;
1145 struct btrfs_path *p;
1146
1147 p = alloc_path_for_send();
1148 if (!p)
1149 return -ENOMEM;
1150
1151 fs_path_reset(path);
1152
1153 key.objectid = ino;
1154 key.type = BTRFS_INODE_REF_KEY;
1155 key.offset = 0;
1156
1157 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1158 if (ret < 0)
1159 goto out;
1160 if (ret) {
1161 ret = 1;
1162 goto out;
1163 }
1164 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1165 if (found_key.objectid != ino ||
96b5bd77
JS		 1166 (found_key.type != BTRFS_INODE_REF_KEY &&
1167 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1168 ret = -ENOENT;
1169 goto out;
1170 }
1171
924794c9
TI		 1172 ret = iterate_inode_ref(root, p, &found_key, 1,
1173 __copy_first_ref, path);
31db9f7c
AB		 1174 if (ret < 0)
1175 goto out;
1176 ret = 0;
1177
1178out:
1179 btrfs_free_path(p);
1180 return ret;
1181}
1182
1183struct backref_ctx {
1184 struct send_ctx *sctx;
1185
1186 /* number of total found references */
1187 u64 found;
1188
1189 /*
1190 * used for clones found in send_root. clones found behind cur_objectid
1191 * and cur_offset are not considered as allowed clones.
1192 */
1193 u64 cur_objectid;
1194 u64 cur_offset;
1195
1196 /* may be truncated in case it's the last extent in a file */
1197 u64 extent_len;
1198
619d8c4e
FM		 1199 /* data offset in the file extent item */
1200 u64 data_offset;
1201
31db9f7c 1202 /* Just to check for bugs in backref resolving */
ee849c04 1203 int found_itself;
31db9f7c
AB		 1204};
1205
1206static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1207{
995e01b7 1208 u64 root = (u64)(uintptr_t)key;
31db9f7c
AB		 1209 struct clone_root *cr = (struct clone_root *)elt;
1210
4fd786e6 1211 if (root < cr->root->root_key.objectid)
31db9f7c 1212 return -1;
4fd786e6 1213 if (root > cr->root->root_key.objectid)
31db9f7c
AB		 1214 return 1;
1215 return 0;
1216}
1217
1218static int __clone_root_cmp_sort(const void *e1, const void *e2)
1219{
1220 struct clone_root *cr1 = (struct clone_root *)e1;
1221 struct clone_root *cr2 = (struct clone_root *)e2;
1222
4fd786e6 1223 if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
31db9f7c 1224 return -1;
4fd786e6 1225 if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
31db9f7c
AB		 1226 return 1;
1227 return 0;
1228}
1229
1230/*
1231 * Called for every backref that is found for the current extent.
766702ef 1232 * Results are collected in sctx->clone_roots->ino/offset/found_refs
31db9f7c
AB		 1233 */
1234static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1235{
1236 struct backref_ctx *bctx = ctx_;
1237 struct clone_root *found;
31db9f7c
AB		 1238
1239 /* First check if the root is in the list of accepted clone sources */
995e01b7 1240 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
31db9f7c
AB		 1241 bctx->sctx->clone_roots_cnt,
1242 sizeof(struct clone_root),
1243 __clone_root_cmp_bsearch);
1244 if (!found)
1245 return 0;
1246
1247 if (found->root == bctx->sctx->send_root &&
1248 ino == bctx->cur_objectid &&
1249 offset == bctx->cur_offset) {
ee849c04 1250 bctx->found_itself = 1;
31db9f7c
AB		 1251 }
1252
31db9f7c
AB		 1253 /*
1254 * Make sure we don't consider clones from send_root that are
1255 * behind the current inode/offset.
1256 */
1257 if (found->root == bctx->sctx->send_root) {
1258 /*
11f2069c
FM		 1259 * If the source inode was not yet processed we can't issue a
1260 * clone operation, as the source extent does not exist yet at
1261 * the destination of the stream.
31db9f7c 1262 */
11f2069c
FM		 1263 if (ino > bctx->cur_objectid)
1264 return 0;
1265 /*
1266 * We clone from the inode currently being sent as long as the
1267 * source extent is already processed, otherwise we could try
1268 * to clone from an extent that does not exist yet at the
1269 * destination of the stream.
1270 */
1271 if (ino == bctx->cur_objectid &&
9722b101
FM		 1272 offset + bctx->extent_len >
1273 bctx->sctx->cur_inode_next_write_offset)
31db9f7c 1274 return 0;
31db9f7c
AB		 1275 }
1276
1277 bctx->found++;
1278 found->found_refs++;
1279 if (ino < found->ino) {
1280 found->ino = ino;
1281 found->offset = offset;
1282 } else if (found->ino == ino) {
1283 /*
1284 * same extent found more then once in the same file.
1285 */
1286 if (found->offset > offset + bctx->extent_len)
1287 found->offset = offset;
1288 }
1289
1290 return 0;
1291}
1292
1293/*
766702ef
AB		 1294 * Given an inode, offset and extent item, it finds a good clone for a clone
1295 * instruction. Returns -ENOENT when none could be found. The function makes
1296 * sure that the returned clone is usable at the point where sending is at the
1297 * moment. This means, that no clones are accepted which lie behind the current
1298 * inode+offset.
1299 *
31db9f7c
AB		 1300 * path must point to the extent item when called.
1301 */
1302static int find_extent_clone(struct send_ctx *sctx,
1303 struct btrfs_path *path,
1304 u64 ino, u64 data_offset,
1305 u64 ino_size,
1306 struct clone_root **found)
1307{
04ab956e 1308 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 1309 int ret;
1310 int extent_type;
1311 u64 logical;
74dd17fb 1312 u64 disk_byte;
31db9f7c
AB		 1313 u64 num_bytes;
1314 u64 extent_item_pos;
69917e43 1315 u64 flags = 0;
31db9f7c
AB		 1316 struct btrfs_file_extent_item *fi;
1317 struct extent_buffer *eb = path->nodes[0];
35075bb0 1318 struct backref_ctx *backref_ctx = NULL;
31db9f7c
AB		 1319 struct clone_root *cur_clone_root;
1320 struct btrfs_key found_key;
1321 struct btrfs_path *tmp_path;
fd0ddbe2 1322 struct btrfs_extent_item *ei;
74dd17fb 1323 int compressed;
31db9f7c
AB		 1324 u32 i;
1325
1326 tmp_path = alloc_path_for_send();
1327 if (!tmp_path)
1328 return -ENOMEM;
1329
3f8a18cc
JB		 1330 /* We only use this path under the commit sem */
1331 tmp_path->need_commit_sem = 0;
1332
e780b0d1 1333 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
35075bb0
AB		 1334 if (!backref_ctx) {
1335 ret = -ENOMEM;
1336 goto out;
1337 }
1338
31db9f7c
AB		 1339 if (data_offset >= ino_size) {
1340 /*
1341 * There may be extents that lie behind the file's size.
1342 * I at least had this in combination with snapshotting while
1343 * writing large files.
1344 */
1345 ret = 0;
1346 goto out;
1347 }
1348
1349 fi = btrfs_item_ptr(eb, path->slots[0],
1350 struct btrfs_file_extent_item);
1351 extent_type = btrfs_file_extent_type(eb, fi);
1352 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1353 ret = -ENOENT;
1354 goto out;
1355 }
74dd17fb 1356 compressed = btrfs_file_extent_compression(eb, fi);
31db9f7c
AB		 1357
1358 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
74dd17fb
CM		 1359 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1360 if (disk_byte == 0) {
31db9f7c
AB		 1361 ret = -ENOENT;
1362 goto out;
1363 }
74dd17fb 1364 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
31db9f7c 1365
04ab956e
JM		 1366 down_read(&fs_info->commit_root_sem);
1367 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
69917e43 1368 &found_key, &flags);
04ab956e 1369 up_read(&fs_info->commit_root_sem);
31db9f7c
AB		 1370
1371 if (ret < 0)
1372 goto out;
69917e43 1373 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
31db9f7c
AB		 1374 ret = -EIO;
1375 goto out;
1376 }
1377
fd0ddbe2
FM		 1378 ei = btrfs_item_ptr(tmp_path->nodes[0], tmp_path->slots[0],
1379 struct btrfs_extent_item);
1380 /*
1381 * Backreference walking (iterate_extent_inodes() below) is currently
1382 * too expensive when an extent has a large number of references, both
1383 * in time spent and used memory. So for now just fallback to write
1384 * operations instead of clone operations when an extent has more than
1385 * a certain amount of references.
1386 */
1387 if (btrfs_extent_refs(tmp_path->nodes[0], ei) > SEND_MAX_EXTENT_REFS) {
1388 ret = -ENOENT;
1389 goto out;
1390 }
1391 btrfs_release_path(tmp_path);
1392
31db9f7c
AB		 1393 /*
1394 * Setup the clone roots.
1395 */
1396 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1397 cur_clone_root = sctx->clone_roots + i;
1398 cur_clone_root->ino = (u64)-1;
1399 cur_clone_root->offset = 0;
1400 cur_clone_root->found_refs = 0;
1401 }
1402
35075bb0
AB		 1403 backref_ctx->sctx = sctx;
1404 backref_ctx->found = 0;
1405 backref_ctx->cur_objectid = ino;
1406 backref_ctx->cur_offset = data_offset;
1407 backref_ctx->found_itself = 0;
1408 backref_ctx->extent_len = num_bytes;
619d8c4e
FM		 1409 /*
1410 * For non-compressed extents iterate_extent_inodes() gives us extent
1411 * offsets that already take into account the data offset, but not for
1412 * compressed extents, since the offset is logical and not relative to
1413 * the physical extent locations. We must take this into account to
1414 * avoid sending clone offsets that go beyond the source file's size,
1415 * which would result in the clone ioctl failing with -EINVAL on the
1416 * receiving end.
1417 */
1418 if (compressed == BTRFS_COMPRESS_NONE)
1419 backref_ctx->data_offset = 0;
1420 else
1421 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
31db9f7c
AB		 1422
1423 /*
1424 * The last extent of a file may be too large due to page alignment.
1425 * We need to adjust extent_len in this case so that the checks in
1426 * __iterate_backrefs work.
1427 */
1428 if (data_offset + num_bytes >= ino_size)
35075bb0 1429 backref_ctx->extent_len = ino_size - data_offset;
31db9f7c
AB		 1430
1431 /*
1432 * Now collect all backrefs.
1433 */
74dd17fb
CM		 1434 if (compressed == BTRFS_COMPRESS_NONE)
1435 extent_item_pos = logical - found_key.objectid;
1436 else
1437 extent_item_pos = 0;
0b246afa
JM		 1438 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1439 extent_item_pos, 1, __iterate_backrefs,
c995ab3c 1440 backref_ctx, false);
74dd17fb 1441
31db9f7c
AB		 1442 if (ret < 0)
1443 goto out;
1444
35075bb0 1445 if (!backref_ctx->found_itself) {
31db9f7c
AB		 1446 /* found a bug in backref code? */
1447 ret = -EIO;
04ab956e 1448 btrfs_err(fs_info,
5d163e0e 1449 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
04ab956e 1450 ino, data_offset, disk_byte, found_key.objectid);
31db9f7c
AB		 1451 goto out;
1452 }
1453
04ab956e
JM		 1454 btrfs_debug(fs_info,
1455 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1456 data_offset, ino, num_bytes, logical);
31db9f7c 1457
35075bb0 1458 if (!backref_ctx->found)
04ab956e 1459 btrfs_debug(fs_info, "no clones found");
31db9f7c
AB		 1460
1461 cur_clone_root = NULL;
1462 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1463 if (sctx->clone_roots[i].found_refs) {
1464 if (!cur_clone_root)
1465 cur_clone_root = sctx->clone_roots + i;
1466 else if (sctx->clone_roots[i].root == sctx->send_root)
1467 /* prefer clones from send_root over others */
1468 cur_clone_root = sctx->clone_roots + i;
31db9f7c
AB		 1469 }
1470
1471 }
1472
1473 if (cur_clone_root) {
1474 *found = cur_clone_root;
1475 ret = 0;
1476 } else {
1477 ret = -ENOENT;
1478 }
1479
1480out:
1481 btrfs_free_path(tmp_path);
35075bb0 1482 kfree(backref_ctx);
31db9f7c
AB		 1483 return ret;
1484}
1485
924794c9 1486static int read_symlink(struct btrfs_root *root,
31db9f7c
AB		 1487 u64 ino,
1488 struct fs_path *dest)
1489{
1490 int ret;
1491 struct btrfs_path *path;
1492 struct btrfs_key key;
1493 struct btrfs_file_extent_item *ei;
1494 u8 type;
1495 u8 compression;
1496 unsigned long off;
1497 int len;
1498
1499 path = alloc_path_for_send();
1500 if (!path)
1501 return -ENOMEM;
1502
1503 key.objectid = ino;
1504 key.type = BTRFS_EXTENT_DATA_KEY;
1505 key.offset = 0;
1506 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1507 if (ret < 0)
1508 goto out;
a879719b
FM		 1509 if (ret) {
1510 /*
1511 * An empty symlink inode. Can happen in rare error paths when
1512 * creating a symlink (transaction committed before the inode
1513 * eviction handler removed the symlink inode items and a crash
1514 * happened in between or the subvol was snapshoted in between).
1515 * Print an informative message to dmesg/syslog so that the user
1516 * can delete the symlink.
1517 */
1518 btrfs_err(root->fs_info,
1519 "Found empty symlink inode %llu at root %llu",
1520 ino, root->root_key.objectid);
1521 ret = -EIO;
1522 goto out;
1523 }
31db9f7c
AB		 1524
1525 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1526 struct btrfs_file_extent_item);
1527 type = btrfs_file_extent_type(path->nodes[0], ei);
1528 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1529 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1530 BUG_ON(compression);
1531
1532 off = btrfs_file_extent_inline_start(ei);
e41ca589 1533 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
31db9f7c
AB		 1534
1535 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
31db9f7c
AB		 1536
1537out:
1538 btrfs_free_path(path);
1539 return ret;
1540}
1541
1542/*
1543 * Helper function to generate a file name that is unique in the root of
1544 * send_root and parent_root. This is used to generate names for orphan inodes.
1545 */
1546static int gen_unique_name(struct send_ctx *sctx,
1547 u64 ino, u64 gen,
1548 struct fs_path *dest)
1549{
1550 int ret = 0;
1551 struct btrfs_path *path;
1552 struct btrfs_dir_item *di;
1553 char tmp[64];
1554 int len;
1555 u64 idx = 0;
1556
1557 path = alloc_path_for_send();
1558 if (!path)
1559 return -ENOMEM;
1560
1561 while (1) {
f74b86d8 1562 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
31db9f7c 1563 ino, gen, idx);
64792f25 1564 ASSERT(len < sizeof(tmp));
31db9f7c
AB		 1565
1566 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1567 path, BTRFS_FIRST_FREE_OBJECTID,
1568 tmp, strlen(tmp), 0);
1569 btrfs_release_path(path);
1570 if (IS_ERR(di)) {
1571 ret = PTR_ERR(di);
1572 goto out;
1573 }
1574 if (di) {
1575 /* not unique, try again */
1576 idx++;
1577 continue;
1578 }
1579
1580 if (!sctx->parent_root) {
1581 /* unique */
1582 ret = 0;
1583 break;
1584 }
1585
1586 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1587 path, BTRFS_FIRST_FREE_OBJECTID,
1588 tmp, strlen(tmp), 0);
1589 btrfs_release_path(path);
1590 if (IS_ERR(di)) {
1591 ret = PTR_ERR(di);
1592 goto out;
1593 }
1594 if (di) {
1595 /* not unique, try again */
1596 idx++;
1597 continue;
1598 }
1599 /* unique */
1600 break;
1601 }
1602
1603 ret = fs_path_add(dest, tmp, strlen(tmp));
1604
1605out:
1606 btrfs_free_path(path);
1607 return ret;
1608}
1609
1610enum inode_state {
1611 inode_state_no_change,
1612 inode_state_will_create,
1613 inode_state_did_create,
1614 inode_state_will_delete,
1615 inode_state_did_delete,
1616};
1617
1618static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1619{
1620 int ret;
1621 int left_ret;
1622 int right_ret;
1623 u64 left_gen;
1624 u64 right_gen;
1625
1626 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
85a7b33b 1627 NULL, NULL);
31db9f7c
AB		 1628 if (ret < 0 && ret != -ENOENT)
1629 goto out;
1630 left_ret = ret;
1631
1632 if (!sctx->parent_root) {
1633 right_ret = -ENOENT;
1634 } else {
1635 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
85a7b33b 1636 NULL, NULL, NULL, NULL);
31db9f7c
AB		 1637 if (ret < 0 && ret != -ENOENT)
1638 goto out;
1639 right_ret = ret;
1640 }
1641
1642 if (!left_ret && !right_ret) {
e938c8ad 1643 if (left_gen == gen && right_gen == gen) {
31db9f7c 1644 ret = inode_state_no_change;
e938c8ad 1645 } else if (left_gen == gen) {
31db9f7c
AB		 1646 if (ino < sctx->send_progress)
1647 ret = inode_state_did_create;
1648 else
1649 ret = inode_state_will_create;
1650 } else if (right_gen == gen) {
1651 if (ino < sctx->send_progress)
1652 ret = inode_state_did_delete;
1653 else
1654 ret = inode_state_will_delete;
1655 } else {
1656 ret = -ENOENT;
1657 }
1658 } else if (!left_ret) {
1659 if (left_gen == gen) {
1660 if (ino < sctx->send_progress)
1661 ret = inode_state_did_create;
1662 else
1663 ret = inode_state_will_create;
1664 } else {
1665 ret = -ENOENT;
1666 }
1667 } else if (!right_ret) {
1668 if (right_gen == gen) {
1669 if (ino < sctx->send_progress)
1670 ret = inode_state_did_delete;
1671 else
1672 ret = inode_state_will_delete;
1673 } else {
1674 ret = -ENOENT;
1675 }
1676 } else {
1677 ret = -ENOENT;
1678 }
1679
1680out:
1681 return ret;
1682}
1683
1684static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1685{
1686 int ret;
1687
4dd9920d
RK		 1688 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1689 return 1;
1690
31db9f7c
AB		 1691 ret = get_cur_inode_state(sctx, ino, gen);
1692 if (ret < 0)
1693 goto out;
1694
1695 if (ret == inode_state_no_change ||
1696 ret == inode_state_did_create ||
1697 ret == inode_state_will_delete)
1698 ret = 1;
1699 else
1700 ret = 0;
1701
1702out:
1703 return ret;
1704}
1705
1706/*
1707 * Helper function to lookup a dir item in a dir.
1708 */
1709static int lookup_dir_item_inode(struct btrfs_root *root,
1710 u64 dir, const char *name, int name_len,
1711 u64 *found_inode,
1712 u8 *found_type)
1713{
1714 int ret = 0;
1715 struct btrfs_dir_item *di;
1716 struct btrfs_key key;
1717 struct btrfs_path *path;
1718
1719 path = alloc_path_for_send();
1720 if (!path)
1721 return -ENOMEM;
1722
1723 di = btrfs_lookup_dir_item(NULL, root, path,
1724 dir, name, name_len, 0);
3cf5068f
LB		 1725 if (IS_ERR_OR_NULL(di)) {
1726 ret = di ? PTR_ERR(di) : -ENOENT;
31db9f7c
AB		 1727 goto out;
1728 }
1729 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1af56070
FM		 1730 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1731 ret = -ENOENT;
1732 goto out;
1733 }
31db9f7c
AB		 1734 *found_inode = key.objectid;
1735 *found_type = btrfs_dir_type(path->nodes[0], di);
1736
1737out:
1738 btrfs_free_path(path);
1739 return ret;
1740}
1741
766702ef
AB		 1742/*
1743 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1744 * generation of the parent dir and the name of the dir entry.
1745 */
924794c9 1746static int get_first_ref(struct btrfs_root *root, u64 ino,
31db9f7c
AB		 1747 u64 *dir, u64 *dir_gen, struct fs_path *name)
1748{
1749 int ret;
1750 struct btrfs_key key;
1751 struct btrfs_key found_key;
1752 struct btrfs_path *path;
31db9f7c 1753 int len;
96b5bd77 1754 u64 parent_dir;
31db9f7c
AB		 1755
1756 path = alloc_path_for_send();
1757 if (!path)
1758 return -ENOMEM;
1759
1760 key.objectid = ino;
1761 key.type = BTRFS_INODE_REF_KEY;
1762 key.offset = 0;
1763
1764 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1765 if (ret < 0)
1766 goto out;
1767 if (!ret)
1768 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1769 path->slots[0]);
96b5bd77
JS		 1770 if (ret || found_key.objectid != ino ||
1771 (found_key.type != BTRFS_INODE_REF_KEY &&
1772 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1773 ret = -ENOENT;
1774 goto out;
1775 }
1776
51a60253 1777 if (found_key.type == BTRFS_INODE_REF_KEY) {
96b5bd77
JS		 1778 struct btrfs_inode_ref *iref;
1779 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1780 struct btrfs_inode_ref);
1781 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1782 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1783 (unsigned long)(iref + 1),
1784 len);
1785 parent_dir = found_key.offset;
1786 } else {
1787 struct btrfs_inode_extref *extref;
1788 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1789 struct btrfs_inode_extref);
1790 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1791 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1792 (unsigned long)&extref->name, len);
1793 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1794 }
31db9f7c
AB		 1795 if (ret < 0)
1796 goto out;
1797 btrfs_release_path(path);
1798
b46ab97b
FM		 1799 if (dir_gen) {
1800 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1801 NULL, NULL, NULL);
1802 if (ret < 0)
1803 goto out;
1804 }
31db9f7c 1805
96b5bd77 1806 *dir = parent_dir;
31db9f7c
AB		 1807
1808out:
1809 btrfs_free_path(path);
1810 return ret;
1811}
1812
924794c9 1813static int is_first_ref(struct btrfs_root *root,
31db9f7c
AB		 1814 u64 ino, u64 dir,
1815 const char *name, int name_len)
1816{
1817 int ret;
1818 struct fs_path *tmp_name;
1819 u64 tmp_dir;
31db9f7c 1820
924794c9 1821 tmp_name = fs_path_alloc();
31db9f7c
AB		 1822 if (!tmp_name)
1823 return -ENOMEM;
1824
b46ab97b 1825 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
31db9f7c
AB		 1826 if (ret < 0)
1827 goto out;
1828
b9291aff 1829 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
31db9f7c
AB		 1830 ret = 0;
1831 goto out;
1832 }
1833
e938c8ad 1834 ret = !memcmp(tmp_name->start, name, name_len);
31db9f7c
AB		 1835
1836out:
924794c9 1837 fs_path_free(tmp_name);
31db9f7c
AB		 1838 return ret;
1839}
1840
766702ef
AB		 1841/*
1842 * Used by process_recorded_refs to determine if a new ref would overwrite an
1843 * already existing ref. In case it detects an overwrite, it returns the
1844 * inode/gen in who_ino/who_gen.
1845 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1846 * to make sure later references to the overwritten inode are possible.
1847 * Orphanizing is however only required for the first ref of an inode.
1848 * process_recorded_refs does an additional is_first_ref check to see if
1849 * orphanizing is really required.
1850 */
31db9f7c
AB		 1851static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1852 const char *name, int name_len,
f5962781 1853 u64 *who_ino, u64 *who_gen, u64 *who_mode)
31db9f7c
AB		 1854{
1855 int ret = 0;
ebdad913 1856 u64 gen;
31db9f7c
AB		 1857 u64 other_inode = 0;
1858 u8 other_type = 0;
1859
1860 if (!sctx->parent_root)
1861 goto out;
1862
1863 ret = is_inode_existent(sctx, dir, dir_gen);
1864 if (ret <= 0)
1865 goto out;
1866
ebdad913
JB		 1867 /*
1868 * If we have a parent root we need to verify that the parent dir was
01327610 1869 * not deleted and then re-created, if it was then we have no overwrite
ebdad913
JB		 1870 * and we can just unlink this entry.
1871 */
4dd9920d 1872 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
ebdad913
JB		 1873 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1874 NULL, NULL, NULL);
1875 if (ret < 0 && ret != -ENOENT)
1876 goto out;
1877 if (ret) {
1878 ret = 0;
1879 goto out;
1880 }
1881 if (gen != dir_gen)
1882 goto out;
1883 }
1884
31db9f7c
AB		 1885 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1886 &other_inode, &other_type);
1887 if (ret < 0 && ret != -ENOENT)
1888 goto out;
1889 if (ret) {
1890 ret = 0;
1891 goto out;
1892 }
1893
766702ef
AB		 1894 /*
1895 * Check if the overwritten ref was already processed. If yes, the ref
1896 * was already unlinked/moved, so we can safely assume that we will not
1897 * overwrite anything at this point in time.
1898 */
801bec36
RK		 1899 if (other_inode > sctx->send_progress ||
1900 is_waiting_for_move(sctx, other_inode)) {
31db9f7c 1901 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
f5962781 1902 who_gen, who_mode, NULL, NULL, NULL);
31db9f7c
AB		 1903 if (ret < 0)
1904 goto out;
1905
1906 ret = 1;
1907 *who_ino = other_inode;
1908 } else {
1909 ret = 0;
1910 }
1911
1912out:
1913 return ret;
1914}
1915
766702ef
AB		 1916/*
1917 * Checks if the ref was overwritten by an already processed inode. This is
1918 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1919 * thus the orphan name needs be used.
1920 * process_recorded_refs also uses it to avoid unlinking of refs that were
1921 * overwritten.
