projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: root->fs_info cleanup, update_block_group{,flags}
[linux-2.6-block.git] / fs / btrfs / send.c
CommitLineData
31db9f7c
AB		 1/*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/bsearch.h>
20#include <linux/fs.h>
21#include <linux/file.h>
22#include <linux/sort.h>
23#include <linux/mount.h>
24#include <linux/xattr.h>
25#include <linux/posix_acl_xattr.h>
26#include <linux/radix-tree.h>
a1857ebe 27#include <linux/vmalloc.h>
ed84885d 28#include <linux/string.h>
31db9f7c
AB		 29
30#include "send.h"
31#include "backref.h"
0b947aff 32#include "hash.h"
31db9f7c
AB		 33#include "locking.h"
34#include "disk-io.h"
35#include "btrfs_inode.h"
36#include "transaction.h"
ebb8765b 37#include "compression.h"
31db9f7c 38
31db9f7c
AB		 39/*
40 * A fs_path is a helper to dynamically build path names with unknown size.
41 * It reallocates the internal buffer on demand.
42 * It allows fast adding of path elements on the right side (normal path) and
43 * fast adding to the left side (reversed path). A reversed path can also be
44 * unreversed if needed.
45 */
46struct fs_path {
47 union {
48 struct {
49 char *start;
50 char *end;
31db9f7c
AB		 51
52 char *buf;
1f5a7ff9
DS		 53 unsigned short buf_len:15;
54 unsigned short reversed:1;
31db9f7c
AB		 55 char inline_buf[];
56 };
ace01050
DS		 57 /*
58 * Average path length does not exceed 200 bytes, we'll have
59 * better packing in the slab and higher chance to satisfy
60 * a allocation later during send.
61 */
62 char pad[256];
31db9f7c
AB		 63 };
64};
65#define FS_PATH_INLINE_SIZE \
66 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
67
68
69/* reused for each extent */
70struct clone_root {
71 struct btrfs_root *root;
72 u64 ino;
73 u64 offset;
74
75 u64 found_refs;
76};
77
78#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
79#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
80
81struct send_ctx {
82 struct file *send_filp;
83 loff_t send_off;
84 char *send_buf;
85 u32 send_size;
86 u32 send_max_size;
87 u64 total_send_size;
88 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
cb95e7bf 89 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
31db9f7c 90
31db9f7c
AB		 91 struct btrfs_root *send_root;
92 struct btrfs_root *parent_root;
93 struct clone_root *clone_roots;
94 int clone_roots_cnt;
95
96 /* current state of the compare_tree call */
97 struct btrfs_path *left_path;
98 struct btrfs_path *right_path;
99 struct btrfs_key *cmp_key;
100
101 /*
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
104 */
105 u64 cur_ino;
106 u64 cur_inode_gen;
107 int cur_inode_new;
108 int cur_inode_new_gen;
109 int cur_inode_deleted;
31db9f7c
AB		 110 u64 cur_inode_size;
111 u64 cur_inode_mode;
644d1940 112 u64 cur_inode_rdev;
16e7549f 113 u64 cur_inode_last_extent;
31db9f7c
AB		 114
115 u64 send_progress;
116
117 struct list_head new_refs;
118 struct list_head deleted_refs;
119
120 struct radix_tree_root name_cache;
121 struct list_head name_cache_list;
122 int name_cache_size;
123
2131bcd3
LB		 124 struct file_ra_state ra;
125
31db9f7c 126 char *read_buf;
9f03740a
FDBM		 127
128 /*
129 * We process inodes by their increasing order, so if before an
130 * incremental send we reverse the parent/child relationship of
131 * directories such that a directory with a lower inode number was
132 * the parent of a directory with a higher inode number, and the one
133 * becoming the new parent got renamed too, we can't rename/move the
134 * directory with lower inode number when we finish processing it - we
135 * must process the directory with higher inode number first, then
136 * rename/move it and then rename/move the directory with lower inode
137 * number. Example follows.
138 *
139 * Tree state when the first send was performed:
140 *
141 * .
142 * |-- a (ino 257)
143 * |-- b (ino 258)
144 * |
145 * |
146 * |-- c (ino 259)
147 * | |-- d (ino 260)
148 * |
149 * |-- c2 (ino 261)
150 *
151 * Tree state when the second (incremental) send is performed:
152 *
153 * .
154 * |-- a (ino 257)
155 * |-- b (ino 258)
156 * |-- c2 (ino 261)
157 * |-- d2 (ino 260)
158 * |-- cc (ino 259)
159 *
160 * The sequence of steps that lead to the second state was:
161 *
162 * mv /a/b/c/d /a/b/c2/d2
163 * mv /a/b/c /a/b/c2/d2/cc
164 *
165 * "c" has lower inode number, but we can't move it (2nd mv operation)
166 * before we move "d", which has higher inode number.
167 *
168 * So we just memorize which move/rename operations must be performed
169 * later when their respective parent is processed and moved/renamed.
170 */
171
172 /* Indexed by parent directory inode number. */
173 struct rb_root pending_dir_moves;
174
175 /*
176 * Reverse index, indexed by the inode number of a directory that
177 * is waiting for the move/rename of its immediate parent before its
178 * own move/rename can be performed.
179 */
180 struct rb_root waiting_dir_moves;
9dc44214
FM		 181
182 /*
183 * A directory that is going to be rm'ed might have a child directory
184 * which is in the pending directory moves index above. In this case,
185 * the directory can only be removed after the move/rename of its child
186 * is performed. Example:
187 *
188 * Parent snapshot:
189 *
190 * . (ino 256)
191 * |-- a/ (ino 257)
192 * |-- b/ (ino 258)
193 * |-- c/ (ino 259)
194 * | |-- x/ (ino 260)
195 * |
196 * |-- y/ (ino 261)
197 *
198 * Send snapshot:
199 *
200 * . (ino 256)
201 * |-- a/ (ino 257)
202 * |-- b/ (ino 258)
203 * |-- YY/ (ino 261)
204 * |-- x/ (ino 260)
205 *
206 * Sequence of steps that lead to the send snapshot:
207 * rm -f /a/b/c/foo.txt
208 * mv /a/b/y /a/b/YY
209 * mv /a/b/c/x /a/b/YY
210 * rmdir /a/b/c
211 *
212 * When the child is processed, its move/rename is delayed until its
213 * parent is processed (as explained above), but all other operations
214 * like update utimes, chown, chgrp, etc, are performed and the paths
215 * that it uses for those operations must use the orphanized name of
216 * its parent (the directory we're going to rm later), so we need to
217 * memorize that name.
218 *
219 * Indexed by the inode number of the directory to be deleted.
220 */
221 struct rb_root orphan_dirs;
9f03740a
FDBM		 222};
223
224struct pending_dir_move {
225 struct rb_node node;
226 struct list_head list;
227 u64 parent_ino;
228 u64 ino;
229 u64 gen;
230 struct list_head update_refs;
231};
232
233struct waiting_dir_move {
234 struct rb_node node;
235 u64 ino;
9dc44214
FM		 236 /*
237 * There might be some directory that could not be removed because it
238 * was waiting for this directory inode to be moved first. Therefore
239 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
240 */
241 u64 rmdir_ino;
8b191a68 242 bool orphanized;
9dc44214
FM		 243};
244
245struct orphan_dir_info {
246 struct rb_node node;
247 u64 ino;
248 u64 gen;
31db9f7c
AB		 249};
250
251struct name_cache_entry {
252 struct list_head list;
7e0926fe
AB		 253 /*
254 * radix_tree has only 32bit entries but we need to handle 64bit inums.
255 * We use the lower 32bit of the 64bit inum to store it in the tree. If
256 * more then one inum would fall into the same entry, we use radix_list
257 * to store the additional entries. radix_list is also used to store
258 * entries where two entries have the same inum but different
259 * generations.
260 */
261 struct list_head radix_list;
31db9f7c
AB		 262 u64 ino;
263 u64 gen;
264 u64 parent_ino;
265 u64 parent_gen;
266 int ret;
267 int need_later_update;
268 int name_len;
269 char name[];
270};
271
95155585
FM		 272static void inconsistent_snapshot_error(struct send_ctx *sctx,
273 enum btrfs_compare_tree_result result,
274 const char *what)
275{
276 const char *result_string;
277
278 switch (result) {
279 case BTRFS_COMPARE_TREE_NEW:
280 result_string = "new";
281 break;
282 case BTRFS_COMPARE_TREE_DELETED:
283 result_string = "deleted";
284 break;
285 case BTRFS_COMPARE_TREE_CHANGED:
286 result_string = "updated";
287 break;
288 case BTRFS_COMPARE_TREE_SAME:
289 ASSERT(0);
290 result_string = "unchanged";
291 break;
292 default:
293 ASSERT(0);
294 result_string = "unexpected";
295 }
296
297 btrfs_err(sctx->send_root->fs_info,
298 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
299 result_string, what, sctx->cmp_key->objectid,
300 sctx->send_root->root_key.objectid,
301 (sctx->parent_root ?
302 sctx->parent_root->root_key.objectid : 0));
303}
304
9f03740a
FDBM		 305static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
306
9dc44214
FM		 307static struct waiting_dir_move *
308get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
309
310static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
311
16e7549f
JB		 312static int need_send_hole(struct send_ctx *sctx)
313{
314 return (sctx->parent_root && !sctx->cur_inode_new &&
315 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
316 S_ISREG(sctx->cur_inode_mode));
317}
318
31db9f7c
AB		 319static void fs_path_reset(struct fs_path *p)
320{
321 if (p->reversed) {
322 p->start = p->buf + p->buf_len - 1;
323 p->end = p->start;
324 *p->start = 0;
325 } else {
326 p->start = p->buf;
327 p->end = p->start;
328 *p->start = 0;
329 }
330}
331
924794c9 332static struct fs_path *fs_path_alloc(void)
31db9f7c
AB		 333{
334 struct fs_path *p;
335
e780b0d1 336 p = kmalloc(sizeof(*p), GFP_KERNEL);
31db9f7c
AB		 337 if (!p)
338 return NULL;
339 p->reversed = 0;
31db9f7c
AB		 340 p->buf = p->inline_buf;
341 p->buf_len = FS_PATH_INLINE_SIZE;
342 fs_path_reset(p);
343 return p;
344}
345
924794c9 346static struct fs_path *fs_path_alloc_reversed(void)
31db9f7c
AB		 347{
348 struct fs_path *p;
349
924794c9 350 p = fs_path_alloc();
31db9f7c
AB		 351 if (!p)
352 return NULL;
353 p->reversed = 1;
354 fs_path_reset(p);
355 return p;
356}
357
924794c9 358static void fs_path_free(struct fs_path *p)
31db9f7c
AB		 359{
360 if (!p)
361 return;
ace01050
DS		 362 if (p->buf != p->inline_buf)
363 kfree(p->buf);
31db9f7c
AB		 364 kfree(p);
365}
366
367static int fs_path_len(struct fs_path *p)
368{
369 return p->end - p->start;
370}
371
372static int fs_path_ensure_buf(struct fs_path *p, int len)
373{
374 char *tmp_buf;
375 int path_len;
376 int old_buf_len;
377
378 len++;
379
380 if (p->buf_len >= len)
381 return 0;
382
cfd4a535
CM		 383 if (len > PATH_MAX) {
384 WARN_ON(1);
385 return -ENOMEM;
386 }
387
1b2782c8
DS		 388 path_len = p->end - p->start;
389 old_buf_len = p->buf_len;
390
ace01050
DS		 391 /*
392 * First time the inline_buf does not suffice
393 */
01a9a8a9 394 if (p->buf == p->inline_buf) {
e780b0d1 395 tmp_buf = kmalloc(len, GFP_KERNEL);
01a9a8a9
FM		 396 if (tmp_buf)
397 memcpy(tmp_buf, p->buf, old_buf_len);
398 } else {
e780b0d1 399 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
01a9a8a9 400 }
9c9ca00b
DS		 401 if (!tmp_buf)
402 return -ENOMEM;
403 p->buf = tmp_buf;
404 /*
405 * The real size of the buffer is bigger, this will let the fast path
406 * happen most of the time
407 */
408 p->buf_len = ksize(p->buf);
ace01050 409
31db9f7c
AB		 410 if (p->reversed) {
411 tmp_buf = p->buf + old_buf_len - path_len - 1;
412 p->end = p->buf + p->buf_len - 1;
413 p->start = p->end - path_len;
414 memmove(p->start, tmp_buf, path_len + 1);
415 } else {
416 p->start = p->buf;
417 p->end = p->start + path_len;
418 }
419 return 0;
420}
421
b23ab57d
DS		 422static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
423 char **prepared)
31db9f7c
AB		 424{
425 int ret;
426 int new_len;
427
428 new_len = p->end - p->start + name_len;
429 if (p->start != p->end)
430 new_len++;
431 ret = fs_path_ensure_buf(p, new_len);
432 if (ret < 0)
433 goto out;
434
435 if (p->reversed) {
436 if (p->start != p->end)
437 *--p->start = '/';
438 p->start -= name_len;
b23ab57d 439 *prepared = p->start;
31db9f7c
AB		 440 } else {
441 if (p->start != p->end)
442 *p->end++ = '/';
b23ab57d 443 *prepared = p->end;
31db9f7c
AB		 444 p->end += name_len;
445 *p->end = 0;
446 }
447
448out:
449 return ret;
450}
451
452static int fs_path_add(struct fs_path *p, const char *name, int name_len)
453{
454 int ret;
b23ab57d 455 char *prepared;
31db9f7c 456
b23ab57d 457 ret = fs_path_prepare_for_add(p, name_len, &prepared);
31db9f7c
AB		 458 if (ret < 0)
459 goto out;
b23ab57d 460 memcpy(prepared, name, name_len);
31db9f7c
AB		 461
462out:
463 return ret;
464}
465
466static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
467{
468 int ret;
b23ab57d 469 char *prepared;
31db9f7c 470
b23ab57d 471 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
31db9f7c
AB		 472 if (ret < 0)
473 goto out;
b23ab57d 474 memcpy(prepared, p2->start, p2->end - p2->start);
31db9f7c
AB		 475
476out:
477 return ret;
478}
479
480static int fs_path_add_from_extent_buffer(struct fs_path *p,
481 struct extent_buffer *eb,
482 unsigned long off, int len)
483{
484 int ret;
b23ab57d 485 char *prepared;
31db9f7c 486
b23ab57d 487 ret = fs_path_prepare_for_add(p, len, &prepared);
31db9f7c
AB		 488 if (ret < 0)
489 goto out;
490
b23ab57d 491 read_extent_buffer(eb, prepared, off, len);
31db9f7c
AB		 492
493out:
494 return ret;
495}
496
31db9f7c
AB		 497static int fs_path_copy(struct fs_path *p, struct fs_path *from)
498{
499 int ret;
500
501 p->reversed = from->reversed;
502 fs_path_reset(p);
503
504 ret = fs_path_add_path(p, from);
505
506 return ret;
507}
508
509
510static void fs_path_unreverse(struct fs_path *p)
511{
512 char *tmp;
513 int len;
514
515 if (!p->reversed)
516 return;
517
518 tmp = p->start;
519 len = p->end - p->start;
520 p->start = p->buf;
521 p->end = p->start + len;
522 memmove(p->start, tmp, len + 1);
523 p->reversed = 0;
524}
525
526static struct btrfs_path *alloc_path_for_send(void)
527{
528 struct btrfs_path *path;
529
530 path = btrfs_alloc_path();
531 if (!path)
532 return NULL;
533 path->search_commit_root = 1;
534 path->skip_locking = 1;
3f8a18cc 535 path->need_commit_sem = 1;
31db9f7c
AB		 536 return path;
537}
538
48a3b636 539static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
31db9f7c
AB		 540{
541 int ret;
542 mm_segment_t old_fs;
543 u32 pos = 0;
544
545 old_fs = get_fs();
546 set_fs(KERNEL_DS);
547
548 while (pos < len) {
d447d0da
FF		 549 ret = vfs_write(filp, (__force const char __user *)buf + pos,
550 len - pos, off);
31db9f7c
AB		 551 /* TODO handle that correctly */
552 /*if (ret == -ERESTARTSYS) {
553 continue;
554 }*/
555 if (ret < 0)
556 goto out;
557 if (ret == 0) {
558 ret = -EIO;
559 goto out;
560 }
561 pos += ret;
562 }
563
564 ret = 0;
565
566out:
567 set_fs(old_fs);
568 return ret;
569}
570
571static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
572{
573 struct btrfs_tlv_header *hdr;
574 int total_len = sizeof(*hdr) + len;
575 int left = sctx->send_max_size - sctx->send_size;
576
577 if (unlikely(left < total_len))
578 return -EOVERFLOW;
579
580 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
581 hdr->tlv_type = cpu_to_le16(attr);
582 hdr->tlv_len = cpu_to_le16(len);
583 memcpy(hdr + 1, data, len);
584 sctx->send_size += total_len;
585
586 return 0;
587}
588
95bc79d5
DS		 589#define TLV_PUT_DEFINE_INT(bits) \
590 static int tlv_put_u##bits(struct send_ctx *sctx, \
591 u##bits attr, u##bits value) \
592 { \
593 __le##bits __tmp = cpu_to_le##bits(value); \
594 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
595 }
31db9f7c 596
95bc79d5 597TLV_PUT_DEFINE_INT(64)
31db9f7c
AB		 598
599static int tlv_put_string(struct send_ctx *sctx, u16 attr,
600 const char *str, int len)
601{
602 if (len == -1)
603 len = strlen(str);
604 return tlv_put(sctx, attr, str, len);
605}
606
607static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
608 const u8 *uuid)
609{
610 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
611}
612
31db9f7c
AB		 613static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
614 struct extent_buffer *eb,
615 struct btrfs_timespec *ts)
616{
617 struct btrfs_timespec bts;
618 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
619 return tlv_put(sctx, attr, &bts, sizeof(bts));
620}
621
622
623#define TLV_PUT(sctx, attrtype, attrlen, data) \
624 do { \
625 ret = tlv_put(sctx, attrtype, attrlen, data); \
626 if (ret < 0) \
627 goto tlv_put_failure; \
628 } while (0)
629
630#define TLV_PUT_INT(sctx, attrtype, bits, value) \
631 do { \
632 ret = tlv_put_u##bits(sctx, attrtype, value); \
633 if (ret < 0) \
634 goto tlv_put_failure; \
635 } while (0)
636
637#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
638#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
639#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
640#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
641#define TLV_PUT_STRING(sctx, attrtype, str, len) \
642 do { \
643 ret = tlv_put_string(sctx, attrtype, str, len); \
644 if (ret < 0) \
645 goto tlv_put_failure; \
646 } while (0)
647#define TLV_PUT_PATH(sctx, attrtype, p) \
648 do { \
649 ret = tlv_put_string(sctx, attrtype, p->start, \
650 p->end - p->start); \
651 if (ret < 0) \
652 goto tlv_put_failure; \
653 } while(0)
654#define TLV_PUT_UUID(sctx, attrtype, uuid) \
655 do { \
656 ret = tlv_put_uuid(sctx, attrtype, uuid); \
657 if (ret < 0) \
658 goto tlv_put_failure; \
659 } while (0)
31db9f7c
AB		 660#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
661 do { \
662 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
663 if (ret < 0) \
664 goto tlv_put_failure; \
665 } while (0)
666
667static int send_header(struct send_ctx *sctx)
668{
669 struct btrfs_stream_header hdr;
670
671 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
672 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
673
1bcea355
AJ		 674 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
675 &sctx->send_off);
31db9f7c
AB		 676}
677
678/*
679 * For each command/item we want to send to userspace, we call this function.
680 */
681static int begin_cmd(struct send_ctx *sctx, int cmd)
682{
683 struct btrfs_cmd_header *hdr;
684
fae7f21c 685 if (WARN_ON(!sctx->send_buf))
31db9f7c 686 return -EINVAL;
31db9f7c
AB		 687
688 BUG_ON(sctx->send_size);
689
690 sctx->send_size += sizeof(*hdr);
691 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
692 hdr->cmd = cpu_to_le16(cmd);
693
694 return 0;
695}
696
697static int send_cmd(struct send_ctx *sctx)
698{
699 int ret;
700 struct btrfs_cmd_header *hdr;
701 u32 crc;
702
703 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
704 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
705 hdr->crc = 0;
706
0b947aff 707 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
31db9f7c
AB		 708 hdr->crc = cpu_to_le32(crc);
709
1bcea355
AJ		 710 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
711 &sctx->send_off);
31db9f7c
AB		 712
713 sctx->total_send_size += sctx->send_size;
714 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
715 sctx->send_size = 0;
716
717 return ret;
718}
719
720/*
721 * Sends a move instruction to user space
722 */
723static int send_rename(struct send_ctx *sctx,
724 struct fs_path *from, struct fs_path *to)
725{
04ab956e 726 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 727 int ret;
728
04ab956e 729 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
31db9f7c
AB		 730
731 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
732 if (ret < 0)
733 goto out;
734
735 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
736 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
737
738 ret = send_cmd(sctx);
739
740tlv_put_failure:
741out:
742 return ret;
743}
744
745/*
746 * Sends a link instruction to user space
747 */
748static int send_link(struct send_ctx *sctx,
749 struct fs_path *path, struct fs_path *lnk)
750{
04ab956e 751 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 752 int ret;
753
04ab956e 754 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
31db9f7c
AB		 755
756 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
757 if (ret < 0)
758 goto out;
759
760 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
761 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
762
763 ret = send_cmd(sctx);
764
765tlv_put_failure:
766out:
767 return ret;
768}
769
770/*
771 * Sends an unlink instruction to user space
772 */
773static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
774{
04ab956e 775 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 776 int ret;
777
04ab956e 778 btrfs_debug(fs_info, "send_unlink %s", path->start);
31db9f7c
AB		 779
780 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
781 if (ret < 0)
782 goto out;
783
784 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
785
786 ret = send_cmd(sctx);
787
788tlv_put_failure:
789out:
790 return ret;
791}
792
793/*
794 * Sends a rmdir instruction to user space
795 */
796static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
797{
04ab956e 798 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 799 int ret;
800
04ab956e 801 btrfs_debug(fs_info, "send_rmdir %s", path->start);
31db9f7c
AB		 802
803 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
804 if (ret < 0)
805 goto out;
806
807 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
808
809 ret = send_cmd(sctx);
810
811tlv_put_failure:
812out:
813 return ret;
814}
815
816/*
817 * Helper function to retrieve some fields from an inode item.
818 */
3f8a18cc
JB		 819static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
820 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
821 u64 *gid, u64 *rdev)
31db9f7c
AB		 822{
823 int ret;
824 struct btrfs_inode_item *ii;
825 struct btrfs_key key;
31db9f7c
AB		 826
827 key.objectid = ino;
828 key.type = BTRFS_INODE_ITEM_KEY;
829 key.offset = 0;
830 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
31db9f7c 831 if (ret) {
3f8a18cc
JB		 832 if (ret > 0)
833 ret = -ENOENT;
834 return ret;
31db9f7c
AB		 835 }
836
837 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
838 struct btrfs_inode_item);
839 if (size)
840 *size = btrfs_inode_size(path->nodes[0], ii);
841 if (gen)
842 *gen = btrfs_inode_generation(path->nodes[0], ii);
843 if (mode)
844 *mode = btrfs_inode_mode(path->nodes[0], ii);
845 if (uid)
846 *uid = btrfs_inode_uid(path->nodes[0], ii);
847 if (gid)
848 *gid = btrfs_inode_gid(path->nodes[0], ii);
85a7b33b
AB		 849 if (rdev)
850 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
31db9f7c 851
3f8a18cc
JB		 852 return ret;
853}
854
855static int get_inode_info(struct btrfs_root *root,
856 u64 ino, u64 *size, u64 *gen,
857 u64 *mode, u64 *uid, u64 *gid,
858 u64 *rdev)
859{
860 struct btrfs_path *path;
861 int ret;
862
863 path = alloc_path_for_send();
864 if (!path)
865 return -ENOMEM;
866 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
867 rdev);
31db9f7c
AB		 868 btrfs_free_path(path);
869 return ret;
870}
871
872typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
873 struct fs_path *p,
874 void *ctx);
875
876/*
96b5bd77
JS		 877 * Helper function to iterate the entries in ONE btrfs_inode_ref or
878 * btrfs_inode_extref.
31db9f7c
AB		 879 * The iterate callback may return a non zero value to stop iteration. This can
880 * be a negative value for error codes or 1 to simply stop it.
881 *
96b5bd77 882 * path must point to the INODE_REF or INODE_EXTREF when called.
31db9f7c 883 */
924794c9 884static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 885 struct btrfs_key *found_key, int resolve,
886 iterate_inode_ref_t iterate, void *ctx)
887{
96b5bd77 888 struct extent_buffer *eb = path->nodes[0];
31db9f7c
AB		 889 struct btrfs_item *item;
890 struct btrfs_inode_ref *iref;
96b5bd77 891 struct btrfs_inode_extref *extref;
31db9f7c
AB		 892 struct btrfs_path *tmp_path;
893 struct fs_path *p;
96b5bd77 894 u32 cur = 0;
31db9f7c 895 u32 total;
96b5bd77 896 int slot = path->slots[0];
31db9f7c
AB		 897 u32 name_len;
898 char *start;
899 int ret = 0;
96b5bd77 900 int num = 0;
31db9f7c 901 int index;
96b5bd77
JS		 902 u64 dir;
903 unsigned long name_off;
904 unsigned long elem_size;
905 unsigned long ptr;
31db9f7c 906
924794c9 907 p = fs_path_alloc_reversed();
31db9f7c
AB		 908 if (!p)
909 return -ENOMEM;
910
911 tmp_path = alloc_path_for_send();
912 if (!tmp_path) {
924794c9 913 fs_path_free(p);
31db9f7c
AB		 914 return -ENOMEM;
915 }
916
31db9f7c 917
96b5bd77
JS		 918 if (found_key->type == BTRFS_INODE_REF_KEY) {
919 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
920 struct btrfs_inode_ref);
dd3cc16b 921 item = btrfs_item_nr(slot);
96b5bd77
JS		 922 total = btrfs_item_size(eb, item);
923 elem_size = sizeof(*iref);
924 } else {
925 ptr = btrfs_item_ptr_offset(eb, slot);
926 total = btrfs_item_size_nr(eb, slot);
927 elem_size = sizeof(*extref);
928 }
929
31db9f7c
AB		 930 while (cur < total) {
931 fs_path_reset(p);
932
96b5bd77
JS		 933 if (found_key->type == BTRFS_INODE_REF_KEY) {
934 iref = (struct btrfs_inode_ref *)(ptr + cur);
935 name_len = btrfs_inode_ref_name_len(eb, iref);
936 name_off = (unsigned long)(iref + 1);
937 index = btrfs_inode_ref_index(eb, iref);
938 dir = found_key->offset;
939 } else {
940 extref = (struct btrfs_inode_extref *)(ptr + cur);
941 name_len = btrfs_inode_extref_name_len(eb, extref);
942 name_off = (unsigned long)&extref->name;
943 index = btrfs_inode_extref_index(eb, extref);
944 dir = btrfs_inode_extref_parent(eb, extref);
945 }
946
31db9f7c 947 if (resolve) {
96b5bd77
JS		 948 start = btrfs_ref_to_path(root, tmp_path, name_len,
949 name_off, eb, dir,
950 p->buf, p->buf_len);
31db9f7c
AB		 951 if (IS_ERR(start)) {
952 ret = PTR_ERR(start);
953 goto out;
954 }
955 if (start < p->buf) {
956 /* overflow , try again with larger buffer */
957 ret = fs_path_ensure_buf(p,
958 p->buf_len + p->buf - start);
959 if (ret < 0)
960 goto out;
96b5bd77
JS		 961 start = btrfs_ref_to_path(root, tmp_path,
962 name_len, name_off,
963 eb, dir,
964 p->buf, p->buf_len);
31db9f7c
AB		 965 if (IS_ERR(start)) {
966 ret = PTR_ERR(start);
967 goto out;
968 }
969 BUG_ON(start < p->buf);
970 }
971 p->start = start;
972 } else {
96b5bd77
JS		 973 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
974 name_len);
31db9f7c
AB		 975 if (ret < 0)
976 goto out;
977 }
978
96b5bd77
JS		 979 cur += elem_size + name_len;
980 ret = iterate(num, dir, index, p, ctx);
31db9f7c
AB		 981 if (ret)
982 goto out;
31db9f7c
AB		 983 num++;
984 }
985
986out:
987 btrfs_free_path(tmp_path);
924794c9 988 fs_path_free(p);
31db9f7c
AB		 989 return ret;
990}
991
992typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
993 const char *name, int name_len,
994 const char *data, int data_len,
995 u8 type, void *ctx);
996
997/*
998 * Helper function to iterate the entries in ONE btrfs_dir_item.
999 * The iterate callback may return a non zero value to stop iteration. This can
1000 * be a negative value for error codes or 1 to simply stop it.
1001 *
1002 * path must point to the dir item when called.
1003 */
924794c9 1004static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
31db9f7c
AB		 1005 struct btrfs_key *found_key,
1006 iterate_dir_item_t iterate, void *ctx)
1007{
1008 int ret = 0;
1009 struct extent_buffer *eb;
1010 struct btrfs_item *item;
1011 struct btrfs_dir_item *di;
31db9f7c
AB		 1012 struct btrfs_key di_key;
1013 char *buf = NULL;
7e3ae33e 1014 int buf_len;
31db9f7c
AB		 1015 u32 name_len;
1016 u32 data_len;
1017 u32 cur;
1018 u32 len;
1019 u32 total;
1020 int slot;
1021 int num;
1022 u8 type;
1023
4395e0c4
FM		 1024 /*
1025 * Start with a small buffer (1 page). If later we end up needing more
1026 * space, which can happen for xattrs on a fs with a leaf size greater
1027 * then the page size, attempt to increase the buffer. Typically xattr
1028 * values are small.
1029 */
1030 buf_len = PATH_MAX;
e780b0d1 1031 buf = kmalloc(buf_len, GFP_KERNEL);
31db9f7c
AB		 1032 if (!buf) {
1033 ret = -ENOMEM;
1034 goto out;
1035 }
1036
31db9f7c
AB		 1037 eb = path->nodes[0];
1038 slot = path->slots[0];
dd3cc16b 1039 item = btrfs_item_nr(slot);
31db9f7c
AB		 1040 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1041 cur = 0;
1042 len = 0;
1043 total = btrfs_item_size(eb, item);
1044
1045 num = 0;
1046 while (cur < total) {
1047 name_len = btrfs_dir_name_len(eb, di);
1048 data_len = btrfs_dir_data_len(eb, di);
1049 type = btrfs_dir_type(eb, di);
1050 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1051
7e3ae33e
FM		 1052 if (type == BTRFS_FT_XATTR) {
1053 if (name_len > XATTR_NAME_MAX) {
1054 ret = -ENAMETOOLONG;
1055 goto out;
1056 }
da17066c
JM		 1057 if (name_len + data_len >
1058 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
7e3ae33e
FM		 1059 ret = -E2BIG;
1060 goto out;
1061 }
1062 } else {
1063 /*
1064 * Path too long
1065 */
4395e0c4 1066 if (name_len + data_len > PATH_MAX) {
7e3ae33e
FM		 1067 ret = -ENAMETOOLONG;
1068 goto out;
1069 }
31db9f7c
AB		 1070 }
1071
4395e0c4
FM		 1072 if (name_len + data_len > buf_len) {
1073 buf_len = name_len + data_len;
1074 if (is_vmalloc_addr(buf)) {
1075 vfree(buf);
1076 buf = NULL;
1077 } else {
1078 char *tmp = krealloc(buf, buf_len,
e780b0d1 1079 GFP_KERNEL | __GFP_NOWARN);
4395e0c4
FM		 1080
1081 if (!tmp)
1082 kfree(buf);
1083 buf = tmp;
1084 }
1085 if (!buf) {
1086 buf = vmalloc(buf_len);
1087 if (!buf) {
1088 ret = -ENOMEM;
1089 goto out;
1090 }
1091 }
1092 }
1093
31db9f7c
AB		 1094 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1095 name_len + data_len);
1096
1097 len = sizeof(*di) + name_len + data_len;
1098 di = (struct btrfs_dir_item *)((char *)di + len);
1099 cur += len;
1100
1101 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1102 data_len, type, ctx);
1103 if (ret < 0)
1104 goto out;
1105 if (ret) {
1106 ret = 0;
1107 goto out;
1108 }
1109
1110 num++;
1111 }
1112
1113out:
4395e0c4 1114 kvfree(buf);
31db9f7c
AB		 1115 return ret;
1116}
1117
1118static int __copy_first_ref(int num, u64 dir, int index,
1119 struct fs_path *p, void *ctx)
1120{
1121 int ret;
1122 struct fs_path *pt = ctx;
1123
1124 ret = fs_path_copy(pt, p);
1125 if (ret < 0)
1126 return ret;
1127
1128 /* we want the first only */
1129 return 1;
1130}
1131
1132/*
1133 * Retrieve the first path of an inode. If an inode has more then one
1134 * ref/hardlink, this is ignored.
1135 */
924794c9 1136static int get_inode_path(struct btrfs_root *root,
31db9f7c
AB		 1137 u64 ino, struct fs_path *path)
1138{
1139 int ret;
1140 struct btrfs_key key, found_key;
1141 struct btrfs_path *p;
1142
1143 p = alloc_path_for_send();
1144 if (!p)
1145 return -ENOMEM;
1146
1147 fs_path_reset(path);
1148
1149 key.objectid = ino;
1150 key.type = BTRFS_INODE_REF_KEY;
1151 key.offset = 0;
1152
1153 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1154 if (ret < 0)
1155 goto out;
1156 if (ret) {
1157 ret = 1;
1158 goto out;
1159 }
1160 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1161 if (found_key.objectid != ino ||
96b5bd77
JS		 1162 (found_key.type != BTRFS_INODE_REF_KEY &&
1163 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1164 ret = -ENOENT;
1165 goto out;
1166 }
1167
924794c9
TI		 1168 ret = iterate_inode_ref(root, p, &found_key, 1,
1169 __copy_first_ref, path);
31db9f7c
AB		 1170 if (ret < 0)
1171 goto out;
1172 ret = 0;
1173
1174out:
1175 btrfs_free_path(p);
1176 return ret;
1177}
1178
1179struct backref_ctx {
1180 struct send_ctx *sctx;
1181
3f8a18cc 1182 struct btrfs_path *path;
31db9f7c
AB		 1183 /* number of total found references */
1184 u64 found;
1185
1186 /*
1187 * used for clones found in send_root. clones found behind cur_objectid
1188 * and cur_offset are not considered as allowed clones.
1189 */
1190 u64 cur_objectid;
1191 u64 cur_offset;
1192
1193 /* may be truncated in case it's the last extent in a file */
1194 u64 extent_len;
1195
619d8c4e
FM		 1196 /* data offset in the file extent item */
1197 u64 data_offset;
1198
31db9f7c 1199 /* Just to check for bugs in backref resolving */
ee849c04 1200 int found_itself;
31db9f7c
AB		 1201};
1202
1203static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1204{
995e01b7 1205 u64 root = (u64)(uintptr_t)key;
31db9f7c
AB		 1206 struct clone_root *cr = (struct clone_root *)elt;
1207
1208 if (root < cr->root->objectid)
1209 return -1;
1210 if (root > cr->root->objectid)
1211 return 1;
1212 return 0;
1213}
1214
1215static int __clone_root_cmp_sort(const void *e1, const void *e2)
1216{
1217 struct clone_root *cr1 = (struct clone_root *)e1;
1218 struct clone_root *cr2 = (struct clone_root *)e2;
1219
1220 if (cr1->root->objectid < cr2->root->objectid)
1221 return -1;
1222 if (cr1->root->objectid > cr2->root->objectid)
1223 return 1;
1224 return 0;
1225}
1226
1227/*
1228 * Called for every backref that is found for the current extent.
766702ef 1229 * Results are collected in sctx->clone_roots->ino/offset/found_refs
31db9f7c
AB		 1230 */
1231static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1232{
1233 struct backref_ctx *bctx = ctx_;
1234 struct clone_root *found;
1235 int ret;
1236 u64 i_size;
1237
1238 /* First check if the root is in the list of accepted clone sources */
995e01b7 1239 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
31db9f7c
AB		 1240 bctx->sctx->clone_roots_cnt,
1241 sizeof(struct clone_root),
1242 __clone_root_cmp_bsearch);
1243 if (!found)
1244 return 0;
1245
1246 if (found->root == bctx->sctx->send_root &&
1247 ino == bctx->cur_objectid &&
1248 offset == bctx->cur_offset) {
ee849c04 1249 bctx->found_itself = 1;
31db9f7c
AB		 1250 }
1251
1252 /*
766702ef 1253 * There are inodes that have extents that lie behind its i_size. Don't
31db9f7c
AB		 1254 * accept clones from these extents.
1255 */
3f8a18cc
JB		 1256 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1257 NULL, NULL, NULL);
1258 btrfs_release_path(bctx->path);
31db9f7c
AB		 1259 if (ret < 0)
1260 return ret;
1261
619d8c4e 1262 if (offset + bctx->data_offset + bctx->extent_len > i_size)
31db9f7c
AB		 1263 return 0;
1264
1265 /*
1266 * Make sure we don't consider clones from send_root that are
1267 * behind the current inode/offset.
1268 */
1269 if (found->root == bctx->sctx->send_root) {
1270 /*
1271 * TODO for the moment we don't accept clones from the inode
1272 * that is currently send. We may change this when
1273 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1274 * file.
1275 */
1276 if (ino >= bctx->cur_objectid)
1277 return 0;
e938c8ad
AB		 1278#if 0
1279 if (ino > bctx->cur_objectid)
1280 return 0;
1281 if (offset + bctx->extent_len > bctx->cur_offset)
31db9f7c 1282 return 0;
e938c8ad 1283#endif
31db9f7c
AB		 1284 }
1285
1286 bctx->found++;
1287 found->found_refs++;
1288 if (ino < found->ino) {
1289 found->ino = ino;
1290 found->offset = offset;
1291 } else if (found->ino == ino) {
1292 /*
1293 * same extent found more then once in the same file.
1294 */
1295 if (found->offset > offset + bctx->extent_len)
1296 found->offset = offset;
1297 }
1298
1299 return 0;
1300}
1301
1302/*
766702ef
AB		 1303 * Given an inode, offset and extent item, it finds a good clone for a clone
1304 * instruction. Returns -ENOENT when none could be found. The function makes
1305 * sure that the returned clone is usable at the point where sending is at the
1306 * moment. This means, that no clones are accepted which lie behind the current
1307 * inode+offset.
1308 *
31db9f7c
AB		 1309 * path must point to the extent item when called.
1310 */
1311static int find_extent_clone(struct send_ctx *sctx,
1312 struct btrfs_path *path,
1313 u64 ino, u64 data_offset,
1314 u64 ino_size,
1315 struct clone_root **found)
1316{
04ab956e 1317 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 1318 int ret;
1319 int extent_type;
1320 u64 logical;
74dd17fb 1321 u64 disk_byte;
31db9f7c
AB		 1322 u64 num_bytes;
1323 u64 extent_item_pos;
69917e43 1324 u64 flags = 0;
31db9f7c
AB		 1325 struct btrfs_file_extent_item *fi;
1326 struct extent_buffer *eb = path->nodes[0];
35075bb0 1327 struct backref_ctx *backref_ctx = NULL;
31db9f7c
AB		 1328 struct clone_root *cur_clone_root;
1329 struct btrfs_key found_key;
1330 struct btrfs_path *tmp_path;
74dd17fb 1331 int compressed;
31db9f7c
AB		 1332 u32 i;
1333
1334 tmp_path = alloc_path_for_send();
1335 if (!tmp_path)
1336 return -ENOMEM;
1337
3f8a18cc
JB		 1338 /* We only use this path under the commit sem */
1339 tmp_path->need_commit_sem = 0;
1340
e780b0d1 1341 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
35075bb0
AB		 1342 if (!backref_ctx) {
1343 ret = -ENOMEM;
1344 goto out;
1345 }
1346
3f8a18cc
JB		 1347 backref_ctx->path = tmp_path;
1348
31db9f7c
AB		 1349 if (data_offset >= ino_size) {
1350 /*
1351 * There may be extents that lie behind the file's size.
1352 * I at least had this in combination with snapshotting while
1353 * writing large files.
1354 */
1355 ret = 0;
1356 goto out;
1357 }
1358
1359 fi = btrfs_item_ptr(eb, path->slots[0],
1360 struct btrfs_file_extent_item);
1361 extent_type = btrfs_file_extent_type(eb, fi);
1362 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1363 ret = -ENOENT;
1364 goto out;
1365 }
74dd17fb 1366 compressed = btrfs_file_extent_compression(eb, fi);
31db9f7c
AB		 1367
1368 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
74dd17fb
CM		 1369 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1370 if (disk_byte == 0) {
31db9f7c
AB		 1371 ret = -ENOENT;
1372 goto out;
1373 }
74dd17fb 1374 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
31db9f7c 1375
04ab956e
JM		 1376 down_read(&fs_info->commit_root_sem);
1377 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
69917e43 1378 &found_key, &flags);
04ab956e 1379 up_read(&fs_info->commit_root_sem);
31db9f7c
AB		 1380 btrfs_release_path(tmp_path);
1381
1382 if (ret < 0)
1383 goto out;
69917e43 1384 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
31db9f7c
AB		 1385 ret = -EIO;
1386 goto out;
1387 }
1388
1389 /*
1390 * Setup the clone roots.
1391 */
1392 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1393 cur_clone_root = sctx->clone_roots + i;
1394 cur_clone_root->ino = (u64)-1;
1395 cur_clone_root->offset = 0;
1396 cur_clone_root->found_refs = 0;
1397 }
1398
35075bb0
AB		 1399 backref_ctx->sctx = sctx;
1400 backref_ctx->found = 0;
1401 backref_ctx->cur_objectid = ino;
1402 backref_ctx->cur_offset = data_offset;
1403 backref_ctx->found_itself = 0;
1404 backref_ctx->extent_len = num_bytes;
619d8c4e
FM		 1405 /*
1406 * For non-compressed extents iterate_extent_inodes() gives us extent
1407 * offsets that already take into account the data offset, but not for
1408 * compressed extents, since the offset is logical and not relative to
1409 * the physical extent locations. We must take this into account to
1410 * avoid sending clone offsets that go beyond the source file's size,
1411 * which would result in the clone ioctl failing with -EINVAL on the
1412 * receiving end.
1413 */
1414 if (compressed == BTRFS_COMPRESS_NONE)
1415 backref_ctx->data_offset = 0;
1416 else
1417 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
31db9f7c
AB		 1418
1419 /*
1420 * The last extent of a file may be too large due to page alignment.
1421 * We need to adjust extent_len in this case so that the checks in
1422 * __iterate_backrefs work.
1423 */
1424 if (data_offset + num_bytes >= ino_size)
35075bb0 1425 backref_ctx->extent_len = ino_size - data_offset;
31db9f7c
AB		 1426
1427 /*
1428 * Now collect all backrefs.
1429 */
74dd17fb
CM		 1430 if (compressed == BTRFS_COMPRESS_NONE)
1431 extent_item_pos = logical - found_key.objectid;
1432 else
1433 extent_item_pos = 0;
04ab956e 1434 ret = iterate_extent_inodes(fs_info,
31db9f7c 1435 found_key.objectid, extent_item_pos, 1,
35075bb0 1436 __iterate_backrefs, backref_ctx);
74dd17fb 1437
31db9f7c
AB		 1438 if (ret < 0)
1439 goto out;
1440
35075bb0 1441 if (!backref_ctx->found_itself) {
31db9f7c
AB		 1442 /* found a bug in backref code? */
1443 ret = -EIO;
04ab956e 1444 btrfs_err(fs_info,
5d163e0e 1445 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
04ab956e 1446 ino, data_offset, disk_byte, found_key.objectid);
31db9f7c
AB		 1447 goto out;
1448 }
1449
04ab956e
JM		 1450 btrfs_debug(fs_info,
1451 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1452 data_offset, ino, num_bytes, logical);
31db9f7c 1453
35075bb0 1454 if (!backref_ctx->found)
04ab956e 1455 btrfs_debug(fs_info, "no clones found");
31db9f7c
AB		 1456
1457 cur_clone_root = NULL;
1458 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1459 if (sctx->clone_roots[i].found_refs) {
1460 if (!cur_clone_root)
1461 cur_clone_root = sctx->clone_roots + i;
1462 else if (sctx->clone_roots[i].root == sctx->send_root)
1463 /* prefer clones from send_root over others */
1464 cur_clone_root = sctx->clone_roots + i;
31db9f7c
AB		 1465 }
1466
1467 }
1468
1469 if (cur_clone_root) {
1470 *found = cur_clone_root;
1471 ret = 0;
1472 } else {
1473 ret = -ENOENT;
1474 }
1475
1476out:
1477 btrfs_free_path(tmp_path);
35075bb0 1478 kfree(backref_ctx);
31db9f7c
AB		 1479 return ret;
1480}
1481
924794c9 1482static int read_symlink(struct btrfs_root *root,
31db9f7c
AB		 1483 u64 ino,
1484 struct fs_path *dest)
1485{
1486 int ret;
1487 struct btrfs_path *path;
1488 struct btrfs_key key;
1489 struct btrfs_file_extent_item *ei;
1490 u8 type;
1491 u8 compression;
1492 unsigned long off;
1493 int len;
1494
1495 path = alloc_path_for_send();
1496 if (!path)
1497 return -ENOMEM;
1498
1499 key.objectid = ino;
1500 key.type = BTRFS_EXTENT_DATA_KEY;
1501 key.offset = 0;
1502 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1503 if (ret < 0)
1504 goto out;
a879719b
FM		 1505 if (ret) {
1506 /*
1507 * An empty symlink inode. Can happen in rare error paths when
1508 * creating a symlink (transaction committed before the inode
1509 * eviction handler removed the symlink inode items and a crash
1510 * happened in between or the subvol was snapshoted in between).
1511 * Print an informative message to dmesg/syslog so that the user
1512 * can delete the symlink.
1513 */
1514 btrfs_err(root->fs_info,
1515 "Found empty symlink inode %llu at root %llu",
1516 ino, root->root_key.objectid);
1517 ret = -EIO;
1518 goto out;
1519 }
31db9f7c
AB		 1520
1521 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1522 struct btrfs_file_extent_item);
1523 type = btrfs_file_extent_type(path->nodes[0], ei);
1524 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1525 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1526 BUG_ON(compression);
1527
1528 off = btrfs_file_extent_inline_start(ei);
514ac8ad 1529 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
31db9f7c
AB		 1530
1531 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
31db9f7c
AB		 1532
1533out:
1534 btrfs_free_path(path);
1535 return ret;
1536}
1537
1538/*
1539 * Helper function to generate a file name that is unique in the root of
1540 * send_root and parent_root. This is used to generate names for orphan inodes.
1541 */
1542static int gen_unique_name(struct send_ctx *sctx,
1543 u64 ino, u64 gen,
1544 struct fs_path *dest)
1545{
1546 int ret = 0;
1547 struct btrfs_path *path;
1548 struct btrfs_dir_item *di;
1549 char tmp[64];
1550 int len;
1551 u64 idx = 0;
1552
1553 path = alloc_path_for_send();
1554 if (!path)
1555 return -ENOMEM;
1556
1557 while (1) {
f74b86d8 1558 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
31db9f7c 1559 ino, gen, idx);
64792f25 1560 ASSERT(len < sizeof(tmp));
31db9f7c
AB		 1561
1562 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1563 path, BTRFS_FIRST_FREE_OBJECTID,
1564 tmp, strlen(tmp), 0);
1565 btrfs_release_path(path);
1566 if (IS_ERR(di)) {
1567 ret = PTR_ERR(di);
1568 goto out;
1569 }
1570 if (di) {
1571 /* not unique, try again */
1572 idx++;
1573 continue;
1574 }
1575
1576 if (!sctx->parent_root) {
1577 /* unique */
1578 ret = 0;
1579 break;
1580 }
1581
1582 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1583 path, BTRFS_FIRST_FREE_OBJECTID,
1584 tmp, strlen(tmp), 0);
1585 btrfs_release_path(path);
1586 if (IS_ERR(di)) {
1587 ret = PTR_ERR(di);
1588 goto out;
1589 }
1590 if (di) {
1591 /* not unique, try again */
1592 idx++;
1593 continue;
1594 }
1595 /* unique */
1596 break;
1597 }
1598
1599 ret = fs_path_add(dest, tmp, strlen(tmp));
1600
1601out:
1602 btrfs_free_path(path);
1603 return ret;
1604}
1605
1606enum inode_state {
1607 inode_state_no_change,
1608 inode_state_will_create,
1609 inode_state_did_create,
1610 inode_state_will_delete,
1611 inode_state_did_delete,
1612};
1613
1614static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1615{
1616 int ret;
1617 int left_ret;
1618 int right_ret;
1619 u64 left_gen;
1620 u64 right_gen;
1621
1622 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
85a7b33b 1623 NULL, NULL);
31db9f7c
AB		 1624 if (ret < 0 && ret != -ENOENT)
1625 goto out;
1626 left_ret = ret;
1627
1628 if (!sctx->parent_root) {
1629 right_ret = -ENOENT;
1630 } else {
1631 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
85a7b33b 1632 NULL, NULL, NULL, NULL);
31db9f7c
AB		 1633 if (ret < 0 && ret != -ENOENT)
1634 goto out;
1635 right_ret = ret;
1636 }
1637
1638 if (!left_ret && !right_ret) {
e938c8ad 1639 if (left_gen == gen && right_gen == gen) {
31db9f7c 1640 ret = inode_state_no_change;
e938c8ad 1641 } else if (left_gen == gen) {
31db9f7c
AB		 1642 if (ino < sctx->send_progress)
1643 ret = inode_state_did_create;
1644 else
1645 ret = inode_state_will_create;
1646 } else if (right_gen == gen) {
1647 if (ino < sctx->send_progress)
1648 ret = inode_state_did_delete;
1649 else
1650 ret = inode_state_will_delete;
1651 } else {
1652 ret = -ENOENT;
1653 }
1654 } else if (!left_ret) {
1655 if (left_gen == gen) {
1656 if (ino < sctx->send_progress)
1657 ret = inode_state_did_create;
1658 else
1659 ret = inode_state_will_create;
1660 } else {
1661 ret = -ENOENT;
1662 }
1663 } else if (!right_ret) {
1664 if (right_gen == gen) {
1665 if (ino < sctx->send_progress)
1666 ret = inode_state_did_delete;
1667 else
1668 ret = inode_state_will_delete;
1669 } else {
1670 ret = -ENOENT;
1671 }
1672 } else {
1673 ret = -ENOENT;
1674 }
1675
1676out:
1677 return ret;
1678}
1679
1680static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1681{
1682 int ret;
1683
1684 ret = get_cur_inode_state(sctx, ino, gen);
1685 if (ret < 0)
1686 goto out;
1687
1688 if (ret == inode_state_no_change ||
1689 ret == inode_state_did_create ||
1690 ret == inode_state_will_delete)
1691 ret = 1;
1692 else
1693 ret = 0;
1694
1695out:
1696 return ret;
1697}
1698
1699/*
1700 * Helper function to lookup a dir item in a dir.
1701 */
1702static int lookup_dir_item_inode(struct btrfs_root *root,
1703 u64 dir, const char *name, int name_len,
1704 u64 *found_inode,
1705 u8 *found_type)
1706{
1707 int ret = 0;
1708 struct btrfs_dir_item *di;
1709 struct btrfs_key key;
1710 struct btrfs_path *path;
1711
1712 path = alloc_path_for_send();
1713 if (!path)
1714 return -ENOMEM;
1715
1716 di = btrfs_lookup_dir_item(NULL, root, path,
1717 dir, name, name_len, 0);
1718 if (!di) {
1719 ret = -ENOENT;
1720 goto out;
1721 }
1722 if (IS_ERR(di)) {
1723 ret = PTR_ERR(di);
1724 goto out;
1725 }
1726 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1af56070
FM		 1727 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1728 ret = -ENOENT;
1729 goto out;
1730 }
31db9f7c
AB		 1731 *found_inode = key.objectid;
1732 *found_type = btrfs_dir_type(path->nodes[0], di);
1733
1734out:
1735 btrfs_free_path(path);
1736 return ret;
1737}
1738
766702ef
AB		 1739/*
1740 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1741 * generation of the parent dir and the name of the dir entry.