1922 */
31db9f7c
AB		 1923static int did_overwrite_ref(struct send_ctx *sctx,
1924 u64 dir, u64 dir_gen,
1925 u64 ino, u64 ino_gen,
1926 const char *name, int name_len)
1927{
1928 int ret = 0;
1929 u64 gen;
1930 u64 ow_inode;
1931 u8 other_type;
1932
1933 if (!sctx->parent_root)
1934 goto out;
1935
1936 ret = is_inode_existent(sctx, dir, dir_gen);
1937 if (ret <= 0)
1938 goto out;
1939
01914101
RK		 1940 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1941 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1942 NULL, NULL, NULL);
1943 if (ret < 0 && ret != -ENOENT)
1944 goto out;
1945 if (ret) {
1946 ret = 0;
1947 goto out;
1948 }
1949 if (gen != dir_gen)
1950 goto out;
1951 }
1952
31db9f7c
AB		 1953 /* check if the ref was overwritten by another ref */
1954 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1955 &ow_inode, &other_type);
1956 if (ret < 0 && ret != -ENOENT)
1957 goto out;
1958 if (ret) {
1959 /* was never and will never be overwritten */
1960 ret = 0;
1961 goto out;
1962 }
1963
1964 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
85a7b33b 1965 NULL, NULL);
31db9f7c
AB		 1966 if (ret < 0)
1967 goto out;
1968
1969 if (ow_inode == ino && gen == ino_gen) {
1970 ret = 0;
1971 goto out;
1972 }
1973
8b191a68
FM		 1974 /*
1975 * We know that it is or will be overwritten. Check this now.
1976 * The current inode being processed might have been the one that caused
b786f16a
FM		 1977 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1978 * the current inode being processed.
8b191a68 1979 */
b786f16a
FM		 1980 if ((ow_inode < sctx->send_progress) ||
1981 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1982 gen == sctx->cur_inode_gen))
31db9f7c
AB		 1983 ret = 1;
1984 else
1985 ret = 0;
1986
1987out:
1988 return ret;
1989}
1990
766702ef
AB		 1991/*
1992 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1993 * that got overwritten. This is used by process_recorded_refs to determine
1994 * if it has to use the path as returned by get_cur_path or the orphan name.
1995 */
31db9f7c
AB		 1996static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1997{
1998 int ret = 0;
1999 struct fs_path *name = NULL;
2000 u64 dir;
2001 u64 dir_gen;
2002
2003 if (!sctx->parent_root)
2004 goto out;
2005
924794c9 2006 name = fs_path_alloc();
31db9f7c
AB		 2007 if (!name)
2008 return -ENOMEM;
2009
924794c9 2010 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
31db9f7c
AB		 2011 if (ret < 0)
2012 goto out;
2013
2014 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
2015 name->start, fs_path_len(name));
31db9f7c
AB		 2016
2017out:
924794c9 2018 fs_path_free(name);
31db9f7c
AB		 2019 return ret;
2020}
2021
766702ef
AB		 2022/*
2023 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2024 * so we need to do some special handling in case we have clashes. This function
2025 * takes care of this with the help of name_cache_entry::radix_list.
5dc67d0b 2026 * In case of error, nce is kfreed.
766702ef 2027 */
31db9f7c
AB		 2028static int name_cache_insert(struct send_ctx *sctx,
2029 struct name_cache_entry *nce)
2030{
2031 int ret = 0;
7e0926fe
AB		 2032 struct list_head *nce_head;
2033
2034 nce_head = radix_tree_lookup(&sctx->name_cache,
2035 (unsigned long)nce->ino);
2036 if (!nce_head) {
e780b0d1 2037 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
cfa7a9cc
TI		 2038 if (!nce_head) {
2039 kfree(nce);
31db9f7c 2040 return -ENOMEM;
cfa7a9cc 2041 }
7e0926fe 2042 INIT_LIST_HEAD(nce_head);
31db9f7c 2043
7e0926fe 2044 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
5dc67d0b
AB		 2045 if (ret < 0) {
2046 kfree(nce_head);
2047 kfree(nce);
31db9f7c 2048 return ret;
5dc67d0b 2049 }
31db9f7c 2050 }
7e0926fe 2051 list_add_tail(&nce->radix_list, nce_head);
31db9f7c
AB		 2052 list_add_tail(&nce->list, &sctx->name_cache_list);
2053 sctx->name_cache_size++;
2054
2055 return ret;
2056}
2057
2058static void name_cache_delete(struct send_ctx *sctx,
2059 struct name_cache_entry *nce)
2060{
7e0926fe 2061 struct list_head *nce_head;
31db9f7c 2062
7e0926fe
AB		 2063 nce_head = radix_tree_lookup(&sctx->name_cache,
2064 (unsigned long)nce->ino);
57fb8910
DS		 2065 if (!nce_head) {
2066 btrfs_err(sctx->send_root->fs_info,
2067 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2068 nce->ino, sctx->name_cache_size);
2069 }
31db9f7c 2070
7e0926fe 2071 list_del(&nce->radix_list);
31db9f7c 2072 list_del(&nce->list);
31db9f7c 2073 sctx->name_cache_size--;
7e0926fe 2074
57fb8910
DS		 2075 /*
2076 * We may not get to the final release of nce_head if the lookup fails
2077 */
2078 if (nce_head && list_empty(nce_head)) {
7e0926fe
AB		 2079 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2080 kfree(nce_head);
2081 }
31db9f7c
AB		 2082}
2083
2084static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2085 u64 ino, u64 gen)
2086{
7e0926fe
AB		 2087 struct list_head *nce_head;
2088 struct name_cache_entry *cur;
31db9f7c 2089
7e0926fe
AB		 2090 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2091 if (!nce_head)
31db9f7c
AB		 2092 return NULL;
2093
7e0926fe
AB		 2094 list_for_each_entry(cur, nce_head, radix_list) {
2095 if (cur->ino == ino && cur->gen == gen)
2096 return cur;
2097 }
31db9f7c
AB		 2098 return NULL;
2099}
2100
766702ef
AB		 2101/*
2102 * Removes the entry from the list and adds it back to the end. This marks the
2103 * entry as recently used so that name_cache_clean_unused does not remove it.
2104 */
31db9f7c
AB		 2105static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2106{
2107 list_del(&nce->list);
2108 list_add_tail(&nce->list, &sctx->name_cache_list);
2109}
2110
766702ef
AB		 2111/*
2112 * Remove some entries from the beginning of name_cache_list.
2113 */
31db9f7c
AB		 2114static void name_cache_clean_unused(struct send_ctx *sctx)
2115{
2116 struct name_cache_entry *nce;
2117
2118 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2119 return;
2120
2121 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2122 nce = list_entry(sctx->name_cache_list.next,
2123 struct name_cache_entry, list);
2124 name_cache_delete(sctx, nce);
2125 kfree(nce);
2126 }
2127}
2128
2129static void name_cache_free(struct send_ctx *sctx)
2130{
2131 struct name_cache_entry *nce;
31db9f7c 2132
e938c8ad
AB		 2133 while (!list_empty(&sctx->name_cache_list)) {
2134 nce = list_entry(sctx->name_cache_list.next,
2135 struct name_cache_entry, list);
31db9f7c 2136 name_cache_delete(sctx, nce);
17589bd9 2137 kfree(nce);
31db9f7c
AB		 2138 }
2139}
2140
766702ef
AB		 2141/*
2142 * Used by get_cur_path for each ref up to the root.
2143 * Returns 0 if it succeeded.
2144 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2145 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2146 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2147 * Returns <0 in case of error.
2148 */
31db9f7c
AB		 2149static int __get_cur_name_and_parent(struct send_ctx *sctx,
2150 u64 ino, u64 gen,
2151 u64 *parent_ino,
2152 u64 *parent_gen,
2153 struct fs_path *dest)
2154{
2155 int ret;
2156 int nce_ret;
31db9f7c
AB		 2157 struct name_cache_entry *nce = NULL;
2158
766702ef
AB		 2159 /*
2160 * First check if we already did a call to this function with the same
2161 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2162 * return the cached result.
2163 */
31db9f7c
AB		 2164 nce = name_cache_search(sctx, ino, gen);
2165 if (nce) {
2166 if (ino < sctx->send_progress && nce->need_later_update) {
2167 name_cache_delete(sctx, nce);
2168 kfree(nce);
2169 nce = NULL;
2170 } else {
2171 name_cache_used(sctx, nce);
2172 *parent_ino = nce->parent_ino;
2173 *parent_gen = nce->parent_gen;
2174 ret = fs_path_add(dest, nce->name, nce->name_len);
2175 if (ret < 0)
2176 goto out;
2177 ret = nce->ret;
2178 goto out;
2179 }
2180 }
2181
766702ef
AB		 2182 /*
2183 * If the inode is not existent yet, add the orphan name and return 1.
2184 * This should only happen for the parent dir that we determine in
2185 * __record_new_ref
2186 */
31db9f7c
AB		 2187 ret = is_inode_existent(sctx, ino, gen);
2188 if (ret < 0)
2189 goto out;
2190
2191 if (!ret) {
2192 ret = gen_unique_name(sctx, ino, gen, dest);
2193 if (ret < 0)
2194 goto out;
2195 ret = 1;
2196 goto out_cache;
2197 }
2198
766702ef
AB		 2199 /*
2200 * Depending on whether the inode was already processed or not, use
2201 * send_root or parent_root for ref lookup.
2202 */
bf0d1f44 2203 if (ino < sctx->send_progress)
924794c9
TI		 2204 ret = get_first_ref(sctx->send_root, ino,
2205 parent_ino, parent_gen, dest);
31db9f7c 2206 else
924794c9
TI		 2207 ret = get_first_ref(sctx->parent_root, ino,
2208 parent_ino, parent_gen, dest);
31db9f7c
AB		 2209 if (ret < 0)
2210 goto out;
2211
766702ef
AB		 2212 /*
2213 * Check if the ref was overwritten by an inode's ref that was processed
2214 * earlier. If yes, treat as orphan and return 1.
2215 */
31db9f7c
AB		 2216 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2217 dest->start, dest->end - dest->start);
2218 if (ret < 0)
2219 goto out;
2220 if (ret) {
2221 fs_path_reset(dest);
2222 ret = gen_unique_name(sctx, ino, gen, dest);
2223 if (ret < 0)
2224 goto out;
2225 ret = 1;
2226 }
2227
2228out_cache:
766702ef
AB		 2229 /*
2230 * Store the result of the lookup in the name cache.
2231 */
e780b0d1 2232 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
31db9f7c
AB		 2233 if (!nce) {
2234 ret = -ENOMEM;
2235 goto out;
2236 }
2237
2238 nce->ino = ino;
2239 nce->gen = gen;
2240 nce->parent_ino = *parent_ino;
2241 nce->parent_gen = *parent_gen;
2242 nce->name_len = fs_path_len(dest);
2243 nce->ret = ret;
2244 strcpy(nce->name, dest->start);
31db9f7c
AB		 2245
2246 if (ino < sctx->send_progress)
2247 nce->need_later_update = 0;
2248 else
2249 nce->need_later_update = 1;
2250
2251 nce_ret = name_cache_insert(sctx, nce);
2252 if (nce_ret < 0)
2253 ret = nce_ret;
2254 name_cache_clean_unused(sctx);
2255
2256out:
31db9f7c
AB		 2257 return ret;
2258}
2259
2260/*
2261 * Magic happens here. This function returns the first ref to an inode as it
2262 * would look like while receiving the stream at this point in time.
2263 * We walk the path up to the root. For every inode in between, we check if it
2264 * was already processed/sent. If yes, we continue with the parent as found
2265 * in send_root. If not, we continue with the parent as found in parent_root.
2266 * If we encounter an inode that was deleted at this point in time, we use the
2267 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2268 * that were not created yet and overwritten inodes/refs.
2269 *
52042d8e 2270 * When do we have orphan inodes:
31db9f7c
AB		 2271 * 1. When an inode is freshly created and thus no valid refs are available yet
2272 * 2. When a directory lost all it's refs (deleted) but still has dir items
2273 * inside which were not processed yet (pending for move/delete). If anyone
2274 * tried to get the path to the dir items, it would get a path inside that
2275 * orphan directory.
2276 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2277 * of an unprocessed inode. If in that case the first ref would be
2278 * overwritten, the overwritten inode gets "orphanized". Later when we
2279 * process this overwritten inode, it is restored at a new place by moving
2280 * the orphan inode.
2281 *
2282 * sctx->send_progress tells this function at which point in time receiving
2283 * would be.
2284 */
2285static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2286 struct fs_path *dest)
2287{
2288 int ret = 0;
2289 struct fs_path *name = NULL;
2290 u64 parent_inode = 0;
2291 u64 parent_gen = 0;
2292 int stop = 0;
2293
924794c9 2294 name = fs_path_alloc();
31db9f7c
AB		 2295 if (!name) {
2296 ret = -ENOMEM;
2297 goto out;
2298 }
2299
2300 dest->reversed = 1;
2301 fs_path_reset(dest);
2302
2303 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
8b191a68
FM		 2304 struct waiting_dir_move *wdm;
2305
31db9f7c
AB		 2306 fs_path_reset(name);
2307
9dc44214
FM		 2308 if (is_waiting_for_rm(sctx, ino)) {
2309 ret = gen_unique_name(sctx, ino, gen, name);
2310 if (ret < 0)
2311 goto out;
2312 ret = fs_path_add_path(dest, name);
2313 break;
2314 }
2315
8b191a68
FM		 2316 wdm = get_waiting_dir_move(sctx, ino);
2317 if (wdm && wdm->orphanized) {
2318 ret = gen_unique_name(sctx, ino, gen, name);
2319 stop = 1;
2320 } else if (wdm) {
bf0d1f44
FM		 2321 ret = get_first_ref(sctx->parent_root, ino,
2322 &parent_inode, &parent_gen, name);
2323 } else {
2324 ret = __get_cur_name_and_parent(sctx, ino, gen,
2325 &parent_inode,
2326 &parent_gen, name);
2327 if (ret)
2328 stop = 1;
2329 }
2330
31db9f7c
AB		 2331 if (ret < 0)
2332 goto out;
9f03740a 2333
31db9f7c
AB		 2334 ret = fs_path_add_path(dest, name);
2335 if (ret < 0)
2336 goto out;
2337
2338 ino = parent_inode;
2339 gen = parent_gen;
2340 }
2341
2342out:
924794c9 2343 fs_path_free(name);
31db9f7c
AB		 2344 if (!ret)
2345 fs_path_unreverse(dest);
2346 return ret;
2347}
2348
31db9f7c
AB		 2349/*
2350 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2351 */
2352static int send_subvol_begin(struct send_ctx *sctx)
2353{
2354 int ret;
2355 struct btrfs_root *send_root = sctx->send_root;
2356 struct btrfs_root *parent_root = sctx->parent_root;
2357 struct btrfs_path *path;
2358 struct btrfs_key key;
2359 struct btrfs_root_ref *ref;
2360 struct extent_buffer *leaf;
2361 char *name = NULL;
2362 int namelen;
2363
ffcfaf81 2364 path = btrfs_alloc_path();
31db9f7c
AB		 2365 if (!path)
2366 return -ENOMEM;
2367
e780b0d1 2368 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
31db9f7c
AB		 2369 if (!name) {
2370 btrfs_free_path(path);
2371 return -ENOMEM;
2372 }
2373
4fd786e6 2374 key.objectid = send_root->root_key.objectid;
31db9f7c
AB		 2375 key.type = BTRFS_ROOT_BACKREF_KEY;
2376 key.offset = 0;
2377
2378 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2379 &key, path, 1, 0);
2380 if (ret < 0)
2381 goto out;
2382 if (ret) {
2383 ret = -ENOENT;
2384 goto out;
2385 }
2386
2387 leaf = path->nodes[0];
2388 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2389 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
4fd786e6 2390 key.objectid != send_root->root_key.objectid) {
31db9f7c
AB		 2391 ret = -ENOENT;
2392 goto out;
2393 }
2394 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2395 namelen = btrfs_root_ref_name_len(leaf, ref);
2396 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2397 btrfs_release_path(path);
2398
31db9f7c
AB		 2399 if (parent_root) {
2400 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2401 if (ret < 0)
2402 goto out;
2403 } else {
2404 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2405 if (ret < 0)
2406 goto out;
2407 }
2408
2409 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
b96b1db0
RR		 2410
2411 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2412 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2413 sctx->send_root->root_item.received_uuid);
2414 else
2415 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2416 sctx->send_root->root_item.uuid);
2417
31db9f7c 2418 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
5a0f4e2c 2419 le64_to_cpu(sctx->send_root->root_item.ctransid));
31db9f7c 2420 if (parent_root) {
37b8d27d
JB		 2421 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2422 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2423 parent_root->root_item.received_uuid);
2424 else
2425 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2426 parent_root->root_item.uuid);
31db9f7c 2427 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
5a0f4e2c 2428 le64_to_cpu(sctx->parent_root->root_item.ctransid));
31db9f7c
AB		 2429 }
2430
2431 ret = send_cmd(sctx);
2432
2433tlv_put_failure:
2434out:
2435 btrfs_free_path(path);
2436 kfree(name);
2437 return ret;
2438}
2439
2440static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2441{
04ab956e 2442 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2443 int ret = 0;
2444 struct fs_path *p;
2445
04ab956e 2446 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
31db9f7c 2447
924794c9 2448 p = fs_path_alloc();
31db9f7c
AB		 2449 if (!p)
2450 return -ENOMEM;
2451
2452 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2453 if (ret < 0)
2454 goto out;
2455
2456 ret = get_cur_path(sctx, ino, gen, p);
2457 if (ret < 0)
2458 goto out;
2459 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2460 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2461
2462 ret = send_cmd(sctx);
2463
2464tlv_put_failure:
2465out:
924794c9 2466 fs_path_free(p);
31db9f7c
AB		 2467 return ret;
2468}
2469
2470static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2471{
04ab956e 2472 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2473 int ret = 0;
2474 struct fs_path *p;
2475
04ab956e 2476 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
31db9f7c 2477
924794c9 2478 p = fs_path_alloc();
31db9f7c
AB		 2479 if (!p)
2480 return -ENOMEM;
2481
2482 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2483 if (ret < 0)
2484 goto out;
2485
2486 ret = get_cur_path(sctx, ino, gen, p);
2487 if (ret < 0)
2488 goto out;
2489 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2490 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2491
2492 ret = send_cmd(sctx);
2493
2494tlv_put_failure:
2495out:
924794c9 2496 fs_path_free(p);
31db9f7c
AB		 2497 return ret;
2498}
2499
2500static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2501{
04ab956e 2502 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2503 int ret = 0;
2504 struct fs_path *p;
2505
04ab956e
JM		 2506 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2507 ino, uid, gid);
31db9f7c 2508
924794c9 2509 p = fs_path_alloc();
31db9f7c
AB		 2510 if (!p)
2511 return -ENOMEM;
2512
2513 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2514 if (ret < 0)
2515 goto out;
2516
2517 ret = get_cur_path(sctx, ino, gen, p);
2518 if (ret < 0)
2519 goto out;
2520 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2521 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2522 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2523
2524 ret = send_cmd(sctx);
2525
2526tlv_put_failure:
2527out:
924794c9 2528 fs_path_free(p);
31db9f7c
AB		 2529 return ret;
2530}
2531
2532static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2533{
04ab956e 2534 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2535 int ret = 0;
2536 struct fs_path *p = NULL;
2537 struct btrfs_inode_item *ii;
2538 struct btrfs_path *path = NULL;
2539 struct extent_buffer *eb;
2540 struct btrfs_key key;
2541 int slot;
2542
04ab956e 2543 btrfs_debug(fs_info, "send_utimes %llu", ino);
31db9f7c 2544
924794c9 2545 p = fs_path_alloc();
31db9f7c
AB		 2546 if (!p)
2547 return -ENOMEM;
2548
2549 path = alloc_path_for_send();
2550 if (!path) {
2551 ret = -ENOMEM;
2552 goto out;
2553 }
2554
2555 key.objectid = ino;
2556 key.type = BTRFS_INODE_ITEM_KEY;
2557 key.offset = 0;
2558 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
15b253ea
FM		 2559 if (ret > 0)
2560 ret = -ENOENT;
31db9f7c
AB		 2561 if (ret < 0)
2562 goto out;
2563
2564 eb = path->nodes[0];
2565 slot = path->slots[0];
2566 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2567
2568 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2569 if (ret < 0)
2570 goto out;
2571
2572 ret = get_cur_path(sctx, ino, gen, p);
2573 if (ret < 0)
2574 goto out;
2575 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
a937b979
DS		 2576 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2577 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2578 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
766702ef 2579 /* TODO Add otime support when the otime patches get into upstream */
31db9f7c
AB		 2580
2581 ret = send_cmd(sctx);
2582
2583tlv_put_failure:
2584out:
924794c9 2585 fs_path_free(p);
31db9f7c
AB		 2586 btrfs_free_path(path);
2587 return ret;
2588}
2589
2590/*
2591 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2592 * a valid path yet because we did not process the refs yet. So, the inode
2593 * is created as orphan.
2594 */
1f4692da 2595static int send_create_inode(struct send_ctx *sctx, u64 ino)
31db9f7c 2596{
04ab956e 2597 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c 2598 int ret = 0;
31db9f7c 2599 struct fs_path *p;
31db9f7c 2600 int cmd;
1f4692da 2601 u64 gen;
31db9f7c 2602 u64 mode;
1f4692da 2603 u64 rdev;
31db9f7c 2604
04ab956e 2605 btrfs_debug(fs_info, "send_create_inode %llu", ino);
31db9f7c 2606
924794c9 2607 p = fs_path_alloc();
31db9f7c
AB		 2608 if (!p)
2609 return -ENOMEM;
2610
644d1940
LB		 2611 if (ino != sctx->cur_ino) {
2612 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2613 NULL, NULL, &rdev);
2614 if (ret < 0)
2615 goto out;
2616 } else {
2617 gen = sctx->cur_inode_gen;
2618 mode = sctx->cur_inode_mode;
2619 rdev = sctx->cur_inode_rdev;
2620 }
31db9f7c 2621
e938c8ad 2622 if (S_ISREG(mode)) {
31db9f7c 2623 cmd = BTRFS_SEND_C_MKFILE;
e938c8ad 2624 } else if (S_ISDIR(mode)) {
31db9f7c 2625 cmd = BTRFS_SEND_C_MKDIR;
e938c8ad 2626 } else if (S_ISLNK(mode)) {
31db9f7c 2627 cmd = BTRFS_SEND_C_SYMLINK;
e938c8ad 2628 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
31db9f7c 2629 cmd = BTRFS_SEND_C_MKNOD;
e938c8ad 2630 } else if (S_ISFIFO(mode)) {
31db9f7c 2631 cmd = BTRFS_SEND_C_MKFIFO;
e938c8ad 2632 } else if (S_ISSOCK(mode)) {
31db9f7c 2633 cmd = BTRFS_SEND_C_MKSOCK;
e938c8ad 2634 } else {
f14d104d 2635 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
31db9f7c 2636 (int)(mode & S_IFMT));
ca6842bf 2637 ret = -EOPNOTSUPP;
31db9f7c
AB		 2638 goto out;
2639 }
2640
2641 ret = begin_cmd(sctx, cmd);
2642 if (ret < 0)
2643 goto out;
2644
1f4692da 2645 ret = gen_unique_name(sctx, ino, gen, p);
31db9f7c
AB		 2646 if (ret < 0)
2647 goto out;
2648
2649 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
1f4692da 2650 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
31db9f7c
AB		 2651
2652 if (S_ISLNK(mode)) {
2653 fs_path_reset(p);
924794c9 2654 ret = read_symlink(sctx->send_root, ino, p);
31db9f7c
AB		 2655 if (ret < 0)
2656 goto out;
2657 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2658 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2659 S_ISFIFO(mode) || S_ISSOCK(mode)) {
d79e5043
AJ		 2660 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2661 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
31db9f7c
AB		 2662 }
2663
2664 ret = send_cmd(sctx);
2665 if (ret < 0)
2666 goto out;
2667
2668
2669tlv_put_failure:
2670out:
924794c9 2671 fs_path_free(p);
31db9f7c
AB		 2672 return ret;
2673}
2674
1f4692da
AB		 2675/*
2676 * We need some special handling for inodes that get processed before the parent
2677 * directory got created. See process_recorded_refs for details.
2678 * This function does the check if we already created the dir out of order.
2679 */
2680static int did_create_dir(struct send_ctx *sctx, u64 dir)
2681{
2682 int ret = 0;
2683 struct btrfs_path *path = NULL;
2684 struct btrfs_key key;
2685 struct btrfs_key found_key;
2686 struct btrfs_key di_key;
2687 struct extent_buffer *eb;
2688 struct btrfs_dir_item *di;
2689 int slot;
2690
2691 path = alloc_path_for_send();
2692 if (!path) {
2693 ret = -ENOMEM;
2694 goto out;
2695 }
2696
2697 key.objectid = dir;
2698 key.type = BTRFS_DIR_INDEX_KEY;
2699 key.offset = 0;
dff6d0ad
FDBM		 2700 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2701 if (ret < 0)
2702 goto out;
2703
1f4692da 2704 while (1) {
dff6d0ad
FDBM		 2705 eb = path->nodes[0];
2706 slot = path->slots[0];
2707 if (slot >= btrfs_header_nritems(eb)) {
2708 ret = btrfs_next_leaf(sctx->send_root, path);
2709 if (ret < 0) {
2710 goto out;
2711 } else if (ret > 0) {
2712 ret = 0;
2713 break;
2714 }
2715 continue;
1f4692da 2716 }
dff6d0ad
FDBM		 2717
2718 btrfs_item_key_to_cpu(eb, &found_key, slot);
2719 if (found_key.objectid != key.objectid ||
1f4692da
AB		 2720 found_key.type != key.type) {
2721 ret = 0;
2722 goto out;
2723 }
2724
2725 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2726 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2727
a0525414
JB		 2728 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2729 di_key.objectid < sctx->send_progress) {
1f4692da
AB		 2730 ret = 1;
2731 goto out;
2732 }
2733
dff6d0ad 2734 path->slots[0]++;
1f4692da
AB		 2735 }
2736
2737out:
2738 btrfs_free_path(path);
2739 return ret;
2740}
2741
2742/*
2743 * Only creates the inode if it is:
2744 * 1. Not a directory
2745 * 2. Or a directory which was not created already due to out of order
2746 * directories. See did_create_dir and process_recorded_refs for details.