1742 */
924794c9 1743static int get_first_ref(struct btrfs_root *root, u64 ino,
31db9f7c
AB		 1744 u64 *dir, u64 *dir_gen, struct fs_path *name)
1745{
1746 int ret;
1747 struct btrfs_key key;
1748 struct btrfs_key found_key;
1749 struct btrfs_path *path;
31db9f7c 1750 int len;
96b5bd77 1751 u64 parent_dir;
31db9f7c
AB		 1752
1753 path = alloc_path_for_send();
1754 if (!path)
1755 return -ENOMEM;
1756
1757 key.objectid = ino;
1758 key.type = BTRFS_INODE_REF_KEY;
1759 key.offset = 0;
1760
1761 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1762 if (ret < 0)
1763 goto out;
1764 if (!ret)
1765 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1766 path->slots[0]);
96b5bd77
JS		 1767 if (ret || found_key.objectid != ino ||
1768 (found_key.type != BTRFS_INODE_REF_KEY &&
1769 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
31db9f7c
AB		 1770 ret = -ENOENT;
1771 goto out;
1772 }
1773
51a60253 1774 if (found_key.type == BTRFS_INODE_REF_KEY) {
96b5bd77
JS		 1775 struct btrfs_inode_ref *iref;
1776 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1777 struct btrfs_inode_ref);
1778 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1779 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1780 (unsigned long)(iref + 1),
1781 len);
1782 parent_dir = found_key.offset;
1783 } else {
1784 struct btrfs_inode_extref *extref;
1785 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1786 struct btrfs_inode_extref);
1787 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1788 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1789 (unsigned long)&extref->name, len);
1790 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1791 }
31db9f7c
AB		 1792 if (ret < 0)
1793 goto out;
1794 btrfs_release_path(path);
1795
b46ab97b
FM		 1796 if (dir_gen) {
1797 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1798 NULL, NULL, NULL);
1799 if (ret < 0)
1800 goto out;
1801 }
31db9f7c 1802
96b5bd77 1803 *dir = parent_dir;
31db9f7c
AB		 1804
1805out:
1806 btrfs_free_path(path);
1807 return ret;
1808}
1809
924794c9 1810static int is_first_ref(struct btrfs_root *root,
31db9f7c
AB		 1811 u64 ino, u64 dir,
1812 const char *name, int name_len)
1813{
1814 int ret;
1815 struct fs_path *tmp_name;
1816 u64 tmp_dir;
31db9f7c 1817
924794c9 1818 tmp_name = fs_path_alloc();
31db9f7c
AB		 1819 if (!tmp_name)
1820 return -ENOMEM;
1821
b46ab97b 1822 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
31db9f7c
AB		 1823 if (ret < 0)
1824 goto out;
1825
b9291aff 1826 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
31db9f7c
AB		 1827 ret = 0;
1828 goto out;
1829 }
1830
e938c8ad 1831 ret = !memcmp(tmp_name->start, name, name_len);
31db9f7c
AB		 1832
1833out:
924794c9 1834 fs_path_free(tmp_name);
31db9f7c
AB		 1835 return ret;
1836}
1837
766702ef
AB		 1838/*
1839 * Used by process_recorded_refs to determine if a new ref would overwrite an
1840 * already existing ref. In case it detects an overwrite, it returns the
1841 * inode/gen in who_ino/who_gen.
1842 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1843 * to make sure later references to the overwritten inode are possible.
1844 * Orphanizing is however only required for the first ref of an inode.
1845 * process_recorded_refs does an additional is_first_ref check to see if
1846 * orphanizing is really required.
1847 */
31db9f7c
AB		 1848static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1849 const char *name, int name_len,
1850 u64 *who_ino, u64 *who_gen)
1851{
1852 int ret = 0;
ebdad913 1853 u64 gen;
31db9f7c
AB		 1854 u64 other_inode = 0;
1855 u8 other_type = 0;
1856
1857 if (!sctx->parent_root)
1858 goto out;
1859
1860 ret = is_inode_existent(sctx, dir, dir_gen);
1861 if (ret <= 0)
1862 goto out;
1863
ebdad913
JB		 1864 /*
1865 * If we have a parent root we need to verify that the parent dir was
01327610 1866 * not deleted and then re-created, if it was then we have no overwrite
ebdad913
JB		 1867 * and we can just unlink this entry.
1868 */
1869 if (sctx->parent_root) {
1870 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1871 NULL, NULL, NULL);
1872 if (ret < 0 && ret != -ENOENT)
1873 goto out;
1874 if (ret) {
1875 ret = 0;
1876 goto out;
1877 }
1878 if (gen != dir_gen)
1879 goto out;
1880 }
1881
31db9f7c
AB		 1882 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1883 &other_inode, &other_type);
1884 if (ret < 0 && ret != -ENOENT)
1885 goto out;
1886 if (ret) {
1887 ret = 0;
1888 goto out;
1889 }
1890
766702ef
AB		 1891 /*
1892 * Check if the overwritten ref was already processed. If yes, the ref
1893 * was already unlinked/moved, so we can safely assume that we will not
1894 * overwrite anything at this point in time.
1895 */
801bec36
RK		 1896 if (other_inode > sctx->send_progress ||
1897 is_waiting_for_move(sctx, other_inode)) {
31db9f7c 1898 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
85a7b33b 1899 who_gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 1900 if (ret < 0)
1901 goto out;
1902
1903 ret = 1;
1904 *who_ino = other_inode;
1905 } else {
1906 ret = 0;
1907 }
1908
1909out:
1910 return ret;
1911}
1912
766702ef
AB		 1913/*
1914 * Checks if the ref was overwritten by an already processed inode. This is
1915 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1916 * thus the orphan name needs be used.
1917 * process_recorded_refs also uses it to avoid unlinking of refs that were
1918 * overwritten.
1919 */
31db9f7c
AB		 1920static int did_overwrite_ref(struct send_ctx *sctx,
1921 u64 dir, u64 dir_gen,
1922 u64 ino, u64 ino_gen,
1923 const char *name, int name_len)
1924{
1925 int ret = 0;
1926 u64 gen;
1927 u64 ow_inode;
1928 u8 other_type;
1929
1930 if (!sctx->parent_root)
1931 goto out;
1932
1933 ret = is_inode_existent(sctx, dir, dir_gen);
1934 if (ret <= 0)
1935 goto out;
1936
1937 /* check if the ref was overwritten by another ref */
1938 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1939 &ow_inode, &other_type);
1940 if (ret < 0 && ret != -ENOENT)
1941 goto out;
1942 if (ret) {
1943 /* was never and will never be overwritten */
1944 ret = 0;
1945 goto out;
1946 }
1947
1948 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
85a7b33b 1949 NULL, NULL);
31db9f7c
AB		 1950 if (ret < 0)
1951 goto out;
1952
1953 if (ow_inode == ino && gen == ino_gen) {
1954 ret = 0;
1955 goto out;
1956 }
1957
8b191a68
FM		 1958 /*
1959 * We know that it is or will be overwritten. Check this now.
1960 * The current inode being processed might have been the one that caused
b786f16a
FM		 1961 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1962 * the current inode being processed.
8b191a68 1963 */
b786f16a
FM		 1964 if ((ow_inode < sctx->send_progress) ||
1965 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1966 gen == sctx->cur_inode_gen))
31db9f7c
AB		 1967 ret = 1;
1968 else
1969 ret = 0;
1970
1971out:
1972 return ret;
1973}
1974
766702ef
AB		 1975/*
1976 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1977 * that got overwritten. This is used by process_recorded_refs to determine
1978 * if it has to use the path as returned by get_cur_path or the orphan name.
1979 */
31db9f7c
AB		 1980static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1981{
1982 int ret = 0;
1983 struct fs_path *name = NULL;
1984 u64 dir;
1985 u64 dir_gen;
1986
1987 if (!sctx->parent_root)
1988 goto out;
1989
924794c9 1990 name = fs_path_alloc();
31db9f7c
AB		 1991 if (!name)
1992 return -ENOMEM;
1993
924794c9 1994 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
31db9f7c
AB		 1995 if (ret < 0)
1996 goto out;
1997
1998 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1999 name->start, fs_path_len(name));
31db9f7c
AB		 2000
2001out:
924794c9 2002 fs_path_free(name);
31db9f7c
AB		 2003 return ret;
2004}
2005
766702ef
AB		 2006/*
2007 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2008 * so we need to do some special handling in case we have clashes. This function
2009 * takes care of this with the help of name_cache_entry::radix_list.
5dc67d0b 2010 * In case of error, nce is kfreed.
766702ef 2011 */
31db9f7c
AB		 2012static int name_cache_insert(struct send_ctx *sctx,
2013 struct name_cache_entry *nce)
2014{
2015 int ret = 0;
7e0926fe
AB		 2016 struct list_head *nce_head;
2017
2018 nce_head = radix_tree_lookup(&sctx->name_cache,
2019 (unsigned long)nce->ino);
2020 if (!nce_head) {
e780b0d1 2021 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
cfa7a9cc
TI		 2022 if (!nce_head) {
2023 kfree(nce);
31db9f7c 2024 return -ENOMEM;
cfa7a9cc 2025 }
7e0926fe 2026 INIT_LIST_HEAD(nce_head);
31db9f7c 2027
7e0926fe 2028 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
5dc67d0b
AB		 2029 if (ret < 0) {
2030 kfree(nce_head);
2031 kfree(nce);
31db9f7c 2032 return ret;
5dc67d0b 2033 }
31db9f7c 2034 }
7e0926fe 2035 list_add_tail(&nce->radix_list, nce_head);
31db9f7c
AB		 2036 list_add_tail(&nce->list, &sctx->name_cache_list);
2037 sctx->name_cache_size++;
2038
2039 return ret;
2040}
2041
2042static void name_cache_delete(struct send_ctx *sctx,
2043 struct name_cache_entry *nce)
2044{
7e0926fe 2045 struct list_head *nce_head;
31db9f7c 2046
7e0926fe
AB		 2047 nce_head = radix_tree_lookup(&sctx->name_cache,
2048 (unsigned long)nce->ino);
57fb8910
DS		 2049 if (!nce_head) {
2050 btrfs_err(sctx->send_root->fs_info,
2051 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2052 nce->ino, sctx->name_cache_size);
2053 }
31db9f7c 2054
7e0926fe 2055 list_del(&nce->radix_list);
31db9f7c 2056 list_del(&nce->list);
31db9f7c 2057 sctx->name_cache_size--;
7e0926fe 2058
57fb8910
DS		 2059 /*
2060 * We may not get to the final release of nce_head if the lookup fails
2061 */
2062 if (nce_head && list_empty(nce_head)) {
7e0926fe
AB		 2063 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2064 kfree(nce_head);
2065 }
31db9f7c
AB		 2066}
2067
2068static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2069 u64 ino, u64 gen)
2070{
7e0926fe
AB		 2071 struct list_head *nce_head;
2072 struct name_cache_entry *cur;
31db9f7c 2073
7e0926fe
AB		 2074 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2075 if (!nce_head)
31db9f7c
AB		 2076 return NULL;
2077
7e0926fe
AB		 2078 list_for_each_entry(cur, nce_head, radix_list) {
2079 if (cur->ino == ino && cur->gen == gen)
2080 return cur;
2081 }
31db9f7c
AB		 2082 return NULL;
2083}
2084
766702ef
AB		 2085/*
2086 * Removes the entry from the list and adds it back to the end. This marks the
2087 * entry as recently used so that name_cache_clean_unused does not remove it.
2088 */
31db9f7c
AB		 2089static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2090{
2091 list_del(&nce->list);
2092 list_add_tail(&nce->list, &sctx->name_cache_list);
2093}
2094
766702ef
AB		 2095/*
2096 * Remove some entries from the beginning of name_cache_list.
2097 */
31db9f7c
AB		 2098static void name_cache_clean_unused(struct send_ctx *sctx)
2099{
2100 struct name_cache_entry *nce;
2101
2102 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2103 return;
2104
2105 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2106 nce = list_entry(sctx->name_cache_list.next,
2107 struct name_cache_entry, list);
2108 name_cache_delete(sctx, nce);
2109 kfree(nce);
2110 }
2111}
2112
2113static void name_cache_free(struct send_ctx *sctx)
2114{
2115 struct name_cache_entry *nce;
31db9f7c 2116
e938c8ad
AB		 2117 while (!list_empty(&sctx->name_cache_list)) {
2118 nce = list_entry(sctx->name_cache_list.next,
2119 struct name_cache_entry, list);
31db9f7c 2120 name_cache_delete(sctx, nce);
17589bd9 2121 kfree(nce);
31db9f7c
AB		 2122 }
2123}
2124
766702ef
AB		 2125/*
2126 * Used by get_cur_path for each ref up to the root.
2127 * Returns 0 if it succeeded.
2128 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2129 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2130 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2131 * Returns <0 in case of error.
2132 */
31db9f7c
AB		 2133static int __get_cur_name_and_parent(struct send_ctx *sctx,
2134 u64 ino, u64 gen,
2135 u64 *parent_ino,
2136 u64 *parent_gen,
2137 struct fs_path *dest)
2138{
2139 int ret;
2140 int nce_ret;
31db9f7c
AB		 2141 struct name_cache_entry *nce = NULL;
2142
766702ef
AB		 2143 /*
2144 * First check if we already did a call to this function with the same
2145 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2146 * return the cached result.
2147 */
31db9f7c
AB		 2148 nce = name_cache_search(sctx, ino, gen);
2149 if (nce) {
2150 if (ino < sctx->send_progress && nce->need_later_update) {
2151 name_cache_delete(sctx, nce);
2152 kfree(nce);
2153 nce = NULL;
2154 } else {
2155 name_cache_used(sctx, nce);
2156 *parent_ino = nce->parent_ino;
2157 *parent_gen = nce->parent_gen;
2158 ret = fs_path_add(dest, nce->name, nce->name_len);
2159 if (ret < 0)
2160 goto out;
2161 ret = nce->ret;
2162 goto out;
2163 }
2164 }
2165
766702ef
AB		 2166 /*
2167 * If the inode is not existent yet, add the orphan name and return 1.
2168 * This should only happen for the parent dir that we determine in
2169 * __record_new_ref
2170 */
31db9f7c
AB		 2171 ret = is_inode_existent(sctx, ino, gen);
2172 if (ret < 0)
2173 goto out;
2174
2175 if (!ret) {
2176 ret = gen_unique_name(sctx, ino, gen, dest);
2177 if (ret < 0)
2178 goto out;
2179 ret = 1;
2180 goto out_cache;
2181 }
2182
766702ef
AB		 2183 /*
2184 * Depending on whether the inode was already processed or not, use
2185 * send_root or parent_root for ref lookup.
2186 */
bf0d1f44 2187 if (ino < sctx->send_progress)
924794c9
TI		 2188 ret = get_first_ref(sctx->send_root, ino,
2189 parent_ino, parent_gen, dest);
31db9f7c 2190 else
924794c9
TI		 2191 ret = get_first_ref(sctx->parent_root, ino,
2192 parent_ino, parent_gen, dest);
31db9f7c
AB		 2193 if (ret < 0)
2194 goto out;
2195
766702ef
AB		 2196 /*
2197 * Check if the ref was overwritten by an inode's ref that was processed
2198 * earlier. If yes, treat as orphan and return 1.
2199 */
31db9f7c
AB		 2200 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2201 dest->start, dest->end - dest->start);
2202 if (ret < 0)
2203 goto out;
2204 if (ret) {
2205 fs_path_reset(dest);
2206 ret = gen_unique_name(sctx, ino, gen, dest);
2207 if (ret < 0)
2208 goto out;
2209 ret = 1;
2210 }
2211
2212out_cache:
766702ef
AB		 2213 /*
2214 * Store the result of the lookup in the name cache.
2215 */
e780b0d1 2216 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
31db9f7c
AB		 2217 if (!nce) {
2218 ret = -ENOMEM;
2219 goto out;
2220 }
2221
2222 nce->ino = ino;
2223 nce->gen = gen;
2224 nce->parent_ino = *parent_ino;
2225 nce->parent_gen = *parent_gen;
2226 nce->name_len = fs_path_len(dest);
2227 nce->ret = ret;
2228 strcpy(nce->name, dest->start);
31db9f7c
AB		 2229
2230 if (ino < sctx->send_progress)
2231 nce->need_later_update = 0;
2232 else
2233 nce->need_later_update = 1;
2234
2235 nce_ret = name_cache_insert(sctx, nce);
2236 if (nce_ret < 0)
2237 ret = nce_ret;
2238 name_cache_clean_unused(sctx);
2239
2240out:
31db9f7c
AB		 2241 return ret;
2242}
2243
2244/*
2245 * Magic happens here. This function returns the first ref to an inode as it
2246 * would look like while receiving the stream at this point in time.
2247 * We walk the path up to the root. For every inode in between, we check if it
2248 * was already processed/sent. If yes, we continue with the parent as found
2249 * in send_root. If not, we continue with the parent as found in parent_root.
2250 * If we encounter an inode that was deleted at this point in time, we use the
2251 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2252 * that were not created yet and overwritten inodes/refs.
2253 *
2254 * When do we have have orphan inodes:
2255 * 1. When an inode is freshly created and thus no valid refs are available yet
2256 * 2. When a directory lost all it's refs (deleted) but still has dir items
2257 * inside which were not processed yet (pending for move/delete). If anyone
2258 * tried to get the path to the dir items, it would get a path inside that
2259 * orphan directory.
2260 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2261 * of an unprocessed inode. If in that case the first ref would be
2262 * overwritten, the overwritten inode gets "orphanized". Later when we
2263 * process this overwritten inode, it is restored at a new place by moving
2264 * the orphan inode.
2265 *
2266 * sctx->send_progress tells this function at which point in time receiving
2267 * would be.
2268 */
2269static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2270 struct fs_path *dest)
2271{
2272 int ret = 0;
2273 struct fs_path *name = NULL;
2274 u64 parent_inode = 0;
2275 u64 parent_gen = 0;
2276 int stop = 0;
2277
924794c9 2278 name = fs_path_alloc();
31db9f7c
AB		 2279 if (!name) {
2280 ret = -ENOMEM;
2281 goto out;
2282 }
2283
2284 dest->reversed = 1;
2285 fs_path_reset(dest);
2286
2287 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
8b191a68
FM		 2288 struct waiting_dir_move *wdm;
2289
31db9f7c
AB		 2290 fs_path_reset(name);
2291
9dc44214
FM		 2292 if (is_waiting_for_rm(sctx, ino)) {
2293 ret = gen_unique_name(sctx, ino, gen, name);
2294 if (ret < 0)
2295 goto out;
2296 ret = fs_path_add_path(dest, name);
2297 break;
2298 }
2299
8b191a68
FM		 2300 wdm = get_waiting_dir_move(sctx, ino);
2301 if (wdm && wdm->orphanized) {
2302 ret = gen_unique_name(sctx, ino, gen, name);
2303 stop = 1;
2304 } else if (wdm) {
bf0d1f44
FM		 2305 ret = get_first_ref(sctx->parent_root, ino,
2306 &parent_inode, &parent_gen, name);
2307 } else {
2308 ret = __get_cur_name_and_parent(sctx, ino, gen,
2309 &parent_inode,
2310 &parent_gen, name);
2311 if (ret)
2312 stop = 1;
2313 }
2314
31db9f7c
AB		 2315 if (ret < 0)
2316 goto out;
9f03740a 2317
31db9f7c
AB		 2318 ret = fs_path_add_path(dest, name);
2319 if (ret < 0)
2320 goto out;
2321
2322 ino = parent_inode;
2323 gen = parent_gen;
2324 }
2325
2326out:
924794c9 2327 fs_path_free(name);
31db9f7c
AB		 2328 if (!ret)
2329 fs_path_unreverse(dest);
2330 return ret;
2331}
2332
31db9f7c
AB		 2333/*
2334 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2335 */
2336static int send_subvol_begin(struct send_ctx *sctx)
2337{
2338 int ret;
2339 struct btrfs_root *send_root = sctx->send_root;
2340 struct btrfs_root *parent_root = sctx->parent_root;
2341 struct btrfs_path *path;
2342 struct btrfs_key key;
2343 struct btrfs_root_ref *ref;
2344 struct extent_buffer *leaf;
2345 char *name = NULL;
2346 int namelen;
2347
ffcfaf81 2348 path = btrfs_alloc_path();
31db9f7c
AB		 2349 if (!path)
2350 return -ENOMEM;
2351
e780b0d1 2352 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
31db9f7c
AB		 2353 if (!name) {
2354 btrfs_free_path(path);
2355 return -ENOMEM;
2356 }
2357
2358 key.objectid = send_root->objectid;
2359 key.type = BTRFS_ROOT_BACKREF_KEY;
2360 key.offset = 0;
2361
2362 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2363 &key, path, 1, 0);
2364 if (ret < 0)
2365 goto out;
2366 if (ret) {
2367 ret = -ENOENT;
2368 goto out;
2369 }
2370
2371 leaf = path->nodes[0];
2372 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2373 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2374 key.objectid != send_root->objectid) {
2375 ret = -ENOENT;
2376 goto out;
2377 }
2378 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2379 namelen = btrfs_root_ref_name_len(leaf, ref);
2380 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2381 btrfs_release_path(path);
2382
31db9f7c
AB		 2383 if (parent_root) {
2384 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2385 if (ret < 0)
2386 goto out;
2387 } else {
2388 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2389 if (ret < 0)
2390 goto out;
2391 }
2392
2393 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
b96b1db0
RR		 2394
2395 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2396 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2397 sctx->send_root->root_item.received_uuid);
2398 else
2399 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2400 sctx->send_root->root_item.uuid);
2401
31db9f7c 2402 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
5a0f4e2c 2403 le64_to_cpu(sctx->send_root->root_item.ctransid));
31db9f7c 2404 if (parent_root) {
37b8d27d
JB		 2405 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2406 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2407 parent_root->root_item.received_uuid);
2408 else
2409 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2410 parent_root->root_item.uuid);
31db9f7c 2411 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
5a0f4e2c 2412 le64_to_cpu(sctx->parent_root->root_item.ctransid));
31db9f7c
AB		 2413 }
2414
2415 ret = send_cmd(sctx);
2416
2417tlv_put_failure:
2418out:
2419 btrfs_free_path(path);
2420 kfree(name);
2421 return ret;
2422}
2423
2424static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2425{
04ab956e 2426 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2427 int ret = 0;
2428 struct fs_path *p;
2429
04ab956e 2430 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
31db9f7c 2431
924794c9 2432 p = fs_path_alloc();
31db9f7c
AB		 2433 if (!p)
2434 return -ENOMEM;
2435
2436 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2437 if (ret < 0)
2438 goto out;
2439
2440 ret = get_cur_path(sctx, ino, gen, p);
2441 if (ret < 0)
2442 goto out;
2443 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2444 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2445
2446 ret = send_cmd(sctx);
2447
2448tlv_put_failure:
2449out:
924794c9 2450 fs_path_free(p);
31db9f7c
AB		 2451 return ret;
2452}
2453
2454static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2455{
04ab956e 2456 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2457 int ret = 0;
2458 struct fs_path *p;
2459
04ab956e 2460 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
31db9f7c 2461
924794c9 2462 p = fs_path_alloc();
31db9f7c
AB		 2463 if (!p)
2464 return -ENOMEM;
2465
2466 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2467 if (ret < 0)
2468 goto out;
2469
2470 ret = get_cur_path(sctx, ino, gen, p);
2471 if (ret < 0)
2472 goto out;
2473 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2474 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2475
2476 ret = send_cmd(sctx);
2477
2478tlv_put_failure:
2479out:
924794c9 2480 fs_path_free(p);
31db9f7c
AB		 2481 return ret;
2482}
2483
2484static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2485{
04ab956e 2486 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2487 int ret = 0;
2488 struct fs_path *p;
2489
04ab956e
JM		 2490 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2491 ino, uid, gid);
31db9f7c 2492
924794c9 2493 p = fs_path_alloc();
31db9f7c
AB		 2494 if (!p)
2495 return -ENOMEM;
2496
2497 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2498 if (ret < 0)
2499 goto out;
2500
2501 ret = get_cur_path(sctx, ino, gen, p);
2502 if (ret < 0)
2503 goto out;
2504 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2505 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2506 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2507
2508 ret = send_cmd(sctx);
2509
2510tlv_put_failure:
2511out:
924794c9 2512 fs_path_free(p);
31db9f7c
AB		 2513 return ret;
2514}
2515
2516static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2517{
04ab956e 2518 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 2519 int ret = 0;
2520 struct fs_path *p = NULL;
2521 struct btrfs_inode_item *ii;
2522 struct btrfs_path *path = NULL;
2523 struct extent_buffer *eb;
2524 struct btrfs_key key;
2525 int slot;
2526
04ab956e 2527 btrfs_debug(fs_info, "send_utimes %llu", ino);
31db9f7c 2528
924794c9 2529 p = fs_path_alloc();
31db9f7c
AB		 2530 if (!p)
2531 return -ENOMEM;
2532
2533 path = alloc_path_for_send();
2534 if (!path) {
2535 ret = -ENOMEM;
2536 goto out;
2537 }
2538
2539 key.objectid = ino;
2540 key.type = BTRFS_INODE_ITEM_KEY;
2541 key.offset = 0;
2542 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
15b253ea
FM		 2543 if (ret > 0)
2544 ret = -ENOENT;
31db9f7c
AB		 2545 if (ret < 0)
2546 goto out;
2547
2548 eb = path->nodes[0];
2549 slot = path->slots[0];
2550 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2551
2552 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2553 if (ret < 0)
2554 goto out;
2555
2556 ret = get_cur_path(sctx, ino, gen, p);
2557 if (ret < 0)
2558 goto out;
2559 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
a937b979
DS		 2560 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2561 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2562 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
766702ef 2563 /* TODO Add otime support when the otime patches get into upstream */
31db9f7c
AB		 2564
2565 ret = send_cmd(sctx);
2566
2567tlv_put_failure:
2568out:
924794c9 2569 fs_path_free(p);
31db9f7c
AB		 2570 btrfs_free_path(path);
2571 return ret;
2572}
2573
2574/*
2575 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2576 * a valid path yet because we did not process the refs yet. So, the inode
2577 * is created as orphan.