2747 */
2748static int send_create_inode_if_needed(struct send_ctx *sctx)
2749{
2750 int ret;
2751
2752 if (S_ISDIR(sctx->cur_inode_mode)) {
2753 ret = did_create_dir(sctx, sctx->cur_ino);
2754 if (ret < 0)
2755 goto out;
2756 if (ret) {
2757 ret = 0;
2758 goto out;
2759 }
2760 }
2761
2762 ret = send_create_inode(sctx, sctx->cur_ino);
2763 if (ret < 0)
2764 goto out;
2765
2766out:
2767 return ret;
2768}
2769
31db9f7c
AB		 2770struct recorded_ref {
2771 struct list_head list;
31db9f7c
AB		 2772 char *name;
2773 struct fs_path *full_path;
2774 u64 dir;
2775 u64 dir_gen;
31db9f7c
AB		 2776 int name_len;
2777};
2778
fdb13889
FM		 2779static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2780{
2781 ref->full_path = path;
2782 ref->name = (char *)kbasename(ref->full_path->start);
2783 ref->name_len = ref->full_path->end - ref->name;
2784}
2785
31db9f7c
AB		 2786/*
2787 * We need to process new refs before deleted refs, but compare_tree gives us
2788 * everything mixed. So we first record all refs and later process them.
2789 * This function is a helper to record one ref.
2790 */
a4d96d62 2791static int __record_ref(struct list_head *head, u64 dir,
31db9f7c
AB		 2792 u64 dir_gen, struct fs_path *path)
2793{
2794 struct recorded_ref *ref;
31db9f7c 2795
e780b0d1 2796 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
31db9f7c
AB		 2797 if (!ref)
2798 return -ENOMEM;
2799
2800 ref->dir = dir;
2801 ref->dir_gen = dir_gen;
fdb13889 2802 set_ref_path(ref, path);
31db9f7c
AB		 2803 list_add_tail(&ref->list, head);
2804 return 0;
2805}
2806
ba5e8f2e
JB		 2807static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2808{
2809 struct recorded_ref *new;
2810
e780b0d1 2811 new = kmalloc(sizeof(*ref), GFP_KERNEL);
ba5e8f2e
JB		 2812 if (!new)
2813 return -ENOMEM;
2814
2815 new->dir = ref->dir;
2816 new->dir_gen = ref->dir_gen;
2817 new->full_path = NULL;
2818 INIT_LIST_HEAD(&new->list);
2819 list_add_tail(&new->list, list);
2820 return 0;
2821}
2822
924794c9 2823static void __free_recorded_refs(struct list_head *head)
31db9f7c
AB		 2824{
2825 struct recorded_ref *cur;
31db9f7c 2826
e938c8ad
AB		 2827 while (!list_empty(head)) {
2828 cur = list_entry(head->next, struct recorded_ref, list);
924794c9 2829 fs_path_free(cur->full_path);
e938c8ad 2830 list_del(&cur->list);
31db9f7c
AB		 2831 kfree(cur);
2832 }
31db9f7c
AB		 2833}
2834
2835static void free_recorded_refs(struct send_ctx *sctx)
2836{
924794c9
TI		 2837 __free_recorded_refs(&sctx->new_refs);
2838 __free_recorded_refs(&sctx->deleted_refs);
31db9f7c
AB		 2839}
2840
2841/*
766702ef 2842 * Renames/moves a file/dir to its orphan name. Used when the first
31db9f7c
AB		 2843 * ref of an unprocessed inode gets overwritten and for all non empty
2844 * directories.
2845 */
2846static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2847 struct fs_path *path)
2848{
2849 int ret;
2850 struct fs_path *orphan;
2851
924794c9 2852 orphan = fs_path_alloc();
31db9f7c
AB		 2853 if (!orphan)
2854 return -ENOMEM;
2855
2856 ret = gen_unique_name(sctx, ino, gen, orphan);
2857 if (ret < 0)
2858 goto out;
2859
2860 ret = send_rename(sctx, path, orphan);
2861
2862out:
924794c9 2863 fs_path_free(orphan);
31db9f7c
AB		 2864 return ret;
2865}
2866
9dc44214
FM		 2867static struct orphan_dir_info *
2868add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2869{
2870 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2871 struct rb_node *parent = NULL;
2872 struct orphan_dir_info *entry, *odi;
2873
9dc44214
FM		 2874 while (*p) {
2875 parent = *p;
2876 entry = rb_entry(parent, struct orphan_dir_info, node);
2877 if (dir_ino < entry->ino) {
2878 p = &(*p)->rb_left;
2879 } else if (dir_ino > entry->ino) {
2880 p = &(*p)->rb_right;
2881 } else {
9dc44214
FM		 2882 return entry;
2883 }
2884 }
2885
35c8eda1
RK		 2886 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2887 if (!odi)
2888 return ERR_PTR(-ENOMEM);
2889 odi->ino = dir_ino;
2890 odi->gen = 0;
0f96f517 2891 odi->last_dir_index_offset = 0;
35c8eda1 2892
9dc44214
FM		 2893 rb_link_node(&odi->node, parent, p);
2894 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2895 return odi;
2896}
2897
2898static struct orphan_dir_info *
2899get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2900{
2901 struct rb_node *n = sctx->orphan_dirs.rb_node;
2902 struct orphan_dir_info *entry;
2903
2904 while (n) {
2905 entry = rb_entry(n, struct orphan_dir_info, node);
2906 if (dir_ino < entry->ino)
2907 n = n->rb_left;
2908 else if (dir_ino > entry->ino)
2909 n = n->rb_right;
2910 else
2911 return entry;
2912 }
2913 return NULL;
2914}
2915
2916static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2917{
2918 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2919
2920 return odi != NULL;
2921}
2922
2923static void free_orphan_dir_info(struct send_ctx *sctx,
2924 struct orphan_dir_info *odi)
2925{
2926 if (!odi)
2927 return;
2928 rb_erase(&odi->node, &sctx->orphan_dirs);
2929 kfree(odi);
2930}
2931
31db9f7c
AB		 2932/*
2933 * Returns 1 if a directory can be removed at this point in time.
2934 * We check this by iterating all dir items and checking if the inode behind
2935 * the dir item was already processed.
2936 */
9dc44214
FM		 2937static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2938 u64 send_progress)
31db9f7c
AB		 2939{
2940 int ret = 0;
2941 struct btrfs_root *root = sctx->parent_root;
2942 struct btrfs_path *path;
2943 struct btrfs_key key;
2944 struct btrfs_key found_key;
2945 struct btrfs_key loc;
2946 struct btrfs_dir_item *di;
0f96f517 2947 struct orphan_dir_info *odi = NULL;
31db9f7c 2948
6d85ed05
AB		 2949 /*
2950 * Don't try to rmdir the top/root subvolume dir.
2951 */
2952 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2953 return 0;
2954
31db9f7c
AB		 2955 path = alloc_path_for_send();
2956 if (!path)
2957 return -ENOMEM;
2958
2959 key.objectid = dir;
2960 key.type = BTRFS_DIR_INDEX_KEY;
2961 key.offset = 0;
0f96f517
RK		 2962
2963 odi = get_orphan_dir_info(sctx, dir);
2964 if (odi)
2965 key.offset = odi->last_dir_index_offset;
2966
dff6d0ad
FDBM		 2967 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2968 if (ret < 0)
2969 goto out;
31db9f7c
AB		 2970
2971 while (1) {
9dc44214
FM		 2972 struct waiting_dir_move *dm;
2973
dff6d0ad
FDBM		 2974 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2975 ret = btrfs_next_leaf(root, path);
2976 if (ret < 0)
2977 goto out;
2978 else if (ret > 0)
2979 break;
2980 continue;
31db9f7c 2981 }
dff6d0ad
FDBM		 2982 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2983 path->slots[0]);
2984 if (found_key.objectid != key.objectid ||
2985 found_key.type != key.type)
31db9f7c 2986 break;
31db9f7c
AB		 2987
2988 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2989 struct btrfs_dir_item);
2990 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2991
9dc44214
FM		 2992 dm = get_waiting_dir_move(sctx, loc.objectid);
2993 if (dm) {
9dc44214
FM		 2994 odi = add_orphan_dir_info(sctx, dir);
2995 if (IS_ERR(odi)) {
2996 ret = PTR_ERR(odi);
2997 goto out;
2998 }
2999 odi->gen = dir_gen;
0f96f517 3000 odi->last_dir_index_offset = found_key.offset;
9dc44214
FM		 3001 dm->rmdir_ino = dir;
3002 ret = 0;
3003 goto out;
3004 }
3005
31db9f7c 3006 if (loc.objectid > send_progress) {
0f96f517
RK		 3007 odi = add_orphan_dir_info(sctx, dir);
3008 if (IS_ERR(odi)) {
3009 ret = PTR_ERR(odi);
3010 goto out;
3011 }
3012 odi->gen = dir_gen;
3013 odi->last_dir_index_offset = found_key.offset;
31db9f7c
AB		 3014 ret = 0;
3015 goto out;
3016 }
3017
dff6d0ad 3018 path->slots[0]++;
31db9f7c 3019 }
0f96f517 3020 free_orphan_dir_info(sctx, odi);
31db9f7c
AB		 3021
3022 ret = 1;
3023
3024out:
3025 btrfs_free_path(path);
3026 return ret;
3027}
3028
9f03740a
FDBM		 3029static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3030{
9dc44214 3031 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
9f03740a 3032
9dc44214 3033 return entry != NULL;
9f03740a
FDBM		 3034}
3035
8b191a68 3036static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
9f03740a
FDBM		 3037{
3038 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3039 struct rb_node *parent = NULL;
3040 struct waiting_dir_move *entry, *dm;
3041
e780b0d1 3042 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
9f03740a
FDBM		 3043 if (!dm)
3044 return -ENOMEM;
3045 dm->ino = ino;
9dc44214 3046 dm->rmdir_ino = 0;
8b191a68 3047 dm->orphanized = orphanized;
9f03740a
FDBM		 3048
3049 while (*p) {
3050 parent = *p;
3051 entry = rb_entry(parent, struct waiting_dir_move, node);
3052 if (ino < entry->ino) {
3053 p = &(*p)->rb_left;
3054 } else if (ino > entry->ino) {
3055 p = &(*p)->rb_right;
3056 } else {
3057 kfree(dm);
3058 return -EEXIST;
3059 }
3060 }
3061
3062 rb_link_node(&dm->node, parent, p);
3063 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3064 return 0;
3065}
3066
9dc44214
FM		 3067static struct waiting_dir_move *
3068get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
9f03740a
FDBM		 3069{
3070 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3071 struct waiting_dir_move *entry;
3072
3073 while (n) {
3074 entry = rb_entry(n, struct waiting_dir_move, node);
9dc44214 3075 if (ino < entry->ino)
9f03740a 3076 n = n->rb_left;
9dc44214 3077 else if (ino > entry->ino)
9f03740a 3078 n = n->rb_right;
9dc44214
FM		 3079 else
3080 return entry;
9f03740a 3081 }
9dc44214
FM		 3082 return NULL;
3083}
3084
3085static void free_waiting_dir_move(struct send_ctx *sctx,
3086 struct waiting_dir_move *dm)
3087{
3088 if (!dm)
3089 return;
3090 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3091 kfree(dm);
9f03740a
FDBM		 3092}
3093
bfa7e1f8
FM		 3094static int add_pending_dir_move(struct send_ctx *sctx,
3095 u64 ino,
3096 u64 ino_gen,
f959492f
FM		 3097 u64 parent_ino,
3098 struct list_head *new_refs,
84471e24
FM		 3099 struct list_head *deleted_refs,
3100 const bool is_orphan)
9f03740a
FDBM		 3101{
3102 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3103 struct rb_node *parent = NULL;
73b802f4 3104 struct pending_dir_move *entry = NULL, *pm;
9f03740a
FDBM		 3105 struct recorded_ref *cur;
3106 int exists = 0;
3107 int ret;
3108
e780b0d1 3109 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
9f03740a
FDBM		 3110 if (!pm)
3111 return -ENOMEM;
3112 pm->parent_ino = parent_ino;
bfa7e1f8
FM		 3113 pm->ino = ino;
3114 pm->gen = ino_gen;
9f03740a
FDBM		 3115 INIT_LIST_HEAD(&pm->list);
3116 INIT_LIST_HEAD(&pm->update_refs);
3117 RB_CLEAR_NODE(&pm->node);
3118
3119 while (*p) {
3120 parent = *p;
3121 entry = rb_entry(parent, struct pending_dir_move, node);
3122 if (parent_ino < entry->parent_ino) {
3123 p = &(*p)->rb_left;
3124 } else if (parent_ino > entry->parent_ino) {
3125 p = &(*p)->rb_right;
3126 } else {
3127 exists = 1;
3128 break;
3129 }
3130 }
3131
f959492f 3132 list_for_each_entry(cur, deleted_refs, list) {
9f03740a
FDBM		 3133 ret = dup_ref(cur, &pm->update_refs);
3134 if (ret < 0)
3135 goto out;
3136 }
f959492f 3137 list_for_each_entry(cur, new_refs, list) {
9f03740a
FDBM		 3138 ret = dup_ref(cur, &pm->update_refs);
3139 if (ret < 0)
3140 goto out;
3141 }
3142
8b191a68 3143 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
9f03740a
FDBM		 3144 if (ret)
3145 goto out;
3146
3147 if (exists) {
3148 list_add_tail(&pm->list, &entry->list);
3149 } else {
3150 rb_link_node(&pm->node, parent, p);
3151 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3152 }
3153 ret = 0;
3154out:
3155 if (ret) {
3156 __free_recorded_refs(&pm->update_refs);
3157 kfree(pm);
3158 }
3159 return ret;
3160}
3161
3162static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3163 u64 parent_ino)
3164{
3165 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3166 struct pending_dir_move *entry;
3167
3168 while (n) {
3169 entry = rb_entry(n, struct pending_dir_move, node);
3170 if (parent_ino < entry->parent_ino)
3171 n = n->rb_left;
3172 else if (parent_ino > entry->parent_ino)
3173 n = n->rb_right;
3174 else
3175 return entry;
3176 }
3177 return NULL;
3178}
3179
801bec36
RK		 3180static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3181 u64 ino, u64 gen, u64 *ancestor_ino)
3182{
3183 int ret = 0;
3184 u64 parent_inode = 0;
3185 u64 parent_gen = 0;
3186 u64 start_ino = ino;
3187
3188 *ancestor_ino = 0;
3189 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3190 fs_path_reset(name);
3191
3192 if (is_waiting_for_rm(sctx, ino))
3193 break;
3194 if (is_waiting_for_move(sctx, ino)) {
3195 if (*ancestor_ino == 0)
3196 *ancestor_ino = ino;
3197 ret = get_first_ref(sctx->parent_root, ino,
3198 &parent_inode, &parent_gen, name);
3199 } else {
3200 ret = __get_cur_name_and_parent(sctx, ino, gen,
3201 &parent_inode,
3202 &parent_gen, name);
3203 if (ret > 0) {
3204 ret = 0;
3205 break;
3206 }
3207 }
3208 if (ret < 0)
3209 break;
3210 if (parent_inode == start_ino) {
3211 ret = 1;
3212 if (*ancestor_ino == 0)
3213 *ancestor_ino = ino;
3214 break;
3215 }
3216 ino = parent_inode;
3217 gen = parent_gen;
3218 }
3219 return ret;
3220}
3221
9f03740a
FDBM		 3222static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3223{
3224 struct fs_path *from_path = NULL;
3225 struct fs_path *to_path = NULL;
2b863a13 3226 struct fs_path *name = NULL;
9f03740a
FDBM		 3227 u64 orig_progress = sctx->send_progress;
3228 struct recorded_ref *cur;
2b863a13 3229 u64 parent_ino, parent_gen;
9dc44214
FM		 3230 struct waiting_dir_move *dm = NULL;
3231 u64 rmdir_ino = 0;
801bec36
RK		 3232 u64 ancestor;
3233 bool is_orphan;
9f03740a
FDBM		 3234 int ret;
3235
2b863a13 3236 name = fs_path_alloc();
9f03740a 3237 from_path = fs_path_alloc();
2b863a13
FM		 3238 if (!name || !from_path) {
3239 ret = -ENOMEM;
3240 goto out;
3241 }
9f03740a 3242
9dc44214
FM		 3243 dm = get_waiting_dir_move(sctx, pm->ino);
3244 ASSERT(dm);
3245 rmdir_ino = dm->rmdir_ino;
801bec36 3246 is_orphan = dm->orphanized;
9dc44214 3247 free_waiting_dir_move(sctx, dm);
2b863a13 3248
801bec36 3249 if (is_orphan) {
84471e24
FM		 3250 ret = gen_unique_name(sctx, pm->ino,
3251 pm->gen, from_path);
3252 } else {
3253 ret = get_first_ref(sctx->parent_root, pm->ino,
3254 &parent_ino, &parent_gen, name);
3255 if (ret < 0)
3256 goto out;
3257 ret = get_cur_path(sctx, parent_ino, parent_gen,
3258 from_path);
3259 if (ret < 0)
3260 goto out;
3261 ret = fs_path_add_path(from_path, name);
3262 }
c992ec94
FM		 3263 if (ret < 0)
3264 goto out;
2b863a13 3265
f959492f 3266 sctx->send_progress = sctx->cur_ino + 1;
801bec36 3267 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
7969e77a
FM		 3268 if (ret < 0)
3269 goto out;
801bec36
RK		 3270 if (ret) {
3271 LIST_HEAD(deleted_refs);
3272 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3273 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3274 &pm->update_refs, &deleted_refs,
3275 is_orphan);
3276 if (ret < 0)
3277 goto out;
3278 if (rmdir_ino) {
3279 dm = get_waiting_dir_move(sctx, pm->ino);
3280 ASSERT(dm);
3281 dm->rmdir_ino = rmdir_ino;
3282 }
3283 goto out;
3284 }
c992ec94
FM		 3285 fs_path_reset(name);
3286 to_path = name;
2b863a13 3287 name = NULL;
9f03740a
FDBM		 3288 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3289 if (ret < 0)
3290 goto out;
3291
3292 ret = send_rename(sctx, from_path, to_path);
3293 if (ret < 0)
3294 goto out;
3295
9dc44214
FM		 3296 if (rmdir_ino) {
3297 struct orphan_dir_info *odi;
0f96f517 3298 u64 gen;
9dc44214
FM		 3299
3300 odi = get_orphan_dir_info(sctx, rmdir_ino);
3301 if (!odi) {
3302 /* already deleted */
3303 goto finish;
3304 }
0f96f517
RK		 3305 gen = odi->gen;
3306
3307 ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino);
9dc44214
FM		 3308 if (ret < 0)
3309 goto out;
3310 if (!ret)
3311 goto finish;
3312
3313 name = fs_path_alloc();
3314 if (!name) {
3315 ret = -ENOMEM;
3316 goto out;
3317 }
0f96f517 3318 ret = get_cur_path(sctx, rmdir_ino, gen, name);
9dc44214
FM		 3319 if (ret < 0)
3320 goto out;
3321 ret = send_rmdir(sctx, name);
3322 if (ret < 0)
3323 goto out;
9dc44214
FM		 3324 }
3325
3326finish:
9f03740a
FDBM		 3327 ret = send_utimes(sctx, pm->ino, pm->gen);
3328 if (ret < 0)
3329 goto out;
3330
3331 /*
3332 * After rename/move, need to update the utimes of both new parent(s)
3333 * and old parent(s).
3334 */
3335 list_for_each_entry(cur, &pm->update_refs, list) {
764433a1
RK		 3336 /*
3337 * The parent inode might have been deleted in the send snapshot
3338 */
3339 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3340 NULL, NULL, NULL, NULL, NULL);
3341 if (ret == -ENOENT) {
3342 ret = 0;
9dc44214 3343 continue;
764433a1
RK		 3344 }
3345 if (ret < 0)
3346 goto out;
3347
9f03740a
FDBM		 3348 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3349 if (ret < 0)
3350 goto out;
3351 }
3352
3353out:
2b863a13 3354 fs_path_free(name);
9f03740a
FDBM		 3355 fs_path_free(from_path);
3356 fs_path_free(to_path);
3357 sctx->send_progress = orig_progress;
3358
3359 return ret;
3360}
3361
3362static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3363{
3364 if (!list_empty(&m->list))
3365 list_del(&m->list);
3366 if (!RB_EMPTY_NODE(&m->node))
3367 rb_erase(&m->node, &sctx->pending_dir_moves);
3368 __free_recorded_refs(&m->update_refs);
3369 kfree(m);
3370}
3371
a4390aee
RK		 3372static void tail_append_pending_moves(struct send_ctx *sctx,
3373 struct pending_dir_move *moves,
9f03740a
FDBM		 3374 struct list_head *stack)
3375{
3376 if (list_empty(&moves->list)) {
3377 list_add_tail(&moves->list, stack);
3378 } else {
3379 LIST_HEAD(list);
3380 list_splice_init(&moves->list, &list);
3381 list_add_tail(&moves->list, stack);
3382 list_splice_tail(&list, stack);
3383 }
a4390aee
RK		 3384 if (!RB_EMPTY_NODE(&moves->node)) {
3385 rb_erase(&moves->node, &sctx->pending_dir_moves);
3386 RB_CLEAR_NODE(&moves->node);
3387 }
9f03740a
FDBM		 3388}
3389
3390static int apply_children_dir_moves(struct send_ctx *sctx)
3391{
3392 struct pending_dir_move *pm;
3393 struct list_head stack;
3394 u64 parent_ino = sctx->cur_ino;
3395 int ret = 0;
3396
3397 pm = get_pending_dir_moves(sctx, parent_ino);
3398 if (!pm)
3399 return 0;
3400
3401 INIT_LIST_HEAD(&stack);
a4390aee 3402 tail_append_pending_moves(sctx, pm, &stack);
9f03740a
FDBM		 3403
3404 while (!list_empty(&stack)) {
3405 pm = list_first_entry(&stack, struct pending_dir_move, list);
3406 parent_ino = pm->ino;
3407 ret = apply_dir_move(sctx, pm);
3408 free_pending_move(sctx, pm);
3409 if (ret)
3410 goto out;
3411 pm = get_pending_dir_moves(sctx, parent_ino);
3412 if (pm)
a4390aee 3413 tail_append_pending_moves(sctx, pm, &stack);
9f03740a
FDBM		 3414 }
3415 return 0;
3416
3417out:
3418 while (!list_empty(&stack)) {
3419 pm = list_first_entry(&stack, struct pending_dir_move, list);
3420 free_pending_move(sctx, pm);
3421 }
3422 return ret;
3423}
3424
84471e24
FM		 3425/*
3426 * We might need to delay a directory rename even when no ancestor directory
3427 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3428 * renamed. This happens when we rename a directory to the old name (the name
3429 * in the parent root) of some other unrelated directory that got its rename
3430 * delayed due to some ancestor with higher number that got renamed.
3431 *
3432 * Example:
3433 *
3434 * Parent snapshot:
3435 * . (ino 256)
3436 * |---- a/ (ino 257)
3437 * | |---- file (ino 260)
3438 * |
3439 * |---- b/ (ino 258)
3440 * |---- c/ (ino 259)
3441 *
3442 * Send snapshot:
3443 * . (ino 256)
3444 * |---- a/ (ino 258)
3445 * |---- x/ (ino 259)
3446 * |---- y/ (ino 257)
3447 * |----- file (ino 260)
3448 *
3449 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3450 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3451 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3452 * must issue is:
3453 *
3454 * 1 - rename 259 from 'c' to 'x'
3455 * 2 - rename 257 from 'a' to 'x/y'
3456 * 3 - rename 258 from 'b' to 'a'
3457 *
3458 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3459 * be done right away and < 0 on error.
3460 */
3461static int wait_for_dest_dir_move(struct send_ctx *sctx,
3462 struct recorded_ref *parent_ref,
3463 const bool is_orphan)
3464{
2ff7e61e 3465 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
84471e24
FM		 3466 struct btrfs_path *path;
3467 struct btrfs_key key;
3468 struct btrfs_key di_key;
3469 struct btrfs_dir_item *di;
3470 u64 left_gen;
3471 u64 right_gen;
3472 int ret = 0;
801bec36 3473 struct waiting_dir_move *wdm;
84471e24
FM		 3474
3475 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3476 return 0;
3477
3478 path = alloc_path_for_send();
3479 if (!path)
3480 return -ENOMEM;
3481
3482 key.objectid = parent_ref->dir;
3483 key.type = BTRFS_DIR_ITEM_KEY;
3484 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3485
3486 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3487 if (ret < 0) {
3488 goto out;
3489 } else if (ret > 0) {
3490 ret = 0;
3491 goto out;
3492 }
3493
2ff7e61e
JM		 3494 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3495 parent_ref->name_len);
84471e24
FM		 3496 if (!di) {
3497 ret = 0;
3498 goto out;
3499 }
3500 /*
3501 * di_key.objectid has the number of the inode that has a dentry in the
3502 * parent directory with the same name that sctx->cur_ino is being
3503 * renamed to. We need to check if that inode is in the send root as
3504 * well and if it is currently marked as an inode with a pending rename,
3505 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3506 * that it happens after that other inode is renamed.