2578 */
1f4692da 2579static int send_create_inode(struct send_ctx *sctx, u64 ino)
31db9f7c 2580{
04ab956e 2581 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c 2582 int ret = 0;
31db9f7c 2583 struct fs_path *p;
31db9f7c 2584 int cmd;
1f4692da 2585 u64 gen;
31db9f7c 2586 u64 mode;
1f4692da 2587 u64 rdev;
31db9f7c 2588
04ab956e 2589 btrfs_debug(fs_info, "send_create_inode %llu", ino);
31db9f7c 2590
924794c9 2591 p = fs_path_alloc();
31db9f7c
AB		 2592 if (!p)
2593 return -ENOMEM;
2594
644d1940
LB		 2595 if (ino != sctx->cur_ino) {
2596 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2597 NULL, NULL, &rdev);
2598 if (ret < 0)
2599 goto out;
2600 } else {
2601 gen = sctx->cur_inode_gen;
2602 mode = sctx->cur_inode_mode;
2603 rdev = sctx->cur_inode_rdev;
2604 }
31db9f7c 2605
e938c8ad 2606 if (S_ISREG(mode)) {
31db9f7c 2607 cmd = BTRFS_SEND_C_MKFILE;
e938c8ad 2608 } else if (S_ISDIR(mode)) {
31db9f7c 2609 cmd = BTRFS_SEND_C_MKDIR;
e938c8ad 2610 } else if (S_ISLNK(mode)) {
31db9f7c 2611 cmd = BTRFS_SEND_C_SYMLINK;
e938c8ad 2612 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
31db9f7c 2613 cmd = BTRFS_SEND_C_MKNOD;
e938c8ad 2614 } else if (S_ISFIFO(mode)) {
31db9f7c 2615 cmd = BTRFS_SEND_C_MKFIFO;
e938c8ad 2616 } else if (S_ISSOCK(mode)) {
31db9f7c 2617 cmd = BTRFS_SEND_C_MKSOCK;
e938c8ad 2618 } else {
f14d104d 2619 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
31db9f7c
AB		 2620 (int)(mode & S_IFMT));
2621 ret = -ENOTSUPP;
2622 goto out;
2623 }
2624
2625 ret = begin_cmd(sctx, cmd);
2626 if (ret < 0)
2627 goto out;
2628
1f4692da 2629 ret = gen_unique_name(sctx, ino, gen, p);
31db9f7c
AB		 2630 if (ret < 0)
2631 goto out;
2632
2633 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
1f4692da 2634 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
31db9f7c
AB		 2635
2636 if (S_ISLNK(mode)) {
2637 fs_path_reset(p);
924794c9 2638 ret = read_symlink(sctx->send_root, ino, p);
31db9f7c
AB		 2639 if (ret < 0)
2640 goto out;
2641 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2642 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2643 S_ISFIFO(mode) || S_ISSOCK(mode)) {
d79e5043
AJ		 2644 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2645 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
31db9f7c
AB		 2646 }
2647
2648 ret = send_cmd(sctx);
2649 if (ret < 0)
2650 goto out;
2651
2652
2653tlv_put_failure:
2654out:
924794c9 2655 fs_path_free(p);
31db9f7c
AB		 2656 return ret;
2657}
2658
1f4692da
AB		 2659/*
2660 * We need some special handling for inodes that get processed before the parent
2661 * directory got created. See process_recorded_refs for details.
2662 * This function does the check if we already created the dir out of order.
2663 */
2664static int did_create_dir(struct send_ctx *sctx, u64 dir)
2665{
2666 int ret = 0;
2667 struct btrfs_path *path = NULL;
2668 struct btrfs_key key;
2669 struct btrfs_key found_key;
2670 struct btrfs_key di_key;
2671 struct extent_buffer *eb;
2672 struct btrfs_dir_item *di;
2673 int slot;
2674
2675 path = alloc_path_for_send();
2676 if (!path) {
2677 ret = -ENOMEM;
2678 goto out;
2679 }
2680
2681 key.objectid = dir;
2682 key.type = BTRFS_DIR_INDEX_KEY;
2683 key.offset = 0;
dff6d0ad
FDBM		 2684 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2685 if (ret < 0)
2686 goto out;
2687
1f4692da 2688 while (1) {
dff6d0ad
FDBM		 2689 eb = path->nodes[0];
2690 slot = path->slots[0];
2691 if (slot >= btrfs_header_nritems(eb)) {
2692 ret = btrfs_next_leaf(sctx->send_root, path);
2693 if (ret < 0) {
2694 goto out;
2695 } else if (ret > 0) {
2696 ret = 0;
2697 break;
2698 }
2699 continue;
1f4692da 2700 }
dff6d0ad
FDBM		 2701
2702 btrfs_item_key_to_cpu(eb, &found_key, slot);
2703 if (found_key.objectid != key.objectid ||
1f4692da
AB		 2704 found_key.type != key.type) {
2705 ret = 0;
2706 goto out;
2707 }
2708
2709 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2710 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2711
a0525414
JB		 2712 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2713 di_key.objectid < sctx->send_progress) {
1f4692da
AB		 2714 ret = 1;
2715 goto out;
2716 }
2717
dff6d0ad 2718 path->slots[0]++;
1f4692da
AB		 2719 }
2720
2721out:
2722 btrfs_free_path(path);
2723 return ret;
2724}
2725
2726/*
2727 * Only creates the inode if it is:
2728 * 1. Not a directory
2729 * 2. Or a directory which was not created already due to out of order
2730 * directories. See did_create_dir and process_recorded_refs for details.
2731 */
2732static int send_create_inode_if_needed(struct send_ctx *sctx)
2733{
2734 int ret;
2735
2736 if (S_ISDIR(sctx->cur_inode_mode)) {
2737 ret = did_create_dir(sctx, sctx->cur_ino);
2738 if (ret < 0)
2739 goto out;
2740 if (ret) {
2741 ret = 0;
2742 goto out;
2743 }
2744 }
2745
2746 ret = send_create_inode(sctx, sctx->cur_ino);
2747 if (ret < 0)
2748 goto out;
2749
2750out:
2751 return ret;
2752}
2753
31db9f7c
AB		 2754struct recorded_ref {
2755 struct list_head list;
2756 char *dir_path;
2757 char *name;
2758 struct fs_path *full_path;
2759 u64 dir;
2760 u64 dir_gen;
2761 int dir_path_len;
2762 int name_len;
2763};
2764
2765/*
2766 * We need to process new refs before deleted refs, but compare_tree gives us
2767 * everything mixed. So we first record all refs and later process them.
2768 * This function is a helper to record one ref.
2769 */
a4d96d62 2770static int __record_ref(struct list_head *head, u64 dir,
31db9f7c
AB		 2771 u64 dir_gen, struct fs_path *path)
2772{
2773 struct recorded_ref *ref;
31db9f7c 2774
e780b0d1 2775 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
31db9f7c
AB		 2776 if (!ref)
2777 return -ENOMEM;
2778
2779 ref->dir = dir;
2780 ref->dir_gen = dir_gen;
2781 ref->full_path = path;
2782
ed84885d
AS		 2783 ref->name = (char *)kbasename(ref->full_path->start);
2784 ref->name_len = ref->full_path->end - ref->name;
2785 ref->dir_path = ref->full_path->start;
2786 if (ref->name == ref->full_path->start)
31db9f7c 2787 ref->dir_path_len = 0;
ed84885d 2788 else
31db9f7c
AB		 2789 ref->dir_path_len = ref->full_path->end -
2790 ref->full_path->start - 1 - ref->name_len;
31db9f7c
AB		 2791
2792 list_add_tail(&ref->list, head);
2793 return 0;
2794}
2795
ba5e8f2e
JB		 2796static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2797{
2798 struct recorded_ref *new;
2799
e780b0d1 2800 new = kmalloc(sizeof(*ref), GFP_KERNEL);
ba5e8f2e
JB		 2801 if (!new)
2802 return -ENOMEM;
2803
2804 new->dir = ref->dir;
2805 new->dir_gen = ref->dir_gen;
2806 new->full_path = NULL;
2807 INIT_LIST_HEAD(&new->list);
2808 list_add_tail(&new->list, list);
2809 return 0;
2810}
2811
924794c9 2812static void __free_recorded_refs(struct list_head *head)
31db9f7c
AB		 2813{
2814 struct recorded_ref *cur;
31db9f7c 2815
e938c8ad
AB		 2816 while (!list_empty(head)) {
2817 cur = list_entry(head->next, struct recorded_ref, list);
924794c9 2818 fs_path_free(cur->full_path);
e938c8ad 2819 list_del(&cur->list);
31db9f7c
AB		 2820 kfree(cur);
2821 }
31db9f7c
AB		 2822}
2823
2824static void free_recorded_refs(struct send_ctx *sctx)
2825{
924794c9
TI		 2826 __free_recorded_refs(&sctx->new_refs);
2827 __free_recorded_refs(&sctx->deleted_refs);
31db9f7c
AB		 2828}
2829
2830/*
766702ef 2831 * Renames/moves a file/dir to its orphan name. Used when the first
31db9f7c
AB		 2832 * ref of an unprocessed inode gets overwritten and for all non empty
2833 * directories.
2834 */
2835static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2836 struct fs_path *path)
2837{
2838 int ret;
2839 struct fs_path *orphan;
2840
924794c9 2841 orphan = fs_path_alloc();
31db9f7c
AB		 2842 if (!orphan)
2843 return -ENOMEM;
2844
2845 ret = gen_unique_name(sctx, ino, gen, orphan);
2846 if (ret < 0)
2847 goto out;
2848
2849 ret = send_rename(sctx, path, orphan);
2850
2851out:
924794c9 2852 fs_path_free(orphan);
31db9f7c
AB		 2853 return ret;
2854}
2855
9dc44214
FM		 2856static struct orphan_dir_info *
2857add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2858{
2859 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2860 struct rb_node *parent = NULL;
2861 struct orphan_dir_info *entry, *odi;
2862
e780b0d1 2863 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
9dc44214
FM		 2864 if (!odi)
2865 return ERR_PTR(-ENOMEM);
2866 odi->ino = dir_ino;
2867 odi->gen = 0;
2868
2869 while (*p) {
2870 parent = *p;
2871 entry = rb_entry(parent, struct orphan_dir_info, node);
2872 if (dir_ino < entry->ino) {
2873 p = &(*p)->rb_left;
2874 } else if (dir_ino > entry->ino) {
2875 p = &(*p)->rb_right;
2876 } else {
2877 kfree(odi);
2878 return entry;
2879 }
2880 }
2881
2882 rb_link_node(&odi->node, parent, p);
2883 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2884 return odi;
2885}
2886
2887static struct orphan_dir_info *
2888get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2889{
2890 struct rb_node *n = sctx->orphan_dirs.rb_node;
2891 struct orphan_dir_info *entry;
2892
2893 while (n) {
2894 entry = rb_entry(n, struct orphan_dir_info, node);
2895 if (dir_ino < entry->ino)
2896 n = n->rb_left;
2897 else if (dir_ino > entry->ino)
2898 n = n->rb_right;
2899 else
2900 return entry;
2901 }
2902 return NULL;
2903}
2904
2905static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2906{
2907 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2908
2909 return odi != NULL;
2910}
2911
2912static void free_orphan_dir_info(struct send_ctx *sctx,
2913 struct orphan_dir_info *odi)
2914{
2915 if (!odi)
2916 return;
2917 rb_erase(&odi->node, &sctx->orphan_dirs);
2918 kfree(odi);
2919}
2920
31db9f7c
AB		 2921/*
2922 * Returns 1 if a directory can be removed at this point in time.
2923 * We check this by iterating all dir items and checking if the inode behind
2924 * the dir item was already processed.
2925 */
9dc44214
FM		 2926static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2927 u64 send_progress)
31db9f7c
AB		 2928{
2929 int ret = 0;
2930 struct btrfs_root *root = sctx->parent_root;
2931 struct btrfs_path *path;
2932 struct btrfs_key key;
2933 struct btrfs_key found_key;
2934 struct btrfs_key loc;
2935 struct btrfs_dir_item *di;
2936
6d85ed05
AB		 2937 /*
2938 * Don't try to rmdir the top/root subvolume dir.
2939 */
2940 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2941 return 0;
2942
31db9f7c
AB		 2943 path = alloc_path_for_send();
2944 if (!path)
2945 return -ENOMEM;
2946
2947 key.objectid = dir;
2948 key.type = BTRFS_DIR_INDEX_KEY;
2949 key.offset = 0;
dff6d0ad
FDBM		 2950 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2951 if (ret < 0)
2952 goto out;
31db9f7c
AB		 2953
2954 while (1) {
9dc44214
FM		 2955 struct waiting_dir_move *dm;
2956
dff6d0ad
FDBM		 2957 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2958 ret = btrfs_next_leaf(root, path);
2959 if (ret < 0)
2960 goto out;
2961 else if (ret > 0)
2962 break;
2963 continue;
31db9f7c 2964 }
dff6d0ad
FDBM		 2965 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2966 path->slots[0]);
2967 if (found_key.objectid != key.objectid ||
2968 found_key.type != key.type)
31db9f7c 2969 break;
31db9f7c
AB		 2970
2971 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2972 struct btrfs_dir_item);
2973 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2974
9dc44214
FM		 2975 dm = get_waiting_dir_move(sctx, loc.objectid);
2976 if (dm) {
2977 struct orphan_dir_info *odi;
2978
2979 odi = add_orphan_dir_info(sctx, dir);
2980 if (IS_ERR(odi)) {
2981 ret = PTR_ERR(odi);
2982 goto out;
2983 }
2984 odi->gen = dir_gen;
2985 dm->rmdir_ino = dir;
2986 ret = 0;
2987 goto out;
2988 }
2989
31db9f7c 2990 if (loc.objectid > send_progress) {
443f9d26
RK		 2991 struct orphan_dir_info *odi;
2992
2993 odi = get_orphan_dir_info(sctx, dir);
2994 free_orphan_dir_info(sctx, odi);
31db9f7c
AB		 2995 ret = 0;
2996 goto out;
2997 }
2998
dff6d0ad 2999 path->slots[0]++;
31db9f7c
AB		 3000 }
3001
3002 ret = 1;
3003
3004out:
3005 btrfs_free_path(path);
3006 return ret;
3007}
3008
9f03740a
FDBM		 3009static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3010{
9dc44214 3011 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
9f03740a 3012
9dc44214 3013 return entry != NULL;
9f03740a
FDBM		 3014}
3015
8b191a68 3016static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
9f03740a
FDBM		 3017{
3018 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3019 struct rb_node *parent = NULL;
3020 struct waiting_dir_move *entry, *dm;
3021
e780b0d1 3022 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
9f03740a
FDBM		 3023 if (!dm)
3024 return -ENOMEM;
3025 dm->ino = ino;
9dc44214 3026 dm->rmdir_ino = 0;
8b191a68 3027 dm->orphanized = orphanized;
9f03740a
FDBM		 3028
3029 while (*p) {
3030 parent = *p;
3031 entry = rb_entry(parent, struct waiting_dir_move, node);
3032 if (ino < entry->ino) {
3033 p = &(*p)->rb_left;
3034 } else if (ino > entry->ino) {
3035 p = &(*p)->rb_right;
3036 } else {
3037 kfree(dm);
3038 return -EEXIST;
3039 }
3040 }
3041
3042 rb_link_node(&dm->node, parent, p);
3043 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3044 return 0;
3045}
3046
9dc44214
FM		 3047static struct waiting_dir_move *
3048get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
9f03740a
FDBM		 3049{
3050 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3051 struct waiting_dir_move *entry;
3052
3053 while (n) {
3054 entry = rb_entry(n, struct waiting_dir_move, node);
9dc44214 3055 if (ino < entry->ino)
9f03740a 3056 n = n->rb_left;
9dc44214 3057 else if (ino > entry->ino)
9f03740a 3058 n = n->rb_right;
9dc44214
FM		 3059 else
3060 return entry;
9f03740a 3061 }
9dc44214
FM		 3062 return NULL;
3063}
3064
3065static void free_waiting_dir_move(struct send_ctx *sctx,
3066 struct waiting_dir_move *dm)
3067{
3068 if (!dm)
3069 return;
3070 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3071 kfree(dm);
9f03740a
FDBM		 3072}
3073
bfa7e1f8
FM		 3074static int add_pending_dir_move(struct send_ctx *sctx,
3075 u64 ino,
3076 u64 ino_gen,
f959492f
FM		 3077 u64 parent_ino,
3078 struct list_head *new_refs,
84471e24
FM		 3079 struct list_head *deleted_refs,
3080 const bool is_orphan)
9f03740a
FDBM		 3081{
3082 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3083 struct rb_node *parent = NULL;
73b802f4 3084 struct pending_dir_move *entry = NULL, *pm;
9f03740a
FDBM		 3085 struct recorded_ref *cur;
3086 int exists = 0;
3087 int ret;
3088
e780b0d1 3089 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
9f03740a
FDBM		 3090 if (!pm)
3091 return -ENOMEM;
3092 pm->parent_ino = parent_ino;
bfa7e1f8
FM		 3093 pm->ino = ino;
3094 pm->gen = ino_gen;
9f03740a
FDBM		 3095 INIT_LIST_HEAD(&pm->list);
3096 INIT_LIST_HEAD(&pm->update_refs);
3097 RB_CLEAR_NODE(&pm->node);
3098
3099 while (*p) {
3100 parent = *p;
3101 entry = rb_entry(parent, struct pending_dir_move, node);
3102 if (parent_ino < entry->parent_ino) {
3103 p = &(*p)->rb_left;
3104 } else if (parent_ino > entry->parent_ino) {
3105 p = &(*p)->rb_right;
3106 } else {
3107 exists = 1;
3108 break;
3109 }
3110 }
3111
f959492f 3112 list_for_each_entry(cur, deleted_refs, list) {
9f03740a
FDBM		 3113 ret = dup_ref(cur, &pm->update_refs);
3114 if (ret < 0)
3115 goto out;
3116 }
f959492f 3117 list_for_each_entry(cur, new_refs, list) {
9f03740a
FDBM		 3118 ret = dup_ref(cur, &pm->update_refs);
3119 if (ret < 0)
3120 goto out;
3121 }
3122
8b191a68 3123 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
9f03740a
FDBM		 3124 if (ret)
3125 goto out;
3126
3127 if (exists) {
3128 list_add_tail(&pm->list, &entry->list);
3129 } else {
3130 rb_link_node(&pm->node, parent, p);
3131 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3132 }
3133 ret = 0;
3134out:
3135 if (ret) {
3136 __free_recorded_refs(&pm->update_refs);
3137 kfree(pm);
3138 }
3139 return ret;
3140}
3141
3142static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3143 u64 parent_ino)
3144{
3145 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3146 struct pending_dir_move *entry;
3147
3148 while (n) {
3149 entry = rb_entry(n, struct pending_dir_move, node);
3150 if (parent_ino < entry->parent_ino)
3151 n = n->rb_left;
3152 else if (parent_ino > entry->parent_ino)
3153 n = n->rb_right;
3154 else
3155 return entry;
3156 }
3157 return NULL;
3158}
3159
801bec36
RK		 3160static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3161 u64 ino, u64 gen, u64 *ancestor_ino)
3162{
3163 int ret = 0;
3164 u64 parent_inode = 0;
3165 u64 parent_gen = 0;
3166 u64 start_ino = ino;
3167
3168 *ancestor_ino = 0;
3169 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3170 fs_path_reset(name);
3171
3172 if (is_waiting_for_rm(sctx, ino))
3173 break;
3174 if (is_waiting_for_move(sctx, ino)) {
3175 if (*ancestor_ino == 0)
3176 *ancestor_ino = ino;
3177 ret = get_first_ref(sctx->parent_root, ino,
3178 &parent_inode, &parent_gen, name);
3179 } else {
3180 ret = __get_cur_name_and_parent(sctx, ino, gen,
3181 &parent_inode,
3182 &parent_gen, name);
3183 if (ret > 0) {
3184 ret = 0;
3185 break;
3186 }
3187 }
3188 if (ret < 0)
3189 break;
3190 if (parent_inode == start_ino) {
3191 ret = 1;
3192 if (*ancestor_ino == 0)
3193 *ancestor_ino = ino;
3194 break;
3195 }
3196 ino = parent_inode;
3197 gen = parent_gen;
3198 }
3199 return ret;
3200}
3201
9f03740a
FDBM		 3202static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3203{
3204 struct fs_path *from_path = NULL;
3205 struct fs_path *to_path = NULL;
2b863a13 3206 struct fs_path *name = NULL;
9f03740a
FDBM		 3207 u64 orig_progress = sctx->send_progress;
3208 struct recorded_ref *cur;
2b863a13 3209 u64 parent_ino, parent_gen;
9dc44214
FM		 3210 struct waiting_dir_move *dm = NULL;
3211 u64 rmdir_ino = 0;
801bec36
RK		 3212 u64 ancestor;
3213 bool is_orphan;
9f03740a
FDBM		 3214 int ret;
3215
2b863a13 3216 name = fs_path_alloc();
9f03740a 3217 from_path = fs_path_alloc();
2b863a13
FM		 3218 if (!name || !from_path) {
3219 ret = -ENOMEM;
3220 goto out;
3221 }
9f03740a 3222
9dc44214
FM		 3223 dm = get_waiting_dir_move(sctx, pm->ino);
3224 ASSERT(dm);
3225 rmdir_ino = dm->rmdir_ino;
801bec36 3226 is_orphan = dm->orphanized;
9dc44214 3227 free_waiting_dir_move(sctx, dm);
2b863a13 3228
801bec36 3229 if (is_orphan) {
84471e24
FM		 3230 ret = gen_unique_name(sctx, pm->ino,
3231 pm->gen, from_path);
3232 } else {
3233 ret = get_first_ref(sctx->parent_root, pm->ino,
3234 &parent_ino, &parent_gen, name);
3235 if (ret < 0)
3236 goto out;
3237 ret = get_cur_path(sctx, parent_ino, parent_gen,
3238 from_path);
3239 if (ret < 0)
3240 goto out;
3241 ret = fs_path_add_path(from_path, name);
3242 }
c992ec94
FM		 3243 if (ret < 0)
3244 goto out;
2b863a13 3245
f959492f 3246 sctx->send_progress = sctx->cur_ino + 1;
801bec36 3247 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
7969e77a
FM		 3248 if (ret < 0)
3249 goto out;
801bec36
RK		 3250 if (ret) {
3251 LIST_HEAD(deleted_refs);
3252 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3253 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3254 &pm->update_refs, &deleted_refs,
3255 is_orphan);
3256 if (ret < 0)
3257 goto out;
3258 if (rmdir_ino) {
3259 dm = get_waiting_dir_move(sctx, pm->ino);
3260 ASSERT(dm);
3261 dm->rmdir_ino = rmdir_ino;
3262 }
3263 goto out;
3264 }
c992ec94
FM		 3265 fs_path_reset(name);
3266 to_path = name;
2b863a13 3267 name = NULL;
9f03740a
FDBM		 3268 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3269 if (ret < 0)
3270 goto out;
3271
3272 ret = send_rename(sctx, from_path, to_path);
3273 if (ret < 0)
3274 goto out;
3275
9dc44214
FM		 3276 if (rmdir_ino) {
3277 struct orphan_dir_info *odi;
3278
3279 odi = get_orphan_dir_info(sctx, rmdir_ino);
3280 if (!odi) {
3281 /* already deleted */
3282 goto finish;
3283 }
99ea42dd 3284 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino);
9dc44214
FM		 3285 if (ret < 0)
3286 goto out;
3287 if (!ret)
3288 goto finish;
3289
3290 name = fs_path_alloc();
3291 if (!name) {
3292 ret = -ENOMEM;
3293 goto out;
3294 }
3295 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3296 if (ret < 0)
3297 goto out;
3298 ret = send_rmdir(sctx, name);
3299 if (ret < 0)
3300 goto out;
3301 free_orphan_dir_info(sctx, odi);
3302 }
3303
3304finish:
9f03740a
FDBM		 3305 ret = send_utimes(sctx, pm->ino, pm->gen);
3306 if (ret < 0)
3307 goto out;
3308
3309 /*
3310 * After rename/move, need to update the utimes of both new parent(s)
3311 * and old parent(s).
3312 */
3313 list_for_each_entry(cur, &pm->update_refs, list) {
764433a1
RK		 3314 /*
3315 * The parent inode might have been deleted in the send snapshot
3316 */
3317 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3318 NULL, NULL, NULL, NULL, NULL);
3319 if (ret == -ENOENT) {
3320 ret = 0;
9dc44214 3321 continue;
764433a1
RK		 3322 }
3323 if (ret < 0)
3324 goto out;
3325
9f03740a
FDBM		 3326 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3327 if (ret < 0)
3328 goto out;
3329 }
3330
3331out:
2b863a13 3332 fs_path_free(name);
9f03740a
FDBM		 3333 fs_path_free(from_path);
3334 fs_path_free(to_path);
3335 sctx->send_progress = orig_progress;
3336
3337 return ret;
3338}
3339
3340static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3341{
3342 if (!list_empty(&m->list))
3343 list_del(&m->list);
3344 if (!RB_EMPTY_NODE(&m->node))
3345 rb_erase(&m->node, &sctx->pending_dir_moves);
3346 __free_recorded_refs(&m->update_refs);
3347 kfree(m);
3348}
3349
3350static void tail_append_pending_moves(struct pending_dir_move *moves,
3351 struct list_head *stack)
3352{
3353 if (list_empty(&moves->list)) {
3354 list_add_tail(&moves->list, stack);
3355 } else {
3356 LIST_HEAD(list);
3357 list_splice_init(&moves->list, &list);
3358 list_add_tail(&moves->list, stack);
3359 list_splice_tail(&list, stack);
3360 }
3361}
3362
3363static int apply_children_dir_moves(struct send_ctx *sctx)
3364{
3365 struct pending_dir_move *pm;
3366 struct list_head stack;
3367 u64 parent_ino = sctx->cur_ino;
3368 int ret = 0;
3369
3370 pm = get_pending_dir_moves(sctx, parent_ino);
3371 if (!pm)
3372 return 0;
3373
3374 INIT_LIST_HEAD(&stack);
3375 tail_append_pending_moves(pm, &stack);
3376
3377 while (!list_empty(&stack)) {
3378 pm = list_first_entry(&stack, struct pending_dir_move, list);
3379 parent_ino = pm->ino;
3380 ret = apply_dir_move(sctx, pm);
3381 free_pending_move(sctx, pm);
3382 if (ret)
3383 goto out;
3384 pm = get_pending_dir_moves(sctx, parent_ino);
3385 if (pm)
3386 tail_append_pending_moves(pm, &stack);
3387 }
3388 return 0;
3389
3390out:
3391 while (!list_empty(&stack)) {
3392 pm = list_first_entry(&stack, struct pending_dir_move, list);
3393 free_pending_move(sctx, pm);
3394 }
3395 return ret;
3396}
3397
84471e24
FM		 3398/*
3399 * We might need to delay a directory rename even when no ancestor directory
3400 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3401 * renamed. This happens when we rename a directory to the old name (the name
3402 * in the parent root) of some other unrelated directory that got its rename
3403 * delayed due to some ancestor with higher number that got renamed.
3404 *
3405 * Example:
3406 *
3407 * Parent snapshot:
3408 * . (ino 256)
3409 * |---- a/ (ino 257)
3410 * | |---- file (ino 260)
3411 * |
3412 * |---- b/ (ino 258)
3413 * |---- c/ (ino 259)
3414 *
3415 * Send snapshot:
3416 * . (ino 256)
3417 * |---- a/ (ino 258)
3418 * |---- x/ (ino 259)
3419 * |---- y/ (ino 257)
3420 * |----- file (ino 260)
3421 *
3422 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3423 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3424 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3425 * must issue is:
3426 *
3427 * 1 - rename 259 from 'c' to 'x'
3428 * 2 - rename 257 from 'a' to 'x/y'
3429 * 3 - rename 258 from 'b' to 'a'
3430 *
3431 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3432 * be done right away and < 0 on error.
3433 */
3434static int wait_for_dest_dir_move(struct send_ctx *sctx,
3435 struct recorded_ref *parent_ref,
3436 const bool is_orphan)
3437{
3438 struct btrfs_path *path;
3439 struct btrfs_key key;
3440 struct btrfs_key di_key;
3441 struct btrfs_dir_item *di;
3442 u64 left_gen;
3443 u64 right_gen;
3444 int ret = 0;
801bec36 3445 struct waiting_dir_move *wdm;
84471e24
FM		 3446
3447 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3448 return 0;
3449
3450 path = alloc_path_for_send();
3451 if (!path)
3452 return -ENOMEM;
3453
3454 key.objectid = parent_ref->dir;
3455 key.type = BTRFS_DIR_ITEM_KEY;
3456 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3457
3458 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3459 if (ret < 0) {
3460 goto out;
3461 } else if (ret > 0) {
3462 ret = 0;
3463 goto out;
3464 }
3465
3466 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3467 parent_ref->name, parent_ref->name_len);
3468 if (!di) {
3469 ret = 0;
3470 goto out;
3471 }
3472 /*
3473 * di_key.objectid has the number of the inode that has a dentry in the
3474 * parent directory with the same name that sctx->cur_ino is being
3475 * renamed to. We need to check if that inode is in the send root as
3476 * well and if it is currently marked as an inode with a pending rename,
3477 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3478 * that it happens after that other inode is renamed.
3479 */
3480 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3481 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3482 ret = 0;
3483 goto out;
3484 }
3485
3486 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3487 &left_gen, NULL, NULL, NULL, NULL);
3488 if (ret < 0)
3489 goto out;
3490 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3491 &right_gen, NULL, NULL, NULL, NULL);
3492 if (ret < 0) {
3493 if (ret == -ENOENT)
3494 ret = 0;
3495 goto out;
3496 }
3497
3498 /* Different inode, no need to delay the rename of sctx->cur_ino */
3499 if (right_gen != left_gen) {
3500 ret = 0;
3501 goto out;
3502 }
3503
801bec36
RK		 3504 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3505 if (wdm && !wdm->orphanized) {
84471e24
FM		 3506 ret = add_pending_dir_move(sctx,
3507 sctx->cur_ino,
3508 sctx->cur_inode_gen,
3509 di_key.objectid,
3510 &sctx->new_refs,
3511 &sctx->deleted_refs,
3512 is_orphan);
3513 if (!ret)
3514 ret = 1;
3515 }
3516out:
3517 btrfs_free_path(path);
3518 return ret;
3519}
3520
80aa6027
FM		 3521/*
3522 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3523 * Return 1 if true, 0 if false and < 0 on error.
3524 */
3525static int is_ancestor(struct btrfs_root *root,
3526 const u64 ino1,
3527 const u64 ino1_gen,
3528 const u64 ino2,
3529 struct fs_path *fs_path)
3530{
3531 u64 ino = ino2;
3532
3533 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3534 int ret;
3535 u64 parent;
3536 u64 parent_gen;
3537
3538 fs_path_reset(fs_path);
3539 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3540 if (ret < 0) {
3541 if (ret == -ENOENT && ino == ino2)
3542 ret = 0;
3543 return ret;
3544 }
3545 if (parent == ino1)
3546 return parent_gen == ino1_gen ? 1 : 0;
3547 ino = parent;
3548 }
3549 return 0;
3550}
3551
9f03740a 3552static int wait_for_parent_move(struct send_ctx *sctx,
8b191a68
FM		 3553 struct recorded_ref *parent_ref,
3554 const bool is_orphan)
9f03740a 3555{
f959492f 3556 int ret = 0;
9f03740a
FDBM		 3557 u64 ino = parent_ref->dir;
3558 u64 parent_ino_before, parent_ino_after;
9f03740a
FDBM		 3559 struct fs_path *path_before = NULL;
3560 struct fs_path *path_after = NULL;
3561 int len1, len2;
9f03740a
FDBM		 3562
3563 path_after = fs_path_alloc();
f959492f
FM		 3564 path_before = fs_path_alloc();
3565 if (!path_after || !path_before) {
9f03740a
FDBM		 3566 ret = -ENOMEM;
3567 goto out;
3568 }
3569
bfa7e1f8 3570 /*
f959492f
FM		 3571 * Our current directory inode may not yet be renamed/moved because some
3572 * ancestor (immediate or not) has to be renamed/moved first. So find if
3573 * such ancestor exists and make sure our own rename/move happens after
80aa6027
FM		 3574 * that ancestor is processed to avoid path build infinite loops (done
3575 * at get_cur_path()).
bfa7e1f8 3576 */
f959492f
FM		 3577 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3578 if (is_waiting_for_move(sctx, ino)) {
80aa6027
FM		 3579 /*
3580 * If the current inode is an ancestor of ino in the
3581 * parent root, we need to delay the rename of the
3582 * current inode, otherwise don't delayed the rename
3583 * because we can end up with a circular dependency
3584 * of renames, resulting in some directories never
3585 * getting the respective rename operations issued in
3586 * the send stream or getting into infinite path build
3587 * loops.
3588 */
3589 ret = is_ancestor(sctx->parent_root,
3590 sctx->cur_ino, sctx->cur_inode_gen,
3591 ino, path_before);
4122ea64
FM		 3592 if (ret)
3593 break;
f959492f 3594 }
bfa7e1f8
FM		 3595
3596 fs_path_reset(path_before);
3597 fs_path_reset(path_after);
3598
3599 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
f959492f 3600 NULL, path_after);
bfa7e1f8
FM		 3601 if (ret < 0)
3602 goto out;
3603 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3604 NULL, path_before);
f959492f 3605 if (ret < 0 && ret != -ENOENT) {
bfa7e1f8 3606 goto out;
f959492f 3607 } else if (ret == -ENOENT) {
bf8e8ca6 3608 ret = 0;
f959492f 3609 break;
bfa7e1f8
FM		 3610 }
3611
3612 len1 = fs_path_len(path_before);
3613 len2 = fs_path_len(path_after);
f959492f
FM		 3614 if (ino > sctx->cur_ino &&
3615 (parent_ino_before != parent_ino_after || len1 != len2 ||
3616 memcmp(path_before->start, path_after->start, len1))) {
bfa7e1f8 3617 ret = 1;
f959492f 3618 break;
bfa7e1f8 3619 }
bfa7e1f8 3620 ino = parent_ino_after;
bfa7e1f8
FM		 3621 }
3622
9f03740a
FDBM		 3623out:
3624 fs_path_free(path_before);
3625 fs_path_free(path_after);
3626
f959492f
FM		 3627 if (ret == 1) {
3628 ret = add_pending_dir_move(sctx,
3629 sctx->cur_ino,
3630 sctx->cur_inode_gen,
3631 ino,
3632 &sctx->new_refs,
84471e24 3633 &sctx->deleted_refs,
8b191a68 3634 is_orphan);
f959492f
FM		 3635 if (!ret)
3636 ret = 1;
3637 }
3638
9f03740a
FDBM		 3639 return ret;
3640}
3641
31db9f7c
AB		 3642/*
3643 * This does all the move/link/unlink/rmdir magic.
3644 */
9f03740a 3645static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
31db9f7c 3646{
04ab956e 3647 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 3648 int ret = 0;
3649 struct recorded_ref *cur;
1f4692da 3650 struct recorded_ref *cur2;
ba5e8f2e 3651 struct list_head check_dirs;
31db9f7c 3652 struct fs_path *valid_path = NULL;
b24baf69 3653 u64 ow_inode = 0;
31db9f7c
AB		 3654 u64 ow_gen;
3655 int did_overwrite = 0;
3656 int is_orphan = 0;
29d6d30f 3657 u64 last_dir_ino_rm = 0;
84471e24 3658 bool can_rename = true;
31db9f7c 3659
04ab956e 3660 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
31db9f7c 3661
6d85ed05
AB		 3662 /*
3663 * This should never happen as the root dir always has the same ref
3664 * which is always '..'
3665 */
3666 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
ba5e8f2e 3667 INIT_LIST_HEAD(&check_dirs);
6d85ed05 3668
924794c9 3669 valid_path = fs_path_alloc();
31db9f7c
AB		 3670 if (!valid_path) {
3671 ret = -ENOMEM;
3672 goto out;
3673 }
3674
31db9f7c
AB		 3675 /*
3676 * First, check if the first ref of the current inode was overwritten
3677 * before. If yes, we know that the current inode was already orphanized
3678 * and thus use the orphan name. If not, we can use get_cur_path to
3679 * get the path of the first ref as it would like while receiving at
3680 * this point in time.
3681 * New inodes are always orphan at the beginning, so force to use the
3682 * orphan name in this case.
3683 * The first ref is stored in valid_path and will be updated if it
3684 * gets moved around.
3685 */
3686 if (!sctx->cur_inode_new) {
3687 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3688 sctx->cur_inode_gen);
3689 if (ret < 0)
3690 goto out;
3691 if (ret)
3692 did_overwrite = 1;
3693 }
3694 if (sctx->cur_inode_new || did_overwrite) {
3695 ret = gen_unique_name(sctx, sctx->cur_ino,
3696 sctx->cur_inode_gen, valid_path);
3697 if (ret < 0)
3698 goto out;
3699 is_orphan = 1;
3700 } else {
3701 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3702 valid_path);
3703 if (ret < 0)
3704 goto out;
3705 }
3706
3707 list_for_each_entry(cur, &sctx->new_refs, list) {
1f4692da
AB		 3708 /*
3709 * We may have refs where the parent directory does not exist
3710 * yet. This happens if the parent directories inum is higher
3711 * the the current inum. To handle this case, we create the
3712 * parent directory out of order. But we need to check if this
3713 * did already happen before due to other refs in the same dir.
3714 */
3715 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3716 if (ret < 0)
3717 goto out;
3718 if (ret == inode_state_will_create) {
3719 ret = 0;
3720 /*
3721 * First check if any of the current inodes refs did
3722 * already create the dir.
3723 */
3724 list_for_each_entry(cur2, &sctx->new_refs, list) {
3725 if (cur == cur2)
3726 break;
3727 if (cur2->dir == cur->dir) {
3728 ret = 1;
3729 break;
3730 }
3731 }
3732
3733 /*
3734 * If that did not happen, check if a previous inode
3735 * did already create the dir.
3736 */
3737 if (!ret)
3738 ret = did_create_dir(sctx, cur->dir);
3739 if (ret < 0)
3740 goto out;
3741 if (!ret) {
3742 ret = send_create_inode(sctx, cur->dir);
3743 if (ret < 0)
3744 goto out;
3745 }
3746 }
3747
31db9f7c
AB		 3748 /*
3749 * Check if this new ref would overwrite the first ref of
3750 * another unprocessed inode. If yes, orphanize the
3751 * overwritten inode. If we find an overwritten ref that is
3752 * not the first ref, simply unlink it.
3753 */
3754 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3755 cur->name, cur->name_len,
3756 &ow_inode, &ow_gen);
3757 if (ret < 0)
3758 goto out;
3759 if (ret) {
924794c9
TI		 3760 ret = is_first_ref(sctx->parent_root,
3761 ow_inode, cur->dir, cur->name,
3762 cur->name_len);
31db9f7c
AB		 3763 if (ret < 0)
3764 goto out;
3765 if (ret) {
8996a48c 3766 struct name_cache_entry *nce;
801bec36 3767 struct waiting_dir_move *wdm;
8996a48c 3768
31db9f7c
AB		 3769 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3770 cur->full_path);
3771 if (ret < 0)
3772 goto out;
801bec36
RK		 3773
3774 /*
3775 * If ow_inode has its rename operation delayed
3776 * make sure that its orphanized name is used in
3777 * the source path when performing its rename
3778 * operation.
3779 */
3780 if (is_waiting_for_move(sctx, ow_inode)) {
3781 wdm = get_waiting_dir_move(sctx,
3782 ow_inode);
3783 ASSERT(wdm);
3784 wdm->orphanized = true;
3785 }
3786
8996a48c
FM		 3787 /*
3788 * Make sure we clear our orphanized inode's
3789 * name from the name cache. This is because the
3790 * inode ow_inode might be an ancestor of some
3791 * other inode that will be orphanized as well
3792 * later and has an inode number greater than
3793 * sctx->send_progress. We need to prevent
3794 * future name lookups from using the old name
3795 * and get instead the orphan name.
3796 */
3797 nce = name_cache_search(sctx, ow_inode, ow_gen);
3798 if (nce) {
3799 name_cache_delete(sctx, nce);
3800 kfree(nce);
3801 }
801bec36
RK		 3802
3803 /*
3804 * ow_inode might currently be an ancestor of
3805 * cur_ino, therefore compute valid_path (the
3806 * current path of cur_ino) again because it
3807 * might contain the pre-orphanization name of
3808 * ow_inode, which is no longer valid.
3809 */
3810 fs_path_reset(valid_path);
3811 ret = get_cur_path(sctx, sctx->cur_ino,
3812 sctx->cur_inode_gen, valid_path);
3813 if (ret < 0)
3814 goto out;
31db9f7c
AB		 3815 } else {
3816 ret = send_unlink(sctx, cur->full_path);
3817 if (ret < 0)
3818 goto out;
3819 }
3820 }
3821
84471e24
FM		 3822 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3823 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3824 if (ret < 0)
3825 goto out;
3826 if (ret == 1) {
3827 can_rename = false;
3828 *pending_move = 1;
3829 }
3830 }
3831
8b191a68
FM		 3832 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3833 can_rename) {
3834 ret = wait_for_parent_move(sctx, cur, is_orphan);
3835 if (ret < 0)
3836 goto out;
3837 if (ret == 1) {
3838 can_rename = false;
3839 *pending_move = 1;
3840 }
3841 }
3842
31db9f7c
AB		 3843 /*
3844 * link/move the ref to the new place. If we have an orphan
3845 * inode, move it and update valid_path. If not, link or move
3846 * it depending on the inode mode.
3847 */
84471e24 3848 if (is_orphan && can_rename) {
31db9f7c
AB		 3849 ret = send_rename(sctx, valid_path, cur->full_path);
3850 if (ret < 0)
3851 goto out;
3852 is_orphan = 0;
3853 ret = fs_path_copy(valid_path, cur->full_path);
3854 if (ret < 0)
3855 goto out;
84471e24 3856 } else if (can_rename) {
31db9f7c
AB		 3857 if (S_ISDIR(sctx->cur_inode_mode)) {
3858 /*
3859 * Dirs can't be linked, so move it. For moved
3860 * dirs, we always have one new and one deleted
3861 * ref. The deleted ref is ignored later.
3862 */
8b191a68
FM		 3863 ret = send_rename(sctx, valid_path,
3864 cur->full_path);
3865 if (!ret)
3866 ret = fs_path_copy(valid_path,
3867 cur->full_path);
31db9f7c
AB		 3868 if (ret < 0)
3869 goto out;
3870 } else {
3871 ret = send_link(sctx, cur->full_path,
3872 valid_path);
3873 if (ret < 0)
3874 goto out;
3875 }
3876 }
ba5e8f2e 3877 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 3878 if (ret < 0)
3879 goto out;
3880 }
3881
3882 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3883 /*
3884 * Check if we can already rmdir the directory. If not,
3885 * orphanize it. For every dir item inside that gets deleted
3886 * later, we do this check again and rmdir it then if possible.
3887 * See the use of check_dirs for more details.
3888 */
9dc44214
FM		 3889 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3890 sctx->cur_ino);
31db9f7c
AB		 3891 if (ret < 0)
3892 goto out;
3893 if (ret) {
3894 ret = send_rmdir(sctx, valid_path);
3895 if (ret < 0)
3896 goto out;
3897 } else if (!is_orphan) {
3898 ret = orphanize_inode(sctx, sctx->cur_ino,
3899 sctx->cur_inode_gen, valid_path);
3900 if (ret < 0)
3901 goto out;
3902 is_orphan = 1;
3903 }
3904
3905 list_for_each_entry(cur, &sctx->deleted_refs, list) {
ba5e8f2e 3906 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 3907 if (ret < 0)
3908 goto out;
3909 }
ccf1626b
AB		 3910 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3911 !list_empty(&sctx->deleted_refs)) {
3912 /*
3913 * We have a moved dir. Add the old parent to check_dirs
3914 */
3915 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3916 list);
ba5e8f2e 3917 ret = dup_ref(cur, &check_dirs);
ccf1626b
AB		 3918 if (ret < 0)
3919 goto out;
31db9f7c
AB		 3920 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3921 /*
3922 * We have a non dir inode. Go through all deleted refs and
3923 * unlink them if they were not already overwritten by other
3924 * inodes.
3925 */
3926 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3927 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3928 sctx->cur_ino, sctx->cur_inode_gen,
3929 cur->name, cur->name_len);
3930 if (ret < 0)
3931 goto out;
3932 if (!ret) {
1f4692da
AB		 3933 ret = send_unlink(sctx, cur->full_path);
3934 if (ret < 0)
3935 goto out;
31db9f7c 3936 }
ba5e8f2e 3937 ret = dup_ref(cur, &check_dirs);
31db9f7c
AB		 3938 if (ret < 0)
3939 goto out;
3940 }
31db9f7c
AB		 3941 /*
3942 * If the inode is still orphan, unlink the orphan. This may
3943 * happen when a previous inode did overwrite the first ref
3944 * of this inode and no new refs were added for the current
766702ef
AB		 3945 * inode. Unlinking does not mean that the inode is deleted in
3946 * all cases. There may still be links to this inode in other
3947 * places.
31db9f7c 3948 */
1f4692da 3949 if (is_orphan) {
31db9f7c
AB		 3950 ret = send_unlink(sctx, valid_path);
3951 if (ret < 0)
3952 goto out;
3953 }
3954 }
3955
3956 /*
3957 * We did collect all parent dirs where cur_inode was once located. We
3958 * now go through all these dirs and check if they are pending for
3959 * deletion and if it's finally possible to perform the rmdir now.
3960 * We also update the inode stats of the parent dirs here.
3961 */
ba5e8f2e 3962 list_for_each_entry(cur, &check_dirs, list) {
766702ef
AB		 3963 /*
3964 * In case we had refs into dirs that were not processed yet,
3965 * we don't need to do the utime and rmdir logic for these dirs.
3966 * The dir will be processed later.
3967 */
ba5e8f2e 3968 if (cur->dir > sctx->cur_ino)
31db9f7c
AB		 3969 continue;
3970
ba5e8f2e 3971 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 3972 if (ret < 0)
3973 goto out;
3974
3975 if (ret == inode_state_did_create ||
3976 ret == inode_state_no_change) {
3977 /* TODO delayed utimes */
ba5e8f2e 3978 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
31db9f7c
AB		 3979 if (ret < 0)
3980 goto out;
29d6d30f
FM		 3981 } else if (ret == inode_state_did_delete &&
3982 cur->dir != last_dir_ino_rm) {
9dc44214
FM		 3983 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3984 sctx->cur_ino);
31db9f7c
AB		 3985 if (ret < 0)
3986 goto out;
3987 if (ret) {
ba5e8f2e
JB		 3988 ret = get_cur_path(sctx, cur->dir,
3989 cur->dir_gen, valid_path);
31db9f7c
AB		 3990 if (ret < 0)
3991 goto out;
3992 ret = send_rmdir(sctx, valid_path);
3993 if (ret < 0)
3994 goto out;
29d6d30f 3995 last_dir_ino_rm = cur->dir;
31db9f7c
AB		 3996 }
3997 }
3998 }
3999
31db9f7c
AB		 4000 ret = 0;
4001
4002out:
ba5e8f2e 4003 __free_recorded_refs(&check_dirs);
31db9f7c 4004 free_recorded_refs(sctx);
924794c9 4005 fs_path_free(valid_path);
31db9f7c
AB		 4006 return ret;
4007}
4008
a4d96d62
LB		 4009static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
4010 struct fs_path *name, void *ctx, struct list_head *refs)
31db9f7c
AB		 4011{
4012 int ret = 0;
4013 struct send_ctx *sctx = ctx;
4014 struct fs_path *p;
4015 u64 gen;
4016
924794c9 4017 p = fs_path_alloc();
31db9f7c
AB		 4018 if (!p)
4019 return -ENOMEM;
4020
a4d96d62 4021 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
85a7b33b 4022 NULL, NULL);
31db9f7c
AB		 4023 if (ret < 0)
4024 goto out;
4025
31db9f7c
AB		 4026 ret = get_cur_path(sctx, dir, gen, p);
4027 if (ret < 0)
4028 goto out;
4029 ret = fs_path_add_path(p, name);
4030 if (ret < 0)
4031 goto out;
4032
a4d96d62 4033 ret = __record_ref(refs, dir, gen, p);
31db9f7c
AB		 4034
4035out:
4036 if (ret)
924794c9 4037 fs_path_free(p);
31db9f7c
AB		 4038 return ret;
4039}
4040
a4d96d62
LB		 4041static int __record_new_ref(int num, u64 dir, int index,
4042 struct fs_path *name,
4043 void *ctx)
4044{
4045 struct send_ctx *sctx = ctx;
4046 return record_ref(sctx->send_root, num, dir, index, name,
4047 ctx, &sctx->new_refs);
4048}
4049
4050
31db9f7c
AB		 4051static int __record_deleted_ref(int num, u64 dir, int index,
4052 struct fs_path *name,
4053 void *ctx)
4054{
31db9f7c 4055 struct send_ctx *sctx = ctx;
a4d96d62
LB		 4056 return record_ref(sctx->parent_root, num, dir, index, name,
4057 ctx, &sctx->deleted_refs);
31db9f7c
AB		 4058}
4059
4060static int record_new_ref(struct send_ctx *sctx)
4061{
4062 int ret;
4063
924794c9
TI		 4064 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4065 sctx->cmp_key, 0, __record_new_ref, sctx);
31db9f7c
AB		 4066 if (ret < 0)
4067 goto out;
4068 ret = 0;
4069
4070out:
4071 return ret;
4072}
4073
4074static int record_deleted_ref(struct send_ctx *sctx)
4075{
4076 int ret;
4077
924794c9
TI		 4078 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4079 sctx->cmp_key, 0, __record_deleted_ref, sctx);
31db9f7c
AB		 4080 if (ret < 0)
4081 goto out;
4082 ret = 0;
4083
4084out:
4085 return ret;
4086}
4087
4088struct find_ref_ctx {
4089 u64 dir;
ba5e8f2e
JB		 4090 u64 dir_gen;
4091 struct btrfs_root *root;
31db9f7c
AB		 4092 struct fs_path *name;
4093 int found_idx;
4094};
4095
4096static int __find_iref(int num, u64 dir, int index,
4097 struct fs_path *name,
4098 void *ctx_)
4099{
4100 struct find_ref_ctx *ctx = ctx_;
ba5e8f2e
JB		 4101 u64 dir_gen;
4102 int ret;
31db9f7c
AB		 4103
4104 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4105 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
ba5e8f2e
JB		 4106 /*
4107 * To avoid doing extra lookups we'll only do this if everything
4108 * else matches.