3507 */
3508 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3509 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3510 ret = 0;
3511 goto out;
3512 }
3513
3514 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3515 &left_gen, NULL, NULL, NULL, NULL);
3516 if (ret < 0)
3517 goto out;
3518 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3519 &right_gen, NULL, NULL, NULL, NULL);
3520 if (ret < 0) {
3521 if (ret == -ENOENT)
3522 ret = 0;
3523 goto out;
3524 }
3525
3526 /* Different inode, no need to delay the rename of sctx->cur_ino */
3527 if (right_gen != left_gen) {
3528 ret = 0;
3529 goto out;
3530 }
3531
801bec36
RK		 3532 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3533 if (wdm && !wdm->orphanized) {
84471e24
FM		 3534 ret = add_pending_dir_move(sctx,
3535 sctx->cur_ino,
3536 sctx->cur_inode_gen,
3537 di_key.objectid,
3538 &sctx->new_refs,
3539 &sctx->deleted_refs,
3540 is_orphan);
3541 if (!ret)
3542 ret = 1;
3543 }
3544out:
3545 btrfs_free_path(path);
3546 return ret;
3547}
3548
80aa6027 3549/*
ea37d599
FM		 3550 * Check if inode ino2, or any of its ancestors, is inode ino1.
3551 * Return 1 if true, 0 if false and < 0 on error.
3552 */
3553static int check_ino_in_path(struct btrfs_root *root,
3554 const u64 ino1,
3555 const u64 ino1_gen,
3556 const u64 ino2,
3557 const u64 ino2_gen,
3558 struct fs_path *fs_path)
3559{
3560 u64 ino = ino2;
3561
3562 if (ino1 == ino2)
3563 return ino1_gen == ino2_gen;
3564
3565 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3566 u64 parent;
3567 u64 parent_gen;
3568 int ret;
3569
3570 fs_path_reset(fs_path);
3571 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3572 if (ret < 0)
3573 return ret;
3574 if (parent == ino1)
3575 return parent_gen == ino1_gen;
3576 ino = parent;
3577 }
3578 return 0;
3579}
3580
3581/*
3582 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3583 * possible path (in case ino2 is not a directory and has multiple hard links).
80aa6027
FM		 3584 * Return 1 if true, 0 if false and < 0 on error.
3585 */
3586static int is_ancestor(struct btrfs_root *root,
3587 const u64 ino1,
3588 const u64 ino1_gen,
3589 const u64 ino2,
3590 struct fs_path *fs_path)
3591{
ea37d599 3592 bool free_fs_path = false;
72c3668f 3593 int ret = 0;
ea37d599
FM		 3594 struct btrfs_path *path = NULL;
3595 struct btrfs_key key;
72c3668f
FM		 3596
3597 if (!fs_path) {
3598 fs_path = fs_path_alloc();
3599 if (!fs_path)
3600 return -ENOMEM;
ea37d599 3601 free_fs_path = true;
72c3668f 3602 }
80aa6027 3603
ea37d599
FM		 3604 path = alloc_path_for_send();
3605 if (!path) {
3606 ret = -ENOMEM;
3607 goto out;
3608 }
80aa6027 3609
ea37d599
FM		 3610 key.objectid = ino2;
3611 key.type = BTRFS_INODE_REF_KEY;
3612 key.offset = 0;
3613
3614 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3615 if (ret < 0)
3616 goto out;
3617
3618 while (true) {
3619 struct extent_buffer *leaf = path->nodes[0];
3620 int slot = path->slots[0];
3621 u32 cur_offset = 0;
3622 u32 item_size;
3623
3624 if (slot >= btrfs_header_nritems(leaf)) {
3625 ret = btrfs_next_leaf(root, path);
3626 if (ret < 0)
3627 goto out;
3628 if (ret > 0)
3629 break;
3630 continue;
72c3668f 3631 }
ea37d599
FM		 3632
3633 btrfs_item_key_to_cpu(leaf, &key, slot);
3634 if (key.objectid != ino2)
3635 break;
3636 if (key.type != BTRFS_INODE_REF_KEY &&
3637 key.type != BTRFS_INODE_EXTREF_KEY)
3638 break;
3639
3640 item_size = btrfs_item_size_nr(leaf, slot);
3641 while (cur_offset < item_size) {
3642 u64 parent;
3643 u64 parent_gen;
3644
3645 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3646 unsigned long ptr;
3647 struct btrfs_inode_extref *extref;
3648
3649 ptr = btrfs_item_ptr_offset(leaf, slot);
3650 extref = (struct btrfs_inode_extref *)
3651 (ptr + cur_offset);
3652 parent = btrfs_inode_extref_parent(leaf,
3653 extref);
3654 cur_offset += sizeof(*extref);
3655 cur_offset += btrfs_inode_extref_name_len(leaf,
3656 extref);
3657 } else {
3658 parent = key.offset;
3659 cur_offset = item_size;
3660 }
3661
3662 ret = get_inode_info(root, parent, NULL, &parent_gen,
3663 NULL, NULL, NULL, NULL);
3664 if (ret < 0)
3665 goto out;
3666 ret = check_ino_in_path(root, ino1, ino1_gen,
3667 parent, parent_gen, fs_path);
3668 if (ret)
3669 goto out;
80aa6027 3670 }
ea37d599 3671 path->slots[0]++;
80aa6027 3672 }
ea37d599 3673 ret = 0;
72c3668f 3674 out:
ea37d599
FM		 3675 btrfs_free_path(path);
3676 if (free_fs_path)
72c3668f
FM		 3677 fs_path_free(fs_path);
3678 return ret;
80aa6027
FM		 3679}
3680
9f03740a 3681static int wait_for_parent_move(struct send_ctx *sctx,
8b191a68
FM		 3682 struct recorded_ref *parent_ref,
3683 const bool is_orphan)
9f03740a 3684{
f959492f 3685 int ret = 0;
9f03740a 3686 u64 ino = parent_ref->dir;
fe9c798d 3687 u64 ino_gen = parent_ref->dir_gen;
9f03740a 3688 u64 parent_ino_before, parent_ino_after;
9f03740a
FDBM		 3689 struct fs_path *path_before = NULL;
3690 struct fs_path *path_after = NULL;
3691 int len1, len2;
9f03740a
FDBM		 3692
3693 path_after = fs_path_alloc();
f959492f
FM		 3694 path_before = fs_path_alloc();
3695 if (!path_after || !path_before) {
9f03740a
FDBM		 3696 ret = -ENOMEM;
3697 goto out;
3698 }
3699
bfa7e1f8 3700 /*
f959492f
FM		 3701 * Our current directory inode may not yet be renamed/moved because some
3702 * ancestor (immediate or not) has to be renamed/moved first. So find if
3703 * such ancestor exists and make sure our own rename/move happens after
80aa6027
FM		 3704 * that ancestor is processed to avoid path build infinite loops (done
3705 * at get_cur_path()).
bfa7e1f8 3706 */
f959492f 3707 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
fe9c798d
FM		 3708 u64 parent_ino_after_gen;
3709
f959492f 3710 if (is_waiting_for_move(sctx, ino)) {
80aa6027
FM		 3711 /*
3712 * If the current inode is an ancestor of ino in the
3713 * parent root, we need to delay the rename of the
3714 * current inode, otherwise don't delayed the rename
3715 * because we can end up with a circular dependency
3716 * of renames, resulting in some directories never
3717 * getting the respective rename operations issued in
3718 * the send stream or getting into infinite path build
3719 * loops.
3720 */
3721 ret = is_ancestor(sctx->parent_root,
3722 sctx->cur_ino, sctx->cur_inode_gen,
3723 ino, path_before);
4122ea64
FM		 3724 if (ret)
3725 break;
f959492f 3726 }
bfa7e1f8
FM		 3727
3728 fs_path_reset(path_before);
3729 fs_path_reset(path_after);
3730
3731 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
fe9c798d 3732 &parent_ino_after_gen, path_after);
bfa7e1f8
FM		 3733 if (ret < 0)
3734 goto out;
3735 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3736 NULL, path_before);
f959492f 3737 if (ret < 0 && ret != -ENOENT) {
bfa7e1f8 3738 goto out;
f959492f 3739 } else if (ret == -ENOENT) {
bf8e8ca6 3740 ret = 0;
f959492f 3741 break;
bfa7e1f8
FM		 3742 }
3743
3744 len1 = fs_path_len(path_before);
3745 len2 = fs_path_len(path_after);
f959492f
FM		 3746 if (ino > sctx->cur_ino &&
3747 (parent_ino_before != parent_ino_after || len1 != len2 ||
3748 memcmp(path_before->start, path_after->start, len1))) {
fe9c798d
FM		 3749 u64 parent_ino_gen;
3750
3751 ret = get_inode_info(sctx->parent_root, ino, NULL,
3752 &parent_ino_gen, NULL, NULL, NULL,
3753 NULL);
3754 if (ret < 0)
3755 goto out;
3756 if (ino_gen == parent_ino_gen) {
3757 ret = 1;
3758 break;
3759 }
bfa7e1f8 3760 }
bfa7e1f8 3761 ino = parent_ino_after;
fe9c798d 3762 ino_gen = parent_ino_after_gen;
bfa7e1f8
FM		 3763 }
3764
9f03740a
FDBM		 3765out:
3766 fs_path_free(path_before);
3767 fs_path_free(path_after);
3768
f959492f
FM		 3769 if (ret == 1) {
3770 ret = add_pending_dir_move(sctx,
3771 sctx->cur_ino,
3772 sctx->cur_inode_gen,
3773 ino,
3774 &sctx->new_refs,
84471e24 3775 &sctx->deleted_refs,
8b191a68 3776 is_orphan);
f959492f
FM		 3777 if (!ret)
3778 ret = 1;
3779 }
3780
9f03740a
FDBM		 3781 return ret;
3782}
3783
f5962781
FM		 3784static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3785{
3786 int ret;
3787 struct fs_path *new_path;
3788
3789 /*
3790 * Our reference's name member points to its full_path member string, so
3791 * we use here a new path.
3792 */
3793 new_path = fs_path_alloc();
3794 if (!new_path)
3795 return -ENOMEM;
3796
3797 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3798 if (ret < 0) {
3799 fs_path_free(new_path);
3800 return ret;
3801 }
3802 ret = fs_path_add(new_path, ref->name, ref->name_len);
3803 if (ret < 0) {
3804 fs_path_free(new_path);
3805 return ret;
3806 }
3807
3808 fs_path_free(ref->full_path);
3809 set_ref_path(ref, new_path);
3810
3811 return 0;
3812}
3813
31db9f7c
AB		 3814/*
3815 * This does all the move/link/unlink/rmdir magic.
3816 */
9f03740a 3817static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
31db9f7c 3818{
04ab956e 3819 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 3820 int ret = 0;
3821 struct recorded_ref *cur;
1f4692da 3822 struct recorded_ref *cur2;
ba5e8f2e 3823 struct list_head check_dirs;
31db9f7c 3824 struct fs_path *valid_path = NULL;
b24baf69 3825 u64 ow_inode = 0;
31db9f7c 3826 u64 ow_gen;
f5962781 3827 u64 ow_mode;
31db9f7c
AB		 3828 int did_overwrite = 0;
3829 int is_orphan = 0;
29d6d30f 3830 u64 last_dir_ino_rm = 0;
84471e24 3831 bool can_rename = true;
f5962781 3832 bool orphanized_dir = false;
fdb13889 3833 bool orphanized_ancestor = false;
31db9f7c 3834
04ab956e 3835 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
31db9f7c 3836
6d85ed05
AB		 3837 /*
3838 * This should never happen as the root dir always has the same ref
3839 * which is always '..'
3840 */
3841 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
ba5e8f2e 3842 INIT_LIST_HEAD(&check_dirs);
6d85ed05 3843
924794c9 3844 valid_path = fs_path_alloc();
31db9f7c
AB		 3845 if (!valid_path) {
3846 ret = -ENOMEM;
3847 goto out;
3848 }
3849
31db9f7c
AB		 3850 /*
3851 * First, check if the first ref of the current inode was overwritten
3852 * before. If yes, we know that the current inode was already orphanized
3853 * and thus use the orphan name. If not, we can use get_cur_path to
3854 * get the path of the first ref as it would like while receiving at
3855 * this point in time.
3856 * New inodes are always orphan at the beginning, so force to use the
3857 * orphan name in this case.
3858 * The first ref is stored in valid_path and will be updated if it
3859 * gets moved around.
3860 */
3861 if (!sctx->cur_inode_new) {
3862 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3863 sctx->cur_inode_gen);
3864 if (ret < 0)
3865 goto out;
3866 if (ret)
3867 did_overwrite = 1;
3868 }
3869 if (sctx->cur_inode_new || did_overwrite) {
3870 ret = gen_unique_name(sctx, sctx->cur_ino,
3871 sctx->cur_inode_gen, valid_path);
3872 if (ret < 0)
3873 goto out;
3874 is_orphan = 1;
3875 } else {
3876 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3877 valid_path);
3878 if (ret < 0)
3879 goto out;
3880 }
3881
3882 list_for_each_entry(cur, &sctx->new_refs, list) {
1f4692da
AB		 3883 /*
3884 * We may have refs where the parent directory does not exist
3885 * yet. This happens if the parent directories inum is higher
52042d8e 3886 * than the current inum. To handle this case, we create the
1f4692da
AB		 3887 * parent directory out of order. But we need to check if this
3888 * did already happen before due to other refs in the same dir.
3889 */
3890 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3891 if (ret < 0)
3892 goto out;
3893 if (ret == inode_state_will_create) {
3894 ret = 0;
3895 /*
3896 * First check if any of the current inodes refs did
3897 * already create the dir.
3898 */
3899 list_for_each_entry(cur2, &sctx->new_refs, list) {
3900 if (cur == cur2)
3901 break;
3902 if (cur2->dir == cur->dir) {
3903 ret = 1;
3904 break;
3905 }
3906 }
3907
3908 /*
3909 * If that did not happen, check if a previous inode
3910 * did already create the dir.
3911 */
3912 if (!ret)
3913 ret = did_create_dir(sctx, cur->dir);
3914 if (ret < 0)
3915 goto out;
3916 if (!ret) {
3917 ret = send_create_inode(sctx, cur->dir);
3918 if (ret < 0)
3919 goto out;
3920 }
3921 }
3922
31db9f7c
AB		 3923 /*
3924 * Check if this new ref would overwrite the first ref of
3925 * another unprocessed inode. If yes, orphanize the
3926 * overwritten inode. If we find an overwritten ref that is
3927 * not the first ref, simply unlink it.
3928 */
3929 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3930 cur->name, cur->name_len,
f5962781 3931 &ow_inode, &ow_gen, &ow_mode);
31db9f7c
AB		 3932 if (ret < 0)
3933 goto out;
3934 if (ret) {
924794c9
TI		 3935 ret = is_first_ref(sctx->parent_root,
3936 ow_inode, cur->dir, cur->name,
3937 cur->name_len);
31db9f7c
AB		 3938 if (ret < 0)
3939 goto out;
3940 if (ret) {
8996a48c 3941 struct name_cache_entry *nce;
801bec36 3942 struct waiting_dir_move *wdm;
8996a48c 3943
31db9f7c
AB		 3944 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3945 cur->full_path);
3946 if (ret < 0)
3947 goto out;
f5962781
FM		 3948 if (S_ISDIR(ow_mode))
3949 orphanized_dir = true;
801bec36
RK		 3950
3951 /*
3952 * If ow_inode has its rename operation delayed
3953 * make sure that its orphanized name is used in
3954 * the source path when performing its rename
3955 * operation.
3956 */
3957 if (is_waiting_for_move(sctx, ow_inode)) {
3958 wdm = get_waiting_dir_move(sctx,
3959 ow_inode);
3960 ASSERT(wdm);
3961 wdm->orphanized = true;
3962 }
3963
8996a48c
FM		 3964 /*
3965 * Make sure we clear our orphanized inode's
3966 * name from the name cache. This is because the
3967 * inode ow_inode might be an ancestor of some
3968 * other inode that will be orphanized as well
3969 * later and has an inode number greater than
3970 * sctx->send_progress. We need to prevent
3971 * future name lookups from using the old name
3972 * and get instead the orphan name.
3973 */
3974 nce = name_cache_search(sctx, ow_inode, ow_gen);
3975 if (nce) {
3976 name_cache_delete(sctx, nce);
3977 kfree(nce);
3978 }
801bec36
RK		 3979
3980 /*
3981 * ow_inode might currently be an ancestor of
3982 * cur_ino, therefore compute valid_path (the
3983 * current path of cur_ino) again because it
3984 * might contain the pre-orphanization name of
3985 * ow_inode, which is no longer valid.
3986 */
72c3668f
FM		 3987 ret = is_ancestor(sctx->parent_root,
3988 ow_inode, ow_gen,
3989 sctx->cur_ino, NULL);
3990 if (ret > 0) {
fdb13889 3991 orphanized_ancestor = true;
72c3668f
FM		 3992 fs_path_reset(valid_path);
3993 ret = get_cur_path(sctx, sctx->cur_ino,
3994 sctx->cur_inode_gen,
3995 valid_path);
3996 }
801bec36
RK		 3997 if (ret < 0)
3998 goto out;
31db9f7c
AB		 3999 } else {
4000 ret = send_unlink(sctx, cur->full_path);
4001 if (ret < 0)
4002 goto out;
4003 }
4004 }
4005
84471e24
FM		 4006 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
4007 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
4008 if (ret < 0)
4009 goto out;
4010 if (ret == 1) {
4011 can_rename = false;
4012 *pending_move = 1;
4013 }
4014 }
4015
8b191a68
FM		 4016 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
4017 can_rename) {
4018 ret = wait_for_parent_move(sctx, cur, is_orphan);
4019 if (ret < 0)
4020 goto out;
4021 if (ret == 1) {
4022 can_rename = false;
4023 *pending_move = 1;
4024 }
4025 }
4026
31db9f7c
AB		 4027 /*
4028 * link/move the ref to the new place. If we have an orphan
4029 * inode, move it and update valid_path. If not, link or move
4030 * it depending on the inode mode.
4031 */
84471e24 4032 if (is_orphan && can_rename) {
31db9f7c
AB		 4033 ret = send_rename(sctx, valid_path, cur->full_path);
4034 if (ret < 0)
4035 goto out;
4036 is_orphan = 0;
4037 ret = fs_path_copy(valid_path, cur->full_path);
4038 if (ret < 0)
4039 goto out;
84471e24 4040 } else if (can_rename) {
31db9f7c
AB		 4041 if (S_ISDIR(sctx->cur_inode_mode)) {
4042 /*
4043 * Dirs can't be linked, so move it. For moved
4044 * dirs, we always have one new and one deleted
4045 * ref. The deleted ref is ignored later.
4046 */
8b191a68
FM		 4047 ret = send_rename(sctx, valid_path,
4048 cur->full_path);
4049 if (!ret)
4050 ret = fs_path_copy(valid_path,
4051 cur->full_path);
31db9f7c
AB		 4052 if (ret < 0)
4053 goto out;
4054 } else {
f5962781
FM		 4055 /*
4056 * We might have previously orphanized an inode
4057 * which is an ancestor of our current inode,
4058 * so our reference's full path, which was
4059 * computed before any such orphanizations, must
4060 * be updated.
4061 */
4062 if (orphanized_dir) {
4063 ret = update_ref_path(sctx, cur);
4064 if (ret < 0)
4065 goto out;
4066 }
31db9f7c
AB		 4067 ret = send_link(sctx, cur->full_path,
4068 valid_path);
4069 if (ret < 0)
4070 goto out;
4071 }
4072 }
ba5e8f2e 4073 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 4074 if (ret < 0)
4075 goto out;
4076 }
4077
4078 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4079 /*
4080 * Check if we can already rmdir the directory. If not,
4081 * orphanize it. For every dir item inside that gets deleted
4082 * later, we do this check again and rmdir it then if possible.
4083 * See the use of check_dirs for more details.
4084 */
9dc44214
FM		 4085 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4086 sctx->cur_ino);
31db9f7c
AB		 4087 if (ret < 0)
4088 goto out;
4089 if (ret) {
4090 ret = send_rmdir(sctx, valid_path);
4091 if (ret < 0)
4092 goto out;
4093 } else if (!is_orphan) {
4094 ret = orphanize_inode(sctx, sctx->cur_ino,
4095 sctx->cur_inode_gen, valid_path);
4096 if (ret < 0)
4097 goto out;
4098 is_orphan = 1;
4099 }
4100
4101 list_for_each_entry(cur, &sctx->deleted_refs, list) {
ba5e8f2e 4102 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 4103 if (ret < 0)
4104 goto out;
4105 }
ccf1626b
AB		 4106 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4107 !list_empty(&sctx->deleted_refs)) {
4108 /*
4109 * We have a moved dir. Add the old parent to check_dirs
4110 */
4111 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4112 list);
ba5e8f2e 4113 ret = dup_ref(cur, &check_dirs);
ccf1626b
AB		 4114 if (ret < 0)
4115 goto out;
31db9f7c
AB		 4116 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4117 /*
4118 * We have a non dir inode. Go through all deleted refs and
4119 * unlink them if they were not already overwritten by other
4120 * inodes.
4121 */
4122 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4123 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4124 sctx->cur_ino, sctx->cur_inode_gen,
4125 cur->name, cur->name_len);
4126 if (ret < 0)
4127 goto out;
4128 if (!ret) {
fdb13889
FM		 4129 /*
4130 * If we orphanized any ancestor before, we need
4131 * to recompute the full path for deleted names,
4132 * since any such path was computed before we
4133 * processed any references and orphanized any
4134 * ancestor inode.
4135 */
4136 if (orphanized_ancestor) {
f5962781
FM		 4137 ret = update_ref_path(sctx, cur);
4138 if (ret < 0)
fdb13889 4139 goto out;
fdb13889 4140 }
1f4692da
AB		 4141 ret = send_unlink(sctx, cur->full_path);
4142 if (ret < 0)
4143 goto out;
31db9f7c 4144 }
ba5e8f2e 4145 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 4146 if (ret < 0)
4147 goto out;
4148 }
31db9f7c
AB		 4149 /*
4150 * If the inode is still orphan, unlink the orphan. This may
4151 * happen when a previous inode did overwrite the first ref
4152 * of this inode and no new refs were added for the current
766702ef
AB		 4153 * inode. Unlinking does not mean that the inode is deleted in
4154 * all cases. There may still be links to this inode in other
4155 * places.
31db9f7c 4156 */
1f4692da 4157 if (is_orphan) {
31db9f7c
AB		 4158 ret = send_unlink(sctx, valid_path);
4159 if (ret < 0)
4160 goto out;
4161 }
4162 }
4163
4164 /*
4165 * We did collect all parent dirs where cur_inode was once located. We
4166 * now go through all these dirs and check if they are pending for
4167 * deletion and if it's finally possible to perform the rmdir now.
4168 * We also update the inode stats of the parent dirs here.
4169 */
ba5e8f2e 4170 list_for_each_entry(cur, &check_dirs, list) {
766702ef
AB		 4171 /*
4172 * In case we had refs into dirs that were not processed yet,
4173 * we don't need to do the utime and rmdir logic for these dirs.
4174 * The dir will be processed later.
4175 */
ba5e8f2e 4176 if (cur->dir > sctx->cur_ino)
31db9f7c
AB		 4177 continue;
4178
ba5e8f2e 4179 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 4180 if (ret < 0)
4181 goto out;
4182
4183 if (ret == inode_state_did_create ||
4184 ret == inode_state_no_change) {
4185 /* TODO delayed utimes */
ba5e8f2e 4186 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 4187 if (ret < 0)
4188 goto out;
29d6d30f
FM		 4189 } else if (ret == inode_state_did_delete &&
4190 cur->dir != last_dir_ino_rm) {
9dc44214
FM		 4191 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4192 sctx->cur_ino);
31db9f7c
AB		 4193 if (ret < 0)
4194 goto out;
4195 if (ret) {
ba5e8f2e
JB		 4196 ret = get_cur_path(sctx, cur->dir,
4197 cur->dir_gen, valid_path);
31db9f7c
AB		 4198 if (ret < 0)
4199 goto out;
4200 ret = send_rmdir(sctx, valid_path);
4201 if (ret < 0)
4202 goto out;
29d6d30f 4203 last_dir_ino_rm = cur->dir;
31db9f7c
AB		 4204 }
4205 }
4206 }
4207
31db9f7c
AB		 4208 ret = 0;
4209
4210out:
ba5e8f2e 4211 __free_recorded_refs(&check_dirs);
31db9f7c 4212 free_recorded_refs(sctx);
924794c9 4213 fs_path_free(valid_path);
31db9f7c
AB		 4214 return ret;
4215}
4216
a0357511
NB		 4217static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4218 void *ctx, struct list_head *refs)
31db9f7c
AB		 4219{
4220 int ret = 0;
4221 struct send_ctx *sctx = ctx;
4222 struct fs_path *p;
4223 u64 gen;
4224
924794c9 4225 p = fs_path_alloc();
31db9f7c
AB		 4226 if (!p)
4227 return -ENOMEM;
4228
a4d96d62 4229 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
85a7b33b 4230 NULL, NULL);
31db9f7c
AB		 4231 if (ret < 0)
4232 goto out;
4233
31db9f7c
AB		 4234 ret = get_cur_path(sctx, dir, gen, p);
4235 if (ret < 0)
4236 goto out;
4237 ret = fs_path_add_path(p, name);
4238 if (ret < 0)
4239 goto out;
4240
a4d96d62 4241 ret = __record_ref(refs, dir, gen, p);
31db9f7c
AB		 4242
4243out:
4244 if (ret)
924794c9 4245 fs_path_free(p);
31db9f7c
AB		 4246 return ret;
4247}
4248
a4d96d62
LB		 4249static int __record_new_ref(int num, u64 dir, int index,
4250 struct fs_path *name,
4251 void *ctx)
4252{
4253 struct send_ctx *sctx = ctx;
a0357511 4254 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
a4d96d62
LB		 4255}
4256
4257
31db9f7c
AB		 4258static int __record_deleted_ref(int num, u64 dir, int index,
4259 struct fs_path *name,
4260 void *ctx)
4261{
31db9f7c 4262 struct send_ctx *sctx = ctx;
a0357511
NB		 4263 return record_ref(sctx->parent_root, dir, name, ctx,
4264 &sctx->deleted_refs);
31db9f7c
AB		 4265}
4266
4267static int record_new_ref(struct send_ctx *sctx)
4268{
4269 int ret;
4270
924794c9
TI		 4271 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4272 sctx->cmp_key, 0, __record_new_ref, sctx);
31db9f7c
AB		 4273 if (ret < 0)
4274 goto out;
4275 ret = 0;
4276
4277out:
4278 return ret;
4279}
4280
4281static int record_deleted_ref(struct send_ctx *sctx)
4282{
4283 int ret;
4284
924794c9
TI		 4285 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4286 sctx->cmp_key, 0, __record_deleted_ref, sctx);
31db9f7c
AB		 4287 if (ret < 0)
4288 goto out;
4289 ret = 0;
4290
4291out:
4292 return ret;
4293}
4294
4295struct find_ref_ctx {
4296 u64 dir;
ba5e8f2e
JB		 4297 u64 dir_gen;
4298 struct btrfs_root *root;
31db9f7c
AB		 4299 struct fs_path *name;
4300 int found_idx;
4301};
4302
4303static int __find_iref(int num, u64 dir, int index,
4304 struct fs_path *name,
4305 void *ctx_)
4306{
4307 struct find_ref_ctx *ctx = ctx_;
ba5e8f2e
JB		 4308 u64 dir_gen;
4309 int ret;
31db9f7c
AB		 4310
4311 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4312 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
ba5e8f2e
JB		 4313 /*
4314 * To avoid doing extra lookups we'll only do this if everything
4315 * else matches.
4316 */
4317 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4318 NULL, NULL, NULL);
4319 if (ret)
4320 return ret;
4321 if (dir_gen != ctx->dir_gen)
4322 return 0;
31db9f7c
AB		 4323 ctx->found_idx = num;
4324 return 1;
4325 }
4326 return 0;
4327}
4328
924794c9 4329static int find_iref(struct btrfs_root *root,
31db9f7c
AB		 4330 struct btrfs_path *path,
4331 struct btrfs_key *key,
ba5e8f2e 4332 u64 dir, u64 dir_gen, struct fs_path *name)
31db9f7c
AB		 4333{
4334 int ret;
4335 struct find_ref_ctx ctx;
4336
4337 ctx.dir = dir;
4338 ctx.name = name;
ba5e8f2e 4339 ctx.dir_gen = dir_gen;
31db9f7c 4340 ctx.found_idx = -1;
ba5e8f2e 4341 ctx.root = root;
31db9f7c 4342
924794c9 4343 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
31db9f7c
AB		 4344 if (ret < 0)
4345 return ret;
4346
4347 if (ctx.found_idx == -1)
4348 return -ENOENT;
4349
4350 return ctx.found_idx;
4351}
4352
4353static int __record_changed_new_ref(int num, u64 dir, int index,
4354 struct fs_path *name,
4355 void *ctx)
4356{
ba5e8f2e 4357 u64 dir_gen;
31db9f7c
AB		 4358 int ret;
4359 struct send_ctx *sctx = ctx;
4360
ba5e8f2e
JB		 4361 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4362 NULL, NULL, NULL);
4363 if (ret)
4364 return ret;
4365
924794c9 4366 ret = find_iref(sctx->parent_root, sctx->right_path,
ba5e8f2e 4367 sctx->cmp_key, dir, dir_gen, name);
31db9f7c
AB		 4368 if (ret == -ENOENT)
4369 ret = __record_new_ref(num, dir, index, name, sctx);
4370 else if (ret > 0)
4371 ret = 0;
4372
4373 return ret;
4374}
4375
4376static int __record_changed_deleted_ref(int num, u64 dir, int index,
4377 struct fs_path *name,
4378 void *ctx)
4379{
ba5e8f2e 4380 u64 dir_gen;
31db9f7c
AB		 4381 int ret;
4382 struct send_ctx *sctx = ctx;
4383
ba5e8f2e
JB		 4384 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4385 NULL, NULL, NULL);
4386 if (ret)
4387 return ret;
4388
924794c9 4389 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
ba5e8f2e 4390 dir, dir_gen, name);
31db9f7c
AB		 4391 if (ret == -ENOENT)
4392 ret = __record_deleted_ref(num, dir, index, name, sctx);
4393 else if (ret > 0)
4394 ret = 0;
4395
4396 return ret;
4397}
4398
4399static int record_changed_ref(struct send_ctx *sctx)
4400{
4401 int ret = 0;
4402
924794c9 4403 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
31db9f7c
AB		 4404 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4405 if (ret < 0)
4406 goto out;
924794c9 4407 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 4408 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4409 if (ret < 0)
4410 goto out;
4411 ret = 0;
4412
4413out:
4414 return ret;
4415}
4416
4417/*
4418 * Record and process all refs at once. Needed when an inode changes the
4419 * generation number, which means that it was deleted and recreated.
4420 */
4421static int process_all_refs(struct send_ctx *sctx,
4422 enum btrfs_compare_tree_result cmd)
4423{
4424 int ret;
4425 struct btrfs_root *root;
4426 struct btrfs_path *path;
4427 struct btrfs_key key;
4428 struct btrfs_key found_key;
4429 struct extent_buffer *eb;
4430 int slot;
4431 iterate_inode_ref_t cb;
9f03740a 4432 int pending_move = 0;
31db9f7c
AB		 4433
4434 path = alloc_path_for_send();
4435 if (!path)
4436 return -ENOMEM;
4437
4438 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4439 root = sctx->send_root;
4440 cb = __record_new_ref;
4441 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4442 root = sctx->parent_root;
4443 cb = __record_deleted_ref;
4444 } else {
4d1a63b2
DS		 4445 btrfs_err(sctx->send_root->fs_info,
4446 "Wrong command %d in process_all_refs", cmd);
4447 ret = -EINVAL;
4448 goto out;
31db9f7c
AB		 4449 }
4450
4451 key.objectid = sctx->cmp_key->objectid;
4452 key.type = BTRFS_INODE_REF_KEY;
4453 key.offset = 0;
dff6d0ad
FDBM		 4454 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4455 if (ret < 0)
4456 goto out;
31db9f7c 4457
dff6d0ad 4458 while (1) {
31db9f7c
AB		 4459 eb = path->nodes[0];
4460 slot = path->slots[0];
dff6d0ad
FDBM		 4461 if (slot >= btrfs_header_nritems(eb)) {
4462 ret = btrfs_next_leaf(root, path);
4463 if (ret < 0)
4464 goto out;
4465 else if (ret > 0)
4466 break;
4467 continue;
4468 }
4469
31db9f7c
AB		 4470 btrfs_item_key_to_cpu(eb, &found_key, slot);
4471
4472 if (found_key.objectid != key.objectid ||
96b5bd77
JS		 4473 (found_key.type != BTRFS_INODE_REF_KEY &&
4474 found_key.type != BTRFS_INODE_EXTREF_KEY))
31db9f7c 4475 break;
31db9f7c 4476
924794c9 4477 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
31db9f7c
AB		 4478 if (ret < 0)
4479 goto out;
4480
dff6d0ad 4481 path->slots[0]++;
31db9f7c 4482 }
e938c8ad 4483 btrfs_release_path(path);
31db9f7c 4484
3dc09ec8
JB		 4485 /*
4486 * We don't actually care about pending_move as we are simply
4487 * re-creating this inode and will be rename'ing it into place once we
4488 * rename the parent directory.
4489 */
9f03740a 4490 ret = process_recorded_refs(sctx, &pending_move);
31db9f7c
AB		 4491out:
4492 btrfs_free_path(path);
4493 return ret;
4494}
4495
4496static int send_set_xattr(struct send_ctx *sctx,
4497 struct fs_path *path,
4498 const char *name, int name_len,
4499 const char *data, int data_len)
4500{
4501 int ret = 0;
4502
4503 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4504 if (ret < 0)
4505 goto out;
4506
4507 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4508 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4509 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4510
4511 ret = send_cmd(sctx);
4512
4513tlv_put_failure:
4514out:
4515 return ret;
4516}
4517
4518static int send_remove_xattr(struct send_ctx *sctx,
4519 struct fs_path *path,
4520 const char *name, int name_len)
4521{
4522 int ret = 0;
4523
4524 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4525 if (ret < 0)
4526 goto out;
4527
4528 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4529 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4530
4531 ret = send_cmd(sctx);
4532
4533tlv_put_failure:
4534out:
4535 return ret;
4536}
4537
4538static int __process_new_xattr(int num, struct btrfs_key *di_key,
4539 const char *name, int name_len,
4540 const char *data, int data_len,
4541 u8 type, void *ctx)
4542{
4543 int ret;
4544 struct send_ctx *sctx = ctx;
4545 struct fs_path *p;
2211d5ba 4546 struct posix_acl_xattr_header dummy_acl;
31db9f7c 4547
924794c9 4548 p = fs_path_alloc();
31db9f7c
AB		 4549 if (!p)
4550 return -ENOMEM;
4551
4552 /*
01327610 4553 * This hack is needed because empty acls are stored as zero byte
31db9f7c 4554 * data in xattrs. Problem with that is, that receiving these zero byte
01327610 4555 * acls will fail later. To fix this, we send a dummy acl list that
31db9f7c
AB		 4556 * only contains the version number and no entries.
4557 */
4558 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4559 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4560 if (data_len == 0) {
4561 dummy_acl.a_version =
4562 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4563 data = (char *)&dummy_acl;
4564 data_len = sizeof(dummy_acl);
4565 }
4566 }
4567
4568 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4569 if (ret < 0)
4570 goto out;
4571
4572 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4573
4574out:
924794c9 4575 fs_path_free(p);
31db9f7c
AB		 4576 return ret;
4577}
4578
4579static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4580 const char *name, int name_len,
4581 const char *data, int data_len,
4582 u8 type, void *ctx)
4583{
4584 int ret;
4585 struct send_ctx *sctx = ctx;
4586 struct fs_path *p;
4587
924794c9 4588 p = fs_path_alloc();
31db9f7c
AB		 4589 if (!p)
4590 return -ENOMEM;
4591
4592 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4593 if (ret < 0)
4594 goto out;
4595
4596 ret = send_remove_xattr(sctx, p, name, name_len);
4597
4598out:
924794c9 4599 fs_path_free(p);
31db9f7c
AB		 4600 return ret;
4601}
4602
4603static int process_new_xattr(struct send_ctx *sctx)
4604{
4605 int ret = 0;
4606
924794c9 4607 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
a0357511 4608 __process_new_xattr, sctx);
31db9f7c
AB		 4609
4610 return ret;
4611}
4612
4613static int process_deleted_xattr(struct send_ctx *sctx)
4614{
e2c89907 4615 return iterate_dir_item(sctx->parent_root, sctx->right_path,
a0357511 4616 __process_deleted_xattr, sctx);
31db9f7c
AB		 4617}
4618
4619struct find_xattr_ctx {
4620 const char *name;
4621 int name_len;
4622 int found_idx;
4623 char *found_data;
4624 int found_data_len;
4625};
4626
4627static int __find_xattr(int num, struct btrfs_key *di_key,
4628 const char *name, int name_len,
4629 const char *data, int data_len,
4630 u8 type, void *vctx)
4631{
4632 struct find_xattr_ctx *ctx = vctx;
4633
4634 if (name_len == ctx->name_len &&
4635 strncmp(name, ctx->name, name_len) == 0) {
4636 ctx->found_idx = num;
4637 ctx->found_data_len = data_len;
e780b0d1 4638 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
31db9f7c
AB		 4639 if (!ctx->found_data)
4640 return -ENOMEM;
31db9f7c
AB		 4641 return 1;
4642 }
4643 return 0;
4644}
4645
924794c9 4646static int find_xattr(struct btrfs_root *root,
31db9f7c
AB		 4647 struct btrfs_path *path,
4648 struct btrfs_key *key,
4649 const char *name, int name_len,
4650 char **data, int *data_len)
4651{
4652 int ret;
4653 struct find_xattr_ctx ctx;
4654
4655 ctx.name = name;
4656 ctx.name_len = name_len;
4657 ctx.found_idx = -1;
4658 ctx.found_data = NULL;
4659 ctx.found_data_len = 0;
4660
a0357511 4661 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
31db9f7c
AB		 4662 if (ret < 0)
4663 return ret;
4664
4665 if (ctx.found_idx == -1)
4666 return -ENOENT;
4667 if (data) {
4668 *data = ctx.found_data;
4669 *data_len = ctx.found_data_len;
4670 } else {
4671 kfree(ctx.found_data);
4672 }
4673 return ctx.found_idx;
4674}
4675
4676
4677static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4678 const char *name, int name_len,
4679 const char *data, int data_len,
4680 u8 type, void *ctx)
4681{
4682 int ret;
4683 struct send_ctx *sctx = ctx;
4684 char *found_data = NULL;
4685 int found_data_len = 0;
31db9f7c 4686
924794c9
TI		 4687 ret = find_xattr(sctx->parent_root, sctx->right_path,
4688 sctx->cmp_key, name, name_len, &found_data,
4689 &found_data_len);
31db9f7c
AB		 4690 if (ret == -ENOENT) {
4691 ret = __process_new_xattr(num, di_key, name, name_len, data,
4692 data_len, type, ctx);
4693 } else if (ret >= 0) {
4694 if (data_len != found_data_len ||
4695 memcmp(data, found_data, data_len)) {
4696 ret = __process_new_xattr(num, di_key, name, name_len,
4697 data, data_len, type, ctx);
4698 } else {
4699 ret = 0;
4700 }
4701 }
4702
4703 kfree(found_data);
31db9f7c
AB		 4704 return ret;
4705}
4706
4707static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4708 const char *name, int name_len,
4709 const char *data, int data_len,
4710 u8 type, void *ctx)
4711{
4712 int ret;
4713 struct send_ctx *sctx = ctx;
4714
924794c9
TI		 4715 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4716 name, name_len, NULL, NULL);
31db9f7c
AB		 4717 if (ret == -ENOENT)
4718 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4719 data_len, type, ctx);
4720 else if (ret >= 0)
4721 ret = 0;
4722
4723 return ret;
4724}
4725
4726static int process_changed_xattr(struct send_ctx *sctx)
4727{
4728 int ret = 0;
4729
924794c9 4730 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
a0357511 4731 __process_changed_new_xattr, sctx);
31db9f7c
AB		 4732 if (ret < 0)
4733 goto out;
924794c9 4734 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
a0357511 4735 __process_changed_deleted_xattr, sctx);
31db9f7c
AB		 4736
4737out:
4738 return ret;
4739}
4740
4741static int process_all_new_xattrs(struct send_ctx *sctx)
4742{
4743 int ret;
4744 struct btrfs_root *root;
4745 struct btrfs_path *path;
4746 struct btrfs_key key;
4747 struct btrfs_key found_key;
4748 struct extent_buffer *eb;
4749 int slot;
4750
4751 path = alloc_path_for_send();
4752 if (!path)
4753 return -ENOMEM;
4754
4755 root = sctx->send_root;
4756
4757 key.objectid = sctx->cmp_key->objectid;
4758 key.type = BTRFS_XATTR_ITEM_KEY;
4759 key.offset = 0;
dff6d0ad
FDBM		 4760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4761 if (ret < 0)
4762 goto out;
31db9f7c 4763
dff6d0ad 4764 while (1) {
31db9f7c
AB		 4765 eb = path->nodes[0];
4766 slot = path->slots[0];
dff6d0ad
FDBM		 4767 if (slot >= btrfs_header_nritems(eb)) {
4768 ret = btrfs_next_leaf(root, path);
4769 if (ret < 0) {
4770 goto out;
4771 } else if (ret > 0) {
4772 ret = 0;
4773 break;
4774 }
4775 continue;
4776 }
31db9f7c 4777
dff6d0ad 4778 btrfs_item_key_to_cpu(eb, &found_key, slot);
31db9f7c
AB		 4779 if (found_key.objectid != key.objectid ||
4780 found_key.type != key.type) {
4781 ret = 0;
4782 goto out;
4783 }
4784
a0357511 4785 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
31db9f7c
AB		 4786 if (ret < 0)
4787 goto out;
4788
dff6d0ad 4789 path->slots[0]++;
31db9f7c
AB		 4790 }
4791
4792out:
4793 btrfs_free_path(path);
4794 return ret;
4795}
4796
ed259095
JB		 4797static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4798{
4799 struct btrfs_root *root = sctx->send_root;
4800 struct btrfs_fs_info *fs_info = root->fs_info;
4801 struct inode *inode;
4802 struct page *page;
4803 char *addr;
4804 struct btrfs_key key;
09cbfeaf 4805 pgoff_t index = offset >> PAGE_SHIFT;
ed259095 4806 pgoff_t last_index;
7073017a 4807 unsigned pg_offset = offset_in_page(offset);
ed259095
JB		 4808 ssize_t ret = 0;
4809
4810 key.objectid = sctx->cur_ino;
4811 key.type = BTRFS_INODE_ITEM_KEY;
4812 key.offset = 0;
4813
4c66e0d4 4814 inode = btrfs_iget(fs_info->sb, &key, root);
ed259095
JB		 4815 if (IS_ERR(inode))
4816 return PTR_ERR(inode);
4817
4818 if (offset + len > i_size_read(inode)) {
4819 if (offset > i_size_read(inode))
4820 len = 0;
4821 else
4822 len = offset - i_size_read(inode);
4823 }
4824 if (len == 0)
4825 goto out;
4826
09cbfeaf 4827 last_index = (offset + len - 1) >> PAGE_SHIFT;
2131bcd3
LB		 4828
4829 /* initial readahead */
4830 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4831 file_ra_state_init(&sctx->ra, inode->i_mapping);
2131bcd3 4832
ed259095
JB		 4833 while (index <= last_index) {
4834 unsigned cur_len = min_t(unsigned, len,
09cbfeaf 4835 PAGE_SIZE - pg_offset);
eef16ba2
KH		 4836
4837 page = find_lock_page(inode->i_mapping, index);
ed259095 4838 if (!page) {
eef16ba2
KH		 4839 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4840 NULL, index, last_index + 1 - index);
4841
4842 page = find_or_create_page(inode->i_mapping, index,
4843 GFP_KERNEL);
4844 if (!page) {
4845 ret = -ENOMEM;
4846 break;
4847 }
4848 }
4849
4850 if (PageReadahead(page)) {
4851 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4852 NULL, page, index, last_index + 1 - index);
ed259095
JB		 4853 }
4854
4855 if (!PageUptodate(page)) {
4856 btrfs_readpage(NULL, page);
4857 lock_page(page);
4858 if (!PageUptodate(page)) {
4859 unlock_page(page);
09cbfeaf 4860 put_page(page);
ed259095
JB		 4861 ret = -EIO;
4862 break;
4863 }
4864 }
4865
4866 addr = kmap(page);
4867 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4868 kunmap(page);
4869 unlock_page(page);
09cbfeaf 4870 put_page(page);
ed259095
JB		 4871 index++;
4872 pg_offset = 0;
4873 len -= cur_len;
4874 ret += cur_len;
4875 }
4876out:
4877 iput(inode);
4878 return ret;
4879}
4880
31db9f7c
AB		 4881/*
4882 * Read some bytes from the current inode/file and send a write command to
4883 * user space.
4884 */
4885static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4886{
04ab956e 4887 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 4888 int ret = 0;
4889 struct fs_path *p;
ed259095 4890 ssize_t num_read = 0;
31db9f7c 4891
924794c9 4892 p = fs_path_alloc();
31db9f7c
AB		 4893 if (!p)
4894 return -ENOMEM;
4895
04ab956e 4896 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
31db9f7c 4897
ed259095
JB		 4898 num_read = fill_read_buf(sctx, offset, len);
4899 if (num_read <= 0) {
4900 if (num_read < 0)
4901 ret = num_read;
31db9f7c 4902 goto out;
ed259095 4903 }
31db9f7c
AB		 4904
4905 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4906 if (ret < 0)
4907 goto out;
4908
4909 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4910 if (ret < 0)
4911 goto out;
4912
4913 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4914 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
e938c8ad 4915 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
31db9f7c
AB		 4916
4917 ret = send_cmd(sctx);
4918
4919tlv_put_failure:
4920out:
924794c9 4921 fs_path_free(p);
31db9f7c
AB		 4922 if (ret < 0)
4923 return ret;
e938c8ad 4924 return num_read;
31db9f7c
AB		 4925}
4926
4927/*
4928 * Send a clone command to user space.
4929 */
4930static int send_clone(struct send_ctx *sctx,
4931 u64 offset, u32 len,
4932 struct clone_root *clone_root)
4933{
4934 int ret = 0;
31db9f7c
AB		 4935 struct fs_path *p;
4936 u64 gen;
4937
04ab956e
JM		 4938 btrfs_debug(sctx->send_root->fs_info,
4939 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4fd786e6
MT		 4940 offset, len, clone_root->root->root_key.objectid,
4941 clone_root->ino, clone_root->offset);
31db9f7c 4942
924794c9 4943 p = fs_path_alloc();
31db9f7c
AB		 4944 if (!p)
4945 return -ENOMEM;
4946
4947 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4948 if (ret < 0)
4949 goto out;
4950
4951 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4952 if (ret < 0)
4953 goto out;
4954
4955 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4956 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4957 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4958
e938c8ad 4959 if (clone_root->root == sctx->send_root) {
31db9f7c 4960 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
85a7b33b 4961 &gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 4962 if (ret < 0)
4963 goto out;
4964 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4965 } else {
924794c9 4966 ret = get_inode_path(clone_root->root, clone_root->ino, p);
31db9f7c
AB		 4967 }
4968 if (ret < 0)
4969 goto out;
4970
37b8d27d
JB		 4971 /*
4972 * If the parent we're using has a received_uuid set then use that as
4973 * our clone source as that is what we will look for when doing a
4974 * receive.
4975 *
4976 * This covers the case that we create a snapshot off of a received
4977 * subvolume and then use that as the parent and try to receive on a
4978 * different host.
4979 */
4980 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4981 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4982 clone_root->root->root_item.received_uuid);
4983 else
4984 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4985 clone_root->root->root_item.uuid);
31db9f7c 4986 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
5a0f4e2c 4987 le64_to_cpu(clone_root->root->root_item.ctransid));
31db9f7c
AB		 4988 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4989 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4990 clone_root->offset);
4991
4992 ret = send_cmd(sctx);
4993
4994tlv_put_failure:
4995out:
924794c9 4996 fs_path_free(p);
31db9f7c
AB		 4997 return ret;
4998}
4999
cb95e7bf
MF		 5000/*
5001 * Send an update extent command to user space.
5002 */
5003static int send_update_extent(struct send_ctx *sctx,
5004 u64 offset, u32 len)
5005{
5006 int ret = 0;
5007 struct fs_path *p;
5008
924794c9 5009 p = fs_path_alloc();
cb95e7bf
MF		 5010 if (!p)
5011 return -ENOMEM;
5012
5013 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
5014 if (ret < 0)
5015 goto out;
5016
5017 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5018 if (ret < 0)
5019 goto out;
5020
5021 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5022 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5023 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
5024
5025 ret = send_cmd(sctx);
5026
5027tlv_put_failure:
5028out:
924794c9 5029 fs_path_free(p);
cb95e7bf
MF		 5030 return ret;
5031}
5032
16e7549f
JB		 5033static int send_hole(struct send_ctx *sctx, u64 end)
5034{
5035 struct fs_path *p = NULL;
5036 u64 offset = sctx->cur_inode_last_extent;
5037 u64 len;
5038 int ret = 0;
5039
22d3151c
FM		 5040 /*
5041 * A hole that starts at EOF or beyond it. Since we do not yet support
5042 * fallocate (for extent preallocation and hole punching), sending a
5043 * write of zeroes starting at EOF or beyond would later require issuing
5044 * a truncate operation which would undo the write and achieve nothing.
5045 */
5046 if (offset >= sctx->cur_inode_size)
5047 return 0;
5048
6b1f72e5
FM		 5049 /*
5050 * Don't go beyond the inode's i_size due to prealloc extents that start
5051 * after the i_size.
5052 */
5053 end = min_t(u64, end, sctx->cur_inode_size);
5054
d4dfc0f4
FM		 5055 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5056 return send_update_extent(sctx, offset, end - offset);
5057
16e7549f
JB		 5058 p = fs_path_alloc();
5059 if (!p)
5060 return -ENOMEM;
c715e155
FM		 5061 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5062 if (ret < 0)
5063 goto tlv_put_failure;
16e7549f
JB		 5064 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5065 while (offset < end) {
5066 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5067
5068 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
16e7549f
JB		 5069 if (ret < 0)
5070 break;
5071 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5072 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5073 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5074 ret = send_cmd(sctx);
5075 if (ret < 0)
5076 break;
5077 offset += len;
5078 }
ffa7c429 5079 sctx->cur_inode_next_write_offset = offset;
16e7549f
JB		 5080tlv_put_failure:
5081 fs_path_free(p);
5082 return ret;
5083}
5084
d906d49f
FM		 5085static int send_extent_data(struct send_ctx *sctx,
5086 const u64 offset,
5087 const u64 len)
5088{
5089 u64 sent = 0;
5090
5091 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5092 return send_update_extent(sctx, offset, len);
5093
5094 while (sent < len) {
5095 u64 size = len - sent;
5096 int ret;
5097
5098 if (size > BTRFS_SEND_READ_SIZE)
5099 size = BTRFS_SEND_READ_SIZE;
5100 ret = send_write(sctx, offset + sent, size);
5101 if (ret < 0)
5102 return ret;
5103 if (!ret)
5104 break;
5105 sent += ret;
5106 }
5107 return 0;
5108}
5109
5110static int clone_range(struct send_ctx *sctx,
5111 struct clone_root *clone_root,
5112 const u64 disk_byte,
5113 u64 data_offset,
5114 u64 offset,
5115 u64 len)
5116{
5117 struct btrfs_path *path;
5118 struct btrfs_key key;
5119 int ret;
431d3988 5120 u64 clone_src_i_size = 0;
d906d49f 5121
72610b1b
FM		 5122 /*
5123 * Prevent cloning from a zero offset with a length matching the sector
5124 * size because in some scenarios this will make the receiver fail.