4109 */
4110 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4111 NULL, NULL, NULL);
4112 if (ret)
4113 return ret;
4114 if (dir_gen != ctx->dir_gen)
4115 return 0;
31db9f7c
AB		 4116 ctx->found_idx = num;
4117 return 1;
4118 }
4119 return 0;
4120}
4121
924794c9 4122static int find_iref(struct btrfs_root *root,
31db9f7c
AB		 4123 struct btrfs_path *path,
4124 struct btrfs_key *key,
ba5e8f2e 4125 u64 dir, u64 dir_gen, struct fs_path *name)
31db9f7c
AB		 4126{
4127 int ret;
4128 struct find_ref_ctx ctx;
4129
4130 ctx.dir = dir;
4131 ctx.name = name;
ba5e8f2e 4132 ctx.dir_gen = dir_gen;
31db9f7c 4133 ctx.found_idx = -1;
ba5e8f2e 4134 ctx.root = root;
31db9f7c 4135
924794c9 4136 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
31db9f7c
AB		 4137 if (ret < 0)
4138 return ret;
4139
4140 if (ctx.found_idx == -1)
4141 return -ENOENT;
4142
4143 return ctx.found_idx;
4144}
4145
4146static int __record_changed_new_ref(int num, u64 dir, int index,
4147 struct fs_path *name,
4148 void *ctx)
4149{
ba5e8f2e 4150 u64 dir_gen;
31db9f7c
AB		 4151 int ret;
4152 struct send_ctx *sctx = ctx;
4153
ba5e8f2e
JB		 4154 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4155 NULL, NULL, NULL);
4156 if (ret)
4157 return ret;
4158
924794c9 4159 ret = find_iref(sctx->parent_root, sctx->right_path,
ba5e8f2e 4160 sctx->cmp_key, dir, dir_gen, name);
31db9f7c
AB		 4161 if (ret == -ENOENT)
4162 ret = __record_new_ref(num, dir, index, name, sctx);
4163 else if (ret > 0)
4164 ret = 0;
4165
4166 return ret;
4167}
4168
4169static int __record_changed_deleted_ref(int num, u64 dir, int index,
4170 struct fs_path *name,
4171 void *ctx)
4172{
ba5e8f2e 4173 u64 dir_gen;
31db9f7c
AB		 4174 int ret;
4175 struct send_ctx *sctx = ctx;
4176
ba5e8f2e
JB		 4177 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4178 NULL, NULL, NULL);
4179 if (ret)
4180 return ret;
4181
924794c9 4182 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
ba5e8f2e 4183 dir, dir_gen, name);
31db9f7c
AB		 4184 if (ret == -ENOENT)
4185 ret = __record_deleted_ref(num, dir, index, name, sctx);
4186 else if (ret > 0)
4187 ret = 0;
4188
4189 return ret;
4190}
4191
4192static int record_changed_ref(struct send_ctx *sctx)
4193{
4194 int ret = 0;
4195
924794c9 4196 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
31db9f7c
AB		 4197 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4198 if (ret < 0)
4199 goto out;
924794c9 4200 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 4201 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4202 if (ret < 0)
4203 goto out;
4204 ret = 0;
4205
4206out:
4207 return ret;
4208}
4209
4210/*
4211 * Record and process all refs at once. Needed when an inode changes the
4212 * generation number, which means that it was deleted and recreated.
4213 */
4214static int process_all_refs(struct send_ctx *sctx,
4215 enum btrfs_compare_tree_result cmd)
4216{
4217 int ret;
4218 struct btrfs_root *root;
4219 struct btrfs_path *path;
4220 struct btrfs_key key;
4221 struct btrfs_key found_key;
4222 struct extent_buffer *eb;
4223 int slot;
4224 iterate_inode_ref_t cb;
9f03740a 4225 int pending_move = 0;
31db9f7c
AB		 4226
4227 path = alloc_path_for_send();
4228 if (!path)
4229 return -ENOMEM;
4230
4231 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4232 root = sctx->send_root;
4233 cb = __record_new_ref;
4234 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4235 root = sctx->parent_root;
4236 cb = __record_deleted_ref;
4237 } else {
4d1a63b2
DS		 4238 btrfs_err(sctx->send_root->fs_info,
4239 "Wrong command %d in process_all_refs", cmd);
4240 ret = -EINVAL;
4241 goto out;
31db9f7c
AB		 4242 }
4243
4244 key.objectid = sctx->cmp_key->objectid;
4245 key.type = BTRFS_INODE_REF_KEY;
4246 key.offset = 0;
dff6d0ad
FDBM		 4247 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4248 if (ret < 0)
4249 goto out;
31db9f7c 4250
dff6d0ad 4251 while (1) {
31db9f7c
AB		 4252 eb = path->nodes[0];
4253 slot = path->slots[0];
dff6d0ad
FDBM		 4254 if (slot >= btrfs_header_nritems(eb)) {
4255 ret = btrfs_next_leaf(root, path);
4256 if (ret < 0)
4257 goto out;
4258 else if (ret > 0)
4259 break;
4260 continue;
4261 }
4262
31db9f7c
AB		 4263 btrfs_item_key_to_cpu(eb, &found_key, slot);
4264
4265 if (found_key.objectid != key.objectid ||
96b5bd77
JS		 4266 (found_key.type != BTRFS_INODE_REF_KEY &&
4267 found_key.type != BTRFS_INODE_EXTREF_KEY))
31db9f7c 4268 break;
31db9f7c 4269
924794c9 4270 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
31db9f7c
AB		 4271 if (ret < 0)
4272 goto out;
4273
dff6d0ad 4274 path->slots[0]++;
31db9f7c 4275 }
e938c8ad 4276 btrfs_release_path(path);
31db9f7c 4277
3dc09ec8
JB		 4278 /*
4279 * We don't actually care about pending_move as we are simply
4280 * re-creating this inode and will be rename'ing it into place once we
4281 * rename the parent directory.
4282 */
9f03740a 4283 ret = process_recorded_refs(sctx, &pending_move);
31db9f7c
AB		 4284out:
4285 btrfs_free_path(path);
4286 return ret;
4287}
4288
4289static int send_set_xattr(struct send_ctx *sctx,
4290 struct fs_path *path,
4291 const char *name, int name_len,
4292 const char *data, int data_len)
4293{
4294 int ret = 0;
4295
4296 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4297 if (ret < 0)
4298 goto out;
4299
4300 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4301 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4302 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4303
4304 ret = send_cmd(sctx);
4305
4306tlv_put_failure:
4307out:
4308 return ret;
4309}
4310
4311static int send_remove_xattr(struct send_ctx *sctx,
4312 struct fs_path *path,
4313 const char *name, int name_len)
4314{
4315 int ret = 0;
4316
4317 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4318 if (ret < 0)
4319 goto out;
4320
4321 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4322 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4323
4324 ret = send_cmd(sctx);
4325
4326tlv_put_failure:
4327out:
4328 return ret;
4329}
4330
4331static int __process_new_xattr(int num, struct btrfs_key *di_key,
4332 const char *name, int name_len,
4333 const char *data, int data_len,
4334 u8 type, void *ctx)
4335{
4336 int ret;
4337 struct send_ctx *sctx = ctx;
4338 struct fs_path *p;
2211d5ba 4339 struct posix_acl_xattr_header dummy_acl;
31db9f7c 4340
924794c9 4341 p = fs_path_alloc();
31db9f7c
AB		 4342 if (!p)
4343 return -ENOMEM;
4344
4345 /*
01327610 4346 * This hack is needed because empty acls are stored as zero byte
31db9f7c 4347 * data in xattrs. Problem with that is, that receiving these zero byte
01327610 4348 * acls will fail later. To fix this, we send a dummy acl list that
31db9f7c
AB		 4349 * only contains the version number and no entries.
4350 */
4351 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4352 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4353 if (data_len == 0) {
4354 dummy_acl.a_version =
4355 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4356 data = (char *)&dummy_acl;
4357 data_len = sizeof(dummy_acl);
4358 }
4359 }
4360
4361 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4362 if (ret < 0)
4363 goto out;
4364
4365 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4366
4367out:
924794c9 4368 fs_path_free(p);
31db9f7c
AB		 4369 return ret;
4370}
4371
4372static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4373 const char *name, int name_len,
4374 const char *data, int data_len,
4375 u8 type, void *ctx)
4376{
4377 int ret;
4378 struct send_ctx *sctx = ctx;
4379 struct fs_path *p;
4380
924794c9 4381 p = fs_path_alloc();
31db9f7c
AB		 4382 if (!p)
4383 return -ENOMEM;
4384
4385 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4386 if (ret < 0)
4387 goto out;
4388
4389 ret = send_remove_xattr(sctx, p, name, name_len);
4390
4391out:
924794c9 4392 fs_path_free(p);
31db9f7c
AB		 4393 return ret;
4394}
4395
4396static int process_new_xattr(struct send_ctx *sctx)
4397{
4398 int ret = 0;
4399
924794c9
TI		 4400 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4401 sctx->cmp_key, __process_new_xattr, sctx);
31db9f7c
AB		 4402
4403 return ret;
4404}
4405
4406static int process_deleted_xattr(struct send_ctx *sctx)
4407{
e2c89907
MY		 4408 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4409 sctx->cmp_key, __process_deleted_xattr, sctx);
31db9f7c
AB		 4410}
4411
4412struct find_xattr_ctx {
4413 const char *name;
4414 int name_len;
4415 int found_idx;
4416 char *found_data;
4417 int found_data_len;
4418};
4419
4420static int __find_xattr(int num, struct btrfs_key *di_key,
4421 const char *name, int name_len,
4422 const char *data, int data_len,
4423 u8 type, void *vctx)
4424{
4425 struct find_xattr_ctx *ctx = vctx;
4426
4427 if (name_len == ctx->name_len &&
4428 strncmp(name, ctx->name, name_len) == 0) {
4429 ctx->found_idx = num;
4430 ctx->found_data_len = data_len;
e780b0d1 4431 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
31db9f7c
AB		 4432 if (!ctx->found_data)
4433 return -ENOMEM;
31db9f7c
AB		 4434 return 1;
4435 }
4436 return 0;
4437}
4438
924794c9 4439static int find_xattr(struct btrfs_root *root,
31db9f7c
AB		 4440 struct btrfs_path *path,
4441 struct btrfs_key *key,
4442 const char *name, int name_len,
4443 char **data, int *data_len)
4444{
4445 int ret;
4446 struct find_xattr_ctx ctx;
4447
4448 ctx.name = name;
4449 ctx.name_len = name_len;
4450 ctx.found_idx = -1;
4451 ctx.found_data = NULL;
4452 ctx.found_data_len = 0;
4453
924794c9 4454 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
31db9f7c
AB		 4455 if (ret < 0)
4456 return ret;
4457
4458 if (ctx.found_idx == -1)
4459 return -ENOENT;
4460 if (data) {
4461 *data = ctx.found_data;
4462 *data_len = ctx.found_data_len;
4463 } else {
4464 kfree(ctx.found_data);
4465 }
4466 return ctx.found_idx;
4467}
4468
4469
4470static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4471 const char *name, int name_len,
4472 const char *data, int data_len,
4473 u8 type, void *ctx)
4474{
4475 int ret;
4476 struct send_ctx *sctx = ctx;
4477 char *found_data = NULL;
4478 int found_data_len = 0;
31db9f7c 4479
924794c9
TI		 4480 ret = find_xattr(sctx->parent_root, sctx->right_path,
4481 sctx->cmp_key, name, name_len, &found_data,
4482 &found_data_len);
31db9f7c
AB		 4483 if (ret == -ENOENT) {
4484 ret = __process_new_xattr(num, di_key, name, name_len, data,
4485 data_len, type, ctx);
4486 } else if (ret >= 0) {
4487 if (data_len != found_data_len ||
4488 memcmp(data, found_data, data_len)) {
4489 ret = __process_new_xattr(num, di_key, name, name_len,
4490 data, data_len, type, ctx);
4491 } else {
4492 ret = 0;
4493 }
4494 }
4495
4496 kfree(found_data);
31db9f7c
AB		 4497 return ret;
4498}
4499
4500static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4501 const char *name, int name_len,
4502 const char *data, int data_len,
4503 u8 type, void *ctx)
4504{
4505 int ret;
4506 struct send_ctx *sctx = ctx;
4507
924794c9
TI		 4508 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4509 name, name_len, NULL, NULL);
31db9f7c
AB		 4510 if (ret == -ENOENT)
4511 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4512 data_len, type, ctx);
4513 else if (ret >= 0)
4514 ret = 0;
4515
4516 return ret;
4517}
4518
4519static int process_changed_xattr(struct send_ctx *sctx)
4520{
4521 int ret = 0;
4522
924794c9 4523 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
31db9f7c
AB		 4524 sctx->cmp_key, __process_changed_new_xattr, sctx);
4525 if (ret < 0)
4526 goto out;
924794c9 4527 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
31db9f7c
AB		 4528 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4529
4530out:
4531 return ret;
4532}
4533
4534static int process_all_new_xattrs(struct send_ctx *sctx)
4535{
4536 int ret;
4537 struct btrfs_root *root;
4538 struct btrfs_path *path;
4539 struct btrfs_key key;
4540 struct btrfs_key found_key;
4541 struct extent_buffer *eb;
4542 int slot;
4543
4544 path = alloc_path_for_send();
4545 if (!path)
4546 return -ENOMEM;
4547
4548 root = sctx->send_root;
4549
4550 key.objectid = sctx->cmp_key->objectid;
4551 key.type = BTRFS_XATTR_ITEM_KEY;
4552 key.offset = 0;
dff6d0ad
FDBM		 4553 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4554 if (ret < 0)
4555 goto out;
31db9f7c 4556
dff6d0ad 4557 while (1) {
31db9f7c
AB		 4558 eb = path->nodes[0];
4559 slot = path->slots[0];
dff6d0ad
FDBM		 4560 if (slot >= btrfs_header_nritems(eb)) {
4561 ret = btrfs_next_leaf(root, path);
4562 if (ret < 0) {
4563 goto out;
4564 } else if (ret > 0) {
4565 ret = 0;
4566 break;
4567 }
4568 continue;
4569 }
31db9f7c 4570
dff6d0ad 4571 btrfs_item_key_to_cpu(eb, &found_key, slot);
31db9f7c
AB		 4572 if (found_key.objectid != key.objectid ||
4573 found_key.type != key.type) {
4574 ret = 0;
4575 goto out;
4576 }
4577
924794c9
TI		 4578 ret = iterate_dir_item(root, path, &found_key,
4579 __process_new_xattr, sctx);
31db9f7c
AB		 4580 if (ret < 0)
4581 goto out;
4582
dff6d0ad 4583 path->slots[0]++;
31db9f7c
AB		 4584 }
4585
4586out:
4587 btrfs_free_path(path);
4588 return ret;
4589}
4590
ed259095
JB		 4591static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4592{
4593 struct btrfs_root *root = sctx->send_root;
4594 struct btrfs_fs_info *fs_info = root->fs_info;
4595 struct inode *inode;
4596 struct page *page;
4597 char *addr;
4598 struct btrfs_key key;
09cbfeaf 4599 pgoff_t index = offset >> PAGE_SHIFT;
ed259095 4600 pgoff_t last_index;
09cbfeaf 4601 unsigned pg_offset = offset & ~PAGE_MASK;
ed259095
JB		 4602 ssize_t ret = 0;
4603
4604 key.objectid = sctx->cur_ino;
4605 key.type = BTRFS_INODE_ITEM_KEY;
4606 key.offset = 0;
4607
4608 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4609 if (IS_ERR(inode))
4610 return PTR_ERR(inode);
4611
4612 if (offset + len > i_size_read(inode)) {
4613 if (offset > i_size_read(inode))
4614 len = 0;
4615 else
4616 len = offset - i_size_read(inode);
4617 }
4618 if (len == 0)
4619 goto out;
4620
09cbfeaf 4621 last_index = (offset + len - 1) >> PAGE_SHIFT;
2131bcd3
LB		 4622
4623 /* initial readahead */
4624 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4625 file_ra_state_init(&sctx->ra, inode->i_mapping);
4626 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4627 last_index - index + 1);
4628
ed259095
JB		 4629 while (index <= last_index) {
4630 unsigned cur_len = min_t(unsigned, len,
09cbfeaf 4631 PAGE_SIZE - pg_offset);
e780b0d1 4632 page = find_or_create_page(inode->i_mapping, index, GFP_KERNEL);
ed259095
JB		 4633 if (!page) {
4634 ret = -ENOMEM;
4635 break;
4636 }
4637
4638 if (!PageUptodate(page)) {
4639 btrfs_readpage(NULL, page);
4640 lock_page(page);
4641 if (!PageUptodate(page)) {
4642 unlock_page(page);
09cbfeaf 4643 put_page(page);
ed259095
JB		 4644 ret = -EIO;
4645 break;
4646 }
4647 }
4648
4649 addr = kmap(page);
4650 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4651 kunmap(page);
4652 unlock_page(page);
09cbfeaf 4653 put_page(page);
ed259095
JB		 4654 index++;
4655 pg_offset = 0;
4656 len -= cur_len;
4657 ret += cur_len;
4658 }
4659out:
4660 iput(inode);
4661 return ret;
4662}
4663
31db9f7c
AB		 4664/*
4665 * Read some bytes from the current inode/file and send a write command to
4666 * user space.
4667 */
4668static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4669{
04ab956e 4670 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
31db9f7c
AB		 4671 int ret = 0;
4672 struct fs_path *p;
ed259095 4673 ssize_t num_read = 0;
31db9f7c 4674
924794c9 4675 p = fs_path_alloc();
31db9f7c
AB		 4676 if (!p)
4677 return -ENOMEM;
4678
04ab956e 4679 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
31db9f7c 4680
ed259095
JB		 4681 num_read = fill_read_buf(sctx, offset, len);
4682 if (num_read <= 0) {
4683 if (num_read < 0)
4684 ret = num_read;
31db9f7c 4685 goto out;
ed259095 4686 }
31db9f7c
AB		 4687
4688 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4689 if (ret < 0)
4690 goto out;
4691
4692 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4693 if (ret < 0)
4694 goto out;
4695
4696 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4697 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
e938c8ad 4698 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
31db9f7c
AB		 4699
4700 ret = send_cmd(sctx);
4701
4702tlv_put_failure:
4703out:
924794c9 4704 fs_path_free(p);
31db9f7c
AB		 4705 if (ret < 0)
4706 return ret;
e938c8ad 4707 return num_read;
31db9f7c
AB		 4708}
4709
4710/*
4711 * Send a clone command to user space.
4712 */
4713static int send_clone(struct send_ctx *sctx,
4714 u64 offset, u32 len,
4715 struct clone_root *clone_root)
4716{
4717 int ret = 0;
31db9f7c
AB		 4718 struct fs_path *p;
4719 u64 gen;
4720
04ab956e
JM		 4721 btrfs_debug(sctx->send_root->fs_info,
4722 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4723 offset, len, clone_root->root->objectid, clone_root->ino,
4724 clone_root->offset);
31db9f7c 4725
924794c9 4726 p = fs_path_alloc();
31db9f7c
AB		 4727 if (!p)
4728 return -ENOMEM;
4729
4730 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4731 if (ret < 0)
4732 goto out;
4733
4734 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4735 if (ret < 0)
4736 goto out;
4737
4738 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4739 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4740 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4741
e938c8ad 4742 if (clone_root->root == sctx->send_root) {
31db9f7c 4743 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
85a7b33b 4744 &gen, NULL, NULL, NULL, NULL);
31db9f7c
AB		 4745 if (ret < 0)
4746 goto out;
4747 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4748 } else {
924794c9 4749 ret = get_inode_path(clone_root->root, clone_root->ino, p);
31db9f7c
AB		 4750 }
4751 if (ret < 0)
4752 goto out;
4753
37b8d27d
JB		 4754 /*
4755 * If the parent we're using has a received_uuid set then use that as
4756 * our clone source as that is what we will look for when doing a
4757 * receive.
4758 *
4759 * This covers the case that we create a snapshot off of a received
4760 * subvolume and then use that as the parent and try to receive on a
4761 * different host.
4762 */
4763 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4764 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4765 clone_root->root->root_item.received_uuid);
4766 else
4767 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4768 clone_root->root->root_item.uuid);
31db9f7c 4769 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
5a0f4e2c 4770 le64_to_cpu(clone_root->root->root_item.ctransid));
31db9f7c
AB		 4771 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4772 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4773 clone_root->offset);
4774
4775 ret = send_cmd(sctx);
4776
4777tlv_put_failure:
4778out:
924794c9 4779 fs_path_free(p);
31db9f7c
AB		 4780 return ret;
4781}
4782
cb95e7bf
MF		 4783/*
4784 * Send an update extent command to user space.
4785 */
4786static int send_update_extent(struct send_ctx *sctx,
4787 u64 offset, u32 len)
4788{
4789 int ret = 0;
4790 struct fs_path *p;
4791
924794c9 4792 p = fs_path_alloc();
cb95e7bf
MF		 4793 if (!p)
4794 return -ENOMEM;
4795
4796 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4797 if (ret < 0)
4798 goto out;
4799
4800 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4801 if (ret < 0)
4802 goto out;
4803
4804 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4805 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4806 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4807
4808 ret = send_cmd(sctx);
4809
4810tlv_put_failure:
4811out:
924794c9 4812 fs_path_free(p);
cb95e7bf
MF		 4813 return ret;
4814}
4815
16e7549f
JB		 4816static int send_hole(struct send_ctx *sctx, u64 end)
4817{
4818 struct fs_path *p = NULL;
4819 u64 offset = sctx->cur_inode_last_extent;
4820 u64 len;
4821 int ret = 0;
4822
4823 p = fs_path_alloc();
4824 if (!p)
4825 return -ENOMEM;
c715e155
FM		 4826 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4827 if (ret < 0)
4828 goto tlv_put_failure;
16e7549f
JB		 4829 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4830 while (offset < end) {
4831 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4832
4833 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
16e7549f
JB		 4834 if (ret < 0)
4835 break;
4836 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4837 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4838 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4839 ret = send_cmd(sctx);
4840 if (ret < 0)
4841 break;
4842 offset += len;
4843 }
4844tlv_put_failure:
4845 fs_path_free(p);
4846 return ret;
4847}
4848
d906d49f
FM		 4849static int send_extent_data(struct send_ctx *sctx,
4850 const u64 offset,
4851 const u64 len)
4852{
4853 u64 sent = 0;
4854
4855 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4856 return send_update_extent(sctx, offset, len);
4857
4858 while (sent < len) {
4859 u64 size = len - sent;
4860 int ret;
4861
4862 if (size > BTRFS_SEND_READ_SIZE)
4863 size = BTRFS_SEND_READ_SIZE;
4864 ret = send_write(sctx, offset + sent, size);
4865 if (ret < 0)
4866 return ret;
4867 if (!ret)
4868 break;
4869 sent += ret;
4870 }
4871 return 0;
4872}
4873
4874static int clone_range(struct send_ctx *sctx,
4875 struct clone_root *clone_root,
4876 const u64 disk_byte,
4877 u64 data_offset,
4878 u64 offset,
4879 u64 len)
4880{
4881 struct btrfs_path *path;
4882 struct btrfs_key key;
4883 int ret;
4884
4885 path = alloc_path_for_send();
4886 if (!path)
4887 return -ENOMEM;
4888
4889 /*
4890 * We can't send a clone operation for the entire range if we find
4891 * extent items in the respective range in the source file that
4892 * refer to different extents or if we find holes.
4893 * So check for that and do a mix of clone and regular write/copy
4894 * operations if needed.
4895 *
4896 * Example:
4897 *
4898 * mkfs.btrfs -f /dev/sda
4899 * mount /dev/sda /mnt
4900 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
4901 * cp --reflink=always /mnt/foo /mnt/bar
4902 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
4903 * btrfs subvolume snapshot -r /mnt /mnt/snap
4904 *
4905 * If when we send the snapshot and we are processing file bar (which
4906 * has a higher inode number than foo) we blindly send a clone operation
4907 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
4908 * a file bar that matches the content of file foo - iow, doesn't match
4909 * the content from bar in the original filesystem.
4910 */
4911 key.objectid = clone_root->ino;
4912 key.type = BTRFS_EXTENT_DATA_KEY;
4913 key.offset = clone_root->offset;
4914 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
4915 if (ret < 0)
4916 goto out;
4917 if (ret > 0 && path->slots[0] > 0) {
4918 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4919 if (key.objectid == clone_root->ino &&
4920 key.type == BTRFS_EXTENT_DATA_KEY)
4921 path->slots[0]--;
4922 }
4923
4924 while (true) {
4925 struct extent_buffer *leaf = path->nodes[0];
4926 int slot = path->slots[0];
4927 struct btrfs_file_extent_item *ei;
4928 u8 type;
4929 u64 ext_len;
4930 u64 clone_len;
4931
4932 if (slot >= btrfs_header_nritems(leaf)) {
4933 ret = btrfs_next_leaf(clone_root->root, path);
4934 if (ret < 0)
4935 goto out;
4936 else if (ret > 0)
4937 break;
4938 continue;
4939 }
4940
4941 btrfs_item_key_to_cpu(leaf, &key, slot);
4942
4943 /*
4944 * We might have an implicit trailing hole (NO_HOLES feature
4945 * enabled). We deal with it after leaving this loop.