5125 *
5126 * For example, if in the source filesystem the extent at offset 0
5127 * has a length of sectorsize and it was written using direct IO, then
5128 * it can never be an inline extent (even if compression is enabled).
5129 * Then this extent can be cloned in the original filesystem to a non
5130 * zero file offset, but it may not be possible to clone in the
5131 * destination filesystem because it can be inlined due to compression
5132 * on the destination filesystem (as the receiver's write operations are
5133 * always done using buffered IO). The same happens when the original
5134 * filesystem does not have compression enabled but the destination
5135 * filesystem has.
5136 */
5137 if (clone_root->offset == 0 &&
5138 len == sctx->send_root->fs_info->sectorsize)
5139 return send_extent_data(sctx, offset, len);
5140
d906d49f
FM		 5141 path = alloc_path_for_send();
5142 if (!path)
5143 return -ENOMEM;
5144
040ee612
RK		 5145 /*
5146 * There are inodes that have extents that lie behind its i_size. Don't
5147 * accept clones from these extents.
5148 */
5149 ret = __get_inode_info(clone_root->root, path, clone_root->ino,
5150 &clone_src_i_size, NULL, NULL, NULL, NULL, NULL);
5151 btrfs_release_path(path);
5152 if (ret < 0)
5153 goto out;
5154
d906d49f
FM		 5155 /*
5156 * We can't send a clone operation for the entire range if we find
5157 * extent items in the respective range in the source file that
5158 * refer to different extents or if we find holes.
5159 * So check for that and do a mix of clone and regular write/copy
5160 * operations if needed.
5161 *
5162 * Example:
5163 *
5164 * mkfs.btrfs -f /dev/sda
5165 * mount /dev/sda /mnt
5166 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5167 * cp --reflink=always /mnt/foo /mnt/bar
5168 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5169 * btrfs subvolume snapshot -r /mnt /mnt/snap
5170 *
5171 * If when we send the snapshot and we are processing file bar (which
5172 * has a higher inode number than foo) we blindly send a clone operation
5173 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5174 * a file bar that matches the content of file foo - iow, doesn't match
5175 * the content from bar in the original filesystem.
5176 */
5177 key.objectid = clone_root->ino;
5178 key.type = BTRFS_EXTENT_DATA_KEY;
5179 key.offset = clone_root->offset;
5180 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5181 if (ret < 0)
5182 goto out;
5183 if (ret > 0 && path->slots[0] > 0) {
5184 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5185 if (key.objectid == clone_root->ino &&
5186 key.type == BTRFS_EXTENT_DATA_KEY)
5187 path->slots[0]--;
5188 }
5189
5190 while (true) {
5191 struct extent_buffer *leaf = path->nodes[0];
5192 int slot = path->slots[0];
5193 struct btrfs_file_extent_item *ei;
5194 u8 type;
5195 u64 ext_len;
5196 u64 clone_len;
040ee612 5197 u64 clone_data_offset;
d906d49f
FM		 5198
5199 if (slot >= btrfs_header_nritems(leaf)) {
5200 ret = btrfs_next_leaf(clone_root->root, path);
5201 if (ret < 0)
5202 goto out;
5203 else if (ret > 0)
5204 break;
5205 continue;
5206 }
5207
5208 btrfs_item_key_to_cpu(leaf, &key, slot);
5209
5210 /*
5211 * We might have an implicit trailing hole (NO_HOLES feature
5212 * enabled). We deal with it after leaving this loop.
5213 */
5214 if (key.objectid != clone_root->ino ||
5215 key.type != BTRFS_EXTENT_DATA_KEY)
5216 break;
5217
5218 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5219 type = btrfs_file_extent_type(leaf, ei);
5220 if (type == BTRFS_FILE_EXTENT_INLINE) {
e41ca589 5221 ext_len = btrfs_file_extent_ram_bytes(leaf, ei);
09cbfeaf 5222 ext_len = PAGE_ALIGN(ext_len);
d906d49f
FM		 5223 } else {
5224 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5225 }
5226
5227 if (key.offset + ext_len <= clone_root->offset)
5228 goto next;
5229
5230 if (key.offset > clone_root->offset) {
5231 /* Implicit hole, NO_HOLES feature enabled. */
5232 u64 hole_len = key.offset - clone_root->offset;
5233
5234 if (hole_len > len)
5235 hole_len = len;
5236 ret = send_extent_data(sctx, offset, hole_len);
5237 if (ret < 0)
5238 goto out;
5239
5240 len -= hole_len;
5241 if (len == 0)
5242 break;
5243 offset += hole_len;
5244 clone_root->offset += hole_len;
5245 data_offset += hole_len;
5246 }
5247
5248 if (key.offset >= clone_root->offset + len)
5249 break;
5250
040ee612
RK		 5251 if (key.offset >= clone_src_i_size)
5252 break;
5253
5254 if (key.offset + ext_len > clone_src_i_size)
5255 ext_len = clone_src_i_size - key.offset;
5256
5257 clone_data_offset = btrfs_file_extent_offset(leaf, ei);
5258 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
5259 clone_root->offset = key.offset;
5260 if (clone_data_offset < data_offset &&
5261 clone_data_offset + ext_len > data_offset) {
5262 u64 extent_offset;
5263
5264 extent_offset = data_offset - clone_data_offset;
5265 ext_len -= extent_offset;
5266 clone_data_offset += extent_offset;
5267 clone_root->offset += extent_offset;
5268 }
5269 }
5270
d906d49f
FM		 5271 clone_len = min_t(u64, ext_len, len);
5272
5273 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
3c850b45
FM		 5274 clone_data_offset == data_offset) {
5275 const u64 src_end = clone_root->offset + clone_len;
5276 const u64 sectorsize = SZ_64K;
5277
5278 /*
5279 * We can't clone the last block, when its size is not
5280 * sector size aligned, into the middle of a file. If we
5281 * do so, the receiver will get a failure (-EINVAL) when
5282 * trying to clone or will silently corrupt the data in
5283 * the destination file if it's on a kernel without the
5284 * fix introduced by commit ac765f83f1397646
5285 * ("Btrfs: fix data corruption due to cloning of eof
5286 * block).
5287 *
5288 * So issue a clone of the aligned down range plus a
5289 * regular write for the eof block, if we hit that case.
5290 *
5291 * Also, we use the maximum possible sector size, 64K,
5292 * because we don't know what's the sector size of the
5293 * filesystem that receives the stream, so we have to
5294 * assume the largest possible sector size.
5295 */
5296 if (src_end == clone_src_i_size &&
5297 !IS_ALIGNED(src_end, sectorsize) &&
5298 offset + clone_len < sctx->cur_inode_size) {
5299 u64 slen;
5300
5301 slen = ALIGN_DOWN(src_end - clone_root->offset,
5302 sectorsize);
5303 if (slen > 0) {
5304 ret = send_clone(sctx, offset, slen,
5305 clone_root);
5306 if (ret < 0)
5307 goto out;
5308 }
5309 ret = send_extent_data(sctx, offset + slen,
5310 clone_len - slen);
5311 } else {
5312 ret = send_clone(sctx, offset, clone_len,
5313 clone_root);
5314 }
5315 } else {
d906d49f 5316 ret = send_extent_data(sctx, offset, clone_len);
3c850b45 5317 }
d906d49f
FM		 5318
5319 if (ret < 0)
5320 goto out;
5321
5322 len -= clone_len;
5323 if (len == 0)
5324 break;
5325 offset += clone_len;
5326 clone_root->offset += clone_len;
5327 data_offset += clone_len;
5328next:
5329 path->slots[0]++;
5330 }
5331
5332 if (len > 0)
5333 ret = send_extent_data(sctx, offset, len);
5334 else
5335 ret = 0;
5336out:
5337 btrfs_free_path(path);
5338 return ret;
5339}
5340
31db9f7c
AB		 5341static int send_write_or_clone(struct send_ctx *sctx,
5342 struct btrfs_path *path,
5343 struct btrfs_key *key,
5344 struct clone_root *clone_root)
5345{
5346 int ret = 0;
5347 struct btrfs_file_extent_item *ei;
5348 u64 offset = key->offset;
31db9f7c 5349 u64 len;
31db9f7c 5350 u8 type;
28e5dd8f 5351 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
31db9f7c
AB		 5352
5353 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5354 struct btrfs_file_extent_item);
5355 type = btrfs_file_extent_type(path->nodes[0], ei);
74dd17fb 5356 if (type == BTRFS_FILE_EXTENT_INLINE) {
e41ca589 5357 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
74dd17fb
CM		 5358 /*
5359 * it is possible the inline item won't cover the whole page,
5360 * but there may be items after this page. Make
5361 * sure to send the whole thing
5362 */
09cbfeaf 5363 len = PAGE_ALIGN(len);
74dd17fb 5364 } else {
31db9f7c 5365 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
74dd17fb 5366 }
31db9f7c 5367
a6aa10c7
FM		 5368 if (offset >= sctx->cur_inode_size) {
5369 ret = 0;
5370 goto out;
5371 }
31db9f7c
AB		 5372 if (offset + len > sctx->cur_inode_size)
5373 len = sctx->cur_inode_size - offset;
5374 if (len == 0) {
5375 ret = 0;
5376 goto out;
5377 }
5378
28e5dd8f 5379 if (clone_root && IS_ALIGNED(offset + len, bs)) {
d906d49f
FM		 5380 u64 disk_byte;
5381 u64 data_offset;
5382
5383 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5384 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5385 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5386 offset, len);
cb95e7bf 5387 } else {
d906d49f 5388 ret = send_extent_data(sctx, offset, len);
31db9f7c 5389 }
ffa7c429 5390 sctx->cur_inode_next_write_offset = offset + len;
31db9f7c
AB		 5391out:
5392 return ret;
5393}
5394
5395static int is_extent_unchanged(struct send_ctx *sctx,
5396 struct btrfs_path *left_path,
5397 struct btrfs_key *ekey)
5398{
5399 int ret = 0;
5400 struct btrfs_key key;
5401 struct btrfs_path *path = NULL;
5402 struct extent_buffer *eb;
5403 int slot;
5404 struct btrfs_key found_key;
5405 struct btrfs_file_extent_item *ei;
5406 u64 left_disknr;
5407 u64 right_disknr;
5408 u64 left_offset;
5409 u64 right_offset;
5410 u64 left_offset_fixed;
5411 u64 left_len;
5412 u64 right_len;
74dd17fb
CM		 5413 u64 left_gen;
5414 u64 right_gen;
31db9f7c
AB		 5415 u8 left_type;
5416 u8 right_type;
5417
5418 path = alloc_path_for_send();
5419 if (!path)
5420 return -ENOMEM;
5421
5422 eb = left_path->nodes[0];
5423 slot = left_path->slots[0];
31db9f7c
AB		 5424 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5425 left_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 5426
5427 if (left_type != BTRFS_FILE_EXTENT_REG) {
5428 ret = 0;
5429 goto out;
5430 }
74dd17fb
CM		 5431 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5432 left_len = btrfs_file_extent_num_bytes(eb, ei);
5433 left_offset = btrfs_file_extent_offset(eb, ei);
5434 left_gen = btrfs_file_extent_generation(eb, ei);
31db9f7c
AB		 5435
5436 /*
5437 * Following comments will refer to these graphics. L is the left
5438 * extents which we are checking at the moment. 1-8 are the right
5439 * extents that we iterate.
5440 *
5441 * |-----L-----|
5442 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5443 *
5444 * |-----L-----|
5445 * |--1--|-2b-|...(same as above)
5446 *
5447 * Alternative situation. Happens on files where extents got split.
5448 * |-----L-----|
5449 * |-----------7-----------|-6-|
5450 *
5451 * Alternative situation. Happens on files which got larger.
5452 * |-----L-----|
5453 * |-8-|
5454 * Nothing follows after 8.
5455 */
5456
5457 key.objectid = ekey->objectid;
5458 key.type = BTRFS_EXTENT_DATA_KEY;
5459 key.offset = ekey->offset;
5460 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5461 if (ret < 0)
5462 goto out;
5463 if (ret) {
5464 ret = 0;
5465 goto out;
5466 }
5467
5468 /*
5469 * Handle special case where the right side has no extents at all.
5470 */
5471 eb = path->nodes[0];
5472 slot = path->slots[0];
5473 btrfs_item_key_to_cpu(eb, &found_key, slot);
5474 if (found_key.objectid != key.objectid ||
5475 found_key.type != key.type) {
57cfd462
JB		 5476 /* If we're a hole then just pretend nothing changed */
5477 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 5478 goto out;
5479 }
5480
5481 /*
5482 * We're now on 2a, 2b or 7.
5483 */
5484 key = found_key;
5485 while (key.offset < ekey->offset + left_len) {
5486 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5487 right_type = btrfs_file_extent_type(eb, ei);
e1cbfd7b
FM		 5488 if (right_type != BTRFS_FILE_EXTENT_REG &&
5489 right_type != BTRFS_FILE_EXTENT_INLINE) {
31db9f7c
AB		 5490 ret = 0;
5491 goto out;
5492 }
5493
e1cbfd7b 5494 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
e41ca589 5495 right_len = btrfs_file_extent_ram_bytes(eb, ei);
e1cbfd7b
FM		 5496 right_len = PAGE_ALIGN(right_len);
5497 } else {
5498 right_len = btrfs_file_extent_num_bytes(eb, ei);
5499 }
007d31f7 5500
31db9f7c
AB		 5501 /*
5502 * Are we at extent 8? If yes, we know the extent is changed.
5503 * This may only happen on the first iteration.
5504 */
d8347fa4 5505 if (found_key.offset + right_len <= ekey->offset) {
57cfd462
JB		 5506 /* If we're a hole just pretend nothing changed */
5507 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 5508 goto out;
5509 }
5510
e1cbfd7b
FM		 5511 /*
5512 * We just wanted to see if when we have an inline extent, what
5513 * follows it is a regular extent (wanted to check the above
5514 * condition for inline extents too). This should normally not
5515 * happen but it's possible for example when we have an inline
5516 * compressed extent representing data with a size matching
5517 * the page size (currently the same as sector size).
5518 */
5519 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5520 ret = 0;
5521 goto out;
5522 }
5523
24e52b11
FM		 5524 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5525 right_offset = btrfs_file_extent_offset(eb, ei);
5526 right_gen = btrfs_file_extent_generation(eb, ei);
5527
31db9f7c
AB		 5528 left_offset_fixed = left_offset;
5529 if (key.offset < ekey->offset) {
5530 /* Fix the right offset for 2a and 7. */
5531 right_offset += ekey->offset - key.offset;
5532 } else {
5533 /* Fix the left offset for all behind 2a and 2b */
5534 left_offset_fixed += key.offset - ekey->offset;
5535 }
5536
5537 /*
5538 * Check if we have the same extent.
5539 */
3954096d 5540 if (left_disknr != right_disknr ||
74dd17fb
CM		 5541 left_offset_fixed != right_offset ||
5542 left_gen != right_gen) {
31db9f7c
AB		 5543 ret = 0;
5544 goto out;
5545 }
5546
5547 /*
5548 * Go to the next extent.
5549 */
5550 ret = btrfs_next_item(sctx->parent_root, path);
5551 if (ret < 0)
5552 goto out;
5553 if (!ret) {
5554 eb = path->nodes[0];
5555 slot = path->slots[0];
5556 btrfs_item_key_to_cpu(eb, &found_key, slot);
5557 }
5558 if (ret || found_key.objectid != key.objectid ||
5559 found_key.type != key.type) {
5560 key.offset += right_len;
5561 break;
adaa4b8e
JS		 5562 }
5563 if (found_key.offset != key.offset + right_len) {
5564 ret = 0;
5565 goto out;
31db9f7c
AB		 5566 }
5567 key = found_key;
5568 }
5569
5570 /*
5571 * We're now behind the left extent (treat as unchanged) or at the end
5572 * of the right side (treat as changed).
5573 */
5574 if (key.offset >= ekey->offset + left_len)
5575 ret = 1;
5576 else
5577 ret = 0;
5578
5579
5580out:
5581 btrfs_free_path(path);
5582 return ret;
5583}
5584
16e7549f
JB		 5585static int get_last_extent(struct send_ctx *sctx, u64 offset)
5586{
5587 struct btrfs_path *path;
5588 struct btrfs_root *root = sctx->send_root;
16e7549f 5589 struct btrfs_key key;
16e7549f
JB		 5590 int ret;
5591
5592 path = alloc_path_for_send();
5593 if (!path)
5594 return -ENOMEM;
5595
5596 sctx->cur_inode_last_extent = 0;
5597
5598 key.objectid = sctx->cur_ino;
5599 key.type = BTRFS_EXTENT_DATA_KEY;
5600 key.offset = offset;
5601 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5602 if (ret < 0)
5603 goto out;
5604 ret = 0;
5605 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5606 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5607 goto out;
5608
a5eeb3d1 5609 sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
16e7549f
JB		 5610out:
5611 btrfs_free_path(path);
5612 return ret;
5613}
5614
82bfb2e7
FM		 5615static int range_is_hole_in_parent(struct send_ctx *sctx,
5616 const u64 start,
5617 const u64 end)
5618{
5619 struct btrfs_path *path;
5620 struct btrfs_key key;
5621 struct btrfs_root *root = sctx->parent_root;
5622 u64 search_start = start;
5623 int ret;
5624
5625 path = alloc_path_for_send();
5626 if (!path)
5627 return -ENOMEM;
5628
5629 key.objectid = sctx->cur_ino;
5630 key.type = BTRFS_EXTENT_DATA_KEY;
5631 key.offset = search_start;
5632 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5633 if (ret < 0)
5634 goto out;
5635 if (ret > 0 && path->slots[0] > 0)
5636 path->slots[0]--;
5637
5638 while (search_start < end) {
5639 struct extent_buffer *leaf = path->nodes[0];
5640 int slot = path->slots[0];
5641 struct btrfs_file_extent_item *fi;
5642 u64 extent_end;
5643
5644 if (slot >= btrfs_header_nritems(leaf)) {
5645 ret = btrfs_next_leaf(root, path);
5646 if (ret < 0)
5647 goto out;
5648 else if (ret > 0)
5649 break;
5650 continue;
5651 }
5652
5653 btrfs_item_key_to_cpu(leaf, &key, slot);
5654 if (key.objectid < sctx->cur_ino ||
5655 key.type < BTRFS_EXTENT_DATA_KEY)
5656 goto next;
5657 if (key.objectid > sctx->cur_ino ||
5658 key.type > BTRFS_EXTENT_DATA_KEY ||
5659 key.offset >= end)
5660 break;
5661
5662 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
a5eeb3d1 5663 extent_end = btrfs_file_extent_end(path);
82bfb2e7
FM		 5664 if (extent_end <= start)
5665 goto next;
5666 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5667 search_start = extent_end;
5668 goto next;
5669 }
5670 ret = 0;
5671 goto out;
5672next:
5673 path->slots[0]++;
5674 }
5675 ret = 1;
5676out:
5677 btrfs_free_path(path);
5678 return ret;
5679}
5680
16e7549f
JB		 5681static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5682 struct btrfs_key *key)
5683{
16e7549f
JB		 5684 int ret = 0;
5685
5686 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5687 return 0;
5688
5689 if (sctx->cur_inode_last_extent == (u64)-1) {
5690 ret = get_last_extent(sctx, key->offset - 1);
5691 if (ret)
5692 return ret;
5693 }
5694
bf54f412
FDBM		 5695 if (path->slots[0] == 0 &&
5696 sctx->cur_inode_last_extent < key->offset) {
5697 /*
5698 * We might have skipped entire leafs that contained only
5699 * file extent items for our current inode. These leafs have
5700 * a generation number smaller (older) than the one in the
5701 * current leaf and the leaf our last extent came from, and
5702 * are located between these 2 leafs.
5703 */
5704 ret = get_last_extent(sctx, key->offset - 1);
5705 if (ret)
5706 return ret;
5707 }
5708
82bfb2e7
FM		 5709 if (sctx->cur_inode_last_extent < key->offset) {
5710 ret = range_is_hole_in_parent(sctx,
5711 sctx->cur_inode_last_extent,
5712 key->offset);
5713 if (ret < 0)
5714 return ret;
5715 else if (ret == 0)
5716 ret = send_hole(sctx, key->offset);
5717 else
5718 ret = 0;
5719 }
a5eeb3d1 5720 sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
16e7549f
JB		 5721 return ret;
5722}
5723
31db9f7c
AB		 5724static int process_extent(struct send_ctx *sctx,
5725 struct btrfs_path *path,
5726 struct btrfs_key *key)
5727{
31db9f7c 5728 struct clone_root *found_clone = NULL;
57cfd462 5729 int ret = 0;
31db9f7c
AB		 5730
5731 if (S_ISLNK(sctx->cur_inode_mode))
5732 return 0;
5733
5734 if (sctx->parent_root && !sctx->cur_inode_new) {
5735 ret = is_extent_unchanged(sctx, path, key);
5736 if (ret < 0)
5737 goto out;
5738 if (ret) {
5739 ret = 0;
16e7549f 5740 goto out_hole;
31db9f7c 5741 }
57cfd462
JB		 5742 } else {
5743 struct btrfs_file_extent_item *ei;
5744 u8 type;
5745
5746 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5747 struct btrfs_file_extent_item);
5748 type = btrfs_file_extent_type(path->nodes[0], ei);
5749 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5750 type == BTRFS_FILE_EXTENT_REG) {
5751 /*
5752 * The send spec does not have a prealloc command yet,
5753 * so just leave a hole for prealloc'ed extents until
5754 * we have enough commands queued up to justify rev'ing
5755 * the send spec.
5756 */
5757 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5758 ret = 0;
5759 goto out;
5760 }
5761
5762 /* Have a hole, just skip it. */
5763 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5764 ret = 0;
5765 goto out;
5766 }
5767 }
31db9f7c
AB		 5768 }
5769
5770 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5771 sctx->cur_inode_size, &found_clone);
5772 if (ret != -ENOENT && ret < 0)
5773 goto out;
5774
5775 ret = send_write_or_clone(sctx, path, key, found_clone);
16e7549f
JB		 5776 if (ret)
5777 goto out;
5778out_hole:
5779 ret = maybe_send_hole(sctx, path, key);
31db9f7c
AB		 5780out:
5781 return ret;
5782}
5783
5784static int process_all_extents(struct send_ctx *sctx)
5785{
5786 int ret;
5787 struct btrfs_root *root;
5788 struct btrfs_path *path;
5789 struct btrfs_key key;
5790 struct btrfs_key found_key;
5791 struct extent_buffer *eb;
5792 int slot;
5793
5794 root = sctx->send_root;
5795 path = alloc_path_for_send();
5796 if (!path)
5797 return -ENOMEM;
5798
5799 key.objectid = sctx->cmp_key->objectid;
5800 key.type = BTRFS_EXTENT_DATA_KEY;
5801 key.offset = 0;
7fdd29d0
FDBM		 5802 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5803 if (ret < 0)
5804 goto out;
31db9f7c 5805
7fdd29d0 5806 while (1) {
31db9f7c
AB		 5807 eb = path->nodes[0];
5808 slot = path->slots[0];
7fdd29d0
FDBM		 5809
5810 if (slot >= btrfs_header_nritems(eb)) {
5811 ret = btrfs_next_leaf(root, path);
5812 if (ret < 0) {
5813 goto out;
5814 } else if (ret > 0) {
5815 ret = 0;
5816 break;
5817 }
5818 continue;
5819 }
5820
31db9f7c
AB		 5821 btrfs_item_key_to_cpu(eb, &found_key, slot);
5822
5823 if (found_key.objectid != key.objectid ||
5824 found_key.type != key.type) {
5825 ret = 0;
5826 goto out;
5827 }
5828
5829 ret = process_extent(sctx, path, &found_key);
5830 if (ret < 0)
5831 goto out;
5832
7fdd29d0 5833 path->slots[0]++;
31db9f7c
AB		 5834 }
5835
5836out:
5837 btrfs_free_path(path);
5838 return ret;
5839}
5840
9f03740a
FDBM		 5841static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5842 int *pending_move,
5843 int *refs_processed)
31db9f7c
AB		 5844{
5845 int ret = 0;
5846
5847 if (sctx->cur_ino == 0)
5848 goto out;
5849 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
96b5bd77 5850 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 5851 goto out;
5852 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5853 goto out;
5854
9f03740a 5855 ret = process_recorded_refs(sctx, pending_move);
e479d9bb
AB		 5856 if (ret < 0)
5857 goto out;
5858
9f03740a 5859 *refs_processed = 1;
31db9f7c
AB		 5860out:
5861 return ret;
5862}
5863
5864static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5865{
5866 int ret = 0;
5867 u64 left_mode;
5868 u64 left_uid;
5869 u64 left_gid;
5870 u64 right_mode;
5871 u64 right_uid;
5872 u64 right_gid;
5873 int need_chmod = 0;
5874 int need_chown = 0;
ffa7c429 5875 int need_truncate = 1;
9f03740a
FDBM		 5876 int pending_move = 0;
5877 int refs_processed = 0;
31db9f7c 5878
46b2f459
FM		 5879 if (sctx->ignore_cur_inode)
5880 return 0;
5881
9f03740a
FDBM		 5882 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5883 &refs_processed);
31db9f7c
AB		 5884 if (ret < 0)
5885 goto out;
5886
9f03740a
FDBM		 5887 /*
5888 * We have processed the refs and thus need to advance send_progress.
5889 * Now, calls to get_cur_xxx will take the updated refs of the current
5890 * inode into account.
5891 *
5892 * On the other hand, if our current inode is a directory and couldn't
5893 * be moved/renamed because its parent was renamed/moved too and it has
5894 * a higher inode number, we can only move/rename our current inode
5895 * after we moved/renamed its parent. Therefore in this case operate on
5896 * the old path (pre move/rename) of our current inode, and the
5897 * move/rename will be performed later.
5898 */
5899 if (refs_processed && !pending_move)
5900 sctx->send_progress = sctx->cur_ino + 1;
5901
31db9f7c
AB		 5902 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5903 goto out;
5904 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5905 goto out;
5906
5907 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
85a7b33b 5908 &left_mode, &left_uid, &left_gid, NULL);
31db9f7c
AB		 5909 if (ret < 0)
5910 goto out;
5911
e2d044fe
AL		 5912 if (!sctx->parent_root || sctx->cur_inode_new) {
5913 need_chown = 1;
5914 if (!S_ISLNK(sctx->cur_inode_mode))
31db9f7c 5915 need_chmod = 1;
ffa7c429
FM		 5916 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5917 need_truncate = 0;
e2d044fe 5918 } else {
ffa7c429
FM		 5919 u64 old_size;
5920
e2d044fe 5921 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
ffa7c429 5922 &old_size, NULL, &right_mode, &right_uid,
e2d044fe
AL		 5923 &right_gid, NULL);
5924 if (ret < 0)
5925 goto out;
31db9f7c 5926
e2d044fe
AL		 5927 if (left_uid != right_uid || left_gid != right_gid)
5928 need_chown = 1;
5929 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5930 need_chmod = 1;
ffa7c429
FM		 5931 if ((old_size == sctx->cur_inode_size) ||
5932 (sctx->cur_inode_size > old_size &&
5933 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5934 need_truncate = 0;
31db9f7c
AB		 5935 }
5936
5937 if (S_ISREG(sctx->cur_inode_mode)) {
16e7549f 5938 if (need_send_hole(sctx)) {
766b5e5a
FM		 5939 if (sctx->cur_inode_last_extent == (u64)-1 ||
5940 sctx->cur_inode_last_extent <
5941 sctx->cur_inode_size) {
16e7549f
JB		 5942 ret = get_last_extent(sctx, (u64)-1);
5943 if (ret)
5944 goto out;
5945 }
5946 if (sctx->cur_inode_last_extent <
5947 sctx->cur_inode_size) {
5948 ret = send_hole(sctx, sctx->cur_inode_size);
5949 if (ret)
5950 goto out;
5951 }
5952 }
ffa7c429
FM		 5953 if (need_truncate) {
5954 ret = send_truncate(sctx, sctx->cur_ino,
5955 sctx->cur_inode_gen,
5956 sctx->cur_inode_size);
5957 if (ret < 0)
5958 goto out;
5959 }
31db9f7c
AB		 5960 }
5961
5962 if (need_chown) {
5963 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5964 left_uid, left_gid);
5965 if (ret < 0)
5966 goto out;
5967 }
5968 if (need_chmod) {
5969 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5970 left_mode);
5971 if (ret < 0)
5972 goto out;
5973 }
5974
5975 /*
9f03740a
FDBM		 5976 * If other directory inodes depended on our current directory
5977 * inode's move/rename, now do their move/rename operations.
31db9f7c 5978 */
9f03740a
FDBM		 5979 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5980 ret = apply_children_dir_moves(sctx);
5981 if (ret)
5982 goto out;
fcbd2154
FM		 5983 /*
5984 * Need to send that every time, no matter if it actually
5985 * changed between the two trees as we have done changes to
5986 * the inode before. If our inode is a directory and it's
5987 * waiting to be moved/renamed, we will send its utimes when
5988 * it's moved/renamed, therefore we don't need to do it here.
5989 */
5990 sctx->send_progress = sctx->cur_ino + 1;
5991 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5992 if (ret < 0)
5993 goto out;
9f03740a
FDBM		 5994 }
5995
31db9f7c
AB		 5996out:
5997 return ret;
5998}
5999
46b2f459
FM		 6000struct parent_paths_ctx {
6001 struct list_head *refs;
6002 struct send_ctx *sctx;
6003};
6004
6005static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name,
6006 void *ctx)
6007{
6008 struct parent_paths_ctx *ppctx = ctx;
6009
6010 return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx,
6011 ppctx->refs);
6012}
6013
6014/*
6015 * Issue unlink operations for all paths of the current inode found in the
6016 * parent snapshot.
6017 */
6018static int btrfs_unlink_all_paths(struct send_ctx *sctx)
6019{
6020 LIST_HEAD(deleted_refs);
6021 struct btrfs_path *path;
6022 struct btrfs_key key;
6023 struct parent_paths_ctx ctx;
6024 int ret;
6025
6026 path = alloc_path_for_send();
6027 if (!path)
6028 return -ENOMEM;
6029
6030 key.objectid = sctx->cur_ino;
6031 key.type = BTRFS_INODE_REF_KEY;
6032 key.offset = 0;
6033 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
6034 if (ret < 0)
6035 goto out;
6036
6037 ctx.refs = &deleted_refs;
6038 ctx.sctx = sctx;
6039
6040 while (true) {
6041 struct extent_buffer *eb = path->nodes[0];
6042 int slot = path->slots[0];
6043
6044 if (slot >= btrfs_header_nritems(eb)) {
6045 ret = btrfs_next_leaf(sctx->parent_root, path);
6046 if (ret < 0)
6047 goto out;
6048 else if (ret > 0)
6049 break;
6050 continue;
6051 }
6052
6053 btrfs_item_key_to_cpu(eb, &key, slot);
6054 if (key.objectid != sctx->cur_ino)
6055 break;
6056 if (key.type != BTRFS_INODE_REF_KEY &&
6057 key.type != BTRFS_INODE_EXTREF_KEY)
6058 break;
6059
6060 ret = iterate_inode_ref(sctx->parent_root, path, &key, 1,
6061 record_parent_ref, &ctx);
6062 if (ret < 0)
6063 goto out;
6064
6065 path->slots[0]++;
6066 }
6067
6068 while (!list_empty(&deleted_refs)) {
6069 struct recorded_ref *ref;
6070
6071 ref = list_first_entry(&deleted_refs, struct recorded_ref, list);
6072 ret = send_unlink(sctx, ref->full_path);
6073 if (ret < 0)
6074 goto out;
6075 fs_path_free(ref->full_path);
6076 list_del(&ref->list);
6077 kfree(ref);
6078 }
6079 ret = 0;
6080out:
6081 btrfs_free_path(path);
6082 if (ret)
6083 __free_recorded_refs(&deleted_refs);
6084 return ret;
6085}
6086
31db9f7c
AB		 6087static int changed_inode(struct send_ctx *sctx,
6088 enum btrfs_compare_tree_result result)
6089{
6090 int ret = 0;
6091 struct btrfs_key *key = sctx->cmp_key;
6092 struct btrfs_inode_item *left_ii = NULL;
6093 struct btrfs_inode_item *right_ii = NULL;
6094 u64 left_gen = 0;
6095 u64 right_gen = 0;
6096
31db9f7c
AB		 6097 sctx->cur_ino = key->objectid;
6098 sctx->cur_inode_new_gen = 0;
16e7549f 6099 sctx->cur_inode_last_extent = (u64)-1;
ffa7c429 6100 sctx->cur_inode_next_write_offset = 0;
46b2f459 6101 sctx->ignore_cur_inode = false;
e479d9bb
AB		 6102
6103 /*
6104 * Set send_progress to current inode. This will tell all get_cur_xxx
6105 * functions that the current inode's refs are not updated yet. Later,
6106 * when process_recorded_refs is finished, it is set to cur_ino + 1.
6107 */
31db9f7c
AB		 6108 sctx->send_progress = sctx->cur_ino;
6109
6110 if (result == BTRFS_COMPARE_TREE_NEW ||
6111 result == BTRFS_COMPARE_TREE_CHANGED) {
6112 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6113 sctx->left_path->slots[0],
6114 struct btrfs_inode_item);
6115 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6116 left_ii);
6117 } else {
6118 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6119 sctx->right_path->slots[0],
6120 struct btrfs_inode_item);
6121 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6122 right_ii);
6123 }
6124 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6125 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6126 sctx->right_path->slots[0],
6127 struct btrfs_inode_item);
6128
6129 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6130 right_ii);
6d85ed05
AB		 6131
6132 /*
6133 * The cur_ino = root dir case is special here. We can't treat
6134 * the inode as deleted+reused because it would generate a
6135 * stream that tries to delete/mkdir the root dir.
6136 */
6137 if (left_gen != right_gen &&
6138 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
31db9f7c
AB		 6139 sctx->cur_inode_new_gen = 1;
6140 }
6141
46b2f459
FM		 6142 /*
6143 * Normally we do not find inodes with a link count of zero (orphans)
6144 * because the most common case is to create a snapshot and use it
6145 * for a send operation. However other less common use cases involve
6146 * using a subvolume and send it after turning it to RO mode just
6147 * after deleting all hard links of a file while holding an open
6148 * file descriptor against it or turning a RO snapshot into RW mode,
6149 * keep an open file descriptor against a file, delete it and then
6150 * turn the snapshot back to RO mode before using it for a send
6151 * operation. So if we find such cases, ignore the inode and all its
6152 * items completely if it's a new inode, or if it's a changed inode
6153 * make sure all its previous paths (from the parent snapshot) are all
6154 * unlinked and all other the inode items are ignored.
6155 */
6156 if (result == BTRFS_COMPARE_TREE_NEW ||
6157 result == BTRFS_COMPARE_TREE_CHANGED) {
6158 u32 nlinks;
6159
6160 nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6161 if (nlinks == 0) {
6162 sctx->ignore_cur_inode = true;
6163 if (result == BTRFS_COMPARE_TREE_CHANGED)
6164 ret = btrfs_unlink_all_paths(sctx);
6165 goto out;
6166 }
6167 }
6168
31db9f7c
AB		 6169 if (result == BTRFS_COMPARE_TREE_NEW) {
6170 sctx->cur_inode_gen = left_gen;
6171 sctx->cur_inode_new = 1;
6172 sctx->cur_inode_deleted = 0;
6173 sctx->cur_inode_size = btrfs_inode_size(
6174 sctx->left_path->nodes[0], left_ii);
6175 sctx->cur_inode_mode = btrfs_inode_mode(
6176 sctx->left_path->nodes[0], left_ii);
644d1940
LB		 6177 sctx->cur_inode_rdev = btrfs_inode_rdev(
6178 sctx->left_path->nodes[0], left_ii);
31db9f7c 6179 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
1f4692da 6180 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 6181 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6182 sctx->cur_inode_gen = right_gen;
6183 sctx->cur_inode_new = 0;
6184 sctx->cur_inode_deleted = 1;
6185 sctx->cur_inode_size = btrfs_inode_size(
6186 sctx->right_path->nodes[0], right_ii);
6187 sctx->cur_inode_mode = btrfs_inode_mode(
6188 sctx->right_path->nodes[0], right_ii);
6189 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
766702ef
AB		 6190 /*
6191 * We need to do some special handling in case the inode was
6192 * reported as changed with a changed generation number. This
6193 * means that the original inode was deleted and new inode
6194 * reused the same inum. So we have to treat the old inode as
6195 * deleted and the new one as new.
6196 */
31db9f7c 6197 if (sctx->cur_inode_new_gen) {
766702ef
AB		 6198 /*
6199 * First, process the inode as if it was deleted.
6200 */
31db9f7c
AB		 6201 sctx->cur_inode_gen = right_gen;
6202 sctx->cur_inode_new = 0;
6203 sctx->cur_inode_deleted = 1;
6204 sctx->cur_inode_size = btrfs_inode_size(
6205 sctx->right_path->nodes[0], right_ii);
6206 sctx->cur_inode_mode = btrfs_inode_mode(
6207 sctx->right_path->nodes[0], right_ii);
6208 ret = process_all_refs(sctx,
6209 BTRFS_COMPARE_TREE_DELETED);
6210 if (ret < 0)
6211 goto out;
6212
766702ef
AB		 6213 /*
6214 * Now process the inode as if it was new.
6215 */
31db9f7c
AB		 6216 sctx->cur_inode_gen = left_gen;
6217 sctx->cur_inode_new = 1;
6218 sctx->cur_inode_deleted = 0;
6219 sctx->cur_inode_size = btrfs_inode_size(
6220 sctx->left_path->nodes[0], left_ii);
6221 sctx->cur_inode_mode = btrfs_inode_mode(
6222 sctx->left_path->nodes[0], left_ii);
644d1940
LB		 6223 sctx->cur_inode_rdev = btrfs_inode_rdev(
6224 sctx->left_path->nodes[0], left_ii);
1f4692da 6225 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 6226 if (ret < 0)
6227 goto out;
6228
6229 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6230 if (ret < 0)
6231 goto out;
e479d9bb
AB		 6232 /*
6233 * Advance send_progress now as we did not get into
6234 * process_recorded_refs_if_needed in the new_gen case.
6235 */
6236 sctx->send_progress = sctx->cur_ino + 1;
766702ef
AB		 6237
6238 /*
6239 * Now process all extents and xattrs of the inode as if
6240 * they were all new.
6241 */
31db9f7c
AB		 6242 ret = process_all_extents(sctx);
6243 if (ret < 0)
6244 goto out;
6245 ret = process_all_new_xattrs(sctx);
6246 if (ret < 0)
6247 goto out;
6248 } else {
6249 sctx->cur_inode_gen = left_gen;
6250 sctx->cur_inode_new = 0;
6251 sctx->cur_inode_new_gen = 0;
6252 sctx->cur_inode_deleted = 0;
6253 sctx->cur_inode_size = btrfs_inode_size(
6254 sctx->left_path->nodes[0], left_ii);
6255 sctx->cur_inode_mode = btrfs_inode_mode(
6256 sctx->left_path->nodes[0], left_ii);
6257 }
6258 }
6259
6260out:
6261 return ret;
6262}
6263
766702ef
AB		 6264/*
6265 * We have to process new refs before deleted refs, but compare_trees gives us
6266 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6267 * first and later process them in process_recorded_refs.
6268 * For the cur_inode_new_gen case, we skip recording completely because
6269 * changed_inode did already initiate processing of refs. The reason for this is
6270 * that in this case, compare_tree actually compares the refs of 2 different
6271 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6272 * refs of the right tree as deleted and all refs of the left tree as new.
6273 */
31db9f7c
AB		 6274static int changed_ref(struct send_ctx *sctx,
6275 enum btrfs_compare_tree_result result)
6276{
6277 int ret = 0;
6278
95155585
FM		 6279 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6280 inconsistent_snapshot_error(sctx, result, "reference");
6281 return -EIO;
6282 }
31db9f7c
AB		 6283
6284 if (!sctx->cur_inode_new_gen &&
6285 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6286 if (result == BTRFS_COMPARE_TREE_NEW)
6287 ret = record_new_ref(sctx);
6288 else if (result == BTRFS_COMPARE_TREE_DELETED)
6289 ret = record_deleted_ref(sctx);
6290 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6291 ret = record_changed_ref(sctx);
6292 }
6293
6294 return ret;
6295}
6296
766702ef
AB		 6297/*
6298 * Process new/deleted/changed xattrs. We skip processing in the
6299 * cur_inode_new_gen case because changed_inode did already initiate processing
6300 * of xattrs. The reason is the same as in changed_ref
6301 */
31db9f7c
AB		 6302static int changed_xattr(struct send_ctx *sctx,
6303 enum btrfs_compare_tree_result result)
6304{
6305 int ret = 0;
6306
95155585
FM		 6307 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6308 inconsistent_snapshot_error(sctx, result, "xattr");
6309 return -EIO;
6310 }
31db9f7c
AB		 6311
6312 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6313 if (result == BTRFS_COMPARE_TREE_NEW)
6314 ret = process_new_xattr(sctx);
6315 else if (result == BTRFS_COMPARE_TREE_DELETED)
6316 ret = process_deleted_xattr(sctx);
6317 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6318 ret = process_changed_xattr(sctx);
6319 }
6320
6321 return ret;
6322}
6323
766702ef
AB		 6324/*
6325 * Process new/deleted/changed extents. We skip processing in the
6326 * cur_inode_new_gen case because changed_inode did already initiate processing
6327 * of extents. The reason is the same as in changed_ref
6328 */
31db9f7c
AB		 6329static int changed_extent(struct send_ctx *sctx,
6330 enum btrfs_compare_tree_result result)
6331{
6332 int ret = 0;
6333
b4f9a1a8
FM		 6334 /*
6335 * We have found an extent item that changed without the inode item
6336 * having changed. This can happen either after relocation (where the
6337 * disk_bytenr of an extent item is replaced at
6338 * relocation.c:replace_file_extents()) or after deduplication into a
6339 * file in both the parent and send snapshots (where an extent item can
6340 * get modified or replaced with a new one). Note that deduplication
6341 * updates the inode item, but it only changes the iversion (sequence
6342 * field in the inode item) of the inode, so if a file is deduplicated
6343 * the same amount of times in both the parent and send snapshots, its
6344 * iversion becames the same in both snapshots, whence the inode item is
6345 * the same on both snapshots.
6346 */
6347 if (sctx->cur_ino != sctx->cmp_key->objectid)
6348 return 0;
31db9f7c
AB		 6349
6350 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6351 if (result != BTRFS_COMPARE_TREE_DELETED)
6352 ret = process_extent(sctx, sctx->left_path,
6353 sctx->cmp_key);
6354 }
6355
6356 return ret;
6357}
6358
ba5e8f2e
JB		 6359static int dir_changed(struct send_ctx *sctx, u64 dir)
6360{
6361 u64 orig_gen, new_gen;
6362 int ret;
6363
6364 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6365 NULL, NULL);
6366 if (ret)
6367 return ret;
6368
6369 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6370 NULL, NULL, NULL);
6371 if (ret)
6372 return ret;
6373
6374 return (orig_gen != new_gen) ? 1 : 0;
6375}
6376
6377static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6378 struct btrfs_key *key)
6379{
6380 struct btrfs_inode_extref *extref;
6381 struct extent_buffer *leaf;
6382 u64 dirid = 0, last_dirid = 0;
6383 unsigned long ptr;
6384 u32 item_size;
6385 u32 cur_offset = 0;
6386 int ref_name_len;
6387 int ret = 0;
6388
6389 /* Easy case, just check this one dirid */
6390 if (key->type == BTRFS_INODE_REF_KEY) {
6391 dirid = key->offset;
6392
6393 ret = dir_changed(sctx, dirid);
6394 goto out;
6395 }
6396
6397 leaf = path->nodes[0];
6398 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6399 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6400 while (cur_offset < item_size) {
6401 extref = (struct btrfs_inode_extref *)(ptr +
6402 cur_offset);
6403 dirid = btrfs_inode_extref_parent(leaf, extref);
6404 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6405 cur_offset += ref_name_len + sizeof(*extref);
6406 if (dirid == last_dirid)
6407 continue;
6408 ret = dir_changed(sctx, dirid);
6409 if (ret)
6410 break;
6411 last_dirid = dirid;
6412 }
6413out:
6414 return ret;
6415}
6416
766702ef
AB		 6417/*
6418 * Updates compare related fields in sctx and simply forwards to the actual
6419 * changed_xxx functions.
6420 */
ee8c494f 6421static int changed_cb(struct btrfs_path *left_path,
31db9f7c
AB		 6422 struct btrfs_path *right_path,
6423 struct btrfs_key *key,
6424 enum btrfs_compare_tree_result result,
6425 void *ctx)
6426{
6427 int ret = 0;
6428 struct send_ctx *sctx = ctx;
6429
ba5e8f2e 6430 if (result == BTRFS_COMPARE_TREE_SAME) {
16e7549f
JB		 6431 if (key->type == BTRFS_INODE_REF_KEY ||
6432 key->type == BTRFS_INODE_EXTREF_KEY) {
6433 ret = compare_refs(sctx, left_path, key);
6434 if (!ret)
6435 return 0;
6436 if (ret < 0)
6437 return ret;
6438 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6439 return maybe_send_hole(sctx, left_path, key);
6440 } else {
ba5e8f2e 6441 return 0;
16e7549f 6442 }
ba5e8f2e
JB		 6443 result = BTRFS_COMPARE_TREE_CHANGED;
6444 ret = 0;
6445 }
6446
31db9f7c
AB		 6447 sctx->left_path = left_path;
6448 sctx->right_path = right_path;
6449 sctx->cmp_key = key;
6450
6451 ret = finish_inode_if_needed(sctx, 0);
6452 if (ret < 0)
6453 goto out;
6454
2981e225
AB		 6455 /* Ignore non-FS objects */
6456 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6457 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6458 goto out;
6459
46b2f459 6460 if (key->type == BTRFS_INODE_ITEM_KEY) {
31db9f7c 6461 ret = changed_inode(sctx, result);
46b2f459
FM		 6462 } else if (!sctx->ignore_cur_inode) {
6463 if (key->type == BTRFS_INODE_REF_KEY ||
6464 key->type == BTRFS_INODE_EXTREF_KEY)
6465 ret = changed_ref(sctx, result);
6466 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6467 ret = changed_xattr(sctx, result);
6468 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6469 ret = changed_extent(sctx, result);
6470 }
31db9f7c
AB		 6471
6472out:
6473 return ret;
6474}
6475
6476static int full_send_tree(struct send_ctx *sctx)
6477{
6478 int ret;
31db9f7c
AB		 6479 struct btrfs_root *send_root = sctx->send_root;
6480 struct btrfs_key key;
31db9f7c
AB		 6481 struct btrfs_path *path;
6482 struct extent_buffer *eb;
6483 int slot;
31db9f7c
AB		 6484
6485 path = alloc_path_for_send();
6486 if (!path)
6487 return -ENOMEM;
6488
31db9f7c
AB		 6489 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6490 key.type = BTRFS_INODE_ITEM_KEY;
6491 key.offset = 0;
6492
31db9f7c
AB		 6493 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6494 if (ret < 0)
6495 goto out;
6496 if (ret)
6497 goto out_finish;
6498
6499 while (1) {
31db9f7c
AB		 6500 eb = path->nodes[0];
6501 slot = path->slots[0];
ca5d2ba1 6502 btrfs_item_key_to_cpu(eb, &key, slot);
31db9f7c 6503
ca5d2ba1 6504 ret = changed_cb(path, NULL, &key,
ee8c494f 6505 BTRFS_COMPARE_TREE_NEW, sctx);
31db9f7c
AB		 6506 if (ret < 0)
6507 goto out;
6508
31db9f7c
AB		 6509 ret = btrfs_next_item(send_root, path);
6510 if (ret < 0)
6511 goto out;
6512 if (ret) {
6513 ret = 0;
6514 break;
6515 }
6516 }
6517
6518out_finish:
6519 ret = finish_inode_if_needed(sctx, 1);
6520
6521out:
6522 btrfs_free_path(path);
31db9f7c
AB		 6523 return ret;
6524}
6525
18d0f5c6
DS		 6526static int tree_move_down(struct btrfs_path *path, int *level)
6527{
6528 struct extent_buffer *eb;
6529
6530 BUG_ON(*level == 0);
6531 eb = btrfs_read_node_slot(path->nodes[*level], path->slots[*level]);
6532 if (IS_ERR(eb))
6533 return PTR_ERR(eb);
6534
6535 path->nodes[*level - 1] = eb;
6536 path->slots[*level - 1] = 0;
6537 (*level)--;
6538 return 0;
6539}
6540
6541static int tree_move_next_or_upnext(struct btrfs_path *path,
6542 int *level, int root_level)
6543{
6544 int ret = 0;
6545 int nritems;
6546 nritems = btrfs_header_nritems(path->nodes[*level]);
6547
6548 path->slots[*level]++;
6549
6550 while (path->slots[*level] >= nritems) {
6551 if (*level == root_level)
6552 return -1;
6553
6554 /* move upnext */
6555 path->slots[*level] = 0;
6556 free_extent_buffer(path->nodes[*level]);
6557 path->nodes[*level] = NULL;
6558 (*level)++;
6559 path->slots[*level]++;
6560
6561 nritems = btrfs_header_nritems(path->nodes[*level]);
6562 ret = 1;
6563 }
6564 return ret;
6565}
6566
6567/*
6568 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
6569 * or down.
6570 */
6571static int tree_advance(struct btrfs_path *path,
6572 int *level, int root_level,
6573 int allow_down,
6574 struct btrfs_key *key)
6575{
6576 int ret;
6577
6578 if (*level == 0 || !allow_down) {
6579 ret = tree_move_next_or_upnext(path, level, root_level);
6580 } else {
6581 ret = tree_move_down(path, level);
6582 }
6583 if (ret >= 0) {
6584 if (*level == 0)
6585 btrfs_item_key_to_cpu(path->nodes[*level], key,
6586 path->slots[*level]);
6587 else
6588 btrfs_node_key_to_cpu(path->nodes[*level], key,
6589 path->slots[*level]);
6590 }
6591 return ret;
6592}
6593
6594static int tree_compare_item(struct btrfs_path *left_path,
6595 struct btrfs_path *right_path,
6596 char *tmp_buf)
6597{
6598 int cmp;
6599 int len1, len2;
6600 unsigned long off1, off2;
6601
6602 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
6603 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
6604 if (len1 != len2)
6605 return 1;
6606
6607 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
6608 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
6609 right_path->slots[0]);
6610
6611 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
6612
6613 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
6614 if (cmp)
6615 return 1;
6616 return 0;
6617}
6618
6619/*
6620 * This function compares two trees and calls the provided callback for
6621 * every changed/new/deleted item it finds.
6622 * If shared tree blocks are encountered, whole subtrees are skipped, making
6623 * the compare pretty fast on snapshotted subvolumes.
6624 *
6625 * This currently works on commit roots only. As commit roots are read only,
6626 * we don't do any locking. The commit roots are protected with transactions.
6627 * Transactions are ended and rejoined when a commit is tried in between.
6628 *
6629 * This function checks for modifications done to the trees while comparing.
6630 * If it detects a change, it aborts immediately.