4946 */
4947 if (key.objectid != clone_root->ino ||
4948 key.type != BTRFS_EXTENT_DATA_KEY)
4949 break;
4950
4951 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4952 type = btrfs_file_extent_type(leaf, ei);
4953 if (type == BTRFS_FILE_EXTENT_INLINE) {
4954 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
09cbfeaf 4955 ext_len = PAGE_ALIGN(ext_len);
d906d49f
FM		 4956 } else {
4957 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
4958 }
4959
4960 if (key.offset + ext_len <= clone_root->offset)
4961 goto next;
4962
4963 if (key.offset > clone_root->offset) {
4964 /* Implicit hole, NO_HOLES feature enabled. */
4965 u64 hole_len = key.offset - clone_root->offset;
4966
4967 if (hole_len > len)
4968 hole_len = len;
4969 ret = send_extent_data(sctx, offset, hole_len);
4970 if (ret < 0)
4971 goto out;
4972
4973 len -= hole_len;
4974 if (len == 0)
4975 break;
4976 offset += hole_len;
4977 clone_root->offset += hole_len;
4978 data_offset += hole_len;
4979 }
4980
4981 if (key.offset >= clone_root->offset + len)
4982 break;
4983
4984 clone_len = min_t(u64, ext_len, len);
4985
4986 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
4987 btrfs_file_extent_offset(leaf, ei) == data_offset)
4988 ret = send_clone(sctx, offset, clone_len, clone_root);
4989 else
4990 ret = send_extent_data(sctx, offset, clone_len);
4991
4992 if (ret < 0)
4993 goto out;
4994
4995 len -= clone_len;
4996 if (len == 0)
4997 break;
4998 offset += clone_len;
4999 clone_root->offset += clone_len;
5000 data_offset += clone_len;
5001next:
5002 path->slots[0]++;
5003 }
5004
5005 if (len > 0)
5006 ret = send_extent_data(sctx, offset, len);
5007 else
5008 ret = 0;
5009out:
5010 btrfs_free_path(path);
5011 return ret;
5012}
5013
31db9f7c
AB		 5014static int send_write_or_clone(struct send_ctx *sctx,
5015 struct btrfs_path *path,
5016 struct btrfs_key *key,
5017 struct clone_root *clone_root)
5018{
5019 int ret = 0;
5020 struct btrfs_file_extent_item *ei;
5021 u64 offset = key->offset;
31db9f7c 5022 u64 len;
31db9f7c 5023 u8 type;
28e5dd8f 5024 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
31db9f7c
AB		 5025
5026 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5027 struct btrfs_file_extent_item);
5028 type = btrfs_file_extent_type(path->nodes[0], ei);
74dd17fb 5029 if (type == BTRFS_FILE_EXTENT_INLINE) {
514ac8ad
CM		 5030 len = btrfs_file_extent_inline_len(path->nodes[0],
5031 path->slots[0], ei);
74dd17fb
CM		 5032 /*
5033 * it is possible the inline item won't cover the whole page,
5034 * but there may be items after this page. Make
5035 * sure to send the whole thing
5036 */
09cbfeaf 5037 len = PAGE_ALIGN(len);
74dd17fb 5038 } else {
31db9f7c 5039 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
74dd17fb 5040 }
31db9f7c
AB		 5041
5042 if (offset + len > sctx->cur_inode_size)
5043 len = sctx->cur_inode_size - offset;
5044 if (len == 0) {
5045 ret = 0;
5046 goto out;
5047 }
5048
28e5dd8f 5049 if (clone_root && IS_ALIGNED(offset + len, bs)) {
d906d49f
FM		 5050 u64 disk_byte;
5051 u64 data_offset;
5052
5053 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5054 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5055 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5056 offset, len);
cb95e7bf 5057 } else {
d906d49f 5058 ret = send_extent_data(sctx, offset, len);
31db9f7c 5059 }
31db9f7c
AB		 5060out:
5061 return ret;
5062}
5063
5064static int is_extent_unchanged(struct send_ctx *sctx,
5065 struct btrfs_path *left_path,
5066 struct btrfs_key *ekey)
5067{
5068 int ret = 0;
5069 struct btrfs_key key;
5070 struct btrfs_path *path = NULL;
5071 struct extent_buffer *eb;
5072 int slot;
5073 struct btrfs_key found_key;
5074 struct btrfs_file_extent_item *ei;
5075 u64 left_disknr;
5076 u64 right_disknr;
5077 u64 left_offset;
5078 u64 right_offset;
5079 u64 left_offset_fixed;
5080 u64 left_len;
5081 u64 right_len;
74dd17fb
CM		 5082 u64 left_gen;
5083 u64 right_gen;
31db9f7c
AB		 5084 u8 left_type;
5085 u8 right_type;
5086
5087 path = alloc_path_for_send();
5088 if (!path)
5089 return -ENOMEM;
5090
5091 eb = left_path->nodes[0];
5092 slot = left_path->slots[0];
31db9f7c
AB		 5093 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5094 left_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 5095
5096 if (left_type != BTRFS_FILE_EXTENT_REG) {
5097 ret = 0;
5098 goto out;
5099 }
74dd17fb
CM		 5100 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5101 left_len = btrfs_file_extent_num_bytes(eb, ei);
5102 left_offset = btrfs_file_extent_offset(eb, ei);
5103 left_gen = btrfs_file_extent_generation(eb, ei);
31db9f7c
AB		 5104
5105 /*
5106 * Following comments will refer to these graphics. L is the left
5107 * extents which we are checking at the moment. 1-8 are the right
5108 * extents that we iterate.
5109 *
5110 * |-----L-----|
5111 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5112 *
5113 * |-----L-----|
5114 * |--1--|-2b-|...(same as above)
5115 *
5116 * Alternative situation. Happens on files where extents got split.
5117 * |-----L-----|
5118 * |-----------7-----------|-6-|
5119 *
5120 * Alternative situation. Happens on files which got larger.
5121 * |-----L-----|
5122 * |-8-|
5123 * Nothing follows after 8.
5124 */
5125
5126 key.objectid = ekey->objectid;
5127 key.type = BTRFS_EXTENT_DATA_KEY;
5128 key.offset = ekey->offset;
5129 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5130 if (ret < 0)
5131 goto out;
5132 if (ret) {
5133 ret = 0;
5134 goto out;
5135 }
5136
5137 /*
5138 * Handle special case where the right side has no extents at all.
5139 */
5140 eb = path->nodes[0];
5141 slot = path->slots[0];
5142 btrfs_item_key_to_cpu(eb, &found_key, slot);
5143 if (found_key.objectid != key.objectid ||
5144 found_key.type != key.type) {
57cfd462
JB		 5145 /* If we're a hole then just pretend nothing changed */
5146 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 5147 goto out;
5148 }
5149
5150 /*
5151 * We're now on 2a, 2b or 7.
5152 */
5153 key = found_key;
5154 while (key.offset < ekey->offset + left_len) {
5155 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5156 right_type = btrfs_file_extent_type(eb, ei);
31db9f7c
AB		 5157 if (right_type != BTRFS_FILE_EXTENT_REG) {
5158 ret = 0;
5159 goto out;
5160 }
5161
007d31f7
JB		 5162 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5163 right_len = btrfs_file_extent_num_bytes(eb, ei);
5164 right_offset = btrfs_file_extent_offset(eb, ei);
5165 right_gen = btrfs_file_extent_generation(eb, ei);
5166
31db9f7c
AB		 5167 /*
5168 * Are we at extent 8? If yes, we know the extent is changed.
5169 * This may only happen on the first iteration.
5170 */
d8347fa4 5171 if (found_key.offset + right_len <= ekey->offset) {
57cfd462
JB		 5172 /* If we're a hole just pretend nothing changed */
5173 ret = (left_disknr) ? 0 : 1;
31db9f7c
AB		 5174 goto out;
5175 }
5176
5177 left_offset_fixed = left_offset;
5178 if (key.offset < ekey->offset) {
5179 /* Fix the right offset for 2a and 7. */
5180 right_offset += ekey->offset - key.offset;
5181 } else {
5182 /* Fix the left offset for all behind 2a and 2b */
5183 left_offset_fixed += key.offset - ekey->offset;
5184 }
5185
5186 /*
5187 * Check if we have the same extent.
5188 */
3954096d 5189 if (left_disknr != right_disknr ||
74dd17fb
CM		 5190 left_offset_fixed != right_offset ||
5191 left_gen != right_gen) {
31db9f7c
AB		 5192 ret = 0;
5193 goto out;
5194 }
5195
5196 /*
5197 * Go to the next extent.
5198 */
5199 ret = btrfs_next_item(sctx->parent_root, path);
5200 if (ret < 0)
5201 goto out;
5202 if (!ret) {
5203 eb = path->nodes[0];
5204 slot = path->slots[0];
5205 btrfs_item_key_to_cpu(eb, &found_key, slot);
5206 }
5207 if (ret || found_key.objectid != key.objectid ||
5208 found_key.type != key.type) {
5209 key.offset += right_len;
5210 break;
adaa4b8e
JS		 5211 }
5212 if (found_key.offset != key.offset + right_len) {
5213 ret = 0;
5214 goto out;
31db9f7c
AB		 5215 }
5216 key = found_key;
5217 }
5218
5219 /*
5220 * We're now behind the left extent (treat as unchanged) or at the end
5221 * of the right side (treat as changed).
5222 */
5223 if (key.offset >= ekey->offset + left_len)
5224 ret = 1;
5225 else
5226 ret = 0;
5227
5228
5229out:
5230 btrfs_free_path(path);
5231 return ret;
5232}
5233
16e7549f
JB		 5234static int get_last_extent(struct send_ctx *sctx, u64 offset)
5235{
5236 struct btrfs_path *path;
5237 struct btrfs_root *root = sctx->send_root;
5238 struct btrfs_file_extent_item *fi;
5239 struct btrfs_key key;
5240 u64 extent_end;
5241 u8 type;
5242 int ret;
5243
5244 path = alloc_path_for_send();
5245 if (!path)
5246 return -ENOMEM;
5247
5248 sctx->cur_inode_last_extent = 0;
5249
5250 key.objectid = sctx->cur_ino;
5251 key.type = BTRFS_EXTENT_DATA_KEY;
5252 key.offset = offset;
5253 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5254 if (ret < 0)
5255 goto out;
5256 ret = 0;
5257 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5258 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5259 goto out;
5260
5261 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5262 struct btrfs_file_extent_item);
5263 type = btrfs_file_extent_type(path->nodes[0], fi);
5264 if (type == BTRFS_FILE_EXTENT_INLINE) {
514ac8ad
CM		 5265 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5266 path->slots[0], fi);
16e7549f 5267 extent_end = ALIGN(key.offset + size,
da17066c 5268 sctx->send_root->fs_info->sectorsize);
16e7549f
JB		 5269 } else {
5270 extent_end = key.offset +
5271 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5272 }
5273 sctx->cur_inode_last_extent = extent_end;
5274out:
5275 btrfs_free_path(path);
5276 return ret;
5277}
5278
5279static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5280 struct btrfs_key *key)
5281{
5282 struct btrfs_file_extent_item *fi;
5283 u64 extent_end;
5284 u8 type;
5285 int ret = 0;
5286
5287 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5288 return 0;
5289
5290 if (sctx->cur_inode_last_extent == (u64)-1) {
5291 ret = get_last_extent(sctx, key->offset - 1);
5292 if (ret)
5293 return ret;
5294 }
5295
5296 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5297 struct btrfs_file_extent_item);
5298 type = btrfs_file_extent_type(path->nodes[0], fi);
5299 if (type == BTRFS_FILE_EXTENT_INLINE) {
514ac8ad
CM		 5300 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5301 path->slots[0], fi);
16e7549f 5302 extent_end = ALIGN(key->offset + size,
da17066c 5303 sctx->send_root->fs_info->sectorsize);
16e7549f
JB		 5304 } else {
5305 extent_end = key->offset +
5306 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5307 }
bf54f412
FDBM		 5308
5309 if (path->slots[0] == 0 &&
5310 sctx->cur_inode_last_extent < key->offset) {
5311 /*
5312 * We might have skipped entire leafs that contained only
5313 * file extent items for our current inode. These leafs have
5314 * a generation number smaller (older) than the one in the
5315 * current leaf and the leaf our last extent came from, and
5316 * are located between these 2 leafs.
5317 */
5318 ret = get_last_extent(sctx, key->offset - 1);
5319 if (ret)
5320 return ret;
5321 }
5322
16e7549f
JB		 5323 if (sctx->cur_inode_last_extent < key->offset)
5324 ret = send_hole(sctx, key->offset);
5325 sctx->cur_inode_last_extent = extent_end;
5326 return ret;
5327}
5328
31db9f7c
AB		 5329static int process_extent(struct send_ctx *sctx,
5330 struct btrfs_path *path,
5331 struct btrfs_key *key)
5332{
31db9f7c 5333 struct clone_root *found_clone = NULL;
57cfd462 5334 int ret = 0;
31db9f7c
AB		 5335
5336 if (S_ISLNK(sctx->cur_inode_mode))
5337 return 0;
5338
5339 if (sctx->parent_root && !sctx->cur_inode_new) {
5340 ret = is_extent_unchanged(sctx, path, key);
5341 if (ret < 0)
5342 goto out;
5343 if (ret) {
5344 ret = 0;
16e7549f 5345 goto out_hole;
31db9f7c 5346 }
57cfd462
JB		 5347 } else {
5348 struct btrfs_file_extent_item *ei;
5349 u8 type;
5350
5351 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5352 struct btrfs_file_extent_item);
5353 type = btrfs_file_extent_type(path->nodes[0], ei);
5354 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5355 type == BTRFS_FILE_EXTENT_REG) {
5356 /*
5357 * The send spec does not have a prealloc command yet,
5358 * so just leave a hole for prealloc'ed extents until
5359 * we have enough commands queued up to justify rev'ing
5360 * the send spec.
5361 */
5362 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5363 ret = 0;
5364 goto out;
5365 }
5366
5367 /* Have a hole, just skip it. */
5368 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5369 ret = 0;
5370 goto out;
5371 }
5372 }
31db9f7c
AB		 5373 }
5374
5375 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5376 sctx->cur_inode_size, &found_clone);
5377 if (ret != -ENOENT && ret < 0)
5378 goto out;
5379
5380 ret = send_write_or_clone(sctx, path, key, found_clone);
16e7549f
JB		 5381 if (ret)
5382 goto out;
5383out_hole:
5384 ret = maybe_send_hole(sctx, path, key);
31db9f7c
AB		 5385out:
5386 return ret;
5387}
5388
5389static int process_all_extents(struct send_ctx *sctx)
5390{
5391 int ret;
5392 struct btrfs_root *root;
5393 struct btrfs_path *path;
5394 struct btrfs_key key;
5395 struct btrfs_key found_key;
5396 struct extent_buffer *eb;
5397 int slot;
5398
5399 root = sctx->send_root;
5400 path = alloc_path_for_send();
5401 if (!path)
5402 return -ENOMEM;
5403
5404 key.objectid = sctx->cmp_key->objectid;
5405 key.type = BTRFS_EXTENT_DATA_KEY;
5406 key.offset = 0;
7fdd29d0
FDBM		 5407 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5408 if (ret < 0)
5409 goto out;
31db9f7c 5410
7fdd29d0 5411 while (1) {
31db9f7c
AB		 5412 eb = path->nodes[0];
5413 slot = path->slots[0];
7fdd29d0
FDBM		 5414
5415 if (slot >= btrfs_header_nritems(eb)) {
5416 ret = btrfs_next_leaf(root, path);
5417 if (ret < 0) {
5418 goto out;
5419 } else if (ret > 0) {
5420 ret = 0;
5421 break;
5422 }
5423 continue;
5424 }
5425
31db9f7c
AB		 5426 btrfs_item_key_to_cpu(eb, &found_key, slot);
5427
5428 if (found_key.objectid != key.objectid ||
5429 found_key.type != key.type) {
5430 ret = 0;
5431 goto out;
5432 }
5433
5434 ret = process_extent(sctx, path, &found_key);
5435 if (ret < 0)
5436 goto out;
5437
7fdd29d0 5438 path->slots[0]++;
31db9f7c
AB		 5439 }
5440
5441out:
5442 btrfs_free_path(path);
5443 return ret;
5444}
5445
9f03740a
FDBM		 5446static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5447 int *pending_move,
5448 int *refs_processed)
31db9f7c
AB		 5449{
5450 int ret = 0;
5451
5452 if (sctx->cur_ino == 0)
5453 goto out;
5454 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
96b5bd77 5455 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 5456 goto out;
5457 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5458 goto out;
5459
9f03740a 5460 ret = process_recorded_refs(sctx, pending_move);
e479d9bb
AB		 5461 if (ret < 0)
5462 goto out;
5463
9f03740a 5464 *refs_processed = 1;
31db9f7c
AB		 5465out:
5466 return ret;
5467}
5468
5469static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5470{
5471 int ret = 0;
5472 u64 left_mode;
5473 u64 left_uid;
5474 u64 left_gid;
5475 u64 right_mode;
5476 u64 right_uid;
5477 u64 right_gid;
5478 int need_chmod = 0;
5479 int need_chown = 0;
9f03740a
FDBM		 5480 int pending_move = 0;
5481 int refs_processed = 0;
31db9f7c 5482
9f03740a
FDBM		 5483 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5484 &refs_processed);
31db9f7c
AB		 5485 if (ret < 0)
5486 goto out;
5487
9f03740a
FDBM		 5488 /*
5489 * We have processed the refs and thus need to advance send_progress.
5490 * Now, calls to get_cur_xxx will take the updated refs of the current
5491 * inode into account.
5492 *
5493 * On the other hand, if our current inode is a directory and couldn't
5494 * be moved/renamed because its parent was renamed/moved too and it has
5495 * a higher inode number, we can only move/rename our current inode
5496 * after we moved/renamed its parent. Therefore in this case operate on
5497 * the old path (pre move/rename) of our current inode, and the
5498 * move/rename will be performed later.
5499 */
5500 if (refs_processed && !pending_move)
5501 sctx->send_progress = sctx->cur_ino + 1;
5502
31db9f7c
AB		 5503 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5504 goto out;
5505 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5506 goto out;
5507
5508 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
85a7b33b 5509 &left_mode, &left_uid, &left_gid, NULL);
31db9f7c
AB		 5510 if (ret < 0)
5511 goto out;
5512
e2d044fe
AL		 5513 if (!sctx->parent_root || sctx->cur_inode_new) {
5514 need_chown = 1;
5515 if (!S_ISLNK(sctx->cur_inode_mode))
31db9f7c 5516 need_chmod = 1;
e2d044fe
AL		 5517 } else {
5518 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5519 NULL, NULL, &right_mode, &right_uid,
5520 &right_gid, NULL);
5521 if (ret < 0)
5522 goto out;
31db9f7c 5523
e2d044fe
AL		 5524 if (left_uid != right_uid || left_gid != right_gid)
5525 need_chown = 1;
5526 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5527 need_chmod = 1;
31db9f7c
AB		 5528 }
5529
5530 if (S_ISREG(sctx->cur_inode_mode)) {
16e7549f 5531 if (need_send_hole(sctx)) {
766b5e5a
FM		 5532 if (sctx->cur_inode_last_extent == (u64)-1 ||
5533 sctx->cur_inode_last_extent <
5534 sctx->cur_inode_size) {
16e7549f
JB		 5535 ret = get_last_extent(sctx, (u64)-1);
5536 if (ret)
5537 goto out;
5538 }
5539 if (sctx->cur_inode_last_extent <
5540 sctx->cur_inode_size) {
5541 ret = send_hole(sctx, sctx->cur_inode_size);
5542 if (ret)
5543 goto out;
5544 }
5545 }
31db9f7c
AB		 5546 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5547 sctx->cur_inode_size);
5548 if (ret < 0)
5549 goto out;
5550 }
5551
5552 if (need_chown) {
5553 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5554 left_uid, left_gid);
5555 if (ret < 0)
5556 goto out;
5557 }
5558 if (need_chmod) {
5559 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5560 left_mode);
5561 if (ret < 0)
5562 goto out;
5563 }
5564
5565 /*
9f03740a
FDBM		 5566 * If other directory inodes depended on our current directory
5567 * inode's move/rename, now do their move/rename operations.
31db9f7c 5568 */
9f03740a
FDBM		 5569 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5570 ret = apply_children_dir_moves(sctx);
5571 if (ret)
5572 goto out;
fcbd2154
FM		 5573 /*
5574 * Need to send that every time, no matter if it actually
5575 * changed between the two trees as we have done changes to
5576 * the inode before. If our inode is a directory and it's
5577 * waiting to be moved/renamed, we will send its utimes when
5578 * it's moved/renamed, therefore we don't need to do it here.
5579 */
5580 sctx->send_progress = sctx->cur_ino + 1;
5581 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5582 if (ret < 0)
5583 goto out;
9f03740a
FDBM		 5584 }
5585
31db9f7c
AB		 5586out:
5587 return ret;
5588}
5589
5590static int changed_inode(struct send_ctx *sctx,
5591 enum btrfs_compare_tree_result result)
5592{
5593 int ret = 0;
5594 struct btrfs_key *key = sctx->cmp_key;
5595 struct btrfs_inode_item *left_ii = NULL;
5596 struct btrfs_inode_item *right_ii = NULL;
5597 u64 left_gen = 0;
5598 u64 right_gen = 0;
5599
31db9f7c
AB		 5600 sctx->cur_ino = key->objectid;
5601 sctx->cur_inode_new_gen = 0;
16e7549f 5602 sctx->cur_inode_last_extent = (u64)-1;
e479d9bb
AB		 5603
5604 /*
5605 * Set send_progress to current inode. This will tell all get_cur_xxx
5606 * functions that the current inode's refs are not updated yet. Later,
5607 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5608 */
31db9f7c
AB		 5609 sctx->send_progress = sctx->cur_ino;
5610
5611 if (result == BTRFS_COMPARE_TREE_NEW ||
5612 result == BTRFS_COMPARE_TREE_CHANGED) {
5613 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5614 sctx->left_path->slots[0],
5615 struct btrfs_inode_item);
5616 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5617 left_ii);
5618 } else {
5619 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5620 sctx->right_path->slots[0],
5621 struct btrfs_inode_item);
5622 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5623 right_ii);
5624 }
5625 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5626 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5627 sctx->right_path->slots[0],
5628 struct btrfs_inode_item);
5629
5630 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5631 right_ii);
6d85ed05
AB		 5632
5633 /*
5634 * The cur_ino = root dir case is special here. We can't treat
5635 * the inode as deleted+reused because it would generate a
5636 * stream that tries to delete/mkdir the root dir.
5637 */
5638 if (left_gen != right_gen &&
5639 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
31db9f7c
AB		 5640 sctx->cur_inode_new_gen = 1;
5641 }
5642
5643 if (result == BTRFS_COMPARE_TREE_NEW) {
5644 sctx->cur_inode_gen = left_gen;
5645 sctx->cur_inode_new = 1;
5646 sctx->cur_inode_deleted = 0;
5647 sctx->cur_inode_size = btrfs_inode_size(
5648 sctx->left_path->nodes[0], left_ii);
5649 sctx->cur_inode_mode = btrfs_inode_mode(
5650 sctx->left_path->nodes[0], left_ii);
644d1940
LB		 5651 sctx->cur_inode_rdev = btrfs_inode_rdev(
5652 sctx->left_path->nodes[0], left_ii);
31db9f7c 5653 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
1f4692da 5654 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 5655 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5656 sctx->cur_inode_gen = right_gen;
5657 sctx->cur_inode_new = 0;
5658 sctx->cur_inode_deleted = 1;
5659 sctx->cur_inode_size = btrfs_inode_size(
5660 sctx->right_path->nodes[0], right_ii);
5661 sctx->cur_inode_mode = btrfs_inode_mode(
5662 sctx->right_path->nodes[0], right_ii);
5663 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
766702ef
AB		 5664 /*
5665 * We need to do some special handling in case the inode was
5666 * reported as changed with a changed generation number. This
5667 * means that the original inode was deleted and new inode
5668 * reused the same inum. So we have to treat the old inode as
5669 * deleted and the new one as new.
5670 */
31db9f7c 5671 if (sctx->cur_inode_new_gen) {
766702ef
AB		 5672 /*
5673 * First, process the inode as if it was deleted.
5674 */
31db9f7c
AB		 5675 sctx->cur_inode_gen = right_gen;
5676 sctx->cur_inode_new = 0;
5677 sctx->cur_inode_deleted = 1;
5678 sctx->cur_inode_size = btrfs_inode_size(
5679 sctx->right_path->nodes[0], right_ii);
5680 sctx->cur_inode_mode = btrfs_inode_mode(
5681 sctx->right_path->nodes[0], right_ii);
5682 ret = process_all_refs(sctx,
5683 BTRFS_COMPARE_TREE_DELETED);
5684 if (ret < 0)
5685 goto out;
5686
766702ef
AB		 5687 /*
5688 * Now process the inode as if it was new.
5689 */
31db9f7c
AB		 5690 sctx->cur_inode_gen = left_gen;
5691 sctx->cur_inode_new = 1;
5692 sctx->cur_inode_deleted = 0;
5693 sctx->cur_inode_size = btrfs_inode_size(
5694 sctx->left_path->nodes[0], left_ii);
5695 sctx->cur_inode_mode = btrfs_inode_mode(
5696 sctx->left_path->nodes[0], left_ii);
644d1940
LB		 5697 sctx->cur_inode_rdev = btrfs_inode_rdev(
5698 sctx->left_path->nodes[0], left_ii);
1f4692da 5699 ret = send_create_inode_if_needed(sctx);
31db9f7c
AB		 5700 if (ret < 0)
5701 goto out;
5702
5703 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5704 if (ret < 0)
5705 goto out;
e479d9bb
AB		 5706 /*
5707 * Advance send_progress now as we did not get into
5708 * process_recorded_refs_if_needed in the new_gen case.
5709 */
5710 sctx->send_progress = sctx->cur_ino + 1;
766702ef
AB		 5711
5712 /*
5713 * Now process all extents and xattrs of the inode as if
5714 * they were all new.
5715 */
31db9f7c
AB		 5716 ret = process_all_extents(sctx);
5717 if (ret < 0)
5718 goto out;
5719 ret = process_all_new_xattrs(sctx);
5720 if (ret < 0)
5721 goto out;
5722 } else {
5723 sctx->cur_inode_gen = left_gen;
5724 sctx->cur_inode_new = 0;
5725 sctx->cur_inode_new_gen = 0;
5726 sctx->cur_inode_deleted = 0;
5727 sctx->cur_inode_size = btrfs_inode_size(
5728 sctx->left_path->nodes[0], left_ii);
5729 sctx->cur_inode_mode = btrfs_inode_mode(
5730 sctx->left_path->nodes[0], left_ii);
5731 }
5732 }
5733
5734out:
5735 return ret;
5736}
5737
766702ef
AB		 5738/*
5739 * We have to process new refs before deleted refs, but compare_trees gives us
5740 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5741 * first and later process them in process_recorded_refs.
5742 * For the cur_inode_new_gen case, we skip recording completely because
5743 * changed_inode did already initiate processing of refs. The reason for this is
5744 * that in this case, compare_tree actually compares the refs of 2 different
5745 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5746 * refs of the right tree as deleted and all refs of the left tree as new.