6631 */
6632static int btrfs_compare_trees(struct btrfs_root *left_root,
6633 struct btrfs_root *right_root,
6634 btrfs_changed_cb_t changed_cb, void *ctx)
6635{
6636 struct btrfs_fs_info *fs_info = left_root->fs_info;
6637 int ret;
6638 int cmp;
6639 struct btrfs_path *left_path = NULL;
6640 struct btrfs_path *right_path = NULL;
6641 struct btrfs_key left_key;
6642 struct btrfs_key right_key;
6643 char *tmp_buf = NULL;
6644 int left_root_level;
6645 int right_root_level;
6646 int left_level;
6647 int right_level;
6648 int left_end_reached;
6649 int right_end_reached;
6650 int advance_left;
6651 int advance_right;
6652 u64 left_blockptr;
6653 u64 right_blockptr;
6654 u64 left_gen;
6655 u64 right_gen;
6656
6657 left_path = btrfs_alloc_path();
6658 if (!left_path) {
6659 ret = -ENOMEM;
6660 goto out;
6661 }
6662 right_path = btrfs_alloc_path();
6663 if (!right_path) {
6664 ret = -ENOMEM;
6665 goto out;
6666 }
6667
6668 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
6669 if (!tmp_buf) {
6670 ret = -ENOMEM;
6671 goto out;
6672 }
6673
6674 left_path->search_commit_root = 1;
6675 left_path->skip_locking = 1;
6676 right_path->search_commit_root = 1;
6677 right_path->skip_locking = 1;
6678
6679 /*
6680 * Strategy: Go to the first items of both trees. Then do
6681 *
6682 * If both trees are at level 0
6683 * Compare keys of current items
6684 * If left < right treat left item as new, advance left tree
6685 * and repeat
6686 * If left > right treat right item as deleted, advance right tree
6687 * and repeat
6688 * If left == right do deep compare of items, treat as changed if
6689 * needed, advance both trees and repeat
6690 * If both trees are at the same level but not at level 0
6691 * Compare keys of current nodes/leafs
6692 * If left < right advance left tree and repeat
6693 * If left > right advance right tree and repeat
6694 * If left == right compare blockptrs of the next nodes/leafs
6695 * If they match advance both trees but stay at the same level
6696 * and repeat
6697 * If they don't match advance both trees while allowing to go
6698 * deeper and repeat
6699 * If tree levels are different
6700 * Advance the tree that needs it and repeat
6701 *
6702 * Advancing a tree means:
6703 * If we are at level 0, try to go to the next slot. If that's not
6704 * possible, go one level up and repeat. Stop when we found a level
6705 * where we could go to the next slot. We may at this point be on a
6706 * node or a leaf.
6707 *
6708 * If we are not at level 0 and not on shared tree blocks, go one
6709 * level deeper.
6710 *
6711 * If we are not at level 0 and on shared tree blocks, go one slot to
6712 * the right if possible or go up and right.
6713 */
6714
6715 down_read(&fs_info->commit_root_sem);
6716 left_level = btrfs_header_level(left_root->commit_root);
6717 left_root_level = left_level;
6718 left_path->nodes[left_level] =
6719 btrfs_clone_extent_buffer(left_root->commit_root);
6720 if (!left_path->nodes[left_level]) {
6721 up_read(&fs_info->commit_root_sem);
6722 ret = -ENOMEM;
6723 goto out;
6724 }
6725
6726 right_level = btrfs_header_level(right_root->commit_root);
6727 right_root_level = right_level;
6728 right_path->nodes[right_level] =
6729 btrfs_clone_extent_buffer(right_root->commit_root);
6730 if (!right_path->nodes[right_level]) {
6731 up_read(&fs_info->commit_root_sem);
6732 ret = -ENOMEM;
6733 goto out;
6734 }
6735 up_read(&fs_info->commit_root_sem);
6736
6737 if (left_level == 0)
6738 btrfs_item_key_to_cpu(left_path->nodes[left_level],
6739 &left_key, left_path->slots[left_level]);
6740 else
6741 btrfs_node_key_to_cpu(left_path->nodes[left_level],
6742 &left_key, left_path->slots[left_level]);
6743 if (right_level == 0)
6744 btrfs_item_key_to_cpu(right_path->nodes[right_level],
6745 &right_key, right_path->slots[right_level]);
6746 else
6747 btrfs_node_key_to_cpu(right_path->nodes[right_level],
6748 &right_key, right_path->slots[right_level]);
6749
6750 left_end_reached = right_end_reached = 0;
6751 advance_left = advance_right = 0;
6752
6753 while (1) {
6af112b1 6754 cond_resched();
18d0f5c6
DS		 6755 if (advance_left && !left_end_reached) {
6756 ret = tree_advance(left_path, &left_level,
6757 left_root_level,
6758 advance_left != ADVANCE_ONLY_NEXT,
6759 &left_key);
6760 if (ret == -1)
6761 left_end_reached = ADVANCE;
6762 else if (ret < 0)
6763 goto out;
6764 advance_left = 0;
6765 }
6766 if (advance_right && !right_end_reached) {
6767 ret = tree_advance(right_path, &right_level,
6768 right_root_level,
6769 advance_right != ADVANCE_ONLY_NEXT,
6770 &right_key);
6771 if (ret == -1)
6772 right_end_reached = ADVANCE;
6773 else if (ret < 0)
6774 goto out;
6775 advance_right = 0;
6776 }
6777
6778 if (left_end_reached && right_end_reached) {
6779 ret = 0;
6780 goto out;
6781 } else if (left_end_reached) {
6782 if (right_level == 0) {
6783 ret = changed_cb(left_path, right_path,
6784 &right_key,
6785 BTRFS_COMPARE_TREE_DELETED,
6786 ctx);
6787 if (ret < 0)
6788 goto out;
6789 }
6790 advance_right = ADVANCE;
6791 continue;
6792 } else if (right_end_reached) {
6793 if (left_level == 0) {
6794 ret = changed_cb(left_path, right_path,
6795 &left_key,
6796 BTRFS_COMPARE_TREE_NEW,
6797 ctx);
6798 if (ret < 0)
6799 goto out;
6800 }
6801 advance_left = ADVANCE;
6802 continue;
6803 }
6804
6805 if (left_level == 0 && right_level == 0) {
6806 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6807 if (cmp < 0) {
6808 ret = changed_cb(left_path, right_path,
6809 &left_key,
6810 BTRFS_COMPARE_TREE_NEW,
6811 ctx);
6812 if (ret < 0)
6813 goto out;
6814 advance_left = ADVANCE;
6815 } else if (cmp > 0) {
6816 ret = changed_cb(left_path, right_path,
6817 &right_key,
6818 BTRFS_COMPARE_TREE_DELETED,
6819 ctx);
6820 if (ret < 0)
6821 goto out;
6822 advance_right = ADVANCE;
6823 } else {
6824 enum btrfs_compare_tree_result result;
6825
6826 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
6827 ret = tree_compare_item(left_path, right_path,
6828 tmp_buf);
6829 if (ret)
6830 result = BTRFS_COMPARE_TREE_CHANGED;
6831 else
6832 result = BTRFS_COMPARE_TREE_SAME;
6833 ret = changed_cb(left_path, right_path,
6834 &left_key, result, ctx);
6835 if (ret < 0)
6836 goto out;
6837 advance_left = ADVANCE;
6838 advance_right = ADVANCE;
6839 }
6840 } else if (left_level == right_level) {
6841 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6842 if (cmp < 0) {
6843 advance_left = ADVANCE;
6844 } else if (cmp > 0) {
6845 advance_right = ADVANCE;
6846 } else {
6847 left_blockptr = btrfs_node_blockptr(
6848 left_path->nodes[left_level],
6849 left_path->slots[left_level]);
6850 right_blockptr = btrfs_node_blockptr(
6851 right_path->nodes[right_level],
6852 right_path->slots[right_level]);
6853 left_gen = btrfs_node_ptr_generation(
6854 left_path->nodes[left_level],
6855 left_path->slots[left_level]);
6856 right_gen = btrfs_node_ptr_generation(
6857 right_path->nodes[right_level],
6858 right_path->slots[right_level]);
6859 if (left_blockptr == right_blockptr &&
6860 left_gen == right_gen) {
6861 /*
6862 * As we're on a shared block, don't
6863 * allow to go deeper.
6864 */
6865 advance_left = ADVANCE_ONLY_NEXT;
6866 advance_right = ADVANCE_ONLY_NEXT;
6867 } else {
6868 advance_left = ADVANCE;
6869 advance_right = ADVANCE;
6870 }
6871 }
6872 } else if (left_level < right_level) {
6873 advance_right = ADVANCE;
6874 } else {
6875 advance_left = ADVANCE;
6876 }
6877 }
6878
6879out:
6880 btrfs_free_path(left_path);
6881 btrfs_free_path(right_path);
6882 kvfree(tmp_buf);
6883 return ret;
6884}
6885
31db9f7c
AB		 6886static int send_subvol(struct send_ctx *sctx)
6887{
6888 int ret;
6889
c2c71324
SB		 6890 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6891 ret = send_header(sctx);
6892 if (ret < 0)
6893 goto out;
6894 }
31db9f7c
AB		 6895
6896 ret = send_subvol_begin(sctx);
6897 if (ret < 0)
6898 goto out;
6899
6900 if (sctx->parent_root) {
6901 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6902 changed_cb, sctx);
6903 if (ret < 0)
6904 goto out;
6905 ret = finish_inode_if_needed(sctx, 1);
6906 if (ret < 0)
6907 goto out;
6908 } else {
6909 ret = full_send_tree(sctx);
6910 if (ret < 0)
6911 goto out;
6912 }
6913
6914out:
31db9f7c
AB		 6915 free_recorded_refs(sctx);
6916 return ret;
6917}
6918
e5fa8f86
FM		 6919/*
6920 * If orphan cleanup did remove any orphans from a root, it means the tree
6921 * was modified and therefore the commit root is not the same as the current
6922 * root anymore. This is a problem, because send uses the commit root and
6923 * therefore can see inode items that don't exist in the current root anymore,
6924 * and for example make calls to btrfs_iget, which will do tree lookups based
6925 * on the current root and not on the commit root. Those lookups will fail,
6926 * returning a -ESTALE error, and making send fail with that error. So make
6927 * sure a send does not see any orphans we have just removed, and that it will
6928 * see the same inodes regardless of whether a transaction commit happened
6929 * before it started (meaning that the commit root will be the same as the
6930 * current root) or not.
6931 */
6932static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6933{
6934 int i;
6935 struct btrfs_trans_handle *trans = NULL;
6936
6937again:
6938 if (sctx->parent_root &&
6939 sctx->parent_root->node != sctx->parent_root->commit_root)
6940 goto commit_trans;
6941
6942 for (i = 0; i < sctx->clone_roots_cnt; i++)
6943 if (sctx->clone_roots[i].root->node !=
6944 sctx->clone_roots[i].root->commit_root)
6945 goto commit_trans;
6946
6947 if (trans)
3a45bb20 6948 return btrfs_end_transaction(trans);
e5fa8f86
FM		 6949
6950 return 0;
6951
6952commit_trans:
6953 /* Use any root, all fs roots will get their commit roots updated. */
6954 if (!trans) {
6955 trans = btrfs_join_transaction(sctx->send_root);
6956 if (IS_ERR(trans))
6957 return PTR_ERR(trans);
6958 goto again;
6959 }
6960
3a45bb20 6961 return btrfs_commit_transaction(trans);
e5fa8f86
FM		 6962}
6963
9f89d5de
FM		 6964/*
6965 * Make sure any existing dellaloc is flushed for any root used by a send
6966 * operation so that we do not miss any data and we do not race with writeback
6967 * finishing and changing a tree while send is using the tree. This could
6968 * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and
6969 * a send operation then uses the subvolume.
6970 * After flushing delalloc ensure_commit_roots_uptodate() must be called.
6971 */
6972static int flush_delalloc_roots(struct send_ctx *sctx)
6973{
6974 struct btrfs_root *root = sctx->parent_root;
6975 int ret;
6976 int i;
6977
6978 if (root) {
6979 ret = btrfs_start_delalloc_snapshot(root);
6980 if (ret)
6981 return ret;
6982 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
6983 }
6984
6985 for (i = 0; i < sctx->clone_roots_cnt; i++) {
6986 root = sctx->clone_roots[i].root;
6987 ret = btrfs_start_delalloc_snapshot(root);
6988 if (ret)
6989 return ret;
6990 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
6991 }
6992
6993 return 0;
6994}
6995
66ef7d65
DS		 6996static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6997{
6998 spin_lock(&root->root_item_lock);
6999 root->send_in_progress--;
7000 /*
7001 * Not much left to do, we don't know why it's unbalanced and
7002 * can't blindly reset it to 0.
7003 */
7004 if (root->send_in_progress < 0)
7005 btrfs_err(root->fs_info,
f5686e3a 7006 "send_in_progress unbalanced %d root %llu",
0b246afa 7007 root->send_in_progress, root->root_key.objectid);
66ef7d65
DS		 7008 spin_unlock(&root->root_item_lock);
7009}
7010
62d54f3a
FM		 7011static void dedupe_in_progress_warn(const struct btrfs_root *root)
7012{
7013 btrfs_warn_rl(root->fs_info,
7014"cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
7015 root->root_key.objectid, root->dedupe_in_progress);
7016}
7017
2351f431 7018long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
31db9f7c
AB		 7019{
7020 int ret = 0;
0b246afa
JM		 7021 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
7022 struct btrfs_fs_info *fs_info = send_root->fs_info;
31db9f7c 7023 struct btrfs_root *clone_root;
31db9f7c 7024 struct btrfs_key key;
31db9f7c
AB		 7025 struct send_ctx *sctx = NULL;
7026 u32 i;
7027 u64 *clone_sources_tmp = NULL;
2c686537 7028 int clone_sources_to_rollback = 0;
e55d1153 7029 unsigned alloc_size;
896c14f9 7030 int sort_clone_roots = 0;
31db9f7c
AB		 7031
7032 if (!capable(CAP_SYS_ADMIN))
7033 return -EPERM;
7034
2c686537
DS		 7035 /*
7036 * The subvolume must remain read-only during send, protect against
521e0546 7037 * making it RW. This also protects against deletion.
2c686537
DS		 7038 */
7039 spin_lock(&send_root->root_item_lock);
62d54f3a
FM		 7040 if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
7041 dedupe_in_progress_warn(send_root);
7042 spin_unlock(&send_root->root_item_lock);
7043 return -EAGAIN;
7044 }
2c686537
DS		 7045 send_root->send_in_progress++;
7046 spin_unlock(&send_root->root_item_lock);
7047
2c686537
DS		 7048 /*
7049 * Userspace tools do the checks and warn the user if it's
7050 * not RO.
7051 */
7052 if (!btrfs_root_readonly(send_root)) {
7053 ret = -EPERM;
7054 goto out;
7055 }
7056
457ae726
DC		 7057 /*
7058 * Check that we don't overflow at later allocations, we request
7059 * clone_sources_count + 1 items, and compare to unsigned long inside
7060 * access_ok.
7061 */
f5ecec3c 7062 if (arg->clone_sources_count >
457ae726 7063 ULONG_MAX / sizeof(struct clone_root) - 1) {
f5ecec3c
DC		 7064 ret = -EINVAL;
7065 goto out;
7066 }
7067
96d4f267 7068 if (!access_ok(arg->clone_sources,
700ff4f0
DC		 7069 sizeof(*arg->clone_sources) *
7070 arg->clone_sources_count)) {
31db9f7c
AB		 7071 ret = -EFAULT;
7072 goto out;
7073 }
7074
c2c71324 7075 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
cb95e7bf
MF		 7076 ret = -EINVAL;
7077 goto out;
7078 }
7079
e780b0d1 7080 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
31db9f7c
AB		 7081 if (!sctx) {
7082 ret = -ENOMEM;
7083 goto out;
7084 }
7085
7086 INIT_LIST_HEAD(&sctx->new_refs);
7087 INIT_LIST_HEAD(&sctx->deleted_refs);
e780b0d1 7088 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
31db9f7c
AB		 7089 INIT_LIST_HEAD(&sctx->name_cache_list);
7090
cb95e7bf
MF		 7091 sctx->flags = arg->flags;
7092
31db9f7c 7093 sctx->send_filp = fget(arg->send_fd);
ecc7ada7
TI		 7094 if (!sctx->send_filp) {
7095 ret = -EBADF;
31db9f7c
AB		 7096 goto out;
7097 }
7098
31db9f7c 7099 sctx->send_root = send_root;
521e0546
DS		 7100 /*
7101 * Unlikely but possible, if the subvolume is marked for deletion but
7102 * is slow to remove the directory entry, send can still be started
7103 */
7104 if (btrfs_root_dead(sctx->send_root)) {
7105 ret = -EPERM;
7106 goto out;
7107 }
7108
31db9f7c
AB		 7109 sctx->clone_roots_cnt = arg->clone_sources_count;
7110
7111 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
752ade68 7112 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
31db9f7c 7113 if (!sctx->send_buf) {
752ade68
MH		 7114 ret = -ENOMEM;
7115 goto out;
31db9f7c
AB		 7116 }
7117
752ade68 7118 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
31db9f7c 7119 if (!sctx->read_buf) {
752ade68
MH		 7120 ret = -ENOMEM;
7121 goto out;
31db9f7c
AB		 7122 }
7123
9f03740a
FDBM		 7124 sctx->pending_dir_moves = RB_ROOT;
7125 sctx->waiting_dir_moves = RB_ROOT;
9dc44214 7126 sctx->orphan_dirs = RB_ROOT;
9f03740a 7127
e55d1153
DS		 7128 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
7129
818e010b 7130 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
31db9f7c 7131 if (!sctx->clone_roots) {
818e010b
DS		 7132 ret = -ENOMEM;
7133 goto out;
31db9f7c
AB		 7134 }
7135
e55d1153
DS		 7136 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
7137
31db9f7c 7138 if (arg->clone_sources_count) {
752ade68 7139 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
31db9f7c 7140 if (!clone_sources_tmp) {
752ade68
MH		 7141 ret = -ENOMEM;
7142 goto out;
31db9f7c
AB		 7143 }
7144
7145 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
e55d1153 7146 alloc_size);
31db9f7c
AB		 7147 if (ret) {
7148 ret = -EFAULT;
7149 goto out;
7150 }
7151
7152 for (i = 0; i < arg->clone_sources_count; i++) {
7153 key.objectid = clone_sources_tmp[i];
7154 key.type = BTRFS_ROOT_ITEM_KEY;
7155 key.offset = (u64)-1;
18f687d5 7156
3619c94f 7157 clone_root = btrfs_get_fs_root(fs_info, &key, true);
31db9f7c
AB		 7158 if (IS_ERR(clone_root)) {
7159 ret = PTR_ERR(clone_root);
7160 goto out;
7161 }
2c686537 7162 spin_lock(&clone_root->root_item_lock);
5cc2b17e
FM		 7163 if (!btrfs_root_readonly(clone_root) ||
7164 btrfs_root_dead(clone_root)) {
2c686537 7165 spin_unlock(&clone_root->root_item_lock);
00246528 7166 btrfs_put_root(clone_root);
2c686537
DS		 7167 ret = -EPERM;
7168 goto out;
7169 }
62d54f3a
FM		 7170 if (clone_root->dedupe_in_progress) {
7171 dedupe_in_progress_warn(clone_root);
7172 spin_unlock(&clone_root->root_item_lock);
00246528 7173 btrfs_put_root(clone_root);
62d54f3a
FM		 7174 ret = -EAGAIN;
7175 goto out;
7176 }
2f1f465a 7177 clone_root->send_in_progress++;
2c686537 7178 spin_unlock(&clone_root->root_item_lock);
18f687d5 7179
31db9f7c 7180 sctx->clone_roots[i].root = clone_root;
2f1f465a 7181 clone_sources_to_rollback = i + 1;
31db9f7c 7182 }
2f91306a 7183 kvfree(clone_sources_tmp);
31db9f7c
AB		 7184 clone_sources_tmp = NULL;
7185 }
7186
7187 if (arg->parent_root) {
7188 key.objectid = arg->parent_root;
7189 key.type = BTRFS_ROOT_ITEM_KEY;
7190 key.offset = (u64)-1;
18f687d5 7191
3619c94f 7192 sctx->parent_root = btrfs_get_fs_root(fs_info, &key, true);
b1b19596
SB		 7193 if (IS_ERR(sctx->parent_root)) {
7194 ret = PTR_ERR(sctx->parent_root);
31db9f7c
AB		 7195 goto out;
7196 }
18f687d5 7197
2c686537
DS		 7198 spin_lock(&sctx->parent_root->root_item_lock);
7199 sctx->parent_root->send_in_progress++;
521e0546
DS		 7200 if (!btrfs_root_readonly(sctx->parent_root) ||
7201 btrfs_root_dead(sctx->parent_root)) {
2c686537
DS		 7202 spin_unlock(&sctx->parent_root->root_item_lock);
7203 ret = -EPERM;
7204 goto out;
7205 }
62d54f3a
FM		 7206 if (sctx->parent_root->dedupe_in_progress) {
7207 dedupe_in_progress_warn(sctx->parent_root);
7208 spin_unlock(&sctx->parent_root->root_item_lock);
62d54f3a
FM		 7209 ret = -EAGAIN;
7210 goto out;
7211 }
2c686537 7212 spin_unlock(&sctx->parent_root->root_item_lock);
31db9f7c
AB		 7213 }
7214
7215 /*
7216 * Clones from send_root are allowed, but only if the clone source
7217 * is behind the current send position. This is checked while searching
7218 * for possible clone sources.
7219 */
6f9a3da5 7220 sctx->clone_roots[sctx->clone_roots_cnt++].root =
00246528 7221 btrfs_grab_root(sctx->send_root);
31db9f7c
AB		 7222
7223 /* We do a bsearch later */
7224 sort(sctx->clone_roots, sctx->clone_roots_cnt,
7225 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
7226 NULL);
896c14f9 7227 sort_clone_roots = 1;
31db9f7c 7228
9f89d5de
FM		 7229 ret = flush_delalloc_roots(sctx);
7230 if (ret)
7231 goto out;
7232
e5fa8f86
FM		 7233 ret = ensure_commit_roots_uptodate(sctx);
7234 if (ret)
7235 goto out;
7236
9e967495
FM		 7237 mutex_lock(&fs_info->balance_mutex);
7238 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
7239 mutex_unlock(&fs_info->balance_mutex);
7240 btrfs_warn_rl(fs_info,
7241 "cannot run send because a balance operation is in progress");
7242 ret = -EAGAIN;
7243 goto out;
7244 }
7245 fs_info->send_in_progress++;
7246 mutex_unlock(&fs_info->balance_mutex);
7247
2755a0de 7248 current->journal_info = BTRFS_SEND_TRANS_STUB;
31db9f7c 7249 ret = send_subvol(sctx);
a26e8c9f 7250 current->journal_info = NULL;
9e967495
FM		 7251 mutex_lock(&fs_info->balance_mutex);
7252 fs_info->send_in_progress--;
7253 mutex_unlock(&fs_info->balance_mutex);
31db9f7c
AB		 7254 if (ret < 0)
7255 goto out;
7256
c2c71324
SB		 7257 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
7258 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
7259 if (ret < 0)
7260 goto out;
7261 ret = send_cmd(sctx);
7262 if (ret < 0)
7263 goto out;
7264 }
31db9f7c
AB		 7265
7266out:
9f03740a
FDBM		 7267 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
7268 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
7269 struct rb_node *n;
7270 struct pending_dir_move *pm;
7271
7272 n = rb_first(&sctx->pending_dir_moves);
7273 pm = rb_entry(n, struct pending_dir_move, node);
7274 while (!list_empty(&pm->list)) {
7275 struct pending_dir_move *pm2;
7276
7277 pm2 = list_first_entry(&pm->list,
7278 struct pending_dir_move, list);
7279 free_pending_move(sctx, pm2);
7280 }
7281 free_pending_move(sctx, pm);
7282 }
7283
7284 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
7285 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
7286 struct rb_node *n;
7287 struct waiting_dir_move *dm;
7288
7289 n = rb_first(&sctx->waiting_dir_moves);
7290 dm = rb_entry(n, struct waiting_dir_move, node);
7291 rb_erase(&dm->node, &sctx->waiting_dir_moves);
7292 kfree(dm);
7293 }
7294
9dc44214
FM		 7295 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
7296 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
7297 struct rb_node *n;
7298 struct orphan_dir_info *odi;
7299
7300 n = rb_first(&sctx->orphan_dirs);
7301 odi = rb_entry(n, struct orphan_dir_info, node);
7302 free_orphan_dir_info(sctx, odi);
7303 }
7304
896c14f9 7305 if (sort_clone_roots) {
6f9a3da5 7306 for (i = 0; i < sctx->clone_roots_cnt; i++) {
896c14f9
WS		 7307 btrfs_root_dec_send_in_progress(
7308 sctx->clone_roots[i].root);
00246528 7309 btrfs_put_root(sctx->clone_roots[i].root);
6f9a3da5 7310 }
896c14f9 7311 } else {
6f9a3da5 7312 for (i = 0; sctx && i < clone_sources_to_rollback; i++) {
896c14f9
WS		 7313 btrfs_root_dec_send_in_progress(
7314 sctx->clone_roots[i].root);
00246528 7315 btrfs_put_root(sctx->clone_roots[i].root);
6f9a3da5 7316 }
896c14f9
WS		 7317
7318 btrfs_root_dec_send_in_progress(send_root);
7319 }
6f9a3da5 7320 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) {
66ef7d65 7321 btrfs_root_dec_send_in_progress(sctx->parent_root);
00246528 7322 btrfs_put_root(sctx->parent_root);
6f9a3da5 7323 }
2c686537 7324
2f91306a 7325 kvfree(clone_sources_tmp);
31db9f7c
AB		 7326
7327 if (sctx) {
7328 if (sctx->send_filp)
7329 fput(sctx->send_filp);
7330
c03d01f3 7331 kvfree(sctx->clone_roots);
6ff48ce0 7332 kvfree(sctx->send_buf);
eb5b75fe 7333 kvfree(sctx->read_buf);
31db9f7c
AB		 7334
7335 name_cache_free(sctx);
7336
7337 kfree(sctx);
7338 }
7339
7340 return ret;
7341}
Linux 6.x block layer and io_uring tree(s)