5747 */
31db9f7c
AB		 5748static int changed_ref(struct send_ctx *sctx,
5749 enum btrfs_compare_tree_result result)
5750{
5751 int ret = 0;
5752
95155585
FM		 5753 if (sctx->cur_ino != sctx->cmp_key->objectid) {
5754 inconsistent_snapshot_error(sctx, result, "reference");
5755 return -EIO;
5756 }
31db9f7c
AB		 5757
5758 if (!sctx->cur_inode_new_gen &&
5759 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5760 if (result == BTRFS_COMPARE_TREE_NEW)
5761 ret = record_new_ref(sctx);
5762 else if (result == BTRFS_COMPARE_TREE_DELETED)
5763 ret = record_deleted_ref(sctx);
5764 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5765 ret = record_changed_ref(sctx);
5766 }
5767
5768 return ret;
5769}
5770
766702ef
AB		 5771/*
5772 * Process new/deleted/changed xattrs. We skip processing in the
5773 * cur_inode_new_gen case because changed_inode did already initiate processing
5774 * of xattrs. The reason is the same as in changed_ref
5775 */
31db9f7c
AB		 5776static int changed_xattr(struct send_ctx *sctx,
5777 enum btrfs_compare_tree_result result)
5778{
5779 int ret = 0;
5780
95155585
FM		 5781 if (sctx->cur_ino != sctx->cmp_key->objectid) {
5782 inconsistent_snapshot_error(sctx, result, "xattr");
5783 return -EIO;
5784 }
31db9f7c
AB		 5785
5786 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5787 if (result == BTRFS_COMPARE_TREE_NEW)
5788 ret = process_new_xattr(sctx);
5789 else if (result == BTRFS_COMPARE_TREE_DELETED)
5790 ret = process_deleted_xattr(sctx);
5791 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5792 ret = process_changed_xattr(sctx);
5793 }
5794
5795 return ret;
5796}
5797
766702ef
AB		 5798/*
5799 * Process new/deleted/changed extents. We skip processing in the
5800 * cur_inode_new_gen case because changed_inode did already initiate processing
5801 * of extents. The reason is the same as in changed_ref
5802 */
31db9f7c
AB		 5803static int changed_extent(struct send_ctx *sctx,
5804 enum btrfs_compare_tree_result result)
5805{
5806 int ret = 0;
5807
95155585 5808 if (sctx->cur_ino != sctx->cmp_key->objectid) {
d5e84fd8
FM		 5809
5810 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5811 struct extent_buffer *leaf_l;
5812 struct extent_buffer *leaf_r;
5813 struct btrfs_file_extent_item *ei_l;
5814 struct btrfs_file_extent_item *ei_r;
5815
5816 leaf_l = sctx->left_path->nodes[0];
5817 leaf_r = sctx->right_path->nodes[0];
5818 ei_l = btrfs_item_ptr(leaf_l,
5819 sctx->left_path->slots[0],
5820 struct btrfs_file_extent_item);
5821 ei_r = btrfs_item_ptr(leaf_r,
5822 sctx->right_path->slots[0],
5823 struct btrfs_file_extent_item);
5824
5825 /*
5826 * We may have found an extent item that has changed
5827 * only its disk_bytenr field and the corresponding
5828 * inode item was not updated. This case happens due to
5829 * very specific timings during relocation when a leaf
5830 * that contains file extent items is COWed while
5831 * relocation is ongoing and its in the stage where it
5832 * updates data pointers. So when this happens we can
5833 * safely ignore it since we know it's the same extent,
5834 * but just at different logical and physical locations
5835 * (when an extent is fully replaced with a new one, we
5836 * know the generation number must have changed too,
5837 * since snapshot creation implies committing the current
5838 * transaction, and the inode item must have been updated
5839 * as well).
5840 * This replacement of the disk_bytenr happens at
5841 * relocation.c:replace_file_extents() through
5842 * relocation.c:btrfs_reloc_cow_block().
5843 */
5844 if (btrfs_file_extent_generation(leaf_l, ei_l) ==
5845 btrfs_file_extent_generation(leaf_r, ei_r) &&
5846 btrfs_file_extent_ram_bytes(leaf_l, ei_l) ==
5847 btrfs_file_extent_ram_bytes(leaf_r, ei_r) &&
5848 btrfs_file_extent_compression(leaf_l, ei_l) ==
5849 btrfs_file_extent_compression(leaf_r, ei_r) &&
5850 btrfs_file_extent_encryption(leaf_l, ei_l) ==
5851 btrfs_file_extent_encryption(leaf_r, ei_r) &&
5852 btrfs_file_extent_other_encoding(leaf_l, ei_l) ==
5853 btrfs_file_extent_other_encoding(leaf_r, ei_r) &&
5854 btrfs_file_extent_type(leaf_l, ei_l) ==
5855 btrfs_file_extent_type(leaf_r, ei_r) &&
5856 btrfs_file_extent_disk_bytenr(leaf_l, ei_l) !=
5857 btrfs_file_extent_disk_bytenr(leaf_r, ei_r) &&
5858 btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) ==
5859 btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) &&
5860 btrfs_file_extent_offset(leaf_l, ei_l) ==
5861 btrfs_file_extent_offset(leaf_r, ei_r) &&
5862 btrfs_file_extent_num_bytes(leaf_l, ei_l) ==
5863 btrfs_file_extent_num_bytes(leaf_r, ei_r))
5864 return 0;
5865 }
5866
95155585
FM		 5867 inconsistent_snapshot_error(sctx, result, "extent");
5868 return -EIO;
5869 }
31db9f7c
AB		 5870
5871 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5872 if (result != BTRFS_COMPARE_TREE_DELETED)
5873 ret = process_extent(sctx, sctx->left_path,
5874 sctx->cmp_key);
5875 }
5876
5877 return ret;
5878}
5879
ba5e8f2e
JB		 5880static int dir_changed(struct send_ctx *sctx, u64 dir)
5881{
5882 u64 orig_gen, new_gen;
5883 int ret;
5884
5885 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5886 NULL, NULL);
5887 if (ret)
5888 return ret;
5889
5890 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5891 NULL, NULL, NULL);
5892 if (ret)
5893 return ret;
5894
5895 return (orig_gen != new_gen) ? 1 : 0;
5896}
5897
5898static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5899 struct btrfs_key *key)
5900{
5901 struct btrfs_inode_extref *extref;
5902 struct extent_buffer *leaf;
5903 u64 dirid = 0, last_dirid = 0;
5904 unsigned long ptr;
5905 u32 item_size;
5906 u32 cur_offset = 0;
5907 int ref_name_len;
5908 int ret = 0;
5909
5910 /* Easy case, just check this one dirid */
5911 if (key->type == BTRFS_INODE_REF_KEY) {
5912 dirid = key->offset;
5913
5914 ret = dir_changed(sctx, dirid);
5915 goto out;
5916 }
5917
5918 leaf = path->nodes[0];
5919 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5920 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5921 while (cur_offset < item_size) {
5922 extref = (struct btrfs_inode_extref *)(ptr +
5923 cur_offset);
5924 dirid = btrfs_inode_extref_parent(leaf, extref);
5925 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5926 cur_offset += ref_name_len + sizeof(*extref);
5927 if (dirid == last_dirid)
5928 continue;
5929 ret = dir_changed(sctx, dirid);
5930 if (ret)
5931 break;
5932 last_dirid = dirid;
5933 }
5934out:
5935 return ret;
5936}
5937
766702ef
AB		 5938/*
5939 * Updates compare related fields in sctx and simply forwards to the actual
5940 * changed_xxx functions.
5941 */
31db9f7c
AB		 5942static int changed_cb(struct btrfs_root *left_root,
5943 struct btrfs_root *right_root,
5944 struct btrfs_path *left_path,
5945 struct btrfs_path *right_path,
5946 struct btrfs_key *key,
5947 enum btrfs_compare_tree_result result,
5948 void *ctx)
5949{
5950 int ret = 0;
5951 struct send_ctx *sctx = ctx;
5952
ba5e8f2e 5953 if (result == BTRFS_COMPARE_TREE_SAME) {
16e7549f
JB		 5954 if (key->type == BTRFS_INODE_REF_KEY ||
5955 key->type == BTRFS_INODE_EXTREF_KEY) {
5956 ret = compare_refs(sctx, left_path, key);
5957 if (!ret)
5958 return 0;
5959 if (ret < 0)
5960 return ret;
5961 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5962 return maybe_send_hole(sctx, left_path, key);
5963 } else {
ba5e8f2e 5964 return 0;
16e7549f 5965 }
ba5e8f2e
JB		 5966 result = BTRFS_COMPARE_TREE_CHANGED;
5967 ret = 0;
5968 }
5969
31db9f7c
AB		 5970 sctx->left_path = left_path;
5971 sctx->right_path = right_path;
5972 sctx->cmp_key = key;
5973
5974 ret = finish_inode_if_needed(sctx, 0);
5975 if (ret < 0)
5976 goto out;
5977
2981e225
AB		 5978 /* Ignore non-FS objects */
5979 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5980 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5981 goto out;
5982
31db9f7c
AB		 5983 if (key->type == BTRFS_INODE_ITEM_KEY)
5984 ret = changed_inode(sctx, result);
96b5bd77
JS		 5985 else if (key->type == BTRFS_INODE_REF_KEY ||
5986 key->type == BTRFS_INODE_EXTREF_KEY)
31db9f7c
AB		 5987 ret = changed_ref(sctx, result);
5988 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5989 ret = changed_xattr(sctx, result);
5990 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5991 ret = changed_extent(sctx, result);
5992
5993out:
5994 return ret;
5995}
5996
5997static int full_send_tree(struct send_ctx *sctx)
5998{
5999 int ret;
31db9f7c
AB		 6000 struct btrfs_root *send_root = sctx->send_root;
6001 struct btrfs_key key;
6002 struct btrfs_key found_key;
6003 struct btrfs_path *path;
6004 struct extent_buffer *eb;
6005 int slot;
31db9f7c
AB		 6006
6007 path = alloc_path_for_send();
6008 if (!path)
6009 return -ENOMEM;
6010
31db9f7c
AB		 6011 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6012 key.type = BTRFS_INODE_ITEM_KEY;
6013 key.offset = 0;
6014
31db9f7c
AB		 6015 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6016 if (ret < 0)
6017 goto out;
6018 if (ret)
6019 goto out_finish;
6020
6021 while (1) {
31db9f7c
AB		 6022 eb = path->nodes[0];
6023 slot = path->slots[0];
6024 btrfs_item_key_to_cpu(eb, &found_key, slot);
6025
6026 ret = changed_cb(send_root, NULL, path, NULL,
6027 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
6028 if (ret < 0)
6029 goto out;
6030
6031 key.objectid = found_key.objectid;
6032 key.type = found_key.type;
6033 key.offset = found_key.offset + 1;
6034
6035 ret = btrfs_next_item(send_root, path);
6036 if (ret < 0)
6037 goto out;
6038 if (ret) {
6039 ret = 0;
6040 break;
6041 }
6042 }
6043
6044out_finish:
6045 ret = finish_inode_if_needed(sctx, 1);
6046
6047out:
6048 btrfs_free_path(path);
31db9f7c
AB		 6049 return ret;
6050}
6051
6052static int send_subvol(struct send_ctx *sctx)
6053{
6054 int ret;
6055
c2c71324
SB		 6056 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6057 ret = send_header(sctx);
6058 if (ret < 0)
6059 goto out;
6060 }
31db9f7c
AB		 6061
6062 ret = send_subvol_begin(sctx);
6063 if (ret < 0)
6064 goto out;
6065
6066 if (sctx->parent_root) {
6067 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6068 changed_cb, sctx);
6069 if (ret < 0)
6070 goto out;
6071 ret = finish_inode_if_needed(sctx, 1);
6072 if (ret < 0)
6073 goto out;
6074 } else {
6075 ret = full_send_tree(sctx);
6076 if (ret < 0)
6077 goto out;
6078 }
6079
6080out:
31db9f7c
AB		 6081 free_recorded_refs(sctx);
6082 return ret;
6083}
6084
e5fa8f86
FM		 6085/*
6086 * If orphan cleanup did remove any orphans from a root, it means the tree
6087 * was modified and therefore the commit root is not the same as the current
6088 * root anymore. This is a problem, because send uses the commit root and
6089 * therefore can see inode items that don't exist in the current root anymore,
6090 * and for example make calls to btrfs_iget, which will do tree lookups based
6091 * on the current root and not on the commit root. Those lookups will fail,
6092 * returning a -ESTALE error, and making send fail with that error. So make
6093 * sure a send does not see any orphans we have just removed, and that it will
6094 * see the same inodes regardless of whether a transaction commit happened
6095 * before it started (meaning that the commit root will be the same as the
6096 * current root) or not.
6097 */
6098static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6099{
6100 int i;
6101 struct btrfs_trans_handle *trans = NULL;
6102
6103again:
6104 if (sctx->parent_root &&
6105 sctx->parent_root->node != sctx->parent_root->commit_root)
6106 goto commit_trans;
6107
6108 for (i = 0; i < sctx->clone_roots_cnt; i++)
6109 if (sctx->clone_roots[i].root->node !=
6110 sctx->clone_roots[i].root->commit_root)
6111 goto commit_trans;
6112
6113 if (trans)
6114 return btrfs_end_transaction(trans, sctx->send_root);
6115
6116 return 0;
6117
6118commit_trans:
6119 /* Use any root, all fs roots will get their commit roots updated. */
6120 if (!trans) {
6121 trans = btrfs_join_transaction(sctx->send_root);
6122 if (IS_ERR(trans))
6123 return PTR_ERR(trans);
6124 goto again;
6125 }
6126
6127 return btrfs_commit_transaction(trans, sctx->send_root);
6128}
6129
66ef7d65
DS		 6130static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6131{
6132 spin_lock(&root->root_item_lock);
6133 root->send_in_progress--;
6134 /*
6135 * Not much left to do, we don't know why it's unbalanced and
6136 * can't blindly reset it to 0.
6137 */
6138 if (root->send_in_progress < 0)
6139 btrfs_err(root->fs_info,
351fd353 6140 "send_in_progres unbalanced %d root %llu",
66ef7d65
DS		 6141 root->send_in_progress, root->root_key.objectid);
6142 spin_unlock(&root->root_item_lock);
6143}
6144
31db9f7c
AB		 6145long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
6146{
6147 int ret = 0;
6148 struct btrfs_root *send_root;
6149 struct btrfs_root *clone_root;
6150 struct btrfs_fs_info *fs_info;
6151 struct btrfs_ioctl_send_args *arg = NULL;
6152 struct btrfs_key key;
31db9f7c
AB		 6153 struct send_ctx *sctx = NULL;
6154 u32 i;
6155 u64 *clone_sources_tmp = NULL;
2c686537 6156 int clone_sources_to_rollback = 0;
e55d1153 6157 unsigned alloc_size;
896c14f9 6158 int sort_clone_roots = 0;
18f687d5 6159 int index;
31db9f7c
AB		 6160
6161 if (!capable(CAP_SYS_ADMIN))
6162 return -EPERM;
6163
496ad9aa 6164 send_root = BTRFS_I(file_inode(mnt_file))->root;
31db9f7c
AB		 6165 fs_info = send_root->fs_info;
6166
2c686537
DS		 6167 /*
6168 * The subvolume must remain read-only during send, protect against
521e0546 6169 * making it RW. This also protects against deletion.
2c686537
DS		 6170 */
6171 spin_lock(&send_root->root_item_lock);
6172 send_root->send_in_progress++;
6173 spin_unlock(&send_root->root_item_lock);
6174
139f807a
JB		 6175 /*
6176 * This is done when we lookup the root, it should already be complete
6177 * by the time we get here.
6178 */
6179 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6180
2c686537
DS		 6181 /*
6182 * Userspace tools do the checks and warn the user if it's
6183 * not RO.
6184 */
6185 if (!btrfs_root_readonly(send_root)) {
6186 ret = -EPERM;
6187 goto out;
6188 }
6189
31db9f7c
AB		 6190 arg = memdup_user(arg_, sizeof(*arg));
6191 if (IS_ERR(arg)) {
6192 ret = PTR_ERR(arg);
6193 arg = NULL;
6194 goto out;
6195 }
6196
f5ecec3c
DC		 6197 if (arg->clone_sources_count >
6198 ULLONG_MAX / sizeof(*arg->clone_sources)) {
6199 ret = -EINVAL;
6200 goto out;
6201 }
6202
31db9f7c 6203 if (!access_ok(VERIFY_READ, arg->clone_sources,
700ff4f0
DC		 6204 sizeof(*arg->clone_sources) *
6205 arg->clone_sources_count)) {
31db9f7c
AB		 6206 ret = -EFAULT;
6207 goto out;
6208 }
6209
c2c71324 6210 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
cb95e7bf
MF		 6211 ret = -EINVAL;
6212 goto out;
6213 }
6214
e780b0d1 6215 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
31db9f7c
AB		 6216 if (!sctx) {
6217 ret = -ENOMEM;
6218 goto out;
6219 }
6220
6221 INIT_LIST_HEAD(&sctx->new_refs);
6222 INIT_LIST_HEAD(&sctx->deleted_refs);
e780b0d1 6223 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
31db9f7c
AB		 6224 INIT_LIST_HEAD(&sctx->name_cache_list);
6225
cb95e7bf
MF		 6226 sctx->flags = arg->flags;
6227
31db9f7c 6228 sctx->send_filp = fget(arg->send_fd);
ecc7ada7
TI		 6229 if (!sctx->send_filp) {
6230 ret = -EBADF;
31db9f7c
AB		 6231 goto out;
6232 }
6233
31db9f7c 6234 sctx->send_root = send_root;
521e0546
DS		 6235 /*
6236 * Unlikely but possible, if the subvolume is marked for deletion but
6237 * is slow to remove the directory entry, send can still be started
6238 */
6239 if (btrfs_root_dead(sctx->send_root)) {
6240 ret = -EPERM;
6241 goto out;
6242 }
6243
31db9f7c
AB		 6244 sctx->clone_roots_cnt = arg->clone_sources_count;
6245
6246 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6ff48ce0 6247 sctx->send_buf = kmalloc(sctx->send_max_size, GFP_KERNEL | __GFP_NOWARN);
31db9f7c 6248 if (!sctx->send_buf) {
6ff48ce0
DS		 6249 sctx->send_buf = vmalloc(sctx->send_max_size);
6250 if (!sctx->send_buf) {
6251 ret = -ENOMEM;
6252 goto out;
6253 }
31db9f7c
AB		 6254 }
6255
eb5b75fe 6256 sctx->read_buf = kmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL | __GFP_NOWARN);
31db9f7c 6257 if (!sctx->read_buf) {
eb5b75fe
DS		 6258 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
6259 if (!sctx->read_buf) {
6260 ret = -ENOMEM;
6261 goto out;
6262 }
31db9f7c
AB		 6263 }
6264
9f03740a
FDBM		 6265 sctx->pending_dir_moves = RB_ROOT;
6266 sctx->waiting_dir_moves = RB_ROOT;
9dc44214 6267 sctx->orphan_dirs = RB_ROOT;
9f03740a 6268
e55d1153
DS		 6269 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6270
c03d01f3 6271 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
31db9f7c 6272 if (!sctx->clone_roots) {
c03d01f3
DS		 6273 sctx->clone_roots = vzalloc(alloc_size);
6274 if (!sctx->clone_roots) {
6275 ret = -ENOMEM;
6276 goto out;
6277 }
31db9f7c
AB		 6278 }
6279
e55d1153
DS		 6280 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6281
31db9f7c 6282 if (arg->clone_sources_count) {
2f91306a 6283 clone_sources_tmp = kmalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
31db9f7c 6284 if (!clone_sources_tmp) {
2f91306a
DS		 6285 clone_sources_tmp = vmalloc(alloc_size);
6286 if (!clone_sources_tmp) {
6287 ret = -ENOMEM;
6288 goto out;
6289 }
31db9f7c
AB		 6290 }
6291
6292 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
e55d1153 6293 alloc_size);
31db9f7c
AB		 6294 if (ret) {
6295 ret = -EFAULT;
6296 goto out;
6297 }
6298
6299 for (i = 0; i < arg->clone_sources_count; i++) {
6300 key.objectid = clone_sources_tmp[i];
6301 key.type = BTRFS_ROOT_ITEM_KEY;
6302 key.offset = (u64)-1;
18f687d5
WS		 6303
6304 index = srcu_read_lock(&fs_info->subvol_srcu);
6305
31db9f7c 6306 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
31db9f7c 6307 if (IS_ERR(clone_root)) {
18f687d5 6308 srcu_read_unlock(&fs_info->subvol_srcu, index);
31db9f7c
AB		 6309 ret = PTR_ERR(clone_root);
6310 goto out;
6311 }
2c686537 6312 spin_lock(&clone_root->root_item_lock);
5cc2b17e
FM		 6313 if (!btrfs_root_readonly(clone_root) ||
6314 btrfs_root_dead(clone_root)) {
2c686537 6315 spin_unlock(&clone_root->root_item_lock);
18f687d5 6316 srcu_read_unlock(&fs_info->subvol_srcu, index);
2c686537
DS		 6317 ret = -EPERM;
6318 goto out;
6319 }
2f1f465a 6320 clone_root->send_in_progress++;
2c686537 6321 spin_unlock(&clone_root->root_item_lock);
18f687d5
WS		 6322 srcu_read_unlock(&fs_info->subvol_srcu, index);
6323
31db9f7c 6324 sctx->clone_roots[i].root = clone_root;
2f1f465a 6325 clone_sources_to_rollback = i + 1;
31db9f7c 6326 }
2f91306a 6327 kvfree(clone_sources_tmp);
31db9f7c
AB		 6328 clone_sources_tmp = NULL;
6329 }
6330
6331 if (arg->parent_root) {
6332 key.objectid = arg->parent_root;
6333 key.type = BTRFS_ROOT_ITEM_KEY;
6334 key.offset = (u64)-1;
18f687d5
WS		 6335
6336 index = srcu_read_lock(&fs_info->subvol_srcu);
6337
31db9f7c 6338 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
b1b19596 6339 if (IS_ERR(sctx->parent_root)) {
18f687d5 6340 srcu_read_unlock(&fs_info->subvol_srcu, index);
b1b19596 6341 ret = PTR_ERR(sctx->parent_root);
31db9f7c
AB		 6342 goto out;
6343 }
18f687d5 6344
2c686537
DS		 6345 spin_lock(&sctx->parent_root->root_item_lock);
6346 sctx->parent_root->send_in_progress++;
521e0546
DS		 6347 if (!btrfs_root_readonly(sctx->parent_root) ||
6348 btrfs_root_dead(sctx->parent_root)) {
2c686537 6349 spin_unlock(&sctx->parent_root->root_item_lock);
18f687d5 6350 srcu_read_unlock(&fs_info->subvol_srcu, index);
2c686537
DS		 6351 ret = -EPERM;
6352 goto out;
6353 }
6354 spin_unlock(&sctx->parent_root->root_item_lock);
18f687d5
WS		 6355
6356 srcu_read_unlock(&fs_info->subvol_srcu, index);
31db9f7c
AB		 6357 }
6358
6359 /*
6360 * Clones from send_root are allowed, but only if the clone source
6361 * is behind the current send position. This is checked while searching
6362 * for possible clone sources.
6363 */
6364 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6365
6366 /* We do a bsearch later */
6367 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6368 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6369 NULL);
896c14f9 6370 sort_clone_roots = 1;
31db9f7c 6371
e5fa8f86
FM		 6372 ret = ensure_commit_roots_uptodate(sctx);
6373 if (ret)
6374 goto out;
6375
2755a0de 6376 current->journal_info = BTRFS_SEND_TRANS_STUB;
31db9f7c 6377 ret = send_subvol(sctx);
a26e8c9f 6378 current->journal_info = NULL;
31db9f7c
AB		 6379 if (ret < 0)
6380 goto out;
6381
c2c71324
SB		 6382 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6383 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6384 if (ret < 0)
6385 goto out;
6386 ret = send_cmd(sctx);
6387 if (ret < 0)
6388 goto out;
6389 }
31db9f7c
AB		 6390
6391out:
9f03740a
FDBM		 6392 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6393 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6394 struct rb_node *n;
6395 struct pending_dir_move *pm;
6396
6397 n = rb_first(&sctx->pending_dir_moves);
6398 pm = rb_entry(n, struct pending_dir_move, node);
6399 while (!list_empty(&pm->list)) {
6400 struct pending_dir_move *pm2;
6401
6402 pm2 = list_first_entry(&pm->list,
6403 struct pending_dir_move, list);
6404 free_pending_move(sctx, pm2);
6405 }
6406 free_pending_move(sctx, pm);
6407 }
6408
6409 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6410 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6411 struct rb_node *n;
6412 struct waiting_dir_move *dm;
6413
6414 n = rb_first(&sctx->waiting_dir_moves);
6415 dm = rb_entry(n, struct waiting_dir_move, node);
6416 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6417 kfree(dm);
6418 }
6419
9dc44214
FM		 6420 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6421 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6422 struct rb_node *n;
6423 struct orphan_dir_info *odi;
6424
6425 n = rb_first(&sctx->orphan_dirs);
6426 odi = rb_entry(n, struct orphan_dir_info, node);
6427 free_orphan_dir_info(sctx, odi);
6428 }
6429
896c14f9
WS		 6430 if (sort_clone_roots) {
6431 for (i = 0; i < sctx->clone_roots_cnt; i++)
6432 btrfs_root_dec_send_in_progress(
6433 sctx->clone_roots[i].root);
6434 } else {
6435 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6436 btrfs_root_dec_send_in_progress(
6437 sctx->clone_roots[i].root);
6438
6439 btrfs_root_dec_send_in_progress(send_root);
6440 }
66ef7d65
DS		 6441 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6442 btrfs_root_dec_send_in_progress(sctx->parent_root);
2c686537 6443
31db9f7c 6444 kfree(arg);
2f91306a 6445 kvfree(clone_sources_tmp);
31db9f7c
AB		 6446
6447 if (sctx) {
6448 if (sctx->send_filp)
6449 fput(sctx->send_filp);
6450
c03d01f3 6451 kvfree(sctx->clone_roots);
6ff48ce0 6452 kvfree(sctx->send_buf);
eb5b75fe 6453 kvfree(sctx->read_buf);
31db9f7c
AB		 6454
6455 name_cache_free(sctx);
6456
6457 kfree(sctx);
6458 }
6459
6460 return ret;
6461}
Linux 6.x block layer and io_uring tree(s)